Efficient Decision Support Systems Pract

EFFICIENT DECISION
SUPPORT SYSTEMS –
PRACTICE AND
CHALLENGES IN
MULTIDISCIPLINARY
DOMAINS
Edited by Chiang S. Jao
Efficient Decision Support Systems –
Practice and Challenges in Multidisciplinary Domains
Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia
Copyright © 2011 InTech

All chapters are Open Access articles distributed under the Creative Commons
Non Commercial Share Alike Attribution 3.0 license, which permits to copy,
distribute, transmit, and adapt the work in any medium, so long as the original
work is properly cited. After this work has been published by InTech, authors
have the right to republish it, in whole or part, in any publication of which they
are the author, and to make other personal use of the work. Any republication,
referencing or personal use of the work must explicitly identify the original source.
Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted
for the accuracy of information contained in the published articles. The publisher
assumes no responsibility for any damage or injury to persons or property arising out
of the use of any materials, instructions, methods or ideas contained in the book.
Publishing Process Manager Petra Zobic

Technical Editor Teodora Smiljanic
Cover Designer Jan Hyrat
Image Copyright zimmytws, 2010. Used under license from Shutterstock.com
First published August, 2011

Printed in Croatia
A free online edition of this book is available at www.intechopen.com

Additional hard copies can be obtained from [email protected]
Efficient Decision Support Systems – Practice and Challenges in Multidisciplinary Domains,

p. cm.
ISBN 978-953-307-441-2
Contents
Preface IX
Part 1 Applications in Business 1
Chapter 1 Application of Decision Support System

in Improving Customer Loyalty:
From the Banking Perspectives 3
Cho-Pu Lin and Yann-Haur Huang
Chapter 2 Intelligent Agent Technology

in Modern Production and Trade Management 21
Serge Parshutin and Arnis Kirshners
Chapter 3 Modeling Stock Analysts Decision Making:

An Intelligent Decision Support System 43
Harry Zhou
Chapter 4 Business Intelligence – CDMB – Implementing

BI-CMDB to Lower Operation Cost Expenses
and Satisfy Increasing User Expectations 67
Vlad Nicolicin-Georgescu, Vincent Bénatier,
Rémi Lehn and Henri Briand
Chapter 5 Decision Support Systems Application

to Business Processes at Enterprises in Russia 83
Konstantin Aksyonov, Eugene Bykov, Leonid Dorosinskiy,
Elena Smoliy, Olga Aksyonova, Anna Antonova and Irina Spitsina
Chapter 6 Intelligence Decision Support Systems in E-commerce 109

Petr Suchánek, Roman Šperka,
Radim Dolák and Martin Miškus
Chapter 7 Quick Response in

a Continuous-Replenishment-Programme
Based Manufacturer Retailer Supply Chain 131
Shu-Lu Hsu and Chih-Ming Lee
VI Contents
Chapter 8 Collaboration in Decision Making:

A Semi-Automated Support for Managing
the Evolution of Virtual Enterprises 147
Marcus Vinicius Drissen-Silva and Ricardo J. Rabelo
Chapter 9 A Lean Balanced Scorecard Using the Delphi Process:

Enhancements for Decision Making 171
Chuo-Hsuan Lee, Edward J. Lusk and Michael Halperin
Part 2 Applications in Water Resource Management 185
Chapter 10 Linking a Developed Decision Support System

with Advanced Methodologies
for Optimized Agricultural Water Delivery 187
Kristoph-Dietrich Kinzli, David Gensler and Ramchand Oad
Chapter 11 Estimating the Impact on Water Quality under Alternate

Land Use Scenarios: A Watershed Level BASINS-SWAT
Modeling in West Georgia, United States 213
Gandhi Bhattarai, Diane Hite and Upton Hatch
Chapter 12 Flood Progression Modelling and Impact Analysis 227

D. Mioc, F. Anton, B. Nickerson, M. Santos, P. Adda, T. Tienaah,
A. Ahmad, M. Mezouaghi,E. MacGillivray, A. Morton and P. Tang
Chapter 13 Providing Efficient Decision Support for Green

Operations Management: An Integrated Perspective 247
Shaofeng Liu and Meili Jiang
Part 3 Applications in Agriculture 271
Chapter 14 Uncertainty Analysis Using Fuzzy Sets

for Decision Support System 273
Mohd Najib Mohd Salleh, Nazri Mohd Nawi and Rozaida Ghazali
Chapter 15 A Web-Based Decision Support System

for Surface Irrigation Design 291
José M. Gonçalves, André P. Muga and Luis Santos Pereira
Part 4 Applications in Spatial Management 319
Chapter 16 A Decision Support System (FMOTS)

for Location Decision of Taxicab Stands 321
Ebru Vesile Ocalir, Ozge Yalciner Ercoskun and Rifat Tur
Chapter 17 Sensor Network and GeoSimulation:

Keystones for Spatial Decision Support Systems 337
Nafaâ Jabeur, Nabil Sahli and Hedi Haddad
Contents VII
Part 5 Applications in Risk and Crisis Management 357
Chapter 18 Emerging Applications of Decision Support Systems (DSS)

in Crisis Management 359
Gabriela Prelipcean and Mircea Boscoianu
Chapter 19 Risk Analysis and Crisis Scenario Evaluation

in Critical Infrastructures Protection 377
Vittorio Rosato, Vincenzo Artale,
Giovanna Pisacane, Gianmaria Sannino,
Maria Vittoria Struglia, Aberto Tofani and Eddy Pascucci
Part 6 Miscellaneous Case Studies 395
Chapter 20 Applications of Decision Support System

in Aviation Maintenance 397
Peng Zhang, Shi-Wei Zhao, Bin Tan, Li-Ming Yu and Ke-Qiang Hua
Chapter 21 Evaluating the Power Consumption

in Carbonate Rock Sawing Process
by Using FDAHP and TOPSIS Techniques 413
Reza Mikaeil, Mohammad Ataei and Reza Yousefi
Chapter 22 A Supporting Decisions Platform for the Design

and Optimization of a Storage Industrial System 437
Riccardo Manzini, Riccardo Accorsi,
Laura Pattitoni and Alberto Regattieri
Chapter 23 Challenges in Climate-Driven

Decision Support Systems in Forestry 459
Oleg Panferov, Bernd Ahrends, Robert S. Nuske,
Jan C. Thiele and Martin Jansen
Preface
Series Preface
This series is directed to diverse managerial professionals who are leading the
transformation of individual domains by using expert information and domain
knowledge to drive decision support systems (DSSs). The series offers a broad range of
subjects addressed in specific areas such as health care, business management,
banking, agriculture, environmental improvement, natural resource and spatial
management, aviation administration, and hybrid applications of information
technology aimed to interdisciplinary issues.
This book series is composed of three volumes: Volume 1 consists of general concepts
and methodology of DSSs; Volume 2 consists of applications of DSSs in the biomedical
domain; Volume 3 consists of hybrid applications of DSSs in multidisciplinary
domains. The book is shaped decision support strategies in the new infrastructure that
assists the readers in full use of the creative technology to manipulate input data and
to transform information into useful decisions for decision makers. This book series is
dedicated to support professionals and series readers in the emerging field of DSS.
Preface
Book Volume 2 extends the applications of decision support systems (DSSs) to

regulate various resources in dealing with business, water resource, agriculture, space,
risks/crisis, and other interdisciplinary issues. Design and development of such hybrid
types of DSSs need to integrate interdisciplinary knowledge, resource data, as well as
a variety of surrounding interdisciplinary parameters (for example behavioral-,
economic-, environmental-, and social-related factors) and to effectively improve
resource management. This book can be used in case-study courses related to decision
support systems (DSSs). It may be used by both undergraduate senior and graduate
students from diverse computer-related fields. After reading this book, the readers
should be able to draw a clear picture about how to apply DSSs in multidisciplinary
fields. It will also assist professionals in business, spatial, agricultural, aviation and
other non-biomedical fields for self-study or reference.
Section 1, including Chapter 1 through 9, illustrates several applications of intelligent

DSSs for business purposes. Chapter 1 focuses on customer relationship management
X Preface
and its adoption in banking industry. Chapter 2 focuses on the improvement of

modern production and trade management using intelligent agent technologies.
Chapter 3 presents a novel DSS performed to analyze stocks, calling market turns and
making recommendations by combining knowledge-based problem solving with case-
based reasoning and fuzzy logic inference. Chapter 4 presents an efficient business
DSS, integrated with the configuration management database, to reduce operational
cost expenses and promote the satisfactory level of client expectations. Chapter 5
focuses on enhancing decision making for managers to re-engineer existing business
processes and enterprise activity analysis. Chapter 6 presents an intelligent DSS that
assists managers in improving customer needs who are using e-commerce systems
based on collected data relating to their behavior. Chapter 7 presents another
intelligent DSS that provides quick responses in a continuous replenishment program
to reduce the stock level throughout the supply chain. Chapter 8 presents a semi-
automated DSS protocol for collaborative decision making in managing virtual
enterprises with business globalization and sharing supply chains. Chapter 9 presents
a lean balanced scorecard method to enhance the organization’s competitive
advantages for decision making in its financial dimension.
Section 2, including Chapter 10 through 13, presents a set of DSSs aimed to water
resource management and planning issues. Chapter 10 focuses on improving water
delivery operations in irrigation systems through the innovative use of water DSSs.
Chapter 11 uses one of the latest biophysical watershed level modeling tools to
estimate the effects of land use change in water quality. Chapter 12 presents a flood
prediction model with visual representation for decision making in early flood
warning and impact analysis. Chapter 13 presents an integrated sustainability analysis
model that provides holistic decision evaluation and support addressed the
environmental and social issues in green operations management.
Section 3, including Chapter 14 through 15, presents two DSSs applied in the
agricultural domain. Chapter 14 integrates agricultural knowledge and data
representation using fuzzy logic methodology that generalizes decision tree
algorithms when an uncertainty (missing data) is existed. Chapter 15 presents Web-
based DSS models in respectively dealing with surface irrigation and sustainable pest
management in the agricultural domain.
Section 4, including Chapter 16 through 17, illustrates two spatial DSSs applied to
multidisciplinary areas. Chapter 16 presents the DSS for location decisions of taxicab
stands in an urban area with the assistance of geographical information system (GIS)
and fuzzy logic techniques. Chapter 17 presents the spatial DSS integrated with GIS
data to assist managers identifying and managing impending crisis situations.
Section 5, including Chapter 18 and 19, emphasizes the importance of DSS applications
in risk analysis and crisis management. In a world experiencing recurrent risks and
crises, it is essential to establish intelligent risk and crisis management systems at the
managerial level that supports appropriate decision making strategies and reduces the
Preface XI
occurrence of uncertainty factors. Chapter 18 illustrates the DSS for industrial

managers to support risk analysis and prediction of critical resource distribution and
infrastructure protection. Chapter 19 focuses on the use of the hybrid DSS in
improving critical decision making process for treating extreme risk and crisis event
management.
This book concludes in Section 6 that covers a set of DSS applications adopted in
aviation, power management, warehousing, and climate monitoring respectively.
Chapter 20 presents an aviation maintenance DSS to promote the levels of
airworthiness, safety and reliability of aircrafts and to reduce indirect costs due to
frequent maintenance.Chapter 21 introduces a fuzzy logic DSS to assist decision
makers in better rankings of power consumption in rock sawing process. Chapter 22
presents a DSS for efficient storage allocation purpose that integrates management
decisions in a warehousing system. Chapter 23 investigates challenges in climate-
driven DSS in forestry. By including prior damage data and forest management
activities caused by changes in weather and forest structure, a DSS model is presents
to assist forest managers in assessing the damage and projecting future risk factors in
monitoring the climate change.
Chiang S. Jao
Transformation, Inc. Rockville
Maryland University of Illinois (Retired) Chicago
Illinois
Part 1
Applications in Business
1
Application of Decision Support System in

Improving Customer Loyalty:
From the Banking Perspectives
Cho-Pu Lin and Yann-Haur Huang
St. John's University/Department of Marketing & Logistics Management, Taipei,
Taiwan
1. Introduction
The focus of this research is on customer relationship management (CRM) and its adoption
in Taiwan’s banking industry. The concept of CRM and its benefits have been widely
acknowledged. Kincaid (2003, p. 47) said, “CRM deliver value because it focuses on
lengthening the duration of the relationship (loyalty).” Motley (2005) found that satisfiers
keep customers with the bank while dissatisfiers eventually chase them out. Earley (2003)
pointed out the necessity of holistic CRM strategies for every company because today even
the most established brands no longer secure lasting customer loyalty. It seems clear that
customer relationship management is critical for all service firms, including the banks.
1.1 Research motivations

Nowadays, for banking industry, one way to keep being profitable is to retain the existing
customers, and one way to keep existing customers is to satisfy them. According to the
80/20 rule in marketing, 80% of sales comes from 20% customers. In addition, Peppers and
Rogers (1993) pointed out that the cost of discovering new customers is six to nine time
higher than that of keeping existing customers. Thus, it is critical to maintain customer
loyalty. According to Lu (2000), through the adoption of CRM systems, companies could (1)
find the best customers, (2) keep existing customers, (3) maximize customer value, and (4)
develop effective risk management. In addition, successful CRM will create huge values for
companies through improved customer retention rate. Therefore, it is worthwhile to
conduct an empirical study on the adoption of CRM systems in Taiwan’s banking industry.
1.2 Statement of problems

A major challenge banks are facing today is to implement new technology solutions that
will provide more responsiveness and flexibility to their business clients. Many corporations
are now conducting their transactions with fewer banks. Dobbins (2006, p. 1) said, “The
challenge for all banks, large and small, is not only to create a centre of excellence with
established international standards of communication, but also to reconstruct and automate
their business processes to maximize efficiency.”In addition, a number of researchers found
that implementation of technology such as CRM do not guarantee that the expected results
will be achieved. In fact, a number of studies indicate that firms have suffered failures
4 Efficient Decision Support Systems – Practice and Challenges in Multidisciplinary Domains
organizational problems (53%) or an inability to access the most relevant information

technologies (40%) (Ernst & Young, 2001). In Taiwan’s banking industry, there is also a CRM
adoption issue. Most banks do not quite understand CRM. F. H. Lin and P. Y. Lin (2002, p.
528) said that according to a CRM-application survey of Taiwan’s industries by ARC
Consulting, 90% of the industry (most of which consists of banks) knew about CRM while
only 64% understood the intension of CRM. Furthermore, only 10% of Taiwan’s industries
have already established the CRM systems. Hence, there is still much room for improving
when it comes to CRM adoption in Taiwan’s banking industry. Huang and Lu (2003, p. 115)
noted that recently, the competition in local banking industry has become more acute as
branches of banks are multiplying and as Taiwan has become a member of World Trade
Organization (WTO). Given these environmental changes, implementing a CRM system is
becoming a pressing item on local banks’ agendas. At this juncture, then, addressing the
importance of CRM adoption in Taiwan’s banking industry is indeed a worthy cause.
Huang and Lu (2003) further suggested that Taiwan’s financial institutions, in this
customer-oriented age, should not be limited to operational strategies that are product-
oriented. Instead, according to these authors, they need to gauge customers’ favorites
accurately and find out the potential needs of their customers. Only by doing so, they would
be able to promote their financial products with their customers. In the future, the focus of
core competitive strategies in Taiwan’s banking industry will shift from “products” to
“customers.” Thus, integrating front and back processes and understanding the intension
and implementation of CRM have become an urgent task for Taiwan’s banking industry.
Therefore, it is imperative to explore the factors that would affect CRM adoption in
Taiwan’s banking industry and to solve the problems arising therein.
2. Literature review
A body of previous studies on this topic lends a solid basis to the present investigation. This
literature covers the following sub-areas: (1) introduction of customer relatioship
management, (2) measurement of success with CRM, (3) CRM technologies and success with
CRM.
2.1 Introduction of Customer Relationship Management

Customer relationship management (CRM) is now a major component of many
organizations’ E-commerce strategy. Trepper (2000) thought that CRM could classified as (1)
operational (e.g., for improving customer service, for online marketing, and for automating
the sales force), (2) analytical (e.g., for building a CRM data warehouse, analyzing customer
and sales data, and continuously improving customer relationships), or (3) collaborative
(e.g., for building Web and online communities, business-to-business customer exchanges
and personalized services).
2.2 Measurement of success with CRM

Every bank, regardless of its size, would pride itself on providing high-quality customer
service. However, the challenge is that the benchmarks for high-quality customer services
are changing dramatically, to the extent that yesterday’s standards will not enable a bank to
win today’s customers. Shermach (2006) considered identifying customer expectation lines
and reaching those lines the most important tasks for the banking industry. Sheshunoff
(1999) likewise argued that banks will need to develop new tools and strategies in an effort
to maintain their reputation and that those tools and strategies will likely involve CRM.
Application of Decision Support System in Improving
Customer Loyalty: From the Banking Perspectives 5
Once a CRM system has been implemented, an ultimate question arises. That is, whether the
CRM adoption can be considered a success. Since there exist many measures to assess IT
adoption success, care must be taken when selecting the appropriate approach for analyzing
CRM implementations with banks. In the present study, the author chose to determine CRM
implementation by process efficiency and IT product quality. These approaches make it
possible to highlight the goals that need to be managed more actively during the CRM
introduction process to make the adoption of CRM systems a success.
2.2.1 Process efficiency

Levitt (1960) believed that customer satisfaction is the ultimate goal for every business
because, for most industries, unsatisfied customers will eventually leave. To the same effect,
Motley (1999, p.44) said, “In most industries, unhappy customers turn to competitors who
promise more. This also happens in banking; it just takes longer.” Cambell (1999, p. 40) said,
“As the competition in banking and financial services industries continues to increase,
achieving the highest possible levels of customer satisfaction will be critical to continued
success.” Along this line of thinking, Jutla and Bodorik (2001) have suggested that three
customer metrics – customer retention, customer satisfaction, and customer profitability—
could be used to measure CRM performance. These assertions have been corroborated by
similar studies conducted in Taiwan. For instance, according to F. H. Lin and P. Y. Lin.
(2002), with CRM systems, businesses could gain higher customer loyalty and higher
customer value. Lu, Hsu and Hsu (2002) also argued that Taiwan’s banking industry should
utilize customer data analyses and multiple communication channels to increase customer
satisfaction. In addition to customer satisfaction, process efficiency has been considered a
norm for successful CRM adoption. Burnett (2004) stated that real-time assessment to CRM
could increase efficiency, responsiveness, and customer loyalty, because it could make
customer information available anytime and anywhere. As long as people continue to
believe that CRM is equal to process efficiency, the real benefits of CRM will stay beyond
reach. Baldock (2001) suggested that banks need to implement additional software that
could do two things: (1) deciding what product or message should be offered to which
customer and (2) delivering these product recommendations in real-time through all of the
bank’s channels allowing CRM to combine efficiency and effectiveness. The relationships
that customers have with banks are becoming increasingly complex, so complex that the
data that a customer’s profile is based on needs to be updated continually, in real time. Luo,
Ye, and Chio (2003) felt that businesses in Taiwan could involve customers in CRM by one-
on-one marketing through Internet. By doing so, businesses could (1) achieve accuracy of
information, (2) increase information value, (3) lower work-force demands, and (4) increase
process efficiency. Chang and Chiu (2006, p. 634) also claimed that “it is also very
meaningful to investigate factors influencing the efficiency of Taiwan banks.” It seems clear,
then, that for Taiwan’s banking industry process efficiency is an important variable in
determining the success of CRM.
2.2.2 Product quality

Research on software engineering has identified certain IT product-quality dimensions, such
as human engineering, portability, reliability, maintainability, and efficiency. A variety of
metrics to assess these dimensions of CRM system quality has also been developed and
validated. As this stream of research continues to evolve, its emphasis has been on the
engineering characteristics of the CRM system while limited attention has been paid to
assessing and enhancing users’ subjective evaluations of the system (Yahaya, Deraman,
and Hamdan, 2008; Ortega, Perez, and Rojas, 2003). A key management objective when
dealing with information products is to understand the value placed by users on these IT
products. In contrast to the technical focus of CRM system’s quality assurance research,
customer satisfaction is an important objective of TQM (total quality management)
initiatives. Customers have specific requirements, and products/services that effectively
meet these requirements are perceived to be of higher quality (Deming, 1986; Juran, 1986). A
similar perspective is evident in the IS management studies, where significant attention has
been paid to understanding user requirements and satisfying them. Research has focused
on identifying the dimensions of developing reliable and valid instruments for the
measurement of this construct (Bailey and Pearson, 1983; Galletta and Lederer, 1989; Ives,
Olson, and Baroudi, 1983). It seems clear that, for Taiwan’s banking industry, product
quality is an important variable in determining the success of CRM.
2.3 CRM technologies and success with CRM

This section presents an exploration of the applications of certain technologies and their
effects on CRM adoption in Taiwan’s banking industry.
2.3.1 Developing information technologies in the banking industry

Chowdhury (2003) stated that banks are widely acknowledged to be heavily dependent on
information technologies. This is so especially in the United States, where banking is
considered a most IT-intensive industry. Such dependence is readily evidenced by the
proportion of computer equipment and software employed by the U.S. banks in their day-
to-day operation. Some researchers found that for the banks the IT spendings were as high
as 8% of the industry’s average revenue, with the average ratio of IT spending to revenue
being approximately 2 to 3% for other industries. In addition, IT spendings represent about
one third of the average operating expenses (Berensmann, 2005; Rebouillon and Muller,
2005). Furthermore, this pattern of extensive IT usage in banking is assumed to be similar
across countries (Zhu, Kraemer, Xu, and Dedrick, 2004). However, this high level of IT has
turned out to be a problem for many banks. Banks have traditionally relied on in-house
development since the early days of electronic data processing. However, the emergent
applications in the 1970s are conceived of as legacy applications today (Moormann, 1998).
These legacy systems have historically been built around the banks’ product lines, e.g.,
loans, deposits and securities, with very limited cross-functional information flow
(Chowdhury, 2003). It is well known that in banking there have been substantial IT-driven
business innovations which not only boost bank efficiency but also benefited the consumers.
Automated teller machines and electronic fund transfer are among these innovations (Dos
Santos and Peffers, 1995). Because of these cutting-edge innovations, banks have not been
motivated to redesign their IT infrastructure in a modular, integrated and more flexible
manner (Betsch, 2005). Consequently, most banks tend to use a number of isolated solutions
to perform even standard business activities, such as loan application processing, instead of
seamlessly integrating business processes and the underlying information systems.
Nevertheless, prevailing legacy systems create problems not merely in data processing.
They are also a major cost driver, as approximately two thirds of a bank’s IT budget
typically goes into the maintenance of legacy applications (Rebouillon & Muller, 2005).
Therefore, existing IT infrastructures in banks are often obstacles to efficiently running a
bank, in spite of a heavy investment in IT. Veitinger and Loschenkohl (2005) asserted that
modular and flexible CRM/ERP systems could mitigate some of the legacy-application
problems with the banks since CRM/ERP systems may enable the banks to align their
business processes more efficiently. Although the CRM systems are able to provide huge
direct and indirect benefits, potential disadvantages, e.g., lack of appropriate CRM/ERP
package, can be enormous and can even negatively affect a bank’s Success with CRM
adoption. Scott and Kaindl (2000) found that approximately 20% of functionalities/modules
needed are not even included in the systems. This rate may arguably be even higher in
banking, possibly creating the impression that CRM/ERP systems with appropriate
functionality coverage are not available at all. In addition to the system package, the
pressure from CRM/ERP vendors to upgrade is an issue. This pressure has become a
problem due to recent architectural changes that systems of major CRM/ERP vendors face.
That is, with the announcement of mySAP CRM as the successor of SAP R/3 and its arrival
in 2004, SAP’s maintenance strategy has been extensively discussed while users fear a strong
pressure to upgrade (Davenport, 2000; Shang & Seddon, 2002).
2.3.2 Customization of CRM functions/modules

Chang and Liu (2005) believed that customization is a key to gaining the emotional promises
from the customers in Taiwan’s banking industry and that such promises can increase
customer loyalty (i.e., decreasing customer churn rate). Due to the fast-changing financial
environment, competition and high customer churn rates, Taiwan’s banking industry
should focus on long-term customer relationships and strive to raise the customer retention
rate. Li (2003) found that, in order to enhance customer services, the CRM systems should be
adjusted when a company adopts them. In other words, companies should “customize”
their CRM systems according to the demands of their own customers. If the customization
of the CRM system cannot satisfy the customers, companies could further consider
“personalizing” their CRM systems. Chang and Liu (2005, p. 510) stated that if a bank has
effective ability to deal with a contingency or emergency and could provide professional
financial plans as well as innovative services with its financial products, these customization
functions could help increase the reorganization of customer values. Liao (2003) suggested
that companies, when adopting CRM systems, should emphasize customer differentiation,
customer loyalty, customer lifetime values, one-on-one marketing, and customization. Born
(2003, p. 1) argued that “the greater the CRM functionality, the less customization of the
CRM system required.” Huang (2004, p. 104) also said, “When the goal of implementing
CRM is to improve operational efficiency, the management should try to minimize the
customization of a CRM system.” Obviously, there is a difference in customization in the
banking industry between the United Sates and Taiwan. Companies in Taiwan would more
likely customize their CRM systems in order to satisfy their customers’ needs. In U.S.,
though, customization is not as widespread. A main goal of the present study is to explore
the relationship between customization of CRM functions/modules and the adoption of
CRM’s in Taiwan’s banking industry.
2.3.3 Selecting the CRM vendors

Once a company starts to implement a CRM system, it will look to the CRM vendors as a
barometer of their CRM system’s health. However, at any given point in time, one may
choose from a host of CRM vendors (e.g., Oracle/People Soft, Siebel, Broadvision, NetSuite
10, SAP, E.pipjany, Rightnow CRM 7.0, Salesforce.com, Salesnet.com, and Microsoft).
According to Hsu and Rogero (2003), in Taiwan, the major CRM vendors are the well-
known brand names, such as IBM, Oracle, Heart-CRM, and SAP. Confronted with so many
vendors, organizations that wish to implement a CRM system will need to choose wisely.It
has been recognized that the service ability of vendors would affect the adoption of
information technologies (Thong and Yap, 1996). Lu (2000) posited that the better the
training programs and technical support supplied by a CRM vendor, the lower the failure
rate will be in adopting a CRM system. It is generally believed that CRM vendors should
emphasize a bank’s CRM adoption factors such as costs, sales, competitors, expertise of
CRM vendors, managers’ support, and operational efficiency (H. P. Lu, Hsu & Hsu, 2002).
For banks that have not yet established a CRM system, CRM vendors should not only
convince them to install CRM systems, but also supply omnifarious solutions and complete
consulting services. For those banks that have already established a CRM system, in
addition to enhancing after-sales services, the CRM vendors should strive to integrate the
CRM systems into the legacy information systems.When companies choose their CRM
vendors, they face many alternatives (Tehrani, 2005). Borck (2005) believed that the current
trend among CRM vendors is moving toward producing tools that satisfy the basic needs of
sales and customer support. The ultimate goal of a bank in choosing a CRM vendor is to
enable the IT application intelligence to drive the revenues to the top line. Based on these
views, it may be concluded that choosing a reputable CRM vendor and benefiting from its
expertise are prerequisite for a successful CRM adoption in Taiwan’s banking industry.
2.3.4 Conducting DSS (Decision Support System)

Since a bank offers countless daily services — offering credit cards, loans, mortgages — it is
very risky for it to offer such services to customers they know nothing about. It is generally
acknowledged that banks need to ensure the reliability of their customers (Hormazi & Giles,
2004). The concept here is simple: the banking industry has a need to reduce the risks from
issuing credit cards or loans to customers who are likely to default. An example, given by
Cocheo (2005), is of a bank that found a borrower appealing until receiving a court notice
saying the customer had filed a bankruptcy. As a solution to problems such as this, the
artificial neural network (ANN) has been widely adopted. According to Fadlalla and Lin
(2001), an ANN is a technology using a pattern-recognition approach that has established
itself in many business applications, including those in the U.S. banking industry. According
to Turban, Aronson, and Liang (2004), an ANN is able to learn patterns in the data
presented during training and will automatically apply what it has learned to new cases.
One important application of ANN is in bank loan approvals because an ANN can learn to
identify potential loan defaulters from the ascertained patterns. Turban et al. (2004) further
observed that one of the most successful applications of ANN is in detecting unusual credit
spending patterns, thereby exposing fraudulent charges. Therefore, conducting DSS
systems, such as ANN technology, to analyze the customer data should be a critical step
toward CRM adoption in Taiwan’s banking industry.
2.4 Summary
In conclusion, this author, based on the previous studies, would like to test the relationship
between CRM technologies and success with CRM. Table 1 sums up the factors selected
from the literature review to be tested in the present study.
Factors
A. CRM deployment.
Measurements of CRM success B. Process efficiency.
C. Product quality.
CT1 Conducting the decision support system (DSS)s
CT2 Customizing CRM functions/modules.
CT3 Choosing reputed CRM vendors.
CT4 Drawing on the expertise of CRM vendors.
CT5 Pressure from CRM(ERP) vendor to upgrade.
CT6 Non-availability of appropriate CRM(ERP) packages.
Table 1. A Summary of Variables Included in the Present Study
3. Research methodology
The objective of this study is to examine the impact of a variety of relevant factors on the
CRM success in Taiwan’s banking industry. This section presents the research question,
hypothesis, and statistical techniques for hypothesis testing.
3.1 The research question and the hypothesis

Since the challenges that the banks are facing today are to implement and support new
technological solutions that will enable them to be more responsive and flexible to their
business clients, the present study seeks to answer the following research question:
Research Question: What are the critical factors that explain the degree of success in the
adoption of a CRM system in Taiwan’s banking industry?
The hypothesis derived from this research question is displayed in the following:
H1: The CRM technology will be positively associated with successful CRM adoption in
Taiwan's Banking industry.
3.2 Research design

First, based on the findings from the literature review, an exploratory study (i.e., focus
group interviews) was conducted to discover the nature of the problems and to generate
possible solutions to the research question. Second, on the basis of the findings from the
focus group interviews, a quantitative analysis was conducted using survey and statistical
methods to identify possible answers to the research question.
3.2.1 Research population and samples

The population for the present survey consists of the local banking industry in Taiwan,
including domestic banks and local branches of foreign banks. The information about these
banks came from the official website of Financial Supervisory Commission, Executive Yuan
in Taiwan. There are 37 domestic banks and 31 local branches of foreign banks in Taiwan.
The research samples are the CRM users in IT or Customer-Service departments in those
banks.
3.3 Statistical methods

Two statistical methods were conducted in this present study: (1) EFA (exploratory factor
analysis) and (2) SEM (structural equation modeling).
3.3.1 Introduction
Figure 1 summaries the processes of two statistical methods in the present study.
Fig. 1. A Summary of Statistical Methods
4. Results
In the present study, this author would like to test the causal relationship between CRM
technologies and success with CRM. The hypothesis of the study, the variables involved and
the statistical methods are described in the following.
Tested hypothesis: H1: The CRM technology will be positively associated with successful
CRM adoption in Taiwan's Banking industry
Observed variables: CT1 (Conducting the decision support system), CT2 (Customizing
CRM functions/modules), CT3 (Choosing reputed CRM vendors), CT4
(Drawing on the expertise of CRM vendors), CT5 (Pressure from
CRM/ERP package), CT6 (Non-availability of appropriate CRM/ERP
packages), CRM deployment, Process efficiency, and Product quality
Latent variables: CRM technologies; success with CRM
Statistical method: Structural-equation modeling
The writing in the remaining part of this section is organized into the following subsections:
(a) offending estimates, (b) construct reliability and average variance extracted and (c)
goodness-of-fit.
a. Offending Estimates
As shown in Table 1, the standard error ranges from 0.001 to 0.006. There is no negative
standard error in this model.
Estimate S.E. C.R. P

CRM technologies .016 .003 4.481 ***
e10 .036 .006 6.453 ***
e1 .034 .003 10.318 ***
e2 .031 .003 10.202 ***
e3 .013 .001 9.132 ***
e4 .002 .001 2.015 .044*
e5 .011 .001 8.380 ***
e6 .025 .003 9.803 ***
e7 .025 .003 7.263 ***
e8 .016 .004 3.952 ***
e9 .029 .003 8.350 ***
Note: Estimate = Unstandardized Coefficients; SE = Standard Error; C.R. = Critical Ration; P =
Significance: *p<0.05, **p<0.01, ***p<0.001
Table 1. Variances: Default model
Also, in Table 2, the standardized regression weight ranges from 0.381 to 0.983. All of the
standardized regression weights are below 1.0. Thus, it is clear that the offending estimates
do not occur in this model.
Estimate
Success with CRM <--- CRM technologies .381
CT1 <--- CRM technologies .563
CRM Deployment <--- Success with CRM .791
Process Efficiency <--- Success with CRM .895
Product Quality <--- Success with CRM .738
Note: Estimate = Standardized Coefficients
Table 2. Standardized Regression Weights: Default model
b. Construct reliability and Average variance extracted
Construct reliability of CRM Technologies was calculated (with a suggested lower limit of
0.70) by using this formula:
∑
∑ ∑
(1)
is the construct reliability of CRM technologies. Let be the standardized loadings (or
the standardized coefficients) for CRM Technologies. Let be the error variance for CRM
Technologies. Based on the data in Table 3, the construct reliability of CRM technologies is:
. . . . . .
= (2)
. . . . . . . . . . . .
= 0.9863
Construct reliability of success with CRM in this model was calculated (with a suggested
lower limit of 0.70) by using this formula:
∑
∑ ∑
(3)
is the construct reliability of success with CRM. Let be the standardized loadings (or
the standardized coefficients) for success with CRM. Let be the error variance for success
with CRM. Based on the data in Table 3, the construct reliability of success with CRM in this
model is:
. . .
= = 0.9882 (4)
. . . . . .
The average variance extracted of CRM Technologies was calculated (with a suggested lower
limit of 0.50) by using this formula:
∑
∑ ∑
(5)
is the average variance extracted of CRM Technologies. Based on the data in Table 3, the
average variance extracted of CRM Technologies is:
. . . . . .
= (6)
. . . . . . . . . . . .
= 0.9708
The average variance extracted of success with CRM in this model was calculated (with a
suggested lower limit of 0.50) by using this formula:
∑
∑ ∑
(7)
is the average variance extracted of success with CRM. Based on the data in Table 3, the
average variance extracted of success with CRM is:
. . .
= = 0.9657 (8)
. . . . . .
To sum up, the construct reliability and average variance extracted in this model are
considered acceptable as all of them are much higher than suggested values (0.70 and 0.50).
This means, the inner quality of this model is acceptable and deserves further analyses.
c. Goodness-of-fit
Factor loadings Error variances

Latent Variable: CRM technologies
Observed Variables: CT1 0.563 0.034
Latent Variable: Success with CRM
Observed Variables: CRM deployment 0.791 0.025
Observed Variables: Process efficiency 0.895 0.016
Observed Variables: Product quality 0.738 0.029
Table 3. Factor Loadings and Error Variances: Default model
Figure 2 is a graphical representation of this model. With 45 distinct sample moments and
21 distinct parameters to be estimated, the total number of degrees of freedom is 24 (45 - 21).
This model fits the hypothesized data structure well. The chi-square is 45.419 with 24
degrees of freedom, and p =. 005. Although the p-value is below 0.05, all other critical fit
index values are above the recommended critical value of .90 (GFI =.958, AGFI =.921, NFI =
.969, TLI = .978) with RMSEA = .064<0.10.
Moreover, a few more findings came out of the testing of this model:
i. The analyzed data in Table 4 indicate that CRM technology has a significant influence
over success with CRM (p<0.001).
ii. The assumption that the six observed variables (i.e., CT1 ~ CT6) would be positively
influenced by CRM technologies was proved. The data in Table 4 indicate that the
relationships between all of the six observed variables and CRM technologies were
significant (p<0.001).
iii. The assumption that three observed variables (i.e., CRM deployment, process efficiency
and product quality) would be positively influenced by success with CRM was proved.
The data in Table 4 indicate the relationships between all of the three observed variables
on the one hand and success with CRM on the other hand were significant (p<0.001).
iv. According to the standard regression weights in Table 2, the relationships among the
variables in this model could displayed with the formulas listed below:
a. η(Success with CRM) = 0.38ξ(CRM technologies) + e10
b. Y1(CRM development) = 0.79η(Success with CRM) + e7
c. Y2(Process efficiency) = 0.89η(Success with CRM) + e8
d. Y3(Product Quality) = 0.74η(Success with CRM) + e9
e. CT6 = 0.77ξ(CRM technologies) + e6
f. CT5 = 0.89ξ(CRM technologies) + e5
g. CT4 = 0.98ξ(CRM technologies) + e4
h. CT3 = 0.86ξ(CRM Technologies) + e3
i. CT2 = 0.66ξ(CRM technologies) + e2
j. CT1 = 0.56ξ(CRM technologies) + e1
v. The covariances in Table 5 indicate that the two-way relationships between the errors of
CT5 & CT6 and CT1 & CT2 should be created. These relationships have statistical
significance.
Note: Chi-Square =Discrepancy; P = Probability Levels; GFI = Goodness of Fit Index; AGFI =
Adjusted Goodness of Fit Index; NFI = Normed Fit Index; TLI = Tucker-Lewis Index; RMSEA =
Root Mean Square Error of Approximation
Fig. 2. Path Diagram
vi. Based on the findings described above, it seems clear that H1 —The CRM technology will
be positively associated with successful CRM adoption in Taiwan's Banking industry — is
accepted, and the null hypothesis is rejected.
4.1 Further discussion

As the coefficient weights in Figure 2 suggest, the variables containing special (high or
negative) correlation coefficient weights need to be singled out as they are worthy of further
discussion. This section will present reasons and descriptions for the phenomenon behind
the correlation between these variables.
4.1.1 Conducting the decision support system (DSS)

The coefficient weight between CT1 (conducting the decision support system) and CT2
(customizing CRM functions/modules) is significantly high (0.59). This result indicates that
when banks in Taiwan try to a deploy a CRM project, they might need to conduct the
decision support system as one of the customized CRM functions/modules. From the
literature review of the present study, a difference in customization of a CRM system

Success with
<--- CRM technologies .627 .134 4.679 ***
CRM
CT1 <--- CRM technologies 1.000
CT2 <--- CRM technologies 1.250 .104 11.962 ***
CRM Deployment <--- Success with CRM 1.000
Process Efficiency <--- Success with CRM 1.216 .099 12.260 ***
Product Quality <--- Success with CRM .911 .082 11.113 ***
Table 4. Regression Weights: Default model

CT5 <--> CT6 .003 .001 2.279 .023*
CT1 <--> CT2 .019 .003 7.306 ***
Table 5. Covariances: Default model
between the United Sates and Taiwan was discussed. In the United States, customization is
not widely applied while companies in Taiwan are more likely to customize their CRM
systems in order to satisfy their customers’ needs. Chang (2003) believes that project
managers, when adopting business intelligence systems such as CRM, should customize the
systems, complete interface requirements, and conduct data conversion . Chang (p. 99) also
suggests that companies establish stable and long-term relationships with their customers
through customization software. Moreover, Lai (2006) argues that CRM systems should
gather and analyze customer related information, develop forecasting models, provide real-
time responses, and effectively customize communication channels by integrating well-
established information technologies in financial service firms. Those discussions echo the
statistical results from the present study. In addition, Turban, Aronson and Liang (2004, p.
457) consider CRM a fundamentally enterprise-level DSS. In the present study, this
author defined CRM as a philosophy to understand and influence customers, to gain new
customers, to retain existing customers, and to enhance the profits provided by customers in
the banking industry. The subsequent empirical investigation makes it clear that a customer
relationship management (CRM) system will provide the technology to do so. Corporations
that are able to achieve high customer retention and high customer profitability are
those that are able to provide the right product (or service), to the right customers, at the
right price, at the right time, through the right channel (see Swift, 2001). Such is the main
goal of CRM. The rise of “E-commerce” has affected the need for quality (product and
service) and accurate customer analysis (Berkowitz, 2001; Kohli et al., 2001). Chou (2006)
takes artificial neural network (ANN) technology as one of the most popular methods on
evaluation of credit-card loan decisions in Taiwan’s banking industry. Support vector
machine (SVM), an ANN technology, is singled out. Chou (2006) took artificial neural
network (ANN) technology as one of the most popular methods for evaluating credit-card
loan decisions in Taiwan’s banking industry. Support vector machine (SVM), an ANN
technology, was singled out by Chou. Both Chou (2006) and Hung (2005) held the
predicting power of the support vector machine as being better than that of the other tools.
Not only does it outplay the traditional statistical approach in precision, it also achieves
better results than the neural network. With the incorporation of a support-vector machine
(SVM) into individual credit-card loan systems, the banks will be to render a highly effective
evaluation of customers credit conditions while decreasing the probability of bad debts.
Equipped with complete mathematical theories, SVM can comply with the principle of risk
minimization and deliver better performance. It is anticipated that properly applying the
SVM theory to loaning decisions of a bank will be a great asset. Clearly, conducting decision
support systems could help Taiwan’s banking industry analyze its customers more
accurately. CRM gathers customer data while tracking the customers. What’s important here
is using data to effectively manage customer relations. The foregoing discussions, in fact,
resonate with the statistical results of the present study. Clearly, conducting decision
support systems could help Taiwan’s banking industry analyze its customers more
accurately. CRM gathers customer data while tracking the customers. What’s important here
is use data to effectively manage customer relations. These further discussion resonate the
statistical results of the present study.
5. Conclusion
In the present study, CRM technologies, containing six observed variables ([i.e., CT1
(conducting the decision support system), CT2 (customizing CRM functions/modules), CT3
(choosing reputed CRM vendors), CT4 (drawing on the expertise of CRM vendors), CT5
(pressure from CRM/ERP packages), and CT6 (non-availability of appropriate CRM/ERP
packages)], indeed help CRM adoption in Taiwan’s banking industry. Moreover, as
mentioned in the research result, four variables contain special (high or negative) correlation
coefficients need to be singled out. They are CT1 (conducting the decision support system)
and CT2 (customizing CRM functions/modules), both with a high coefficient weight (0.59).
Thus, if the banks in Taiwan wish to deploy a CRM program, they should consider the
following suggestion:
Conducting the decision support systems (DSS): If the goal of implementing a CRM
system is to improve the process efficiency and IT product quality, managers in Taiwan’s
banking industry should emphasize more on conducting the decision support systems (DSS)
(e.g., artificial neural network (ANN) technology) as one of the customized module in the
whole CRM system.
6. References
Bailey, J. E. & Pearson, S. W. (1983). Development of a tool for measuring and
analyzingcomputer user satisfaction. Management Science, Vol.29, pp. 519-529.
Baldock, R. (2001). Cold comfort on the CRM front. The Banker, Vol.151, pp. 114.
Berensmann, D. (2005). The role of IT in the industrialization of banking. In Z. Sokolovsky

&S. Loschenkohl (Eds.), Handbuch Industrialisierung der Finanzwirtschaft (pp. 83-93).
Wiesbaden: Gabler Verlag.
Berkowitz, J. (2001). Customer relationship management (CRM): The defining business initiative
ofthe new millennium. Available from http://www.bi-bestpractices.com/view-
articles/4662.
Betsch, O. (2005). The future of the banking industry. In Z. Sokolovsky, & S. Loschenkohl
(Eds.), Handbuch Industrialisierung der Finanzwirtschaft (pp. 3-19). Wiesbaden: Gabler
Verlag.
Borck, J. B. (2005). CRM on demand. InfoWorld, Vol.27, pp. 30-39.
Born, C. (2003). What makes a good CRM package? Available from
http://www.computerworld.com/action/article.do?command=viewArticleBasic&
articleId=81971
Burnett, A. (2004). Mobilizing CRM: Routes to success. Customer Inter@ction Solutions, Vol.23,
pp. 40-42.
Cambell, M. (1999). Affordable measurement at last! Bank Marketing, Vol. 31, pp. 40.
Chang, B. S. (2003). The advantages and critical successful factors of introducing customer
relationship management system. Unpublished Master’s thesis, Chung Yuan Christian
University, Chung Li, Taiwan, R.O.C.
Chang, T. C. & Chiu, Y. H. (2006). Affecting factors on risk-adjusted efficiency in Taiwan’s
banking industry. Contemporary Economic Policy, Vol.24, pp. 634-648.
Chang, K. & Liu, N. (2005). The effects of customer relationship programs on customer
loyalty—An empirical study of Taiwan financial institutions. Commerce and
Management Quarterly, Vol.6, pp. 491-514.
Chou, C. H. (2006). An application of support vector machines and neural networks on credit card
loan decision. Unpublished Master’s thesis, National Taiwan University of Science
and Technology, Taipei, Taiwan, R.O.C.
Chowdhury, A. (2003). Information technology and productivity payoff in the banking
industry: Evidence from the emerging markets. Journal of International Development,
Vol.15, pp. 693-708.
Cocheo, S. (2005). Back in balance. American Bankers Association. ABA Banking Journal, Vol.97,
pp. 7-87.
Davenport, T. H. (2000). Mission critical: Realizing the promise of enterprise systems. Harvard
Business School Press, Boston, Massachusetts.
Deming, E. (1986). Out of the Crisis. MIT Center for Advanced Engineering, Cambridge, MA.
Dobbins, J. (2006). The future of banking technology? Available from
http://www.bcs.org/server.php?show=ConWebDoc.5851
Dos Santos, B. L., & Peffers, K. (1995). Rewards to investors in innovative information
technology applications: First movers and early followers in ATMs. Organization
Science, Vol.6, pp. 241-259.
Earley, R. M. (2003). Foreshore: A vision of CRM outsourcing. Customer Inter@ction Solutions,
Vol.22, pp. 42-46.
Ernst & Young. (2001). Eighth annual special report on technology in banking and financial
services.
Fadlalla, A. & Lin, C. (2001). An analysis of the applications of neural networks in finance.
Interfaces, Vol.31, pp. 112-122.
Galletta, F. G. & Lederer, A. L. (1989). Some cautions on the measurement of user
information satisfaction. Decision Sciences, Vol.20, pp. 419-438.
Hormazi, A. M. & Giles, S. (2004). Data mining: a competitive weapon for banking and retail
industries. Information Systems Management, Vol.21, pp. 62-71.
Huang, Y. H. (2004). E-commerce technology – CRM implementation and its integration
with enterprise systems. Unpublished Doctoral dissertation, Golden Gate
University, California.
Huang, C. C. & Lu, C. L. (2003). The processing functions and implementation factors of
Customer Relationship Management Systems in the local banking industry. Journal
of Information Management, Vol.5, pp. 115-127.
Hung, L. M. (2005). An application of support vector machines and neural networks on
banks loan evaluation. Unpublished Master’s thesis, National Taiwan University of
Science and Technology, Taipei, Taiwan, R.O.C.
Hsu, B. & Rogero, M. (2003). Taiwan CRM implementation survey report--2002. Available
from http://www.rogero.com/articles/taiwan-crm-report-article1.htm
Ives, B., Olson, M.H. & Baroudi, J. J. (1983). The measurement of user information
satisfaction. Communications of the ACM, Vol. 26, pp. 785-793.
Julta, D. & Bodorik, P. (2001). Enabling and measuring electronic customer relationship
management readiness. In The 34th Hawaii International Conference on System
Science, 3-6 January 2001. Hawaii, U.S.A.
Juran, J. M. (1986). Quality control handbook. McGraw-Hill, New York.
Kincaid, J. W. (2003). Customer relationship management getting it right! Prentice Hall, NJ.
Kohli, R., Piontek, F., Ellington, T., VanOsdol, T., Shepard, M. & Brazel, G. (2001). Managing
customer relationships through E-business decision support applications: A case of
hospital-physician collaboration. Decision Support System, Vol. 32, pp. 171-187.
Lai, K. Y. (2006). A study of CRM system from its current situation and effective factors
perspectives. Unpublished Master’s thesis, National Central University, Chung Li,
Taiwan.
Levitt, T. (1960). Marketing myopia. Harvard Business Review, July-August, pp. 3-23.
Li, M. C. (2003). Object-oriented analysis and design for service-oriented customer
relationship management information systems. Unpublished Master’s thesis,
Chaoyang University of Technology, Taichung, Taiwan, R.O.C.
Liao, S. J. (2003). The study of decision factors for customer relationship management
implementation and customer satisfaction in Taiwan banking industry.
Unpublished Master’s thesis, National Taipei University, Taipei, Taiwan, R.O.C.
Lin, F. H. & Lin, P. Y. (2002). A Case study of CRM implementation: Cathay United Bank as
an example. In The 13th International Conference on Information Management, 25
May 2002. Taipei, Taiwan, Republic of China.
Lu, K. L. (2000). The study in CRM systems adoption effective factors in Taiwan’s business.
Unpublished Master’s thesis, National Taiwan University, Taipei, Taiwan, R.O.C.
Lu, H. P., Hsu, H. H. & Hsu, C. L. (2002). A study of the bank’s establishment on customer
relationship management. In The 13th International Conference on Information
Management, 25 May 2002. Taipei, Taiwan, Republic of China.
Luo, W. S., Ye, R. J. & Chio, J. C. (2003). The integration of CRM and Web-ERP system. In
The 2003 Conference of Information Technology and Applications in Outlying
Islands, 30 May 2003. Peng-Hu, Taiwan, Republic of China.
Moormann, J. (1998). Status quo and perspectives of information processing in banking,
Working Paper, Hochschule for Bankwirtschaft, Frankfurt, Germany.
Motley, L. B. (1999). Are your customer as satisfied as you bank’s shareholders? Bank
Marketing, Vol.31, pp. 44.
Motley, L. B. (2005). The benefits of listening to customers. ABA Banking Marketing, Vol. 37,
pp. 43.
Ortega, M., Perez, M. A., & Rojas, T. (2003). Construction of a systemic quality model for
evaluating a software product. Software Quality Journal, Vol.11, pp. 219-242.
Peppers, D. & Rogers, M. (1993). The one to one future: Building relationships one customer at a
time. Bantam Doubleday Dell Publishing Group, New York.
Rebouillon, J., & Muller, C. W. (2005). Cost management as a strategic challenge. In Z.
Sokolovsky & S. Loschenkohl (Eds.), Handbuch Industrialisierung der
Finanzwirtschaft (pp. 693-710). Wiesbaden: Gabler Verlag.
Scott, J. E., & Kaindl, L. (2000). Enhancing functionality in an enterprise software package.
Information and Management, Vol.37, pp. 111-122.
Shang, S., & Seddon, P. B. (2002). Assessing and managing the benefits of enterprise
systems: The business manager’s perspective. Information Systems Journal, Vol.12,
pp. 271-299.
Shermach, K. (2006). Online banking: Going above and beyond security. Available from
http://www.crmbuyer.com/story/53901.html
Sheshunoff, A. (1999). Winning CRM strategies. American Bankers Association ABA Banking
Journal, Vol.91, pp. 54-66.
Swift, R. S. (2001). Accelerating customer relationships: Using CRM and relationship technologies.
Prentice Hall PTR, NJ.
Tehrani, R. (2005). A host of opinions on hosted CRM. Customer Inter@ction Solutions,
Vol.23, pp. 14-18.
Thong, J. Y. L. & Yap, C. S. (1995). CEO characteristics, organizational characteristics and
information technology adoption in small business. Omega, Vol.23, pp. 429-442.
Trepper, C. (2000). Customer care goes end-to-end. Information Week, March, pp. 55-73.
Turban, E., Aronson, J. E. & Liang, T. P. (2004). Decision support systems and intelligent systems
(7th ed.). Prentice Hall PTR, NJ.
Veitinger, M., & Lbschenkohl, S. (2005). Applicability of industrial practices to the business
processes of banks. In Z. Sokolovsky & S. Loschenkohl (Eds.), Handbuch
Industrialisierung der Finanzwirtschaft, (pp. 397-408). Wiesbaden: Gabler Verlag.
Yahaya, J. H., Deraman, A., & Hamdan, A. R. (2008). Software quality from behavioural and
human perspectives. International Journal of Computer Science and Network Security,
Vol.8, pp. 53-63.
0
2
Intelligent Agent Technology in Modern

Production and Trade Management
Serge Parshutin, Arnis Kirshners
Institute of Information Technology, Riga Technical University
Riga, Latvia
1. Introduction
The technological progress is constantly evolving and each year presents new growth
opportunities for different people and companies. Modern production and trade management
are processing terabytes of gathered statistical data to gain useful information for different
management tasks. Most common tasks that have arisen for solving are forecasting the
demand for a new product, knowing only the product descriptive data and creating a
production plan for a product on different product life cycle phases. As the amount of data
grows, it becomes impossible to analyse it without using modern intelligent technologies.
Trade management is strictly connected with sales forecasting. Achieving accurate sales
forecasting will always remain on the top of the most important tasks. Due to that different
solutions are introduced each year in order to overcome the efficiency of the traditional sales
forecasting methods. Let us discuss several possible solutions recently introduced. A sales
forecasting model based on clustering methods and fuzzy neural networks was proposed by
(Chang et al., 2009). The model groups demand time series applying clustering algorithms,
then a fuzzy neural network is created for each cluster and used to forecast sales of a new
product. Another interesting solution for sales forecasting was proposed by (Ni & Fan, 2011).
The model combines the Auto-Regressive trees and artificial neural networks to dynamically
forecast a demand for a new product, using historical data. Both models apply modern
intelligent information technologies and, as authors state, are more efficient and accurate
comparing to traditional forecasting methods. Nevertheless, the models preserve that a
product is already introduced on the market and some sales data are already available. This
rises a question that may be important for retailers, supplying market with items, but not
producing them personally. The question may be stated as "How to forecast a demand for
a product before purchasing it for resale?". A system, capable of making such forecasts, will
bring to retailers a useful and valuable information, enabling them to create a list of stock
products before any investments are made. A possible solution to such task was proposed by
(Thomassey & Fioraliso, 2006). The authors proposed a hybrid sales forecasting system based
on a combination of clustering and classification methods. The system applies clustering
methods to gain product demand profiles from historical data, then, using product descriptive
data, a classifier is built and used to forecast a sales curve for a new product, for which only
descriptive data are available. The system was tested on textile item sales and, as authors
state, the results were relatively good. Authors continued their research and in (Thomassey
& Happiette, 2007) a new system based on idea of combining clustering and classification
methods was introduced. Authors used Artificial Neural Networks technologies to cluster
22
Will-be-set-by-IN-TECH
and classify sales time series. As they state, the proposed system increases accuracy of
mid-term forecasting in comparison with the mean sales profile predictor.
The production management itself combines several tasks of forecasting and planning, one
of which is a product life cycle (PLC) management. A product life cycle is a time of
product existence on the market and can be viewed as a sequence of known phases with
bounds. Each PLC phase differs by sales profiles, due to what different production planning
strategies are used and what is more important - different advertising strategies are applied.
Traditionally a product life cycle has four phases, namely "Growth", "Introduction", "Maturity"
and "End-of-Life". The models with a larger number of phases are also possible and are used
(Aitken et al., 2003). For products with a significantly long PLC, growth and introduction
phases can be merged into one phase - "Introduction". This PLC phase imply high investments
in advertising, as potential customers need to know about the new product. This gives rise to
the task of defining bounds for the PLC phases, which to be one of the important tasks in the
PLC management. Let us discuss several possible solutions introduced for this task. Most of
the solutions concentrate around the idea of forecasting the sales curve for a product and then
defining in which particular PLC phase product currently is. Often the Bass diffusion model or
its modifications are used to forecast sales, as it was proposed by (Chien et al., 2010). Authors
present a demand forecasting framework for accurately forecasting product demand, thus
providing valuable information to assist managers in decision-making regarding capacity
planning. The framework is based on the diffusion model and, as authors state, has a practical
viability to forecast demands and provide valuable forecasting information for supporting
capacity planning decisions and manufacturing strategies. Other researchers (Venkatesan &
Kumar, 2002) applied a genetic algorithm to estimate the Bass diffusion model and forecast
sales for products, when only a few data points are available. Both solutions make accurate
forecasts in the scope of the tasks solved, but still give an answer to what the demand will
be, but not to when the PLC phase will be changed. To answer that, the forecasted sales
curves need to be analysed by a manager, who will set PLC phase bounds, using his personal
experience. Due to that, the efficiency of the discussed models may fall as the amount of
monitored products growths.
The present research is oriented towards bringing intelligent agent technology for supporting
manager’s decisions in such tasks as forecasting demand for a new product and defining the
end of the introduction/beginning of the maturity PLC phase for the monitored products. The
chapter proposes two multi-agent systems designed directly for solving the tasks defined.
Intelligent agent technology was chosen as it gives a new modern look at the forecasting
and planning tasks and brings new possibilities in designing solutions to support modern
production and trade management.
The conceptual and functional aspects of the proposed systems are revealed in Sections
2 and 3, but first, some agent terminology needs to be defined to eliminate possible
misunderstandings. There are no unified definitions of what an agent is (Wooldridge, 2005).
One of the reasons for that is that agents are mostly defined as a task specific concept.
Nevertheless, some definitions are useful (Weiss, 1999; Wooldridge, 2005) and also applicable
to present research:
• Agent is a computer program that exists in some environment and is capable of performing
an autonomous activity in order to gain its desired objectives.
• Intelligent agent is an agent capable of learning, capable of mining and storing knowledge
about the environment it is situated in.
• Multi-agent system contains agents and intelligent agents capable of interaction through
an information exchange.
Intelligent Agent Technology
Intelligent Agent Technology inandModern
in Modern Production Production and Trade Management
Trade Management 233
• Agent community is an indivisible unit of agents with similar attributes, abilities and
behaviour, aimed at gaining the desired objectives. It differs from a Multi-agent system
in that agents are not interacting with one another.
The proposed definitions are task specific and are oriented to making the ideas proposed in
the multi-agent systems simple and clear.
2. Agent Technology in PLC Management

The product life cycle management, as it was stated in the introduction, is an important
objective for company growth, as most of the decisions taken are dependent on the current
demand. A proper product life cycle management strategy gives opportunities for precise
resource planning, lessening expanses and negative profit.
The present section reveals the concepts of the proposed PLC management support
multi-agent system, and describes both structural and functional aspects of it. An analysis
of the obtained experimental results concludes this section.
The main task of the proposed system is to forecast a period - transition point, when a
product will change an introduction phase to maturity phase. A specific feature of the system
is that the forecast is made having only the first several data points, and a scalar value -
period, is forecasted. From the viewpoint of Data Mining this task is a classification task
and formally can be stated as follows. Assume that the available sales data is a dataset
D = {d1 , . . . , di , . . . , dn }, where each record d = { x1 , . . . , x j , . . . , xl } is a discrete time series
with l periods, representing a sales profile of a specific product during the defined phase of
a product life cycle. The length of a time series l is not a constant value. It may take a finite
number of values defined by the set L, l ∈ L = {l1 , . . . , lh , . . . , ls } and may vary from record
to record.
Each record di ∈ D has a specific marker p containing the number of a period after which
a PLC phase, represented by record di , was finished. Marker p may take a finite number of
values defined by the set P, p ∈ P = { p1 , . . . , pk , . . . , pm }. Having such assumptions the
forecasting of a transition point for a specific new product d ∈ / D should start with building a
model of an implication between datasets D and P, f : D → P, representing relations between
sales profiles and the value of a specified marker. As the model is built, it may be applied to
forecast a transition point for a new record, when it reaches the minimal length lmin ∈ L.
2.1 Structure of the proposed system

The proposed product life cycle management support multi-agent system is displayed in
Figure 1. The system contains three interacting agents, namely Data Management Agent, Data
Mining Agent and Decision Support Agent. The Data Management Agent was included as the
system should be capable of processing a large amount of sales data. This condition asks for
an agent capable of autonomously performing general and specific operations with data, one
of which is pre-processing. As it was stated before, first, the model of the relations between
the sales profiles and a target attribute should be made. The Data Mining Agent implements
an intelligent data processing methods and is capable of autonomously creating a knowledge
base with desired relations. The application of the knowledge base to the forecasting of a
transition point for a new product is designated to the Decision Support Agent.
2.1.1 Data Management Agent

The Data Management Agent is an agent as it uses a predefined logic and cannot change
it over time. The agent handles several data management tasks, including data cleaning,
normalization and preparing datasets. The environment the Data Management Agent is
24
DATA MANAGEMENT AGENT
1. Querying data
from database
2.1. Data cleaning
with a switching point
Demand time series
2. Pre-processing 2.2. Data

data normalization
Pre-processed
data
3. Dataset formation
DATA MINING AGENT

1. Clustering different duration demand time series
CONTROLLED AGENT COMMUNITY
Clustering agent 1
Clustering agent 2
...
Clustering agent n
2. Extracting profiles from clusters and mapping with

PLC phase switching points
Centroid C1 + period 4
Centroid C2 + period 6 Knowledge
… base
Centroid Cn + period 17
DECISION SUPPORT AGENT

1. Forecasting a switching point for a new product using
only first m periods, available in the current period
2. Monitoring selected and new products
3. Creating a list of products for which it would be

reasonable to reconsider a production planning policy
due to PLC phase changing process
Fig. 1. PLC Management Support Multi-Agent System

situated in is a raw data repository. The environment affects the agent with new data, forcing
the Data Management Agent to perform the data pre-processing actions. By performing those
actions, the agent affects the environment by changing the raw data to the pre-processed data.
The agent functioning algorithm, displayed in Figure 1, contains three main processes - first
get data from a database, then pre-process it and prepare datasets for system training and
testing. The first two processes are performed in the autonomous mode, the actions in the
third process are performed reactively by responding to the Data Mining Agent’s requests.
The data repository contains data of the unique structure that may not fit the desired data
format. The processes in the first block of an Data Management Agent algorithm return
the set of sales data, obtained during a specified product life cycle phase, as a time series,
summing sales by each period (day, week, month), which is defined by the user. For each
sales time series a value of a target attribute (transition point) is supplied, excluding the new
data, for what the information on the transition point is not available yet. The next step is
pre-processing, it becomes necessary as the raw data may contain noise, missing or conflicting
data. The defined task speaks for performing several steps of data pre-processing:
• Excluding the outliers. The definition of an outlier is dynamic and changes according to
the desired tasks. This makes the step of outlier exclusion highly task specific. The present
research foresees the following actions to be completed. The time series with the number
of periods (length) less than the user defined minimum lmin or with missing values to be
excluded from the raw dataset.
• Data normalization. This step is performed in order to bring sales time series to one level,
lessening possible domination occurrences. One of the normalization methods that suites
the sales time series normalization is the Z-score normalization with standard deviation

method. To calculate the normalized value xi for i-th period of a time series X, Equation 1
is used, where X is the average of an X.
xi − X
xi = (1)
sx
To calculate the standard deviation s x , Equation 2 is used. As the number of periods in
the normalized sales time series is not large, it is recommended to use the (n − 1) as a
denominator in order to obtain an unbiased standard deviation. As an option, the standard
deviation can be replaced with a mean absolute deviation.

n
1
sx = ∑ ( x − x )2
n − 1 i =1 i
(2)
• Imitation of a data flow. The sales data become available over time and this moment
should be taken into account in order to obtain an accurate transition point forecast. This
step is very important as the proposed system will be limited to forecast a binary result
- either the transition is occurred or not, if the data flow imitation will not be performed.
The imitation of the data flow process ensures that the proposed system will be able to
forecast transition points in different time periods having only first lmin periods available.
2.1.2 Data Mining Agent

The Data Mining Agent (see Fig. 1) is an intelligent agent that performs an intellectual data
processing with data mining methods. The environment the agent is situated in is the datasets
created by the Data Management Agent. This environment is highly dynamic as it changes
26
each time a new data is obtained. The agent itself may induce changes in the environment
by requesting the Data Management Agent to produce new datasets. The Data Mining Agent
changes the environment by creating the knowledge base, that contains relations between the
sales time series and transition point values. The proposed structure of a Data Mining Agent
contains two main steps - first clustering the time series with different duration and then
extracting profiles from clusters and mapping profiles with PLC phase switching points.
The time series clustering is performed by the clustering agents in the agent community
controlled by the Data Mining Agent (see Fig. 1). Each clustering agent implements a
clustering algorithm and processes a defined part of the input data. The number of the
clustering agents as well as the part of dataset it will process, is defined by the parameter
Q called the load of a clustering agent. It is an integer parameter containing a number of
different time series lengths li ∈ L each clustering agent can handle. The set L contains all
possible lengths of the time series in the input data, as it was defined in the task statement at
the beginning of this section. For example, assume that lmin = 4 periods and Q = 3, then the
first clustering agent will handle the part of an input data with 4, 5 and 6 periods long time
series. The load of a clustering agent is defined by the user and is also used for calculating nca
- the number of the clustering agents in the Controlled Agent Community (CAC). Equation 3
is used for obtaining the value of nca .

| L|
nca = Roundup (3)
Q
The input data is distributed among nca clustering agents cai ∈ CAC according to the defined
value of Q. Given a uniform clustering agent load distribution, Equation 4 can be used for
splitting the input data among clustering agents.

i = 1, li,min = lmin
i > 1, li,min = li−1,max + 1 (4)
li,max = li,min + Q − 1
where the lmin and lmax are the bounds for the number of periods in the time series the
clustering agent cai will process. One of the proposed system objectives is to support
clustering of time series with different number of periods. A number of clustering methods
exist, but majority use an Euclidean distance to calculate the similarity of two objects. The
simple Euclidean distance will not allow a clustering agent to process time series with
different number of periods, but still may be applied while Q = 1. Two clustering methods
- Self Organising Maps (SOM) (Kohonen, 2001) and Gravitational Clustering algorithm (GC)
(Gomez et al., 2003; Wright, 1977), were chosen to support the data mining process. To gain
an ability to cluster time series with different duration we suggest using different distance
measures in each of the methods. The proposed system implements two different distance
measures - Dynamic Time Warping (DTW) (Keogh & Pazzani, 2001; Salvador & Chan, 2007)
and a modified Euclidean distance MEuclidean.
2.1.2.1 Distance measures
The DTW creates a warping path W ( X, Y ) between two time series X and Y and uses it to
calculate the distance, which occurs as follows. Let X and Y have durations equal to n and m
periods, respectively. First a distance matrix n × m is created, where each cell (i, j) contains
distance d( xi , y j ) between two points xi and y j (Keogh & Pazzani, 2001). The warping path
W ( X, Y ) is a sequence of the steps wk in the distance matrix starting with a cell (1, 1) and
ending in (n, m). Each time performing a new step wk , a direction with minimal distance is
chosen. Other strategies for choosing the next step are possible and widely used (Salvador &
Chan, 2007). An example of a warping path is graphically displayed in Figure 2.
y4 (1,4)
Y: y1 y2 y3 y4
y3
y2 W(X,Y)
y1 (6,1) X: x1 x2 x3 x4 x5 x6
x1 x2 x3 x4 x5 x6
(a) Warping path in the distance matrix (b) A warping map M
Fig. 2. Example of a DTW warping path W ( X, Y )
The total path distance d( X, Y ) between two time series X and Y using the Dynamic Time
Warping is calculated as shown in Equation 5. The denominator K representing the total
number of steps wk in the warping path W ( X, Y ) is used in order to calculate the distance
proportionally for one step wk , as the number of steps may be different in different warping
paths.
∑kK=1 w2k
d( X, Y ) = (5)
K
The second proposed distance measure is MEuclidean - Modified Euclidean. The concept of
this measure is to calculate the distance d( X, Y ) between two time series X = x1 , . . . , xi , . . . , xn
and Y = y1 , . . . , y j , . . . , ym only by first z periods of each time series, where z = min{n, m}.
Speaking in the terminology of DTW, the warping path W ( X, Y ) in the case of MEuclidean will
always begin in the cell (1, 1) and will always end in the cell (z, z). The example of a warping
path for a distance measure MEuclidean is displayed in Figure 3.
y4 (1,4) (6,4)
Y: y1 y2 y3 y4
y3
y2 W(X,Y)
y1 (6,1) X: x1 x2 x3 x4 x5 x6
x1 x2 x3 x4 x5 x6
(a) Warping path in the distance matrix (b) A warping map M
Fig. 3. Example of a MEuclidean warping path W ( X, Y )
The total path distance d( X, Y ) between two time series X and Y using the MEuclidean distance
measure is calculated using Equation 6.

z
d( X, Y ) = ∑ ( x i − y i )2 (6)
i =1
28
2.1.2.2 Clustering algorithms

The SOM model introduced by T.Kohonen is one of the vector-coding algorithms, whose main
objective is to transform an input data into a discrete map of neurons. The SOM algorithm
contains four main steps - network initialization, neuron competition, neuron cooperation and
synoptic adaptation (Kohonen, 2001). Executing the initialization,weights of each neuron are
assigned random values small enough to eliminate any possible organization in the network
at the beginning of the system training.
During the neuron competition step, for each time series in the input dataset the winning
neuron, as the one with most similar vector of weights, is defined. This is where the proposed
distance measures are applied to find the winning neuron, which will be set as the centre of the
topological neighbourhood in order to perform the neuron cooperation step. The cooperating
neurons are neurons that are stimulated by the activity in the winning neuron. Executing
the cooperation step each neuron in the SOM is assigned a neighbourhood level, using the
neighbourhood function h j,i( X ) , where i ( X ) is the index of a winning neuron for a time series
X and j is an index of a cooperating neuron. The neighbourhood level is also affected by the
chosen neural network topology. The classical SOM algorithm uses Equation 7 for synoptic
adaptation of weights.
Wj (n + 1) = Wj (n) + η (n) · h j,i( X ) (n) · ( X − Wj (n)) , (7)
where n is the iteration number; W (n) - vector of weights of a j-th neuron; η is a learning
coefficient.
The learning coefficient decreases over time, but remains higher than a defined minimal value
ηmin . This can be reached by using Equation 8 (Kohonen, 2001), where τ2 is a time constant,
which can be assigned total iterations number. The value of a topological neighbourhood
function h j,i( X ) (n) at the moment n is calculated by Equation 9.

n
η (n) = η0 · exp − (8)
τ2

d2j,i( X )
h j,i( X ) (n) = exp − 2 , (9)
2σ (n)
where d is the lateral distance between the j-th neuron and the winning neuron; parameter σ
defines the width of a topological neighbourhood function.
In Subsection 2.1.2.1 it was shown that both of the proposed distance measures can produce
a warping path in order to compare two time series. If a distance measure MEuclidean is
used, then Equation 7 is modified and Equation 10 is used for synoptic adaptation of neuron
weights.
∀wk ∈ Wj , k ≤ lmin : wk (n + 1) = wk (n) + η (n) · h j,i(X ) (n) · ( xk − wk (n)) , (10)
where Wj is the weight vector of a j-th neuron and lmin = min(lX , lW ).

Using distance measure MEucliden two cases during weight adaptation are possible:
1. The length lW of a weight vector Wj of the j-th neuron is less or equal to the lX - the length
of a time series X. In this case all weights of the j-th neuron will be changed according to
the warping map M, example of which is shown in Figure 3.b.
2. The length lW of a weight vector Wj of the j-th neuron is greater than the lX - the length of
a time series X. In this case only the first lX weights wk ∈ Wj will be adopted according to
the warping map M. The rest of neuron weights will remain unchanged.
If the distance measure DTW is used, then cases are possible, when one weight wk should be
adopted to several values of a time series X. An example of this case is shown in Figure 2.b.
Assuming that Y is a weight vector, the weights y1 , y2 and y4 must be adopted to two or three
different values at one time. In this case each of the mentioned weights will be adopted to the
average of all time series X values it must be adopted to, according to the warping map M.
This concludes the modifications of a SOM algorithm, the rest part of the subsection is devoted
to the second chosen clustering algorithm - gravitational clustering algorithm. This algorithm
is an unsupervised clustering algorithm (Gomez et al., 2003; Wright, 1977). Each clustering
agent in the CAC represents a multidimensional space, where each time series from the input
dataset is represented as an object in the n-dimensional space, where each object can be
moved by the gravitational force of the other objects. If two or more objects are close enough,
that is distance d between them is less or equal to the dmax , then those objects are merged
into the cumulative object. By default, masses of all non-cumulative objects are equal to 1.
Depending on user choice, masses of the merged objects can be summed or remain equal to
1. A movement of an object X, induced by the gravitational force of an object Y at the time
moment t, can be calculated by Equation 11.
→ G · mY · Δt2
X (t + Δt) = X (t) + v(t) · Δt+ d(t) · → 3 , (11)

2·
d(t)
where v(t) is the object velocity; G - gravitational constant, which can also be set by the user;
→
mY - the mass of an object Y; vector d(t) represents the direction of a gravity force, induced by
→
the object Y and is calculated as d(t)= X (t) − Y (t).
Setting velocity v(t) as a null vector and having Δt = 1, Equation 11 can be simplified as
shown in Equation 12.
→ G·m
X (t + Δt) = X (t)+ d(t) · Y 3 , (12)
→
2·
d ( t )

The GC uses the Euclidean distance to measure spacing between to objects. The support of
clustering of time series with different number of periods can be gained by using the distance
measures, proposed in Subsection 2.1.2.1, but the first of the gravitational clustering algorithm
concepts should be redefined. In the classical algorithm this concept is defined as follows: "In
the n-dimensional space all objects have values in all n dimensions". The proposed redefined
version is "In the n-dimensional space can exist m-dimensional objects, where m ≤ n". This
means that in a 10-dimensional space can exist an object with four dimensions, which will be
present only in the first four dimensions.
In the case of using the distance measure MEuclidean two possible cases can occur while
calculating the induced gravitational force:
1. The number of the object‘s X dimensions n X is less or equal to the number of the inducting
object‘s Y dimensions nY . In this case the gravitational force is calculated for all dimensions
of an object X and it is moved according to the calculated gravitational force.
30
2. The number of the object‘s X dimensions n X is greater than the number of the inducting
object‘s Y dimensions nY . In this case the gravitational force for X is calculated only for the
first nY dimensions of the X. The object X is moved only in those dimensions for which
the gravitational force was calculated.
Using the dynamic time warping, an m-dimensional object X will always be moved in all m
dimensions. Situations, when for some dimension xi a gravitational force should be calculated
using several dimensions of an object Y, may take place. This is similar to the case with
SOM when a single weight should be adopted to multiple time series values. Having a such
situation, the gravitational force, induced by the object Y, is calculated using an average from
all y j dimensions inducing gravitational force to the object‘s X dimension xi .
2.1.2.3 Creating a knowledge base
The knowledge base contains a model, representing the connections between sales
profiles and the transition points, induced from the input data. The knowledge base is
level-structured, the number of levels coincides with the number of the clustering agents in
the clustering agent community. Each level contains clusters from a corresponding clustering
agent. Let us define the meaning of a cluster for each of a clustering algorithm described in
the previous subsection.
The cluster extraction in the SOM algorithm begins after network organisation and
convergence processes are finished, and contains next three steps:
1. Each record X from a training set is sent to the system.
2. The winning neuron for each record X is found. For each winning neuron a transition point
statistics S j is collected, containing a transition point value and frequency of it‘s appearing
in this neuron, which is used as a rank of a transition point.
3. The cluster is a neuron that at least once became a winning neuron during the cluster
extraction process. The weight vector of a neuron is taken as a centroid of a cluster.
For the gravitational clustering algorithm a cluster is defined as follows: "The cluster is a
cumulative object having at least two objects merged". Using the merged objects, a transition
point statistics S j is collected for each cluster c j . Each transition point, included in S j , is
assigned a rank as the frequency of it‘s appearing in the cluster c j . The centroid of a cluster
is set as a position of a cumulative object in the clustering space at the moment of cluster
extraction.
2.1.3 Decision Support Agent

The Decision Support Agent is an agent that forecasts a transition point value for a new data,
monitors selected and new products and creates a list of the products, for which it would
be reasonable to reconsider a production planning and advertising policies. The DSA can
perform its actions either by receiving a request from a user, or in autonomous mode, with
defined interval of time (at the end of each period) reporting the decision analysis results.
The transition point for a new data is forecasted by using the data from a cluster,
best-matching the new time series. The best-matching cluster is obtained by sending
an appropriate request to the Data Mining Agent, which finds the closest cluster in the
knowledge base. The transition point with a higher rank is taken as the value that will be
forecasted for the new time series. If several transition points have equal rank, then the
minimal transition point is chosen.
The list of the noteworthy products (LNP) is created by applying this strategy:
1. If l < p and p − l > θ Then: Product remains monitored and is not included in the LNP;
Trade Management 31
11
2. If l < p and p − l ≤ θ Then: Product is included in the LNP;

3. If l ≥ p Then: Product is included in the LNP.
where l is the duration of the demand time series in periods; p - a forecasted transition point;
and variable θ stores the minimal threshold of interest for either including the product in the
list of the noteworthy products or not.
2.2 Experimental results

The proposed product life cycle management support system was trained and tested on a
dataset, received within the international project ECLIPS. The dataset contains sales profiles
of different products during the PLC introduction phase. For each record in the dataset
a transition point value is supplied. The length of a time series in the dataset varies
from 4 to 23 periods, giving 20 possible values for a clustering agent load Q parameter,
Q = {1, 2, . . . , qi , . . . , 20}. The dataset was pre-processed according to the steps described
in Subsection 2.1.1 and normalized by the Z-score normalization method.
The precision of the proposed system is measured, using the Mean Absolute Error, measured
in periods. To measure the precision of the system, first the learning error for all combinations
of the proposed clustering methods and distance measures was measured. The system with
gravitational clustering algorithm was tested with both options for calculating the mass of a
cumulative object after merging two or more other objects. The obtained results are displayed
in Figure 4.
2.00 2.00
1.60 1.60
1.20 1.20
0.80 0.80
0.40 0.40
0.00 0.00
1 3 5 7 9 11 13 15 17 19 1 3 5 7 9 11 13 15 17 19
Clustering agent load, Q Clustering agent load, Q
DTW MEuclidean DTW MEuclidean
(a) Masses are summed (b) Object mass remains equal to 1
Fig. 4. Learning error (MAE) in periods for the system with gravitational clustering algorithm
As may be seen from Figure 4, the results, received using the distance measure MEuclidean,
are more precise than those obtained with the DTW. The MEuclidean dominates DTW within
all possible Q values, and that is the reason to exclude the DTW distance measure from further
experiments with testing the system with gravitational clustering algorithm. As can also be
seen, the results, obtained using MEuclidean in Figures 4.a and 4.b, are close to be equal.
This indicates that both strategies for calculating the mass of a cumulative object are equal.
However the strategy, where masses are summed after merging the objects, has a weakness.
Merging the objects results in the growth of a mass of a cumulative object, consequently
increasing the gravity force of that object. This could lead to the situation, when merging of
objects will be affected more by the masses of an objects either by the closeness of those objects.
32
This could be the reason for that some potentially good combinations of clusters will not
appear during the clustering process. Due to that, the strategy with summing objects masses
will not be used in further experiments while using the gravitational clustering algorithm.
The transition point forecast error (using MEuclidean) decreases while the clustering agent
load increases. This is the reason for performing further experiments by using only five values
of Q that returned the smallest learning error. Those values of a clustering agent load are 16,
17, 18, 19 and 20, that is Q = {16, 17, 18, 19, 20}.
The system with SOM algorithm was tested using three neural network topologies - quadratic
topology with eight neighbours in the first lateral level, cross-type topology with four
neighbours and linear topology with two neighbours. The neural network contained 100
neurons, organized as a matrix 10×10 neurons for quadratic and cross-type topologies and
as a one-dimensional line for the linear topology. The obtained results with learning error for
each topology and distance measure MEuclidean are displayed in Figure 5.
3.9
3.7
3.5
3.3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Clustering agent load, Q
Linear Cross-type Quadratic
Fig. 5. Learning error (MAE) in periods for the system with SOM algorithm, using
MEuclidean
The SOM algorithm returned a much larger learning error than the gravitational clustering
algorithm. Due to that, only three values of Q, Q = {5, 6, 7} will be used in further
experiments. Table 1 contains the learning error obtained with SOM algorithm while using
DTW and selected Q values.
Q=5 Q=6 Q=7
Linear 3.381 3.320 3.323
Cross-type 3.366 3.317 3.323
Quadratic 3.354 3.317 3.323
Table 1. Learning error (MAE) in periods for SOM algorithm, using DTW
Comparing the efficiency of the chosen topologies (see Figure 5) it may be concluded that none
of three topologies dominates the others two. For further experiments with SOM algorithm
all three topologies were chosen.
The learning error evaluation was used to lessen the number of experiments needed for
system strong evaluation. To evaluate systems the 10-fold crossvalidation method was
applied. The proposed system with gravitational clustering was tested, applying only the
MEuclidean distance measure and using the five values of the clustering agent load, selected
at the beginning of this section, while evaluating the system learning error. Table 2 shows
the obtained system testing results, each cell contains a 10-fold average Mean Absolute Error in
periods. Figure 6 displays the data from Table 2.
Trade Management 33
13
Q = 16 Q = 17 Q = 18 Q = 19 Q = 20
MAE 2.021 1.995 2.026 2.019 2.010
Table 2. Mean absolute error in periods for the system with GC algorithm
As can be seen, the error remains at the level of two periods, the best result was gained with
the clustering agent load Q = 17 and is 1.995 periods.
2.30
2.20
2.10
2.00
1.90
1.80
1.70
16 17 18 19 20
Clustering agent load, Q
Fig. 6. MAE in periods for system with gravitational clustering, using MEuclidean
The same testing method and data were used for testing the system with SOM algorithm.
While testing the system, all three neural network topologies - linear, cross-type and quadratic,
were applied, combining with two proposed distance measures (see Subsection 2.1.2.1). The
system with SOM was tested by using only three clustering agent load values - Q = {5, 6, 7},
selected while evaluating the learning error. Table 3 summarises the obtained testing results
and Figure 7 displays them.
Linear topology Cross-type topology Quadratic topology
Q DTW MEuclidean DTW MEuclidean DTW MEuclidean
5 3.475 3.447 3.362 3.437 3.434 3.578
6 3.267 3.267 3.332 3.404 3.353 3.353
7 3.344 3.411 3.372 3.513 3.371 3.509
Table 3. Mean absolute error in periods for the system with SOM algorithm
The best results for all three topologies were obtained while testing the system with clustering
agent load Q = 6 and remains equal about 3.3 periods. That supports the conclusion, stated
while evaluating the learning error that none of three topologies dominates other two. The
best result was obtained while using the linear topology and is equal to 3.267 periods.
Comparing the efficiency of applied distance measures we may conclude that for the proposed
system with SOM algorithm it is not crucial whether to choose DTW or MEuclidean distance
measure, as the DTW returned better result only once, using the cross-type topology. For other
topologies both distance measures returned equal results.
34
3.5 3.6
3.5
3.4
3.4
3.3
3.3
3.2 3.2
5 6 7 5 6 7
Clustering agent load, Q Clustering agent load, Q
Linear Cross-type Quadr. Linear Cross-type Quadr.
(a) Using distance measure DTW (b) Using distance measure MEuclidean
Fig. 7. Test error in periods for system with SOM algorithm
2.3 Results analysis and summary

Analysing the obtained results it may be concluded that for the task of forecasting the
period when the product will change its PLC phase, the efficiency of the hierarchical
clustering algorithms, a part of which is the gravitational clustering algorithm, comparing
to the self-organising maps. Comparing the learning and test errors for both of clustering
algorithms, it can be seen that the GC test error (1.995 periods) is about 10 times greater than
the learning error (0.2 periods), but the SOM learning and test errors remain on the same level
of 3.3. From that it may be concluded that the gravitational clustering algorithm is less robust
to the noise in data than the SOM algorithm.
The smallest transition point forecasting error - 1.995 periods, was gained by the multi-agent
system with the gravitational clustering algorithm, while using the distance measure
MEuclidean. The test error of two periods may be taken as a relatively high, but, viewing
it from the position of the end user, the ability of gaining the additional useful information
about the bounds of a PLC phase, makes this error level acceptable.
3. Intelligent Sales Forecasting

Traditional forecasting is mostly based on finding relations in time series, representing
changes in economic indexes, system technical parameters etc. It is assumed that relations are
mined from a stationary time series and the longer it is, the higher is the probability to find
some relations in it (Armstrong et al., 2005). However, in most of real-life tasks the length
of a time series is relatively short and mining relations will not bring the desired results.
An example of such task is a textile sales analysis and forecasting, which is the main task
the proposed system was designed for. As it was stated, clustering textile sales short time
series will return sales profiles, but how precise they will be? An additional information can
be gained from a descriptive data of each textile product. This sets a task of combinations
of several data mining methods - cluster analysis and classification, in order to obtain an
additional useful information.
The proposed system is aimed at forecasting textile product sales profiles using only
descriptive data, such as price, size, colour, type etc. Subsection 3.1 reveals the structure and
conceptual aspects of the proposed system. Figure 8 displays the proposed system.
Trade Management 35
15
3.1 Concepts of Sales Forecasting Multi-Agent System

The proposed sales forecasting multi-agent system contains three agents - Data Management
Agent, Data Mining Agent and the Sales Forecasting Agent. The Data Management Agent
supports receiving data from a data repository and data preparation for further analysis. The
Data Mining Agent mines relations from sales data, merging those with product descriptive
data and inducing classification rules that are included in the knowledge base. The Sales
Forecasting Agent uses the knowledge base to forecast a sales profile for a new product,
having only descriptive data available.
3.1.1 Data Management Agent

The Data Management Agent is an agent that is similar to the one described in Subsection
2.1.2. It receives raw data from a data repository, each record containing product descriptive
data and one year sales data, and performs data pre-processing - this two steps are performed
in the autonomous mode. The Data Management Agent cleans the uninformative profiles
like those containing one or two sales facts during all 12 months or have a large number of
missing values. The rest of sales profiles that were not excluded, are normalized with life curve
normalization method (Thomassey & Fioraliso, 2006). Normalizing by the life curve, the new
value xi‘ of each period of a time series X = x1 , . . . , xi , . . . , xn is calculated by Equation 13.
xi
xi‘ = (13)
∑nj=1 x j
The third step of pre-processing is feature selection. During this process a set of the attributes
with descriptive data is produced as a an aggregation of available attributes. The objective of
this process is to lessen the attributes negative impact to the classification results (Thomassey
& Fioraliso, 2006). Attributes can be selected using the information gain or by simple choosing
based on user experience. The third - Dataset formation, process is performed reactively by
responding to the Data Mining Agent’s requests.
3.1.2 Data Mining Agent

The Data Mining Agent is an intelligent agent. The environment it is situated in is the
data from the datasets received from the Data Management Agent. The Data Mining Agent
performs data clustering and creates classification models, changing the environment into the
classification rule sets. The processes in the Data Mining Agent are divided into two main
steps - data clustering and the classification rules induction.
The k-means algorithm (Tan et al., 2006) was chosen to support the data clustering,
though other clustering algorithms can also be applied. The k-means is a simple and
easy-to-understand algorithm, but it has a sufficient weakness - the number of clusters must
be defined before the clustering process begins. Due to that, the clustering process should be
repeated several times with different number of clusters and the best number of clusters may
be chosen by the smallest clustering error. To lessen the time of searching for the appropriate
number √ of clusters an assumption that the maximal number of clusters should be less or equal
to the n, where n is the number of records in the dataset, can be used (Tan et al., 2006). The
k-means algorithm is executed as follows:
1. The number of clusters m is defined, for each cluster centroid random values are assigned.
2. By using a distance measure (the Euclidean distance is used by default) each record in the
dataset is assigned to the closest cluster. If none of records changed its cluster, then the
clustering process finishes.
36
DATA MANAGEMENT AGENT
1. Querying data from 2.1. Data cleaning

database
2.2. Profile
product descriptive data
Historical sales data and
normalization
2.3. Feature
2. Pre-processing data
selection
Pre-processed
data
3. Dataset formation
DATA MINING AGENT

1. Selecting number of clusters and creating sales profiles
C2
Abs. error
C4
C1
Clust. 2 3 4 5 C3
2. Mapping descriptive data with created sales profiles and

building a Classification Decision Tree
Seasonality
Descriptive data + C2
... 1 2
Price Type
Descriptive data + C4 3.99 12.0 6 2
...
Descriptive data + C1 C2 C4 C1 C2
SALES FORECASTING AGENT

Forecasting a sales profile for a new product, using only product
descriptive data
Seasonality
1 2
Price Type
C2 C4
3.99 12.00 6 2
C2 C4 C1 C2 C1 C3
Fig. 8. Multi-Agent System for Sales Forecasting

Trade Management 37
17
3. Cluster centroids are recalculated using the records in the appropriate cluster. Then the
algorithm returns to Step 2.
Having all data distributed among the defined number of clusters, the clustering error should
be calculated in order to show the efficiency of the chosen number of clusters. The clustering
error is calculated as a mean absolute error. First, Equation 14 is used to calculate the absolute
error AEi for each cluster ci . Then the mean absolute error is calculated by Equation 15.
n(c ) i
1
AEi = · ∑ dj , (14)
n ( c i ) j =1
where n(ci ) is the number of records in the cluster ci ; d j - the absolute distance between j-th
record in the cluster ci and the centroid of this cluster.
1 m
m i∑
MAE = · AEi (15)
=1
The result of clustering the sales data is the set P, containing a number of sales profiles -
cluster centroids, example of which is displayed in Figure 9, where the centroid is marked
with a bold line. Each sales profile p is assigned a unique identification number, which will
be used in the classification rule set induction step. The data clustering step finishes with the
0.90
0.80
Normalized sales value
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
-0.10
T1 T2 T3 T4 T5 T6 T7 T8 T9 T10 T11 T12
Periods
Fig. 9. An example of a cluster and a centroid

process of merging the descriptive data with found sales profiles. During this process each
record is merged with an identification number of a closest sales profile. In other words - each
record is assigned an appropriate target attribute value.
In order to forecast a sales curve for a product by having only descriptive data, a model of
the relations between found sales profiles and descriptive data must be created. Assuming
that each sales profile p ∈ P is a class, the classification methods can be applied for building
the desired model. A number of classification methods exist, each having its positive and
negative sights. The proposed sales forecasting system applies the inductive decision trees
for building the classification model. The decision trees not only create a classification model,
which can be represented as a rule set, but also support a graphical tree-like representation of
a model, which makes it more understandable for a user.
38
The Data Mining Agent builds the inductive decision tree using the training dataset, merged
with found sales profiles. It uses the C4.5 classification algorithm (Quinlan, 1993). The C4.5
algorithm is able to proceed with discrete and numeric attributes and to handle the missing
values. The decision tree built is the model representing the relations between descriptive
attributes and the sales profiles. The internal nodes represent the descriptive attributes and
leafs are pointing to one of sales profiles found.
3.1.3 Sales Forecasting Agent

The Sales Forecasting Agent is an agent that uses the model (knowledge base) created by the
Data Mining Agent. The environment the Sales Forecasting Agent is situated in is the set of
new data, containing only descriptive attributes. The environment may include the records
from a test set that have the sales profile attached, while the system is in the evaluation stage.
The Sales Forecasting Agent changes the environment by forecasting a sales profile for the
new data, using the product descriptive data and the model built by the Data Mining Agent.
The sales profile forecasting process can be described as a sequence of the next steps:
1. Chose a record X, containing the descriptive data of a new product.
2. By following the structure of the classification tree, created by the Data Mining Agent, find
a tree leaf l j for the record Xi .
3. Take the sales profile pk , represented by the determined leaf l j , as a forecasted one for the
record Xi .
Figure 10 displays a part of the classification tree and shows the path to the leaf for a record
with the following descriptive data: Kind = 2, Type = 1 and Price = 65.4.
Kind
≤1 >1
... Type
≤8 >8
Type С8
≤7 >7
Price C6
≤ 55 >55
С4 Type
≤1 >1
С4 С8
Fig. 10. Example of a classification decision tree
3.2 Experimental results

The proposed sales forecasting system was trained using sales data of a local retailer of textile
products, gathered in the year 2005. After all necessary pre-processing steps were finished, the
dataset contained 423 records. The 149 sales records from the year 2006 were used for system
Trade Management 39
19
testing. As a descriptive data, the following three parameters were chosen: Kind, showing
either the product is seasonal or not, the Price of the product and the Type, showing what the
product actually is - jeans, t-shirt, bag, shorts, belt etc.
The data clustering was powered by the k-means algorithm and the number of clusters was
chosen by evaluating clustering error with different cluster number starting with two and
continuing till 21, as the square root of 423 records is approximately equal to 20.57. Figure
11 shows the results of clustering error evaluation with different number of clusters. The
0.41
0.39
0.37
0.35
0.33
0.31
0.29
0.27
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21
The number of clusters
Fig. 11. The clustering error for different number of clusters

first noteworthy improvement in clustering error was while using 10 clusters. For larger
numbers of clusters the clustering error was not sufficiently improving, comparing to the
result with 10 clusters. Taking this into account, the number of clusters was set to 10 for all
further experiments.
For building the classification model, the C4.5 classification algorithm was chosen among
other classification methods by comparing six classifiers using such measures as Root Mean
Square Error - RMSE (see Equation 16) and the Mean Absolute Error - MAE (see Equation 17).
The RMSE and MAE were used to measure an error between the forecasted sales profile and
the actual sales profile.
1 n
n i∑
RMSE = · ( f i − x i )2 (16)
=1
1 n
n i∑
· MAE =
| f i − xi | , (17)
=1
where n is the number of periods in the forecasted sales profile F and also in the actual sales
profile X.
Figure 12 displays the classification tree that was built by the C4.5 classification algorithm.
Table 4 shows the results of different classifier evaluation. As can be seen, the C4.5 decision
tree algorithm dominates others in both error measures.
20
40
Kind
≤1 >1
Type Type
≤9 >9
≤8 >8
Type С8 Type Price

≤5 >5
≤7 >7 >16.5 ≤16.5
Type С3 Type Price С10 Price

≤3 >3 ≤40 >40 ≤14.4
≤6 >6 >14.4
Price С8 Type Type Price Price С10 Price

≤55 >55 ≤1 >1 >4 ≤4 ≤36.7 >36.7 >67 ≤67 ≤16 >16
С4 Type Price Price Price Price Price Price С4 С4 С7
Price
Fig. 12. Decision tree for sales profile forecasting

≤1 >1 ≤80 >80 ≤62 >62 ≤30.6 >30.6 ≤39 >39
≤28.15 >28.15 ≤49.5 >49.5
С4 С8 С6 С10 С4 С10 Price С3 Price С5 С3 С1 Price С3
≤48.38 >48.38 ≤36.79 >36.79 ≤30 >30 >46.2

≤46.2
Price Price С10 Price С3 С4 Price С10
≤42.99 >42.99 ≤58 >58

≤41 >41 >41.36 ≤41.36
С7 Price Price Price
С7 С2 Price С5
≤44.25 >44.25 ≤56.9 >56.9 ≤67 >67
≤41.86 >41.86
С1 С8 Price С7 С1 С5 С3 Price
≤53 >53 ≤45 >45

Efficient Decision Support Systems – Practice and Challenges in Multidisciplinary
С6 С5 С5 С8
Domains
Trade Management 41
21
Classifiers
Error measures ZeroR OneR JRip Naive Bayes IBk C4.5
RMSE 0.295 0.375 0.284 0.282 0.298 0.264
MAE 0.174 0.141 0.161 0.156 0.160 0.129
Table 4. Error measures for various classification methods
3.3 Results analysis and summary

The results obtained while testing the proposed multi-agent sales forecasting system show
that a combination of the intelligent agent technology and the data mining methods -
clustering and classification, can be efficiently used for solving such task as the sales
forecasting. The system showed good results in clustering the sales data and building the
decision tree for forecasting a sales profiles for new products, having only descriptive data
available, which is important for retailers while planning the investments for the new stock
products. One of the advances of the system is that a user is able to personally view the
model - classification decision tree that was used for forecasting to find the relations between
the sales profile and the descriptive data, which makes the model more understandable.
Using the intelligent agent technology, the system was made modular, which simplifies the
implementation of new clustering and classification methods.
4. Conclusions
The modern production and trade management remains a complicated research field and
as technology is evolving, becomes even more complicated. This field combines different
tasks of forecasting, planning, management etc. In the present research two of them - PLC
management and sales forecasting, were discussed and possible solutions proposed. Looking
back in time, those tasks were efficiently solved by standard statistical methods and a single
user was able to perform that. Today the level of technological evolution, the amount of
gathered data and complicated planning objectives require the solutions that are able to lessen
the human’s load and are capable to perform different data analysis tasks in the autonomous
mode. In order to gain the above possibilities, the agent technology and combinations of the
intelligent data analysis methods were chosen as the basis. Using the mentioned technologies,
the structures of the PLC management support multi-agent system and the multi-agent sales
forecasting system were proposed and the concepts of each system were described and
discussed. The structures of the proposed systems were intentionally made general in order
to demonstrate the advances of the agent technology. The modularity of the proposed systems
allows one to adapt individual processes to the specific field of application, without breaking
the structure of the system.
Both systems were tested using the real-life data and the results obtained show the
effectiveness of the proposed multi-agent systems and the technologies, chosen for creating
those systems. The conclusion of the present research is that the intelligent agent technology
and other intelligent data analysis technologies can be efficiently applied in tasks connected
with supporting the modern production and trade management. The two proposed
multi-agent systems give an example of such applications.
42
5. Acknowledgement
This work has been supported by the European Social Fund within the project "Support for
the implementation of doctoral studies at Riga Technical University".
6. References
Aitken, J., Childerhouse, P. & Towill, D. (2003). The impact of product life cycle on supply
chain strategy, International Journal of Production Economics Vol. 85(2): 127–140.
Armstrong, J. S., Collopy, F. & Yokum, J. T. (2005). Decomposition by causal forces: A
procedure for forecasting complex time series, International Journal of Forecasting Vol.
21(1): 25–36.
Chang, P.-C., Liu, C.-H. & Fan, C.-Y. (2009). Data clustering and fuzzy neural network for sale
forecasting: A case study in printed circuit board industry, Knowledge-Based Systems
Vol. 22(5): 344–355.
Chien, C.-F., Chen, Y.-J. & Peng, J.-T. (2010). Manufacturing intelligence for semiconductor
demand forecast based on technology diffusion and product life cycle, International
Journal of Production Economics Vol. 128(2): 496–509.
Gomez, J., Dasgupta, D. & Nasraoui, O. (2003). A new gravitational clustering algorithm,
Proceedings of the Third SIAM International Conference on Data Mining, pp. 83–94.
Keogh, E. J. & Pazzani, M. J. (2001). Derivative dynamic timewarping, Proceedings of the First
SIAM International Conference on Data Mining, pp. 1–11.
Kohonen, T. (2001). Self-Organizing Maps, 3rd edn, Springer-Verlag Berlin Heidelberg.
Ni, Y. & Fan, F. (2011). A two-stage dynamic sales forecasting model for the fashion retail,
Expert Systems with Applications Vol. 38(3): 1529–1536.
Quinlan, J. R. (1993). C4.5: Programs for Machine Learning, Morgan Kaufmann publishers.
Salvador, S. & Chan, P. (2007). Fastdtw: Toward accurate dynamic time warping in linear time
and space, Intelligent Data Analysis Vol. 11(5): 561–580.
Tan, P. N., Steibach, M. & Kumar, V. (2006). Introduction to Data Mining, Addison-Wesley.
Thomassey, S. & Fioraliso, A. (2006). A hybrid sales forecasting system based on clustering
and decision trees, Decision Support Systems Vol. 42(1): 408–421.
Thomassey, S. & Happiette, M. (2007). A neural clustering and classification system for sales
forecasting of new apparel items, Applied Soft Computing Vol. 7(4): 1177–1187.
Venkatesan, R. & Kumar, V. (2002). A genetic algorithms approach to growth phase forecasting
of wireless subscribers, International Journal of Forecasting Vol. 18(4): 625–646.
Weiss, G. (ed.) (1999). Multiagent Systems. A Modern Approach to Distributed Artificial
Intelligence, MIT Press.
Wooldridge, M. (2005). An Introduction to MultiAgent Systems, 3rd edn, John Wiley & Sons,
Ltd.
Wright, W. E. (1977). Gravitational clustering, Pattern recognition Vol. 9: 151–166.
3
Modeling Stock Analysts Decision Making:

An Intelligent Decision Support System
Harry Zhou
Department of Computer and Information Sciences
Towson University
MD. USA
1. Introduction
Over 60% of the families in the United States have billions of dollars invested in mutual
funds. Consequently portfolio managers are under tremendous pressure to make critical
investment decisions in dynamically changing financial markets. Experts have been
forecasting and trading financial markets for decades, using their knowledge and expertise
in recognizing patterns and interpreting current financial data. This paper describes a
knowledge based decision support system with the analogical and fuzzy reasoning
capabilities to be used in financial forecasting and trading. In an attempt to maximize a
portfolio’s return and avoid costly losses, the portfolio manager must decide when to enter
trades as well as when to exit and thus, must predict the duration as well as the direction of
stock price movement. A portfolio manager is faced with a daunting information
management task and voluminous amounts of rapidly changing data. It is simply an
impossible task to manually and simultaneously follow many investment vehicles, stocks
and market moves effectively.
To assist portfolio managers’ decision making processes, tens of millions of dollars are spent
every year on Wall Street to automate the process of stock analysis and selection. Many
attempts have been made at building decision support systems by employing mathematical
models and databases(Liu et al, 2010; Tan 2010). These traditional decision support systems
are composed of three components: a database, a user interface, and a model base. The user
interface allows the end user (in most cases, a manager) to select appropriate data and
models to perform an analysis for a given stock. The database contains data necessary to
perform the analysis, such as market capitalization, shares outstanding, price ratio, return
on equity, and price to book ratio. The model base consists of mathematical models to deal
with the present value of expected future payoffs from owning the stock, such as the
dividend discount model, price-to-earning ratio(P/E) models, and their variants. Standard
decision support systems employ quantitative approaches to decision making. Specific
quantities in the form of payoffs and probabilities are used to arrive at a quantitative
expected value. The decision maker simply selects the alternative that has the highest
expected value. Despite the mathematical soundness and systematic approach of traditional
decision support systems, researchers have discovered compelling reasons for developing
qualitative approaches in decision making. It is inappropriate to suggest that people reason
about decisions in a purely quantitative fashion. In the domain of security analysis and
trading, if we expect a stock to go up in value, we are likely to invest in the stock despite not
knowing quantitatively how much or how soon a stock will rise. A qualitative analysis of a
decision problem is usually logically prior to a quantitative analysis(Sen, 2009).
In order to develop a decision support system with the capabilities of qualitative analysis,
considerable progress has been made toward developing “knowledge-based decision
support systems”, sometimes known as expert systems (Silverman, 1992). An expert system
designed in the domain of security analysis is an intelligent decision support system whose
behavior duplicates, in some sense, the ability of security analysts. Expert systems are
computer programs with characteristics that include the ability to perform at an expert
level, representing domain-specific knowledge and incorporating an explanation of
conclusions reached. Traditional expert systems consist of five major components: an end
user interface, an inference engine, a working memory, a knowledge base, and an
explanation mechanism. The interface allows the end user to input data to and receive
explanations or conclusions from the expert system. The working memory keeps track of
specific data, facts and inferences relevant to the current status of the problem solving. The
knowledge base contains problem-solving knowledge of a particular domain collected from
an expert. The inference engine matches data in the working memory with knowledge in the
knowledge base, and acts as the interpreter for the knowledge base. By combining the five
components together, an expert system is able to act as a knowledgeable assistant and
provide “expert quality” advice to a decision maker in a specific domain.
Despite the impressive performance of expert systems, they have a number of inherent
flaws, especially when operating in a dynamically changing environment. An expert system
solves problems by taking input specifications and then “chaining” together an appropriate
set of rules from the knowledge base to arrive at a solution. Given exactly the same problem
situation, the system will go through the same amount of work to come up with a solution.
In addition, traditional expert systems are “brittle” in the sense that they require substantial
human intervention to compensate for even slight variations in descriptions, and break
down easily when they reach the edge of their knowledge(Zhou, 1999).
In response to weaknesses associated with traditional decision support systems and expert
systems, this paper presents a knowledge-based and case-based intelligent decision support
system with fuzzy reasoning capabilities called TradeExpert, designed for assisting portfolio
managers to make investment decisions based not only mathematical models, but also facts,
knowledge, experiences, and prior episodes.
2. Efficient market hypotheses and trading opportunities

Since the 1960’s, the theory of Efficient Market Hypothesis (EMH) has been popular and
well-known in the investment community(Eugene,1965; Malkiel, 1987). The basic idea
behind EMH is that at any point in time, prices of stocks in an efficient market reflect the
information from a variety of sources available to all investors involved in the market.
According to EMH, past stock prices, movements, and patterns provide little or no guidance
as to the direction of future prices. EMH asserts that stock markets are "informationally
efficient". That is, one cannot consistently achieve returns in excess of average market
returns on a risk-adjusted basis, given the information publicly available at the time the
investment is made. The validity of the hypothesis has been questioned by critics who blame
the belief in rational markets for much of the financial crisis of 2007–2010(Fox, 2009; Nocera,
2009). Since the 1980’s, there was a resurgence of interest in questions of the accuracy of
Modeling Stock Analysts Decision Making: An Intelligent Decision Support System 45
EMH. A variety of empirical tests developed new and often elaborate theories that sought to
explain subtle security-pricing irregularities and other anomalies. Market strategist Jeremy
Grantham has stated flatly that the EMH is responsible for the current financial crisis,
claiming that belief in the hypothesis caused financial leaders to have a "chronic
underestimation of the dangers of asset bubbles breaking"(Nocera, 2009).
More recently, institutional investors and computer-driven trading program appeared to be
resulting in more anomalies. Anomalies may also arise from the market’s incomplete
assimilation of news about particular securities, lack of attention by market participants,
difference of opinion about the meaning and significance of available data, response inertia
and methodological flaws in valuation(Robertson & Wolff, 2006). Findings like the
preceding suggest that the EMH is not perfect representation of market realities that behave
as complex systems. Because security markets are complex, time-variant, and probably
nonlinear dynamic systems, one simple theory cannot adequately represent their behavior;
From time to time there may arise anomalies that are exploitable. In the design of
TradeExpert, technical analysis rules, trend analysis rules, and fundamental analysis rules
are used to exploit whatever inefficiencies may exist and unearth profitable situations.
Technical analysis is the systematic evaluation of price, volume and patterns for price
forecasting(Wilder 1978). There are hundreds of thousands of market participants buying
and selling securities for a wide variety of reasons: hope of gain, fear of loss, tax
consequences, short-covering, hedging, stop-loss triggers, price target triggers, broker
recommendations and a few dozen more. Trying to figure out why participants are buying
and selling can be a daunting process. Technical analysis puts all buying and selling into
perspective by consolidating the forces of supply and demand into a concise picture. As a
complete pictorial record of all trading, technical analysis provides a framework to analyze
the battle raging between bulls and bears. More importantly, it can help determine who is
winning the battle and allow traders and investors to position themselves accordingly.
Technical analysis uses trading indicators and stock trading prices to analyze the stock
trend. Trading indicators, e.g. Moving Average, Bollinger Bands, Relative Strength Index,
and Stochastic Oscillator, provide trading signals which can be used to determine when to
trade stocks.
Trend analysis, also known as Momentum Trading(Simple, 2003), is the process by which
major and minor trends in security prices are identified over time. The exact process of trend
analysis provides investors with the ability to identify possible changes in price trends. As
securities prices change over time it will exhibit one of three states: uptrend, downtrend or
sideways consolidation. It is the job of the technical analyst to identify when a security may
change from one trend to another as these may provide valid trading signals. Trend analysis
uses a fixed trading mechanism in order to take advantages from long-term market moves
without regards to past price performance. Trend analysis is a simplistic trading strategy
that tries to take advantage of stock price movements that seem to play out in various
markets. Trend analysis aims to work on the market trend mechanism and take benefits
from both sides of the market. It gains profits from ups and downs of the stock market.
Traders who use this approach can use current market price calculation, moving averages
and channel breakouts to determine the general direction of the market and to generate
trade signals.
Fundamental analysis maintains that markets may misprice a security in the short run but
that the "correct" price will eventually be reached. Profits can be made by trading the
mispriced security and then waiting for the market to recognize its "mistake" and reprice the
security. One of the most important areas for any investor to look when researching a
company is the financial statement of that company. Fundamental analysis is a technique
that attempts to determine a security’s value by focusing on underlying factors that affect a
company’s actual business and its future price.
To summarize, technical analysis, fundamental analysis, and trend analysis do not really
follow the theory of Efficient Market hypothesis. They attribute the imperfections in
financial markets to a combination of cognitive biases such as overconfidence, overreaction,
representative bias, information bias, and various other predictable human errors in
reasoning and information processing.
3. System architecture
TradeExpert is a hybrid system that uses both quantitative and qualitative analysis to
exploit changing market conditions and to grind through mountains of economic, technical
and fundamental data on the market and on individual stocks. Furthermore, TradeExpert
incorporates the mechanism of case-based reasoning which enables it to recall similar episodes
from the past trading experience, and the consequences of that action. The goal, of course, is to
ferret out patterns from which future price action can be deduced, to meaningfully assimilate
vast quantities of information automatically, and to provide support intelligently for portfolio
managers in a convenient, timely and cost-effective manner.
TradeExpert assumes the role of a hypothetical securities analyst who makes a
recommendation for a particular stock in terms of strong buy, buy, hold, sell and strong sell
based on chart reading, fundamental analysis, investment climate, and historical
performance. TradeExpert is designed and implemented based on principles consistent with
those used by a securities analyst(Kaufman 2005). It is capable of reasoning, analyzing,
explaining, and drawing inferences. In terms of system architecture, TradeExpert consists of
the following major components: a user interface, databases, knowledge bases, a case base, a
similarity evaluator, an explanation synthesizer, a working memory, and an inference
engine, shown in the following diagram 1. The user interface provides communication
between the user and the system. Input data from the user to the system contains the
parameters of a stock under consideration, such as Yield, Growth rate, P/E, Return on net
worth, Industry sector, Profit margins, Shares outstanding, Estimated earning per share,
Price to book ratio, Market capital, Five-year dividend growth rate, Five-year EPS growth
rate, and Cash flow per share. If a stock being analyzed is already stored in the database,
TradeExpert retrieves data and fills in the parameter fields on the input screen
automatically. The fuzzy inference component employs fuzzy logic and reasoning to
measure partial truth values of matched rules and data. It makes reasoning processes more
robust and accurate. The output from the system is a trading recommendation with a list of
reasons justifying the conclusion. With human expertise in the knowledge base, past trading
experience in the case base, and stock’s data in the database, TradeExpert is capable of
analyzing a stock, forecasting future price movements, and justifying the conclusion
reached.
4. Knowledge bases and organization

The knowledge bases capture the knowledge of human securities analysts in the form of If-
Then rules. In the design of TradeExpert, the knowledge base is divided into three separate
knowledge bases: technical analysis knowledge base, fundamental analysis knowledge base
and trend analysis knowledge base. TradeExpert can use any or all of these different but
somewhat complementary methods for stock picking. For example it uses technical analysis
for deciding entry and exit points, uses trend analysis for deciding the timing of a trade, and
uses fundamental analysis for limiting its universe of possible stock to solid companies. As a
justification of the expert system serving as a model of human thought processes, it is
believed that a single rule corresponds to a unit of human knowledge. In the design of an
intelligent decision support system, knowledge representation for human problem solving
expertise is a critical and complex task.
Case Bases Databases
Industry Economic
Stock Company Industry
Group Politcal
History Performance Impacts Data Data
User
Inference engine Working Memory Similarity Evaluatior
Interface
Trend Technical Fundamental

Explanation Conflict
Analysis Analysis Analysis
Synthesizer Rules Rules Rules Resolution
Knowledge Bases
TradeExpert Architecture
Fig. 1.
In what follows, the knowledge representation of stock analysis and selection in
TradeExpert is described along with a rule frame as an example (Trippi & Lee 1992).
RULE <ID>
(INVESTMENT HORIZON: <statement>)
(ASSUMPTION: <statement>)
(INVESTMENT OBJECTIVES: <statement>)
(IF <statement> <statement> ..... <statement>)
THEN <statement>)
(REASONS: <statement>)
(UNLESS: <statement>)
(CERTAINTY FACTOR: <real number>)
The key word INVESTMENT HORIZON designates the period of the investment: day
trading, short term or long term. The key word INVESTMENT OBJECTIVES indicates the
investment objective of the user, such as trading for profits, income, and capital
appreciation. The key word REASONS provides justifications and explanations of how a
rule arrives at a conclusion. The key word UNLESS excludes the stock even when the
hypothesis(condition) part of a rule is satisfied. The key word CERTAINTY FACTOR

contains a real number in the range of (0..1) measuring the strength of evidence in support
of the rule’s hypothesis. In order to invoke a rule, both the hypothesis part and the
ASSUMPTION part must be met. Given a situation, investors may reach different
conclusions and select different trading strategies based on different assumptions. Some
examples are: “Can the Asian financial crisis be over in the next 12 months?”, “Will the U.S.
dollar be stronger or weaker over the next 3 months?”, or “Will the FED cut interest rates at
its next meeting?” Of course, different answers to these questions lead to different trading
strategies. Since investing is a subjective process, TradeExpert allows the user to determine
the future direction of financial markets and then select stocks accordingly. One of the rules
in the knowledge bases is listed below:
RULE <31>
(INVESTMENT HORIZON: <day trading>)
(ASSUMPTION: <bull market>)
(INVESTMENT OBJECTIVES: <trading profits>)
(IF
<Company is in internet-sector> AND
<internet-sector = Hot> AND
<Avg-peer-earnings = beat the estimate> AND
<estimated EPS = increase over last two months> AND
<product = high demand> AND
<average-daily-volume >= 1.5 average) AND
< price-range = new 52-week high> AND
< liquidity = good>
THEN <Short term Strong Buy> )
(REASONS:
<technical analysis signals a new high> AND
<price movement trend is favorable> AND
<stock momentum carries over> AND
<small number of outstanding shares> AND
<history performance of hot stocks> )
(UNLESS:
<downgrades-by-analysts > 2 in last 5 days> OR
<earning-warning-by-peers > 1 in last 5 days>)
(CERTAINTY-FACTOR: 0.85)
With a set of reasons associated with the conclusion of a rule, TradeExpert is able to justify
its reasoning by providing evidence and rational of its conclusion in addition to a certainty
factor. In order to retrieve appropriate rules efficiently, the rules in TradeExpert are
organized in a hierarchical structure. They are classified in terms of their scope and
attributes. One rule designed for day trading may not be important at all for long-term
investment and vice versa. A technical break out of stock price and daily price fluctuation
may not have impact on long-holding investors. Classifying rules into different categories
makes logical sense in terms of knowledge retrieval and maintenance. Of course, one rule
may be indexed by multiple categories so that the match and evaluation steps during the
inference process can be completed efficiently. Another factor for consideration is an
investor’s subjective judgmental call, such as the assessment and long term view of a
particular market. Given current economic data and political environment, different
investors market be up or down over the next 6 months?” will definitely elicit several
answers from the investor community. A partial diagram that shows the knowledge
organization (Trippi & Lee 1992) in TradeExpert is presented in figure 2.
Fig. 2. Knowledge Organization

It is worth noting the above diagram is a simplified version, many levels and branches of
indices have been left out. Hot stocks that satisfy certain criteria such as high daily trading
volume, large percentage price gain and belonging to a best-performing industrial sector are
identified and collected daily.
4.1 Fundamental analysis knowledge base

There are two models of investing: growth-oriented and value-based. Growth-oriented
investing seeks companies that show consistent earnings and sales growth, usually 20% or
more per annum for the past three to five years. These stocks usually carry a high PE and
have a record of better-than-average earnings growth. They usually beat analyst estimates
by a large percentage quarter after quarter. Acceleration in earning growth often indicates
the future upward price movements of a stock. Value-based investing looks for bargains and
gravitates to stocks with a low price/ratio(P/E) and a low price-to-book value. As a
foundation of growth-oriented and value-based investing, fundamental analysis relies on
the following factors to make an evaluation: P/E, price-to-cash flow, market value-to-book,
sales-to-price, debt ratios, yield, and projected earnings. It is a method of evaluating
securities that attempts to measure intrinsic values by examining related economic, financial
and other qualitative and quantitative factors. Fundamental analysis views the stock market
as a relatively ordered system and each company is represented by factors such as financial
ratios and performance measures, which are used to draw inferences. The rules in the
fundamental analysis knowledge base are concerned with analysis of the company itself and
deriving estimates of net worth per share, current assets, future earnings and dividends, and
other related measures of value. The fundamental analysis rules attempt to estimate the
future earning of a stock and therefore define a stock’s true value. It performs analysis on
industries or the economy as a whole. The term refers to the analysis of the economic well-
being of a financial entity as opposed to its price movements. It does not pay attention to
daily price movements and short term fluctuations. The investment horizon is often in the
range of multiple years, not days and weeks.
Fundamental analysis serves to answer questions, such as:

 Is the company’s revenue growing?
 Is it actually making profits?
 Is it in a strong-enough position to beat out its competitors in the future?
 Is it able to repay its debts?
One of the rules in the fundamental knowledge base is listed below:
RULE <16>
(INVESTMENT HORIZON: <long term>)
(ASSUMPTION: <bull or bear market>)
(INVESTMENT OBJECTIVES: <return on capital>)
(IF
<PE < similar_stock_average> AND
<earning_history = stable> AND
<dividend pay-out history = consistent> AND
<market_value_to_book < 1.05> AND
<Company is believed to be profitable> AND
<credit-rating > A or better> AND
<debt rations = low> AND
THEN <Long term Strong Buy> )
(REASONS:
<good credit rating> AND
<its intrinsic value = attractive> AND
<consistent dividend payout>
(UNLESS:
<law suit = pending> OR
<In the middle of financial_scandal> OR
< Bankruptcy = likely>
Fundamental analysis is a technique that attempts to determine a security’s value by
focusing on underlying factors that affect a company's actual business and its future
prospects. It believes that the intrinsic value of the stock eventually drives long-term stock
prices. It looks at revenue, expenses, assets, liabilities and all the other financial aspects of a
company to gain insight on a company's future performance and therefore, the stock’s
future price.
4.2 Technical analysis knowledge base

Fundamental analysis tells us what stocks are capable of but does not tell us just how high
or low stock prices will go or when they will peak or bottom out. Technical analysis works
in opposite principle of the fundamental analysis(Edwards & Magee, 1992). The underlying
philosophy for this approach is that market prices well reflect all known factors at all times.
So the price is already a solid performance indicator as a result of the supply and demand
for that particular stock. Therefore technical analysis evaluates solely on market prices
themselves rather than on any fundamental factor outside the market. The rules in the
technical analysis knowledge base are concerned with studies of supply and demand for a
stock and statistics which are related to price and volume. Technical indicators are tools
invented on concept of technical analysis. There are many kinds of technical indicators that
have been developed to gain insight about the market behavior. Each of them may have
different concept about when to buy and sell (trading signals). Some well-known and
widely used technical indicators are listed below briefly. Moving Average (MA) is used to
remove market noise and find the direction of prices. It is calculated by sum up the stock
prices over n days divided by n. Relative Strength Index (RSI) compares the magnitude of
the stock’s recent gains to the recent losses then turns into the number ranged from 0 to 100.
It compares the magnitude of recent gains to recent losses in an attempt to determine
overbought and oversold conditions of an asset. It is calculated using the following formula:
RSI = 100 - 100/(1 + RS*), where RS = Average of x days' up closes / Average of x days'
down closes. If RSI is less than 30, it shows an oversold market condition. Stochastic
Oscillator (SO) is an oscillator that measures the relative position of the closing price within
a past high-low range. Volatility Index(VIX) is derived from the prices of ALL near-term at-
the-money call and out-of-the-money call and put options. It is a measure of fear and
optimism among option writers and buyers. When a large number of traders become
fearful, then the VIX rises. Conversely, when complacency reigns, the VIX falls. The VIX is
sometimes used as a contrarian indicator and therefore can be used as a measurement of
how oversold or overbought the market is. Usually it increases as the market decreases. Its
usefulness is in showing when a trend reversal is about to take place.
One popular form of technical analysis is charting, in which the graphic display of past price
performance, moving-average trends, cycles, and intro or inter-day stock price ranges are
studied in an attempt to discern cues for profitable trading. Stock price pattern analysis is
the basis of the technical analysis of stocks. It relies on the interpretation of some typical
configurations of the ups and downs of price movements like “head and shoulders”, “top
and bottom formations” or resistance lines. Stock price pattern analysis comes down to
comparing known patterns with what is evolving on the chart. Charting makes use of
techniques such as moving indices, trend analysis, turning-point indicators, cyclical spectral
analysis, and the recognition of various formations or patterns in prices in order to forecast
subsequent price behavior. Some examples of patterns are assigned names such as “flag”,
“triangle”, “double bottom or top”, “symmetrical triangles”, and “head-and-shoulders”.
The rules in the technical analysis knowledge base represent the expertise of chart reading
without concerning the fundamentals of a company. By analyzing MA, RSI, SO, VIX, and
trading volumes of a stock, the technical analysis rules suggest entry points as well as exit
points, taking advantage of market fluctuations instead of being victimized by them. One
technical analysis rule used by TradeExpert is presented below:
RULE <92>
(INVESTMENT HORIZON: <short term>)
(INVESTMENT OBJECTIVES: <trading for profits>)
(IF
<RSI < 35> AND
<10_day MA > 30_day MA> AND
<VIX > 70) AND
< Chart reading shows a near complete W shape> AND
<average-daily-volume >= 1.5 average)
(REASONS:
<price moving average shows an uptrend direction> AND

<technical analysis signals a breakout> AND
< Volatility index shows a reversal is imminent> AND
<stock relative strength shows the market is oversold> AND
<Daily volume > daily average>
)
(UNLESS:
<Overall market is extremely weak> OR
<down_grades > 2 in last 7 days>)
Technical analysis maintains that all information is reflected in the stock price. Trends 'are
your friend' and sentiment changes predate and predict trend changes. Investors' emotional
responses to price movements lead to recognizable price chart patterns. Technical analysis
does not care what the 'value' of a stock is. Their price predictions are only extrapolations
from historical price patterns. Like x-rays and brain scans are used to diagnose diseases and
seismic data is used to exploit oil, technical analysis is used to help determine the behavior
and future price movements of a stock. TradeExpert as well as investors can use technical
analysis to buy stocks and sell stocks at the most advantageous and profitable prices.
4.3 Trend analysis knowledge base

Instead of striving to predict a market direction, Trend analysis, also known as Momentum
Trading, reacts to the market’s movements whenever they occur(Eng 1993). It responds
meticulously to what has recently happened and what is currently happening, rather than
anticipating what will happen.
Unlike the strategy “buy low and sell high” often employed in value-based investing, the
trend analysis rules are based on a different approach: “buy high and sell higher.” The basic
rational behind this approach is based on the well known Wall Street statement: “go with
the tape”. That is, if you try to short a stock in an up trend, you will get killed. The same is
true if you try to go long on a stock in a down trend. Since investing is an art not a science, a
reasonable guess is that tomorrow will likely follow the trend established over the past few
days. The stocks which have been strong relative to all other stocks should continue to be
relatively stronger in the near future. Trend analysis rules reflect and implement this idea in
TradeExpert. The rules in the trend analysis knowledge are designed and selected to
recognize some short-term persistence in markets and to identify the stocks that may
generate significant returns in a short period of time. Fortunes are made by those who
recognize trends and ride them. In the period of 1997-1998, the internet sector was white
hot. A company would have done much better if it could create a website than buying back
its own stock to pop up its stock’s price. The trend then was internet-related stocks. For
example, the 400 plus percentage gain of Amazon and Yahoo, the gravity-defying internet
stocks, was made in the first 10 months of 1998. These rules attempt to unearth stocks whose
prices move far beyond any reasonable estimate of intrinsic value and often outperform the
market index averages by a big margin in a short period of time.
In what follows, one formula used in trend analysis rules and momentum trading is
presented (Schulmeister, 2002). It consists of a short-term moving average (MASj) and a
long-term moving average (MALk) of past prices. The length j of MAS usually varies
between 1 day (in this case the original price series serves as the shortest possible MAS) and
10 days, the length k of MAL usually lies between 10 and 30 days. (if one uses 30 minutes
data, then Buy (go long) when the short-term (faster) moving average crosses the long-term
(slower) moving average from below and sell (go short) when the converse occurs. Or
equivalently: Open a long position when the difference (MASj−MALk) becomes positive,
otherwise open a short position. If one expresses this difference as percentage of MALk one
gets the moving average oscillator:
MONTENTUM(j,k)t = [(MASj,t – MALk,t)/ MALk,t] * 100

This type of representation facilitates a (graphical) comparison of the signal generation
between moving average models and momentum models. Another way to express the basic
trading rule is then: Hold a long position when MONTENTUM is positive, hold a short
position when MONTENTUM is negative. The second type of model works with the relative
difference (rate of change in %) between the current price and that i days ago:
M(i) = (Ptcurrent – Pti)

The basic trading rule of momentum models is as follows(Carter, 2006):
Buy (go long) when the momentum M(i) turns from negative into positive and sell (go
short) in the opposite case. Or equivalently: Hold a long position when M is positive, hold a
short position when M is negative. The variables MONMENTUM(j,k) or M(i) are called
“oscillators” because they fluctuate around zero. The basic trading rule of moving average
models and momentum models is trend-following since MONMENTUM(j,k)t and M(i)t,
respectively, are positive (negative) only if an upward (downward) price movement has
persisted for some days (or some 30 minutes intervals). When and how often
MONMENTUM(j,k)t and M(i)t, respectively, cross the zero line depends not only on the
persistence of recent price movements but also on the lengths of moving averages and the
time span i in the case of momentum models, respectively.
As an example, a rule in the trend analysis knowledge base is presented below:
RULE <45>
(INVESTMENT OBJECTIVES: <trading for profits>)
(IF
<growth_rate > PE> AND
<Momentum > 0) AND
<the industrial sector = Hot> AND
<reported-peer-earnings = beat the estimate> AND
<floating-shares = small> AND
<estimated EPS = increase over last two months> AND
<earning_release_day < 5 days> AND
<average-daily-volume >= 1.5 average)
(REASONS:
<possible short squeeze> AND
<Stock price is on uptrend> AND
<momentum before earning release> AND
<expectation of good earning report> AND
<historical performance of hot stocks before
earning> )
(UNLESS:
<downgrades-by-analysts > 2 in last 5 days> OR
<earning-warning-by-peers > 1 in last 5 days>)
Trend analysis observes a security price movement, times an exit and entry point, analyzes
price change directions, and jumps on a trend and rides it for the purpose of making quick
profits. It tries to profitably exploit the frequent occurrence of asset price trends (“the trend
is your friend”).Hence, this trading technique derives buy and sell signals from the most
recent price movements which indicate the continuation of a trend or its reversal.
5. Object-oriented databases
The objected-oriented database contains financial data of industries and companies, such as
the company name, debt ratio, price-earning ratio, annual sales, value to book value and
projected earnings(Trippi & Lee 1992). An object in the database is defined by a vector of
three components: <I, P, M>. The component I is a unique ID or name for the object, which
may represent an industry or a company. The component P is a set of properties belonging
to an object represented as a vector. The vector characterizes an object in terms of its
attributes. The component M is a set of methods implemented by member functions. In
addition to direct input, these functions are the only means of accessing and changing the
attributes of objects. There are two kinds of objects in our database: industry-based objects
and company-based objects. The company-based objects contain company-specific data,
such as P/E ratio, projected growth rate, debt ratio, and annual sales. The industry-based
objects contain data and information related to a specific industry, such as tax benefits, stage
of life cycle, sensitivity to interest rates, legal liability, and industry strength.
Examples of such objects are shown below:
An industry-based object:
(<ID> :- industry-name (xyz))

(<ATTIBUTES> :-
(Tax-Benefits: (direct-input))
(Stage-in-life-cycle:(direct-input))
(total-sales: (method1))
(daily-volumes: (method2))
(Sensitivity-to-interest-rate: (direct-input))
(Potential-legal-liability: (direct-input))
(industry-strength: (method3))
(<METHODS> :-
(method1: sum up sales from companies)
(method2: sum up from related stocks)
(method3: average up from companies)
))
A company-based object:
{(<ID> :- company-name <abc>)
(<Attributes> :-
(Sales: <database retrieval>)
(Sales-growth-rate: <database retrieval>)
(Debt-ratio: <database retrieval>)

(Tax-benefits: <inherit from industry>)
(High/low price: method1)
(daily volume: method2)
(industry-strength: <inherit from industry>)
(number-shares-outstanding: <database retrieval>)
(shares-sold-by-insiders-recently: <database retrieval>)
(relative-strength: method3)
(gross-margin-on-sales: <database retrieval>)
(<Methods> :-
(method1: update-daily-price)
(method2: update-daily-volume)
(method3: calculate-average-strength)
)}
In the object-oriented database, company objects and industry objects are organized in a
hierarchy where industry objects are at higher levels and related company objects are
considered subclasses. Each object represents either an industry sector or a company.
Specific objects have descriptive attributes and methods that they can invoke to perform
certain tasks. These attributes and methods are inherited from super classes to subclasses.
Thus, the principles of inheritance, average up, and sum up will apply. For example, to
obtain the total sales figure, the function Sum-up-sales-from-companies would be invoked
to calculate the total sales from all companies indexed by the industry object. On the other
hand, all the indexed company objects would inherit the parent object’s attributes, such as
industry-strength, tax-benefits, and sensitivity-to-interest-rate which are applicable to every
company within the same industry sector.
6. Case base and case-based reasoning

The process of developing an effective trading system must take into consideration the
current situation and past data. History often repeats itself. Similar behaviors may occur
again despite not being in the same domain or industry. Lessons learned in the past can
provide useful guidance in terms of a stock price changes over time. The difference between
a recent business school graduate and a seasoned securities analyst is that the former has
learned everything from textbooks and knows the rules of securities analysis and trading,
while the latter knows all this but in addition, has experienced up markets as well down
markets. Therefore, the seasoned analyst has witnessed thousands of trading episodes
which supply a rich context for recognizing and analyzing new problems. TradeExpert
employing case-based reasoning(CBR) provides a contrast to traditional expert systems and
is analogous to comparing an experienced analyst to a new MBA. Being a cognitive model of
human securities analysts, TradeExpert is capable of recalling similar cases seen in the past,
retrieving and modifying the results of those cases and then deducing the reasoning behind
those results(Kolodner 1983). If no solution is found, TradeExpert does not give up like a
traditional expert system, instead it employs a heuristic approach in which reasoning can be
guided, past experiences can be recalled, and solutions to new and similar problems can be
constructed.
A case in the case base can be described as follows:
Case base {C} := {<C1><C2> . . . <Ci> . . <Cp>}

Case Ci := {<Ai><Ri><Pi><Ji><Mi>}
Attributes Ai := {<ai1><ai2> . . <aij> .. <aim>}
Recommendation Ri := {<strong-buy>, <buy>, <hold>,
<sell>, <strong-sell>}
Performance Pi := {<direction> <percentage> <time-frame>}
Justification Ji := {<Ji1><Ji2> . . <Jij> .. <Jis>}
Method Mi := {<mi1><mi2> .. <mij> .. <mik>}
A case Ci represented by an object consists of a set of attributes Ai, a recommendation Ri
made by a security analyst, a performance figure Pi that measures the security's
performance after the recommendation, and a list of methods that perform related
computation. The attributes vector Ai represents financial data and characteristics of the
security under consideration. The performance value Pi measures the performance of the
security after recommendation in terms of the price change percentage. A methods Mi is
implemented by a function that provides means to access databases or updates the data of
the case. The justification Ji lists reasons and explanations on how and why the analyst
arrived at the recommendation.
Analogical problem solving is one of the most complex techniques in human cognition. Yet,
people have no difficulties in recalling similar episodes as new problems occur and applying
them where appropriate. The design of similarity evaluation in TradeExpert is
psychologically motivated and experimentally tested. The similarity evaluator, as shown in
the system architecture diagram, consists of two parts: an external impact evaluator and a
feature evaluator. The external impact evaluator is concerned with the impact of external
forces on the performance of stocks and is not directly related to the fundamentals of
companies, such as political pressures from the federal government, international events, or
economical conditions. An example of the cases in the case base is presented below:
CASE 24:
(
{ATTRIBUTES:
<time-frame: 1962>
<industry-sector: steel makers>
<R/D spending: insignificant>
<price-competition: fierce>
<long-term growth rate: < 10% >
<management/worker-relation: intense>
.
.
}
{RECOMMENDATION: Sell}
{PERFORMANCE: down 50% on average in 6 months}
{JUSTIFICATIONS: Caused by JFK’s policy}
{METHODS: calculating industry-related data}
)
In 1962, President John Kennedy declared war on steel companies blaming the higher labor
cost and management. The sudden change of federal policy caused the steel stock prices to
drop more than 50% in a very short period of time and remained low for months to come.
This case remembers the consequences of an unfavorable drastic change in federal policy
toward a particular industrial sector, and the dramatic price decline in the months ahead. In
1992, President Bill Clinton proposed a reform to national healthcare widely seen as
criticism of pharmaceutical industry’s practice and pricing. Consider the following pairs:
presidents(John Kennedy, Bill Clinton) federal policies(against the steel industry, against the
pharmaceutical industry), one could easily see the consequence of political pressure on the
price of affected stocks. When given a drug manufacturer’s stock, TradeExpert is able to
draw inferences from its past experience and to make a SELL recommendation based on the
historical lessons learned. The days that follow show the significant price decline of all
pharmaceutical stocks in the next 12 to 18 months. Without the case-based reasoning
mechanism, a decision support system would not be able to make such a predication since
these two industry sectors are very different in terms of basic attributes, such as industry-
sector, R/D spending, price-competition and long-term growth rate.
In addition to the external impact evaluator, TradeExpert also employs a feature matching
process. The similarity score is determined by a combination of features in common and the
relative importance, expressed by weights, of these features(Simpson, 1985). Formally, the
feature evaluator can be described as follows.
Let N be a case with m features:
N = {n1 n2 .. nm}
and O is an old case with k features:
O = {o1 o2 .. ok}
CF denotes a common feature set
CF = {c1 c2 .. ct}
More specifically,
CF =  c  CF where c  N  c  O 
Thus, a similarity score, S(N,O), of a new case N with respect to an old case O is given as:
 k 
  i  c i   k
S(N,O) =  i 1  1ik
m
where i is a weight assigned to the ith feature of a case.
With the case-based reasoning mechanism built in, TradeExpert is able to accumulate vast
and specialized expertise in the domain of stock analysis and trading, and to apply it to new
situations.
A case in point is the similar price movements of casino stocks in the early 90’s and internet
stocks from the period of 1997-2000. Both sectors have many attributes in common, such as
expected high growth rates, high P/E, daily volumes far exceeding average trading
volumes, a record of better-than-average earnings, and investors’ favorable sentiments. By
employing the feature similarity evaluator, TradeExpert is able to identify the upward price
movements of internet-related stocks, while ignoring unbelievably high P/Es associated
with them. As hot stocks in different periods of 90’s, many stocks in these two sectors
generated 3-digit returns in a matter of 12 months. Another example of analogical reasoning
can be found in the banking sectors both in Great Depression of 30’s and in Great Recession
of 2007-2010. During these years, the stocks of all banks went down and stayed low for a
long period of time. With regard to particular stocks, let us consider the examples of Google
(GOOG) and Baidu (BIDU), a Chinese search engine. Having observed an impressive run by
GOOG, it can be predicated with confidence that an IPO of BIDU would do very well
considering the dominant market presence and the customer base. Indeed, Bidu actually
went from $12(split adjusted) in 2005 to a high of $118 in 2011.
In general, given a stock, TradeExpert searches for similar cases in the case base, establishes
correspondence between past experience and the stock currently under consideration, and
then transforms related experience to a recommendation for the stock.
7. Fuzzy reasoning and inference

Many decision-making tasks of investors are too complex to be understood quantitatively.
However, humans succeed by using knowledge that is imprecise rather than precise. Fuzzy
logic refers to a logic system which represents knowledge and reasons in an imprecise or fuzzy
manner for reasoning under uncertainty (Baldwin, 1981). Unlike classical logic which requires
a deep understanding of a system, exact equations, and precise numeric values, fuzzy logic
incorporates an alternative way of thinking, which allows modeling complex systems using a
higher level of abstraction originating from our knowledge and experience. Fuzzy systems,
including fuzzy logic and fuzzy set theory, provide a rich and meaningful addition to standard
logic(Zadeh,1963). The applications which may be generated from or adapted to fuzzy logic
are wide-ranging, and provide the opportunity for modeling of conditions which are
inherently imprecisely defined. TradeExpert with fuzzy reasoning capabilities allows
expressing this knowledge through subjective concepts such as very expensive and a significant
below the market average which are mapped into exact numeric ranges. Since knowledge can be
expressed more naturally by using fuzzy sets, many decision making problems can be greatly
simplified. Fuzzy logic provides an inference morphology that enables approximate human
reasoning capabilities to be applied to knowledge-based systems. The theory of fuzzy logic
provides mathematical strength to capture the uncertainties associated with human cognitive
processes, such as thinking and reasoning. The conventional approaches to knowledge
representation lack the means for representing the meaning of fuzzy concepts. As a
consequence, approaches based on first order logic do not provide an appropriate conceptual
framework for dealing with the representation of commonsense knowledge, since such
knowledge is by its nature both lexically imprecise and non categorical. The development of
fuzzy logic was motivated in large measure by the need for a conceptual framework which can
address the issue of lexical imprecision. Decision support systems have been the most obvious
recipients of the benefits of fuzzy logic, since their domain is often inherently fuzzy.
To summarize, in fuzzy logic:
 exact reasoning is viewed as a limiting case of approximate reasoning.
 everything is a matter of degree.
 knowledge is interpreted a collection of elastic or, equivalently, fuzzy constraint on
a collection of variables.
 Inference is viewed as a process of propagation of elastic constraints.
 Any logical system can be fuzzified.
7.1 Fuzzy logic and degree of truth

English abounds with vague and imprecise concepts, such as "The p/e(ratio of price and
earning per share) of this stock is low" or "The outstanding share of this company is small."
Such statements are difficult to translate into more precise language without losing some of
their semantic value: for example, the statement "p/e is 6.9." does not explicitly state that the
stock is traded at a cheap price. It is fraught with difficulties. Let us assume the average p/e
of all public traded stocks is 20. The p/e of 5 is considered low and the p/e of 40 is high.
How would a stock with the p/e of 16 be ranked? What about a stock with the p/e of 25?
The use of fuzzy logic allows for more gradual changes between categories and allows for a
representation of certainty in the rule consequence through the ability to fire rules with
varying strength dependent on the antecedents. One of the major tasks in the design of
TradeExpert is to codify the investor’s decision-making process. Exact reasoning strategies
that make use of standard probability theory. A common approach is the certainty
factor(CF) in rule-based systems. CF has a value between -1 and +1, representing 100% false
and 100% true respectively. Degrees of truth are often confused with probabilities.
However, they are conceptually distinct; fuzzy truth represents membership in vaguely
defined sets, not the likelihood of some event or condition. Approximate reasoning is
needed when the assumptions necessary to apply a probability based approach cannot be
met. Fuzzy reasoning uses a collection of fuzzy membership functions and rules (instead of
Boolean logic) to reason about data.
There is an important distinction between fuzzy logic and probability. Both operate over the
same numeric range, and at first glance both have similar values: 0.0 representing False (or
non-membership), and 1.0 representing True (or membership). However, there is a
distinction to be made between the two statements: The probabilistic approach yields the
natural-language statement, "There is an 80% chance that the stock looks attractive" while
the fuzzy terminology corresponds to "The stock's degree of membership within the set of
attractiveness is 0.80." The semantic difference is significant: the first view supposes that the
stock is attractive or not; it is just that we only have an 80% chance of knowing it. By
contrast, fuzzy terminology supposes that the stock is "more or less" attractive, or some
other term corresponding to the value of 0.80. Further distinctions arising out of the
operations will be noted below.
For independent events, the probabilistic operation for AND is multiplication, which (it can
be argued) is counterintuitive for fuzzy systems. For example, let us presume that x is a
company, S is the fuzzy set of high_p/e companies, and T is the fuzzy set of investor-
preferred companies. Then, if S(x) = 0.90 and T(x) = 0.90, the probabilistic result would be:
S(x) * T(x) = 0.81

whereas the fuzzy result would be:
MIN{S(x), T(x)} = 0.90

The probabilistic calculation yields a result that is lower than either of the two initial values,
which when viewed as "the chance of knowing" makes good sense.
However, in fuzzy terms the two membership functions would read something like "x is a
high-growth company" and "x is an investor-preferred company." If we presume for the
sake of argument that "very" is a stronger term than "fairly," and that we would correlate
"fairly" with the value 0.81, then the semantic difference becomes obvious. The probabilistic
calculation would yield the statement
If x is a high-growth company and x is an investor-preferred company, then x is a fairly high-
growth, investor-preferred company.
The fuzzy calculation, however, would yield
If x is a high-growth company and x is an investor-preferred company, then x is a very high-growth,
investor-preferred company.
Another problem arises as we incorporate more factors into our equations (such as the fuzzy
set of actively-traded companies, etc.). We find that the ultimate result of a series of AND's
approaches 0.0, even if all factors are initially high. Fuzzy theorists argue that this is wrong:
that five factors of the value 0.90 (let us say, "very") AND'ed together, should yield a value
of 0.90 (again, "very"), not 0.59 (perhaps equivalent to "somewhat").
Similarly, the probabilistic version of A OR B is (A+B - A*B), which approaches 1.0 as
additional factors are considered. Fuzzy theorists argue that a sting of low membership
grades should not produce a high membership grade. Instead, the limit of the resultant
membership grade should be the strongest membership value in the collection.
Another important feature of fuzzy systems is the ability to define "hedges," or modifier of
fuzzy values(Radecki, 1982). These operations are provided in an effort to maintain close
ties to natural language, and to allow for the generation of fuzzy statements through
mathematical calculations. As such, the initial definition of hedges and operations upon
them will be a subjective process and may vary from one statement to another. Nonetheless,
the system ultimately derived operates with the same formality as classic logic.
For example, let us assume x is a company. To transform the statement “x is an expensive
company in terms of its p/e” to the statement “x is a very expensive company in terms of its
p/e. The hedge “very” can be defined as follows:
"very"A(x) = A(x)^2
Thus, if Expensive(x) = 0.8, then Very_Expensive(x) = 0.64. Similarly, the word “more or
less” can be defined as Sqrt(Expensive(x)). Other common hedges such as "somewhat,"
"rather," and "sort of," can be done in a similar way. Again, their definition is entirely
subjective, but their operation is consistent: they serve to transform membership/truth
values in a systematic manner according to standard mathematical functions. From the
above discussion, it is clear that fuzzy logic can describe the investor’s decision making
process in a more natural and accurate way than the probability theory.
7.2 Fuzzy linguistic variables and membership

Fuzzy logic allows for set membership values to range (inclusively) between 0 and 1, and
anything in between representing linguistic and imprecise terms like "slightly", "quite" and
"very". Specifically, it allows partial membership in a set. It is related to fuzzy sets and
possibility theory. Fuzzy logic is a form of multi-valued logic derived from
0 < < 10
0 < < 10 10 < < 20
PEverylow (x) = PElow (x) =
0 0 ≥ 10 0 ≥ 20
0 ≤ 10 0 ≤ 20
10 < ≤ 20 20 < ≤ 30
PEavg (x) =
20 < < 30 PEhigh(x) = 30 < < 40
0 ≥ 30 0 ≥ 40
0 ≤ 30
PEveryhign(x) = 30 < < 40
1 ≥ 40
The memberships of the above formulas can be shown below:
v low low avg high v high
Fuzzy value
0.5
0 5 10 15 20 25 30 35 40 45 50 60 70
Input variable
Fuzzy logic is a form of mathematics that let computers deal with shades of gray. It handles
the concept of partial truth – truth values between completely true and completely false. As
an example, consider a p/e of 17, it yields the following fuzzy values:
PElow(17) = 0.3 and PEavg(17) = 0.7

A p/e of 17 is considered 30% in the low range and 70% in the middle range. It shows a
linguistic term can simultaneously belong to more than one category while exhibiting
different degrees. By converting linguistic terms into numerical values, it permits
TradeExpert to conduct reasoning and inference quantitatively, and to allow conflicting
rules to be fired jointly.
Fuzzy set theory provides a host of attractive aggregation connectives for integrating
membership values representing uncertain information. These connectives can be
categorized into the following three classes union, intersection and complement connectives.
Union produces a high output whenever any one of the input values representing degrees
of satisfaction of different features or criteria is high. Intersection connectives produce a
high output only when all of the inputs have high values. Complement connectives are
defined as 1 minus the fuzzy value of its argument. More formally, these three basic fuzzy
operators are defined below:
Not F(x) = (1 – F(x))
F(x) And F(y) = minimum(F(x), F(y))

F(x) Or F(y) = maximum(F(x), F(y))

Some of these operators provide compensation that has the property in which a higher
degree of satisfaction of one criterion can compensate for a lower degree of satisfaction of
another criterion to a certain extent. In that sense, union connectives provide full
compensation and intersection connectives provide no compensation. In a decision process
the idea of trade-offs corresponds to viewing the global evaluation of an action as lying
between the worst and the best local ratings. This occurs in the presence of conflicting goals
when compensation between the corresponding compatibilities is allowed. More operators
known as Contain, Empty and Equal are listed below:
F(x) is Empty if and only if for all x, A(x) = 0.0.
Fa(x) Equal Fb(x) if and only if for all x: Fa(x) = Fb(x)
Fa(x) is Contained in Fb(x) if and only if for all x: Fa(x) <= Fb(x).
With these fuzzy operators, a decision support system can provide structured ways of
handling uncertainties, ambiguities, and contradictions.
7.3 Fuzzy rule reasoning and deduction

TradeExpert is a rule-based decision support system that uses a collection of fuzzy sets and
rules for the reasoning of data. Fuzzy logic in TradeExpert allows conclusion(s) to be
reached from premise(s) with a gradation of truth. The rule's premises describe to what
degree the rule applies while the conclusion assigns a membership function to each of one
or more output variables. Compared to standard rule-bases systems, it allows multiple rules
to be fired jointly as long as they exceed the predetermined threshold. When TradeExpert is
used to solve real problems, the following steps are generally followed:
First, one needs to provide investment parameters, such as objectives, assumption,
investment horizon and etc in a linguistic or mathematical form.
 Under FUZZIFICATION, Convert numeric data to literate words using fuzzy
membership functions, and determine the degree of truth for the word. It calculates
the degree to which the input data match the condition of the fuzzy rules.
 Under INFERENCE, the truth value for the condition of each rule is computed
using AND, NOT or OR, and applied to the conclusion part of each rule. The result
is one fuzzy subset to be assigned to the output variable for each rule. The output
of each rule is scaled by the rule condition’s computed degree of truth.
 Under COMPOSITION, all of the fuzzy subsets assigned to the output variable are
combined together to form a single fuzzy. The operation SUM takes the point wise
sum over all of the fuzzy subsets.
 DEFUZZIFICATION: convert the fuzzy output set to a numeric value. TradeExpert
uses the MAXIMUM method. It selects the maximum value of the fuzzy sets as the
crisp value for the output variable.
Unlike most rule-based systems, threshold values define the minimum required
membership of the premises an investor would expect for that particular rule to be fired.
The minimum requirements are generally defined by subjective criteria. With fuzzy
functions, it is possible that several rules may be fired concurrently with different certainty
factors.
The following diagram shows how the fuzzy reasoning component works in TradeExpert:
Input data
Fuzzification
Case base(fuzzy)
Fuzzy inference
Fuzzy membership and fuzzy match
Rule Base(fuzzy)
Evaluate rules
and cases
difuzzification
Crisp output
To illustrate this process, let us consider the following simplified rules stored in the
fundamental knowledge base and the fuzzy functions PE(x) as described before and Yield(x)
which is defined below:
Rule 1: IF p/e is low AND yield is high THEN strong buy

Rule 2: IF p/e is high AND yield is low THEN strong sell
Rule 3: IF p/e is average AND yield is high THEN buy
Rule 4: IF p/e is high AND yield is average THEN hold
.
0 < < 1.5
Yieldlow(x) .
0 0 ≥ 1.5
0 < < 1.5

.
Yieldavg(x) = 1.5 < <3

.
0 ≥ 3
0 ≤ 1.5
.
Yieldhigh(x) = . 1.5 < < 3
1 ≥ 3
Given a stock with a p/e of 13 and dividend yield of 2.7%, the following membership
functions return non-zero values in the process of fuzzification:
Yieldavg(2.7) = 0.2
Yieldhigh(2.7) = 0.8
P/E low(13) = 0.7
P/E avg(13) = 0.3

TradeExpert believes that the yield of the given stock is considered 20% in the average
range and 80% in the high range compared to all publicly traded stocks. Its P/E is ranked
70% in the low range and 30% in the average range. By matching these words to the rules
listed above and applying the operator AND, the non-zero truth value of each rule is listed
below: Rule 1 has the truth value of 0.7 and Rule 3 has a truth value of 0.3. In other words,
TradeExpert would issue a Strong Buy recommendation with a 70% certainty based on the
experts’ opinions and expertise stored in the fundamental knowledge base. Rule 3 was fired
too but its conclusion had not been accepted due to a predetermined minimal threshold of
55%.
8. Conclusion
The development of TradeExpert demonstrates the effectiveness and value of using analogy
in an intelligent decision support system, particularly in situations of imprecision,
dynamics, or lack of perfectly-matched knowledge. TradeExpert shows its ability to
manipulate two kinds of knowledge over time: episodic information and evolving
experience. The system becomes “wiser” as more cases are added to the case base. With a
research area as new as case-based reasoning in intelligent decision support systems, it often
raises more questions than answers. Future research includes but is not limited to the
following topics: how to dynamically adjust the weights associated with the features of
cases, how to automatically index and store cases to facilitate efficient retrievals and
searches, and how to modify the rules in knowledge bases with minimal intervention from
human experts. In conclusion, the work reported in this paper discussed several important
issues in the design of intelligent decision support systems, proposed and implemented
solutions to these problems, and demonstrated the usefulness and feasibility of these
solutions through the design of TradeExpert.
9. References
Baldwin, J.F. "Fuzzy logic and fuzzy reasoning," in Fuzzy Reasoning and Its Applications, E.H.
Mamdani and B.R. Gaines (eds.), London: Academic Press, 1981.
Carter, John. “Mastering the Trade”. McGraw_Hill. 2006.
Edwards, Robert and Magee, John, “Technical Analysis of Stock Trends”. John Magee Inc.
1992.
Eng, W. F. “The Day Trader’s Manual”. John Wiley & Sons, Inc. 1993. New York.
Fama, Eugene (1965). "The Behavior of Stock Market Prices". Journal of Business 38: 34–105.
Fox, Justin, “The Myth of the Rational Market”. E-book. ISBN: 9780061885778.
Liu, Shaofeng; Duffy, Alex; Whitfield, Robert; Boyle, Iain. Knowledge & Information Systems,
Mar2010, Vol. 22 Issue 3, p261-286, 26p.
Nocera, Joe , “Poking holes in a Theory on Markets”. June 5,2009. New York Times. Nocera.
http://www.nytimes.com/2009/06/06/business/06nocera.html?scp=1&sq=efficie
nt%20market&st=cse. Retrieved 8 June 2009.
Kaufman, P.J. “New Trading System and Methods”, John Wiley & Sons, Inc., 2005.
Kolodner, Janet. "Reconstructive Memory: A Computer Model," Cognitive Science 7 (1983): 4
Malkiel, Burton G.(1987). "Efficient market hypothesis," The New Palgrave: A Dictionary of
Economics, v. 2, pp. 120–23.
Covel, Michael, Trend Following: “How Great Traders Make Millions in Up or Down Markets,
Financial Times”, Prentice Hall Books, 2004
Radecki, T. "An evaluation of the fuzzy set theory approach to information retrieval," in R. Trappl,
N.V. Findler, and W. Horn, “Progress in Cybernetics and System Research”, Vol.
11: Proceedings of a Symposium Organized by the Austrian Society for Cybernetic
Studies, Hemisphere Publ. Co., NY: 1982.
Robertson, C. S. Geva, R. Wolff, "What types of events provide the strongest evidence that the stock
market is affected by company specific news", Proceedings of the fifth Australasian
conference on Data mining and analystics, Vol. 61, Sydney, Australia, 2006, pp.145-
153.
Şen, Ceyda Güngör; Baraçlı, Hayri; Şen, Selçuk; Başlıgil, Hüseyin. “An integrated decision
support system dealing with qualitative and quantitative objectives for enterprise software
selection”. Detail Only Available By: Expert Systems with Applications, Apr2009 Part
1, Vol. 36 Issue 3, p5272-5283.
Silveman, Barry G.” Survey of Expert Critiquing Systems: Practical and Theoretical Frontiers”.
Communications of the ACM, Apr92, Vol. 35 Issue 4, p106-127.
Simple,J. James, “Trend-following strategies in currency trading”. Quantitative Finance 3, 2003,
C75-C77.
Simpson, R. L. “A Computer Model of Case-based Reasoning in Problem Solving: An Investigation
In The Domain of Dispute Mediation”. Ph.D. Dissertation, Georgia Tech. 1985
Schulmeister, Stephan. “Technical Trading Systems and Stock Price Dynamics”. WIFO-Studie
mit Unterstützung des Jubiläumsfonds der Österreichischen Nationalbank, 2002
Schulmeister. Stephan. Review of Financial Economics 18 (2009) 190–201.
Tan; Hock-Hai Teo; Izak Benbasat. “Assessing Screening and Evaluation Decision Support
System: A Resource-Matching Approach”. Information Systems Research, Jun2010, Vol.
21 Issue 2, p305-326.
4
Business Intelligence – CDMB – Implementing

BI-CMDB to Lower Operation Cost Expenses
and Satisfy Increasing User Expectations
Vlad Nicolicin-Georgescu1,2, Vincent Bénatier1,
Rémi Lehn2 and Henri Briand2
2Polytechnic School of Nantes University
1SP2 Solutions
France
1. Introduction
Since the beginning of Decision Support Systems (DSS), benchmarks published by hardware
and software vendors show that DSS technologies are more efficient by offering better
performance for a stable cost and are aiming at lowering operations costs with resources
mutualisation functionalities. Nevertheless, as a paradox, whereas data quality and
company politics improves significantly, user dissatisfaction with poor performances
increases constantly (Pendse, 2007).
In reality, this surprising result points the difficulties in globally defining the efficiency of a DSS.
As a matter of fact, we see that whereas a “performance/cost” ratio is good for a technician, from
the user perspective the “real performance/ expected performance” ratio is bad. Therefore,
efficient may mean all or nothing, due to the characteristic of user perspective: a DSS may be very
efficient for a developers needs, and completely inefficient for production objectives.
In order to improve the global DSS efficiency, companies can act upon (i) cost and/or (ii)
expectations. They must be capable of well defining and exhaustively taking into account
the costs and expectations relative to their DSS.
Concerning the first aspect of cost, the Total Cost of Ownership (TCO) provides a good
assessment. It contains two principal aspects: hardware/software costs and cost of
operations. With hardware/software, companies are limited by market offers and
technological evolutions. The cost of operations, in addition to the potential gain by resource
mutualisation, strongly influences the organization costs in rapport with its pertinence.
Over this problematic, companies have implemented rationalization based on best practice
guidelines, such as the Information Technology Integration Library (ITIL). Nevertheless, an
internal study we have conducted at SP2 Solutions in 2007 showed that these guidelines are
little or not-known to the DSS experts within or outside company IT departments. Our
experience shows that this hasn’t changed almost at all since then.
Relating with the second aspect, we state that user expectations cannot be ignored when
evaluating the efficiency of a DSS. We estimate that the expected results must be based on
the Quality of Service (QoS) and not on the raw technical performances. This way we define
the DSS efficiency as a ratio between the DSS QoS and the DSS TCO, which is the
foundation of our propositions.
Having described the context in which we position ourselves, we have developed and we
offer several solutions to improve the DSS efficiency.
(1) First, we note that an improvement process cannot exist without a prior monitoring
process. We propose to adapt and apply best practices, already deployed for operational
systems, by implementing a system which permits the measuring of the DSS efficiency.
Following ITIL guidelines, this kind of system is known as the Configuration Management
Data Base (CMDB), and relates with ITIL specifications for: (i) service support and (ii) service
offering (Dumont, 2007). The CMDB integrates all the information required to manage an
information system. This information varies from incidents and known problems,
configuration and change management, to management of availability, capacity, continuity
and levels of service.
(2) Second, we show the necessity of taking into account DSS specifics (in contrast with
operational systems), with the purpose of elaborating the right measures for the efficiency of
a DSS. These measures are different from the ones used with operational systems, which
focus on the problematic of real time utilization. DSS are often defined in comparison with
operational information systems as the distinction between On-Line Analytical Processing
(OLAP) and On-Line Transaction Processing (OLTP) shows it (Inmon, 2005), (Inmon, 2010).
The two are part of fundamentally different worlds over aspects such as: the contained data
(raw/aggregated), data organization (data bases/data warehouses), orientation
(application/subject) or utilization (continuous/periods of utilization). DSS are classically
aimed at managers and rely on analytical data, which is derived from the primitive data
provided by operational systems.
Our major proposition presented here is the elaboration of the Business Intelligence CMDB
(BI-CMDB) for the supervision of a DSS. It serves as a “decisional of the decisional”
solution, and, to our knowledge and experience, it has never been approached so far with an
integrated proposition. The BI-CMDB follows the principles of ITIL’s CMDB and contains
all the elements required for managing the DSS: architecture, configuration, performances,
service levels, known problems, best practices etc. It integrates information both structured
(e.g. configuration) and non-structured (e.g. best practices). With the help of the BI-CMDB,
SLA/Os (Service Level Agreements / Objectives) are taken into account when computing
performance, thus focusing on improving user satisfaction over technical performance.
Following this, a second contribution is to optimize the value offered by the CMDB. We
combine BI, OLAP and semantic web technologies in order to improve DSS Management
efficiency, while offering autonomic self management capabilities with references from
(Nicolicin-Georgescu et al., 2009).
The chapter is organized as follows. Section 2 presents how monitoring of the DSS is done
so far and what elements are followed in concordance with the ITIL guidelines. In Section 3
we focus on the description of the DSS specifics in comparison with the operational systems,
and how performance measurement changes with these specifics. The subjective aspect of
user perceived performance is equally discussed. Further, we present our proposition, the
BI-CMDB, in Section 4. The A-CMDB and the integrated monitoring and analysis solutions
for DSS management are some of the described elements. We equally note that the
presented work has already been implemented and the solutions area available for use (SP2
Solutions, 2010). Section 5 consists of a series of possible shortcomings and limitations of the
BI-CMDB, so we can conclude in Section 6 with an overview of the chapter and the future
doors that BI-CMDB opens for DSS management.
Business Intelligence – CDMB – Implementing BI-CMDB
to Lower Operation Cost Expenses and Satisfy Increasing User Expectations 69
2. Monitoring Decision Support Systems

As with any type of system, prior to any analysis and assessment, monitoring of the
interesting elements is required. “You can’t manage what you don’t measure” is an old well
known management adage, which seems to apply more than ever nowadays.
2.1 Monitoring challenges and guidelines

There are two challenges with monitoring, particularly underlined by the DSS: (i) getting
some data and (ii) getting the right data.
The first aspect is obvious, and challenges here usually relate with technical barriers. For
example, consider an interactive reporting application, which builds reports based on data
from various data warehouses. If we are interested in the functioning of this application, we
need to trace its running, usually from application and operation system logs (files, data
bases etc.). The organization of these sources is usually very specific and requires
specialized Extraction Transform Load (ETL) software. Therefore, the process is purely
technical.
The second aspect of getting the interesting data proves to be much more subjective-
oriented, as a natural question arises: Who needs this data? Continuing with our example, if
the interesting party is a technician, he will most likely want to know if the available
resources are enough for the application to work properly. If the interesting party is a
business department responsible, he is more likely interested in knowing how often the
application is used and if it offers any added value to the business process. Choosing the
right data based on the specific needs is either done directly when monitoring is performed
(i.e. gather only the interesting data), or is filtered afterwards from a pool of potentially
interesting data.
Getting the right data to the right party is strongly related with the elaborated SLAs. In the
context of information systems, an SLA is defined as a compact between a customer and a
provider of an IT service that specifies the levels of availability, serviceability, performance,
reporting, security or other attributes of a service, often established via negotiation from the
two parts. It basically identifies the client’s needs. IT organizations are measured on end-to-
end business system response times, the degree of efficiency with which the system is used,
and their ability to adapt to dynamic workloads (Ganek & Corbi, 2003). Breaking the SLAs
may lead to financial and even legal problems, hence the need of their integration with the
monitoring processes.
A very clear and detailed example of this aspect is shown by (Agrwala et al., 2006). The
authors explain that Quality of Service is a particular concern for enterprise monitoring,
exemplifying with an online airplane booking system. Formally, the QoS is defined as the
level of current obtained data (usually performance) in rapport with what is expected. They
describe QMON a QoS-capable monitoring system, which acts upon the system according to
the importance of the monitored elements, importance expressed through the SLAs. The
provided conclusion underlines the fact that QoS should be dynamically configurable and
supported at monitoring level, allowing balance between monitoring overheads and
improvement in application utility.
With DSSs, integration of SLAs and elaboration of QoS metrics is a central focus point.
Unfortunately, best practices and guidelines to do so are not that elaborated as they are for
operational system. One of the common mistakes done when managing a DSS is to use
existing practices for operational systems and adopt them for DSSs without taking into
account their particularities (Inmon, 2005). Nevertheless, ITIL provides a good foundation
with the CMDB over service sustainability and offering (Dumont, 2007).
Service sustainability corresponds to the interface between the service user and the system
providing the service. The CMDB integrates monitored data from: incidents, known
problems and errors, configuration elements, changing and new versions. Each of the five
data types corresponds to a management module (e.g. incident management, configuration
management etc.). In turn, each of the management modules is retaken by the user to ensure
service maintenance.
Service offering formalizes the interface between the clients and the system. This time, it
refers to offering new services in rapport with the defined policies. Starting from
management and infrastructure software (like the CMDB), the service levels are assessed
based on SLA/Os. Elaboration of these agreements and objectives include availability
management, capacity management, financial service management, service continuity,
management of the levels of service and the assurance of a high QoS.
2.2 Monitoring solutions with DSS software

The described ITIL CMDB is general for an IS. As DSSs lack proper monitoring, elaborating
a CMDB for a DSS is the first step towards a more efficient DSS. At this point a discussion is
required over how monitoring is achieved with various DSS software solutions. The three
most common ways are: (i) logging files, (ii) logging databases and (iii) console specific
commands.
Logging files are the most used source of monitoring information. They contain all messages
regarding the functioning of the DSS. For example, IBM’s Cognos 8 Financial Performance
Management (IBM, 2011) solutions offer a list of various log files and tracking levels by
purpose. These include: the transfer log file for recording activity from the installation
wizard with the transferred files; the startup configuration file for logging configuration
choices; the run-time log for tracking all events from the server while at run time. The
information from the logging files is normally semi-structured, and in order to be useful,
additional parsing and analysis must be performed.
Logging databases are preferred over log files as they provide a better information structure,
but are equally harder to implement. A good example of this is the audit data base (the
Audit_event table) from the Business Objects Enterprise solution (Business Objects, 2008).
The table contains information about run time events, such as the event id, the name of the
user, the timestamp etc. Logging of these events can be activated or deactivated by the user,
as it slightly affects performance. This also comes from the fact that the audit table is
constructed by accessing a list of services provided by the BO server.
Console specific commands are especially used when recovering data concerning configuration
parameters, which are isolated within logs, rendering them harder to identify. Sometimes,
they are even impossible to recover, if the respective log that contains the last parameter
modification has been archived and is no longer part of the current logs. To exemplify, the
Oracle Hyperion Essbase solution employs a specific command line utility console, the
“essmsh”, to retrieve information such as the file sizes of the data marts or the cache
memory configuration parameters (Oracle, 2010). Even if this solution is the hardest to use,
from the monitoring point of view, it proves useful in situations such as above.
Either of the data sources used, the monitored data is usually raw basic data. Nevertheless, a
distinction is done with this data at a first level of analysis which results in a combined
monitored data. Some of the solutions, such as the BO monitoring data bases, offer a list of
combined data parameters. To provide with a very simple example, consider that a data
mart with two size indicators: the index file size and the size of the data file. We can obtain
the two separately, but we have no indicator of the total size of the data mart. This is
obtained by adding the two and ‘creating’ a new combined monitored indicator.
3. Intelligent performance with DSS specifics

Having the monitored data, the next natural phase is to use it in order to measure the
performance levels of the DSS, consequently allowing the assessment of its efficiency. There
are several particularities that render the process more delicate than for operational systems.
3.1 DSS specifics

We make a presentation of the most notable DSS particularities, as taken from (Inmon,
2005).
The data warehouse is the heart of a DSS, and represents the architecture used for storing the
analytical data. The core of data warehouses is the data mart, which is a repository of data
gathered with the help of ETL software from operational data and other sources, with the
purpose of serving a particular community of knowledge workers (SAP, 2010).
Metaphorically speaking, data marts are the bricks that are used to construct the data
warehouse building. Inmon identifies four main considerations for making the separation
between the analytical and operational data: (i) analysis data is physically different from
operational data, (ii) the technology used for operational systems is fundamentally different
from the one used with DSS, (iii) the processing characteristics for the two are different and
(iv) the analytical and operational communities are two different worlds.
DSS are subject oriented, with the goal towards business objectives, whereas operational
systems are application specific focusing on application requirements. That is why
measuring the performance of a DSS is harder as it is based on user expectations.
The resource utilization patterns of DSS are completely different from the ones of operational
systems. There is a notion of time relaxation, as data warehouses are not used constantly
with the same load. There are periods of maximum load, which are also known as critical
periods and periods of small load or inactivity. For example, a budget planning application
will only be used the first week at the beginning of each month, thus requiring the
maximum of available resources during this period only. The other three weeks, it can
function with the minimum needed resources. We identify two major criteria for the
distinction of the utilization patterns: (i) the utilization periods and (iii) the utilization purpose.
The utilization periods reflect the critical periods and are usually specified with the SLAs. The
performance objectives per utilization period and per user type, alongside the specification
of the critical functioning periods, are regular elements of the agreement documents.
The utilization purpose formalizes the two main data cycles of the data warehouse: (i) the
usage of the data for the generation of reports, and, (ii) the recalculation of the data for data
warehouse update. They are also known as the day (production) and the night (batch)
usage. During day, users access the various data from the data warehouses in order to build
analysis reports, which help them with the process of decision making. The data
warehouses are exposed to intense data retrieval operations. Even if the response times are
very small (e.g. 80% of OLAP queries must be under a second (Codd et al., 1993)) the user
activity is very high (even thousands of users). During night, batch operations are executed
to integrate the new operational data (from the day that has just passed) into the data
warehouse. The executed calculation and restructuration operations are very resource and
time consuming. Moreover, the new data must be integrated before the start of the next day
in order for the users to have the correct reports.
One last DSS characteristic we want to discuss here is the motto of the decision expert: “Give
me what I say I want, so I can tell you what I really want” (Inmon et al., 1999). This expresses the
fact that data warehouse users access a large variety of data from the data warehouse,
always being in search of explanations over the observed reports. Therefore anticipation and
prediction of the requested data, data usage or system charge is very hard to implement,
leading to (not necessarily a good thing) more relaxed performance requirements.
3.2 Performance
Performance measurement with DSSs brings into discussion, along the technical aspect
(from operational systems), the user satisfaction aspect (form the business objective oriented
point of view of analytical systems). There are two main types of performance (Agarwala et
al., 2006):
Raw technical performance – related to the technical indicators for measuring system
performance (such as response times for recovering data from the data warehouse, report
generation, data calculation etc.)
Objective performances – relating to user expectations. These measures are based on
predefined policies and rules that indicate weather or not satisfactory levels are achieved by
the system (e.g. response time relative to the importance of the data warehouse or with the
user activity).
Retaking the example shown by the authors of (Agarwala et al., 2006), a detailing of the QoS
formula is shown below.
MonitoredValue
QoS  (1)
ObjectiveValue
and
n
 QoSi * ScalingFactori
1
QoSglobal  n
(2)
 ScalingFactori
1
The basic single parameter QoS is computed by the ratio between the monitored values and
the specified expected values. If several parameters are responsible for a global QoS, then a
scaling factor is multiplied with each individual QoS, and, in the end, an average of all the
scaled QoS is computed to obtain the overall system QoS value.
The data warehouse quality model is goal oriented and roots from the goal-question-
metric model. Whether metrics satisfy or not the goal determinates the quality of the data
warehouse (Vassiliadis et al., 1999). Data warehouses must provide high QoS, translated
into features such as coherency, accessibility and performance, by building a quality
meta-model and establishing goal metrics through quality questions. The main advantage
of a quality driven model is the increase in the service levels offered to the user and
implicitly increase in the user satisfaction. QoS specifications are usually described by the
SLAs and the SLOs.
We have seen what SLAs stand for. In a similar manner, SLOs describe service objectives,
but on a more specific levels. For example, a SLA would describe that an application is
critical on the first week of each month, whereas the SLO would state that during critical
periods the query response time on the data marts used by the application should not be
greater than 1s.
With DSSs and data warehouses, one important area where service levels are crucial is the
utilization periods, in direct link with the answer to the question: what is used, by whom and
when? Data warehouse high charge periods are specified with SLAs. When building a new
DSS and putting in place the data warehouse architecture, the utilization periods are
specified for the various applications. To paraphrase a study from Oracle, the key to enterprise
performance management is to identify an organization’s value drivers, focus on them, and align the
organization to drive results (Oracle, 2008).
4. The BI-CMDB
In this section we present, in an exhaustively manner, our view and proposition for building
a CMDB adapted to decisional systems, which we consequently call the BI-CMDB. In order
to understand how this is built, we present a schema of the data flow and the organization
of five modules that are part of our approach.
Fig. 1. The SP2 BI-CMDB simplified schema

On the bottom level we have the data sources that provide the monitored data. As
mentioned, these are distributed, non-centralized and under different formats. Next, there is
a data collector module, which gathers and archives the data sources. This is the first level of
centralization while greatly reducing the size of this data by archiving. Following, an
Operational CMDB is build with the data collector centralized information. It is the first level
of filtering (parsing and interpretation) of the monitored data, which allows to keep only the
interesting data for further analysis.
Analysis of the O-CMDB is brought climbing up to the Analytical CMDB, which builds a
data warehouse architecture, by specific objective data marts, with the data from the O-
CMDB. We note that the elements of the O-CMDB and of the A-CMDB correspond to ITIL’s
guidelines, with the addition of elements that correspond to the specifics of DSS. Once the
A-CMDB has been constructed, two more levels of information access (by user) are added,
with a module for publishing the reports built on the A-CMDB data warehouse and a Web
Portal for consulting these reports and for dynamically exploring the data from the CMDBs.
These last two are not of interest for this chapter, but equally make a challenge, from the
study of computer-human interaction and GUI design.
A detailed description of the first three modules that are interesting to the BI-CMDB is done
in the following. Examples for each of them are provided, in the light of the differences with
the traditional CMDB and on the characteristics of DSS.
4.1 The data collector

The data collector is the first module in the data flow towards the construction of the BI
CMDB. Its main objective is to gather and archive the various interesting monitoring
information.
The core of the data collector architecture is the connector. A connector represents a generic
interface for defining the elements that allow the gathering of information from a data
source and the methods used for this process. The parameters of a connector allow him to
connect to a target machine (i.e. name of the machine, connection port, data source location
and type etc.), recover and archive the data from the sources and send these archives via ftp
connections to the machine which is in charge of archive analysis and construction of the O-
CMDB. Depending of the target machine installed BI software solution, various
implementations of the generic connector are required (e.g. an Oracle Hyperion Essabse
connector or a BO connector).
As we have seen previously in the monitoring section, there are three types of data sources:
logging files, logging databases and specific console commands. A connector can specify
any (or all) of these types.
Following the ITIL guidelines, a separation of the data fluxes is done for each connector
(where possible). There are five types of data fluxes: (i) logging (log), (ii) exception (xcp), (iii)
parameters (param), (iv) performance (perf) and (v) errors (err). This separation persists
further on with the construction of both the O-CDMB and the A-CMDB. Another argument
for this split is the fact that some solutions use different data sources for different fluxes. For
instance, with Essbase, the monitored information is usually stored in plain text files (.log
and .xcp). These files allow the identification of all the required data with the exception of
the configuration parameters. Due to the fact that they change rarely, their identification
through log files is very difficult or even impossible. This is why for this task, specific
console commands are used for the param flux.
Example. We show below an example of data collection for a connector implementation for
the Oracle Hyperion Essbase server, with the various connector data fluxes.
The Essbase server log files are under the /log repository of the Essbase installation. The
connector will specify the machine and port (for connection to the Essbase server) and the
place of the Essbase installation (e.g. C:/Hyperion/logs/essbase). This repository contains a
suite of .log and .xcp files, that are organized by application (here the term application
indicates the regrouping several data marts which suit the same business objective)
The .log file contains various information about the application and its data marts, which is
interesting for the log, perf and err data fluxes. An extract of the file is shown below for a
data retrieval operation on a data mart from the application:
[Mon Jan 31 17:57:51 2011]Local/budgetApplication///Info(1013210)
User [admin] set active on database [budDM]
[Mon Jan 31 17:57:51 2011]Local/budgetApplication/budDM/admin/Info(1013164)
Received Command [Report] from user [admin]
[Mon Jan 31 17:58:21 2011]Local/budgetApplication/budDM/admin/Info(1001065)
Regular Extractor Elapsed Time : [29.64] seconds
These lines permit to identify that there was an Essbase report request (data retrieval) on the
budDM data mart, part of the budget application on the on 31st of January at 17:57:51 from
the admin user, which lasted 29.64 seconds. These elements represent the interesting
monitored information, and are recovered from the last four lines. The first two lines are
useless. This filtering process is done at the construction of the O-CMDB.
The .xcp file contains the information needed for the xcp data flux, logging the fatal crashes
of the Essbase server. An extract of this file could be:
Current Date & Time: Fri Sep 19 11:29:01 2008
Application Name: budgetApplication
Exception Log File: /app/hyp/Hyperion/AnalyticServices/app/budgetApplication/log00001.xcp
Current Thread Id: 2
Signal Number: 0x15=Termination By Kill
Last, in order to get the configuration parameters of the application data marts, the esscmd
command line console is used (i.e. GETDBINFO, GETAPPINFO etc.). The results of these
commands are written into customized text files, which assure the perf data flux.
The role of the example was to show that the monitored information is very different, and
data sources and formats vary even within the same implementation. The data collector has
the role of assuring a first centralized level of information, with no data analysis intelligence.
The only intelligence in the module allows an incremental data collection, such that data is
not redundant (e.g. if data is monitored once each day, each collection takes only the
information from the logs corresponding to the last past day). The following levels of
information centralization are shown next with the construction of the O-CMDB and the A-
CMDB.
4.2 The O-CMDB

The Operational CMDB is the first intelligent unification of the DSS information. It suits the
role of an Operational Data Store (ODS), and it is build from the data collected by the data
collector, by a series of filtering and analysis operations. The O-CMDB contains only the
interesting information, rejecting any other unrequited data from the collector archives (as
exemplified earlier with the Essbase report generation).
At first glance, the O-CMDB may seem as a simple log analyzer. This cannot be further from
the truth. In reality, the process of building the O-CMDB makes a sweep of the information
available, holding only the useful one (through interpretation of the log data), and
constructing a series of purpose objective data stores with this filtered data. The
organization of the data stores retakes the main ITIL groups for the CMDB and continues in
the line of the data collector connector data fluxes. Technically, our implementation of these
data stores is via relational databases, for both the O and the A CMDB, term which we will
further use for a better understanding.
The following O-CMDB databases are constructed.
The CI (Configuration Items) tables, holding information about the architectural organization
of the target decision system. These tables are different by construction, as they are not built
with the data recovered from the data collector, but they have as source specific defined files
or databases which we use with our system to let the user ‘complete’ his DSS architecture. In
specific cases some of the CIs can be automatically built from the data from the logs. This
operation of DSS entity specification must be performed before the construction of the O-
CMDB and is compulsory, as the rest of the tables are based on the DSS entities.
In our conception, we identify six levels of vertical hierarchy: (i) the entire DSS, (ii) the
physical servers, (iii) the logical servers, (iv) the applications, (v) the bases and (vi) the
programs. In order to keep this as generic as possible we note that depending on the specific
BI software implementation, some of these levels disappear or are remapped (e.g. an
application for Essbase is well defined with its own parameters while for a BO solution it
becomes a report folder as a logical organization of BO reports).
The PARAM tables contain information about the configuration indicators. This is raw data,
organized by indicator type and by CI type. A reference measure table (configuration and
performance), lists the interesting monitored indicators. The data collector archives are
analyzed for patterns that allow the identification of these indicators, and then extracted and
written in the O-CMDB PARAM table. An example is shown in Table 1. We have mentioned
that the data contained here is raw data, meaning that there is no treatment done on it. This
is what we have earlier seen as basic monitored structured information. However an
exception from this rule is done with combined monitored configuration indicators (i.e.
deduced from existing indicators). The given example was on the size of an Essbase data
mart. There are two basic configuration indicators (data file size and index size) which are
extracted directly from the data collector archives and stored in the O-CMDB. Yet, an
interesting measure is the total size of the data mart, which expresses as the sum of the two
indicators. In this case, when constructing the O-CMDB a second passage is made to
compute these combined indicators.
Data collector archive O-CMDB CONFIG table extract

Field name Field Type Data value
CI_CODE INT 2
[Tue May 11 17:57:51 2010]
MEASURE_CODE INT 10
Local/budgetApplication/budDM
MEASURE_VALUE VARCHAR 100
Index Cache Size : [100] Mo
MEASURE_TYPE VARCHAR Mo
DATE_CHANGE DATE 11/05/2010
Table 1. O-CMDG PARAM table extract
The PERF tables are similar to the PARAM tables with the difference that the contained
indicators express performance measures. The distinction between the two is made for the
further analytical purposes, and in concordance with ITIL guidelines. Still, a series of
conceptual differences are identified, mainly related to the timeline intervals. Configuration
indicators are recorded in relation with the date at which they last change, operations that
are rare. On the other hand, performance indicators have a high apparition frequency, and
depend on several factors (from the network lag to the machine processor charge). On busy
periods, we have met up to several of tens of thousands performance entries in the log files,
which is a totally different scale from the configuration case.
In addition to these three ‘main’ tables, a series of additional tables are build, such as the
XCP tables with data concerning fatal crashes, the ERR tables with data from (non-lethal)
warnings and errors or the MSG tables with the various functioning messages.
One of the biggest benefits of constructing the O-CMDB is the reduction of the size of the
monitored information. By comparing the size of the initial raw monitored data sources
with the size of the O-CMDB database after integration of the data from those sources, we
have obtained a gain of 1 to 99 (i.e. for each 99Mb of raw data we have an equivalent of 1Mb
of integrated interesting data). In addition, this reduction allows keeping track of the data
for both short and medium time periods (even long in some cases), which permits further
prediction and tendency assessment.
4.3 The A-CDMB

The Analytical CMDB is the third and most advanced level of data integration. Unlike the
O-CMDB which contained only raw data, the information in the A-CMDB is derived from
this data. It represents the data warehouse of the O-CMDB, with analytical data suitable to
business objectives. Considering the fact that the data from the O-CMDB regards a DSS
implementation, we call the A-CMDB as the data warehouse of the DSS, or the support for
‘the decisional of the decisional’.
The A-CMDB consists of tables (i.e. AGG_PERF, AGG_PARAM) which aggregate the O-
CMDB tables on several axes of analysis. These axes are the fundamental characteristic of
the A-CMDB, and reflect the characteristics of a DSS while suiting the DSS user needs. The
main analysis axes are: (i) temporal intervals, (ii) utilization periods, (iii) utilization purpose,
(iv) user type, (v) SLA/SLO and, obviously, (vi) the CIs. Data is aggregated on each of these
axes, and their intersection provides the DSS user with any valuable information that he
requires. We detail each of these axes, as they represent key points of our proposition.
The CIs represent the first implicit analysis axis, as it provides the vertical aggregation on
the DSS hierarchy. Aggregation operations can either be averages, sums, standard
deviations or min/max (e.g. the total size occupied by the entire DSS, the sum of all sizes of
its data warehouses and the correspondent data marts, the maximum response time etc.).
Temporal intervals are the correspondent to the fundamental characteristic of the data
warehouse, the time aggregation of the operational data. In the same manner, when
building the O-CMDB we make a separation of six time intervals: year, month, weak, day,
hour, min:sec. Buy using these six pivots, the A-CMDB aggregates the averages over each of
them (e.g. the average response time for each week for a specific CI). The CI axis is the base
axis which is always found in the aggregation for any of the rest of the axes.
Utilization periods are integrated in the A-CMDB in concordance with the specified data
warehouse functioning patterns. By using the temporal intervals we are able to deduce if a
specific CI is in its utilization period. The integration of the utilization periods in the AGG_
tables will further be used for the assessment of the level of user satisfaction and the
computation of the Quality of Service, in rapport with the given performance objectives.
Usage purposes are similar to the utilization periods, and are formalized under the same
form. By using the hour time axis, the day and night intervals are defined (i.e. from 6:00 to
22:00 there is day, from 22:01 to 05:59 is night). Integrating this axis allows the user to
identify or not whether a certain performance analysis is pertinent (i.e. following the
restitution times during night periods makes little sense).
The user type axis represents the main focus axis when assessing user satisfaction. A
response time that is acceptable for the developer who is testing the data warehouse can be
completely unacceptable for a DSS user who wants his rapport as soon as possible. In the O-
CDMB, the tables have columns which specify the user who triggered the actions (e.g. the
user who asked for a report, the user that modified the configuration parameters etc.). When
integrated in the A-CMDB, users are organized logically by areas of interest, such as: (i) data
warehouse administrator, developer or user; (ii) area of responsibility (entire system,
department, application); (iii) type of responsibility (technical, business objective). By
intersecting these axes, the A-CMDB can rapidly respond at any questions concerning the
user activity, thus covering a total ‘surveillance’ of who is using what and when.
Lastly, the SLA/SLO axis integrates all the elements related to business / technical
objectives. For instance, defining the time intervals for the utilization and usage periods as
business objectives and specifying the performance thresholds for each of these periods as
technical objectives. By having these values we can afterwards compute the QoS indicators,
with regards to any intersection of the analysis axes, as the A-CMDB offers all the data
required to do so. The SLA and SLO data are not obtained from the O-CMDB and are not
part of the data recovered by the data collector. They are usually integrated when building
the A-CMDB, either by being manually specified or by using ETLs to load them from the
source where they are specified. Even it may seem a paradox, in many enterprises today
SLAs and SLOs are specified in semi-formatted documents (e.g. word documents) and their
integration is only possible manually. This represents a big drawback for automatic
integration, and by respecting the A-CMDB specifications, enterprises are ‘forced’ to render
these elements into formats that allow automatic manipulation (such as spreadsheets, .xml
or even data bases). Even if this alignment operation poses an effort in start, the return over
investment makes its implementation worth.
We provide an example with an extract of some of the columns from the AGG_PERF table,
which aggregates the O-CMDB performance tables.
Field name CI_CODE YEAR WEEK MEAS_CODE

(Field type) (INT) (INT) (INT) (INT)
Data value 5 2010 23 15
SLO_OBJECTIVE
Field name USER USAGE QOS
(FLOAT)
(Field type) (VARCHAR) (VARCHAR) (FLOAT)
(user specified)
Data value admin Day 3 0.56
Field name AGG_VALUE AGG_TYPE UTILIZATION
(Field type) (FLOAT) (VARCHAR) (VARCHAR)
Data value 5.3 avg high
Table 2. AGG_PERF tables extract
The table shows the fact that for the CI with code 5 (e.g. is an Essbase logical server), during
the 23rd week of 2010, the measure response time (code = 15), has an average of 5.3 seconds.
Moreover, this average is computed for the admin user, during a high utilization periods
and during day usage. Finally, there is an SLO objective of 3 seconds for the admin user
under the same conditions, which leads to a quality of service of 0.56 (=3 / 5.3), meaning
that the admin user is only 56% satisfied with the logical server response times. If we
consider that under a level of satisfaction of 80%, more details are required, automatically
we have the need to drill into the data in order to know exactly what are the applications
and further the bases and the programs that generate this dissatisfaction. This functioning is
characteristic for any data warehouse (from the DSS user motto).
Therefore, the A-CMDB contains all the information needed for the analysis of the
functioning of the DSS. It is a question of vision and development of the ITIL’s CMDB as a
decisional component. Moreover, it is about applying the specific DSS characteristics for its
construction, such that the relevant measures and indicators are well defined and that DSS
supervisors have a very clear view over the performance of their system from the
perspective of user and their satisfaction levels when using the DSS.
4.4 Advanced practices for the evolution of the BI-CMDB

Having seen how the BI-CMDB is built by processes from data collection, to construction of
the operational CMDB and finally of the analytical CMDB, a series of advanced practices
and topics are presented next, most of which open doors to future evolution of the model.
The benefice of the BI-CMDB is that it allows a complete (on every level) view of the
functioning of an enterprise DSS.
One first very good application of this view is prediction and prevention. Using prediction
algorithms and analyzing the data from the A-CMDB, we can see the potential bad
tendencies of a DSS. For instance a slow but constant higher memory usage can lead to a
system crash when the RAM memory is no longer sufficient. Moreover, by looking at past
evolutions, prevention can be enabled by avoiding the previously done ‘mistakes’. For
example, modifying the available cache values of a data mart which leads to disastrous
results can be avoided from repeating itself.
The aspects of prediction and prevention lead to a second application case, which is change
impact analysis. What happens to my DSS if I change this or that parameter, if I buy
additional resources, if I change the version of my BI software solution etc.? are commonly
asked questions. So far, answers to these questions were practically impossible to provide
specifically. With the help of the BI-CMDB detail levels, impact analysis can be rapidly
performed by comparing the various situations and intersecting the analysis axes (e.g. the
average response times per logical servers with different data mart cache values and
different RAM memory available on the server). By intersecting the various configuration
axes of the A-CMDB with the performance axis, one can isolate and explain performance
variations. Moreover, by building a knowledge base around the BI-CMDB, the question of
change management and migration can be responded with the integration of all the external
‘meta’ knowledge regarding the DSS management.
This is where the third advanced topic is discussed: providing meta-data and semantics
with the BI-CMDB. For example, integrating all the information from the release Readme
documents of the software, in order to identify the editor specified known and solved bugs
or software compatibilities. When performing a migration such information is vital, and
having it integrated with the BI-CMDB under a unified machine understandable form
would simplify a lot the processes. In order to achieve this, solutions that imply the usage of
web semantic technologies, specifically ontologies (Gruber, 1992), have been recently
proposed for exploration (Nicolicin-Georgescu et al., 2010), and have been the result of our
work over the last years. Using ontologies to formalize the DSS management aspects (such
as best practices, subjective quality measurement, business rules etc.) go hand in hand with
the BI-CMDB elaboration. Moreover, by using ontologies, it is possible to specify horizontal
relations between the CI, as an addition to the hierarchical vertical relations of the DSS
architecture, thus allowing a full description of the elements and their connections.
Last, but not least, having a complete integrated view of the data and information describing
the DSS, automatic behaviors can be adopted, such that the DSS human expert is helped
with managing his system. To this end, solutions as the Autonomic Computing (IBM, 2006)
have been developed, with the declared purpose of helping IT professionals with low level,
repetitive tasks, so they can focus on high level, business objectives. Inspired by the
functioning of the human body, this autonomy model proposes four principles to assure
that a system can function by ‘itself’, self: configuration, healing, optimization and
protection. An intelligent control loop implemented by several Autonomic Computing
Managers, assure these functions. It has four phases, which correspond to a self sufficient
process: monitor, analyze, plan and execute, all around a central knowledge base.
The problem with autonomic computing is that it has been elaborated with the real time
operational systems as targets. In order to apply the autonomic model for DSS, similar to the
passage from the CMDB to the BI-CMDB, the characteristics of DSS must be taken into
consideration. The works of (Nicolicin-Georgescu et al., 2009) have shown an example of
optimizing shared resources allocations with data warehouses, by focusing on the data
warehouse importance and priority to the user rather then on raw technical indicators.
Moreover, to comply with the DSS functioning, changes in the traditional loop functioning
were proposed, via the description of heuristics targeted at the SLAs and the improvement
of the QoS.
5. Possible limitations and shortcomings

Before concluding this chapter, an overview of the limitations and shortcomings of the BI-
CMDB is done. We note that with some of them we have already been confronted, whereas
others are presented only as possible to arrive.
The first limitation when building the BI-CMDB is generics, specifically the difference
between CI modeling of the different BI software solutions. As presented, the CI tables in
the O-CMDB consist of six hierarchy levels, which were thought to fit a maximum number
of software implementations. Yet, with every one of them, the connotations of a specific CI
table change (e.g. a base table in the case of OLAP software becomes a report table in the
case of reporting software). Moreover, the elements composing the DSS are not always
logically organized hierarchically (not on all levels). For instance the reports and data fluxes
with BO, are related through horizontal links. So vertically aggregating on the CI hierarchic
sometimes doesn’t make sense. Nevertheless, what is important is that the model offers the
specification capacity, thus finally generics is assured through small (of semantic nature)
changes.
Another limitation of the solution is related to the construction process time of the O and A
CMDBs. One may object the fact that due to the times needed to build these databases, real
time analysis cannot be achieved. Yet, is there a real need for real time? The answer may be
yes in the case of technical alerts (which simplifies a lot the processes), but no in the case of
DSS analysis. In this case, the ‘standard’ time interval is an operational day, following the
DSS usage purpose characteristic (i.e. the day/night usage cycles). At the end of each
operational day, the data collector recovers the new information via the configured
connectors, archives it and passes it for integration into the O-CMDB. This interval evidently
is configurable, based on the specific needs (e.g. once each week, once each 3 days etc.), and
user needs for this interval are mostly higher than the time needed to integrate the data into
the BI-CMDB.
Following, one shortcoming may be the unpredictability of usage of the system, which
makes scaling effects harder to measure. We have here a decisional of the decisional
solution, which metaphorically is like doubling the DSS expert motto. A user of the BI-
CMDB will constantly require new rapports and find new analysis axes to intersect. This
means, that alongside the list of the main analysis axis that are implemented, depending on
the users needs and objectives a series of new analysis pivots are described. As these needs
vary from case to case, it is hard to asses the usage process of the BI-CMDB. Its
performances depend strongly on the chosen client implementation. Moreover, through the
web portal interface the user has the liberty to define his custom queries towards the BI-
CMDB, a liberty which can make the system overcharge (similar to the data warehouse user
freedom problematic).
Also, as we have spoken of the combination of the BI-CMDB with ontologies, this is
regarded as a future evolution. For now, assuring model consistency and completeness
(from the contained data point of view) is done ‘semi-manually’, after the report generation
(therefore at the final stage). We try to move this verification at database construction level,
as to avoid any unnecessary report generation (or worst give the bad report to the bad
receiver). Implementing an ontology model and by benefitting from the inference engines
capacities, such verifications can be done at the construction stage, before any utilization of
the BI-CMDB data. A classic example is the mixing of the CIs from the hierarchy. The CI
specification is build from the manual user specification of his system, thus is prone to
human error. A physical server which becomes an application by accident would lead to
completely incorrect rapports (from what the user was expecting).
Finally, we note that semantic and autonomic evolutions of the BI-CMDB arise a series of
questions, from knowledge integration to supervision of autonomic behaviors and rule
consistency.
6. Conclusion
We have detailed in this chapter an implementation of the Business Intelligence – CMDB,
with the purpose of improving the efficiency of a DSS. Based on the fact that despite the
technological advancements, user satisfaction with DSS performance keeps decreasing, we
have isolated some of the elements that are responsible for this paradox, among which the
Quality of Service, the Total Cost of Ownership and the integration of Service Level
Agreements policies when computing DSS performance. We state that user perception is a
vital factor that cannot be ignored in the detriment of the raw technical performance
indicators.
In conclusion, implementing the BI-CMDB for DSS offers a series of elements we consider
compulsory to an efficient DSSs: a CMDB for DSS, a complete TCO, an integration of the
SLA/Os, an adoption of ITIL guidelines and best practices, all these with lower costs and the
improvement of the user’s satisfaction levels. The BI-CMDB proposition for the ‘decisional of
the decisional’, integrated with cutting edge semantic and autonomic technologies opens the
doors to a whole new area of research towards the future of efficient DSS.
Nevertheless, developing the BI-CMDB is a first successful step towards the future of DSS.
With the development of semantics and autonomic technologies, we strongly believe that
future steps include these aspects, towards the development of a complete BI-CMDB,
integrated with all pieces of knowledge required for managing a DSS. The ultimate
autonomic semantic based BI-CMDB for DSS is a planet which seems “light years” away, but
currently technology evolution speeds rapidly approach it.
7. Acknowledgment
We would like to thank Mr. Eng. Adrien Letournel and Mr. Eng. Guillaume St-Criq, both
employees of SP2 Solutions, for their support with providing the data and the elements for
the examples and for their feedback and support with this chapter.
8. References
Agarwala, S., Chen, Y., Milojicic, D. & Schwan, K. (2006), Qmon: Qos-and utility-aware
monitoring in enterprise systems, in ‘IEEE International Conference on Autonomic
Computing’, pp. 124–133.
Codd, E. F., Codd, S. & Salley, C. (1993), ‘Providing olap to user-analysts: An it mandate’.
Dumont, C. (2007), ITIL pour un service informatique optimal 2e édition, Eyrolles.
Ganek, A. G. & Corbi, T. A. (2003), ‘The dawning of the autonomic computing era’, IBM
Systems Journal 43(1), 5–18.
Gruber, T. (1992), What is an ontology?, Academic Press Pub.
IBM (2006), An architectural blueprint for autonomic computing, 4th Edition, IBM Corporation.
(2011), ‘Ibm cognos 8 financial performance management’. Last accessed - January 2011.
http://publib.boulder.ibm.com/infocenter/c8fpm/v8r4m0/index.jsp?topic=/com.
ibm.swg.im.cognos.ug_cra.8.4.0.doc/ug_cra_id30642LogFiles.html
Inmon, W.H. (2010), ‘Data warehousing: Kimball vs. inmon (by inmon)’. last accessed july
2010. http://www.b-eye-network.com/view/14115/
Inmon, W. H. (2005), Building the data warehouse, fourth edition, Wiley Publishing.
Inmon, W. H., Rudin, K., Buss, C. K. & Sousa, R. (1999), Data Warehouse Performance, John
WIley & Sons, Inc.
Nicolicin-Georgescu, V., Benatier, V., Lehn, R. & Briand, H. (2009), An ontology-based
autonomic system for improving data warehouse performances, in ‘Knowledge-
Based and Intelligent Information and Engineering Systems, 13th International
Conference, KES2009’, pp. 261–268.
Nicolicin-Georgescu, V., Benatier, V., Lehn, R. & Briand, H. (2010), Ontology-based
autonomic computing for resource sharing between data warehouses in decision
support systems, in ‘12th International Conference on Enterprise Information
Systems, ICEIS 2010’, pp. 199–206.
Objects, B. (2008), Business Objects enterprise xi white paper, Technical report, SAP Company.
Oracle (2008), Management excellence: The metrics reloaded, Technical report, Oracle.
Oracle (2010), ‘Oracle Hyperion Essbase’. last accessed July 2010.
http://www.oracle.com/technology/products/bi/essbase/index.html
Pendse, N. (2007), ‘Performance matters - the OLAP surveys hold some performance
surprises’. www.olapreport.com
SAP (2010), Data, data everywhere: A special report on managing information, Technical
report, SAP America, Inc.
SP2Solutions (2010), ‘SP2 Solutions homepage’. last accessed December 2010, www.sp2.fr
Vassiliadis, P., Bouzeghoub, M. & Quix, C. (1999), Towards quality-oriented data warehouse
usage and evolution, in ‘Proceedings of the 11th International Conference on
Advanced Information Systems Engineering, CAISE 99’, pp. 164–179.
5
Decision Support Systems Application to

Business Processes at Enterprises in Russia
Konstantin Aksyonov, Eugene Bykov,
Leonid Dorosinskiy, Elena Smoliy, Olga Aksyonova,
Anna Antonova and Irina Spitsina
Ural Federal University
Russian Federation
1. Introduction
The chapter focuses on one of possible approaches to decision support problem, which is
based on multi-agent simulation modelling. Most decision support cases generally consist of
a set of available alternatives definition, estimation and selection of the best one. When
choosing a solution one needs to consider a large number of conflicting objectives and, thus,
to estimate possible solutions on multiple criteria.
System analysis and synthesis constitute a management aid, targeting more effective system
organization considering its limitations and objectives. In the limits of decision support
process, analysis and synthesis methods are used for forecast and estimation of the
consequences of taken decision. In such case available situation development scenarios are
designed and analysed, scenario synthesis is performed, results are being evaluated in order
to develop the policy for the most desired results. Computer aid for analysis and synthesis
problems provides the most perspective and rational system development strategy making.
Formalized analysis and synthesis problems solving might be interesting from the point of
view of workflow automation for analysts and decision making people.
The following analysis and synthesis problems of business systems are discussed further:
virtual enterprise establishment; business process re-engineering; business process
benchmarking (improvement and enhancement). In Russia major research relates to
business systems analysis. This concept includes business processes, management,
production and manufacture, logistics, technological processes and decision making. The
task of virtual enterprise establishment (Andreichikov & Andreichikova, 2004) relates to a
class of complex systems structural synthesis tasks. The main objective of virtual enterprise
establishment is cooperation of legally independent companies and individuals,
manufacturing certain product or providing services in common business process. Main
goal of business processes re-engineering is reorganization of material, financial and
informational flows, targeting organizational structure simplification; resource use
redistribution and minimization; client needs satisfaction time decrease; quality of client
service improvement.
Business process re-engineering offers solution for the following tasks: existing business
process and enterprise activity analysis, decomposition to sub-processes; analysis of
bottlenecks in business processes structure, e.g. related to resources underload or overload,
queuing; business process re-organization (synthesis) for elimination of problems and reach
of set effectiveness criteria; design and development of information system for business
process support. Problem of business process structure bottlenecks analysis can be solved
with aid of static or dynamic modeling methods. Static modeling implies definition and
analysis of business process structure, as well as cost analysis of process functions,
identification of the most demanding and unprofitable functions, or those ones with low
coefficient of resource use. Dynamic simulation modeling allows implementation of
multiple business process operations in continuous period of time, offering statistics
gathering of process operation and identification of bottlenecks in their structure. Business
process re-organization relying on static modeling is based on heuristic approach and
demands high qualification and experience from analyst. Simulation modeling in business
processes re-engineering allows automation of re-engineering rules and their application.
Business process benchmarking closely relates to re-engineering. Main goal of
benchmarking is business process re-organization in accordance with master model. Master
model is based on a combination of best business processes of various enterprises, identified
with comparative analysis.
The main idea of the chapter is situational, multi-agent, simulation and expert modeling
methods and tools integration in order to increase the decision support effectiveness in
situational control of resource conversion.
Practical application of business process re-engineering approaches as management tools
has its limitations: need for a highly qualified analyst, his deep understanding of problem
domain; unavailability of synthesis of business process re-engineering decisions and check
of computed solutions on real management object. Simulation and AI methods applied to
automation of business process re-engineering allow to: 1. Reduce demand for analyst
experience in system analysis and synthesis by using formalized expert knowledge and
applying mathematical modelling methods; 2. Estimate computed re-engineering solutions
and select the most effective solution on management object model.
Use of multi-agent approach on stage of business process model formalization is caused by
the presence of decision making people in the system; their behaviour is motivated, they
cooperate with each other, accumulate knowledge of problem domain and management
tasks solution scenarios. Intelligent agent model provides design of decision making person
model, which is the basis of information technology implementation for model analysis and
decision making.
2. Analysis of business processes multi-agent dynamic models

Consider the problem of business process formalization model selection. The following
models support agent representation of business processes: Gaia model, Bugaichenko’s
model, Masloboev’s model, simulation model of intelligent agents interaction (SMIAI),
Resource-Activity-Operation model (RAO), multi-agent resource conversion processes
model (MRCP).
Gaia model of multi-agent system was offered by Michael Wooldridge and Nicholas
Jennings (Wooldridge et al., 2000). It defines a system as an artificial structure that consists
of heterogeneous agents, interacting with each other in order to achieve common goal that
in turn consists of agents’ sub-goals. Note that authors do not consider agents behaviour in
situations with competitive goals.
The most abstract entity in concepts hierarchy of Gaia model is a system. Next hierarchy
level contains roles (Fig. 1). Model design begins with analysis stage.
Decision Support Systems Application to Business Processes at Enterprises in Russia 85
Fig. 1. Gaia methodology base concepts

A role has 4 attributes (Fig. 2) (Wooldridge et. Al, 2000): Responsibility determines
functionality of role and can be divided into 2 kinds – Liveliness properties contain system
reaction to external exposure (reactive component) and can be defined with temporal logic
that has activity and role protocols as arguments; Safety properties define a set of activities
that should not occur during role functioning. Permissions include rights associated with a
role, for resources that are available. Resources in Gaia model may include information
objects that may be used in ways of generation, reading and modification. Activities are
computations to be carried out by the agents. These computations are carried out without
interactions with other roles. Ways of interaction with other roes are defined by Protocols.
Fig. 2. Role attributes in Gaia model

Interactions model includes definition of the attributes of interaction protocols: interaction
goal, interaction initiator role, interaction respondent role, interaction input and output
resources and text definition of interaction process. Authors imply limitation of system
structure being static in frames of roles interaction model definition.
Design phase starts with agent model definition. This incudes system agents types definition
and agent types correlation with defined roles. During simulation a defined number of
agent instances, in accordance with instance generation rules, are dynamically created for
each type.
Services model is provided for agent fuctionality definition, when each agent service is
mapped to a role defined on role activity analysis phase. Service has information inputs and
outputs (corresponding to attributes of roles interaction protocol), as well as start and finish
conditions in form of role safety properties. Here service represents a resource convertor,
active when input resources are present and start condition is fulfilled; it generates output
resources when finish condition is fulfilled.
Acquaintance model represents a directed graph with agent types as graph nodes and
communications between agent types as arcs.
Bugaichenko’s model (Bugaichenko, 2007) defines intelligent agents with mental (Belief-
Desire-Intention, BDI) architecture. Multi-agent system properties are defined with aid of
developed formal logic – MASL – multi-agent systems logical specification method with
time limitations; these systems are considered to be capable of experience accumulation and
analysis.
Multi-agent system model is represented with a triplet:
MAS=(S, AG, env), where

S – finite collection of external environment states;
AG={ag1,…,agn} – finite collection of agents;
env – function, defining a possible reaction of external environment to activity of all system
agents.
Each agent is represented with a collection:
ag=(S, A, env, see, Ib, bel, brf, ID, des, drf, filter, plan, prf), where
A – nonempty finite collection of agent activities;
env – external environment behavior function, correlating a collection of next possible states
of external environment to the current state of external environment and selected action of
an agent;
see – correct perception of external environment states by an agent, setting collection P of
equivalence classes on S;
Ib – collection of agent beliefs, dependent on agent perception of external environment and
his own activity;
bel – collection of current agent beliefs;
brf – beliefs update function;
ID – collection of agent desires; depends on goals (criteria functions);
des – collection of current agent desires;
drf – agent desires update function;
filter – agent desires filtration function;
plan – current agent plan, represented by finite state machine with input alphabet P, output
alphabet A, states set Ipln, transition relation σpln and initial state of ipln,0;
prf – plan update function.
Bugaichenko’s agent has mental (BDI) architecture, featuring three components: beliefs,
desires and intentions (Bugaichenko, 2005).
Beliefs contain information about regularities and current state of external environment of an
agent. This information may be erroneous or incomplete, thus it may be considered as belief,
not as reliable knowledge. Note that prognosis function is also considered an agent belief.
Agent desires make up a collection of all agent goals. It is unlikely that an agent, limited with
resources, will be able to achieve all his goals.
Intentions make up a collection of goals that an agent decided to achieve. Satisfiability of
intentions (in case of planning agent) may be defined as possession of a plan, leading to goal
achievement.
One of the most complex stages of decision support process is selection (planning) of
activities for goals achievement. Bugaichenko offers agent plans representation in form of a
network of interacting finite state machines (Bugaichenko, 2007).
Problems of accumulating experience by the system and agent self-prognosis of own activity
within Bugaichenko model are also solved with use of symbolic data representation in form
of resolving diagrams.
Multi-agent system properties are described with aid of developed formal logic MASL –
method for logical specification of multi-agent systems with time limitations; these systems
are considered to be capable of experience accumulation and analysis. MASL logic is the
interconnection of capabilities of the following temporal logics: Propositional Dynamic
Logic PDL, Real-Time Computation Tree Logic RTCTL and Alternating-Time Temporal
Logic ATL (Bugaichenko, 2006). Adaptation of these logics for specification of properties of
multi-agent system mathematical model allowed Bugaichenko formalize definition of
cooperation and agent competitiveness, define nondeterministic behaviour of external
environment, extend expressive power of specification language.
Key concept of agent interaction modeling is coalition. According to Bugaichenko, coalition
C is a certain subset of system agents that act together in order to achieve personal and
common goals. Agents of the coalition trust each other, exchange information and
coordinate activities. At the same time they do not trust the agents for outside the coalition,
and do not affect their behavior. So, formally, coalition may be considered a single
intelligent agent, that may have sense of external environment and mental behavior defined
as a combination of senses and mental states of coalition agents.
Masloboev’s multi-agent model (Masloboev, 2008) was developed for information support
of innovative activity in the region together with support and estimation of potentially
effective innovative structures.
Innovative activity subjects are represented in form of software agents, operating and
interacting with each other in common information environment (virtual business
environment, VBE) for the benefit of their owners, forming an open multi-agent system with
decentralized architrecture.
According to Masloboev (Masloboev, 2008), VBE model has the form of
EVBE={S, P, I, A, R, Atr}, where

S – collection of business process subjects;
P – collection of business processes;
I – set of relations on object models (business process subjects);
A – collection of agents;
R={BI, BPL} – collection of innovation resources, incuding business ideas (BI) and business
plans (BPL);
Atr – collection of model objects attributes.
Agent model has the form of:
A={S, BI, ORGA, CA}, where

ORGA – agent organizational structure,
CA – agent inner structure.
Organizational structure is defined in form:
ORGA={G, RL, CP, ACT, STR, L, ST, SL, T}, where

G – agent goals tree;
RL – collection of agent roles that it needs to operate in order to achieve the goals;
CP – collection of agent skills and capabilities that it needs to possess in order to operate the
roles;
ACT – collection of actions;
STR – collection of agent behavior strategies for goal achievement;

L – collection of languages, including agents interaction language, local planning language,
and execution level language, defined by used network services;
ST – set of agent states;
SL – collection of agent operation rules;
T – general transfer function on multitudes ST, SL and ACT (Masloboev, 2009).
Agent inner structure defines its functional design:
CA={KA, MA, PA, RA, IA, CA}, where

KA – mental subsystem;
MA – modeling subsystem;
PA – analysis and planning sybsystem;
RA – reactive subsystem;
IA – coordination and interaction subsystem;
CA – communication subsystem (Masloboev, 2009).
Masloboev’s hybrid agent architecture is an extension of existing InteRRaP architecture with
problem-oriented sybsystem of continuous simulation modeling (system dynamics models
complex), used by the agents for simulation of single innovation projects development
scenarios, and behavior of companions and competitors on innovative services market.
Thus, use of simulation apparatus lets the agent forecast the results of its activity (risks and
economical effect of investments into innovations). Hybrid architecture consists of these
elements:
 Knowledge control subsystem, based on descriptive logic, provides temporal reasoning
and some other processing mechanisms. Three levels of knowledge play key role in
agent architecture: problem domain knowledge, interaction knowledge (general
declarative behavior rules together with rules for problem domain knowledge
replenishment and update), and control knowledge (knowledge, applying interaction
knowledge to problem domain knowledge for replenishing and update of operational
memory).
 Operational memory is used as a temporary data storage for control subsystem
knowledge, user or communication control data. Operates in form global message
board.
 Decision support system is implemented on the basis of OLAP technology.
 Communications control module compiles and sends messages to other agents, also
receives message delivery confirmations. ACL (agent communication language), based
on theory of speech activity, is used for agent interaction.
Masloboev’s model is designed for creation of virtual enterprise for innovation project
implementation by selecting the most effective one (from the point of view of innovation
project investments) and its correlation with the most advantageous business structures
capable of project implementation. Innovation project is selected on the basis of effectiveness
criteria mathematical models (Masloboev, 2008), including such project indicators as
economical effectiveness, estimated project implementation time, partners reliability
(competence).
Business structures within the model are implemented as agent coalitions with common
goals of corresponding business ideas implementation. Business ideas are used for
automatic generation of agent coalition on the basis of semantic matchmaking of
corresponding business idea parameters.
Model of intelligent agent with simulation apparatus is implemented in multi-agent system

of innovation activity information support, providing joint use of innovation portals
information databases during automated generation of potentially effective business
structures, targeting innovation projects implementation.
Simulation model of intelligent agents interaction (SMIAI) was developed by Rybina and
Paronjanov with the purpose of formalizing communicative activity of intelligent agents by
modelling separate communication components: interaction participants, communicative
environment, problem region, interaction language, dialog scenarios (Rybina & Paronjanov,
2008a).
Model of intelligent agents interaction is defined in form (Rybina & Paronjanov, 2008a):
MI={SA, SE, SP, DI, L, SRA}, where

SA – collection of agents Ai;
SE – collection of communicative environments Ek;
SP – collection of problem regions, set by problem domains PR (collections of classes, their
instances and relations between them), collection of current tasks SТ (represented with five
typical tasks: diagnostic, engineering, planning, control and study) and relations of
compliance of elements PR with current tasks from collection ST;
DI – collection of dialog scenarios between interaction participants within multi-agent
system;
L – interaction language, represented with lexical, syntactical and semantical components;
SRA – collection of relations, indicating possibility of interaction between a pair of
interaction participants Aj and Ai in environment Ek.
Intelligent agent model was received by adaptation of Cetnorovich’s model K with a
purpose of having intelligent agents as participants of interaction (Rybina & Paronjanov,
2008a):
A={M, Q, R, P, Eff, MI, Plan, Aim, Time}, where

M – collection of environment models, available to the agent;
Q – collection of agent goals;
R – collection of actions, acceptable for the agent;
P – collection of tasks decomposition options (partial plans library);
Eff – collection of acceptable impacts on the agent from environment;
MI – agents communication model;
Plan – function of agent activity plan generation based on current goal and environment
model;
Aim – agent goal-setting mechanism based on external impacts;
Time – collection of activities reflection into their duration based on environment state.
Interacting agents must have knowledge of communication language and problem domain
that have to be the same for each agent or at least have common areas (Rybina &
Paronjanov, 2008b).
Interation of intelligent agents in SMIAI model is performed in form of a dialog (polidialog),
including global structure, dependent only on goals of interaction participants, thematical
structure, dependent only on the the current problem, and local structure, used for
representation of communication actions sequence on local level (Rybina & Paronjanov,
2008b). Authors offer use of painted Petri nets for local interaction steps definition (i.e.
actions and reactions of partner agents on each step of interaction).
Model of dialog local structure has finite set of dialog states (Pi positions) and finite set of
transfers tj, each of which has a corresponding expression (transfer condition) in interaction
language L.Model makes use of KIF and KQML languages for language interaction. The
main concept here is speech acts theory, which is the basis of KQML language, as well as
large amount of problem domains, defined on the basis of KIF language.
Graphical notation of local structure model for agents dialog is presented on Fig. 3.
Fig. 3. Graphical notation of local structure model for agents dialog

Each dialog position is defined in form:
 execut _ action  
   
P  agent _ state exec _ reactions , color _ position, actions _ collection 
 
 end _ of _ dialg  
 
Each dialog transfer has a structure like:
t  Pi  1 ,Pi , transfer _ condition( pa nting _ Pi  1 ), actions _ with _ colors , where
Delk  position _ with _ deleted _ color ; deleted _ color ,

actions _ with _ colors  
 Addk  position _ with _ added _ color ; added _ color .
Sequece of agents communicative actions is defined on the Petri net by output track
analysis, determination of passed positions and inclusion of those actions into final sequence
that are defined in actions list for current position.
General model of intelligent agents interaction is implemented in SMIAI, developed with
aid of Gensym Corp. G2 and Microsoft Visual Studio. System has been tested in problem
domains of online billing, investment projects management, control of state for chemically-
dangerous objects anf other (Rybina & Paronjanov, 2009).
Resource-Activity-Operation model (RAO) is used for definition of complex discrete
systems (CDS) and activities within these systems in order to study static and dynamic
features of events and activities (Emelyanov & Yasinovskiy, 1998). System distrecity is
defined with two properties:
 CDS content may be defined with countable set of resources, each of which relates to
certain type;
 CDS state changes occur in countable moments of time (events) and C e+i = C e-i+1 (system
state after ei event is identical to system state before ei+1 event).
Conceptually CDS may be represented with a set of resources, that have certain parameters.
Parameter structure is inherited from resource type. Resource state is defined with a vector
of all its parameter values. System state is defined with all its resource parameter values.
There are two types of resources – permanent (always exist during model simulation) and
temporary (dynamically created and destroyed during model simulation). By analogy with
GPSS language, RAO resources may be called transacts, that are dynamically generated in
certain model object, pass the other objects and are finally destroyed. Note that resource
database in RAO is implemented on the basis of queueing system apparatus, having
automatic support of system operation statistic gathering: average value of examined
indicator, minimum and maximum values, standard deviation.
Resources perform specific activities by interacting with each other. Activities are defined
with events of activity start and activity end, the event in general is considered a signal,
transferring data (control) about certain CDS state for certain activity. All events are divided
into regular and irregular. Regular events reflect logic of resource interaction between each
other (interaction sequence). Irregular events define changes of the system, unpredictable in
production model. Thus, CDS operation may be considered as timely sequence of activities
and interrupting irregular events.
Activities are defined with operations, that are basically modified production rules that
consider timely relations:
IF (condition) THEN1 (event1) WAIT (timely interval) THEN2 (event2).

Operation defines pre-conditions (operation resources state should fulfill those) and state
change rules for resources in the beginning and end of corresponding activity. Operation
with temporal interval of 0 in RAO model is considered a decision point, and is a usual
production rule. RAO decision points are equivalent to reactive agents that store
environment response data in knowledge base.
Intelligent modeling system RAO-simulator has been developed, its structure is shown on
Fig. 4.
Fig. 4. Structure of RAO-simulator

Main elements of RAO-simulator include dynamic production system and events apparatus.
During simulation system state changes in accordance with irregular event definition or
activity that started or ended. Output system is called after each state change. It scans
knowledge base for all operations and checks pre-conditions for availability of operation
start. If located, the start events are raised for corresponding actions. Trace system displays
detailed information on events to dedicated file, which is further processed for process
detailed analysis and information representation in convenient form.
Dynamic model of multi-agent resource conversion processes (MRCP) (Aksyonov &
Goncharova, 2006) was developed on the basis of resource conversion process (RCP) model
(Aksyonov, 2003) and targets modeling of business processes and decision support for
management and control pricesses.
Multi-agent resource conversion process model was developed on the basis of several
approaches integration: simulation and situational modeling, expert and multi-agent
systems, object-oriented approach.
Key concept of the RCP model is a resource convertor that has input, launch condition,
conversion, control, output. Launch condition defines a moment in time when a convertor
starts activity on the basis of such factors as state of conversion process, state of
input/output resources, control commands, tools for conversion process and other events in
external environment. Conversion duration is defined immediately before conversion based
on control command parameters and active resource limitations.
MRCP model may be considered an extension to base RCP model, adding functionality of
intelligent agents.
The main objects of discrete Multi-agent RCP are: operations (Op), resources (Res), control
commands (U), conversion devices (Mech), processes (PR), sources (Sender) and resource
receivers (Receiver), junctions (Junction), parameters (P), agents (Agent) and coalitions (C).
Process parameters are set by the object characteristics function. Relations between resources
and conversion device are set by link object (Relation). The agents and coalitions existence
resumes availability of the situations (Situation) and decisions (action plan) (Decision).
MRCP model has hierarchical structure, defined with system graphs of high-level
integration.
Agents control the RCP objects. There is a model of the decision-making person for every
agent. An agent (software or hardware entity) is defined as an autonomous artificial object,
demonstrating active motivated behavior and capable of interaction with other objects in
dynamic virtual environment. In every point of system time a modeled agent performs the
following operations (Aksyonov & Goncharova, 2006): environment (current system state)
analysis; state diagnosis; knowledge base access (knowledge base (KB) and data base (DB)
interaction); decision-making. Thus the functions of analysis, situations structuring and
abstraction, as well as resource conversion process control commands generation are
performed by agents.
Coalition is generated consequently after several agents union. Agent coalition has the
following structure:
С = <Name, {A1,…,Am}, GС, KBС , M_In, M_Out, SPC, Control_O >, where
Name – coalition name;
{A1,…, Am} – a collection of agents, forming a coalition;
GС – coalition goal;
KBС – coalition knowledge base;
M_In – a collection of incoming messages;
M_Out – a collection of outgoing messages;
SPC – a collection of behaviour scenarious acceptable within coalition;
Control_O – a collection of controlled objects of resource conversion process.
Fig. 5 shows an example of C1 coalition formation after union of A2 and A3 agents. Here C1
coalition controls agents A2 and A3, but A1 agent acts independently.
SPA1
Receiver1
U1
SPC1
Sender1 U4
Res1, Res4,
Res2 PR1 Res5,
Res6 U5
A1
Op1
Res7,
С1 Mech1,
Res8 KBA1
Mech2, Op2
Mech3,
KBС1 Res9,
Res10
MsgС1,A3 Mech4,
MsgС1,A2 Mech5
MsgA2,С1 MsgA3,С1
Res1 SPA2 SPA3

U2 U3
PR2 Res4 Msg1 Op3 Res5

Res2 ...
A2 A3
Res3 Res3 Res6
KBA2 Msgn
KBA3
С1
Mech Mech Mech
1 2 3
Fig. 5. Coalition formation

Simulation engine algorithm of agent-containing model consists of the following main
stages: current point of system time identification SysTime  min Tj ; agent and coalition
jRULE
actions processing (state diagnosis, control commands generation); conversion rules queue
generation; conversion rules execution and operation memory state (i.e. resources and
mechanisms values) modification. Simulator makes use of expert system unit for situations
diagnosis and control commands generation (Aksyonov et al., 2008a).
Each agent possesses its knowledge base, set of goals that are needed for behavior
configuration setting, and priority that defines agent order in control gaining queue.
Generally in case of any corresponding to agent’s activity situation an agent tries to find a
decision (action scenario) in the knowledge base or work it out itself; makes a decision;
controls goals achievement; delegates the goals to its own or another agent's RCP objects;
exchanges messages with others.
Multi-agent resource conversion process agent may have hybrid nature and contain two
components (Fig. 6):
 Intelligent (production rules and/or frame-based expert system access).
 Reactive (agent activity is defined on UML activity diagram)
Two main agent architecture classes are distinguished. They are:
1. Deliberative agent architecture (Wooldridge, 2005), based on artificial intelligence
principles and methods, i.e. knowledge-based systems;
2. Reactive architecture, based on system reaction to external environment events.
All currently existing architectures cannot be defined as purely behavioral or purely

knowledge-based. Any designed architecture is hybrid, offering features of both types.
Multi-agent resource conversion process architecture is based on InteRRaP (Muller &
Pischel, 1993; Aksyonov et al., 2009a) architecture, as the most appropriate for problem
domain.
In accordance with InteRRaP architecture common concept, multi-agent RCP agent model is
represented in four levels:
1. External environment model corresponds to the following MRCP elements: convertors,
resources, tools, parameters, goals. External environment performs the following
actions: generates tasks, transfers messages between agents, processes agent commands
(performs resource conversion), alters current state of external environment (transfers
situation Sn into state Sn+1).
2. External environment interface and reactive behavior components are implemented in
form of agent productional rules base and inference machine (simulation algorithm).
3. Reactive behavior components performs the following actions: receives tasks from
external environment, places tasks in goal stack, collates goal stack in accordance with
adopted goal ranging strategy, selects top goal from stack, searches knowledge base. If
appropriate rule is located, component transfers control to corresponding resource
convertor from external environment. Otherwise, component queries local planning
sub-system.
4. Local planning level purpose is effective search for solutions in complex situations (e.g.
when goal achievement requires several steps or several ways for goal achievement are
available). Local planning component is based on frame expert system. Frame-concept
and conceptual-graph based approach is utilized for knowledge formalization.
Fig. 6. MRCP agent hybrid architecture

Problem domain conceptual model and agent local planning knowledge base design is
based on UML class diagram extension. Semantically this notion may be interpreted as
definition of full decision search graph, containing all available goal achievement ways (pre-
defined by experts). Current knowledge base inference machine is implemented in decision
search diagram (Fig. 7), based on UML sequence diagram. Each decision represents agent
activity plan. Each plan consists of a set of rules from reactive component knowledge base.
Based on located decision, current agent plan is updated. Examination of all available
options, contained in knowledge base, generates agent plans library.
Fig. 7. General decision search diagram in decision support system BPsim.DSS

If an agent, when processing task or message received from external environment, is unable
to locate appropriate rule in its knowledge base (e.g., select an option from several ones), the
reactive behavior component queries plans library, indicating goal (i.e. task to execute, or
external environment state to bring into). Planning sub-system searches plans library, selects
appropriate plan and places first rule of selected plan into reactive component goals stack.
Comparison of described models in presented in Table 1. The following comparison criteria
has been selected: resource convertor model (discrete business process model); queue
system model – tool for analysis of discrete stochastic systems and business process; reactive
agent and intelligent agent models (tool for decision making people definition).
Full potential of intelligent agents is implemented in Bugaichenko, Maslovboev and MRCP
models. At the same time Buhgaichenko’s agent is not equivalent to decision making
person, but is a software entity that works independent from an expert (analyst). Masloboev
uses a complex of continuous system dynamics models for agent actions forecast and has no
support for discrete-event processes. SMIAI model is oriented on agent communication
methods research and does not integrate with simulation modeling system. Gaia model has
support for information resources convertor and intelligent agent, but has less functional
capabilities, than RAO and MRCP.
RAO model is similar to MRCP in areas of decision support and has powerful tools for
business processes formalization. A serious disadvantage is missing support of intelligent
agents. MRCP model includes a hybrid agent model (intelligent and reactive), model of
resource convertor and queue system, allowing the analyst to analyze dynamic features of
business processes. Hybrid agent model allows analysis of various decision making people
behavior scenarios.
Masloboev’s
Bugaichen-
ko’s model
Rybina’s
Comparison criteria
MRCP
model
model
model
model
model
RAO
Gaia
Resource convertor model
input/output/start ●/●/ ●/●/ ●/●/
conditions/duration ●/○ ●/● ●/●
hierarchical model
○ ○ ○ ○ ○ ●
of convertor
operations timely
○ ○ ●
interrupts
Queue system ○ ○ ○ ○ ● ●
Reactive agent
model MASL Painted
Producti- Producti- Producti-
(knowledge temporal Petri ○
ons ons ons
representation logic nets
form)
Intelligent agent model
agent goals ● ● ● ● ●
actions forecast ● ● ○ ○ ●
action analysis and
● ● ● ● ●
planning
agents self-study ● ● ● ○ ●
planning, forecast, Resolving
System Frame-
self-study diagrams ○
dynamics Expert Expert based
implementation -based
simula- system system expert
technology expert
tion system
system
message exchange TPA /
TPA TPA Signals Signals
language Signals
agents cooperation
● ● ● ● ●
model
Software
Proto- Proto-
implementation Beta ○ Beta RTM
type type
of the model
Table 1. Comparison of dynamic situation modeling models

Thus, MRCP model and BPsim products have the fullest functionality in area of business
processes formalization.
3. Current state of dynamic situations modeling systems

Dynamic situations modeling systems area state analysis reveals unavailability of resource
conversion processes oriented systems. Nearest functionality analogs (Table 2) include
simulation and expert modeling tools, particularly real-time expert system G2 (G), multi-
agent simulation system AnyLogic (L), business-processes modeling system ARIS (T),
simulation system Arena (A) , simulation model of intelligent agents interaction SMIAI (S),
Masloboev’s innovations support system (M), Resource-Activity-Operation model (R). The
last three are unavailable for purchase, that is the reason for zero in retail price line.
Comparison criteria T G L A S M R
Problem domain conceptual model design ○ ○ ○ ○ ○ ○ ●
RCP description language ● ● ● ● ● ● ●
Systems goals definition:
 Graphical ● ● ○ ○ ○ ○ ○
 Balanced ScoreCard based ● ○ ○ ○ ○ ○ ○
Hierarchical process model ● ● ● ● ● ○ ○
Commands description language ○ ● ○ ○ ○ ○ ○
Use of natural language for model definition ○ ● ○ ○ ● ○ ○
Multi-agent modeling availability
 “Agent” element ○ ○ ● ○ ● ● ○
 Agents behavior models ○ ○ ● ○ ● ● ○
 Agent’s knowledge base support ○ ○ ○ ○ ● ● ○
 Message exchange language ○ ○ ○ ○ ● ○ ○
Discrete event Simulation modelling ● ● ● ● ● ○ ●
Expert modeling ○ ● ○ ○ ● ○ ○
Situational modeling ○ ● ○ ○ ● ○ ○
Object-oriented approach
 Use of UML language ● ○ ○ ○ ○ ○ ○
 Object-oriented programming ○ ● ● ○ ● ○ ●
 Object-oriented simulation ○ ● ● ○ ● ○ ●
 Problem domain conceptual model and object-
○ ○ ○ ○ ○ ○ ●
oriented simulation integration
Retail price, ths $ 50 70 8 4 0 0 0
Table 2. Modeling tools comparison
As we can see, all current systems lack support of some features that might be useful in
effective simulation. For example, problem domain conceptual model design and agent-
based approach implementation is limited, except RAO system that makes use of internal
programming language. Another disadvantage of two most powerful systems, ARIS ToolSet
and G2, is a very high retail price, which might stop a potential customer. Also systems such
as AnyLogic, G2, SMIAI and RAO require programming skills from users. So, from a non-
programming user’s point of view, no system has convenient multi-agent resource
conversion process definition aids. Again, AnyLogic and G2 make use of high-level
programming language, which results in these products being highly functional.
Simulation, situational modeling and expert systems are used in modeling, analysis and
synthesis of organizational-technical systems and business processes. Multi-agent resource
conversion processes theory (Aksyonov & Goncharova, 2006) may be used for
organizational-technical systems definition from decision support point of view, a dynamic
component of business processes, expert systems, situational modeling and multi-agent
systems.
Next section presents development principles and technical decisions of designed object-
oriented multi-agent resource conversion processes based decision support system, relying on
above-stated multi-agent resource conversion process model and multi-agent architecture.
4. Multi-agent systems simulation and engineering systems integration

Object-oriented decision support system BPsim.DSS (“Business Processes Simulation –
Decision Support System”) is implemented on basis of dynamic situations modeling system
BPsim.MAS (“Multi-Agent Simulation”), software engineering system BPsim.SD (“Software
Designer”) and technical economical engineering system BPsim.MSN (“Multi-Service
Network”) integration.
The following program packages are being used during multi-agent resource conversion
processes problem domain business process modeling and software design (www.bpsim.ru),
offering a comprehensive solution for business modeling and techno-economic engineering
problems, which in turn considerably simplifies and speeds analysts’ work:
 BPsim.MAS – multi-agent dynamic situations modeling system (Aksyonov et al., 2008a).
BPsim.MAS offers the following functionality:
a. Multi-agent resource conversion process model design;
b. Dynamic simulation;
c. Experiment results analysis;
d. Model- and results-based reporting;
e. Experiment data export to Microsoft Office family products.
 BPsim.SD – Software Developer CASE tool
BPsim.SD offers automation on the following phases of software development:
a. DFD diagrams design is not automated. As in every CASE tool a DFD diagram
needs to be designed manually;
b. Use-case diagrams design is fully automated, use-case diagrams are achieved by a
transition from a DFD diagram. This process lets us keep our business objects;
c. Classes diagram design is partially automated. The core classes frames are
generated automatically, that greatly simplifies work on the final classes diagram.
Benefit is estimated in 10-15%;
d. Sequence diagram design is semi-automatic.
e. Database structure generation is automated.
BPsim.SD offers an opportunity of forms design. This allows the end-user place the controls
on the form as he wants them to be positioned. Some of the controls can be associated with
data on the phase of GUI design before passing the project to the developers. After this
phase a developer receives GUI forms in an appropriate format, i.e. the forms are saved in a
software development file format (Aksyonov et al., 2008b).
 BPsim.MSN – techno-economic engineering intellectual system (Aksyonov et al., 2009b)

automates the following functions:
a. Problem domain conceptual model engineering;
b. Filling the knowledge base data;
c. Decision search diagrams design, setting up dialog-based expert system;
d. Decision search.
 BPsim Wizard Technology – a framework of intelligent software assistants for step-by-
step model definition. A wizard is a dialog-based program assistant targeting
information integration and conversion from one system (BPsim.DSS / BPsim.MAS /
BPsim.SD) to another. BPsim Wizard Technology performs the following functions:
a. Transfers information between simulation, decision support and software
engineering modules in the framework of a single complex problem;
b. Simplifies a non-programming user experience when getting started with BPsim
products family;
c. Validates data on various stages of simulation model design, problem domain
conceptual model and information system engineering.
Various tools and methods use on all stages of organizational technical systems analysis and
synthesis and their support by BPsim products is presented in Table 3.
BPsim.DSS agent model is represented with four levels in compliance with InteRRaP
architecture general concept. External interface and reactive behavior components together
with external environment model are implemented in BPsim.MAS tool. Local planning
component is based on BPsim.MSN expert system module. Expert system shell visual
output mechanism builder is based on decision search diagrams (UML sequence diagram
extension). Cooperation level is based on both modules.
No Stage Tool Support in BPsim

1. Processes definition IDEF0 notation SD, MSN
DFD notation SD, MSN
UML use-case diagram SD, MSN
Multi-agent resource conversion
MAS
processes notation
2. Software engineering DFD use-case, class, sequence
SD, MSN
diagrams
3. Knowledge Semantic networks MSN
formalization Frames MSN
Production rules MAS
4. Decision support Simulation MAS
Multi-agent simulation MAS
Situational control MAS, MSN
Dialog-based expert systems MSN
Table 3. Methods used in BPsim products
An example, illustrating decision search diagram workflow, is presented on Fig. 8. For
simplification the dialog form classes are not shown on the diagram. The example illustrates
work of expert system for real estate agency. The figure shows available house/apartment
search in the database on the basis of user set criteria. The search is run in form of decision
search diagram.
Object-oriented decision support system BPsim.DSS allows the following features
implementation:
1. Problem domain conceptual model definition
2. Multi-agent resource conversion process dynamic model design
3. Dynamic simulation
4. Experiment results analysis
5. Reporting on models and experiment results
6. Data export to MS Excel and MS Project
Decision support system visual output mechanism builder, based on decision search
diagrams (Fig. 7), as well represents agent knowledge base, based on frame-concepts. So,
agent knowledge base may be defined in two ways: productive and frame-concept – based.
Here is an example of wizard implementation. This one focuses on analysis an re-
engineering of business processes.
System analysis in area of business process management implies design of various system
models (or views) by the analyst, each of which reflects certain aspects of its behavior. Re-
engineering of these models targets the search of alternative ways of system processes
development by means of consolidation of certain model sub-elements into a single process.
There are three ways of achieving this. One is parametric synthesis, when initial model is
converted into series of new models by modifying separate parameters. Another is structural
synthesis, this is a way of forming new models by modifying model structure according
certain rules. Mixed synthesis uses features of both options. An actual task is development of
algorithm for model analysis and mixed synthesis.
Fig. 8. Decision search tree for decision search diagram

Consider initial model M0 of multi-agent resource conversion process. Carrying out of

experiments gathers statistical data of operation execution, agent activity and tools
utilization in operations. Analysis and mixed synthesis algorithm is shown on Fig. 9 in form
of decision search graph. Graph nodes stand for: 0 – zero value, L – low value, A – average
value, H – high value – of corresponding graph object (queue size, load or idling). Directed
graph arcs connect fulfilled conditions of model analysis (nodes) into a general solution of
synthesis operator application. For example, solution for application of operator “Delete
parallel operation” is the following chain execution: model statistics contains zero values of
average transact queue for operation, average operation load, operation idling caused by
missing resources or occupied tools. Dashed arcs correspond to solutions with zero-length
or short transact queue for operation. Solid arcs correspond to solutions with long transact
queue for operation.
Fig. 9. Decision search graph for synthesis operators application, implemented in intelligent
agent
There are several examples demonstrating BPsim.DSS system application. They are
presented in the next section.
5. BPsim.DSS system application

5.1 BPsim.DSS application to IT projects management
Decision support system BPsim.DSS was used on various stages of Ural State Technical
University Common Information System (CIS) development and deployment, starting with
educational process analysis stage, performing re-engineering, and ending with separate
CIS units deployment efficiency estimation.
Model of an agent (decision making person), controlling software development process in
Ural State Technical University, was developed in decision support system BPsim.DSS.
Model consists of simulation model “Educational process software development” and
decision support models, including the main model “CIS implementation options selection”.
Model knowledge base contains information on networking, hardware and software,
information systems, IT-projects, teams of IT-specialists.
Expert system module is used for project alternatives and effective alternative search
algorithms knowledge base development. Simulation model is used for separate project
stages monitoring, detection of errors and conflicts, occurred on initial planning stage,
solution of vis major (i.e. search of decision in a force majeure situation that occurs under
irresistible force or compulsion and may drastically change system certain parameters), that
happens during development project control and CIS deployment. Simulation model is
based on Spiral model of software lifecycle and is designed in BPsim.DSS.
BPsim line products were used for business processes analysis and requirements specification
development for Common Information System (CIS) of Ural Federal University. University
has complex organizational structure (faculties, departments, subsidiaries and representative
offices), which means that success of University business processes optimization depends on
quality of survey and enterprise model comprehensiveness. Survey revealed non-optimal
implementation of certain processes at the University, e.g. movement of personnel (Fig. 10).
Fig. 10. Movement of personnel model

Simulation model “as-is” of the process was designed in dynamic situations modeling
system BPsim.MAS. Model data has been achieved from questioning employees of
personnel office, IT department, and four selected dean offices. Model nodes represent
document processing stages. Nodes 1-10 correspond to document processing within faculty,
nodes 11-14 – processing of faculties weekly documents in personnel office. Use of
intelligent agent for business process analysis and re-engineering allowed fixing certain
problems.
1. Intelligent agent discovered two identical transacts within the model. Analyst decided
that dean office employees prepared two similar documents on different stages –
application and first statement. Recommendation – unification of these documents.
2. Intelligent agent discovered a feedback loop, which is a negative indicator for
document flow business process. Analysis of highly-loaded personnel affairs-involved
employees revealed that dean offices allowed generation of long document queues.
Also asst. rector was required to approve 3 similar documents at different stages.
Recommendation – process automation, use of electronic documents.
3. Analyst discovered that IT personnel assistance is required for database data
modification. Recommendation – process automation, data changes should be carried
out by information-specific personnel, IT dept. employees should not perform
uncommon functions.
The following process updates were offered after analysis and re-engineering:
1. When employee signs an application, the dean office representative creates its electronic
copy in common database;
2. First statement is not necessary, all information needed for the order is stored in
common database, so, approvals need to be gathered only once;
3. All process participants work directly with electronic application, without requirement
of sending documents to dean office;
4. Availability of storing files electronically;
5. Asst. rector approves only application and order;
6. Availability of tracing document route;
7. Employees of IT department no longer perform uncommon function of updating
database;
8. In case of urgent order the corresponding information objects are generated by the dean
office employee.
As a result, the “as-to-be” model no longer has nodes 7, 9, 10, 15 (Fig. 10). New business
process definition became a basis for CIS module “Movement of personnel” requirements
specification, which as well used diagrams, designed in BPsim.SD CASE tool. The
deployment effectiveness was estimated after deployment and half-year use of the module
based on simulation models “as-is” (old model) and “as-to-be” (new model). Results are
presented in Table 4.
Thus, due to “Movement of personnel” process improvement and automation dean’s office
employees work efficiency was raised by 27%, student desk employees work efficiency was
raised by 229%. Deployment economical effect is estimated by about 25 thousand euro per
calendar year. Economical effect is achieved in shortening and automation of unnecessary
document processing stages, information double input prevention and employee load
decrease.
“AS-IS” “AS-TO-BE”
Indicators
model model
Documents processed, per month 390 1264
Documents lost, per month 12 0
Dean office employees performance,
0.4 0.5
documents per hour
Personnel office employees performance,
2.4 7.9
documents per hour
Table 4. “Movement of personnel” deployment effect estimation
5.2 BPsim.DSS application to Subaru auto dealer logistical processes simulation

Finally, BPsim.DSS was applied to Subaru auto dealer sale process. Simulation result
analysis helped this process be optimized, i.e. certain initial parameters being modified,
resulting in effective logistics and warehouse processes. Initial data for simulation included
sales statistics for each model, average retail pricing and dealer price markup, together with
sales statistics depending on initial car location (at warehouse on location, at official Subary
representative’s warehouse in Moscow, at Japanese warehouse ready for delivery),
including number of contracts and average delivery time from the order date. The main
purpose was to estimate the necessary number of cars of each model at the warehouses on
location and in Moscow, in order to achieve sales results of 20 to 40 cars per month.
Another model for Subaru auto included simulation of car repair process. The model
considered main repair process stages, resulting in effective search of repair strategy.
Model was designed to examine, analyse and improve repair department activity of two
dealers in Siberian region of Russia, and was based on the statistical data from the dealers.
The model can be used by other enterprises, provided that it is adapted accordingly.
5.3 BPsim.DSS application to multi-service telecommunication networks technical

economical engineering
Another application of BPsim.DSS included multi-service telecommunication network
models design and telecommunication services area business processes dynamic simulation.
Currently leading Russian region cellular carriers engineers polling revealed, that carriers’
development departments use their own experimental knowledge base when engineering
data-communication networks, while data-communication implementation engineering
solutions are foisted by hardware vendors. No operator either makes use of data-
communication networks automated design aids, or models various designed/existing
network behaviour situations when developing new regions, introducing new services or
modifying data-communication network topology.
Development of automated design and modelling methods and aids requires large quantity
of primary data for qualitative MSN technical and economical engineering, which includes:
telecommunication hardware and technologies types and parameters; engineers,
economists, project managers, marketers, and lawyers’ level of knowledge.
Decision support systems fit most for MSN technical and economical engineering problem
solution. Decision support systems can make use of simulation, expert and situational
modelling (Aksyonov & Goncharova, 2006). Decision support systems development and
deployment within cellular communication operators is a pressing and needed problem.
The following mathematical methods are used in MSN and business processes modelling,
analysis and synthesis tasks: teletraffic theory may be used on all MSN levels except services
level; simulation, situational and expert modelling methods are used for business processes
analysis and synthesis tasks. Expert and situational modelling methods, neural networks,
multi-agent and evolutionary modelling methods can be used in RCP formalization.
Multi-agent resource conversion processes theory is applied for MSN definition from
decision support point of view.
Frame-concept and conceptual graphs based approach, offered by A. N. Shvetsov and
implemented in form of «Frame systems constructor» expert system shell (FSC), is used as a
means of knowledge formalization (Shvetsov, 2004). A frame-based semantic network,
representing feasible relations between frame-concepts, is defined in form of extended UML
classes diagram, at the stage of system analysis.
UML sequence diagram is used for visual FSC output mechanism builder implementation.
This approach allows visual (in form of flowchart) problem solution flow definition, when
solution turns into a sequence of procedure (method/daemon) calls from one frame to
another. Hereby, this approach allowed visual object-oriented ontology and knowledge-
based output mechanism constructor implementation in form of decision search diagrams.
BPsim.DSS was applied for MSN technical economical engineering in Ural region, covering
metropolis Ekaterinburg, Russia, and satellites. Designed model is shown on Fig. 11.
Fig. 11. MSN Ekaterinburg – Sergi model view

This constructor, provided that being filled with MSN subject area knowledge and technical
and economical engineering rules, represents an intelligent MSN automated engineering
system.
Graphical implementation of the model is presented on Fig. 12. Model allows switching on
and off Base Stations (Access network elements) and Transport Networks, as well as
changing elements parameters and allowing to select from options of renting or constructing
a specific element.
Fig. 12. Modelled MSN graphical model

The model is designed with a main purpose of MSN technical economical engineering with
a centre in the metropolis and covering surrounding towns. Main goal is to estimate
available MSN deployment options for provision of cellular and data transfer services.
Synthesized model allows estimation of main investment indicators (IRR, EBI, Payback
Period), that are required for substantiated decision making in MSN engineering.
Live, one of the experiments that performed best, was implemented and performance
indicators were measured after a certain while of performance. The real indicators were
close to ones estimated with aid of decision support system BPsim.DSS.
6. Conclusion
In this chapter we have presented the following keynote features.
Some popular dynamic situations modelling systems including AnyLogic, ARIS, G2, Arena,
RAO and models/prototypes of other systems were compared. This comparison revealed
the necessity of a new system development, for it to be focused on multi-agent resource
conversion processes. Among the disadvantages of the named systems we can name an
incomplete set of features for dynamic situations modelling system; no support for problem
domain conceptual model engineering and multi-agent models, containing intelligent
agents, design; incomplete multi-agent resource conversion processes problem orientation;
programming user orientation; high retail price.
Multi-agent resource conversion process situational mathematical model requirements were
designed. The model must provide the following functions: dynamic resource conversion
processes modelling; definition of intelligent agent communities, controlling the resource
conversion process; situational approach application.
System development required multi-agent resource conversion process model definition.
The following features of the model were designed:
 Multi-agent resource conversion process main objects;

 Graphical notation;
 System graphs apparatus was applied to hierarchical process structure definition;
 Frame-semantic representation, based on frame-concepts and semantic graphs, was
selected for knowledge representation model, which allowed problem domain
conceptual model definition;
 InteRRaP hybrid architecture was selected as a basis of multi-agent system;
 Multi-agent resource conversion process output mechanism, rule types, intelligent
agent activity algorithm and situational simulation modelling algorithm were designed.
Based on the model and multi-agent resource conversion process system analysis, a
software family of BPsim products was developed. It offers the full list of functional
features, required from a problem-oriented dynamic simulation modelling systems and
implements the following specific features:
 Problem domain conceptual model definition;
 Multi-agent models definition, including both reactive and intelligent agents;
 Multi-agent resource conversion processes problem orientation;
 Balanced scorecard methodology integration;
 Significantly lower retail price.
Simulation, expert, situational and multi-agent modeling integration with object-oriented
approach allowed implementation of new object-oriented multi-agent resource conversion
processes simulation and decision support method, reflected in development of object-
oriented decision support system BPsim.DSS, deployed at companies in Ural region of
Russia.
The mathematical model is based on discrete resource conversion process model. Within its
framework the problem of transition between the knowledge representation, conceptual
model and their technical implementation on relational database level, was solved. This
approach allows Transact-SQL language to be used for problem domain models design,
data and knowledge input, logical output mechanism implementation.
7. Acknowledgment
This work was supported by the State Contract No. 02.740.11.0512.
8. References
Aksyonov K.A., Sholina I.I. & Sufrygina E.M. (2009a). Multi-agent resource conversion
process object-oriented simulation and decision support system development and
application, Scientific and technical bulletin, Vol. 3 (80), Informatics.
Telecommunication. Control, St.Petersburg, pp.87-96.
Aksyonov K.A., Bykov E. A., Smoliy E. F. & Khrenov A. A. (2008a). Industrial enterprises
business processes simulateon with BPsim.MAS, Proceedings of the 2008 Winter
Simulation Conference, Pages 1669-1677
Aksyonov K.A., Spitsina I.A. & Goncharova N.V. (2008b). Enterprise information systems
engineering method based on semantic models of multi-agent resource conversion
processes and software, Proceedings of 2008 IADIS International Conference on
Intelligent Systems and Agents, part of Multi Conference on Computer Science and
Information Systems (MCCSIS), July 22-27, 2008, Amsterdam, Netherlands, pp. 225-
227.
Aksyonov, K.A. & Goncharova N.V. (2006). Multi-agent resource conversion processes dynamic
simulation, Ekaterinburg, USTU
Aksyonov, K.A. (2003). Research and development of tools for discrete resource conversion
processes simulation, DPhil research paper. Ural State Technical University,
Ekaterinburg, Russia
Andreichikov, A.V. & O.N.Andreichikova (2004). Intelligent information systems, Moscow:
Finance and statistics.
Bugaichenko, D.Y. (2007). Development and implementation of methods for formally logical
specification of self-tuning multi-agent systems with timely limitations. DPhil research
paper. St. Petersburg, Russia.
Bugaichenko, D.Y. (2006). Mathematical model and specification of intelligent agent systems
/ Bugaichenko. System Programming. No 2.
Bugaichenko, D.Y. (2005). Abstract architecture of intelligent agent and methods of its
implementation / Bugaichenko, Soloviev. System Programming. No 1.
Emelyanov V.V., Yasinovskiy S.I. (1998). Introduction to intelligent simulation of complex
discrete systems and processes. RAO language. Moscow, Russia
Masloboev A.V. (2009). Hybrid architecture of intelligent agent with simulation apparatus /
Masloboev, MGTU bulletin. Vol. 12, No 1. Murmansk State Technical University,
Murmansk, Russia
Masloboev A.V. (2008). Method of superposed generation and effectiveness estimation for
regional innovation structures / Masloboev, MGTU bulleting. Vol. 11, No 2.
Murmansk State Technical University, Murmansk, Russia
Muller J.P. & M.Pischel (1993). The Agent Architecture InteRRap: Concept and Application,
German Research Center for Artificial Intelligence (DFKI)
Rybina G.V., Paronjanov S.S. (2009). SMIAI system and its application experience for
building multi-agent systems / Rybina, Paronjanov. Software products and systems.
No 4.
Rybina G.V., Paronjanov S.S. (2008a). General model of intelligent agents interaction and
methods of its implementation / XI National Conference on Artificial Intelligence.
http://www.raai.org
Rybina G.V., Paronjanov S.S. (2008b). Modeling intelligent agents interaction in multi-agent
systems / Rybina, Paronjanov. Arificial intelligence and decision making. No 3.
Shvetsov, A.N. (2004). Corporate intellectual decision support systems design models and methods,
DPhil research paper, St.Petersburg, Russia
Wooldridge M. (1995). Intelligent Agent: Theory and Practice, Knowledge Engineering Review,
Vol. 10 (2).
Wooldridge M., Jennings N., Kinny D. (2000). The Gaia Methodology for Agent-Oriented
Analysis and Design. Journal of Autonomous Agents and Multi-Agent Systems, 3. p.
285-312.
6
Intelligence Decision Support

Systems in E-commerce
Petr Suchánek, Roman Šperka, Radim Dolák and Martin Miškus
Silesian University in Opava, School of Business Administration in Karviná
Czech Republic
1. Introduction
The present state of world economy urges managers to look for new methods which can
help to start the economic growth. To achieve this goal, managers use standard as well as
new procedures and tools. Development of information society and so-called new economy
has created a whole new business environment that is more extensive and rapidly changing.
Become a standard the use of modern information and communication technologies (ICT),
which enable faster and cheaper communication and an increase in the number of business
activities. Today in the world's economy, a major role plays electronic business (e-business).
Basic support for the e-business transactions are so-called e-commerce systems. E-commerce
systems became standard interface between sellers (or suppliers and manufactures) and
customers. E-commerce systems can be considered for systems with large geographical
distribution. This follows from their nature, when they are based on the Internet. Statistics
show an increasing interest in cross-border online shopping. This is also dependent on the
efforts of manufacturers and retailers to establish themselves on foreign markets. E-
commerce systems allow them to do it quickly and at relatively low financial cost.
One of basic features of efficient e-commerce is correct definition and description of all
internal and external processes. All the management activities and decision making has to
be targeted to customers´ needs and requirements. The optimal and most exact way how to
obtain and find optimal solution of e-commerce system and its procedural structure is
modelling and simulation. Organizations developing e-business and e-commerce solutions
consider business modelling as a central part of their projects. The latest theoretical and
practical experiences are testaments to how great strategic advantage for organizations
doing business is creation and use of new e-business and e-commerce models. The
importance and necessity of the creation and use of new models is described as the most
recent publications, for example in (Laudon & Traver, 2009) or (Laudon & Traver, 2010) as
well as in older, for example (Barnes & Hunt, 2000). Next to the business model, a big
attention must be paid to models of e-commerce systems from a technical point of view.
(Rajput, 2000)
Expanded scope, creation and use of new models create the need for increased demands on
decision-making. Social and business environment is changing very rapidly and this has an
impact on managerial approaches. ICT became basic decision-making support.
Technological bases of e-commerce systems are ERP and CRM systems, which constitute the
core of enterprise informatics. Now almost all of them contain decision-making support
modules and tools, which fall into the decision support system (DSS) or business
intelligence (BI) category. In this context it is appropriate to ask, what is the difference
between DSS and BI, how can these systems be used in e-commerce, which of these systems
is more appropriate for decision support and what are the conditions that these systems
could be classified as so-called intelligent decision support systems.
On the basis of these facts main goal of this chapter is to present new approaches to the
creation of e-commerce systems models and new approaches to managing e-commerce
systems using modern software tools for decision support.
2. E-commerce system
E-commerce systems are fundamental aids of online shopping. According to (Velmurugan
& Narayanasamy, 2008) e-commerce is defined as an attempt to increase transactional
efficiency and effectiveness in all aspects of the design, production, marketing and sales of
products or services for existing and developing marketplaces through the utilization of
current and emerging electronic technologies. In the globalization era, understanding the
adoption of information communication technology, including e-commerce by developing
countries is becoming important to improve its adoption success. This, in turn, enables
developed countries to trade with developing countries more efficiently.
Generally and simply, e-commerce system can be defined as a web server linked by
company´s information system. Detailed definition of e-commerce system appears from
definition of information system whose basic components are information and
communication technologies and users. Information systems are developed mainly for
management support. Managers are in a way, a special group of users. Information system
for them is on the one hand, a tool to support management activities, on the other hand, the
source of current information describing the current state of managed objects. It is
a principle of feedback, which is one of the fundamental principles of management.
Some authors (in some publications) consider as an e-commerce system only web server that
contains all the necessary functionality (for example (Garcia et al., 2002)). Main model of e-
commerce system based on process-oriented approach is shown for example in (Rajput,
2000). This model can be extended and then we can define main components of e-commerce
systems which are (Fig. 1):
 customers;
 internet;
 web server (web interface);
 CRM (Customer Relationship Management);
 ERP (Enterprise Resource Planning);
 LAN (Local Network Area);
 payment system;
 delivery of goods;
 after-delivery (after-sales) services;
 information systems of cooperating suppliers and customers.
E-commerce systems are developed to support business activities. Customers (buyers) have
their own requirements and corporate managers have to find all the ways, methods and
resources to meet their needs and requirements. Great emphasis must be placed on all
management control systems and systems to support the decision-making processes. The
Intelligence Decision Support System in E-commerce 111
deployment of e-commerce could be seen mainly in the CRM (Customer Relationship

Management), SCM (Supply Chain Management), FRM (Financial Resource Management),
HRM (Human Resource Management), MRP (Manufacturing Resource Planning) and CPM
(Composite Product Mapping).
Fig. 1. E-commerce system

When implementing e-commerce system it is important to link new system with company’s
information system. All changes in e-commerce system will be immediately performed into
the information system to ensure the data consistency. Enterprise Resource Planning (ERP)
is suitable for global operations as it encompasses all the domestic jargons, currency
conversions, diverse accounting standards, and multilingual facilities. ERP software
attempts to integrate business processes across departments onto a single enterprise-wide
information system. The major benefits of ERP are improved coordination across functional
departments and increased efficiencies of doing business. The goal of CRM is to collect
information gained by doing business in a central repository, analyze it, and make it
available to all departments. In many cases, CRM are integrated into ERP, but it is not the
rule. CRM and ERP benefits can be measured and quantified. With usage of ERP, company
can gain following benefits:
 improved alignment of strategies and operations;
 improved productivity and insight;
 reduced costs through increased flexibility;
 supported changing industry requirements;
 reduced risk;
 improved financial management and corporate governance;
 optimized IT spending;
 gained higher ROI (Return on Investment) faster;

 retained top performers;
 provided immediate access to enterprise information.
A major benefit of CRM can be the development of better relations with the existing and
new customers, which can lead to:
 increased sales through better timing due to anticipating needs based on historic trends;
 identifying needs more effectively by understanding specific customer requirements;
 cross-selling of other products by highlighting and suggesting alternatives or
enhancements;
 identifying which of your customers are profitable and which are not.
The condition of an entry to global markets through e-commerce deployment is an
adjustment of the information system to global information and business system
management standards. Management standards are defined for three basic sectors, which
are customer relationship management, supply chain management and operating
management. These sectors can be further refined into information system management,
business processes management, management of logistics and production logistics,
management of human resources, legal rules in relation to international law etc. (Vymětal &
Suchánek, 2009)
Payment systems are often one of the problem areas of e-commerce. It can be seen especially
in the context of cross-border online shopping. There are numerous different payments
systems available for online merchants. These include the traditional credit, debit and
charge card but also new technologies such as digital-wallets, e-cash, mobile payment, e-
checks, cash on delivery. To support the cross-border online shopping development, it is
necessary to provide the customers safe payment environment and to give support to the
international bank clearing. It is one of the tasks of e-commerce system to ensure the
correctness of financial flows e.g. from the sales when converting and storing the business
results to the company’s account information system. Dispatch is supported and controlled
by SCM. SCM is the oversight of materials, information, and finances as they move in a
process from supplier to manufacturer to wholesaler to retailer to consumer. Supply chain
management involves coordinating and integrating these flows both within and among
companies. Supply-chain management, at least in the largest multi-national corporations, is
a global endeavor (Murillo, 2001). More about efficient SCM is discussed in (Šperka, 2010).
Like every area of business in these days, e-commerce is surrounded by a maze of red tape,
rules and regulations. In fact, selling online tends to be worse because of the international
dimension. On the one hand, legislature can help to online shopping and cross-border
online shopping, on the other hand it can scant e-commerce development (Suchánek, 2010a).
Following part of the text is about requirements management. It will be explained why it’s
so important for companies to meet customers’ requirements when developing an e-
commerce system.
3. Customer requirements and management needs

When starting to develop an e-commerce system it is important to think about future
functions of the system from the customer’s point of view. Development process of the most
recent projects initiates with requirements analysis. In terms of e-commerce, customer
requirements can be separated up two groups. The first group of customer requirements
results from exploitation of IS/IT as the main technological support of e-business and e-
commerce environment. In 2003, ANEC Policy Statement on Design for All called upon the
standard-makers to take the following generic consumer requirements into account when
designing, selecting, commissioning, modifying and standardizing ICT systems.
Requirements for IS/IT were summarized as accessibility/design for all, adaptability, child
safety issues, comprehensible standards, consistent user interface, cost transparency, easily
adaptable access and content control, ease of use, environmental issues, error tolerance and
system stability, exportability, functionality of solution, health and safety issues,
information supply for first-time user set-up procedure, interoperability and compatibility,
multi-cultural and multi-lingual aspects, provision of system status information, privacy
and security of information, quality of service, system reliability and durability, rating and
grading systems, reliability of information, terminology. (ANEC, 2005)
The second group of customer requirements is closely associated with business transactions
(Suchánek, 2010a). Customers want to find what they want easily and in short time, to get
sufficient number of information, to place an order easily, payment system to be secured
and failsafe, to get goods in quality of service and in short time, goods to be guaranteed by
sellers (producers) and to get benefits in dependence on a number of purchases. To be
reliable in an uncertain and changing environment companies must be able to respond to
the changes quickly. To obtain it, management needs actual information. The most
important condition of customer satisfaction is feedback. Suppliers and producers have to
monitor market environment and all have to be targeted to the customers. All customer
requirements have to be monitored for ever and company information system with the all
company´ processes have to be formed to ensure quality and rapid processing of the all
customer feedback information, needs and requirements. Feedback information can be
getting by the communication channels which are usually integrated in CRM (Customer
Relationship Management). Feedback is the most important condition of getting
information. If managers want to satisfy all customer requirements, they should:
 get precision information;
 get information in time;
 get information in required form;
 get information in visual form;
 know information they want to.
Managers need besides information:
 to develop the ability to apply information technology in complex and sustained
situations and to understand the consequences of doing so;
 to learn the foundations on which information technology and applications are built;
 and current or contemporary skills.
4. Decision-making processes in e-commerce system environment

When the company, implementing e-commerce system wants to use full potential of new
solution it is necessary to decide according to the relevant information. E-commerce systems
produce amount of raw data. Why not to use these data to improve the decision-making?
Decision Support System (DSS) is an umbrella term used to describe any computer
application that enhances the user’s ability to make decisions. More specifically, the term is
usually used to describe a computer based system designed to help decision makers use
data, knowledge and communications technology to identify problems and make decisions
to solve those problems. DSS systems can be separated into seven broad categories namely
Communications Driven DSS, Data Driven DSS, Document Driven DSS, Knowledge Driven
DSS, Model Driven DSS, Spreadsheet based DSS and Web-based DSS. Communications
Driven DSS is a type of DSS that enhances decision-making by enabling communication and
sharing of information between groups of people. Data Driven DSS are a form of support
system that focuses on the provision of internal (and sometimes external) data to aid
decision making. Document Driven DSS are support systems designed to convert
documents into valuable business data. Knowledge Driven DSS are systems designed to
recommend actions to users. Model Driven DSS support systems incorporate the ability to
manipulate data to generate statistical and financial reports, as well as simulation models, to
aid decision-makers. Spreadsheet based DSS offer decision-makers easy to understand
representations of large amounts of data. Web-based DSS system is operated through the
interface of a web browser, even if the data used for decision support remains confined to a
legacy system such as a data warehouse. (Velmurugan & Narayanasamy, 2008)
Decision making processes need combination of skills, creativity, recognition of the
problems, and lucidity of judgment, determination, and effective implementation in
operational plan. Generally, decision making process has five stages: (Harrison, 1998)
 problem determination (definition of objectives);
 collection of information (identification of alternatives);
 choosing of optimal decision;
 implementation of a decision;
 evaluation of decision.
To adopt a right decision, managers have to get correct information in right time. In
connection with e-commerce source system of data set is extended. With a view to
minimization of failure during the domestic and especially cross-border online selling, it is
necessary to allow for many factors. Besides typically economic indicators, source
information of management systems have to be for example legislature, culture, conventions
etc. (Fig. 2)
Decision making processes are also proceed on the side of customers. The customers´
decision-making process is the process they go through when they decide to purchase
something. (Olsen, 2003) Research suggests that customers go through a five-stage decision-
making process in any purchase:
 need recognition and problem awareness;
 information search;
 evaluation of alternatives;
 purchase decision;
 post-purchase evaluation.
Managers´ decisions should lead to make the customers´ decision-making process easier. All
decision making processes have to be targeted to the customers and their needs and
requirements. Customers´ needs and requirements are usually different in a number of
countries. This fact is always a cause of unsuccessfully cross-border online selling
transactions. Only the way leading to reduce the number of unsuccessfully cross-border
online selling transactions is an optimal management system making use of all necessary
source information. To obtain an efficient decision-making, there are used mathematical
models of allocation processes (Bucki, 2008). More about mathematical model of e-
commerce simulation example is written at the end of this chapter. Supranational character
of e-commerce systems evokes the need to process an extensive set of information and urges
the managers to look for the new methods leading to maintenance and improvement of
position in domestic and especially foreign markets. This is possible only with the aid of
modern information technologies. Current trend is oriented to the development and usage
of systems with business intelligence tools. (Suchánek et al., 2010b)
Fig. 2. Management system and its source information areas
5. E-commerce systems modelling to assess the added value

Modelling and system design techniques are currently widely required. This is caused by
information technologies support needed to achieve and sustain necessary business
flexibility imposed by market fluctuations. There are two substantial ways to use the model.
During the system design, the model serves as a conceptual description of the future system.
After the new system is put into operation, the management can use it also for decision
support. Typical management control loop incorporates the measurement of the controlled
system outputs (“what has happened” - facts) which are evaluated and compared with the
company objectives (targets). The controller – manager then takes corrective action (“what
should or must be done”) in order to reach the specified targets. In case of decision support
by simulation the outputs of the controlled subsystem and the data concerning company
environment have to be included in the model structure. The results of simulation are
presented to decision maker – manager for evaluating possible decision alternatives. While
conceptual modelling forms the base of new system design, the simulation modelling can be
seen as operations support. This is why both ways of modelling are important for achieving
flexibility of data processing in particular, the company management effectiveness in
general.
E-commerce systems to work properly should support and cover all company’s core
processes. Correct processes identifying, mapping and implementing into e-commerce
system is the main condition of successful business. E-commerce dramatically and
strategically changes traditional business models. Companies are now pursuing more
intensive and interactive relationships with their business partners: suppliers and
customers. Competitive conditions and pressures on global market are forcing companies to
search for strategies of streamlining the entire value chain. To compete effectively,
companies must structurally transform its internal and external processes. These goals could
be reached by simultaneous renovation of business processes and implementation of
electronic business solutions. Business Process Reengineering (BPR) is an organizational
method demanding radical redesign of business processes in order to achieve more
efficiency, better quality and more competitive production. BPR has become one of the most
popular topics in organisational management creating new ways of doing business. Many
leading organisations have conducted BPR in order to improve productivity and gain
competitive advantage. However, regardless of the number of companies involved in re-
engineering, the rate of re-engineering projects success is less than 50%. Some of the
frequently mentioned problems related to BPR include the inability to accurately predict the
outcome of a radical change, difficulty in capturing existing processes in a structured way,
shortage of creativity in process redesign, the level of costs incurred by implementing the
new process, or inability to recognize the dynamic nature of the processes. An e-commerce
model generally means the adoption of company’s current business model to the Internet
economy. Main purpose of developing and analysing business models is to find revenue
and value generators inside reversible value chain or business model's value network. BPR
in 90-s has been focused on internal benefits such as a cost reduction, downsizing of
company and operational efficiency which is rather tactical then strategic focus. Nowadays,
e-business renovation strategies put their focus on the processes between business partners
and the applications supporting these processes. These strategies are designed to address
different types of processes with the emphasis on different aspects: customer relationship
management (CRM), supply chain management (SCM), selling-chain management and
enterprise resource planning (ERP).
It is well known that e-commerce might bring several advantages to the company. However,
existing practical business applications have not always been able to deliver the benefits
they promise in the theory. Prior to adopting e-business, companies need to assess the costs
needed for setting up and maintaining the necessary infrastructure and applications and
compare it with the expected benefits. Although the evaluation of alternative solutions
might be difficult, it is essential in order to reduce some of the risks associated with BPR
projects.
Before implementing the e-commerce model we should know if the designed model works
properly. For this verification process is often used the simulation. Simulation has an
important role in modelling and analysing the activities in introducing BPR since it enables
quantitative estimations of influence of the redesigned process on system performances.
Simulation of business processes represents one of the most widely used applications of
operational research as it allows understanding the essence of business systems, identifying
opportunities for change, and evaluating the impact of proposed changes on key
performance indicators. The design of business simulation models that will incorporate the
costs and effects of e-commerce implementation and will allow for experimentation and
analysis of alternative investments is proposed as a suitable tool for BPR projects. Some of
the benefits can be directly evaluated and predicted, but the others are difficult to measure
(intangible benefits).
5.1 Process oriented approach

Often used approaches to model the e-commerce systems are process and business oriented
systems. Process oriented model will be used to describe the basic structure and data flow.
Process modelling is one of the most cost-effective and rewarding ideas to come along in
years. Many different techniques can be used for modelling business processes in order to
give an understanding of possible scenarios for improvement. Flowcharting, IDEF0, IDEF3,
Petri Nets, System Dynamics, Knowledge-based Techniques, Activity Based Costing and
Discrete-Event Simulation are only some examples of business process modelling
techniques widely used. There are also many software tools on the market using these
modelling techniques. In terms of methods, we can use three ways of modelling. E-
commerce systems can be modeled using process-oriented, value-oriented and/or multi-
agents-oriented approaches.
The goal for process-oriented modelling is a specification of activities and their structure.
Each activity can be looked upon as a process and decomposed into further ones until the
set of single activities is reached. The activities do not come into being or exist
independently from the environment. They are instigated by defined external or internal
impulses or reasons. The basic elements of a process model include usually (Řepa, 2006):
 process;
 activity;
 initiative or instigation;
 relations among processes and activities.
Process oriented approach can be used for many purposes. As an example we can specify
the model describing the activities taking place during the execution of transactions in the
on-line shopping. A diagram of existing process could be seen on Fig. 3.
In many cases customers find and select a product in web-stores and then buy it later in
traditional (physical) stores. The technique should be capable of representing one of more of
the following modelling perspectives: functional (represents what activities are being
performed), behavioural (represents when and how activities are performed), organizational
(represents where and by whom activities are performed) and informational (represents the
informational entities – data). (Bosilj-Vukšic et al., 2010)
After-sales
services
Product and/
Product (-s)
Customer or services Cart (basket)
selection
catalog
Login
Registration
Back to catalog or guide CRM
of ordering ERP
False
Delivery of goods Registered and logged

user
Guide of ordering
Method of delivery
Method of payment
True Data checking
Store inventory Other additional information
Payment system
(removal from storage)
Fig. 3. Process model of e-commerce system
5.2 Value-chain oriented approach

Business oriented model works with value chains of business processes. It is used to define
the methods and possibilities of e-commerce system. A Value Chain perspective, developed
and introduced by Michael Porter (Porter, 1980) can be arranged as a series of input-output
business processes with resource flows between them. (Dunn et. al., 2005), (Hruby, 2006).
Value chain models deal with the participants of the value flow (customer, enterprise etc.);
define the objects of value flow (like goods, services, cash) and types of economic events
accomplishing the actual value flow itself. The basic difference to traditional business
process models explains how processes happen by means of general process description
concepts such as the activity, entity, process state, is that value chain models explain why
processes happen in terms of the added value to the participants’ entrepreneurial goals.
5.3 Multi-agents oriented approach

Another attitude to this activity brings the advantages of the model based on multi-agents
oriented approach to model the e-commerce system and to validate it with simulation.
(Vymětal, 2009) gives perfect example for using multi-agents simulation in business
processes of a corporation. (Šperka, 2010) explains the advantages of multi-agents
simulation methods for the increasing of the competitiveness of the enterprises. A multi-
agent system is a system composed of multiple interacting intelligent agents. Multi-agent
systems can be used to solve problems which are difficult or impossible for an individual
agent or monolithic system to solve. Examples of problems which are appropriate to multi-
agent systems research include online trading, (Rogers, 2007) disaster response, (Schurr,
2005) and modelling social structures. (Ron & Naveh, 2004) The agents in a multi-agent
system have several important characteristics: (Wooldridge, 2004)
 autonomy - the agents are at least partially autonomous;
 local views - no agent has a full global view of the system, or the system is too complex
for an agent to make practical use of such knowledge;
 decentralization - there is no designated controlling agent (or the system is effectively
reduced to a monolithic system). (Panait & Luke, 2005)
Typically multi-agent systems research refers to software agents. However, the agents in
a multi-agent system could equally well be robots, (Kaminka, 2004) humans or human
teams. A multi-agent system may contain combined human-agent teams. Multi-agent
systems can manifest self-organization and complex behaviors even when the individual
strategies of all their agents are simple. Model based on multi-agents-oriented approach can
be used to define the methods and possibilities of e-commerce systems simulation.
As above, in the area of e-commerce, multi-agent system (MAS) consists of various number
of software components, called agents. An agent is an entity that is designed to undertake
autonomous action on behalf of a user in pursuit of his desired goal. This implies that agents
are intelligent and autonomous objects that are capable of behaving like a human being.
Therefore agent technology is a suitable means for expanding business activities and saving
cost. Agents possess some form of basic intelligence to allow decision-making or have
a structured response to a given set of circumstances. These behaviour patterns are acquired
from or given to the agent by the user, this enables the agent to be capable of flexible actions.
Allowing it to exhibit good oriented and opportunistic behaviour to meet it’s objective.
Agents exist in an environment and can also respond to changes therein. Complex tasks can
be broken down into smaller components that can each be the responsibility of an agent.
They concentrate on their individual components, find optimal solution and combine their
efforts by inter-agent communication. (Folorunso et al., 2006)
6. Parameters for describing models

Models can be described by a number of parameters which are often called key indicators.
There are a number of indicators that can be measured and calculated and by which we can
make a description of current e-commerce system performance. If we consider the sales
system in the form of an internet shop (e-shop), fundamental question is what are the most
important indicators and key numbers that can be used to measure the success or failure of
internet-based sales, rather e-commerce system, and how can they be measured?
Measurement is a very important thing, because in general, what can´t be measured, can´t
be improved. In this context, objective will be the systemization of measurable key
indicators, which are input values of DSS and BI.
Within e-commerce systems, measurements can be divided into the following types:
 performance measurement - measurement of technical parameters (for example web
page load time, speed of processing user commands, speed reporting, etc.);
 usability testing – usability testing is done in laboratory conditions (for experienced
users are set various tasks);
 measurement of the success – measurement of the success should answer the question
how e-commerce system helps to meet company objectives. Outcomes of this
measurement are used primarily to optimize and control.
Source data can be divided into next groups:
 operational characteristics – on the web interface there are, for example, number of
displayed web pages, number of website visitors, in connection with the information
system there can be, for example, number of successfully completed transactions,
number of failed transactions (may be associated with system disturbances), etc.;
 customer data – data relate primarily to demographic characteristics of visitors and
their preferences (city listed in the order, city listed in the query or demand, gender of
customer, etc.);
 transactional data – data directly related to the sale of goods or services. These data are
the basis for financial analysis. (average order value);
 data from other sources - other data relating to consumer behaviour in the website.
6.1 Key indicators of websites success and performance

One of the technologies used for the web indicators measurement is so called clickstream.
(Sweiger et al., 2002) The result of clickstream analysis may be two groups of characteristics.
Outputs belonging to the first group are data representing the operating characteristics (for
example number of pages viewed, number of unique IP addresses, etc.), the second group
consists of characteristics of a commercial nature (cross-selling analyses, abandonment of
the shopping cart, number of the orders, etc.). Other outputs, which are the numbers of
orders, inquiries, etc. are processed and analyzed in the context of CRM and ERP. In these
systems some specific business cases and business activities are already processed.
Generally, on the website, measurements can be realized with help of traffic volume
measurement from panels, TCP/IP packet sniffing, direct server measurement, browser
based measurement, server-based measurement or internet traffic measurement.
The first group of parameters includes the key indicators for the measurement of the e-
commerce system success. In this case, the input data are the number of days (time period),
site visits (unique visits), number of demands received from the web, number of orders
received from the web, number of users who have made an order, and data from CRM and
ERP (amount of orders – revenue, margin orders).
Based on these input data, there can be made the calculations, from which we can obtain
sales (total orders), margin (after deduction of all costs), conversion ratio of demands,
conversion ratio of orders, average value per order, average returns per customer, average
number of order per customer, average margin of one order, average margin per customer,
average returns per visit, average margin per demand, and average margin per visit. These
indicators are used for the basic analysis and are very important for the management
activities related to the sale, planning and finance.
Web interface is usually a part of a CRM system that is connected with the ERP. CRM and
ERP are the information systems that are at the core of the enterprise informatics. The
indicators from this group belongs to the area of IS/IT. The basic indicators for the proper
operation of the information system are: system availability, average response time,
breakdown intensity, breakdown rate, average download time, failure rate.
The integral part of the enterprise information system is created by its users at all levels of
the corporate activities, and in particular, of the management. Each user carries out various
activities, while each of them constitutes interference in the system. In this context, there are
indicators that can be included in the administrative field, partly in the area of security and
personnel. In this group of indicators, there can be included: number of correct user
intervention, number of incorrect user intervention (for example, it may be related to poor
secure system, inappropriate user interface, incompetence users), number of system
administrator intervention (corrections system) and number of users involved in processing
by one commercial transaction.
Another important area is the logistics. The logistics is a channel of the supply chain which
adds the value of time and place utility. The basic indicators tracked in this area are: product
availability, number of successfully delivered products, number of unsuccessfully delivered
products (for example customer entered incorrect address or error occurred at the side of
vendor or carrier), average length of warranty service, storage costs, cost to deliver goods,
average delivery time of goods, number of cooperating suppliers offering the same goods.
The Internet has allowed the development of the new methods of marketing that provide
very important information useful for the management support and the planning. In this
area, the important key indicators are: number of unique customers, average visit frequency,
number of repeat customers, margin per customer, number of first buyer, average sales per
repeat customers, average order value per repeat customer, market share, percentage share
of new vs. returning visitors, average conversion rate, average time spent on the website,
average number of pages displayed per visit, percentage share of returns (one page visits),
average number of clicks on adverts. One of the key methods of e-marketing is a campaign.
For example, in the campaigns we can trace percentage share of visits according to the type
of campaign and percentage share of conversion rate according to the type of campaign.
These values are the basis for the determination of index of campaign quality.
The e-commerce systems are implemented with a goal to expanse the sales channels, and
thus to increase the profits. In this context, the important general indicator is the return on
investments (ROI). To calculate ROI in the web projects, we can use, inter alia, an online
calculator on the website. In this case, the source data can include: number of days (time
period), regular fixed cost (the cost of webhosting, salary, etc.), cost per customer
acquisition, the current daily visit site, the current conversion ratio of orders, the percentage
increase of conversion ratio due to investments, percentage increase of average order value
due to investments, estimated average order value increase after, the current average
margins, the percentage increase of average margins due to investment, and estimated
average margin value increase after.
Using these data, we can calculate: number of visitors, number of orders, yields, margins,
fixed costs, direct cost of acquiring visitors, profit, profit per days, and return on initial
investments.
6.2 Data in e-commerce systems

E-commerce system produces a huge amount of data collection. Data has potentially great
value. Data must be accepted, processed and appropriately presented. This is the main
prerequisite for successful management. To adopt a right decision, managers have to get
correct information in right time. In connection with e-commerce and especially online
selling, source system of data set is extended. With a view to minimization of failure during
the domestic and cross-border online selling, it is necessary to allow for many factors.
Besides typically economic indicators, source information of management systems have to
be for example legislation, culture, conventions etc. (see Fig. 2)
Data can be processed by the using of ICT and then may be obtained from them valuable
information for decision-making. We can find new customer segments, analyze trends or
uncover new business opportunities. Business intelligence solutions are used for the
purposes of advanced data analysis and search for hidden dependencies. Business
intelligence aims to support better business decision-making and also can be helpful in
process of retaining existing customers and acquiring new customers. In next part of this
chapter will be mentioned more about BI.
7. Business intelligence
It's common knowledge that Decision Support Systems are a specific class of computerized
information system that supports business and organizational decision-making activities.
A properly designed DSS is an interactive software-based system intended to help decision
makers compile useful information from raw data, documents, personal knowledge, and/or
business models to identify and solve problems and make decisions. But what about so-
called intelligence decision support system? What should this type of systems meet the
requirements? From our perspective, an intelligent decision support system is a system
containing BI tools. Business Intelligence is a term that refers to the sum total, or effect, of
gathering and processing data, building rich and relevant information, and feeding it back
into daily operations so that managers can make timely, effective decisions and better plans
for the future. Generally business intelligence brings to managers a quite number of
advantages. The advantages enjoyed by market leaders and made possible by business
intelligence include the high responsiveness of the company to the needs of its customers,
recognition of customer needs, ability to act on market changes, optimization of operations,
cost-effectiveness, quality analysis as the basis for future projections, the best possible
utilization of resources etc.
Business intelligence is oriented to the management needs and decision making support.
Optimal setting of control processes is a prerequisite of the planned and expected aims.
Business processes are the collections of activities designed to produce a specific output for
a particular customer or market. It implies a strong emphasis on how the work is done
within an organization, in contrast to a product's focus on what. To do right decisions,
managers need information. Data that is relevant to a business decision may come from
anywhere. The most important sources of data include: (Suchánek, 2010a)
 Master data - this is data collected (usually once and once only) to define the entities in
an e-business system (customer file, product file, account codes, pricing codes, etc...).
Scores of the time we can meet with term of Master Data Management (MDM). MDM
comprises a set of processes and tools that consistently define and manage the non-
transactional data entities of an organization (which may include reference data).
 Configuration data - as the term implies this is data defining the nature of the system
itself. The system is configured to reflect the nature and needs of the business.
 Operations data (OLTP - Online Transaction Processing) - also known as activity. This
data is generated by daily business activities such as sales orders, purchase orders,
invoices, accounting entries, and so on. OLTP refers to a class of system that facilitates
and manages transaction-oriented applications, typically for data entry and retrieval
transaction processing.
Information systems (OLAP - Online Analytical Processing) - these are sophisticated
applications collecting information from various internal and external sources to analyze
data and distill meaningful information. OLAP software is used for the real-time analysis of
data stored in a database. The OLAP server is normally a separate component that contains
specialized algorithms and indexing tools to efficiently process data mining tasks with
minimal impact on database performance.
Almost all requisite data for the decision making support in e-commerce comes from CRM
and ERP systems. Business intelligence is closely related to data warehousing. Data has to
be processed (data selection, data analysis, data clearing etc.) and sent in right time and in
required form to the competent person usually acting in management system. (Fig. 4)
Obtained data are basis for decision making support at all levels and kinds of management.
Staging database
CRM
(primary layer of DW)
Communication Operational
channels CRM Management
Business (www, phone, branch
environment office, dealers, Analytical SCM (Supply Chain Management)
promotional flyer, and CRM (Customer Relationship Management)
so on) CRM FRM (Financial Resource Management)
HMR (Human Resource Management)
MRP (Manufacturing Resource Planning)
CPM (Composite Product Mapping)
Data warehouse
ETL ...
Orders & Contracts & Payments
ERP Reporting
Applications Ad-hoc reporting
Accounting Planning
OLAP
Inventory control Project management SharePoint
Production control Claim control MS Excel
BIRT Project
Order management Receivable management The Pentaho BI Project
Human resource Agata Report
Property administration
management More others
Fig. 4. Business intelligence system. Source: Suchánek, 2010a.

Data processing is supported by ETL (Extraction, Transformation and Loading). The ETL
process is also very often referred to as Data Integration process and ETL tool as a Data
Integration platform. The terms closely related to and managed by ETL processes are: data
migration, data management, data cleansing, data synchronization and data consolidation.
(ETL-Tools.Info, 2009)
The purpose of the Data Warehouse in the overall Data Warehousing Architecture is to
integrate corporate data. Data warehousing evolved through 1990s to environments that
typically featured batch-oriented “stocking” and “restocking” of the data warehouse,
primarily for purposes of after-the-fact reporting and analysis. (Simon & Shaffer, 2001) The
amount of data in the data warehouse is massive. Data is stored at a very granular level of
detail. For example, every sale that has ever occurred in the organization is recorded and
related to dimensions of interest. This allows data to be sliced and diced, summed and
grouped in unimaginable ways. Recent technological advancements in data warehousing
have been contributing to the emergence of business intelligence useful for managerial
decision making. (Taniar, 2009)
BI applications are for example SharePoint, MS Excel, BIRT Project, The Pentaho BI Project,
Agata Report and many others (some of applications can be interconnected). These
applications are fundamental interface between system and managers. With the aid of
shown application users can choose and manipulate a wide variety of visual
representations, including map-based data displays to review business intelligence
reporting results geographically, multidimensional scatter plots to view data statistically,
and bar charts, pie charts, line graphs, profile charts and more. At present, when the world
economic conditions are not affable, companies look for all information which would help
them to start economic growth. In this respect e-commerce companies have recently started
to capture data on the social interaction between consumers in their websites, with the
potential objective of understanding and leveraging social influence in customers' purchase
decision making to improve customer relationship management and increase sales. (Young
& Srivastava, 2007)
Business intelligence systems are able to provide the managers quite a number of statistics
dealing with customers and their environment. As important customer statistics can be
considered, for example, matching sales revenues with site visitor activity, by week and
month, in total and by product line, matching weekly and monthly sales with site visitor
activity over time (trend analysis), in total and by product line, matching sales revenues
with site visitor activity, by day and hour, in total and by product line (to measure the
effectiveness of advertising campaigns), matching sales revenues with site visitor activity
from main referrers, by week and month, in total and by product line. Where the referrer is
a search engine, also matching the search query with sales revenues. These statistics respond
the managers to questions:
 Who did buy?
 How much did they buy?
 When did they buy?
 What did they buy?
 From where customers arrived at the site?
 In which region customer are located?
 How they arrived at the site (e.g. by what search engine query)?
 From which page customer entered the site?
 What is their path through the site?
 From which page customer left the site?
 On a weekly and monthly basis and the trends, over time.
Advantages and benefits of business intelligence in the sphere of e-commerce can be
summarized as:
 Business intelligence gives any firm the specific view of corporate data that is required
for progress (quickly access sales, product and organizational data in any database).
 In sales and marketing, business intelligence offers new tools for understanding
customers’ needs and responding to market opportunities
 By providing financial planners with immediate access to real-time data, Business
Intelligence builds new value into all financial operations including budgeting &
forecasting.
 Business intelligence supports decision-making with automatic alerts and automatically
refreshed data.
 Business intelligence provides performance monitoring for accelerated action and
decision making.
 Business intelligence makes companies possible to receive and process data from cross-
border business activities (above all from cross-border online shopping).
 Business intelligence can bring to companies competitive advantage.
Besides data from business intelligence systems, companies can use services of the
consultation companies. There are many consultation companies providing information and
doing analyses. Correct source information are necessary, but usually have to be paid for.
There are many sources providing statistical data on the Internet, but these are usually not
adequate. Managers can acquire correct information from the many sources. (Molnár, 2009)
Following sources supply the attested information usable for manager´s decision making in
the global markets. Foreign stock market information can be found on (for example):
 http://www.bloomberg.com;
 http://www.dowjones.com;
 http://www.nyse.com;
 http://www.nasdaq.com;
 http://www.reuters.com;
 http://money.cnn.com;
 http://www.imrworld.org/;
 http://finance.yahoo.com/;
 http://www.marketwatch.com/;
 http://www.indiainfoline.com/.
Marketing information can be found on (for example):
 http://www.formacompany.ie;
 http://reports.mintel.com/;
 http://www.eiu.com;
 http://www.frost.com;
 http://www.adlittle.com;
 http://www.datamonitor.com;
 http://www.euromonitor.com/.
Law and legislative information can be found on (for example):
 http://www.iblc.com/home/;
 http://www.iblc.com/;
 http://www.ibls.com/;
 http://www.enterweb.org/law.htm;
 http://euro.ecom.cmu.edu/resources/elibrary/ecllinks.shtml.
Properly implemented BI tools into the DSS in conjunction with multi-agents-oriented
approach enable the implementation of simulation methods. Simulation can be described as
a technologically and logically the highest level of decision-making support. DSS allows
simulation can fit neatly into the category of intelligent DSS.
The reasons for the introduction of simulation modelling into process modelling can be
summarized as follows (Bosilj-Vukšic et al., 2010):
 simulation enables modelling of process dynamics;
 influence of random variables on process development can be investigated;
 anticipation of reengineering effects can be specified in a quantitative way;
 process visualization and animation are provided;
 simulation models facilitate communication between clients and an analyst.
In this context, in the following section, we will present an example of simulations that can
be carried out within the e-commerce system.
8. Simulation of e-commerce systems

Simulation is generally defined as a method to analyze the behavior of complex systems by
monitoring the behavior of the model (currently in particular computer model). Simulation
system can be described as intelligent decision support systems. Simulation system must
take into account all the areas described in the preceding sections. Simulation system should
allow to enter input parameters which are analyzed by the system. The resulting output
parameters describe the state of the system under the conditions of entry.
8.1 Control loop as the default model of simulation system

A suitable way can be a system simulation based on the control loop (Fig. 5).
Fig. 5. Control loop

Key to figure: w(t) – desired value; e(t) – controller error (w(t) – value of feedback); u(t) –
manipulated variable; y(t) – process variable (output value); v(t) - disturbances
Diagram in Figure 5 can be used as a starting model for a generic model of e-commerce
system (Fig. 6). A similar, much simpler model can be found for example in (Wolf, 2006).
Model in Figure 6 can be used to create realistic simulation models based on software
support. Input variables (desired value) represent the input data of simulation. This data
can be entered by the user manually or automatically generated for example by genetic
algorithms when the user only sets the initial conditions (criteria).
8.2 Integration of simulation into the management structure

For a comprehensive description of the architecture and design of simulation system three
layers can be used (Fig. 7). There is a layer of management control activities primarily
focused on real activity. The default values used in the simulations as a comparative are
contained in the plan. Real regulatory interventions are implemented after their verification
in the simulation layer. In the layer of simulations we can perform simulations based on all
types of models (process oriented, value-chain oriented, multi-agents). Assuming the correct
model and mathematical description, this system may help managers to test some of the
assumptions on which it can be effective decisions.
As we can see in Figure 7, simulation system should be integrated into the overall
management system. Whereas, for the purposes of simulation software tools are used,
integration of simulation into the management system is becoming increasingly easier.
Economic system simulations are used increasingly. Reason is the advantages of simulation,
which can be facts, that simulations are able to provide users with practical feedback when
designing real world systems; simulations permit system designers to study a problem at
several different levels of abstraction and/or simulators can be used as an effective means
for teaching or demonstrating concepts to students.
Control Actuating
Disturbances Disturbances
deviation signal
Desired value Controlled value

Differential component Regulator Controlled member
- Production - Production
- Management Management E-commerce system value
value
- Profit
- Corporate strategy - Corporate strategy - Architecture - Profit
- Financial indicators definition - ERP - Return on
- Return on
- Analyses - Corporate strategy - CRM investment
investment
- Legislation reengineering - Hardware - Other
- Other
- Administrator IS/IT - Investment Components economic
economic
- Accounting - Human resource - Software indicators
indicators
management components - Functionality
- Functionality Comparison of the - Co-operation with - Services - Dependability
- Dependability current stage, desired customers, suppliers - Human Factor - Performance
- Performance values and controlled and business partners - Processes - Availability
- Availability values. - Support to IS/IT - Financial flow - Security
- Security
- Process management - Logistics - Stability
- Stability
- Marketing - User-
- User-
friendliness friendliness
Auxiliary feedback (the fundamental information

about an actual stage of e-commerce system.
Measuring element
Management
Economics Competent
IS/IT department
department persons
- Number of proper closured business
transactions
- Number of unsuccessful business transactions
- Return on investments
- Time of incidental system breakdown
- Number of conversions on e-shop web sites
- Fruitfulness of SEO or PPC
- Number of customers´ feedback
- Number of people failures
- Profit figure
- Quite a few of others
Fig. 6. Generic e-commerce model
9. Conclusion
E-commerce systems are fundamental aids of online shopping. These systems are focused
on customer needs and requirements. E-commerce systems are large systems and produce
a huge amount of data collection particularly data related to the behavior of customers. Data
has potentially great value for management and decision making support. Data must be
accepted, processed and appropriately presented using appropriate tools. Only modern
software tools can provide data quickly and with the required quality. Modern software
tools for processing large data collections are BI and DSS systems. Development of e-
commerce systems places increasing emphasis on the need to create models of these
systems. The modelling of e-commerce systems can be effectively done by use of a process-
oriented approach, value-oriented approach or approach based on multi-agents systems.
The main difference between process modelling and value chain modelling is that process
modelling specifies "How" a process is realized and implemented, while value chain
modelling specifies "Why" the process occurs in terms of added value to the process
participants. Multi-agents approach is primarily used to support simulation.
Simulations more often help the managers implement viable plans and adequate
management activities. Simulation system can be described as intelligent decision support
systems. Important is that the simulation system would be integrated into the management
system as an automatic or semiautomatic system.
Fig. 7. The model links management layers, real activities and simulations
10. Acknowledgment
Chapter's content is based on the outputs of the project SGS/24/2010 - Use of BI and BPM to
support effective management (project is carried out at The Silesian University in Opava,
School of Business Administration in Karviná – Czech Republic).
11. References
Agrawal, P. (2006). E-Business Measurements and Analytics: (Measuralytics). iUniverse, Inc., 91
p., ISBN 978-0595398386, Bloomington, USA
ANEC. (2005). Consumer Requirements in Standardisation relating to the Information Society.
05.05.2009, Available from http://www.anec.org/attachments/it008-03rev1.pdf
Arai, T., Kragic, D. (1999). Variability of Wind and Wind Power, In: Wind Power, S.M.
Muyeen, (Ed.), 289-321, Scyio, ISBN 978-953-7619-81-7, Vukovar, Croatia
Barnes, S., Hunt. B. (2000). E-Commerce and V-Business: Business Models for Global Success,
Butterworth-Heinemann, ISBN 978-0750645324, Oxford, United Kingdom
Bosilj-Vukšič, V., Stemberger, M., I., Jaklič, J. (2010). Simulation Modelling to Assess the Added
Value of Electronic Commerce. Available from http://bib.irb.hr/datoteka/
89704.MIPRO-Vesna-Bosilj.pdf
Bucki, R. (2008). Mathematical Modelling of Allocation Processes as an Effective Tool to Support
Decision Making. Information and Telecommunication Systems, Polish Information
Processing Society, The Beskidy Group, Academy of Computer Science and
Management, No. 17, Bielsko-Biała, Poland
Dunn, C., L., Cherrington, J., O., Hollander A., S. (2005). Enterprise Information Systems. A
pattern based approach. McGraw-Hill, 397 p., ISBN 0-07-240429-9, New York, USA
ETL-Tools.Info. (2009). Definition and concepts of the ETL processes. 15.11.2009, Available from
http://etl-tools.info/en/bi/etl_process.htm
Folorunso, O., Sharma, S., K., Longe, H., O., D., Lasaki, K. (2006). An Agent-based Model for
Agriculture E-commerce System. Information Technology Journal 5(2). 2006, ISSN:
1812-5638, pp. 230-234, Available from http://www.docsdrive.com/
pdfs/ansinet/itj/2006/230-234.pdf?sess=jJghHkjfd76K8JKHgh76JG7FHGD
redhgJgh7GkjH7Gkjg57KJhT
Garcia, F.J., Paterno, F., Gil, A.B. (2002). An Adaptive e-Commerce System Definition. Springer,
Heidelberg, 2002, ISBN 978-3-540-43737-6, Berlin, Germany
Harrison, E. F. (1998). The Managerial Decision-Making Process. Cincinnati: South-Western
College Pub. 576 p., ISBN 978-0395908211, Boston, Massachusetts, United States
Hruby, P. (2006). Model Driven Design Using Business Patterns. Heidelberg: Springer
Verlag, ISBN 978-3-540-30154-7, Berlin, Germany
Kaminka, G. A. (2004). Robots are Agents, Too! AgentLink News, pp. 16–17, ISBN: 978-81-
904262-7-5, New York, USA
Laudon, K., Traver, C.G. (2009). E-Commerce 2010 (6th edition), Prentice Hall, ISBN 978-
0136100577, New Jersey, USA
Laudon, K., Traver, C.G. (2010). E-Commerce 2011 (7th edition), Prentice Hall, ISBN 978-
0136100577, New Jersey, USA
Lima, P., Bonarini, A., Mataric, M. (2004). Application of Machine Learning, InTech, ISBN 978-
953-7619-34-3, Vienna, Austria
Li, B., Xu, Y., Choi, J. (1996). Applying Machine Learning Techniques, Proceedings of ASME
2010 4th International Conference on Energy Sustainability, pp. 14-17, ISBN 842-6508-
23-3, Phoenix, Arizona, USA, May 17-22, 2010
Molnár, Z. (2009). Internetové informační zdroje pro strategické plánování. International
Conference Internet, Competitivenes and Organisational Security in Knowledge Society.
Tomas Bata University in Zlin, ISBN 978-80-7318-828-3, Zlín, Czech Republic
Murillo, L. (2001). Supply chain management and the international dissemination of e-
commerce. Industrial Management & Data Systems. Volume 101, Issue 7, Page 370 –
377, ISSN 0263-5577, San Francisco, California, USA
Olsen, H. (2003). Supporting customers' decision-making process. [on-line] Retrieved 09.12.2009.
URL: <http://www.guuui.com/issues/02_03.php >
Panait, L., Luke, S. (2005). Cooperative Multi-Agent Learning: The State of the Art.
Autonomous Agents and Multi-Agent Systems, pp 387-434, [on-line] Retrieved
January 3, 2011, <http://cs.gmu.edu/~eclab/papers/panait05cooperative.pdf>
Porter, M. (1980). Competitive Strategy: Techniques for Analyzing Industries and
Competitors. Free Press, 398 p., ISBN 978-0684841489, New York, USA
Rajput, W. (2000). E-Commerce Systems Architecture and Applications, Artech House
Publishers, ISBN 978-1580530859, London, United Kingdom
Řepa, V. (2006). Podnikové procesy: Procesní řízení a modelování. Grada Publishing, 288 p., ISBN
80-247-1281-4, Prague, Czech Republic
Rogers, A., David, E., Schiff, J. and Jennings, N. R. (2007). The Effects of Proxy Bidding and
Minimum Bid Increments within eBay Auctions. ACM Transactions on the Web, 1
(2). article 9-(28 pages).
Simon, A., Shaffer, S. (2001). Data Warehousing and Business Intelligence for E-commerce.
Academic press, 320 p., ISBN 1-55860-713-7, London, England
Sun, R., Naveh, I. (2004). Simulating Organizational Decision-Making Using a Cognitively

Realistic Agent Model, Journal of Artificial Societies and Social Simulation, Journal of
Artificial Societies and Social Simulation vol. 7, no. 3
<http://jasss.soc.surrey.ac.uk/7/3/5.html>
Schurr, N., Marecki, J., Tambe, M., Scerri, P. (2005). The Future of Disaster Response:
Humans Working with Multiagent Teams using DEFACTO, [on-line] Retrieved
January 3, 2011, URL: <http://www.aaai.org/Papers/Symposia/Spring/2005/SS-
05-01/SS05-01-002.pdf>
Siegwart, R. (2001). Indirect Manipulation of a Sphere on a Flat Disk Using Force
Information. International Journal of Advanced Robotic Systems. InTech, Vol.6, No.4,
(December 2009), pp. 12-16, ISSN 1729-8806
Suchánek, P. (2010a). Business Intelligence as a Support of E-commerce Systems in Connection
with Decision Making and Cross-Border Online Shopping. In Journal of Applied
Economic Science, Volume V, Issue 1, Spring 2010, ISSN 1843-6110, Dolj, Romania
Suchánek, P., Slaninová, K. & Bucki, R. (2010b). Business Intelligence as the Support of
decision-Making processes in E-commerce Systems Environment. Proceedings of
the Workshop Methods and Applications of Artificial Intelligence, The College of
Informatics and Management, pp. 5 – 20, ISBN 978-83-62466-02-3, Bielsko-Biala,
Poland
Sweiger, M., Madsen, M.R., Langston, J., Lombard, H. (2002). Clickstream Data Warehousing. 1
st edition. Wiley, 456 p., ISBN 978-0471083771, New Jersey, USA
Šperka, R. (2010). Improving the Competitiveness of Enterprises through Multi-agent
Systems. Science & Education Foundation: www.science-journals.eu [online]. 2010,
Economy & Business, Volume 4, Part 3. Available from http://www.science-
journals.eu/economy/2010/ISP-EB-Vol4-Part3.swf, ISSN 1313-2555, Bulgaria
Taniar, D. (2009). Progressive Methods in Data Warehousing and Business Intelligence:
Concepts and Competitive Analytics. Information Science Reference. 390 p. ISBN
978-1605662329, Toronto, Canada
Van der Linden, S. (June 2010). Integrating Wind Turbine Generators (WTG’s) with
Energy Storage, In: Wind Power, 17.06.2010, Available from
http://sciyo.com/articles/show/title/wind-power-integrating-wind-turbine-
generators-wtg-s-with-energy-storage
Velmurugan, M., S., Narayanasamy, K. (2008). Aplication of Decision Support System in E-
commerce. Communications of the IBIMA, pp. 156-169, Volume 5, 2008, Available
from http://www.ibimapublishing.com/journals/CIBIMA/volume5/v5n19.pdf
Vymětal, D. (2009). Nástin simulace podnikového řídicího systému s použitím agentů, In
Informační technologie pro praxi, Ekonomická fakulta VŠB, October 1–2, pp. 158–166,
Ostrava, Czech Republic
Vymětal, D., Suchánek, P. (2009). E-business environment in the Global Information Society.
V G. w. Korfantego, Podnikání v podmínkách Evropské unie. Górnośląska wyższa
szkoła Handlova im. Wojciecha Korfantego, ISBN 978-83-60953-24-2, Katowice,
Poland
Wolf, P. (2006). Úspěšný podnik na globálním trhu. CS Profi - Public, 240 p., ISBN 80-969546-5-
2, Bratislava, Slovakia
Wooldridge, M. (2002). An Introduction to MultiAgent Systems, John Wiley & Sons Ltd, 348 p.,
ISBN 978-0471496915, New Jersey, USA
Young, K., Srivastava, J. (2007). Impact of social influence in e-commerce decision making.
In ACM International Conference Proceeding Series, Vol. 258, p. 293 – 302, ISBN 978-1-
59593-700-1, Minneapolis, USA
7
Quick Response in
a Continuous-Replenishment-Programme
Based Manufacturer-Retailer Supply Chain
Shu-Lu Hsu1 and Chih-Ming Lee2
1Department of Management Information Systems,
National Chiayi University
2Department of Business Administration,
Soochow University
Taiwan
1. Introduction
With the widespread concept of partnership in supply chain management, the traditional
replenishment process is fast giving way to the Quick Response (QR) and Continuous
Replenishment Programme (CRP). QR is a movement in industries to shorten the
replenishment lead time which is critical to reduce inventory level and improve the levels of
customer service. Wal-Mart, Seven-Eleven Japan, and many other retailers apply
tremendous pressure on their suppliers to reduce the replenishment lead time (Chopra and
Mendil 2007). CRP is an efficient replenishment initiative which focuses on removing excess
inventory throughout the pipeline and synchronizes demand with production (EAN
International 2000). In CRP, the inventory of retailer is planned, monitored, and replenished
by the supplier on behalf of the consumers. To enable CRP, sales data and inventory level of
the retailer must be provided to the supplier via Electronic Data Interchange (EDI) or other
electronic means. Thus, to successfully implement CRP requires the supplier and the retailer
to work in a cooperative manner based on mutual trust and joint gains.
The lead time is needed for several operations among trading parties, such as ordering,
manufacturing, delivering, and handling. In practice, lead time can be shortened with extra
crashing costs. Many researchers have studied the inventory decision incorporating lead
time reduction under various assumptions. For example, Liao and Shyu (1991) first
developed an inventory model in which lead time was the unique decision variable. Ben-
Daya and Raouf (1994) extended the previous model by including both lead time and order
quantity as decision variables. Pan and Yang (2002) and Pan and Hsiao (2005) considered
lead time crashing cost as a function of both the order quantity and the reduced lead time.
Since the reduction of lead time may involve with the vendor and the buyer to improve the
related operations, a dyadic vendor-buyer viewpoint for shortening lead time is often
suggested. For instance, Iyer and Bergen (1997) compare the profits of a single vendor and a
single buyer supply chain before and after QR based on a newsboy inventory model. Ben-
Daya and Hariga (2004), Ouyang et al. (2004), Chang et al. (2006), and Ouyang et al. (2007)
investigated the decision of lead time reduction in a context of a single-vendor-single-

retailer integrated inventory system, in which the possibility of stockout of the supplier was
ignored.
The study of inventory decision of CRP based inventory systems was initiated by Goyal
(1976) by presenting a joint economic lot-size model for a supplier and a retailer.
Subsequently, many researchers investigated this issue under various assumptions. For
example, Banerjee (1986) generalized the model of Goyal (1976) by incorporating a finite
replenishment rate for the supplier. Goyal (1988) extended Banerjee’s work by relaxing the
lot-for-lot policy for the supplier and assumed that the supplier’s lot size is an integral
multiple of retailer’s order quantity. Goyal and Srinivasan (1992) further extended the
model by relaxing the assumption that the supplier can supply to the retailer only after
completing the entire production lot. At the same time, Banerjee and Banerjee (1992)
extended the study of integrated inventory control to a multiple-retailer case by considering
that the supplier delivers items to several buyers at a coordinated common cycle time.
Banerjee and Banerjee (1994) generalized their previous model by dealing with the normally
distributed demand case and assuming a fixed shortage cost attributable to each stockout
incident.
To understand the effects of QR on the CRP based supply chains, we build a formal model
of lead time reduction and replenishment decisions for a supply chain consisting of a
manufacturer and multiple retailers in which the inventory throughout the supply chain is
managed in a CRP context. The model extends the work of Banerjee and Banerjee (1994) by
including the ordering cost and adopting QR with the expenditure regarded as a function of
the reduced lead time. In stead of explicitly estimating the shortage cost, a service level
constraint (SLC) approach is applied to the problem for which may skirt the difficult
practical issue of explicitly determining the shortage cost (Chen and Krass 2001). According
to the definition in Ouyang and Chuang (2000), the service level in this study is measured
by the expected fraction of demand met from inventory on hand in an inventory cycle. The
objective of the model is to determine the common shipment cycle time and the
replenishment lead time for the retailers, the manufacturer’s production cycle time, and the
target levels of replenishment for the manufacturer and each retailer so that the expected
total system cost can be minimized under the service level constraint.
2. Notations and assumptions

To develop the proposed models, we adopt the following notation which is principally the
same as that in Banerjee and Banerjee (1994):
n = total number of retailers;
di = demand rate of retailer i, which follows a probability density function (p.d.f.) fi(di)
with mean Di and variance  i2 , i =1, 2,…, n;
hi = carrying cost per unit per year for retailer i ($/unit/year);
hv = carrying cost per unit per year for the manufacturer ($/unit/year);
l = length of lead time for retailers (year), a decision variable;
xi = demand during the protection period (T+l) on retailer i, which has a p.d.f.
f i T L (xi ) with the mean Di T  l  and the variance  i2 (T+l);
Quick Response in
a Continuous-Replenishment-Programme Based Manufacturer-Retailer Supply Chain 133
y = demand during the production cycle time of KT on manufacturer, which has a

n n
p.d.f. g(y) with the mean KT  Di and the variance KT   i2 ;
i 1 i 1
zi = safety factor for retailer i with zi  0;
zv = safety factor for the manufacturer with zv  0;
αi = the threshold of retailer i’s service level, 0  αi 1;
αv = the threshold of manufacturer’s service level, 0  αv 1;
A = setup cost for the manufacturer ($/per setup);
C = common ordering cost shared by all retailers ($/per order);
Ci = individual ordering cost for retailer i ($/per order);
Cr(l) = lead-time crashing cost per order;
P = production rate of manufacturer, which is a known constant;
n
D =  di , total demand from all the retailers on the manufacturer per year, which
i 1
n n n
has mean  Di and variance  i2 . Note that, of necessity, P >  Di ;
i 1 i 1 i 1
K = integral number of shipments to retailers per manufacturer’s production cycle,
which is a decision variable with K  1;
Si = replenish-up-to level after placing a new order for retailer i, a decision variable;
Sv = produce-up-to level after placing a new manufacture order for the manufacturer,
a decision variable;
T = common shipment cycle time to all retailers (year), a decision variable;
ECi = the expected annual inventory cost for retailer i;
ECv = the expected annual inventory cost for the manufacturer;
ETC = the expected annual total cost for the system.
The major assumptions made in this study are the following:
1. In the CRP context, the manufacturer plans, monitors, and replenishes the inventory of
retailers based on the information of sales data and inventory level provided by all the
retailers to minimize the expected annual total cost of the supply chain.
2. The demand rate of each retailer is independently distributed.
3. Each production lot of the manufacturer will be delivered in an integral number, K, of
shipments to all retailers periodically.
4. The target level of replenishment of Si units for retailer i is equal to the sum of retailer
i’s expected demand during the protection period, Di(T+l), and the safety stock, where
safety stock = ziσi T  l .
5. The target level of production of Sv units for the manufacturer is equal to the sum of
n
manufacturer’s expected demand during the production cycle, KT  Di , and safety
i 1
n
stock, where safety stock = zv KT   i2 .
i 1
6. For each order, a common ordering cost C is incurred to and shared by all retailers and
an individual ordering cost Ci is incurred to the retailer i.
7. The manufacturer incurs a setup cost A for each production run.
8. The service level measures, i.e., fraction of demand met per cycle, for the retailer i and
E(xi  Si ) E(y  Sv )
for the manufacturer are defined as 1  and 1  , respectively.
Di T  l  n
KT  Di
i 1
9. The expenditure for implementing QR is modeled as the lead-time crashing cost per
order, Cr(l), and is assumed to be a non-increasing stairstep function of l, such as:
0 l  l0 ,
r l1  l  l0 ,
1
C r (l )  
  
rb lb  l  lb  1 ,
where l0 and lb represent the existing and the minimal length of lead times, respectively.
This cost could be expenditure on equipment improvement, order expediting, or special
shipping and handling.
3. Model development and solution

Figure 1 depicting the inventory time plots for a manufacturer and n retailers, illustrates the
CRP based inventory control system proposed in this work. For the retailer i, there is a
target level of replenishment of Si units. For the manufacturer, there is a target level of
production of Sv units. Given a common shipment cycle time of T years, the manufacturer
produces a batch of item every KT years. After the Kth shipment cycle since the beginning of
the last production cycle, the manufacturer begins the production of another lot DT/P years
prior to the next scheduled shipment time, as shown in Figure 1, which is illustrated the case
of K =3. In order to cope with a stochastic demand, all of the trading parties carry safety
stocks.
The demand during the protection period for retailer i is assumed to be normally distributed
with mean Di T  l  and standard deviation  i T  l . Then, the total demand from all
retailers on the manufacturer during a production cycle of KT is normally distributed with
n n
the mean KT  Di and the standard deviation KT   i2 . For the retailer i, with a cycle of
i 1 i 1
T, a replenishment lead time of l, and a target level of Si, the expected net inventory level
just before receipt of an order is given by Si–Dil–DiT and the expected net inventory level
immediately after the successive order is Si–Dil. Then, the average inventory over the cycle is
1
approximated by  Si  Di l  DiT    Si  Di l   . According to the fourth assumption, the
2
target levels of replenishments, Si, is set as
Si = Di T  l   zi i T  l . (1)
So that the retailer i’s expected annual carrying cost is
hi  DT   DT 
 Si  Di l  DiT    Si  Di l    hi  Si  Di l  i  = hi  i  zi i T  l  .
2  2   2 
Quick Response in
The annual individual ordering cost incurred to retailer i is Ci/T. Then, the expected annual
inventory cost for all retailers, consisting of common ordering cost, individual ordering cost,
and carrying cost, is given by
n
1 n  n
 DiT 
 ECi  T  C   Ci    hi  2
 zi i T  l  .

i 1  i 1  i 1
Inventory
Retailer 1 S1
level
Time
Inventory
level
Retailer n Sn
l Time
Inventory
level
Manufacturer Sv
Time
Fig. 1. Inventory levels for a manufacturer and multiple retailers under a coordinated
common shipment cycle time
Based on the fifth assumption, the target levels of replenishments, Sv, is set as
n n
Sv = KT  Di  zv KT   i2 . (2)
i 1 i 1
According to Banerjee and Banerjee (1994) with Eq. (2), the manufacturer’s average
inventory can be derived as
n  n  n   n  n 
 DiT   Di  n  DiT    Di  2 Di  n
 i 1  2  K   K  1  S  KT D  K  1  i 1   i 1  1  z KT  2 .
i 1
2  P  v  i i 1
2   P  P  v  i
i 1 i 1
     
     
Then, the manufacturer’s expected annual inventory cost is given by
 n   n  n  
  DiT    Di  2  Di  n

A   K  1  i  1   i  1  1  z KT  2  ,
EC v 
KT
 hv  i  1
  P   v  i 
 2   
P
 i 1 
   
 
   
 
where the first term represents the average annual setup cost and the second term is the
average carrying cost.
Based on the service level measure defined in Ouyang and Chuang (2000), the following
service level constraints are specified:

S (xi  Si )f i T l (xi )dxi
service level constraint on retailer i = 1  i
 αi, i =1, 2,…, n, (3)
Di (T  l )

S  y  Sv  g(y )dy
service level constraint on the manufacturer = 1  v
n
 αv. (4)
KT  Di
i 1
As the demand is normally distributed, (3) and (4) can be rewritten as (see Appendix for
proof):
service level constraint on retailer i =  (zi )  zi (zi ) 

 1   i  Di T l
 0 , i =1, 2,…, n, (5)
i
n
1  v  KT  Di
i 1
service level constraint on the manufacturer =  (zv )  zv(zv )  0, (6)
n
  i2
i 1
where (z) and (z) represent the standard normal distribution and the complementary
cumulative normal distribution function, respectively. The expected annual total cost for the
system, comprising the expected annual inventory costs for all parties and the lead time
crashing cost, is given by
Quick Response in
n
C r (l )
ETC(K, T, l, z, zv)=  EC i  EC v 
i 1 T
n
A  n   n  n  
C   Ci   C r (l )  hv  Di    Di  2  Di  n D h 
K   K  1  i  1   i  1  1 
  2 
i 1
=  T  i 1 i i
T   P 
 2   
P
 i 1 
     
 
n n
+ hv zv KT   i2   hi zi i T  l , (7)
i 1 i 1
where the vector z ={z1, z2,…, zn}. Then, the decision is to minimize ETC(K, T, l, z, zv) under
the constraints (5) and (6).
In order to solve the nonlinear programming problem, the following propositions are
needed.
Proposition 1. For any given K, T, z, and zv, the minimal ETC will occur at the end points of
the interval [lj+1, lj] when Cr(l) is a stairstep function.
Proof:
 2C r (l )
As Cr(l) is a stairstep function,  0 for any l in a continuous interval [lj+1, lj]. The
l 2
second derivative of (7) with respect to l is equal to
2
 ETC 1  C r (l ) 2  hi zi i
  i 1 0
l 2 T l 2 4 T  l 
3
for any l in a continuous interval [lj+1, lj]. Therefore, the proposition holds.
Proposition 2. For any fixed K, l, z, and zv, there is a unique optimal T to ETC in (7).
Proof:
For any given K, l, z, and zv in a continuous interval [lj+1, lj], a necessary condition for a
solution of T to be optimal for (7) is
n
A n   n  n 
C   Ci   C r (l ) hv  Di    Di  2 Di 
ETC K
 i 1
 i  1 K  1  i  1   i  1  1
T T 2 2   P  P 
   
   
n n n
 Di hi hv zv K  i2  hi zi σ i
i 1 i 1 i 1
    0.
2 2 T 2 T l
The above equation can be rearranged as
n
A n   n  n 
C   Ci   C r (l ) hv  Di    Di  2 Di 
K
i 1
 i  1  K  1  i  1   i  1  1
T2 2   P  P 
   
   
n n
n
hv zv K   i2  hi zi i
Di hi
  i 1
 i 1
. (8)
i 1 2 2 T 2 T l
Let Tf be the solution to (8). Then, the second derivative of (7) at Tf satisfies
 n
A  n n
 2 ETC
2  C   C i 
K
 C r (l )  h v z v K   i2  hi zi i
  i 1   i 1
 i 1
T 2 (T f ) T f3 3 3
4T 2 f 
4 Tf  l 2
 n   n  n  n n 
 hv  Di  Di  2  Di  n D h hv zv K  σ i  hi zi σ i 
2
 
2  i 1  K  1  i  1   i  1  1 
  2
i 1
 
i i
  i 1 
Tf  2   P  P 2 Tf 2 Tf  l 
i 1
   
     
 
n n
hv zv K  σ i2  hi zi σ i
i 1 i 1
 3
 3
4T f 2

4 Tf  l 2
n   n  n  n n
hv  Di    Di  2 Di  n D h 3hv zv K  i 3T f  4l  hi zi i
2
 
 i  1  K  1  i  1   i  1  1   i i  i 1
 i 1
 0.
  P   3
Tf
  
P
 i 1 T f 4T f T f
Tf Tf  l 2  
   
That is, the second derivative of ETC at each Tf is positive. The result implies that, for given
K, l, z, and zv, an only solution for T, corresponding to the minimum of ETC, can be derived
from (8). This completes the proof.
Proposition 3. For any given K, T, and l in a continuous interval [lj+1, lj], the boundary
solution for z and zv derived from (5) and (6) is the optimal solution to this nonlinear
programming problem.
Quick Response in
Proof:
By using the method of Lagrange multipliers, the problem can be transformed into
n
A  n   n  n  
C  Ci   Cr (l)  hv  Di    Di  2 Di  n 
K  D h 
ETC(K, T, l, z, zv, , s)= i 1
 T  i  1  K  1  i  1   i  1  1   i i 
T   P  
 2   
P
 i 1 2 
  
     
n n n 
 hv zv KT   i2   hi zi i T  l   i  (zi )  zi (zi ) 
 1   i  Di T  l  s2 
i 
i 1 i 1 i 1  i 
 n 
  1   v  KT  Di 
 i 1 2
 v  (zv )  zv (zv )   sv  , (9)
n
 
  i  2

 i 1 
where λ is a vector of Lagrange multipliers with λ={λ1, λ2,…, λn, λv} 0, and s is a vector of
slack variables with s = {s12, s22,…, sn2, sv2}. According to Kuhn-Tucker theorem (Taha, 2002),
for any given l in a continuous interval [lj+1, lj], it can be shown that s = {0,0,…,0} is a
necessary condition for a solution in (9) to be optimal. Then, for any given l in a continuous
interval [lj+1, lj], the necessary conditions for (9) to be minimized are
ETC
 hi i T  l  i (zi )  0 , i =1, 2,.., n, (10)
zi
n
ETC
 hv KT   i2  v (zv )  0 , (11)
zv i 1
ETC T l
  (zi )  zi (zi )   1   i  Di 0, (12)
i i
n
1  v  KT  Di
ETC i 1
  (zv )  zv (zv )  0, (13)
v n
  i2
i 1
From (10) and (11), we have
n
hv KT   i2
hi i T  l i 1
i* (zi )   0, i = 1, 2,…, n, and v* (zv )   0.
(zi ) (zv )
 2 ETC
Therefore, constraints (5) and (6) are binding. In addition, = i (zi ) >0 (i=1, 2,…, n,
zi2
m), then the z and zv derived from (12) and (13) (i.e., the boundary solutions to (5) and (6))
are optimal to (9) with given K, T, and l.
Based on the above propositions, the optimal solution for K, T, l, S, and Sv to this nonlinear
programming problem can be obtained by the following algorithm:
Algorithm
10. Set K=1.
11. Find ETC * (K , l j | j  0,1,..., b )
2.1 Let j =0
2.2 Set l = lj. zi (i =1, 2,…, n) = 0, and zv = 0.
2.3 Compute T* by substituting K, lj, z and zv into (8). Set Tkj = T*.
2.4 Compute z* by substituting lj and Tkj into (12) and compute z*v by substituting K
and Tkj into (13).
2.5 Compute T* by substituting K, lj, z*, and zv* into (8).
2.6 If T* = Tkj, go to step 2.7; otherwise, set Tkj = T* and go to step 2.4.
2.7 Set T*(K,lj)=Tkj, z * (K , l j ) =z*, and zv* (K , l j ) = zv* . Compute ETC * (K , l j ) by substituting
K, lj, T*(K,lj), z * (K , l j ) , and zv* (K , l j ) into (7).

2.8 Let j = j+1 and go back to step 2.2 until j = b.
12. Set ETC(K,l) =Min[ETC(K,lj)|j = 0,1,…,b].
13. If K = 1, set ETCs = ETC(K,l), K = K+1, zi (i =1, 2,…, n) = 0, zv = 0 and go back to step 2;
otherwise go to step 5.
14. If ETC(K,l) < ETCs, let ETCs = ETC(K,l), K = K+1, zi (i =1, 2,…, n) = 0, zv = 0, and go back
to step 2; otherwise let ETC*= ETCs along with the corresponding values of K and l
being the value of K* and l*, respectively.
15. Find T* from (8) with the obtained K* and l*. Find z* from (12) and zv* from (13) with the
obtained K*, l*, and T*. Find S* from (1) and Sv* from (2) with the obtained K*, l*, T*, z*,
and zv* . Output the values of K*, l*, T*, S*, Sv* , ETC* and Stop.
4. Computer implementation
The algorithm proposed in Section 3 has been implemented as a decision support system on
a personal computer in which Intel Pentium D 2.8 GHz CPU with 1024 MB RAM inside it.
Visual Basic 2005 is utilized as the software platform to develop the decision support
system. Figure 2 shows the window to input the numbers of retailers, e.g., n =3 in this case,
and steps of the lead-time crashing cost function, e.g., three steps in this case, involved in
the replenishment and lead time reduction decisions. Once the decision maker inputs the
numbers, a window to specify the relevant parameters will be displayed as shown in Figure
2. If all of the relevant parameters are input for the system, the decision maker then clicks
the label “Find Solution” to select the lead time policy and to set the options of output.
Figure 3 illustrates the result that the system derives the solution in which K* = 2, l* = 0.005
years, T* = 0.0709 years, S1* =708, S2* = 760, S3* = 1276, Sv* = 3574, and ETC* = $19455.5 for the
case of implementing QR (representing by the controllable lead time option) after the
Quick Response in
Fig. 2. The system window to input parameters
Fig. 3. The replenishment and lead time reduction decisions under CRP provided by the
decision support system
Fig. 4. The coordinated replenishment decision under CRP provided by the decision support
system
Fig. 5. The “what-if analysis” function of the decision support system

Quick Response in
decision maker clicks “Perform”. From the value of “Saving”, the benefit derived from
implementing QR can be learned. As shown in Figure 4, the system also provides the
inventory decision before QR (represented by the constant lead time option). One advantage
of the system is that the decision maker can learn the sensitivity of replenishment and lead
time decisions and the expected annual total cost of the chain by specifying the parameter
with the variation range in the Window of “What-if Analysis”. Figure 5 illustrates the
output of the what-if analysis of K*, l*, T*, S*, Sv* , ETC*, and the cost saving after QR against
the variation of standard deviation of demand.
K* T* l* S1* S2* S3* S*v ETC* K0a T0a ΔETC(%)b
Base example 2 0.0709 0.005 708 760 1276 3574 19455.3 2 0.0711 0.73
σi
+100% 2 0.0568 0.002 895 1086 1384 3789 27806.2 2 0.0577 2.59
(i=1,2,3)
P +100% 1 0.0877 0.005 827 874 1320 2357 18822.3 1 0.0880 0.71
C +100% 2 0.0779 0.005 758 808 1208 3885 20799.7 2 0.0781 0.67
Ci
+100% 1 0.1024 0.005 930 970 1488 2691 23410.0 1 0.1027 0.54
(i=1,2,3)
A +100% 3 0.0705 0.005 705 757 1291 5108 20696.8 2 0.0781 1.16
hi
+100% 3 0.0507 0.005 560 616 886 3803 26567.3 3 0.0513 1.88
(i=1,2,3)
hv +100% 1 0.0704 0.005 704 756 1120 1958 23078.6 1 0.0705 0.62
1-αi
+100% 2 0.0733 0.01 705 741 1128 3681 18548.2 2 0.0735 0.32
(i=1,2,3)
1-αv +100% 2 0.0727 0.005 721 772 1147 3467 18897.2 2 0.0728 0.75
a: K0 and T0 represent the optimal K and T, respectively, before QR is implemented.
b: ΔETC(%)=(1－ETC*/ETC0)×100%, where ETC0 represents the optimal ETC before QR is implemented.
Table 1. Sensitivity analysis for parameters

The sensitivity of the optimal solution has been further examined by conducting several
numerical experiments. The results are illustrated by a base case and are shown in Table 1.
Some findings are summarized as follows:
1. The amount of ΔETC(%) measures the improvement rate of expected annual total
system cost after implementing QR. The behavior of ΔETC(%) reveals that the supply
chain may benefit from shortening the replenishment lead time.
2. The value of ΔETC(%) is especially sensitive to the variation of σi or hi. As σi or hi
increases, the value of ΔETC(%) increases. The results mean that the return of
investment in QR is especially significant for a system with high uncertainty in demand
or with high carrying costs incurred to the retailers.
3. The common shipment cycle time is more sensitive to the variance of demand, the
variation of manufacturer’s production rate, the individual ordering cost, and the
retailer’s carrying cost.
4. The shipment cycle time, T*, will be reduced after implementing QR in which the
replenishment lead time for retailers has been shortened. Moreover, the target levels of
replenishments for the manufacturer and the retailers may become lower after bringing
QR into practice.
5. The number of shipments per production cycle, K*, after implementing QR is always no
less than that before implementing QR. When the values of K for the two situations are
equal, the optimal shipment cycle time, T* as well as the protection period, T*+l*, after
QR is always no large than those before QR. The result implies that under a fixed
number of shipments during a production cycle, the protection period for the retailer
will be reduced after implementing QR.
6. The increase of production rate will result in a longer shipment cycle time and higher
target levels of replenishments for the retailers. In contrast, as the value of P increases,
the number of shipment per production cycle and the target level of production for the
manufacturer will become smaller.
7. When the ordering cost increases, the shipment cycle time and the target levels of
replenishments for each party will increase.
8. The value of αi specifies the minimal fraction of demand met per cycle for the retailer
and directly relates with the length of protection period (T+l). It can be found that the
amount of ΔETC(%) decreases as the retailer’s maximal fraction of demand unfilled per
cycle increases. The result implies that the benefit from implementing QR is
significantly related to the retailer’s service level threshold and the benefit is substantial
for a supply chain requesting a high customer service level.
9. The numerical example shows that as 1－αv increases from 1% to 2%, the length of
common shipment cycle time increases by 2.54% (from 0.0709 to 0.0727) but the lead
time is unaffected. The result implies that the effect of manufacturer’s service level
threshold is more significant on the common shipment cycle time than on the reduction
of lead time.
5. Conclusions
Speed, service, and supply chain management have been core capabilities for business
competition. The reduction of overall system response time with a satisfied service level has
received a great deal of attention from researchers and practitioners. In this study, we
investigated the effect of investing in QR on a CRP based supply chain where a
manufacturer produces and delivers items to multiple retailers at a coordinated common
cycle time with the minimized expected total system cost and satisfied service levels.
Extending the work of Banerjee and Banerjee (1994) by involving ordering costs and the
reducible replenishment lead time, a model and an algorithm are proposed to
simultaneously determine the optimal shipment cycle time, target levels of replenishments,
lead time, and number of shipments pre production cycle under service level constraints for
the supply chain.
A numerical experiment along with sensitivity analysis was performed and the results
explain the effect of QR on the replenishment decisions and the total system cost. The results
provide the following findings about our model:
1. The system can reduce the stocks throughout the pipeline and remain service levels of
retailers via investing in QR initiative.
2. The benefit from implementing QR is especially significant for a supply chain with high
uncertainty in demand or the retailers requesting high service levels or incurring high
carrying costs.
Quick Response in
3. The shipment cycle time will decrease after QR implemented. Additionally, the shipment
cycle time is especially sensitive to the variation of manufacturer’s production rate, the
individual ordering cost, the variance of demand, and the retailer’s carrying cost.
4. The decision of adopting QR is mainly influenced by the variance of demand and the
retailer’s service level threshold. The higher the demand uncertainty or the higher the
retailer’s service level threshold, the more beneficial to implement QR in supply chains.
Appendix
 xi   i
Let G    x  Si  f (xi )dxi and zi  , where f(xi) represents a normal distribution with
Si i
the mean i and the standard deviation i. By replacing xi with i+zii, we have
 
G  S    i  zi i  Si   (zi )dzi =  i Si  i zi (zi )dzi   Si  i  Si  i  (zi )dzi , (A.1)
i i
i i i
where (zi) is the standard normal distribution and (zi) is the complementary cumulative
normal distribution function defined as
w2
 1  
(zi )    (w )dw  z e 2 dw .
zi 2 i
zi2
Let v  , then there is dv  zi dzi and
2
2
 1  Si  i 
 1  1  v 1  2   i 

 1  Si  i  2 e dv  2 e 1 Si  i  2  2 e 
v
Si  i zi (zi )dz    (zi ) (A.2)
i 2 
2   i 
 2
 i 
Next,

 Si  i  S    (zi )dzi   Si  i  (zi ) .
i i
(A.3)
i
By substituting (A.2) and (A.3) into (A.1), we have

G
Si
 x  Si  f (xi )dxi  i (zi )   Si  i  (zi ) . (A.4)
Let i = Di(T+l) and i =  i T  l and apply (A.4) to (3), (3) can be rewritten as
 i T  l (zi )  Si  Di T  l   (zi )

1  αi, i =1, 2, 3. (A.5)
Di T  l 
By substituting (1) in to (A.5) and rearranging the inequality, we have
T  l  0, i =1, 2, 3.
 (zi )  zi (zi )   1   i  Di
i
This completes the proof for (5). Accordingly, (6) can be proved.
6. Acknowledgment
This research was partially supported by the National Science Council, Taiwan ROC (Plan
No. NSC 98-2410-H-415 -007).
7. References
Banerjee A (1986) A joint economic lot size model for purchaser and vendor. Decis Sci
17(3):92-311.
Banerjee A, Banerjee S (1992) Coordinated, orderless inventory replenishment for a single
supplier and multiple buyers through electronic data interchange. Int J Technol
Manage 79(4-5):328-336.
Banerjee A, Banerjee S (1994) A coordinated order-up-to inventory control policy for a single
supplier and multiple buyers using electronic data interchange. Int J Prod Econ
35(1-3):85-91.
Ben-Daya M, Hariga M (2004) Integrated single vendor single buyer model with stochastic
demand and variable lead time. Int J Prod Econ 92(1):75-80.
Ben-Daya M, Raouf A (1994) Inventory models involving lead time as a decision variable. J
Oper Res Soc 45(5):579-582.
Chang HC, Ouyang LY, Wu KS, Ho CH (2006) Integrated vendor-buyer cooperative
inventory models with controllable lead time and ordering cost reduction. Eur J
Oper Res 170(2):481-495.
Chen FY, Krass D (2001) Inventory models with minimal service level constraints. Eur J
Oper Res 134(1):120-140.
Chopra S, Meindl P (2007) Supply chain management. Pearson Education Inc., New Jersey.
EAN International (2000) Continuous replenishment: how to use the EAN.UCC standard.
http://www.gs1.org/docs/EDI0002.pdf.
Goyal SK (1976) An integrated inventory model for a single-supplier-single- customer
problem. Int J Prod Res 15:107-111.
Goyal SK (1988) A joint economic-lot-size model for purchaser and vendor: a comment.
Decis Sci 19(1):236-241.
Goyal SK, Srinvasan G (1992) The individually responsible and rational comment. Decis Sci
23(3):777-784.
Liao CJ, Shyu CH (1991) An analytical determination of lead time with normal demand. Int J
Oper Prod Manage 7(4):115-124.
Iyer AV, Bergen ME (1997) Quick response in manufacturer-retailer channels. Manage Sci
43(4):559-570.
Ouyang LY, Chuang BR (2000) Stochastic inventory model involving variable lead time with
a service level. Yugosl J Oper Res 10(1):81-98.
Ouyang LY, Wu KS, Ho CH (2004) Integrated vendor-buyer cooperative models with
stochastic demand in controllable lead time. Int J Prod Econ 92(3):255-266.
Ouyang LY, Wu, KS, Ho, CH (2007) An integrated vendor-buyer inventory model with
quality improvement and lead time reduction. Int J Prod Econ 108(1-2):349-358.
Pan JC, Hsiao YC (2005) Integrated inventory models with controllable lead time and
backorder discount consolidations. Int J Prod Econ 93-94:387-397.
Pan JC, Yang JS (2002) A study of an integrated inventory with controllable lead time. Int J
Prod Res 40(5):1263-1273.
Taha HA (2002) Operations research: an introduction. Prentice Hall, Upper Sadle River,
New Jersey.
8
Collaboration in Decision Making:

A Semi-Automated Support for Managing
the Evolution of Virtual Enterprises
Marcus Vinicius Drissen-Silva and Ricardo J. Rabelo
Federal University of Santa Catarina
Brazil
1. Introduction
Collaborative Networked Organizations (CNO) has become one of the most prominent
strategic paradigms that companies have sought as a mean to face the challenges imposed
by globalization (Camarinha-Matos et al., 2005). There are several types of CNOs, like as
supply chain, virtual labs, virtual organizations breeding environment (VBE), extended
enterprises, virtual organizations and virtual enterprises. The common rationale behind
such alliances is that they rely on collaboration with other companies to be more
competitive. This work focuses on virtual enterprise.
A Virtual Enterprise (VE) can be generally defined as a temporary alliance of autonomous
and heterogeneous enterprises that dynamically joint together to cope with a given business
opportunity, acting as one single enterprise. A VE dismiss itself after accomplishing its goal
(Rabelo et al., 2004).
Managing the VE life cycle very efficiently is crucial for the business realization. This
involves the creation, the operation, the evolution and the dissolution of a VE. This work
focuses on the VE evolution phase. In general, the VE evolution phase comprises activities
related to managing changes and adaptations in the VE’s plan (i.e. the VE operation phase)
in order to guarantee the achievement of its goals and duties. This can comprehend actions
like simple modifications in some technical specification, passing by changes in and/or
negotiations on the VE’s schedule, or more drastically the replacement of some of its
members.
VEs have, however, some intrinsic and particular characteristics which impose respecting a
number of requirements in decision making. The most important one is that decision should
be performed in a collaborative, decentralized, distributed and transparent way, considering
that VE members are autonomous, independent and geographically dispersed. Besides that,
the fact that each VE is per definition completely different from one to another (in terms of
number of partners, their skills, culture, local regulations, specificities determined by the
given client, etc.) makes the solution of some problems not necessarily deterministic and the
use of previous decisions for equivalent problems not necessarily useful. As such, managing
the VE evolution requires additional approaches in order to be properly handled (Drissen-
Silva & Rabelo, 2009b).
Figure 1 presents a general vision of the aspects related to the management approach and
the decision making process in a centralized and decentralized ways, trying to expose the
necessary requirements for offering a new decentralized and collaborative decision making
model for the Virtual Enterprise evolution, as offered by this work.
Source: Extended from Ollus et al., 2009 and Loss, 2007.

Fig. 1. Requirements for a decentralized decision with collaborative work.
Considering that a VE is usually a grasp-driven alliance for a short-term business
opportunity, the major challenge for a decision making that respects those requirements is
to be agile. This means that, once identified, a problem should be solved as fast as possible,
with high quality and feasibility, and relying on trustful information. It has to be also taken
into account that enterprises are often involved in several VEs simultaneously and that some
of them are inter-related. Therefore, managing the evolution of a VE requires an ample
spectrum of issues that make decision-making extremely complex, making phone calls or
chatting among VE members by far insufficient to solve a problem.
This work proposes a novel decision-making approach and framework as a contribution to
face those requirements. It is represented by an integrated collaborative decision making
framework that assist VE managers along the entire decision making process, including the
possibility of evaluating decision feasibility and its impact over each VE that members are
involved in.
This chapter is organized as follows: Section 1 presented a general analysis of the
requirements for VE management and the evolution phase. Section 2 discusses the problem
related to a collaborative decision making and its requirements for the offered framework.
Section 3 introduces the proposed framework for managing the virtual enterprise evolution
with a collaborative discussion. Section 4 presents the results of the framework considering
A Semi-Automated Support for Managing the Evolution of Virtual Enterprises 149
a prototype development. Section 5 provides a general evaluation with contributions,

limitations and future research. Finally, section 6 discuss around the conclusions reached on
termination this work.
2. Collaborative decision making

Distributed decision-making is not a new research topic and many works have been
developed along the last decade concerning this matter, especially in the form of distributed
decision support systems (Bostrom et al., 2003). Actually, the work presented in this chapter
follows the same line but it adds diverse elements and requirements from the VE area. The
approach addressed in this paper is anchored in the following scenario:
“Partners, although being distributed and autonomous, belong to a long-term alliance of
type VBE (Virtual Organization Breeding Environment), so sharing common operating
principles. One of the main principles is that they trust on each other and they should
collaborate towards reaching a globally feasible solution for a problem that takes place
during the VE operation. Partners should solve the problem related to the VE they are
involved in and hence they should discuss about it through a computing network (e.g.
Internet). The discussion should be structured in order to get focused and to have
potentially better quality, making use of a shared distributed decision-making environment
guided by a decision protocol. This structure should be connected to business processes
management. Besides that, it should be flexible and adaptive regarding the problem and the
VE characteristics. However, there is not one very pre-defined protocol for every single
problem. Each VE tends to be so unique and problems so particular for a given business
context and VE composition. Partners should have some freedom to exchange ideas while
they evaluate possibilities against their availabilities. This evaluation should be made via an
easy access to the most common managerial supporting software tools in order to facilitate
Small and Medium Enterprises (SME) managers’ activities. After this, they should have
means to evaluate the impact of their decisions before acting. All this should be supported
by adequate ICT infrastructures, which can also provide the necessary security in the
communications and access rights.”
In order to cope with this scenario and with those requirements previously mentioned, six
aspects have to be supported by a comprehensive decision-making environment for the VE
evolution: 1) Partners’ discussion; 2) Methodological guidance; 3) Modular and flexible execution of
decision protocols, aligning business and processes; 4) Performance measurement; 5) Performance
evaluation; and 6) ICT Infrastructure. Next sections provide a resumed revision of the most
important techniques used to support these six aspects. The methodological approach
followed in this work is to use existing theoretical and software results related to those
aspects and to combine and adapt them respecting the envisaged environment.
2.1 Partners’ discussion

This issue is related to endowing partners with a collaborative environment where they can
exchange information towards the problem resolution. In this sense, Groupware or CSCW
tools (Wulf et al., 2008) have been largely used to support multiple users working on related
tasks in local and remote networks. However, they cope with a partial – and perhaps less
complex – part of the problem. The issue here is not only to make partners interact with each
other, but also to globally coordinate their discussions about each identified problem, and as
fast as possible. Besides that, it is necessary to integrate information for further auditing,
giving transparency to the whole process, as well as to regulate partners’ involvement and
information access as long as decisions are taken. After a review in the literature, three
works have been found out that offer elements for this desired environment.
HERMES (Karacapilidis & Papadias, 2001) is a support system used for collaborative
decision-making via argumentation. It helps in the solution of non-structured problems,
coordinating a joint discussion among decision makers. It offers an online discussion about
one or more specific subjects, where each participant can suggest alternatives to the problem
or simply point out their pros and cons in relation to current alternatives. There is an
association of weights that considers the positioning in favor of or against the suggestions,
hence providing a global vision of the opinions.
DELPHI is a classical method (Dalkey & Helmer, 1963) created with the purpose of finding a
consensus about a given topic of discussion but without confrontation. Essentially, a
summary of the opinions is elaborated along diverse rounds and it is sent back to the
participants keeping the names anonymous. The process continues until the consensus /
final decision or opinion is reached.
Woelfel et al. (described in Rabelo et al., 2008) developed an integrated suite of web-based
groupware services that has considered CNOs requirements. This suite includes the services
of instant messaging, mailing, discussion forum, calendar, wiki, content management
system, and news & announcement. One interesting feature of the instant messaging service
is the possibility of having private discussions rooms, allowing having several parallel
discussions involving all partners and some rooms only available for authorized partners.
2.2 Methodological guidance

The methodology’s goal is to prevent partners from dealing with the problem without any
guidance, losing time and resources, which can hazard the VE’s business. An approach for
that is to see a VE as a project, making use of project management reference models. This
means making partners to be guided along the problem resolution through a set of steps
grounded on project management foundations.
One of the most relevant foundations to support VE as a project is the Project Management
Body of Knowledge, or just PMBOK (PMBOK, 2004). PMBOK states that “a project is a
temporary effort to create a unique product or service”. Respecting the VE evolution phase
and the VE definition (see section 1), it is argued that a VE can be seen as a project as both
are temporary and unique in view of the creation of a product or service or to cope with a
specific collaboration need. Jansson and Eschenbaecher (2005) advocate that managing a VE
is more than managing a project as the creation of VEs requires a long and previous
preparation. However, this also embraces the VE creation phase, whereas the focus here is
the VE evolution phase, i.e. when the VE is already in execution.
In spite of being is a very comprehensive model, PMBOK is too general for handling
changes in projects subjected to constant changes - which is the case of VE - and for which
other models have been proposed.
The model called as Capability Maturity Model Integration (CMMI, 2006) has been
fundamentally used in the area of software development. It presents a decision and
resolution analysis with more details (compared to PMBOK), and gives a strong foundation
to assist organizations in the improvement of their processes and in their capacity to manage
the development, acquisition and maintenance of products and services. Some CMMI steps
can be useful in the development of an agile method for the VE concept, but it is too focused
on software development business processes.
The Agile Project Management (APM) model sees the changing need as an adaptation in the
exploration of alternatives that can fit to new scenes. APM was essentially created for
projects which demand more agility and dynamism (Leite, 2004), presenting a deeper set of
actions for handling changes. There are other management models that handle changes in a
project, namely ECM - Engineering Change Management (Tavčar & Duhovnik, 2005), CC -
Configuration Control (Military Handbook, 2001) and CM – Change Management (Weerd,
2007). In general, they organize the phases of change management in four macro phases: i)
need of change identification, where the causes of the problem and the affected members are
identified in order to prepare a change solicitation; ii) change proposal, where the members
that are going to participate in the change analysis are defined; iii) change planning, where
the different possible scenarios to solve the problem are evaluated via general evaluations,
and; iv) Implementation, where the most suitable alternative for the problem is settled and the
new project’s parameters are reconfigured.
All these reference models are very general and they can be instantiated to any type of VE
topology. As such, any managerial style and model, different support techniques,
management tools and performance evaluation methods can be applied in each case
(Karvonen et al., 2005). Considering their generality the models are not ready used in VE
evolution scenarios. Therefore, in spite of the extremely importance of their foundations,
they should be adapted for that.
2.3 Decision protocols

Decision protocols are seen in this work as an instrument to: i) systemize a set of actions
where there is a strong human intervention, ii) to standardize and iii) to enhance their
execution efficiency. In the context of VE, three works were found out in the literature those
offer some computer assistance for handling decision protocols.
ILMSS system (Rabelo et al., 1998) was developed to systemize logistic actions in Extended
Enterprises following a pre-defined and general decision-protocol. DBPMS system (Rabelo
et al., 2000) was an evolution of ILMSS, and it was developed to coordinate conflicts among
partners in a Supply Chain applying a modular but fixed approach to generate decision
protocols. SC2 (Rabelo & Pereira-Klen 2002) was a multi-agent system developed as an
evolution of DBPMS. One of its agents was responsible for managing conflicts that took
place along the execution of tasks in dynamic Supply Chains. The decision “blocks” were
chosen by an agent but the blocks had a high granularity. Another relevant particularity –
and limitation – of these three works is that they assumed that the main coordinator was the
only one who could trigger the process of looking for solutions close to members when
problems arose, as well as the only one who could make suggestions, who could have access
to the others’ information, and who took the decision. Besides that, these works only dealt
with rescheduling and basic actions towards partners’ replacement. As it was stressed in the
previous section, managing VE evolution requires several other features and types of
actions.
The VOM Toolkit (Pěchouček & Hodík 2007) is an integrated environment that has been
developed to help the VE coordinator in doing several activities, such as VE performance
monitoring, alerting about changes in the expected performance, and rescheduling and
reconfiguration simulation to optimize the VE performance. However, likewise in those
other three systems, it leaves totally to the VE coordinator to implement the corrections to
solve the conflict. No guidelines or supporting methodology are offered to help in these
activities.
Another perspective is that all these works - and some other more recent ones (e.g. Hodík &
Stach, 2008; Negretto et al., 2008; Muller et al., 2008) – are however disconnected from the
global operation ambient of the companies. This means that the decision-making process is
carried out separated of the other processes. In practice, this obliges managers to switch
from one environment to another and to cope with different sources of information (this is a
problem as SMEs usually have several basic problems of systems integration). A sound
alternative for that is the BPM (Business Process Management) approach (Grefen et al., 2009)
and the SOA (Service Oriented Architecture) paradigm (Ordanini & Pasini, 2008). BPM
provides foundations for a loose-coupled, modular, composite and integrated definition of
business processes. From the process execution point of view, BPM tools generate BPEL
(Business Process Execution Language) files as output, allowing a direct integration among
business level (BPM) and execution level (SOA / web services). There are both commercial
and academic supporting tools for that (e.g. Oracle BPEL Designer, IBM WebSphere). This
combination can provide the notion of flexible and modular decision protocols.
2.4 Performance monitoring and measurement

Performance monitoring and measurement look to the current situation of the production
system, treating the problem in the (VE) operation phase. The goal of this aspect from the
VE evolution management point of view is to offer conditions for the VE partners to
measure their own performance and to check their capacity in order to get more confidence
when deciding about how to do respecting the given problem. This involves, therefore,
monitoring (i.e. gathering of internal information) and further analysis (performance
measurement). There are a number of performance measurement models. Two of the most
relevant ones are the Balanced Scorecard (BSC) and SCOR (Supply Chain Operation
Reference).
BSC is a method that “translates the mission and the view of companies in a wide group of
performance measures which is a foundation to a measurement system and strategic
management” (Kaplan & Norton, 1997). It allows managers to identify which of the
activities could be considered as critical for the well functioning of the organization that are
directly responsible for the generation of value to the shareholders, clients, partners,
providers and to the community.
SCOR is as the cross-industry de facto standard diagnostic tool for supply chain management
grounded on three fundamental perspectives: processes, performance indicators and best
practices (Supply Chain Council, 2005). SCOR is based on five distinct management
processes: Plan, Source, Make, Deliver, and Return, which have many standard performance
indicators associated to. The goal is to optimize and integrate processes and logistics while
attending client needs.
In terms of performance indicators, Baldo et al. (2008) has developed a framework to identify
the most relevant performance indicators that should be applied to a VE regarding the
characteristics of the business opportunity and involved partners.
In terms of techniques, OLAP (On-line Analytical Processing) is an approach to quickly
provide answers to analytical queries that are multi-dimensional in nature (sales,
marketing, budgeting, forecasting, capacity, etc.). Via the so-called OLAP cube, it allows
for complex analytical and ad-hoc queries with a rapid execution time based on historical
data facilitating decision-making (adapted from Lechtenborger & Vossenm, 2003; Moon et
al., 2007).
2.5 Performance evaluation

Performance evaluation aims at providing decision elements based on performance
measurement results. The goal of this aspect in the context of VE evolution management is
to provide partners with techniques that help them to evaluate the impact of their decisions
at their companies along the discussion process. At the same time, it allows the VE
coordinator to evaluate the global solution before validating the final decision.
According to Raj Jain (1991), the three techniques of performance evaluation are: (i)
analytical modeling, (ii) simulation, and (iii) direct measurement. Each one has pros and
cons, and there are several considerations to decide which technique is better to use, like
modeling time, data acquisition, model complexity, execution time, required skills, among
others. Actually, simulation has been attracting a large number of users due to its intrinsic
capability for creating and evaluating what-if scenarios, capturing the dynamic behavior of
the system (Johnsson & Johanson, 2003). On the other hand, analytical models are more
adequate when near-optimum solutions are needed. Capacity planning is a sensible part as
most of the problems that use to happen in the VE operation requires changes in the
companies’ production capacity.
2.6 Information and Communication Technology Infrastructure

Information and Communication Technology (ICT) Infrastructures are a mean to support all
(or almost all) the transactions among partners in a CNO. Actually, this is one of the
conditions to work as such. In the context of VE evolution and decision-making, ICT
infrastructures are responsible for providing the necessary functionalities to allow partners
in making all the previously mentioned tasks: partners’ discussion, methodological
guidance, decision protocols, performance measurement and monitoring and performance
evaluation.
Security is a crucial issue to provide the required trust building in CNOs. Sowa and
Sniezynsky (2007) developed a security framework that controls information access
dynamically according to partners’ roles in a VE. This guarantees that all the sensible
information can be accessed only by authorized partners. Yet, that the information comes
from recognized and authenticated partners and sources.
Rabelo et al. (2008) developed an integrated, web-based and on-demand ICT infrastructure
devoted to cope with CNO requirements. Although it currently does not have implemented
all those necessary functionalities to support the VE evolution phase (at least in the way the
problem is approached in this work), it is opened to receive new functions.
The combination and some adaptations in all these mentioned works are seen as a feasible
starting point to support the envisaged distributed and collaborative decision support
scenario.
3. Collaborative decision support for the virtual enterprise evolution

Previous section has presented the scenario for the management of VE evolution associated
to the collaborative decision making as well as the aspects to support it. In order to cope
with all of them, a framework has been conceived. This framework gathers such aspects and
groups them into four categories, or pillars: Human, Organizational, Knowledge and
Technological. The essential rationale of these four pillars is to enable humans to discuss and
to decide about a problem related to a given organizational process, applying a set of
organizational procedures and methods, using information and knowledge available in the
VBE’s repositories, all this supported by a sort of ICT (technological) tools and infrastructures
(Drissen-Silva & Rabelo, 2009a). That discussion is framed by a decision protocol (conceived
using project management foundations) and is carried out within a distributed and
collaborative decision support environment. The decision protocol is the mechanism which
“links” the four pillars according to the particular problem to be solved within the VE
evolution phase. Figure 2 shows the framework.
Fig. 2. Framework for VE Evolution Management.

The Human pillar represents people, i.e. the VE companies’ managers who use their tacit
knowledge and collaborative attitude to help solving the problem came from the VE
operation phase. It embraces the empowered managers and the experts that can
(re)configure the decision protocols. The Organizational pillar comprises intra and inter-
enterprises processes, ontologies as well as working methods, techniques and procedures
that should be involved in a distributed and collaborative decision-making process. It
embraces companies’ business processes themselves, project management, performance
measurement and performance evaluation methods and techniques, and decision
procedures and rules to be followed. The Knowledge pillar comprises explicit information
and knowledge that are available in the VBE’s repositories and that managers can have
access to for helping in the decision-making process. This embraces lessons learned, best
practices as well as information about partners, regulations, historical data, etc. The
Technological pillar refers to all kind of ICT tools, platforms and security artifacts that should
be available to support managers in managing the processes accessing suitable methods. It
embraces OLAP, BPM, simulators and groupware tools, besides databases, ontology
management systems and the general computing infrastructure for systems deployment,
security, communication, interoperation and services management.
It is important to point out an issue about the human pillar. SMEs have many difficulties in
terms of management skills, whereas working in collaborative networks requires several
levels of additional preparedness (Afsarmanesh & Camarinha-Matos, 2005). In fact, it seems

unrealistic to assume that VE partners are already prepared and know the most relevant
managerial and performance evaluation techniques and methods that can help them during
the discussions and decision-making. If on one hand their experience and knowledge are of
extremely importance for that, on the other hand they are insufficient for dealing with all
the intrinsic complexity that managing the VE evolution represents. Therefore, in order to
effectively support the use of the proposed framework, it is essential that VE partners are
also empowered with adequate training. Klen et al. (2008) has proposed a methodology for
training VBE members relying on governance and individual competences on VE
management.
The proposed approach relies on and combines two basic areas: Project Management (PM)
and Decision Support Systems (DSS). One of the most important framework’s elements is a
decision protocol. It corresponds to a mechanism that coordinates the problem solving and
that is based on an adaptation of the ECM (Engineering Change Management) model
(Rozenfeld et al., 2006) for agile and change management. This protocol has the aim of
guiding decision-makers towards more effective solutions in a methodological way. In
essence, this all aims at offering a Collaborative and Distributed DSS for VE’s partners to get
together to discuss about necessary changes but guided by a decision protocol that consider
the most relevant VE characteristics. A set of performance evaluation and knowledge
mechanisms completes the framework, providing a previous analysis before decisions are
implemented. A database model saves all the discussed information for further auditing.
The management of collaborative projects deals within distributed environments. Activities
and process are distributed through partners and organizations on different locations and
countries, with different cultures but the management could be done in centralized or
distributed way (Ollus et al., 2009). Collaborative work has been imposing the conception of
a new kind of tools for supporting its management offering lessons learned and knowledge
for future decisions (Loss, 2007). Differently from extended enterprises where there is a
dominant enterprise (O’Neill, 1995), managing the VE evolution implies to consider that all
partners are autonomous and have to participate in the decision making process
transforming the evolution management a complex process.
3.1 Framework architecture

The four framework’s pillars are operated through three concrete elements: the decision
protocol, the distributed and collaborative decision support computing environment, and
the ICT Toolbox. They all form the Distributed Collaborative Decision Support System for
the Management of VE Evolution (DDSS-VE). Figure 3 presents the framework architecture,
also illustrating the relation of the elements with the pillars. Yet, it shows the three different
types of actors that are involved in the discussions about the problem detected in the VE
operation. To be highlighted the fact that all transactions – involving both humans and
systems – are carried out over computing networks making use an adequate ICT supporting
infrastructure.
3.1.1 Decision protocol

The decision protocol is a sequence of steps that defines the activities that have be executed
in given situations within a given context to solve a problem. Conceptually, it should
indicate what has to be done, why, by whom, where, when, how, and with which resources.
Fig. 3. Framework Architecture.

The conception of the proposed protocol has considered three aspects: its generality, its
underlying foundation, and its execution automation. As far as the generality is concerned,
the protocol is not seen as a reference protocol that would be generic enough to comprise all
possibilities of how every single different problem should be solved by/at certain
companies related to a certain VE. Instead, it is seen as a basis on which particular protocols
can be derived, grounded on project management reference models, considering the VBE
policies and operation rules. Figure 4 shows the proposed decision protocol.
This particularization means that new steps can be added, some modified / adapted and
some disabled (Figure 5). The whole approach can be seen under three layers: basis protocol,
specific protocol, and computer aided. As said before, the basis protocol layer for the VE
evolution phase is the one showed in the figure 4, where the box outside the main square
contains activities within the VE operation phase. The specific protocol layer represents the one
that would have been customized for a given VBE and that would be effectively applied in
the VEs created from it. The computer aided layer contains digital information repositories
and very concrete ICT tools and infrastructure that are used to support the diverse actions in
a decision-making process. This is made available via an ICT toolbox (see section 3.1.3).
A modification in the basis protocol is however not hard coded made. Thanks to a BPM tool,
the protocol is flexibly modeled and directed connected to software services that execute the
protocol’s steps themselves. Therefore, if a modification is required, the user modifies the
processes at BPM level (or even at the SOA level), but not at programming code level.
However, a protocol particularization has some restrictions. This refers to the second aspect
of its design, which is the underlying foundation. Actually, the steps of the basis protocol
comprehend the most typical ones presented in the changes management reference models,
and ECM in particular (see section 2.2). Thus, users are not allowed to change its essential
logical structure (the macro steps Need of Change Identification, Change Proposal, Change
Planning, and Implementation, as well as some of their sub-steps). Besides using ECM and
adapting it to the VE evolution context, this work has also used some ideas proposed in
O’Neill (1995) when determining the most significant events to handle in more strategic
decisions. In resume, this proposed decision protocol represents the mentioned framework’s
methodology and it is modeled via BPM and SOA-based tools.
Fig. 4. Basis Protocol for the VE Evolution Management.
Fig. 5. Multi-layered conceptual scenario of the proposed framework.

3.1.2 Partners’ discussion environment

This second element of the framework corresponds to the Distributed Collaborative
Decision environment and it is the main element to support partners’ discussion over the
network.
It is important to point out that VE partners are fundamentally composed of SMEs.
Therefore, it is important to offer an easy and low cost way to support the access to
management methods, techniques and tools to help the involved people in the discussions,
performance measurement and evaluation.
This underlying vision of this environment relies on the assumption that there is no sense to
develop a wizard-like, expert systems or agent-based decision support systems solution that
are used for modeling closed world problems (e.g. Baffo et al., 2008). Instead, partners
should have room – with a methodological support and integrated within the companies’
business process environment – for exchanging ideas, exercising their creativity, and
reasoning about particular cases based on the very current status of the involved companies.
In this environment, the actors involved in a discussion are (figure 3) the VE partners: 1) the
VE coordinator, who owns the business and who is, at last, responsible for it; 2) the VE
members, who are the companies’ representatives in the given VE and; 3) an invited expert
(e.g. the broker, a specialized technician, a VBE’s representative), an ad-hoc member who
may participate in the discussions and whose role is defined for each case.
This environment is controlled by the DDSS-VE. Based on different classifications for
decision support systems (Turban & Aronson, 1998; Phillips-Wren & Forgionne, 2001) this
model for distributed decisions support system with argumentation and moderation for the
VE evolution (DDSS-VE) is of type:
 Negotiation: decision about a problem is reached via a negotiation process, where the
reaching of the solution involves relaxations of constraints and changes in the plan;
 Decentralized: the VE coordinator coordinates the discussion but the decision itself
emerge from the discussions;
 Partially hierarchical: the VE coordinator has the power to validate the final decision
achieved after a (non-hierarchical) discussion;
 Multi-stage: a decision can be reached after several rounds of discussion;
 With semi-structured tasks: the problem and related information is partially made
available by the DDSS-VE system and VBE’s information repositories, and the
discussion is generally assisted. The other part of the information and knowledge come
from the tacit knowledge of the own participants;
 Multi-participant: several members can participate simultaneously in the discussion;
 Team-based: although autonomous and independent, VE members act collaboratively
as they share the same goal.
In order to give an overview on how the framework works, figure 6 illustrates an abstract
discussion scenario to be supported by the DDSS-VE where partners would exchange their
opinions about a given problem. Actually, DDSS-VE will manage the interaction among
three entities. One entity is the companies’ representatives, each one having a DDSS-VE’s
graphical interface to interact with. Another entity is the set of ICT and network
infrastructure, tools (the ones common to all VBE members, and the local ones, accessible
only by each company) and VBE’s information repositories (see next section). The third
entity is the decision protocol, which will help guiding the discussions.
After the problem has been detected, DDSS-VE starts the protocol steps (Figure 4), within
the Need of Change Identification phase. In this phase the goal is to identify the problem
reasons and to check if it can be solved by the own partner, without impacting the other VE
members. This reveals the strategy to involve the other partners only if the problem cannot
be solved at a “local” level. For this, the VE coordinator and the partner that has generated
the conflict (illustrated as Partner 1) discuss together (e.g. via chat and file transfer), initially.
After discussions and evaluations, if the problem is considered solved without needing the
other partners, the protocol’s flow goes through another phases, the Change Proposal,
Planning and Implementation phases. In the case the problem could not be solved, it is
necessary to evaluate which partners were affected and that should then be involved in the
collaborative discussion and decision-making. In the Change Proposal phase, the discussion
is supported by the services that combine the ideas of HERMES and Delphi methods (see
section 2.1). The part inspired in HERMES aims to organize partners’ arguments in a concise
structure, using an appropriate semantic, communicating their suggestions but in a
compiled way, including an association of weights to the most important arguments. This
aims at finding the better (and faster) consensus about the problem. The part inspired in the
Delphi method aims at avoiding direct confrontations among participants, which could
generate counterproductive discussions. In this sense, all the arguments are gathered by the
VE Coordinator who, in a first moment, acts as the moderator selecting, deleting, changing
or suggesting changes in the arguments received before they can be published to all
participants. Actually, it is not the aim to restrain partners conversation and information
exchange, but rather to guarantee a faster discussion and, mainly, that some sensible
information (e.g. the precise level of capacity of a given partner) can be disclosed to
everybody. In this way, the VE coordinator have the option to just say to the others that the
given partner has “enough” capacity. This discussion round, with the compiled opinions, is
illustrated as gray frames in figure 6, at each member’s side. The white frames illustrate the
argumentation console where partners expresses their opinions as well as where the VE
coordinator receives them. He moderates the discussion via this console. After the
arguments have been sent out to the other participants, they can reevaluate their
considerations and make other suggestions. This process continues until a consensus is
reached (within the Change Planning phase).
The protocol is not fixed in its inner actions. Regarding VE uniqueness and topology, and
the natural partners’ heterogeneity, the protocol can be different for each situation. There
are many possible scenarios that could influence the decision to be taken in order to solve
the current problem. In this way, the protocol acts as a reminder of some more important
questions so that partners can recall they should check them. For example, if an item is
delayed and the final customer is very important or the fine is too high, partners can agree
on subcontracting part of the production in order to keep the delivery date. If the client has
a very rigorous quality control and he manages the suppliers’ certification level quite
tightly, perhaps is not possible to hire any company, but one equivalent, and so forth. In the
case of any other particular issue, partners should handle this, managed by the VE
coordinator. Once the problem is solved, the new VE’s parameters are set up
(Implementation phase) and the control flow goes back to the VE operation phase.
This hypothetical argumentation scenario would be based on the results achieved helped by
a pool of tools for performance evaluation modeling, monitoring and tasks rescheduling,
which can also involve the invited expert’s opinion (Change Planning). Some participants
could use their own tools or the common toolbox (including the access to the VBE database)
available to all participants to help in the discussions.
Fig. 6. Illustration of the collaborative decision support environment.
3.1.3 The tool box

Traditionally, SMEs have many difficulties to access, use and maintain software, mainly due
to its costs and to the required expertise to do that. The toolbox concept was introduced in
(Bernhard, 1992) with the goal of providing a pool of industrial software to help users from
all departments of a single company to implement the CIM philosophy (Computer
Integrated Manufacturing).
This concept was largely extended in Rabelo et al. (2008) through the development of a web
based distributed ICT infrastructure (ICT-I) devoted to CNOs. The access to ICT-I is totally
made in the form of services, which are invoked either by the user or by other software
services. Besides integrating many CNO supporting tools, it provides the access to the VBE’s
information repositories. These tools cover the VE creation (Afsarmanesh et al., 2008) and
operation phases (Negretto et al., 2008). However, there are no specialized services for the
VE evolution and dissolution phases yet.
Taking the ICT-I scalability facilities into account, the framework for the VE evolution and
associated protocol was added to and seen as another class of services of ICT-I. This also
involves some non directed CNO-related services, such as simulators, spread sheets, CSCW,
assisted methods and other supporting tools that help VE members along the protocol
execution. This all corresponds to the computer aided layer illustrated in figure 5.
In this work, these supporting tools are group into a logically centralized repository of ICT
tools called ICT Toolbox. ICT Toolbox is therefore a pool of common tools that are accessed
via ICT-I – hence via the network – facilitating members’ acceptance and use of
management methods. This however does not cover the existing local tools used by each
member at their companies. The Toolbox’s tools themselves can congregate both the set of
tools previously agreed (or existing) in the VBE and tools that can be accessed on demand
from other providers.
4. Prototype implementation
This section presents the results of the implementation of the DDSS-VE framework, which is
concentrated in three different functionalities: the Decision Protocol, the Partners’
Discussion Environment and a Tool for previous evaluation scenarios. The decision
protocol once started will help manager to do actions in the right moment in the decision
making process. It was used an adapted VBE database in order to access the competences of
all partner in the usage scenario. Partner’s Discussion Environment is implemented
considering ideas from HERMES System and Delphi method, applying a collaborative
discussion with voting and comparing suggestions all on supervision by the moderator. The
Toolbox is populated with a tool for capacity planning using the performance evaluation
method applied in advanced dashboards. Within a controlled testing environment, the
problems detected in the VE operation phase are manually introduced and the discussions
are simulated in a distributed scenario using a number of PCs.
As already said, the Collaborative Discussion Environment has the goal to combine
HERMES system and Delphi method, and to adapt them to the desired decision philosophy.
In other words, it aimed at facing the partners’ autonomy and transparency requirements as
well as the need for a more structured way of deciding. The main adaptations include:
 The creation of a moderator (role), who is responsible to evaluate and to make available
the arguments sent by members. Depending on the case, the moderator can be the own
VE coordinator;
 The comparison of two different arguments using different connectors (better than;
worse than; equal to; as bad as; as good as). Each comparison assigns negative and/or
positive points to each argument, depending on the connector;
 Voting: Partners can vote pro or against to each argument;
 During the discussion, partner are guided by the Decision Protocol;
 Is possible to use a previous evaluation decision tool, in order to evaluated the impact
of a new scenario into the VE operation.
4.1 Usage scenario

In order to evaluate the collaborative discussion using the DDSS-VE, a VE scenario has been
created. This VE would be responsible to develop a new helmet style for racing, involving
four partners from different countries (Drissen-Silva & Rabelo, 2009b).
Considering the decision protocol showed in figure 4, it is assumed that the phase “Need of
Change Identification” has been passed. Figure 7 illustrates in a general way how the
discussion would try to solve the conflict resolution from the protocol‘s phase “Change
Proposal” on. All this have been implemented in a web portal, on top of Liferay web
application server (www.liferay.com). In this example, the VE Coordinator (Mr. Ricardo) has
concluded that it is necessary to start a discussion with two members (Mr. Marcus and Mr.
Rui) due to a problem detected in the specification of the first lot. After starting the
collaborative discussion, the protocol gets in the “Changing Planning” phase where
different scenarios are evaluated using tools form the tool box. “Changing Planning“ phase
ends when the best alternative has been chosen in the “Implementation” phase, where the
new scenario is put on practice. The sequence described below quickly explains figure 7.
1. Starting the discussion (to be conducted via the DDSS-VE):
 The protocol ask some questions to delineate the better attitude for each case (e. g. if it is a
rigorous client constraint that avoids from choosing another supplier);
 Each participant can use some tools to preview which different scenarios could be acceptable
to reschedule the activities that have to be done, choosing the best one, and publishing it as a
suggestion for the problem resolution:
a. Mr. Rui posts the first suggestion: ‘Buy from another supplier’ (Figure 7a);
b. Each partner can vote pro or against it (bottom Figure 7a);
c. Each suggestion can be compared with other suggestions using ‘COMPARE’ button
(Figure 7a). Figure 7b presents the list of suggestions and the possible logical
connectors. For example, a comparison using ‘is better than’ as the connector assigns
+1 point to the best suggestion and -1 to the worst;
d. Figure 7c shows a tree (associated to the detected problem: helmet strip allotment) with
the three posted suggestions (plus authors) and four comparisons among them. One of
them is not yet evaluated as it is ‘awaiting approval’;
 The moderator (Mr. Ricardo) evaluates the different suggestions and the comparisons,
mainly to see if there is some confrontation among the participants:
a. Figure 7d shows the Moderator’s view. He can modify and/or simply approve Mr.
Rui‘s opinion (“RE: buy from another supplier is as good as …”) and send them to the
group;
b. Figure 7e represents the vision seen by the other two members before Mr. Rui‘s
opinion approval. Thus, they only see ‘message awaiting approval‘;
 In what the final voting result is concerned:
a. It is possible to see the number of votes of each suggestion, which is +3 in relation to
the Mr. Rui’s one (Figure 7a), also meaning that the three consulted members
(including the VE coordinator) have agreed on it;
b. Figure 7c shows a signaled number beside each suggestion expressing the final sum of
voting with the weights of comparisons. In this case, ‘Buy from another supplier’ has
more positions in favor (+3 from direct voting) that is added to more 2 points from two
positive comparisons, resulting 5 points in favor;
2. Once agreed, the most suitable solution is settled on the VE plan and partners (re)start to work
based on it. This means that VE evolution is ended and the VE management goes back to the
operation phase.
4.2 Previous evaluation tool for decision making

Performance evaluation needs the selection of the most important factors for the best
system’s performance. For each factor is necessary to set some levels (in terms of numbers)
they could assume. In a manufacture environment the factors could be machines or
employees, for example, and levels could be the quantity each one could be available. The
performance evaluation could indicate which is the most important factor in the system’s
performance effect.
In order to offer a tool for previous evaluation of the decision’s impact using performance
evaluation it was developed a module adequate to the conceptual model previous
described. This tool uses different spreadsheets compounding a dashboard that offers the
possibility to see each partner’s competence, production scheduling, available resources,
number of resources looking for the integrations of the scheduling in order to calculate
another scenario for solving the problem in the discussion on DDSS-VE. Figure 8 shows the
developed dashboard.
Fig. 7. Some snapshots of the Partner’s Discussion Environment.
5. General evaluation
The developed prototype passed through a sequence of exhaustive tests for the verification
and validation of the conceptual model for the collaborative discussion around a problem
emerged in the Virtual Enterprise operation phase forcing it to go on the evolution phase.
The conceptual model and the prototype were evaluated by experts on the main areas
studied in the model development process. In the evaluating average all experts agreed in
the contribution, relevance and attending the needs of the scientific problem to be solved: ‘to
find a more transparent and collaborative environment that puts autonomous partners in a
discussion around the partnership conflict using a set of computable tools and a decision
protocol to support the decision’. The methodology used to evaluate this work followed
three main steps: i) prototype evaluation in a sequence of stressed tests; ii) explanation the
conceptual model by a scientific article to the experts with a number of questions answered
with their evaluation; iii) explanation the prototype functionalities in an example execution
with another number of questions answered with their opinions.
Fig. 8. Previous Evaluation Scenarios Tool using Dashboards for Tasks Rescheduling.
5.1 Contributions
Main scientific contribution of this work is centered in using different techniques, tools and
methods already acceptable in an adequate semi-automated system that help managers in
the decision making process around a problem in the VE operation phase. The integration of
those different methods can offer a distributed and collaborative discussion with
transparency, controlled by moderation using previous analysis of the decision’s impact.
Central element is the human, who has the ability to feel and to decide what is the best
scenario respecting his knowledge. The framework can only support his decision offering
flexibility, calculus tools and communication availability through partners.
Compared with the state-of-the-art in the area, this work covers different aspects, which are
showed in the Table 1.
Considering the flexibility offered by the decision protocol, this framework could be
adapted to other strategic alliances models and also to the management of virtual
organization operation phase, only making the necessary modification on the some phases
and processes in the base protocol in order to attend different cases needs.
5.2 Limitations
The main limitation of this work is related to CNO concept that assumes each partner are
autonomous and has to participate in a collaborative way trying to help other partners in
difficulties. Some aspects related to VE concept is difficult to reach in the reality because
trust among partners has to be strong, and also it is necessary a well developed ICT
infrastructure to put this environment on work. But on the other hand, there are a number
of VBE in execution in the world that feed expectative of a strong dissemination of the VE
concepts to these kinds of enterprises collaborative environment.
CNOs / VEs
Traditional CNOs / VEs
(proposed
Management Model (current approaches)
approach)
Decision Centralized Centralized Decentralized
Information sharing
No or eventual Yes Yes
between partners
Transparency in the
No or partial Partial Yes
decision
Decision quality
No Low and Eventual Yes
evaluation
Decision scope Intra-organizational Inter-organizational Inter-organizational
Flexible /
Decision process Inflexible / “Data Inflexible / “Data
Systemized /
Rigidity flow” flow”
Adaptable
Information
integration between Low / Medium Medium / High High / Very high
partners
Trust between Explicit /
Implicit Explicit
partners Reinforced
Good global results
Decision objective Best global results Good global results with previous
analysis
Full-fledged
Mutual help between Punctual
Cooperation Collaboration along
partners Collaboration
decision making
Methodological aid / Low efficiency and
No or partial Yes
Assisted decision without assistance
Source: Adapted from Drissen-Silva & Rabelo, 2008.
Table 1. Comparison between traditional management model, current and proposed
CNOs/VEs approaches.
Considering the prototype it was developed only with one tool for supporting previous
impact analysis of the decision, but the conceptual model can consider a big number of
available tools those could be put in a collaborative access environment for all partners.
5.3 Future research

Considering the high complexity of the problem presented in this work, there were another
themes to be researched to better develop the ideas described in the Distributed and
Collaborative Decision Making Environment for the Virtual Enterprise Evolution (DDSS-
VE), for example:
 Development of a model that consider aspects of hierarchy, power and governance
between VBE and VE partners. A model that also consider the moderator’s competence
and his position during the decision process;
 Adequate the collaborative discussion environment, that uses ideas from HERMES
system and Delphi method to the Moodle system;
 Creation of an ontology that describes formally the relations, hierarchies and concepts
associated to the explored domain on decision making in the Collaborative Networked
Organizations (CNO).
6. Conclusion
This chapter has presented a framework to support a collaborative discussion among VE
members for solving problems during the VE evolution phase. It is essentially composed of
a decision protocol, a distributed and collaborative decision support system, and of ICT
supporting tools and communication infrastructure. It was designed to cope with the VE
requirements, mainly in what members’ autonomy and decision transparency is concerned.
Developed based on project management methodologies, discussions are guided and
assisted by the system but preserving and counting on the members’ experience and
knowledge in order to reach a suitable/feasible solution for the given problem.
The proposed framework groups such requirements and organizes them into four pillars:
Human, Organizational, Knowledge and Technological. The essential rationale of these four
pillars is to enable humans to discuss and to decide about a problem related to a given
organizational process, applying a set of organizational procedures and methods, using
information and knowledge available in the VBE’s repositories, supported by a sort of ICT
tools. A crucial aspect in the proposed approach is the human intervention, i.e. the problem
is so complex that is unfeasible to try to automate the decisions. Instead, the approach is to
put the managers in the centre of the process, surrounding them with adequate tools and
methods.
All the framework’s elements operates in a methodological way by the human element, on
a democratic, transparent, decentralized, systematized and moderated basis, considering
their geographical distribution.
In order to offer more quality in the suggestions made by each partner during a discussion
around a problem resolution, different tools, techniques and methods for performance
evaluation are offered to provide a vision for a future capacity planning in order to evaluate
different scenarios for solving the problem in discussion. In this way, the participants have
conditions to previously evaluate the impact of the decision to be taken. This evaluation can
be made isolated by each participant during the conflict resolution process.
A software prototype has been implemented to evaluate the framework, and it was tested in
an open but controlled environment. The implementation copes with the required flexibility
and adaptability of the decision protocol to different VEs, applying BPM (Business Process
Management) e SOA (Service Oriented Architecture) technologies as a support for. The
developed framework fundamentally assumes that VE partners are all members of a kind of
cluster of companies. This presupposes the presence of a reasonable degree of trust among
members, of an adequate computing infrastructure, of common organization vision (in
terms of collaboration and enterprise networking) and operational procedures to be
followed when problems take place, and that VE managers are trained for that.
The implementation results have showed that the proposed mechanisms for supporting
partners’ autonomy, Internet-based decentralized decision-making, voting and transparency
have worked out in a controlled environment. During the discussions, selected partners can
have access to the problem, can freely exchange opinions about how to solve it, and can
express their preferences via voting. This guarantees that the solution emerges from the
collaboration and trust among partners. The decision protocol helps participants to take the
right action on the right moment. The scenarios evaluation tools is capable to offer a pre-
evaluation of the decision impact.
This work’s evaluation was composed of a set of procedures that offers conditions to affirm
the final general research conclusion: “A semi-automated decision protocol, flexible and adaptable,
integrated with scenarios analysis tools and a collaborative discussion environment makes better the
quality and trust in the decision around a problem in a VE”.
7. Acknowledgment
This work is partially supported by CNPq – The Brazilian Council for Research and
Scientific Development (www.cnpq.br). The authors would like to thanks Ms. Cindy
Dalfovo, Mr. Leonardo G. Bilck and Mr. André C. Brunelli for the software implementation.
8. References
Afsarmanesh, H. & Camarinha-Matos, L. M. (2005). A Framework for Management of Virtual
Organization Breeding Environments. Proceedings 6th IFIP Working Conf. on Virtual
Enterprises, Kluwer Acad. Publishers, pp. 35-48.
Afsarmanesh, H.; Msanjila, S.; Erminova, E.; Wiesner, S.; Woelfel, W. & Seifert, M. (2008).
VBE Management System, in Methods and Tools for Collaborative Networked
Organizations, Eds. L.M. Camarinha-Matos, H. Afsarmanesh and M. Ollus,
Springer, pp. 119-154.
Baffo, I.; Confessore, G. & Liotta, G. (2008). A Reference Model for Distributed Decision Making
through a Multiagent Approach, in Pervasive Collab. Networks, Eds. L.M.
Camarinha-Matos and W. Picard., Springer, pp. 285-292.
Baldo, F.; Rabelo. R. J. & Vallejos, R. V. (2008). Modeling Performance Indicators’ Selection
Process for VO Partners’ Suggestions, in Proceedings BASYS’2008 – 8th IFIP Int. Conf.
on Information Technology for Balance Automation Systems, Springer, pp. 67-76.
Bernhard, R. (1992). CIM system planning toolbox, CIM-PLATO Project Survey and
Demonstrator, in Proceed. CIM-PLATO Workshop, Karlsruhe, Germany, pp. 94-107.
Camarinha-Matos, L. M.; Afsarmanesh, H. & Ollus, M. (2005). ECOLEAD: A Holistic
Approach to Creation and Management of Dynamic Virtual Organizations. In:
Collaborative Networks and Their Breeding Environments. Eds. L. M. Camarinha-
Matos, H. Afsarmanesh e A. Ortiz. Springer, pp. 3-16.
Bostrom, R.; Anson, R. & Clawson, V. (2003). Group facilitation and group support systems.
Group Support Systems: New Perspectives, Ed. Macmillan.
CMMI (2006). CMMI for Development Version 1.2. Tech. Report DEV, V1.2. Pittsburgh:
Carnegie Mellon – Software Engineering Institute.
Dalkey, N. C. & Helmer, O. (1963). An experimental application of the Delphi method to the case
of experts, Management Science; 9, pp. 458-467.
Drissen-Silva, M. V., Rabelo, R. J., (2008). A Model for Dynamic Generation of Collaborative
Decision Protocols for Managing the Evolution of Virtual Enterprises, in Proceedings
BASYS’2008 – 8th IFIP International Conference on Information Technology for
Balance Automation Systems, Springer, pp. 105-114.
Drissen-Silva, M. V. & Rabelo, R. J. (2009a). A Collaborative Decision Support Framework for

Managing the Evolution of Virtual Enterprises, International Journal of Production
Research, Vol. 47, No. 17, pp. 4833-4854.
Drissen-Silva, M. V. & Rabelo, R. J. (2009b). Managing Decisions on Changes in the Virtual
Enterprise Evolution, in Proceedings PRO-VE’ 2009 –, Leveraging knowledge for
Innovation in Collaborative Networks, Springer, pp. 463-475.
Grefen, P.; Mehandjiev, N.; Kouvas, G.; Weichhart, G. & Eshuis, R. (2009). Dynamic business
network process management in instant virtual enterprises, Computers in Industry; 60
(2), pp. 86-103.
Hodík, J. & Stach, J. (2008). Virtual Organization Simulation for Operational Management, in
2008 IEEE CSM Int. Conf. on Distributed Human-Machine Systems, Czech
Technical University in Prague, ISBN 978-80-01-04027.
Jain, R. (1991). The Art of Computer Systems Performance Analysis: Techniques for Experimental
Design, Measurement, Simulation and Modeling. New York: John Wiley & Sons, Inc.
Jansson, K. & Eschenbaecher, J. (2005). Challenges in Virtual Organisations Management –
Report on methods for distributed business process management. Tech. Report D32.1.
ECOLEAD – European Collaborative networked Organizations LEADership
initiative. FP6 IP 506958.
Johnsson, J. & Johansson, B. (2003). Discrete Event Simulation in a Virtual Enterprise
Environment: a Case Study of Multiple Developers, 15th European Simulation
Symposium and Exhibition, October 26-29, Delft, the Netherlands.
Kaplan, R. S. & Norton, D. P. (1997). The Strategy in Action. [in Portuguese], Rio de Janeiro:
Campus.
Karacapilidis, N. & Papadias, D. (2001). Computer supported argumentation and collaborative
decision making: the HERMES system, Information Systems; 26 (4), pp. 259-277.
Karvonen, I.; Salkari, I. & Ollus, M. (2005). Characterizing Virtual Organizations and Their
Management. In: Collab. Networks and Their Breeding Environments. Eds. L. M.
Camarinha-Matos, H. Afsarmanesh and A. Ortiz. U. States: Springer; pp. 193-204.
Klen, E. R.; Pereira-Klen, A. A. & Loss, L. (2008). Selection of a virtual organization coordinator,
in Collaborative Networks: Reference Modeling, Eds. L.M. Camarinha-Matos and
H. Afsarmanesh, Springer, pp. 297-310.
Lechtenborger, J. & Vossenm G. (2003). Multidimensional normal forms for data warehouse
design, Information Systems 28 (5), pp. 415-434.
Leite, M. M. (2004). Implementation requirements of CRM strategies in SMEs: an approach based
on Project Management [in Portuguese], PhD Thesis, Federal University of Santa
Catarina.
Loss, L. (2007). A Framework for Collaborative Networked Organizations Learning: An Approach
Based on Organizational Learning and Knowledge Management [in Portuguese], PhD
Thesis, Federal University of Santa Catarina.
Camarinha-Matos, H. Afsarmanesh and P. Novaes, Springer, pp. 243-252.
Military Handbook (2001). Configuration Management Guidance. MIL-HDBK-61A(SE) Dept. of
Defense –USA.
Moon, S. W.; Kim, J. S. & Kwon, K. N. (2007). Effectiveness of OLAP-based cost data management
in construction cost estimate, Automation in Construction; 16 (3), pp. 336-344.
Muller, E.; Horbach, S. & Ackermann, J. (2008). Decentralized Decision Making in Non-
Hierarchical Networks, in Pervasive Collaborative Networks, Eds. L.M. Camarinha-
Matos and W. Picard., Springer, pp. 277-284.
Negretto, H.; Hodík, J.; Mulder, W.; Ollus, M.; Pondrelli, P. & Westphal, I. (2008). VO
Management Solutions: VO management e-services, in Methods and Tools for
Collaborative Networked Organizations, Eds. L.M. Camarinha-Matos, H.
Afsarmanesh and M. Ollus, Springer, pp. 257-274.
Ollus, M.; Jansson, K.; Karvonen, I.; Uoti, M. & Riikonen, H. (2009). On Services for
Collaborative Project Management. In: Leveraging Knowledge for Innovation in
Collaborative Networks, Eds. Luis M. Camarinha-Matos, Iraklis Paraskakis and
Hamideh Afsarmanesh, Springer, pp. 451-462.
OŃeill, H. (1995). Decision Support in the Extended Enterprise, Ph.D. Thesis, Cranfield
University, The CIM Institute.
Ordanini, A. & Pasisni, P. (2008). Service co-production and value co-creation: The case for a
service-oriented architecture (SOA), European Management Journal; 26 (5), pp. 289-
297.
PMBOK (2004). A Guide to the Project Management Body of Knowledge. PMI Standards
Committee.
Pěchouček, M. & Hodík, J. (2007). Virtual Organisation Management eServices version 1. Tech.
Report D34.5. ECOLEAD Project, www.ecolead.org.
Phillips--Wren, G. E. & Forgionne, G. A. (2001). Aided Search Strategy Enable by Decision
Support, Information Processing and Management, Vol. 42, No. 2, pp. 503-518.
Piccard, W. (2007). Support for power adaptation of social protocols for professional virtual
communities, in Establishing the Foundation of Collaborative Networks, Eds. L.M.
Camarinha-Matos, H. Afsarmanesh and P. Novaes, Springer, pp. 363-370.
Rabelo, R. J.; Pereira-Klen, A. A.; Spinosa, L. M. & Ferreira, A. C. (1998). Integrated Logistics
Management Support System: An Advanced Coordination Functionality for the Virtual
Environment, in Proceedings IMS'98 - 5th IFAC Workshop on Intelligent
Manufacturing Systems, pp. 195-202.
Rabelo, R. J.; Pereira-Klen, A. A. & Ferreira, A. C. (2000). For a Smart Coordination of
Distributed Business Processes, in Proceedings 4th IEEE/IFIP Int. Conf. on Balanced
Automation Systems, Berlin, Germany; pp. 378-385.
Rabelo, R. J. & Pereira-Klen, A. A. (2002). A Multi-agent System for Smart Co-ordination of
Dynamic Supply Chains, in Proceedings PRO-VE´2002; pp. 312-319.
Rabelo, R. J.; Pereira-Klen, A. A. & Klen, E. R. (2004). Effective management of dynamic supply
chains, Int. J. Networking and Virtual Organisations; Vol. 2, No. 3, pp. 193–208.
Rabelo, R. J.; Castro, M. R.; Conconi, A. & Sesana, M. (2008). The ECOLEAD Plug & Play
Collaborative Business Infrastructure, in Methods and Tools for Collaborative
Networked Organizations, Eds. L.M. Camarinha-Matos, H. Afsarmanesh and M.
Ollus, Springer, pp. 371-394.
Rozenfeld, H.; Forcellini, F. A.; Amaral, D. C.; Toledo, J. C.; Silva, S. L.; Alliprandini, D. H. &
Scalice, R. K. (2006). Products Development Management – A Reference for Process
Improvement. [in Portuguese] – 1st edition - São Paulo: Saraiva.
Sowa, G. & Sniezynski, T. (2007). Configurable multi-level security architecture for CNOs.
Technical Report Deliverable D64.1b, in www.ecolead.org.
9
A Lean Balanced Scorecard

Using the Delphi Process:
Enhancements for Decision Making
Chuo-Hsuan Lee1, Edward J. Lusk1 and Michael Halperin2
1The State University of New York at Plattsburgh
2The Wharton School, University of Pennsylvania
USA
1. Introduction
Kaplan & Norton’s Balanced Scorecard (BSC) first appeared in the Harvard Business Review
in 1992 (Kaplan and Norton, 1992). It described in general terms a method by which
management may improve the organization’s competitive advantage by broadening the
scope of evaluation from the usual Financial dimension to include: the organization’s
Customer base, the constitution and functioning of the firm’s Internal Business processes, and
the necessity of Innovation and Learning as a condition for growth. Over the years, these four
constituent elements of the BSC have remained largely unchanged, with the exception of a
modification in 1996 when Innovation and Learning was changed to Learning and Growth
(Kaplan and Norton, 1996).
As the BSC is now in its second decade of use, there have been a number of articles
suggesting that the BSC is in need of refocusing. Lusk, Halperin & Zhang (2006) and Van
der Woerd, F. & Van den Brink (2004). This refocusing suggests two changes in the Financial
Dimension of the BSC. First, the Financial Dimension of the BSC needs to be broadened from
measures that only address internal financial performance to those that are more market-
oriented. For example, according to the Hackett Group (2004: 67), the majority of Balanced
Scorecards are “out of balance because they are overweight with internal financial
measures”. Second, there seems to be a tendency to add evaluation variables with little
regard for their relationship to existing variables. This “piecemeal approach” results in a
lack of coherence and often causes Information Overload. This tendency is underscored by
Robert Paladino (2005), former vice president and global leader of the Telecommunications
and Utility Practice for Norton’s company, the Balanced Scorecard Collaborative,
http://www.bscol.com/, who suggests that a major failure of the BSC is that many
organizations adopt a piecemeal approach that generates an information set consisting of
too many highly associated variables without independent associational linkages to the
firm’s evaluation system. The observation is consistent with one of the classic issues first
addressed by Shannon (1948) often labelled as the Shannon-Weaver Communication Theory
which in turn lead to the concept of Information Overload.
Information Overload is particularly germane, because from the beginning of the Internet
Era, management has succumbed to the irresistible temptation to collect too many financial
performance variables due to the plethora of data sources offering simple downloads of
hundreds of financial performance variables. For example, Standard & Poors™ and
Bloomberg™, to mention a few, have data on thousands of organisations for hundreds of
variables.
In this study, we answer the call for refocusing the Financial Dimension of the BSC. The
purpose of this study is twofold. First, as suggested by Lusk, Halperin & Zhang (2006) we
add two market-oriented variables - the standard CAPM Market Beta and Tobin’s q to
broaden the financial dimension of the BSC. And second, we suggest, in detail, a simple
modelling procedure to avoid the “Information Overload and Variable Redundancy” so
often found in the Financial Dimension of the BSC. Although this study is focused only on
the Financial Dimension of the BSC, the modelling process that we suggest in this study
may be also applied to the other dimensions of the BSC.
This study is organized as follows:
1. First, we suggest a simple procedure for generating a “lean” i.e.,—parsimonious,
variable characterisation of the firm’s financial profile and then use that lean or
consolidated variable set as an input to a standard Delphi process.
2. Second, we present an empirical study to illustrate the lean BSC modelling system.
3. Lastly, we offer some concluding remarks as to use of our refined information set.
2. Our simple procedures leading to parsimony in the BSC

As an overview, as discussed above the BSC has been criticized as being “too endowed”
with financial performance variables due to the convenience of downloading financial
variables using the standard databases. Simply put, many of these financial variables are
merely expressing the same firm process characterization but in a different variable voice.
To deal with this “redundancy” issue, we are following the suggestion of Jones (2006) who
recommends factor analysis, essentially based upon the standard Harmon (1960) Factor
model [SHFM], as the best technique to develop non-overlapping categorizations of impact
variables for risk assessment. His model forms the basis of the Information Risk
Management module of the Certified Information Security Manager manual (2009, Ch 2).
The essential idea is that factor analysis reduces the “over-endowed” variable space to its
factor equivalent; this will achieve parsimony in the variable space and thus enable firm
decision-makers to better understand their processes as characterisations of independent
factors. This factor reduction thus addresses the issues of Variable Redundancy and
Information Overload discussed above. Consider now how factor analysis and
benchmarking can be used to develop the Lean BSC.
2.1 Three variable contexts of factor analysis and the Delphi Process: our lean
modelling system
In the process of using factor analysis to develop lean variable set characterizations, we
suggest performing factor analyses in three variable contexts: (1) the Industry, (2) a
Benchmarked Comparison Organisation and (3) the Particular Firm. These three variable
contexts1 taken together help management refocus on variables that may be productively
1The above lean modelling framework and benchmarking procedure was used by one of the authors as
a member of the Busch Center of the Wharton School of the University of Pennsylvania, the consulting
arm of the Department of Social Systems Science, as a way to focus decision-makers’ attention on
differences between their organisation and (1) the related industry as well as (2) a selected firm in the
A Lean Balanced Scorecard Using the Delphi Process: Enhancements for Decision Making 173
used in planning and executing the navigation of the firm. Specifically, the feedback from
these three variable contexts will be organized as a Delphi Process as the information
processing logic of the BSC. Consider now the details of the Information Processing logic for
the BSC as organized through the Delphi process.
2.2 The two-stage Delphi Process in our lean modelling system

Because the principal information link in the Delphi Process is generated by factor analysis,
the decision makers [DM] must have a practical understanding of the output of the factor
model—i.e., a working familiarity with the output of the factor model is essential to the
successful creation of a Lean BSC. Accordingly, we suggest the Delphi Process to be
achieved in the following two stages.
2.2.1 Stage 1: unfreezing stage –familiarize the decision makers with “factors”
The goal of the Unfreezing stage is to have the DM feel comfortable with the logic of factor
analysis as a statistical technique that groups BSC Variables into Factors. To effect the
Unfreezing stage we recommend using an intuitive example dealing with the simple
question: What is a Computer? It is based upon an example, the data of which is presented
following in Table 1, used by Paul Green and Donald Tull (1975) in their classic text on
marketing research; they used it to illustrate the logic of Factor Analysis.
Advanced Maximum
Computers Basic Minimum Add-Ons Cycle
Processing Non-swap
n=15 Processing Storage Non-buffer Time
Storage
1 -0.28 -0.36 -0.49 -0.52 -0.48 -0.27
2 3.51 3.61 -0.55 -0.6 -0.87 3.74
3 -0.39 -0.34 -0.55 -0.53 -0.59 -0.27
4 -0.06 -0.28 -0.55 -1.07 -0.83 -0.26
5 0.38 -0.27 -0.46 -0.50 -0.88 -0.27
6 -0.43 -0.38 -0.55 -0.52 -0.48 -0.27
7 -0.26 0.37 -0.55 -0.52 -0.59 -0.27
8 0.70 0.68 -0.60 -0.61 -0.92 -0.27
9 -0.47 -0.39 -0.37 -0.52 -0.48 -0.27
10 -0.28 -0.23 -0.02 -0.14 -0.77 -0.27
11 -0.49 -0.39 -0.13 0.16 1.71 -0.27
12 -0.50 -0.39 1.32 2.47 1.08 -0.27
13 -0.51 -0.39 0.36 0.16 1.70 -0.27
14 -0.51 -0.39 3.26 2.47 1.08 -0.27
15 -0.52 -0.12 -0.13 -0.23 1.33 -0.27
Table 1. The Green and Tull Computer Factor Dataset
industry used as a positive benchmark—that is, a best practices case that the study firm would like to
emulate. Sometimes, a negative benchmark was also used—in this case an organisation from which the
study firm wished to distance itself.
To start the Unfreezing process, we suggest presenting to the DM the following information
processing task:
Assume that you do not know anything about Computers, and you wish to better understand the
essential functioning of Computers. Unfortunately, as you are not gifted “technically, it is not
possible for you to reverse-engineer a computer and tinker with it so as to understand its essential
features. So, you are left with an “empirical” or analytic option. Therefore, with the aid of your tech-
friends you collect six performance Variables on 15 different models of computers: (1) Basic
Processing Speed, (2) Advanced Processing Speed, (3) Minimum Storage, (4) Maximum Non-
Swap Storage, (5) Add-Ons Non-buffer Capacity, and (6) Cycle Time. So now you have a
dataset with 15 computers measured on six performance variables—i.e., a matrix of size 15 rows and
6 columns as presented in Table 1.
As we continue with the Unfreezing stage, we distribute to all the DM a copy of the dataset
in Table 1 for anchoring their processing of the variable reduction from this dataset. We, the
group conducting the Unfreezing, enter the dataset into the Standard Harmon Factor Model
[SHFM] statistical program; the results produced are presented in Table 2. We recommend
that Table 2 be displayed either: (1) on the computers of the DM, (2) on a projection screen
for all to see or (3) distributed to the DM as printed copies. They need to be able to see and
discuss the information in Table 1 and the factor results as presented in Table 2.
Variables Factor A Factor B

Basic Processing 0.96 -0.22
Advanced Processing 0.98 -0.16
Minimum Storage -0.07 0.94
Maximum Non-Swap Storage -0.07 0.96
Add-on Non Buffer -0.25 0.75
Cycle Time 0.98 -0.06
Table 2. Two Factor Rotation of the Six Variables of the Computer Dataset
Using the above information and the usual Factor Definition Rule that: any variable score
greater than 0.71 is an important variable descriptor of that factor—here bolded in Table 2, we
discuss with the DM that the six variables really describe just two Factors2. Factor A is
described predominately by the variables: Basic Processing, Advanced Processing and Cycle
Time all of which have variable scores greater than 0.71. These are all measures of Speed.
Factor B, on the other hand, is described predominately by Minimum Storage, Maximum Non-
Swap Storage and Add-on Non Buffer which are all measures of Storage. The results in Table
2 provide simple and intuitive information which speaks in a straightforward way to the
question which started the Unfreezing analysis “What is a Computer?” The DM now
recognizes and, of course, probably knew at the outset that Computers are devices that have
Speed of Processing and Storage Capacity as their essential profiling characteristics. Also, they
2 We do not discuss with the DM the 0.5 loading rule because this concept requires a relatively
sophisticated understanding of the mathematical statistics of factor models. Rather we use the simple
loading rule that if a Variable had a weight of 0.71 or greater then that variable is a meaningful
descriptor of that Factor. In our experience, this level of detail is sufficient for the DM to use the results
of the Factor Model. This does assume, of course, that there will be someone in the firm with sufficient
training in statistical modeling to actually use the factor software that is to be used in the Delphi
process. This is usually the case.
understand, via the comparison of the data in Table 1 and the results in Table 2, that there
were not really six variables but rather two variables: Speed and Storage with each of them
measured in three different ways.
In summary, to wind-down the Unfreezing stage, we emphasize that the initial set of six
variables was by definition “over-weight” or redundant and so the six variables were
consolidation by the SHFM to form only two dimensions: Speed and Storage in the lean-
version characterization of the dataset.
This simple computer example is critical to the understanding of the use of factor analysis to
reduce variable redundancy and so deals with “over-weight” variable characterizations. We
find that this simple and intuitive example unfreezes the DM in the sense that they knew
that computers were fast storage computation devices and this is exactly what the factor
analysis shows; this reinforcing their belief that the Factor Analysis can be “trusted” to
replicate the reality of the associated variable space with fewer variables.
2.2.2 Stage 2: Factor analysis in three contexts and brainstorming

Following the first or Unfreezing Stage, the next stage is where we engage the Delphi
process. We recommend that the Delphi process be used in its EDI-mode (See Jung-Erceg,
Pandza, Armbruster & Dreher (2007)) where the DM discuss, in a Chat-Loop-Context, the
various information sets until they are satisfied with their insights and then they propose
the Action Plan for the firm derived from the Lean-BSC.
Specifically we recommend that the firm, given its understanding of the Mission, Goals and
Objectives, engages the Delphi Process to generate the Lean-BSC using the following steps:
1. The firm will identify the set of DM who intend to navigate their firm using the BSC.
2. This group of DM will select the longitudinal panel consisting of (i) a sample of firms
from their industry, possibly all, and (ii) a particular firm that could be a positive or
negative benchmark—meaning that the DM judge the benchmark to be a firm that they
wish to emulated or a firm from which they wish to distance themselves.
3. Then the DM will select a Comprehensive Variable Set [CVS] that they believe are the
firm performance variables that can be used to best profile their firm. In our experience
the best profiling variables may include those financial performance variables that (i)
are simple to measure, (ii) have operational measures that are sensitive indicators of
change, and (iii) are themselves considered as direct measures of effects relative to the
Mission, Goals and Objectives of the firm. This will be important in constructing the
necessary reward linkages which is one of the principal reasons to use the BSC. The
CVS, itself, is very likely to be a variable set that the firm has been using in the past and
so may be characterized as the “over-weight” variable set. This is not a problem as the
intention at this stage is to incorporate all the variables that the DM feel to be important.
We expect that the CVS will be formidably large.
4. The “Overweight” CVS will be inputted to the SHFM so as to consolidate the variable
set to the factors—i.e., the lean variable set. This process will be repeated for the
Industry and the Firm-benchmark respectively (i.e., the other two contexts of factor
analysis).
5. And finally, these three lean-variable sets, one for: the Firm, the Industry and the Firm-
benchmark, will be sent to the Chat-Loop-Context-Delphi-Space and the DM will begin
the convergent process of deciding the Financial Action Plan that will be integrated into
the full BSC evaluation process. This process will lead to the final action plan of the firm
considering all four of the BSC dimensions.
To enrich understanding of how these five steps will be used for a particular firm, we will
now present a detailed and comprehensive example of all the steps that are needed to create
the Lean BSC.
Study Firm: A.D.A.M. Inc.

(ADAM). [http://www.adam.com] The principal activity of A.D.A.M.,Inc is to provide
health information services and technology solutions to healthcare organizations, group
insurance brokers, employers, consumers, and educational institutions. The products of the
Group are used for learning about health, wellness, disease, treatments, alternative
medicine, anatomy, nutrition and general medical reference in both the healthcare and
education markets. The products contain physician-reviewed text, in-house developed
medical graphics and multimedia to create health information that offers visual learning
experience. The Group provides information on annual licensing agreements to healthcare
organizations, Internet websites and educational institutions.
Benchmark: Amdocs
(DOX). [http://www.amdocs.com/Site/AmdocsCom.htm] Amdocs,Inc is a leading
provider of customer care, billing and order management systems for communications and
Internet services. Amdocs has an unparalleled success record in project delivery of its
mission-critical products. With human resources of over 5,900 information systems
professionals, Amdocs has an installed base of successful projects with more than 75 service
providers throughout the world. In April 2000, Amdocs completed the acquisition of Solect
Technology Group Inc., a leading provider of customer care and billing systems for IP
providers.
Table 3. Brief Profiles of the Study Firm and its Benchmark
3. Illustrative example – pre-packaged software: SIC 7372

We next present an illustrative example of the BSC Delphi Factor Procedure [Delphi BSC]
using an actual firm: A.D.A.M. Inc. in the SIC 7372. It is not our intention to suggest that
this selected dataset speaks to actual recommendations drawn from the BSC analysis as this
is clearly the domain of the DM of the firm using the Delphi BSC. We offer this example as a
detailed illustration of the guidance through the process. To this end we, the authors, have
assumed the role of the DM for A.D.A.M. Inc. and will (1) discuss our reaction to the
information that we have generated using the Delphi BSC and (2) how this may be used to
develop the BSC navigation information for the sample firm. Our assuming the roles of
decision makers is only to illustrate the possible functioning of the Delphi BSC—i.e., our
Lean-BSC navigation recommendations for the actual firm that we have selected are not
normative in nature.
3.1 Selection of three lean variable contexts

We assume the role of DM for A.D.A.M. Inc. (Ticker: ADAM; NASDAQ) which is in the
software and related devices industry in the SIC: 7372. The principal activity of our
company, A.D.A.M. Inc., is to provide health information services and technology solutions
to healthcare organizations, group insurance brokers, employers, consumers, and
educational institutions. Our positive benchmark is Amdocs (Ticker: DOX; NYSE), a leading
provider of customer care, billing and order management systems for communications and
Internet services (See Table 3 above for a brief description of these firms and their URL-
links). Our Industry Benchmark includes all 7372 SIC Firms that had data reported in
COMPUSTAT from 2003 to and including 2006. We selected this time period as it was after
Sarbanes-Oxley: 2002, and before the lead-up to the sub-prime debacle in 2008, which we
view as event partitions of reported firm data. This accrual yielded 411 firms and netted a
variable-panel of 1,161 observations.
3.2 Selection of a comprehensive variable set [CVS]

We, the DM for A.D.A.M. Inc., began the Delphi BSC process by selecting from the extensive
COMPUSTAT™ menu 15 variables which we believe are useful in portraying the financial
profile of market traded organizations. Further, we have used these variables in
characterizing our operating profile in that, over time, these variables have been
instrumental change variables for us, and have been used in the evaluation of our firm.
These DM-selected variables are presented in Table 4. The seven (7) variables that were not
downloaded from COMPUSTAT™, but rather calculated from COMPUSTAT™ variables
are noted in boldface. The computation of these variables is detailed in the Table 5. The
other definitions are available from COMPUSTAT™. Finally, we downloaded Beta from
CRSP™. Therefore, in total, we have 23 variables as our financial profile, two of which are
Tobin’s q and the CAPM β as suggested by Lusk, Halperin & Zhang (2006).
Cash & Short-Term Diluted EPS before

Net Sales Current Ratio
Investments extraordinary items
Market Value: Fiscal
Receivables Total Net Income (Loss) Quick Ratio
Year End
Number of Accounts
Cash from
Cash Common Shares Receivable
Operations
outstanding Turnover
Depreciation &
Current Assets Net PPE Tobin’s Q
Amortization
Current Liabilities Beta Gross Margin EPS Growth
Total Assets Cost of Goods Sold ROA
Table 4. The 23 Judgmental Variables Selected by the Decision-makers of A.D.A.M. Inc.
Effectively we are saying that these 23 variables would be important in profiling our
organization as we play the role of DM for A.D.A.M. Inc.; other DM as well as other firms
may, and probably will, select other variables. This is a positive feature of the Delphi BSC in
that it provides the needed idiosyncratic flexibility in selecting the variables that will be
inputted into the Factor Analysis, and so constitutes the judgmental factor set critical in the
analysis.
3.3 The results of the factor study

All of the results of the factor study reported in Tables 6, 7 and 8 were created using the
SHFM; the standard Varimax rotation on the Pearson correlation matrix, as programmed in
JMP of the SAS Institute, version 6.0 (see Sall, Lehman & Creighton, 2005). The number of
factors selected was the number of factors in the un-rotated factor space—i.e., the correlation
matrix—for which the eigenvalue was greater than 1.0. Finally, variable loadings greater
than √0.5 were used in the description of the factors. For ease of reading, the variable
loadings are presented only to the second decimal, and those loading greater than √0.5 are
bolded. For the industry factor analysis, we excluded A.D.A.M. Inc. and Amdocs as they are
the study firm and the benchmark firm respectively.
Tobin’s q =(A25xA199+A130+A9)/A6
Current Ratio =A4/A5
Quick Ratio =(A1+A2)/A5
ROA =(A172-A19)/A6
EPS Growth =((A57t-A57t-1)/|A57t-1|
Gross Margin =A12-A41
Accounts Receivable Turnover =A12/((A2t+A2t-1)/2)
Where:
A1- Cash and short term investments
A2- Receivables
A4- Current Assets
A5- Current liabilities
A6- Total assets
A9- Total long-term liabilities
A12- Net sales
A19- Preferred Dividends
A25- Common shares outstanding
A41- Cost of goods sold
A57- Diluted earnings per share excluding extraordinary items
A130- Preferred stock-carrying value
A172- Net income (loss)
A199- Common Stock Price: Fiscal year end
Table 5. Computation of Ratios Based on COMPUSTAT Data
3.3.1 The Delphi Process judgmental interpretation of the representative variables for
the various factors
We, in the role of DM for A.D.A.M. Inc. were guided by the factor loading results and also
by those variables that did not load across the final factors. These latter variables are
interesting in that they are independent, in the strong scene, as they did not achieve
association in the rotated space greater than √0.5. We will note these non-associative
variables in the Tables 6, 7 and 8 using Bold-Italics. Therefore, we will have two groups of
variables that exhaust the Factor/Variable Space: Those variables that have loaded on a
factor such that the rotate loading is greater than √0.5 and those that did not exhibit such an
association. Both groups have guided our interpretation of the Delphi BSC.
3.3.2 The industry factor profile and its relation to the BSC
The industry Factor analysis is presented in Table 6. We remark that both Beta and Tobin’s q
do not align in association with any of the COMPUSTAT™ financial performance profile
variables. This suggests that relative market volatility and stockholder preference are
independent measures for the Pre-Packaged Software industry—in and of itself an interesting
result. One strong implication of Table 6 for us as DM for A.D.A.M. is that insofar as the BSC
analysis is concerned the industry is a mixed portfolio with both disparate hedge and market
sub-groupings. See two recent articles that treat these topical relationships in the hedge fund
context: Vincent (2008) and Grene (2008). Confirmatory information is also
Beta 0.02 0 0.01 0 -0.1 0 0.03 0.98 0
Tobin's Q 0.02 0.11 0.1 0.03 0 0.04 0.99 0.03 0.03
Current Ratio 0 0.99 0.01 0 0.03 0.07 0.06 0 0.01
Quick Ratio 0 0.99 0 0 0.04 0.06 0.06 0 0
ROA 0.11 0.08 0.23 0.2 0.92 0.04 0 -0.1 0.15
Gross Margin 0.99 0 0.08 0 0.04 0.02 0.03 0 0.04
A/R Turnover 0.02 0.12 0.05 -0.1 0.03 0.99 0.04 0 0
EPS Growth 0.01 0 0.07 0.97 0.17 -0.1 0.03 0 0.12
Cash 0.94 0 0.05 0 0.04 0 0.01 0.02 0.03
Cash & Short-term
0.89 0.05 0.17 0.03 0.02 0.07 0 0.04 0.03
Investment
Cash from operations 0.98 0 0.05 0 0.04 0.02 0.05 0 0.06
Receivables Total 0.96 0 0 0.01 0.04 -0.1 0 0.01 0.04
Current Assets 0.96 0.02 0.14 0.02 0.03 0.04 0 0.02 0.03
Current Liabilities 0.98 -0.1 0 0 0.03 0 0 0.01 0.02
Total Assets 0.98 0 0.06 0.01 0.03 0 0 0.03 0
PPE Net 0.94 0 0.1 0.01 0.03 0 0 0.01 -0.1
Net Sales 0.98 0 0.09 0 0.04 0.04 0.01 0 0.04
Depreciation &
0.90 -0.1 0.2 0.07 0.03 0.02 -0.1 0.02 -0.2
Amortization
Common Shares
0.94 0 -0.2 0 0.02 0 0.04 0.01 0.06
Outstanding
COGS 0.82 0 0.09 0 0.03 0.08 0 0.01 0.05
Net Income(Loss) 0.92 0 0.01 0.01 0.05 0.03 0.08 0 0.18
Diluted EPS 0.10 0 0.48 0.29 0.34 0 0.06 0 0.73
Market Value- Fiscal Year
0.22 0.01 0.92 0.06 0.2 0.06 0.11 0.02 0.16
end
Table 6. Industry Factor Analysis: Mid-Range Year Randomly Selected 2004
provided by the fact that EPS Growth and Market Value are independent variables with
respect to the other factor defined variables. For this reason, we note: It will be important to
understand that A.D.A.M. Inc., our firm, does not have to compete on a market relative basis—i.e.,
against the industry as a portfolio. One possible reaction to this, that the decision-makers may
decide, is that it would be useful in a strategic planning context to partition the industry into
various profile groupings and then re-start the Delphi BSC using these industry sub-
groupings as additional benchmarks. We have opted for the other approach that is to use
the Amdocs benchmark and continue with the Delphi BSC.
3.3.3 The study firm—A.D.A.M., Inc. and the selected benchmark: Amdocs
We will now concentrate on the relative analysis of the study firm and Amdocs, our positive
benchmark. One of the underlying assumptions of this comparative analysis is the stability
of the factors in the panel. As we have an auto-correlated panel for these two firms, there is
no statistical test for stability for the particular factor arrangement that was produced. If one
seeks to have a demonstration of this stability then a simple boot-strapping test will give the
required information. See Tamhane and Dunlop (2000, p. 600). For our data test, we
conducted a relational test for this data set and found it to be stable over the two following
partitions: years 2005 and 2006 compared to 2003 to 2006 and this argues for factor stability.
Beta -0.81 0.55 0.20

Tobin's Q -0.59 0.75 0.30
Current Ratio -0.82 0.52 -0.24
Quick Ratio -0.84 0.49 -0.23
ROA -0.42 0.91 -0.03
Gross Margin 0.98 0.22 0.00
A/R Turnover 0.88 -0.19 -0.43
EPS Growth -0.88 0.47 0.00
Cash 0.88 -0.40 -0.26
Cash & Short-term Investment 0.42 0.91 -0.02
Cash from Operations 0.11 0.86 0.50
Receivables Total 0.92 0.11 0.38
Current Assets 0.65 0.75 0.06
Current Liabilities 0.98 0.16 0.12
Total Assets 0.98 0.20 -0.01
PPE Net 0.99 0.10 -0.05
Net Sales 0.98 0.22 -0.01
Depreciation & Amortization -0.54 -0.84 -0.10
Common Shares Outstanding 0.92 0.40 0.02
COGS 0.97 0.22 -0.03
Net Income(Loss) -0.08 0.99 -0.09
Diluted EPS -0.13 0.99 -0.07
Market Value-Fiscal Year End 0.28 0.96 0.09
Table 7. Factor Analysis: A.D.A.M., Inc. [ADAM]
As part of the Delhi process as it relates to the interchange of information among the DM so
as to create the BCS navigation we, the authors, using the information in Tables 6, 7 and 8
exchanged our ideas as to the interpretation of the information generated by the factor
analyses as presented in Tables 6, 7 and 8 by email and also in face-to-face meetings. This
was done over a two week period; after that time we felt that we had closure and developed
the following two Observations: One using the information in Table 7 above for A.D.A.M
and One using the information in Table 8 following for Amdocs. Finally from these two
observations we developed the Navigation information for A.D.A.M. Inc. Consider these
Observations and the related Navigation Imperatives next.
Observation I: Consider these relative to A.D.A.M. Inc.
For our organization, A.D.A.M. Inc., in the first column of Table 7 (i.e., Factor 1) we note that
Beta is inversely associated with the direct—i.e., not computed—balance sheet variables
representing the resource configuration of the firm such as: Total Assets, Net PPE, and Total
Receivables. This suggests that more resources are associated with lower market volatility.
Further, we observe a positive association between Gross Margin (i.e., Net Sales less COGS)
and certain balance sheet variables such as Total Assets and Liabilities, but we do not
observe any relationship between Reported Net Income [RNI] and these balance sheet
variables, implying that although our resource configuration effort is influencing the gross
margin it is not aligned with our bottom line RNI improvement. It is also interesting to note
that Factor 1 and Beta are not associated with Factor 2 which seems to be best characterized
as the growth dimension of A.D.A.M. Inc. For the second factor, our RNI movement is
consistent with our cash management (Cash & Short-term Investments) and assets
management (ROA) as well as our potential to grow as measured by Tobin’s q.
Beta -0.94 -0.22 0.25
Tobin's Q 0.99 -0.07 0.16
Current Ratio -0.05 -0.99 -0.09
Quick Ratio -0.20 -0.98 -0.10
ROA 0.84 -0.50 -0.24
Gross Margin 0.98 0.05 0.17
A/R Turnover 0.30 -0.95 0.09
EPS Growth -0.73 0.67 -0.09
Cash -0.48 0.69 -0.54
Cash & Short-term Investment -0.96 -0.02 -0.28
Cash from Operations 0.71 0.70 0.07
Receivables Total 0.97 0.08 0.24
Current Assets 0.34 0.88 -0.32
Current Liabilities 0.03 0.99 0.09
Total Assets 0.91 0.34 0.23
PPE Net 0.37 0.81 0.45
Net Sales 0.99 0.01 0.16
Depreciation & Amortization 0.62 0.30 0.73
Common Shares Outstanding -0.48 0.86 0.17
COGS 0.99 -0.01 0.16
Net Income(Loss) 0.98 -0.20 -0.01
Diluted EPS 0.97 -0.23 -0.01
Market Value-Fiscal Year End 0.96 0.19 0.21
Table 8. Factor Analysis: Amdocs,Inc. [DOX]
Observation II : Relative to Amdocs,Inc.
In comparison, Table 8 presents a very different profile of Amdocs, Inc., our positive
benchmark. The first factor of Table 8 has many of the same resource configuration aspects
except that both Beta and Tobin's q are featured in Factor 1 as well as is RNI! This strongly
indicates that the resource configuration is aligned linearly with the income producing
potential. In this way, Amdocs, our positive benchmark, seems to have worked out the asset
employment relation to RNI that is not found for our firm, A.D.A.M. Inc. Further, Beta is
inversely associated with the “income machine” as reflected by Tobin's q, RNI and Net Sales
for Amdocs meaning that the more assets the more they are converted linearly to Reported
Net Income and this has a dampening effect on Market volatility. This certainly is a
positive/desirable profile and is probably why Tobin's q is positively associated with this
profile; we do not see this for A.D.A.M. Inc. as Tobin’s q is only associated with growth and
is invariant to Beta and the resources employed as discussed above.
3.3.4 BSC implications: input to the navigation plan

Navigation Imperative I
The benchmarking with Amdocs, Inc. suggests that if we, A.D.A.M. Inc., value the
performance configuration as profiled in Table 8 of Amdocs Inc., then we should endeavour
to align our resource employment with Reported Net Income (RNI) generation. Based on the
benchmarking results, we seem to be successful in utilizing resources to improve Gross
Margin but fail to make our resource configuration deliver with respect to the bottom line:
Reported Net Income [RNI] as we observed for Amdocs Inc. our positive benchmark.
To align our resource employment with Net Income, we need to examine our policies as
they relate to project acceptance and management to make sure that we effectively manage
our resources to achieve a higher net income, leading to a higher return on assets (ROA).
Accordingly, the principal Financial Profiling action we need to take is to: Attend to ROA.
This objective can be accomplished in a variety of ways—either by generating relatively
abnormally high RNI but doing so by moderately increasing the Asset generating base or by
achieving average or below average RNI but doing so with a highly efficient Asset
configuration—i.e., a reduction in the asset base.
For example, consider an example taken from the A.D.A.M. Inc. website
http://www.adam.com/ we learn that we have developed:
“The A.D.A.M. Symptom Navigator for the iPhone that is an innovative web application designed
and developed specifically for the iPhone, and runs using Apple’s Safari browser. This unique tool
allows consumers to quickly and easily find important information about health symptoms, all with
just a few taps on their iPhone. With a home screen icon branded specifically to your organization,
the application also allows your facility to be seen as a leading provider of health information and
services among today’s growing market of mobile users.”
Let us assume that there were two projects which have equal expectation on profitability—
i.e., RNI. However, the iPhone Symptom Navigator has double the ROA compared to, let us
assume, a PDA-Telemetry Download System—i.e., Project B. In this case, given the Delphi
BSC navigation information generated, the DM would prefer the iPhone project on the basis
of ROA.
Navigation Imperative II
Our Tobin’s q, which acts as a proxy of growth potential perceived by the market
participants, is aligned with the ROA and RNI but cannot line up with our resource
configuration variables such as total assets, receivables and net sales. In other words, if we
cannot use our resources in a way to boost our RNI and ROA, we will fail to improve our
growth profile as well in the eyes of the market participants. Accordingly, this missing
connection between our resource configuration variables (e.g., our assets) and our Tobin’s q
(i.e., our growth potential) indicates that the principal Market Profiling action we need to
take is to: Manage our Growth Profile While Attending to ROA. This action requires us to
(1) consider new investment projects that will not only improve short-term financial
performance such as ROA but also offer strategic profiling information for informing the
market, and (2) pay attention to the impact of the new investment projects as they impact
our recorded book value as we consider the return from the new investment projects.
To continue with the simple two project example, if two projects are the same in terms of
profitability (e.g., ROA), we should consider the project that will offer richer strategic
opportunities. Also, if these two projects permit various asset contracting possibilities such
as Operating Leases, Capital Leasing, or Purchasing and there are relatively wide ranges for:
Useful Economic Life, Resale Market Valuation as well as the methods of depreciating the
asset, then we should consider the project that will have the more favorable impact on our
recorded book value given the same performance in profitability.
4. Conclusion
To “close-the-loop” we wish to note that the Financial Dimension information developed
from the Delphi BSC or Lean Modeling approach will be included in the BSC along with
information on the other three dimensions of the BSC: Customers, Internal Processes, and
Learning and Growth. These BSC dimensional encodings are the input to the firm DSS needed
to develop priority information for the various projects that the firm may be considering—
i.e., project prioritization is the fundamental reason that firms use the BSC or budgeting
models. This is to say that the BSC information is “intermediate” information in that this
BSC information will be used to characterize the projects that are the planned future of the
firm. The final task in the Delphi BSC process then is selecting from the BSC-encoded
projects the actual projects to be funded by the firm.
In this regard, to conclude our research report, we wish to note a simple way that one may
“prioritize” the navigation information from all four dimensions of the BSC as they are
encoded in the projects. For example, in our simple illustrative example where we have
focused on the Financial Dimension, we have proposed that there were two projects and the
iPhone Project dominated on both of the criteria variables: ROA and Growth Profile
Management. It is more likely the case that there will be multiple criteria that result from
the Delphi BSC and that the preference weights considering all four of the BSC dimensions
will be distributed over the various projects so that there would be no clear dominance. To
determine the actual project-action plan from the output of the Lean-version of the Delphi
BSC is indeed a daunting task even with a relatively small number of performance criteria.
In this regard, it is necessary for the firm to select a method to determine the final set of
projects in which the firm will invest so as to satisfy the Overall BSC navigational
imperatives for the firm.
There are many such preference or priority processing methods. Based upon our consulting
work, we prefer the Expert Choice Model™ [http://www.expertchoice.com/] developed by
Saaty (1980 and 1990). We find this Alternative/Project Ranking methodology to be the
simplest, easiest to communicate to the DM, and most researched Ranking Model according
the average number of annual citations in the literature. As a disclose note: we have no
financial interest in Expert Choice Inc. the firm which has developed the application software
for the Expert Choice Model™.
5. Acknowledgements
We wish to thank the following individuals for their detailed and constructive comments on
previous versions of the paper: The participants in Technical Session 9 of the 2010 Global
Accounting and Organizational Change conference at Babson College, Babson Park, MA
USA, in particular Professor Ruzita Jusoh, University of Malaya, Malaysia; the participants
at the 2009 National Meeting of the Decision Science Institute, New Orleans LA, USA in
particular Professor Sorensen, University of Denver, and additionally Professor Neuhauser
of the Department of Finance, Lamar University, Beaumont Texas, USA and Professor
Razvan Pascalau of the Department of Finance and Economics, SUNY; Plattsburgh,
Plattsburgh, NY, USA.
6. References
Certified Information Security Manager. (2009). CISM Review Manual, ISACA, Rolling
Meadows, IL
Green, P. E. & Tull, D.S. (1975). Research for Marketing Decisions, Upper Saddlewood, NJ,
Prentice Hall
Grene, S. (December 2008). Clarity to Top Agenda. London: Financial Times, 01.12.2010
Hackett Company Report. (2004). Balanced Scorecards: Are their 15 Minutes of Fame Over?
The Hackett Group Press, Report AGF/213
Harmon, H. H. (1960). Modern Factor Analysis, University of Chicago Press, Chicago, IL
Jung-Erceg, P.; Pandza, K.; Armbruster, A. & Dreher, C. (2007). Absorptive Capacity in
European Manufacturing: A Delphi study. Industrial Management + Data Systems,
Vol. 107, No. 1, pp. 37-51
Jones, J. A. (2006). Introduction to Factor Analysis of Information Risk. Risk Management
Insight [online], Available from
http://www.riskmanagementinsight.com/media/documents/FAIR_Introduction.
pdf
Kaplan, R & Norton, D. (1992). The Balanced Scorecard—Measures that Drive Performance.
Harvard Business Review, Vol. 83, No. 7/8, pp. 71-79
Kaplan, R. & Norton, D. (1996). Linking the Balanced Scorecard to Strategy. California
Management Review, Vol. 39, No. 1, pp. 53-79
Lusk, E.J.; Halperin, M.& Zhang, B-D. (2006). The Balanced Scorecard: Suggestions for
Rebalancing. Problems and Perspectives in Management, Vol. 4, No. 3, pp.100-114
Paladino, R. (2005). Balanced Forecasts Drive Value. Strategic Finance, Vol. 86, No. 7, pp. 36-
42
Saaty, T. L. (1980). The Analytic Hierarchical Process, McGraw-Hill, New York, NY
Saaty, T. L. (1990). How to Make a Decision: The Analytic Hierarchy Process. European
Journal. of Operational Research, Vol. 48, No. 1, pp. 9-26
Sall, J.; Lehman, A. & Creighton, L. (2005). JMPTM Start Statistic-Version 5, 2nd Ed, Duxbury
Press, Belmont CA
Shannon, C. (1948). A Mathematical Theory of Communication. Bell System Technical Journal,
No. 27 (July and October), pp. 379-423, 623-656, Available from
http://plan9.bell-labs.com/cm/ms/what/shannonday/shannon1948.pdf
Tamhane, A. C. & Dunlop, D. D. (2000). Statistics and Data Analysis, Prentice Hall, Upper
Saddle River, NJ
Van der Woerd, F. & Van den Brink, T. (2004). Feasibility of a Responsive Scorecard-A Pilot
Study. Journal of Business Ethics, Vol. 55, No. 2, pp. 173-187
Vincent, M. (2008). Ivy League Asset Allocation Excites Wall Street. London: Financial Times,
8 April: 9.
10
Linking a Developed Decision Support System

with Advanced Methodologies for Optimized
Agricultural Water Delivery
Kristoph-Dietrich Kinzli, David Gensler and Ramchand Oad
Florida Gulf Coast University, Middle Rio Grande Conservancy District,
and Colorado State University
United States of America
1. Introduction
Water is the lifeblood of the American West and the foundation of its economy, but it
remains its scarcest resource. The explosive population growth in the Western United States,
the emerging additional need for water for environmental uses, and the national importance
of the domestic food production are driving major conflicts between these competing water
uses. Irrigated agriculture in particular is by far the largest water user of diverted water –
80% country wide and 90% in the Western U.S – and since it is perceived to be a
comparatively inefficient user, it is frequently asked to decrease its water consumption. The
case of the Middle Rio Grande illustrates the problem very well. The Rio Grande is the
ecological backbone of the Chihuahuan Desert region in the western United States, and
supports its dynamic and diverse ecology, including the fish and wildlife habitat. The Rio
Grande Silvery Minnow is a federally listed as endangered species, and irrigated agriculture
in the Middle Rio Grande has come under increasing pressure to reduce its water
consumption, while maintaining the desired level of service to its water users.
Irrigated agriculture in the Western United States has traditionally been the backbone of the
rural economy. The climate in the American West with low annual rainfall of 10 to 14 inches
is not conducive to dry land farming. Topography in the West is characterized by the Rocky
Mountains which accumulate significant snowfall, and the peaks of the snowmelt
hydrograph are stored in reservoirs allowing for irrigation throughout the summer crop
growing season. Of the total available surface water irrigated agriculture uses roughly 80 to
90% (Oad and Kullman, 2006). The combined demands of agriculture, urban, and industrial
sectors in the past have left little water for fish and wildlife. Since irrigated agriculture uses
roughly 80 to 90% of surface water in the West, it is often targeted to decrease diversions.
Due to wildlife concerns and demands from an ever growing urban population, the
pressure for flow reductions on irrigated agriculture increases every year. In order to sustain
itself and deal with external pressure for reduced river diversions irrigated agriculture has
to become more efficient in its water consumption. This chapter focuses on research
regarding improving water delivery operations in the Middle Rio Grande Conservancy
District system through the use of a decision support system linked with advanced
infrastructure, methodologies, and technology.
The Middle Rio Grande (MRG) Valley runs north to south through central New Mexico
from Cochiti Reservoir to the headwaters of Elephant Butte Reservoir, a distance of
approximately 175 miles. The MRG Valley is displayed in Figure 1.
Fig. 1. Middle Rio Grande Valley (Barta, 2003)

The valley is narrow, with the majority of water use occurring within five miles on either
side of the river. The bosque, or riverside forest of cottonwood and salt cedar, is supported
by waters of the Rio Grande; the bosque being surrounded by widespread irrigated farming.
The Cities of Albuquerque, Rio Rancho, Belen and several smaller communities are located
in and adjacent to the MRG Valley. Although the valley receives less than 10 inches of
rainfall annually, it supports a rich and diverse ecosystem of fish and wildlife and is a
common outdoor resource for communities in the region. Water supply available for use in
the MRG Valley includes: native flow of the Rio Grande and its tributaries, allocated
according to the Rio Grande Compact of 1938; San Juan-Chama (SJC) project water, obtained
via a trans-mountain diversion from the Colorado River system; and groundwater. Water is
fully appropriated in the MRG Valley and its utilization is limited by the Rio Grande
Compact, which sets forth a schedule of deliveries of native Rio Grande water from
Colorado to New Mexico and from New Mexico to Texas (Rio Grande Compact
Commission, 1997). Water demand in the MRG Valley includes irrigated agriculture in the
Middle Rio Grande Conservancy District (MRGCD) Indian Lands, and municipal and
industrial consumption. In addition to these demands, there are significant consumptive
uses associated with riparian vegetation, reservoir evaporation, and river flow targets
associated with two federally-listed endangered species, the Rio Grande silvery minnow
with Advanced Methodologies for Optimized Agricultural Water Delivery 189
(Hybognathus amarus), and the southwestern willow fly catcher (Empidonax traillii extimus)
(USFWS, 2003). Figure 2 displays the Rio Grande silvery minnow and the southwestern willow fly
catcher.
Fig. 2. Federally-listed endangered species, the silvery minnow (Hybognathus amarus), and
the southwestern willow fly catcher (Empidonax traillii extimus).
This chapter will focus on research for improving water delivery operations in irrigation
systems through the innovative use of water delivery decision support systems (DSS) linked
with SCADA technology and infrastructure modernization. The chapter will present
decision support modeling of irrigation systems in a broad sense and present the model
development and structure of a decision support system developed specifically for the
Middle Rio Grande Conservancy District to provide for more efficient and effective
management of water supplies, and more timely delivery of irrigation water to agricultural
users. The developed DSS will be presented in detail and all three modules that function
together to create water delivery schedules will be explained and examined. The chapter
will address the development of the DSS and will also present the utility of linking the
developed DSS to the SCADA (Supervisory Control And Data Acquisition) system and
automated structure network that the Middle Rio Grande Conservancy District utilizes to
manage water delivery. Linking the DSS and SCADA allows water managers to implement
DSS water delivery schedules at the touch of a button by remotely controlling water
delivery structures. Linking the DSS water distribution recommendations with the MRGCD
SCADA provides a simple and effective medium for managers to implement DSS
recommended water delivery schedules. Additionally, the combination of both programs
allowed for real-time management that converged river diversions and the water required
by the DSS. As the demand for water increases globally throughout the next decades, water
managers will be faced with significant challenges. The use of decision support systems
linked with SCADA and automated structures presents an advanced, efficient, and
innovative approach for meeting water management challenges in the future.
2. Middle Rio Grande Conservancy District

The Middle Rio Grande Conservancy District (MRGCD) may be one of the oldest operating
irrigation systems in North America (Gensler et al. 2009). Prior to Spanish settlement in the
1600s the area was being flood irrigated by the native Pueblo Indians. At the time of
Albuquerque’s founding in 1706 the ditches that now constitute the MRGCD were in
already existence and were operating as independent acequia (tertiary canal) associations
(Gensler et al. 2009). Acequias consisted of farmer groups that maintained individual
irrigation canals. The acequia system was introduced to the MRG Valley by Spanish settlers.
In acequia communities, each farmer was responsible for maintaining a certain length of
canal and would in return receive irrigation water. The use of irrigation water was managed
by an elected mayordomo (Gensler et al. 2009).
Irrigated agriculture in the MRG Valley reached its greatest extent in the 1880s, but
thereafter underwent a significant decline partially caused by an overabundance of water.
By the early 1920s inadequate drainage and periodic flooding resulted in water logging
throughout the MRG Valley. Swamps, seeps, and salinization of agricultural lands were the
result. In 1925, the State of New Mexico passed the Conservancy Act, which allowed for the
creation of the MRGCD, by combining 79 independent acequia associations into a single
entity (Gensler et al. 2009; Shah, 2001). Over the next twenty years the MRGCD provided
benefits of irrigation, drainage, and flood control; however, by the late 1940’s, the MRGCD
was financially unstable and further rehabilitation of structures was required. In 1950, the
MRGCD established a 50-year contract termed the Middle Rio Grande Project with the
USBR to provide financial assistance, system rehabilitation, and system improvement.
System improvements and oversight from the USBR continued until 1975 when the MRGCD
resumed operation and maintenance of the system. The loan from the USBR to the MRGCD
for improvements and operational expenses was repaid in 1999 (Shah, 2001). Currently the
MRGCD operates and maintains nearly 1,200 miles of canals and drains throughout the
valley in addition to nearly 200 miles of levees for flood protection.
Water use in the MRG Valley has not been adjudicated but the MRGCD holds various water
rights and permits for irrigation (Oad and Kullman, 2006). Some users in the MRGCD hold
vested water rights that are surface rights claimed by land owners who irrigated prior to
1907 (SSPA, 2002). Most water users in the MRGCD receive water through state permits
held by the MRGCD. In 1930, the MRGCD filed two permits (#0620 and #1690) with the
Office of the State Engineer that allow for storage of water in El Vado reservoir (180,000 acre
feet capacity), release of the water to meet irrigation demand, and diversion rights from the
Rio Grande to irrigate lands served by the MRGCD. The permits allow the MRGCD to
irrigate 123,000 acres although only about 70,000 acres are currently served (MRGCD, 2007).
This acreage includes roughly 10,000 acres irrigated by pueblo farmers. The MRGCD
charges water users an annual service charge per acre to operate and maintain the irrigation
system. In 2000 the MRGCD charged $28 per acre per year for the right to irrigate land
within the district (Barta, 2003). The MRGCD services irrigators from Cochiti Reservoir to
the Bosque del Apache National Wildlife Refuge. An overview map of the MRGCD is
displayed in Figure 3.
Irrigation structures managed by the MRGCD divert water from the Rio Grande to service
agricultural lands that include both small urban landscapes and large scale production of
alfalfa, corn, vegetable crops such as chile, orchards, and grass pasture. The majority of the
planted acreage, approximately 85%, consists of alfalfa, grass hay, and corn. In the period
from 1991 to 1998, USBR crop production and water utilization data indicate that the
average irrigated acreage in the MRGCD, excluding pueblo lands, was 53,400 acres (21,600
ha) (SSPA, 2002). Analysis from 2003 through 2009 indicates that roughly 50,000 acres
(20,200 ha) are irrigated as non-pueblo or privately owned lands and 10,000 acres (4,000 ha)
are irrigated within the six Indian Pueblos (Cochiti, San Felipe, Santo Domingo, Santa Ana,
Sandia, and Isleta). Agriculture in the MRGCD is a $142 million a year industry (MRGCD,
2007). Water users in the MRGCD include large farmers, community ditch associations, six
Native American pueblos, independent acequia communities and urban landscape
irrigators. The MRGCD supplies water to its four divisions -- Cochiti, Albuquerque, Belen
and Socorro -- through Cochiti Dam and Angostura, Isleta and San Acacia diversion weirs,
respectively (Oad et al. 2009; Oad et al. 2006; Oad and Kinzli, 2006). In addition to
diversions, all divisions except Cochiti receive return flow from upstream divisions.
Fig. 3. Overview Map of MRGCD (MRGCD, 2009)

Return flows are conveyed through interior and riverside drains. From the drains, excess
water is diverted into main canals in the downstream divisions for reuse or eventual return
to the Rio Grande. Drains were originally designed to collect excess irrigation water and
drain agricultural lands, but are currently used as interceptors of return flow and as water
conveyance canals that allow for interdivisional supply.
During the later part of the irrigation season, the MRGCD operates using released storage
water from the high mountain reservoirs of El Vado, Heron, and Abiquiu. Water stored in
these reservoirs consists of snowmelt runoff captured during the early summer months and
water from the San Juan-Chama trans-mountain diversion. These reservoirs are located 98
river miles upstream and water delivery is associated with a significant time lag, which can
approach seven days to reach the southern portion of the district.
The Cochiti Division consists primarily of Native American pueblo land. The pueblo and
non-pueblo lands in the Cochiti Division are managed by four MRGCD ditch-riders and
represent 4800 acres. The Albuquerque Division services many small urban irrigators, but
also provides irrigation water to pueblo irrigators at the northern and southern boundaries
of the division. The Albuquerque Division is managed by one water master and 12 ditch-
riders to oversee the complex irrigation scheme. The Albuquerque Division acreage is 10,300
acres. The Belen Division is the largest in terms of overall service area with a total irrigated
acreage of 31,400 acres. Irrigation in the Belen Division is comprised of large farms, pueblo
irrigators, and urban water users. In Belen the MRGCD employs one water master and 11
ditch-riders. The Socorro Division consists of mostly large parcel irrigators with a total
irrigated acreage of 13,600 acres. Water distribution in Socorro is straightforward when
compared to the Albuquerque and Belen Division, and is managed by one water master and
four ditch-riders. Water availability in Socorro can become problematic since the division
depends on return flows from upstream users.
Water in the MRGCD is delivered in hierarchical fashion; first, it is diverted from the river
into a main canal, then to a secondary canal or lateral, and eventually to an acequia or small
ditch. Figure 4 displays the organization of water delivery in the MRGCD. Conveyance
canals in the MRGCD are primarily earthen canals but concrete lined canals exist in areas
where bank stability and seepage are of special concern. After water is conveyed through
laterals or acequias it is delivered to the farm turnouts, with the aid of check structures
when necessary. Once water passes the farm turnout it is the responsibility of individual
farmers to apply water and it is applied to fields generally using basin or furrow irrigation
techniques.
Fig. 4. Representation of MRGCD Irrigation System

Water in the MRGCD is delivered to users through management and administration
provided at a central office and four divisional offices. The central office in Albuquerque
provides oversight of the four divisional offices and assesses service charges for water use.
Each division office includes administrative, field and equipment maintenance, and water
operations personnel. Water operations in each division are managed by a division manger,
a water master, and ditch-riders in each division. The division managers oversee all aspects
of the division, and water masters coordinate ditch-rider operations. Ditch-riders are
responsible for managing water delivery in a particular service area. Ditch-riders have
anywhere from 250 to 900 irrigators they are responsible for in their service area. Check
structures and head gates are controlled by ditch-riders to deliver irrigation water in their
service area to meet user demand. Water delivery and water use conditions are monitored
by ditch-riders through the physical riding of ditches and through communication with
water users. Ditch-riders generally cover all of the ditches in their service area twice a day
and are in constant contact with water users via cellular phones. Ditch-riders are on call 24
hours a day to deal with emergencies and water disputes, in addition to daily operations.
Water delivery in the MRGCD is not metered at individual farm turnouts. To determine
water delivery the ditch-riders estimate the time required for irrigation. The historic practice
in the MRGCD was to operate main canals and laterals as full as possible throughout the
entire irrigation season. This practice provided for flexible and reliable water delivery with
minimal managerial and financial ramifications; also known as on-demand water delivery.
On-demand or continuous water delivery, however resulted in large diversions from the Rio
Grande. During the past decade, the MRGCD has voluntarily reduced river diversions by
switching to scheduled water delivery. The drawback to this approach is the increased
managerial involvement and the overall cost of water delivery. To aid with the operational
and managerial challenges posed by scheduled water delivery, the MRGCD has developed
and implemented a Decision Support System (DSS) to aid in facilitating scheduled water
delivery. Additionally, the MRGCD has begun to replace aging water delivery
infrastructure with automated control gates that allow for precise control of canal flow rates,
a requirement for scheduled water delivery.
3. Decision support system for the Middle Rio Grande Conservancy District
The New Mexico Interstate Stream Commission and the MRGCD, in cooperation with
Colorado State Univeristy have sponsored a research project from 2003 to 2010 to develop a
decision support system (DSS), to model and assist implementation of scheduled water
delivery for efficiency improvements in the MRGCD’s service area. A DSS is a logical
arrangement of information including engineering models, field data, GIS and graphical
user interfaces, and is used by managers to make informed decisions. In irrigation systems,
a DSS can organize information about water demand in the service area and then schedule
available water supplies to efficiently fulfill the demand.
The conceptual problem addressed by a DSS for an irrigation system, then, is: how best to
route water supply in a main canal to its laterals so that the required river water diversion is
minimized. The desirable solution to this problem should be “demand-driven”, in the sense
that it should be based on a realistic estimation of water demand. The water demand in a
lateral canal service area, or for an irrigated parcel, can be predicted throughout the season
through analysis of information on the irrigated area, crop type and soil characteristics. The
important demand concepts are: When is water supply needed to meet crop demand
(Irrigation Timing), How long is the water supply needed during an irrigation event
(Irrigation Duration), and How often must irrigation events occur for given service area
(Frequency of Irrigation). Decision support systems have found implementation throughout
the American West and are mostly used to regulate river flow. Decision support systems on
the river level are linked to gauging stations and are used to administer water rights at
diversions points. Although decision support systems have proved their worth in river
management, few have been implemented for modeling irrigation canals and laterals and
improving water delivery (NMISC, 2006). The following section will focus on developing a
decision support system capable of modeling flow on a canal and lateral level, with the
overall goal of efficient irrigation water delivery.
The DSS has been formulated to model and manage water delivery in the MRGCD. The DSS
was designed to optimize water scheduling and delivery to meet crop water demand, and,
specifically, to aid in the implementation of scheduled water delivery (Oad et al. 2009;
NMISC, 2006). The DSS consists of three elements, or modules:
 A water demand module that calculates crop consumptive use and soil moisture
storage, aggregated by lateral service area;
 A water supply network module that represents the layout of the conveyance system,
main canal inflow, conveyance system physical properties, and the relative location of
diversions for lateral service area; and,
 A scheduling module that routes water through the supply network to meet irrigation
demand using a mass-balance approach and based on a ranking system that depends
on the existing water deficit in the root-zone.
A Graphical User Interface (GUI) was designed to link the three modules of the DSS
together allowing users to access data and output for the system. A schematic of the three
modules and the way that they relate within the DSS framework is shown in Figure 5.
Figure 6 displays a simplified view of the DSS. GIS information and data obtained from the
MRGCD were used to develop input for both the water demand and the supply network
modules. Some of the input is directly linked through the GUI and some is handled
externally (NMISC, 2006).
Fig. 5. DSS Model Structure Showing Modules

Fig. 6. Simplified View of the DSS

The DSS has two modes of operation: planning mode and operation mode. In planning
mode, the user inputs an anticipated cropping pattern for the season and other related data,
and the model calculates the required main canal diversions as a function of time based on
the calculated demand. In operation mode, the user inputs the available main canal flows,
and the model recommends a water delivery schedule for the lateral canal service areas
within the main canal that optimizes the use of available water (Oad et al. 2009; NMISC,
2006).
3.1 Model structure

The MRGCD DSS was developed using three modules. The three modules that comprise the
MRGCD DSS are the water demand module, the supply network module, and the irrigation
scheduling module. The following subsections describe the three modules in detail.
3.1.1 Water demand module

The water demand module of the MRGCD DSS is implemented through the Integrated
Decision Support Consumptive Use (IDSCU) model code or the ET Toolbox. The IDSCU
model was developed over a period of years by Colorado State University. The IDSCU code
consists of a GUI written in Visual C++ and program calculations implemented with
FORTRAN (Oad et al. 2009; NMISC, 2006). The ET Toolbox was developed specifically for
the Middle Rio Grande and calculates the crop consumptive use using a system of weather
stations throughout the valley. Using either the IDSCU or the ET Toolbox, the following
variables are determined in the water demand module:
 Crop consumptive use (CU);

 Crop irrigation requirement (CIR); and,
 Readily available (soil) moisture (RAM), as a capacity.
CIR and RAM as a capacity, are used in the supply network module. Required data for the
water demand module, the source of these data, and the spatial unit for which the
information is aggregated, are shown in Table 1.
Table 1. Required Data for Water Demand Module (NMISC, 2006)

The Penman-Monteith method suggested by FAO is used to determine the crop
consumptive use for the system when using the IDSCU. The Penman-Monteith equation is
discussed in detail in Chapter 3. When using the IDSCU, weather data required for the
calculation are obtained from the USBR ET Toolbox (USBR, 2003). The USBR ET Toolbox
obtains weather data from 16 weather stations throughout the MRG Valley. In the water
demand module, crop coefficients using growing degree days are combined with the
Penman-Monteith based ET to obtain a consumptive use for each crop type throughout the
growing season. Crop coefficients using growing degree days were obtained from the New
Mexico Climate Center and are based on work done by (King et al. 2000). The water demand
module performs calculations to obtain a spatially-averaged CU at the lateral service area
level using the distribution of crop types within each service area. When the ET Toolbox is
used in the demand module, the consumptive use is already calculated within the ET
Toolbox using grid cells throughout the valley.
The Penman-Monteith-based CU is adjusted to account for effective precipitation, and is
used to obtain a CIR for each lateral service area. Effective precipitation is calculated
according to the Soil Conservation Service Method and is subtracted from the Penman-
Monteith-based CU. The CIR is calculated on a daily basis, corresponding to the water
needed to directly satisfy crop needs for all acres in a lateral service area. The CIR for each
lateral is subsequently passed to the supply network module, where it is divided by an
efficiency factor to obtain a lateral delivery requirement or LDR (Lateral Delivery
Requirement)(NMISC, 2006).
RAM is determined by first calculating the total available water (TAW) per unit area, which
is the available water-holding capacity (AWC) for a given soil type, multiplied by the root-
zone depth. The calculation of TAW is displayed in Equation 1 (Oad et al. 2009; NMISC,
2006).
TAW = AWC * root zone depth (1)

Once the TAW is calculated, a management allowed depletion factor (MAD) for each crop
type is applied to determine the RAM. The values employed for MAD in the Water Demand
Module are crop specific and range from 30% to 60% (NMISC, 2006; FAO, 1998). RAM is
calculated using Equation 2.
RAM = MAD * TAW (2)

Based on acreage, crop type, and soil type within each lateral service area, a value for RAM
is calculated. The RAM calculated in this context represents a storage capacity to be filled
and depleted over several irrigation cycles during the course of the irrigation season
(NMISC, 2006). During each irrigation event, it is expected that an amount of water equal to
the RAM will be stored in soils. Then, as crops utilize water, the RAM will become depleted
(Oad et al. 2009; NMISC, 2006).
3.1.2 Supply network module

The supply network module consists of data relating to water supply, water demand, and
physical information relating to the conveyance network. It represents the layout of the
conveyance system, its physical properties, and the relative location of diversions from the
network to individual lateral service areas (Oad et al. 2009; NMISC, 2006). The layout of the
conveyance system is specified through a user-designed link-node network. Through the
DSS GUI, a user can drag and drop in different types of nodes such as inflows, demands,
and return flow nodes into the program (NMISC, 2006). Developing the link-node network
is straightforward and can be completed by ditch-riders and water master after some short
training. The link-node network represents the connections between canals or laterals and
water demands for each service area. The GUI also contains information on ditch-rider
service area, and includes photographs and locations of all major water control structures.
Figure 7 provides a view of the link-node network.
The supply network module obtains the CIR and the RAM from the water demand module,
and then associates these parameters with a demand node on the link-node network (Oad et
al. 2009; NMISC, 2006). An efficiency factor is applied to the CIR to account for on-farm
application efficiency and to account for conveyance losses within the service area. This
results in a lateral delivery requirement (LDR) that applies to all acreage served by a given
lateral. The efficiency factor can be specified by the user for each lateral service area
(NMISC, 2006).
The link-node network representing the layout of the conveyance system within the supply
network module consists of inflow nodes, stream nodes, and return-flow nodes, connected
by links that represent canals and laterals (NMISC, 2006). Inflow nodes contain information
about inflow volume. The user may either type the inflow values into the inflow node GUI,
or import them from an inflow table. Stream nodes require the user to provide information
about individual lateral service areas, including turnout capacity, the number of days
needed to completely irrigate the service area, the lateral service area efficiency factor,
which days of the week the service area can be irrigated, the minimum flow required in the
canal on a daily basis, and to which “sub-system” a given stream segment belongs (NMISC,
2006). Sub-systems are used to preferentially rank laterals that should be irrigated together.
Fig. 7. DSS Link-Node Network

Return flow nodes are used to capture and route drain flow. The percentage of water
applied to farms that can be captured by drains is set by the user for the entire project. Each
drain has a daily schedule that indicates how much of that return flow is available as inflow
in the successive days (NMISC, 2006). Using the information specified, the supply network
module calculates a demand-based flow rate associated with each diversion to a lateral
service area. The flow rate is calculated as the LDR divided by the irrigation duration.
However, if the calculated flow rate exceeds the lateral capacity specified by the user, then
the flow rate is set at the user-input lateral capacity. In either case, when irrigation occurs, as
determined by the scheduling module, the amount of water removed from the stream link
and delivered to the lateral service area for the daily time step is set equal to the volume of
water that would be delivered at this flow rate over a one-day period (NMISC, 2006). This
volume of water, or time-adjusted portion of LDR, is then reduced by the efficiency factor
and added to the daily RAM for the service area. If the LDR is not fully delivered in the one-
day irrigation, the irrigation may continue into subsequent days depending on the
remaining need within the given lateral, the need of other laterals, and the assigned ranking
system in the scheduler (NMISC, 2006).
3.1.3 Irrigation scheduling module

The irrigation scheduling module uses the information provided by the water demand and
supply network modules to schedule water deliveries to meet crop demand at the lateral
level. The irrigation scheduling module calculates and displays a schedule for the laterals on
a given main canal. This schedule indicates how many laterals can be run at a time, how
long each lateral should run, and how often (NMISC, 2006). Figure 8 displays an irrigation
schedule created using the DSS. The module can also create monthly water delivery
calendars for individual laterals.
The module is set to run on a daily time step. The irrigation scheduling module calculates
the daily irrigation schedule using mass balance equations and a linear programming solver.
The module writes out an input file for the solver, executes the solver, and reads the solver
output. In the present model version, the module displays results in tabular form. Mass
balance calculations used to schedule irrigation timing and duration for lateral canal service
areas are based on the consideration that the farm soil root-zone is a reservoir for water
storage, for which irrigation applications are inflows and CIR is an outflow (NMISC, 2006).
The mass balance approach is displayed in Equation 3:
RAMt 1  It  1  Ot  1  RAMt (3)
Where I is inflow, O is outflow (which includes return flow), and t is time.
Fig. 8. Irrigation Schedule Created Using the DSS

A linear programming approach is used to calculate flows to the service areas by posing the
problem as a minimum cost flow optimization. The model uses the projected number of
days until the soil moisture storage is depleted using a reverse-ranking system to prioritize
the need for irrigation among service areas (NMISC, 2006). Based on observations of the
water delivery operations, it appears that water delivery can be changed from one set of
laterals to another set within one day.
3.2 Model programming

Model programming for scheduling water supplies to lateral service areas is described in
the following section. To obtain the optimum irrigation schedule a linear programming
approach is utilized. Linear programming is a method for optimizing a quantity that is
defined with a mathematical expression or objective function. Constraints on variables
within the objective function are also specified and must be satisfied in determining the
optimum solution. This process favors water delivery to laterals with more immediate water
needs, and minimizes delivery to laterals that have sufficient water (Oad et al. 2009; NMISC,
2006).
For illustration purposes, the scheduling problem is described using a hypothetical network.
Figure 9 shows a simple irrigation network with a main supply canal and a number of
laterals that represent crop water demand. The problem is similar to a transportation
problem, where the service areas are demand nodes and the inflows are supply nodes
(NMISC, 2006). Links are created between nodes where water can be routed. In a
transportation problem, the supply needs to equal the demand; in this case, however, both
under-supply (excess demand) and excess supply are possible. Therefore, to ensure that the
system balances, a “dummy” source node is added that computes the water shortage in the
event the system is water-short. Note that in a water-rich scenario, the dummy node is not
used because it calculates only water shortage (NMISC, 2006)
Fig. 9. Hypothetical Irrigation Network (NMISC, 2006)

The scheduling problem is cast as a minimization problem, for which the goal is to provide
water to the nodes with the greatest need for water (NMISC, 2006). This is achieved through
the use of a ranking system based on water need, the use of water delivery from the dummy
supply, and a set of constraints that capture mass balance conditions through the stream
network. The objective function is displayed by Equation 4.
Minimize Z = MPD-0 XD-0 + MP D-1 XD-1 + MP D-2 XD-2 + MPD-3 XD-3 (4)
In this equation Z is the sum of a modified priority (MP) multiplied by the amount of
supply (X) from the dummy supply to each demand node. The subscripts refer to the nodal
points between which flow occurs, i.e., X D-1 refers to flow from the Dummy supply to Check
1, and MP D-1 refers to the modified priority of demand to be satisfied at Check 1 from the
Dummy supply node. The MP value reflects the need-based ranking system where demand
nodes with lower available soil moisture are favored for irrigation (NMISC, 2006). The
objective function is solved in conjunction with a system of mass balance equations
representing the actual water (and dummy water) delivered to demand nodes, along with
other physically-based constraints.
The variables in the objective function represent the links in the network between the
dummy supply and the demand nodes. The coefficient of each variable represents the flow
“cost” of that link. In other words, delivery of water to a node without a need for water
results in a higher “cost”. As further discussed below, the ranking system has been assigned
such that minimization of this objective function will result in minimization of water
delivery to demand nodes that already have sufficient RAM (NMISC, 2006).
Constraints on the objective function solution reflect the mass balance relationships
throughout the link-node network and the capacity limits on flow (NMISC, 2006). A mass-
balance constraint is created for each node (including the dummy) that establishes the
inflow and outflow to that node. The coefficients of the variables for each constraint (each
row) are represented as a matrix, with a column for every variable in the objective function
and a row for every node (NMISC, 2006). Inflows are represented as negative values and
outflows as positive values. Outflow coefficients are always one, and inflow coefficients
equal the conveyance loss of the connection.
The objective function is subject to the following constraints:
XI-0 <= I
-XI-0 + X0-1 - XD-0 = R0
- L1X0-1 + X1-2 - XD-1 = R1
- L2X1-2+ X2-3 - XD-2 = R2
- L3X2-3 - XD-3 = R3
XD-1 + XD-2 + XD-3 < ∞
Where
X0-1 <= C0-1 X1-2 <= C1-2 X2-3 <= C2-3 All Xi-j >= 0
The variables used are:
 I is the total available inflow
 Xi-j is the flow in a canal reach between points i and j
 Ci-j is the maximum capacity of the canal reach between points i and j
 D refers to a dummy supply node that is used to force the demands and supplies to
balance. The subscript 0 refers to the inflow node, and subscripts 1, 2, 3, … refer to
nodal points, typically located at check structures
 Li is the conveyance loss between in the canal reach
 Ri is the demand (water requirement) at the nodal point indicated by the subscript
(can be zero if not associated with a lateral diversion point)
For example, the third row refers to activity at check 1. There is an inflow from the headgate
(- L1X0-1), and it is given a negative sign since by convention all inflows are negative. The
conveyance loss is represented by the coefficient L1. There is an outflow to check 2 (+ X1-2)
(positive sign, since by convention all outflows are positive). To ensure that the system
balances, there is also an inflow from the dummy source (- XD-1). Because this node
represents a demand, the solution for this row is constrained to be exactly the demand (R1).
If a node represented a source, then the solution for the row would be constrained to fall
between zero and the inflow. That allows the use of less than the total amount of water
available if the demands are less than the supplies, or if at some point in the network the
capacity is insufficient to route the inflow (NMISC, 2006). The first row in the constraint
equations represents this type of node.
The conveyance loss factor specified in the supply network module is a fractional value of
flow per mile. The conveyance loss (L) to be applied in the mass balance equation is
calculated by subtracting the fractional value from one and raising it to the power of the
number of miles of the canal segment between nodes. For example, a 3-mile reach with a
0.015 conveyance loss factor would have a loss of [1 – (1-0.015)3], or a loss of 0.0443 of the in-
stream flow to this reach.
The ranking system used to derive the modified priority (MP) values for the objective
function is a two-step process, involving assignment of a priority (P) based on the irrigation
need at demand nodes, and then a modified priority that effectively reverses the ranking so
that nodes with the least need are the preferred recipients for dummy water (NMISC, 2006).
This results in the actual available water being delivered to the demand nodes with highest
irrigation need. First, a priority (P) is assigned to each of the demand nodes, with smaller
values indicating higher needs for irrigation. The priority is based on the number of days
until the service area runs out of RAM (NMISC, 2006). If the service area is not being
irrigated, 100 is added to the priority, which forces the system to favor areas being irrigated
until the RAM is full again. Subsystems were added to give priority to remaining canals
within a group on the assumption that if one canal service area in a subsystem is being
irrigated then it is desirable that the remaining canal services areas in the same group be
irrigated as well. If a service area is not being irrigated, but is in a subsystem that is being
irrigated, 50, rather than 100, is added to the priority. This makes it a higher priority than
service areas that are not being irrigated but are not in the subsystem. Normally a service
area is irrigated only once during a schedule. However, when excess water is available,
service areas in need of water are added back into the scheduling algorithm with a higher
priority. The ranking system is implemented by modifying the priorities with respect to the
dummy connections, effectively reversing the priorities (NMISC, 2006). Currently the
modified priority (MP) for the “dummy -> node x” connection is 100,000/Px. For example, if
the node has a priority of 105, then the priority assigned to the connection is 100,000/105 or
952.38. This will force dummy water to be delivered first to the lower priority nodes, leaving
real water for the higher priority nodes. The modified priority (MP) values are represented
by the MP variables in the objective function. The linear programming software utilized in
the DSS is a package called GLPK (GNU Linear Programming Kit).
4. Supervisory Control and Data Acquisition (SCADA) technology and

infrastructure modernization
One of the most critical components to realizing schedule water delivery and utilizing real
time recommendations from the DSS to deliver water was the MRGCD SCADA network
and infrastructure modernization. The MRGCD SCADA system can be broken into the
following components:
 Flow Measurement Structures
 Automated Control Structures
 Instrumentation
 Telemetry
 Software
4.1 Flow measurement structures

Water measurement is the single most important component of the MRGCD's SCADA
system, since all operational decisions require sound knowledge of available water supplies
and the demand throughout the system. In 1996, crisis struck the MRGCD in the form of
drought, endangered species flow requirements, and development of municipal water
supplies. At the time, the MRGCD was operating only 15 gauges on 1200 miles of canals.
The following year, MRGCD officially embarked upon its modernization program. The
construction of new flow gauges was the first step in this program. New gauges were
constructed at key points in the canal system, notably at diversion structures and at return
flow points. Along with the increase in numbers of gauging stations, efforts were made to
improve the quality of measurement. Open channel gauging sites with no control structures
gave way to site specific measuring structures. A variety of flow measurement structures
were built in the MRGCD and include sharp crested weirs, broad crested weirs, adjustable
weirs and Parshall flumes. Currently, the MRGCD is operating over 100 gauges. Figure 10
displays a broad crested with telemetry utilized for flow measurement
Fig. 10. Broad Crested Weir Gauging Station with Radio Telemetry
4.2 Automated control structures

With the advent of better data collection, it became apparent that automated control was
necessary. Data from gauges revealed that many operational problems occurred because
canal operators could not be physically present at all times. Automation followed shortly
thereafter with an experimental effort at a wasteway that had been fitted with an automated
Langemann gate for water measurement, and was therefore a practical starting point. The
MRGCD built the electronic controller and created the control software for this first
automated gate, borrowing heavily from Bureau of Reclamation experience in Utah. Success
with the first automated structure led to installation of over 60 additional automated
structures. Most of MRGCD’s recent automation efforts have involved the installation of
Langemann overshot gates (Aqua Systems, 2006). The majority of these can be easily
retrofitted to existing structures, though some involve the construction of new check or
heading structures. The Langemann Gate has the capability to maintain a constant upstream
water level as a check structure or it can provide a constant flow rate to downstream users
(Figure 11). The Langemann gate is equipped with solar panels to power both gate operation
and telemetry units. The gates employ integrated electronic controllers built around IC Tech
radio terminal units (RTU’s) and Aqua Systems 2000 software. Langemann gates in the
MRGCD are used as checks, turnouts, spillways, and diversion structures.
Fig. 11. Langemann Gate
4.3 Instrumentation
Flow measurement and automated control must include some level of instrumentation. In
the 1930’s, a float in a stilling well driving a pen across a revolving strip of paper was
adequate. In fact, at the beginning of modernization efforts, the MRGCD was still using 15
Stevens A-71 stage recorders. Diversions into the canal system were only known after the
strip charts were collected and reduced at the end of the irrigation season. Modernization
meant a device was needed to generate an electrical or electronic output that could be
digitally stored or transmitted. Initially, shaft encoders were used for this purpose,
providing input for electronic data loggers. Experimentation with submersible pressure
sensors soon followed, and these have been adopted, although a number of shaft encoders
are still in use. Sonar sensors have been used satisfactorily at a number of sites and recently
compressed air bubblers have been utilized. Different situations call for specific sensor types
and sensors are selected for applications where they are most appropriate.
4.4 Telemetry
Data from electronic data-loggers was initially downloaded manually and proved to be
only a minimal improvement over strip chart recording, though processing was much
faster. To address data downloading concerns telemetry was adopted to bring the
recorded data back to MRGCD headquarters at regular intervals (Figure 12). MRGCD's
initial exposure to telemetry was through the addition of GOES satellite transmitters to
existing electronic data loggers. This method worked, but presented limitations. Data
could only be transmitted periodically, and at regularly scheduled intervals. Of greater
consequence was that the GOES system, at least as used by MRGCD, was a one-way link.
Data could be received from gauging stations, but not sent back to them. As experiments
with automation progressed, it was clear that reliable 2-way communication would be a
necessity. To address the rising cost of phone service, experiments with FM radio
telemetry were conducted. These began as a way to bring multiple stream gage sites to a
central data logger, which would then be relayed via GOES to MRGCD. First attempts
with FM radio were not encouraging; however a successful system was eventually
developed. As this use of FM radio telemetry (licensed 450 mHz) expanded, and
knowledge of radio telemetry grew, it was soon realized that data could be directly
transmitted to MRGCD headquarters without using the GOES system. Full conversion to
FM radio based telemetry also resulted in very low recurring costs.
The shift to FM radio produced what is one of the more unique features of the MRGCD
SCADA system. The data link proved so reliable, that there was no longer a need to store
data on site, and the use of data loggers was mostly discontinued, the exception being
weather stations. In effect, a single desktop computer at the MRGCD headquarters has
become the data-logger for the entire stream gauge and gate system, being connected to
sensors in the field through the FM radio link. Three repeater sites are used to relay data up
and down the length of the valley, with transmission being up to 75 miles. Also, this has the
benefit of being a 2-way link, so various setup and control parameters can be transmitted to
devices along the canals. This 2-way link allowed for the system to be linked to the DSS in
order to improve water delivery operations. The MRGCD telemetry network consists
exclusively of Control Design RTU’s. Several different types of these units are used,
depending on the application. The simplest units contain only a modem and radio, and
transmit collected and processed weather station data from Campbell Scientific CR10X
dataloggers.
The majority of the RTU’s contain a modem, radio, and an input/output (I/O) board
packaged into a single unit. Sensors can be connected directly to these and read remotely
over the radio link. A variety of analog (4-20ma, 0-20ma, 0-5v) and digital (SDI-12, RS-485)
output devices can be accommodated this way. Another type includes a programmable (RP-
52 BASIC) controller in the modem/radio/(I/O) unit. This style is used for all automatic
control sites and places where unusual processing of sensor outputs such as averaging
values, combining values, or timed functions, are required. At the present time, the MRGCD
telemetry network gathers data from over 100 stream flow gages and 18 ag-met stations, and
controls 60 automated gates (Figure 12).
4.5 Software
Measurement, automation, and telemetry components were developed simultaneously, but
largely independent of one another. While each component functioned as expected,
components did not exist as a harmonious whole, or what could truly be called a SCADA
system. The missing component was software to tie all the processes together. There are a
variety of commercially available software packages for such use and MRGCD
experimented with several. Ultimately, the MRGCD chose to purchase the commercial
software package Vsystem and to employ the vendor Vista Controls to develop new
features specific to the control of a canal network. Installation and setup was done by the
MRGCD.
This system, known as the Supervisory Hydro-data Acquisition and Handling System
(SHAHS), gathers data from RTU's on a regular basis. With the capability to define both
timed and event driven poll routines, and specify a virtually unlimited number of RTU’s
and MODBUS registers to collect, virtually any piece of information can be collected at any
desired time. The Vsystem software can process data through a myriad of mathematical
functions, and combine outputs from multiple stations. Vsystem also incorporates the ability
to permanently store data in its own internal database, MS Sequel databases, or export data
in other formats. Data can be displayed in a user-created graphical user interface (GUI)
which MRGCD water operations personnel use to monitor water movement. The screens
can also execute scripts to generate data, control parameters, control gate set points, and
monitor alarm conditions for automated control structures. Finally, the GUI’s can be used to
control automated structures by transmitting new parameters and setpoints.
Fig. 12. MRGCD Telemetry Network
5. Linking the decision support system to SCADA network

One of the key components to improving water delivery efficiency was linking the DSS
water delivery recommendations to the MRGCD SCADA network. This was done in 2009 so
that water operations personnel could closely monitor the actual canal diversions and
compare those diversions to the DSS recommendations on a real-time basis. The overall goal
was to match the diversions from the Rio Grande to the real-time crop water requirement
calculated using the DSS.
Incorporating the DSS with the MRGCD SCADA system involved three distinct steps. First,
the DSS was installed on the MRGCD main water operations computer. Second, the DSS
output was converted to a format that the Vista software could recognize. The Vista
software uses the SHEF.A. file format for data coming into the SCADA system, and the
entire MRGCD canal network is set to function on this format. The data for each individual
gate or measuring site are characterized by a distinct data stream in the Vista Software, the
data and are linked to the appropriate node in the SCADA GUI. The data stream for each
node in the GUI is user specified; and therefore, nodes were created that display DSS
recommended flow rates. In order to link the DSS to this SCADA software it was necessary
to create DSS output files in SHEF.A. format. This was accomplished through cooperative
work between Colorado State University, the MRGCD, and Vista Systems. A subroutine in
the MRGCD SCADA Vsystem software converts the output from the DSS into the SHEF.A.
format. This subroutine allows for the creation of separate data streams from the DSS
schedule for each lateral canal service area. These data streams are the same data streams
that are used throughout the MRGCD SCADA network.
The third step in linking the DSS was to create a node for the DSS recommended flowrate
for each lateral service area in the SCADA GUI that also contains actual flowrate data. This
consisted of creating nodes in the SCADA GUI that display the actual flow passing into a
lateral on the left side of the node and the recommended flowrate from the DSS in
parentheses. The DSS recommended flow was linked to the correct data stream from the
DSS output in SHEF.A format. For the 2009 irrigation season actual flow and the DSS
recommended flow were displayed side by side for each lateral on the Peralta Main Canal in
the SCADA GUI. Figure 13 displays the revised SCADA GUI with the DSS
recommendations.
Fig. 13. MRGCD SCADA Screen with Actual Deliveries and DSS Recommendations
Whenever the recommendations for DSS flowrates are updated the DSS is run and the
scheduler output saves as a CSV file in Excel. This file is then converted using the Vsystems
subroutine and available as a SHEF.A. data stream for each individual lateral. Linking the
DSS water distribution recommendations with the MRGCD SCADA provides a simple and
effective medium for managers to implement and monitor DSS water delivery schedules.
Additionally, the combination of both programs allows for real-time management and the
convergence of river diversions and the water required based on crop demand.
6. Results
The DSS linked with SCADA has multiple benefits for the MRGCD. These benefits include
providing a method for predicting anticipated water use, sustaining efficient agriculture in
the valley, minimizing the difficulty of meeting flow requirements for environmental
concerns, and improved reservoir operations. The DSS in concert with SCADA also
provides the MRGCD with a powerful tool to manage water during periods of drought.
One of the main benefits that the DSS has for the MRGCD is that the DSS provides a tool to
determine anticipated water use. The need for this arises out of the fact that water delivery
to farms in the MRGCD is not metered and scheduling is inconsistent. This makes it
difficult, if not impossible, to place the appropriate amount of water in a canal at the proper
time. The MRGCD dealt with this in the past by typically operating all canals at maximum
capacity throughout the irrigation season. This provided a high level of convenience for
water users, and made the lack of scheduling unimportant. But this practice has had
significant negative consequences. Not least of these consequences was the public
perception of irrigated agriculture. The MRGCD practice of operating canals at maximum
capacity resulted in diversion rates from the Rio Grande that were large. Through the
creation of a thorough and systematic database in the DSS of cropping patterns, irrigated
acreage, automated processing of climate data into ET values, and incorporation of a flow-
routing component incorporating the physical characteristics of the MRGCD canals, it
became possible to predict how much water users will need. The DSS, with the myriad of
calculations it performs, provides the MRGCD water operations manager a method to
determine water requirements in advance and manage water using SCADA. Problems of
timing and crop stress are addressed and the MRGCD can operate at reduced diversion
levels, while serving its water users.
Use of the DSS linked with SCADA also has the benefit that it can aid the MRGCD in
sustaining irrigated agriculture in the Middle Rio Grande Valley. The agricultural tradition
in the Middle Rio Grande Valley dates back over 500 years and one of the main goals of
MRGCD is to sustain the agricultural culture, lifestyle and heritage of irrigation. The
problem facing the MRGCD is how to sustain irrigated agriculture amidst drought and
increased demands for water from the urban and environmental sectors. These demands for
water from other sectors will increase as the population in the Middle Rio Grande Valley
grows and expands. Additionally, the MRGCD will be faced with dealing with periodic
drought and climate change with the possibility of reduced snow melt runoff in the future.
The concept of scheduled water delivery, implemented through the use of the DSS linked
with SCADA provides the MRGCD with the ability to sustain irrigated agriculture with
reduced river diversions. Scheduled water delivery that is based on crop demand calculated
using the DSS and delivered through a highly efficient modernized system will allow the
MRGCD to continue supplying farmers with adequate water for irrigation, even though the
available water supply may be reduced due to natural or societal constraints.
Scheduled water delivery implemented through the use of the DSS and SCADA will also
benefit the MRGCD by easing competition with environmental water needs. Due to the
Endangered Species Act, water operations and river maintenance procedures have been
developed in the Middle Rio Grande Valley to ensure the survival and recovery of the Rio
Grande silvery minnow. These procedures include timing of flow requirements to initiate
spawning and continuous flow throughout the year to provide suitable habitat.
Additionally, the entire Rio Grande in the MRGCD has been designated as critical habitat
for the RGSM. The use of the DSS, linked with SCADA provides the MRGCD with the
ability to reduce river diversions at certain times during the irrigation season. Reduced river
diversions from the MRGCD main canals may at times leave more water in the Rio Grande
for the benefit of the RGSM with credit toward the MRGCD for providing the flow
requirements for the recovery of the species. The DSS may also be useful in providing
deliveries of water specifically for the RGSM. Since the listing of the RGSM, the MRGCD
canal system has been used to deliver a specific volume of water to points along the Rio
Grande to meet flow requirements for the species. At certain times of the year, this is the
only efficient way to maintain RGSM flow targets. While not presently incorporated in the
DSS, it would be straightforward to specify delivery volumes at specific points in the
MRGCD system. These delivery volumes for the RGSM would be scheduled and routed in a
similar fashion to agricultural deliveries. Depending on the outcome of the current process
of developing RGSM management strategies, this could someday become a very important
component and benefit of the DSS.
One of the significant benefits of using the DSS with SCADA in the MRGCD is improved
reservoir operations. The storage reservoirs in the MRGCD are located in the high
mountains and it takes up to seven days for released water to reach irrigators. Prior to
scheduled water delivery utilizing the DSS, the MRGCD water manager had to guess at
what the demand for a main canal would be in advance and then release stored water to
meet the assumed demand. Through the use of the DSS and SCADA the water operations
manager can utilize historical climate data to predict what the agricultural demand will be
in the future. This allows for an accurate calculation of the required release from reservoir
storage and minimizes superfluous releases. These reduced reservoir releases have
significant benefits for the MRGCD. Since less water is released from storage reservoirs
during the irrigation season it allows for increased storage throughout the season. This
allows the MRGCD to stretch the limited storage further and minimize the impacts of
drought. Decreases in reservoir releases also have the added benefit of providing more
carryover storage for the following irrigation season, providing greater certainty for water
users in subsequent years. Larger carryover storage also translates to less empty space to fill
during spring runoff. This leads to three benefits for the MRGCD. The first is that reservoirs
can still be filled, even in a year when runoff is below average. The second benefit is that in
above average runoff years the reservoirs will fill quickly and much of the runoff will go
downstream, mimicking the hydrograph before the construction of upstream storage
reservoirs. This is a subtle but significant environmental benefit to the Middle Rio Grande
and RGSM as peaks in the hydrograph induce spawning, provide river channel equilibrium
and affect various other ecosystem processes. The third benefit is that increased downstream
movement of water during the spring will aid the state of New Mexico in meeting Rio
Grande compact obligations to Texas. Overall, scheduled water delivery in the MRGCD
provides significant benefits to reservoir operations and will allow the MRGCD to reduce
reservoir releases, provide more reliable deliveries, increase certainty of full deliveries, and
sustain irrigated agriculture.
In times of surplus water the MRGCD may not utilize the DSS for scheduled water delivery
because of the associated increase in management intensity and higher operational costs. In
drought years the available water supply will be lower and result in tighter and more
stringent management requirements to equitably distribute water to users. It is during times

of drought and shortage that the DSS linked with SCADA will provide the necessary
management to distribute water based on crop demand. In the future the DSS will be a
powerful tool that the MRGCD can utilize to equitably distribute water and sustain irrigated
agriculture through times of low water supply resulting from drought.
7. Conclusions
Water delivery through the use of a DSS linked with SCADA has applications and benefits
that can be realized throughout the arid regions of the world. The main benefit of scheduled
water delivery utilizing a DSS and SCADA is that diversions can more closely match crop
water requirements eliminating spills and drainage problems. In the American West urban
growth and environmental concerns have forced irrigated agriculture to reduce diversions.
In many irrigated systems in the United States that traditionally used surface application,
agriculture has opted to improve water use efficiency by changing water application
methods to sprinkler and drip irrigation systems. Irrigated agriculture throughout most of
the world still relies on surface application and cannot afford to upgrade to systems such as
sprinkler or drip irrigation. Therefore, water delivery utilizing a DSS with SCADA has the
potential to reduce diversions and sustain agriculture.
The DSS, SCADA, and scheduled water delivery also have the potential of meeting future
agricultural demands in developing regions throughout the world. As the world population
continues to grow there will be an increased demand for food production and in many cases
water resources available for agriculture are already fully utilized. Utilizing a DSS linked to
SCADA would allow water users in developing countries with surface application systems
to conserve water from their current practices and apply the saved water to increased food
production.
The DSS could also be used to refine water delivery scheduling. Many arid regions have
been dealing with water shortages for decades and have already implemented scheduled
water delivery. In most cases, water delivery schedules are based on a set interval of time
and do not coincide with crop demand. In areas where this type of scheduling is practiced
the DSS could be used to refine scheduling protocols to include crop demand. Scheduling
water deliveries based on crop demands would provide additional saving in areas where
scheduled water delivery is already implemented. The developed DSS could be utilized in
any irrigation system worldwide that practices surface irrigation techniques. Through
scheduling based on crop demand, overall diversions could be significantly reduced.
Reduced diversions could help irrigators deal with drought, and climate change by allowing
for increased utilization of stored water. Additionally, reduced diversions could be utilized
to grow supplementary crops to supply the needs of a growing population in the future.
8. Acknowledgements
The authors would like to thank Subhas Shah, the MRGCD Board of Directors and staff, the
New Mexico Interstate Stream Commission, the New Mexico Office of the State Engineer,
the Middle Rio Grande Endangered Species Act Collaborative Program, the United States
Army Corp of Engineers, and the United States Bureau of Reclamation for the assistance and
the financial support to undertake this project. Also, the exceptional support of Jim Conley
at Control Design, Gerald Robinson and Lee Allen at Aqua Systems 2000, and Cathy
Laughlin and Peter Clout of Vista Control Systems is graciously recognized. The authors
would also like to acknowledge the support of the National Science Foundation
9. References
Barta, R. 2003. Improving Irrigation System Performance in the Middle Rio Grande
Conservancy District. Master’s Thesis. Department of Civil Engineering, Colorado
State University. Fort Collins, Colorado
Gensler, D. Oad, R. and Kinzli, K-D. 2009. Irrigation System Modernization: A Case Study
of the Middle Rio Grande Valley. ASCE Journal of Irrigation and Drainage. 135(2):
169-176.
King, J.P., A.S. Bawazir, and T.W. Sammis. 2000. Evapotranspiration Crop Coefficients as a
Function of Heat Units for Some Agricultural Crops in New Mexico. New Mexico
Water Resources Research Institute. Technical Completion Report. Project No.
01-4-23955.
MRGCD. 2009. http://www.mrgcd.com
MRGCD. 2007. Middle Rio Grande Conservancy District – Keeping the Valley
Green. Flyer of the Middle Rio Grande Conservancy District. Middle Rio Grande
Conservancy District. Albuquerque, New Mexico.
New Mexico Interstate Stream Commission (NMISC); 2006 (authors Ramchand Oad, Luis
Garcia, Dave Patterson and Kristoph-Dietrich Kinzli of Colorado State University,
and Deborah Hathaway, Dagmar Llewellyn and Rick Young of S. S. Papadopulos
& Assoc.); Water Management Decision-Support System for Irrigation in the Middle Rio
Grande Conservancy District; March, 2006.
Oad, R. Garcia, L. Kinzli, K-D, Patterson, D and N. Shafike. 2009. Decision Support Systems
for Efficient Irrigation in the Middle Rio Grande Valley. ASCE Journal of Irrigation
and Drainage. 135(2): 177-185.
Oad, Ramchand and K. Kinzli. 2006. SCADA Employed in Middle Rio Grande Valley to
Help Deliver Water Efficiently. Colorado Water – Neswletter of the Water Center
of Colorado State University. April 2006,
Oad, R. Garcia, L. Kinzli, K-D. Patterson, D. 2006. Decision Support Systems for Efficient
Irrigated Agriculture. (2006). Sustainable Irrigation Management, Technologies and
Policies. WIT Transactions on Ecology and Environment. Wessex Institute of
Technology, Southampton UK.
Oad, Ramchand and R. Kullman (2006). Managing Irrigated Agriculture for Better River
Ecosystems—A Case study of the Middle Rio Grande. Journal of Irrigation and
Drainage Engineering, Vol. 132, No. 6: 579-586. American Society of Civil
Engineers, Reston, VA.
Shah, S.K. 2001. The Middle Rio Grande Conservancy District. Water, Watersheds, and Land
use in New Mexico: Impacts of Population Growth on Natural Resources, New
Mexico Bureau of Mines and Mineral Resources. Socorro, New Mexico. 123-125.
S.S. Papadopulos and Associates, Inc. (SSPA). 2000. Middle Rio Grande Water Supply
Study. Boulder, Colorado.
S.S. Papadopulos and Associates, Inc. (SSPA). 2002. Evaluation of the Middle Rio Grande
Conservancy District Irrigation System and Measurement Program. Boulder,
Colorado.
United States Fish and Wildlife Service (USFWS). 2003. Federal Register 50 CFR Part 17.
Endangered and Threatened Wildlife and Plants; Designation of Critical Habitat
for the Rio Grande Silvery Minnow; Final Rule. Vol. 68, No. 33. U.S. Department of
Interior.
11
Estimating the Impact on Water Quality

under Alternate Land Use Scenarios:
A Watershed Level BASINS-SWAT
Modeling in West Georgia, United States
Gandhi Bhattarai1, Diane Hite2 and Upton Hatch3
1HealthCare
Innovation and Information, Ingenix
2Agricultural
Economics and Rural Sociology, Auburn University
3Agricultural and Resource Economics, North Carolina State University
USA
1. Introduction
Non-point source pollution (NPP) is caused by the movement of water over and through the
ground, generally after each rainfall. The runoff picks up and carries away natural and man-
made pollutants, eventually depositing them in water bodies like lakes, rivers and coastal
waters. Thus the pollutants left on the surface from various sources accumulate in receiving
water bodies. For example, crop cultivation requires more use of chemicals and nutrients
than natural vegetative cover like forest and grasslands. Tillage operations affect the soil
structure and the level of chemicals in the soil. Such activities make the nutrient rich topsoil
fragile and cause it to lose more chemicals and soil particles during rainfall. Lands in
residential and development uses, such as lawns and gardens are managed more
intensively, which encourages the generation of even more pollutants. On the other hand,
urban areas have higher percentages of impervious to porous surfaces that result in low
percolation and higher runoff. During precipitation, the runoff carries more nutrients and
sediments from agricultural and residential lands resulting in higher chemical level and
turbidity in the water. Thus increasing urbanization coupled with increasing use of
nutrients and chemicals in agricultural lands create significant challenges to maintain water
quality.
Recent water quality studies have focused on developing and successfully applying various
biophysical simulation methods to model levels of NPP and to identify critical locations
from which these pollutants originate (Bhuyan et al., 2001; Mankin et al. 1999; Marzen et al.,
2000). These models collect and use various geospatial data, facilitating the spatial analysis
of sources and effects of point and non-point pollutants with reference to their origin and
geographical locations. The findings of such models help the environmental policy planners
to understand both the short-term and long-term effects of changes from alternative land
management scenarios and simulate how the pollution can be reduced effectively through
the institutionalization of best management practices. This study uses the BASINS-SWAT
modeling framework, one of the latest biophysical modeling tools to estimate the effects of
land use change in the quality of water. Simulations are done to estimate the level of
nitrogen, phosphorus and sediment loads using two time period land use maps. This study
demonstrates the use of geospatial technologies to gather and organize reliable and current
data for inputs into the BASINS-SWAT model runs.
2. The modeling approach

Better Assessment Science Integrating Point and Non-point Sources (BASINS) is a
multipurpose environmental analysis system developed by U.S. Environmental Protection
Agency’s Office of Water. It is used in performing watershed-and water-quality-based
studies and can support analysis at a variety of scales using different tools. BASINS can
support implementation of TMDLs by state agencies using watershed-based point and
nonpoint source analysis for a variety of pollutants including the alternative land
management practices. It comprises a suite of interrelated components for performing the
various aspects of environmental analysis including data extraction, assessment, watershed
delineation, classifying digital elevation models (DEM), land use, soils and water quality
observations, and watershed characterization reports (United States Environmental
Projection Agency [USEPA], 2001).
Soil and Water Assessment Tool (SWAT) is a river basin or watershed scale model
developed by the USDA Agricultural Research Service’s Grassland, Soil and Water Research
Laboratory which has become an integral component of BASINS. It was designed to assist
resource managers in the long-term assessment of sediment and chemical yields in large
watersheds and river basins. The model can predict the average impact of land use and
management practices on water, sediment, and agricultural chemical yields in large,
complex watersheds with varying soils, land uses, and management conditions over long
periods of time (DiLuzio et al., 2002; Neitsch et al., 2002). In comparative studies using
hydrologic and non-point source pollution models, SWAT has been shown to be among the
most promising for simulating long-run NPP in agricultural watersheds (Borah and Bera,
2003).
BASINS-SWAT uses an ArcGIS Geographic Information System interface to derive the
model input parameters and simulation. Hydrological modeling relies on elevation data in
which watershed and subwatersheds are delineated based on the terrain and flow direction
of raindrops. The watershed drains at the lowest point of the catchment area and contains
several sequential subwatersheds with directional flow (raindrop flow) to the main channel
based on the topography of land and user supplied intermediate outlet points.
Subwatersheds are grouped based on climate, hydrologic response units (HRUs), ponds,
ground water, and main channels (Borah & Bera, 2003). Each subwatershed is virtually
divided into several hydrological response units (HRUs) which are uniquely lumped areas
within the subwatershed based on weighted land cover, soil type, and management
combinations at a certain threshold level, generally 10% in practice (Saleh et al., 2000).
SWAT model simulation requires weather parameter inputs (precipitation, wind,
temperature) and management parameter inputs (irrigation, tillage, chemical and fertilizer
application). These input variables are converted to standard SWAT input files within the
model. The model then simulates the runoff levels of nutrients, chemicals and sediment
loadings, as well as crop yields, under a particular combination of land management
scenarios. SWAT output can be traced across the watersheds both for short and long period
of times.
Estimating the Impact on Water Quality under Alternate Land Use Scenarios:
A Watershed Level BASINS-SWAT Modeling in West Georgia, United States 215
3. Study area
This study is based on the Mulberry Creek Watershed a 10-digit coded Hydrological Unit
(HUC #0313000212) in Harris County, Georgia, nested within the Middle Chattahoochee
watershed (HUC# 03130002). The area is situated to the north of Columbus, GA and covers
approximately 50% of the county surface area. The map of study area is given in Figure 1.
Location of Study Area in West Georgia
Paulding
Cobb
k
Douglas Fulton
Fort
ee
k
ee
Cr
Cree
Cr
G in
ell
wd
k
Pa
Do
Bell Creek
Carroll
lm
ett
oC
Ba
N rne
r ee
sC k
reek Cree
k
ree
k yC erry
lberr Mulb
Mu
Wh
Coweta
ets
ek
ton
W E Randolph Heard erry
Cre
k
e
Mulb chie Cree
Bra
Ossahat
nc h
Meriwether
S
Troup (HUC #0331000212)
HUC # 03310002
5 0 5 Miles
Chambers Harris
Talbot
Lee
Legend
Muscogee
Stream Network (NHD)
Reach File V1
Study area sub-watershed
Main Watershed
County Boundary
Fig. 1. Map of Study Area

Harris County has experienced a tremendous market influence due to the expansion of the
Columbus Metropolitan Area in the last two decades. Limited by the Chattahoochee River
on the west and by a military base on the south, the suburban area is expanding mostly to
the north and northeast. Harris County experienced a 33.2% increase in total population
between 1990 and 2000 census years (United States Census Bureau, 1990, 2000). In the same
time, the population in Troup and Muscogee counties increased only by 5.8% and 3.9%
along the north and south boarder of the county respectively. On the other hand, Talbot
County, which is mostly rural and lies to the east of Harris County did not see any increase
in its population. As the demographic structure and other socioeconomic conditions in the
area changed between 1990 and 2000 census years, land use distribution has also changed.
This provided a very good experiment site to estimate the water quality impact of land use
change. Figure 2 compares the land use scenarios within the Mulberry Creek watershed
using the National Land Cover Datasets 1992 and 2001 (NLCD-1992 and NLCD-2001).
Land Use Change in Mulberry Creek Subwatershed

1992 - 2001
(a) Reclassified NLCD-1992 (a) Reclassified NLCD-2001
Legend
Developed Land
Agricultural Land
Forest Land
Other Land
Subwatershed Boundary
Fig. 2. Land use distribution in Mulberry Creek sub-watershed

Delineated watershed covered 58,323 hectares in size, roughly half of the total area in Harris
County, GA. There has been significant change in the distribution of land use in the
Mulberry Creek sub-watershed between 1992 and 2001. For comparison, land use grids
were broadly reclassified into developed, agricultural, forest and other lands for both years.
In NLCD-2001, forest land covered 71.2% of total area, down from 87.8% in NLCD-1992.
Share of agricultural land increased to 17.2% in NLCD-2001, about two times bigger than in
NLCD-1992.Developed land share increased by nine-fold in NLCD-2001. Broadly
reclassified land use distribution is given in Figure 2.
The change in the land use pattern in the area is expected to bring changes in water quality
through different levels of point and non-point source pollution. According to the EPA’s list
of impaired water, 13 total maximum daily loads (TMDLs) had been approved in the
Middle Chattahoochee Watershed between 1996 and 2004 (EPA, Surf Your Watershed).
4. Methodology
4.1 Data collection and processing
The study uses secondary data obtained from various sources. The sources and processing
of the individual data components is described below.
Core BASINS data: Once the BASINS project was opened, basic datasets were extracted from
various sources using the data download extension within the BASINS program. These files
included various boundary shape files at watershed, county and state levels, soil and land
use grids, locations of water quality stations, bacteria stations, USGS gage stations, permit
compliance systems and National Hydrography Data (NHD). The BASINS view was
projected to Albers Equal-Area Conic projection system. All subsequent data were projected
in the same system using grid projector tool in BASINS.
NEM data: The 1:24,000 scale 30x30m resolution national elevation model (NEM) data for the
entire area was downloaded from Seamless Data Distribution System of USGS Web Server.
The dataset was converted to grid files using ArcView Spatial Analysis tool.
Land use data: Two sets of spatial land use data were obtained from Seamless Data
Distribution System of USGS Web Server. The first land use map was the National Land
Cover Database 1992 (NLCD-1992), which was prepared based on satellite images taken
circa 1990. The second land use map was the National Land Cover Database 2001 (NLCD-
2001). This map was prepared from satellite images taken circa 2000. Both of the land cover
maps had 30x30m resolution at the scale of 1:24,000 (USGS MRLC metadata information).
Land use grids were re-projected to match with the NEM grids projection.
Climate Data: SWAT model uses climate information including precipitation, temperature,
wind speed, solar radiation and relative humidity data. However, when no such data are
available, the model has capability to simulate data based on historical weather observations
from the nearest of more than 800 climate stations (SWAT Users’ Manual). Observed daily
precipitation and minimum/maximum temperature data were obtained from the National
Climate Data Center (NCDC) database for ten nearby climate stations between January 1945
and December 2004. Raw precipitation and temperature data were converted into SWAT
readable individual database files for each station.
4.2 Watershed modeling process

Once all data were collected and the geospatial data layers were re-projected and matched
to a single view, the BASINS process started with preprocessing of DEM to remove any
sinks and holes in the DEM coverage. A masking grid covering the anticipated area of
watershed was used as focus area and a Reach File used as a burn-in option to enhance the
watershed delineation process. After the preprocessing of DEM, digital stream networks
were created with a 100-hectare headwater threshold area, which defines the minimum area
required to begin a stream flowing out of the area in any part of the watershed (Figure 3). As
a standard, the typical value is 5000 cells (450 hectares) when using National Elevation
Dataset. Although any values can be used, threshold under 1000 cells or 90 hectares
becomes questionable in the areas of flat terrain (Maidment, 2002).
The lowest outlet point at the stream network was selected as the drainage outlet to define
the watershed boundary. Spurious nodes were removed and very long river segments were
divided into outlet nodes during the calculation of sub-basin parameters to ensure a
uniform size distribution of sub-basins. Sub-basin parameters were calculated within the
model that makes the use of DEM and stream network (Figure 4-5).
Fig. 3. Overlaying elevation grids, watershed boundary and NHD networks
Fig. 4. Digitization of stream networks and nodes

Fig. 5. Creation of sub-basins

Land use grids were loaded in the program and reclassified to standard SWAT codes using the
user defined land use look up code created outside of the program. A land use look up code
table was prepared to match the land use class to the standard SWAT classification (Table 1).
The state soil layer was also loaded and reclassified using the standard STATSGO soils look
up codes. Once the reclassification was complete, soils and land use layers were spatially
overlaid to facilitate the creation of hydrological response units (HRUs, Figure 6). An HRU
is a result of interaction of land use and soils thus creating a unique land area within each
sub-basin. Later, several components of the SWAT results, for example, crop yields, biomass
would be based at the HRU level. Multiple HRUs were created by setting 5% threshold on
land use and 10% threshold on soils.
Upon the successful creation of HRUs the SWAT module was loaded by the system. It
started by adding the SWAT database into the model that included the weather and
management information. The weather station databases, which included daily precipitation
record and daily minimum-maximum temperature records, were created for 10 nearby
weather stations outside of the program extracted from the NCDC climate database. During
the model simulation, the program assigns the recorded values from the nearest weather
station. When data is unavailable for any period of time, the program uses the simulated
values based on the information stored for more than 1041 weather stations nationwide (De
Luzio et. al., 2000). The other inputs fed into the program were solar radiation, wind speed
and relative humidity data. Since none of the weather stations contained these information,
it was left to the program to simulate those values from within the internal database. Once
the weather station databases were loaded into the system, all necessary SWAT inputs were
written using the elevation, land use, soil and weather station databases. This process built
the database files containing the information needed to generate default inputs for SWAT.
Land management practices, including average nutrients and chemical uses for each
NLCD 1992 NLCD 2001

GRID Code NLCD Description GRID Code NLCD Description
11 WATR Open Water 11 WATR Open Water
12 WATR Perennial Snow/Ice 12 WATR Perennial Snow/Ice
21 URML Low Intensity Residential 21 URML Developed, Open Space
22 URHD High Intensity Residential 22 URML Developed, Low Intensity
23 UIDU Commercial/Transportation 23 URMD Developed, Medium Intensity
31 UIDU Rocks/Sand/Clay 24 URHD Developed, High Intensity
32 UIDU Quarries/Strip mines/Pits 31 UIDU Barren Land (Rock/Sand/Clay)
33 FRST Clear Cut/Sparse Vegetation 32 UIDU Quarries/Strip Mines/Pits
41 FRSD Deciduous Forest 41 FRSD Deciduous Forest
42 FRSE Evergreen Forest 42 FRSE Evergreen Forest
43 FRST Mixed Forest 43 FRST Mixed Forest
51 RNGB Shrublands/Brushes 52 RNGB Shrub/Scrub
81 HAY Pasture And Hay 71 RNGE Grasslands/Herbaceous
82 AGRR Row Crops 81 HAY Pasture And Hay
83 AGRC Small Grains 82 AGRR Cultivated Crops
84 AGRL Fallow 90 WETF Forested Wetlands
85 URML Recreational Grasses 95 WETN Herbaceous Wetlands
91 WETF Woody Wetlands
92 WETN Herbaceous Wetlands
Table 1. Reclassification of land use grids to standard SWAT codes
Fig. 6. Land use and soil reclassification and spatial overlay

type of land use were set to common practices reported in various published extension
service papers. Sedimentation and nutrient deposition were evaluated for different land use
scenarios using NLCD-1992 and NLCD-2001. Since the data on wind, solar radiation and
humidity were not available, built-in data from the SWAT model were used. Curve number
and Preistley-Taylor methods were used for modeling the rainfall and evapotranspiration
respectively.
5. Results
The watershed was divided into 331 sub-basins. The mean elevation of all sub-basins was
216 meters above sea level ranging from 128 to 368 meters. The mean and median sizes of
sub-basins were 176.2 hectares and 174.4 hectares respectively. This sub-basins area ranged
from 14.3 hectares to 438.9 hectares.
HRUs are determined by the combination of land use and topographic parameters.
Therefore, the number of HRUs was different when two land use maps were used. A total of
2089 HRUs were identified by overlaying soils and land use map for land use condition in
1992. On average each sub-basin contained 6.3 HRUs with mean area of each HRU as 27.9
hectares. When NLCD-2001 land use map was overlaid on the same sub-basins with same
threshold level, a total of 2158 HRUs were identified. On average each sub-basin contained
6.52 HRUs with the mean area of 27.0 hectares.
Table 2 contains the results from the SWAT run with two sets of land cover data NLCD-1992
and NLCD-2001. Simulations were run for 30 years from 1975 to 2004. SWAT results for the
initial years are generally taken as warm up results. Results from the first 10 years were
used for calibration and the later twenty years of results (1985-2004) were used for
comparative analysis.
NLCD-1992 NLCD-2001
Variable Change p-value^
Mean Std. Dev. Mean Std. Dev.
Precipitation (mm) 1204.29 194.46 1204.29 194.46 n/a n/a
Water Yield (mm)* 218.58 75.57 224.82 75.12 2.85% 0.001
Sediment Yield (t/ha) 13.56 7.89 15.60 8.46 15.04% 0.000
Organic Nitrogen (kg/ha) 1.89 1.03 1.98 1.00 4.76% 0.021
Organic Phosphorus (kg/ha) 0.22 0.12 0.23 0.12 4.55% 0.027
^ p-value are for Pair-wise t-test for the comparison of means

* Water Yield = Surface Flow + Lateral Flow + Ground Water Flow–Transportation Loss
Table 2. Average Annual Basin Values (1985–2004)
The modeled area received 1204.29 mm of precipitation per year. When the land use
distribution changed from that of 1992 to 2001, there was 2.85% increase on water yield
(p=0.001) that washed away 15.04% more sediment (p<0.001). The sediment level increased
from 13.56 t/ha to 15.60 t/ha, equivalent of additional 119,096 tons of sedimentation per
year from the entire watershed area. Similarly organic nitrogen runoff increased by 4.76%
(p=0.021) and organic phosphorus increased by 4.55% (p=0.027). The p-values are based on
pairwise t-test for the comparison of means and are significant at 5% level of significance.
Graphical comparison of sediment loadings and organic nitrogen and phosphorus runoff
under two land use scenarios are given in figures 7 – 9.
Average Annual Organic Nitrogen Runoff (kg/ha)
5.0
4.5
4.0
Nitrogen (kg/ha)
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
NLCD-1992 NLCD-2001
Fig. 7. Simulated average organic nitrogen runoff (1985-2004)
Average Annual Organic Phosphorus Runoff (kg/ha)
0.6
0.5
Phosphorus (kg/ha)
0.4
0.3
0.2
0.1
0.0
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
NLCD-1992 NLCD-2001
Fig. 8. Simulated average organic phosphorus runoff (1985-2004)

Average Annual Sedimentation (t/ha)
40
35
30
Sediment (t/ha)
25
20
15
10
0
1985 1987 1989 1991 1993 1995 1997 1999 2001 2003
NLCD-1992 NLCD-2001
Fig. 9. Simulated average sediment runoff (1985-2004)
6. Discussion
This study demonstrates how changes in land cover scenarios can affect the water quality in
the catchment area with quantitative findings. These results are basic to the understanding
of water quality impacts of land use change which ultimately helps regional planners and
watershed management policy makers by providing the estimates of changes in water
quality when land use changes in an area over long run. King and Balogh (2001) applied the
methodology to simulate water quality impacts associated with converting farmland and
forests to turfgrass. They found SWAT to be a useful tool in evaluating risk assessments
associated with land use conversions.
The application of the model can be extended to include much detailed land use and soil
distribution facilitating the short-run assessment of point and non-point source pollution.
Using recent satellite images to create maps with details of cropping patterns will help to
understand the impact of alternative best management practices such as minimum or no-
tillage practices and reduced use of fertilizers and pesticides (e.g. Santhi et al., 2001). The
fact that the watershed level results can be disaggregated to the individual hydrological
response units helps to achieve precise estimate of water quality impact at much smaller
level. While these biophysical models are extremely valuable in assessing the physical
impacts of BMPs on quantity and quality of water bodies, the results can be combined into a
more complex bioeconomic modeling to estimate the impacts of land use changes with
respect to economic profitability and the water quality. Bhattarai et al. (2008) demonstrated
that a reduction in cropland improved the water quality with an overall loss in agricultural
returns. Results from these biophysical results combined with economic analysis helps in
setting up a policy that maintains the balance between the water quality and economic
incentives among the stakeholders.
7. Limitations of the study

The land in the study area is predominantly forested. The shares of other land uses such as
developed land, agricultural land, and wetland, are relatively small. During the HRU
distribution process, the model picks up the dominant land use as determined by the
threshold level. Thus unique land uses with very small areas may be ignored, leaving out
the effects of specific land use in a localized area. The selected sub-watershed does not have
enough water quality observation sites and USGS gage stations to calibrate and validate
simulated results with the real-time observed daily streamflow data and water quality data.
The objective of the study was to compare the levels of nitrogen, phosphorus and sediment
achieved under alternative land use scenarios at the aggregated level, ceteris paribus, the
results are still valid for comparative analysis.
8. Conclusion
Change in land use distribution had a significant impact on the levels of organic nitrogen
and phosphorus runoff and sediment loadings coming out of the watershed. The forest land
decreased by 16.6% and mostly the developmental use and agricultural cultivation took that
share resulting in overall average annual nutrient runoff increase by 4.76% for nitrogen and
4.55% for phosphorus. Average annual sediment loadings at the basin level increased by
more than 15.04%. With the same amount of rainfall and climate conditions, water flowing
to the main channel from the catchment area was 2.85% higher. This result confirms to
hypothesis that with less vegetative cover, more impervious surfaces in urban areas and
increasing fragile agricultural land, there would be less percolation and higher runoff. This
resulted in higher organic nitrogen and phosphorus concentration in the discharged water.
This study also indicates SWAT’s relative strength in quantifying the effect of changes in
land use distribution and land management scenarios on water quality.
9. Acknowledgement
Partial funding for this research was provided by the Center for Forest Sustainability
through the Peaks of Excellence program at Auburn University. Additional funds were also
provided by Auburn University’s Environmental Institute. The views expressed in this
article are of authors and do not necessarily represent the views of the funding institution or
the institution the authors represent.
10. References
Bhattarai, G., P. Srivastava, L. Marzen, D. Hite, and U. Hatch. 2008. Assessment of Economic
and Water Quality Impacts of Land Use Change Using a Simple Bioeconomic
Model. Environmental Management 42(1): 122-131.
Bhuyan, S.J., L.J. Marzen, J.K. Koelliker, J.A. Harrington Jr., and P.L. Barnes. 2001.
“Assessment of Runoff and Sediment Yield Using Remote Sensing, GIS and
AGNPS.” Journal of Soil and Water Conservation 57(5):351-364.
Borah, D.K., and M. Bera. 2003. SWAT Model Background and Application Reviews. Paper
presented at American Society for Engineering in Agricultural, Food, and
Biological Systems Annual International Meeting, Las Vegas, Nevada, USA. July
27-July 30, 2003.
DiLuzio, M., R. Srinivasan, J.G. Arnold, and S.L. Neitsch. 2002. ArcView Interface for
SWAT2000: User’s Guide. TWRI Report TR-193, Texas Water Resources Institute,
Texas. 2002.
Intarapapong, W., D. Hite, and L. Reinschmiedt. 2002. “Water Quality Impacts of
Conservation Agricultural Practices in the Mississippi Delta.” Journal of the
American Water Resources Association, 38(2):507-515, April 2002.
King, K. W. and J. C. Balogh. 2001. Water Quality Impacts Associated with Converting
Farmland and Forests to Turfgrass. Transactions of the ASAE 44(3): 569-576.
Liao, H., and U.S. Tim. 1997. “An Interactive Modeling Environment for Non-point Source
Pollution Control." Journal of the American Water Resources Association 33(3):1-13.
Maidment, D. R. (eds.). 2002. Arc Hydro: GIS for Water Resources. ESRI Press, Redlands,
California (pp. 73).
Mankin, K. R., J. K. Koelliker, and P. K. Kalita. 1999. “Watershed and Lake Water Quality
Assessment: An Integrated Modeling Approach.” Journal of the American Water
Resources Association 35(5):1069-1080.
Marzen, L.J., S.J.Bhuyan, J.A. Harrington, J.K. Koelliker, L.D. Frees, and C.G. Volkman. 2000.
“Water Quality Modeling in the Red Rock Creek Watershed, Kansas.” Proceedings of
the Applied Geography Conference 23: 175-182.
Neitsch, S.L., J.G. Arnold, J.R. Kiniry, J.R. Williams, and K.W. King. 2002. Soil and Water
Assessment Tool: Theoretical Documentation, Version 2000. Texas Water Resources
Institute, College Station, Texas, TWRI Report TR-191.
Neitsch, S.L., J.G. Arnold, J.R. Kiniry, R. Srinivasan, and J.R. Williams. 2002. Soil and Water
Assessment Tool: User’s Guide, Version 2000. Texas Water Resources Institute,
College Station, Texas, TWRI Report TR-192.
Saleh, A., J.G. Arnold, P.W. Gassman, L.W. Hauck, W.D. Rosenthal, R.R. Williams, and
A.M.S. McFarland. 2000. Application of SWAT for the upper north Bosque
Watershed. Transactions of the ASAE 43(5):1077-1087.
Santhi, C., J. G. Arnold, J. R. Williams, L. M. Hauck, and W. A. Dugas. 2001. Applications of
a Watershed Model to Evaluated Management Effects on Point and Nonpoint
Source Pollution. Transactions of the ASAE 44(6): 1559-1570.
U.S. Department of Agriculture Natural Resources Conservation Service (USDA-NRCS).
2002. Major Land Resource Area Regional #15. Auburn, AL. Available online at
http://www.mo15.nrcs.usda.gov/alsoilnt/almlramp.html.
United States Census Bureau, 1990. States and County Quick Facts. Online at:
http://quickfacts.census.gov/qfd/index.html
United States Census Bureau, 2000. States and County Quick Facts. Online at:
http://quickfacts.census.gov/qfd/index.html
United States Environment Protection Agency (USEPA). Total Maximum Daily Loads
Factsheet. Available online at
http://www.epa.gov/owow/tmdl/ (accessed 11/18/2003)
United States Environmental Protection Agency (USEPA). 1998. “National Water Quality
Inventory: 1996 Report to Congress.” Washington, DC.
12
Flood Progression Modelling

and Impact Analysis
D. Mioc1 et al.,*
1National Space Institute, Technical University of Denmark,
Denmark
1. Introduction
People living in the lower valley of the St. John River, New Brunswick, Canada, frequently
experience flooding when the river overflows its banks during spring ice melt and rain.
Media reports reveal devastating effects of the latest floods that hit New Brunswick, Canada
during the summer of 2008 (CTV, 2008). The rising water levels forced the closure of the
New Brunswick Legislature, and also resulted in the temporary closures of the international
bridge that links the Province to the United States of America. It also closed the operation of
the Gagetown ferry. Fredericton experienced its worst floods in 1973, when the St John River
reached the 8.6 m mark (ENB-MAL, 1979).
The 2008 flood was recorded as one of the major floods experienced in Fredericton after the
1973 flood (see Figures 1 and 2). On April 24th 2008, due to rapid melting of snow set by an
unusually severe winter and combined with intense rainfall, the water level of St. John River
reached 7.2 m (TC, 2008). The water levels in Fredericton raised by a meter overnight to 8.33
m on May 1st. Raising St. John River levels peaked at 8.36 m on May 2nd, almost reaching the
previous record of 8.61 m set in 1973 (CIWD, 1974).
The closure of roads, government buildings followed by the evacuation of people and their
possessions during floods is necessary to avoid the loss of life and property. Raising river
water levels could affect electrical, water and telecommunication facilities. It could also
affect the sewage system. In such a situation, the public buildings washrooms cannot be
used. The question that the government officials are facing is: “When can an office be
declared risky to occupy?” The decisions to close public utilities require strong reasons.
Such decisions have social, economic and political impacts on the community.
City Managers will require reliable support for the decision to close down government
infrastructure. To facilitate this process, we propose the use of 3D flood modelling,
*F. Anton3, B. Nickerson2, M. Santos6, P. Adda6, T. Tienaah6, A. Ahmad6, M. Mezouaghi6,

E. MacGillivray4, A. Morton4 and P. Tang5
2Faculty of Computer Science, University of New Brunswick, Canada
3 Department of Informatics and Mathematical Modelling, Technical University of Denmark, Denmark
4Emergency Measures Organization, Victoria Health Centre, 65 Brunswick Street, Fredericton, NB, Canada
5New Brunswick Department of Environment, Marysville Place, Fredericton, NB, Canada
6Department of Geodesy and Geomatics Engineering, University of New Brunswick, Canada

embedded on 3D terrain together with infrastructure (electrical power, water,

telecommunications) and buildings to make correct decisions on when buildings and
infrastructure are declared unsafe to use.
Fig. 1. Flood monitoring using satellite imagery

Flood Progression Modelling and Impact Analysis 229
Fig. 2. Evacuation of diary cows during the flood in 2008
2. Previous research
Since ancient times, humans have developed means to monitor flood levels and to some
extent, predict the rate of flood rise. People in medieval times marked animals and pushed
them down the river to see how deep the river was. According to the director of EMO,
Ernest McGillivray, his grandfather had a self-calibrated stick that he used to insert in the
river to compare and forecast river flood levels.
Today, more innovative technologies have been developed to study floodings (Jones, 2004;
Marks & Bates 2000). These include satellite remote sensing, aerial photogrammetry and
LiDAR. Such technologies are combined with computer terrain modelling tools to create
scenarios for analysis, as in the case of Geographic Information Systems (GIS).
Previously, the New Brunswick Emergency Measures Organization (EMO) in collaboration
with the University of New Brunswick, Canada developed a flood model (available from
http://www.gnb.ca/public/Riverwatch/index-e.asp) for the area of lower St. John River,
New Brunswick, Canada (EMO, 2008).
The research done so far on early mapping and flood monitoring in the lower St. John
produced a web-based system for flood warning (Mioc et al., 2008; Mioc et al., 2010),
publicly available to the population living in this area (see Figures 3 and 4). However, the
accuracy of existing elevation data was a major problem in this project. The maps we were
able to produce were not accurate enough to effectively warn the population and to plan the
evacuation. Another problem we faced in the 2008 flood was that even though the buildings
were not affected by the flood the electrical and water facilities were not functioning, so the
Fig. 3. Fly through simulation of the extreme flood in 1973
Fig. 4. Riverwatch website for flood warning

buildings and houses were inhabitable. The elevated groundwater levels would affect the
power cables causing power outages and the old plumbing installations would cause sewer
back-propagation due to the pressure of the rising water from St. John River.
Current visualization methods cannot adequately represent the different perspectives of the
affected infrastructure. It can be seen that the building models are represented by polygons
only and the detailed 3D views of the buildings and terrain affected by the flooding do not
exist. As a result, the public may not have adequate technical or analytical know-how to
analyze the polygons. Therefore, they may not find the River watch website very useful (see
Figure 4). To overcome these problems we had to develop a new digital terrain model and
new techniques for flood modelling. In order to improve the existing digital elevation model
we needed to use new technologies for data acquisition (Moore et al., 2005).
3. Data collection and processing

In our research efforts to improve the accuracy of the elevation data, we used new Light
Detection and Ranging (LiDAR) data acquisition and LiDAR data processing. Furthermore,
the hydrological modelling was integrated with processed LiDAR data within a 3D GIS
application.
Fig. 5. LiDAR data processing

Light Detection and Ranging (LiDAR) (see Figure 5) has become a useful technology for
collecting point cloud data for three-dimensional surface representations. The ability to
quantify vegetation height, the roughness of the ground surface, ground elevation, and
nearby buildings has greatly improved the parameterization of the two-dimensional
hydraulic models, currently at the forefront of flood impact assessment studies (Kraus &
Pfeifer, 2001; Hodgson & Bresnahan, 2004).
3.1 Available data

In this research, the newly acquired LiDAR data and Tidal gauge readings from May 1st to
May 4th, 2008 were used to delineate the extent of flood during the 2008 flood in Fredericton.
Elevation data of Fredericton were obtained using LiDAR techniques and have the vertical
accuracy of 15 cm on the hard surface. For this section, only data from Fredericton
downtown area are used, as it was the most active area exposed to the flood. The coverage
area is shown in Figure 6. Coordinates of river gauges and recorded river levels for four
cross-sections over a period of four days (May 1st to May 4th, 2008) are shown in Table 1.
Tidal gauge readings were extended to a profile across the river in order to facilitate spatial
analysis.
Table 1. Tidal Gauge Readings

Light Detection and Ranging (LiDAR) is a data collection technique that uses a beam of light
to make range-resolved remote measurement of features within the path of a reflected beam
of light. The New Brunswick government, through the Department of the Environment and
Emergency Measures Organization (EMO), collected LiDAR data sets for great part of the
province of New Brunswick after the flood in 2008.
100
95
90
85
80
75
70
65
60
55
50
45
40
35
30
25
20
15
10
5
0
-5
-10
-15
Fig. 6. Downtown Fredericton LiDAR data

Fig. 7. LiDAR data; First Return, part of Fredericton, NB
3.2 Data processing

For this project, accurate planimetric coordinates and orthometric heights were obtained for
the 3D flood modelling. In order to process LiDAR data (Elaksher & Bethel, 2002a, 2002b),
we applied the following workflow of tasks:
1. Find LiDAR strip for the project area and extract returns of LiDAR beams (Figures 6
and 7).
2. Extract ground surface, i.e. ground DTM of the area of interest (Figure 8).
3. Extract building footprints from the returns of the LiDAR data (Figure 9).
4. Edit footprints by comparing them with cadastral data. The cadastral map does not
contain height values and is not recent. However, based on different predefined
parameters, the algorithm used to extract the building footprints produce varying
results. The best fit is shown in Figure 10. The results were compared to the digital
cadastral map of the study area. Further editing was necessary to correct the
planimetric errors from the building footprints obtained from extraction process.
5. Proceed with other modelling tasks (shown in Figure 11) and add attributes and other
data necessary for analysis (see Figures 12 and 13). The buildings and utilities,
including electrical power, water and telecommunications are all modelled in 3D. Their
attributes are also included in the modelling.
LiDAR data is usually collected with reference to ground control points using GPS methods.
Subsequently, the output coordinates are ellipsoidal. To obtain accurate flood modelling
together with Digital Terrain Model, the data originally available in the geographic
coordinates using the ellipsoid defined by the WGS84 geodetic datum are finally
transformed to UTM and the CGVD28 vertical datum. To reduce the dataset and improve
the processing speed, orthometric heights above 20m were filtered out, since we did not
expect flood levels to go above this height. The transformed coordinates are used as the
input to the GIS system in order to model a Digital Terrain Model (DTM). Layers containing
flood profiles and polygons are created for each day of the extreme flood (1st to 4th of May)
in 2008. In addition the extreme flood from 1973 was modelled as well and used for
comparison. Existing utilities and buildings in downtown Fredericton we modelled in 3D
and geo-referenced to the WGS 84 geodetic datum and the CGVD28 vertical datum. All the
developed models and DTMs are then combined under the same coordinate system.
For the efficient decision support system, different flood scenarios were modelled for
different flood levels (Sanders et al., 2005; Dal Cin et al., 2009). The technologies for 3D
modelling are available in commercial and public domain software. One such tool that was
used in this project was Google SketchUp (Chopra, 2007). Although limited in its interaction
with other spatial features, Google SketchUp provided the tools needed to model the main
features of the buildings (buildings geometry and the texture for the façade and the roof)
and surrounding utilities for the test area. As part of future work, advanced tools in the
CityGML (Kolbe et al., 2008) could be used for more interactive results.
Fig. 8. Extracted ground surface from LiDAR data

Fig. 9. Extracted Building polygons from LiDAR data
Fig. 10. Extracted building polygons (thin brown lines) compared with Cadastral buildings
blueprints (thick red lines)
Fig. 11. The workflow of detailed methods and techniques applied for modelling of flooded
buildings and infrastructure.
Fig. 12. Flood progression modelling

Fig. 13. The 3D model of buildings overlaid over the water and electrical utilities
4. Flood modelling and forecasts

The advanced modelling method we used in this research is based on 3D modelling of
flooding, buildings and government infrastructure. The 3D visualizations of the floodplain
and its effect on utilities can be discussed in more comprehensive terms, among engineers,
government executives and the public. To showcase such a possibility, buildings of
downtown Fredericton were extruded in 3D and modelled in detail. Figure 11 shows the
processing steps in obtaining 3D models. Embedding the 3D models with the utilities and
DTM, through the process of modeling different water rise situations, gives different
scenarios of flooding, which can be used for further analysis. Based on this analysis, we
show flood levels that could render utilities unavailable and make the government
infrastructure risky to occupy.
4.1 Floodplain computations

To accurately compute the floodplain it is important to obtain topography of floodplain
areas; the bathymetry of rivers; snow information; storm surges (rainfall forecast); and
temperature information to create hydraulic models for effective prediction of inundation
areas and risk probabilities. However, water levels obtained by hydraulic modelling do not
tell much about the severity and extent of a flood. This motivates the modelling and a
visualization of predicted flood areas using GIS. The spatial delineation of flood zones using
GIS has become a new research area following the advancement of technologies for data
collection (Noman et al, 2001, 2003).
This research uses different spatial analysis tools to create floodplains from LiDAR data in
Fredericton area and water gauges for Saint John River. For hydrological modelling we
used DWOPER (Fread, 1992, 1993 ; Fread & Lewis, 1998) (as described in Mioc et al., 2010).
The results of hydrological modelling were then used within a GIS for 3-dimensional
modelling of flood extents (Mioc et al., 2010).
Following our processing workflow (see Figure 11), there were two main objectives in this
part of the research:
1. Compare the DTM resulting from LiDAR data with DTMs resulting from water gauges
to find the flood extents for Fredericton downtown area.
2. The second objective is to create a single TIN for both LiDAR data and water gauges to
calculate the difference of the volume, which represents the floodplain.
The flood progression from May 1st to May 4th, 2008 is analyzed (see Figure 12). Using the
results of spatial analysis, a flood prediction model was developed for emergency planning
during future floods. In this phase of our research, we were able to obtain the delineation of
flood zones using LiDAR and Tidal height information (available from water gauges). In
addition, we were able to integrate a number of processes that make flood forecast possible:
the acquisition and processing of the elevation data; the use of hydrological software to
simulate models of flow across floodplains; the use of spatial analysis software (GIS) that
turns the modelling results into maps and overlays them on other layers (thematic maps or
aerial photographs); and software that makes these models and predictions available on the
Internet in a flexible and user-friendly way.
From the computed floodplain, displayed as superimposed polygon layers, we visualize
that the major flood extent occurred from May 1st to May 2nd, with the peak for flood on
May 2nd. Furthermore, the flood subsided from May 3rd to May 4th. The system we
developed allows the computation of floodplain for predicted flood peak (shown in red on
Figure 12) that is critical for emergency managers. Furthermore, the flood modelling results
are used to develop a three dimensional model of flooded buildings combined with some
city infrastructure, roads, water and electrical utilities (see Figures 14, 15 and 16).
4.2 3D modelling of flooded buildings and infrastructure

The method of simulating and predicting floods and its effects on buildings and utilities
provides powerful visual representation for decision making on when the buildings in the
flood zone may be safe for people to occupy. Traditional paper maps and digital maps may
not give us the possibility to do a 3D visualization in order to study the detailed effect of a
flood situation on utilities.
In this research, we used LiDAR data and the application of 3D modelling in order to
provide an analysis of the risk of floods on government buildings and utilities. LiDAR data
provides a cheaper, faster and denser 3D coverage of features for 3D mapping. LiDAR data
was acquired for the city of Fredericton after the flood in 2008 and processed to generate 3D
animated views. To further enhance visual perception, 3D buildings, infrastructure and
utilities were integrated within a 3D GIS application. The resulting 3D view does not only
display a clear-to-nature scenario, but provides also a more realistic outlook of the buildings
and infrastructure during floods. Finally, a flood scene was produced for each of the
forecasted flood levels for visualization via Web interface.
Using different extrusion heights to represent different flood scenes, it is possible to
simulate different flood progression events. The pictorial scene, representation of the
building, floods and its effects can be clearly visualized and analyzed. Figure 15 shows a 3D
view of the flooding in May 2008. The 3D buildings and utilities can be seen consecutively
as a result of applying transparency to the thematic layers. In Figure 16, the utility lines are
embedded in the 3D models. Figures 17 and 18 present a 3D visual model of the submerged
newly built Public Washroom at the Military Guard Square, in Fredericton, Canada in an
extreme flood scenario. At this level of water rise, the electrical boxes would be flooded. It is
visible from the model that during flooding, surrounding areas including the Public Library,
the Armory and the Military depots will be out of use and certainly inaccessible.
Fig. 14. The 3D model of buildings overlaid over the water and electrical utilities that will be
affected by the flood
The DTMs in Figures 15 and 16 show the natural critical point at which water will begin to
flow upwards. When ground water rises up to this level, waste matter from the sewage
system will flow upward, under pressure from rising water from the river. The washroom
facility may not be used under these circumstances. Computing the floodplain for the 20
year statistical flood showed that many parts of Fredericton may have been build on a
floodplain. The new analysis of DTMs combined with the groundwater levels shows that, if
ground water levels rise across the city, many homes and governmental buildings will be
flooded. The electrical utilities and sewage system that are laid in the underground will be
affected as well resulting with the sewer back-propagation and the electric power outages.
The situation is worse in the downtown area, which has the lowest heights. Priority
emergency decisions can be made in this situation to close the downtown offices and
infrastructure first, at the start of rising water levels. Based on the results of the overlaying
the floodplains with existing utilities and infrastructure, it can be decided when it is risky to
occupy or use the buildings or the infrastructure.
Fig. 15. 3D models of selected buildings in Fredericton integrated with DTM and flood
model
Daily automatic generation of flood polygons from the data provided by the existing online
River Watch application (see http://www.gnb.ca/public/Riverwatch/index-e.asp) can
produce an animated 3D video, which can be uploaded on the Riverwatch website to
provide updated 3D models to residents.
It can be seen clearly from the comparison of the model and the picture (shown in the Figure
19) captured during the 2008 flood, that the levels of flooding are the same. The water just
touches the back gates of the Fredericton Public library on both of these. The accuracy of the
flood model depends on the vertical accuracy of the LiDAR datasets and the accuracy of the
hydrological modelling.
Fig. 16. 3D models of selected buildings in Fredericton affected by the flood

The 3D GIS application provides a better platform for visualizing flood situations than
previously done in 2D maps. All exterior parts of a building could be visualized in detail
during a flood event. This provides a better tool for analyzing and preparing for emergency
measures. It also presents a near to reality situation that can easily be understood. Provincial
Ministers and decision makers who may not be familiar with GIS analytical tools and Query
Languages can now understand technical discussions on flood analysis through the use of
3D flood models, which are close to reality. It is also possible to simulate floodplain
polygons for different river water levels in order to produce different flood scenarios.
Simulation can also be used to trace and analyze underground utilities by making thematic
layers transparent. Flood scene animations can be published to website for public access.
Fig. 17. 3D Washroom, with power box, in the extreme100 year simulated flood
Fig. 18. 3D model of a washroom, with electrical power box, in 1973 simulated flood
Fig. 19. 3D Model compared with Photograph taken during the flood in 2008
5. Conclusions
The new flood prediction model that computes accurately floodplain polygons directly from
the results of hydrological modelling allows emergency managers to access the impact of the
flood before it occurs and better prepare for evacuation of the population and flood rescue.
The method of simulating and predicting floods and its effects on utilities provides
powerful visual representation for decision making on when buildings in the flood zone
may be safe for people to occupy. Traditional paper maps and digital maps may not give us
the possibility to do a 3D visualization of the detailed effect of a flood on utilities and
infrastructure.
This research explores the application of 3D modeling using LiDAR data to provide an
analysis of the risk of floods on government buildings and utilities. LiDAR data provides a
cheaper, faster and denser coverage of features for 3D mapping. LiDAR data was processed
to generate 3D maps. By employing accurate coordinate conversion and transformations
with respect to the geoid, a Digital Terrain Model (DTM) was created. Floodplain
delineation was computed by intersecting the Digital Terrain Model with the simulated
water levels. Furthermore, to enhance visual perception of the upcoming flood, 3D
buildings, infrastructure and utilities were modelled for the city downtown area. The DTM
and the 3D models of the government buildings, infrastructure and utilities were overlaid
and presented as a 3D animation. The resulting 3D view does not only register a clear-to-
nature scenario, but also provides a more discerning outlook of the buildings and
infrastructure during floods. Finally, in this research we have clearly shown that GIS and
LiDAR technologies combined with hydrological modelling can significantly improve the
decision making and visualization of flood impact needed for early emergency planning
and flood rescue.
6. Acknowledgment
The authors would like to acknowledge the generous contribution of time, materials and
resources to this project by the New Brunswick Department of Transportation.
This project was financially supported in part by the N.B. Emergency Measures
Organization and the Canadian Department of Natural Resources Geoconnections program
as well as the University of New Brunswick and the New Brunswick Innovation Foundation
(NBIF). The authors would also like to acknowledge the financial contribution of EMO to
the acquisition of new LiDAR datasets (available for this project) after the flood in 2008.
7. References
Alharthy, A. & Bethel, J. (2002). Heuristic filtering and 3D feature extraction from LiDAR
data. International Archives Of Photogrammetry, Remote Sensing and Spatial Information
Sciences, Vol. 34, No.3/A, pp. 29-34.
Canada Inland Waters Directorate, Atlantic Region, (CIWD), (1974), New Brunswick Flood,
April-May, 1973, Ottawa: Inland Waters Directorate, Atlantic Region, 1974, 114
pages.
Chopra, A., (2007). Google SketchUp for Dummies, John Wiley & Sons Ltd., Hoboken, US, 398
pages.
Canadian Television, CTV, (CTV), (2008). News Headlines, Accessed on May 1, 2008. from
http://www.ctv.ca.
Dal Cin, C., Moens, L., Dierickx, PH., Bastin, G., and Zech, Y. (2005). “An Integrated
Approach for Realtime Floodmap Forecasting on the Belgian Meuse River”. Natural
Hazards 36 (1-2), 237-256, Springer Netherlands. [Online] June 10, 2009.URL:
http://springerlink.com/content/tq643818443361k8/fulltext.pdf
Elaksher, A. F. & Bethel J. S. (2002a). Building extraction using LiDAR data. Proceedings of
ASPRS-ACSM annual conference and FIG XXII congress, 22 pages.
Elaksher, A.F. & Bethel J.S. (2002b). Reconstructing 3D buildings from LiDAR data.
International Archives Of Photogrammetry Remote Sensing and Spatial Information
Sciences Vol. 34, No. 3/A, pp. 102-107.
EMO Fredericton, (EMO), (2008). Flood Warning website, Available from:
http://www.gnb.ca/public/Riverwatch/index-e.asp.
Environment New Brunswick & MacLaren Atlantic Limited, New Brunswick, (ENB-MAL),
(1979). Canada-New Brunswick Flood Damage Reduction Program:
Hydrotechnical Studies of the Saint John River from McKinley Ferry to Lower
Jemseg, Fredericton, 116 pages.
Fread, D.L. (1992). Flow Routing, Chapter 10, Handbook of Hydrology. Editor E.R.
Maidment, pp. 10.1-10.36.
Fread, D.L. (1993). NWS FLDWAV Model: The Replacement of DAMBRK for Dam-Break
Flood Prediction, Dam Safety’93. Proceedings of the 10th Annual ASDSO Conference,
Kansas City, Missouri, pp. 177-184.
Fread, D.L. & Lewis, J.M. (1998). NWS FLDWAV MODEL: Theoretical description and User
documentation, Hydrologic Research Laboratory, Office of Hydrology, National
Weather Service (NWS), Silver Spring, Maryland USA, November, 335 pages.
Hodgson, M.E. & Bresnahan, P. (2004). Accuracy of airborne LiDAR-derived elevation:
Empirical assessment and error budget. Photogrammetric Engineering & Remote
Sensing, Vol. 70, No. 3, pp. 331-40.
Jones, J. L. (2004). "Mapping a Flood...Before It Happens". U.S. Geological Survey Fact Sheet
2004-3060. [Online] June 5, 2009., Available from
http://pubs.usgs.gov/fs/2004/3060/pdf/fs20043060.pdf
Kraus, K. & Pfeifer, N. (2001). Advanced DTM generation from LiDAR data. International
Archives Of Photogrammetry Remote Sensing And Spatial Information Sciences
34(3/W4):23-30.
Kolbe, T. H., Gröger, G., & Plümer, L. (2008). CityGML - 3D City Models and their Potential
for Emergency Response. In: Zlatanova, S. and Li, J. (Eds.) Geo-Information
technology for emergency response. ISPRS book series, Taylor&Francis, pp. 257-
274.
Marks, K. & Bates, P. (2000). Integration of high-resolution topographic data with floodplain
flow models. Hydrological Processes, Vol. 14, No. 11, pp. 2109-2122.
Mioc, D; Nickerson B.; McGillivray, E.; Morton A.; Anton, F.; Fraser, D.; Tang, P. & Liang, G.
(2008). Early warning and mapping for flood disasters, Proceedings of 21st ISPRS
Conference, China, Bejing, 2008, 6 pages.
Mioc, D., Nickerson, B., Anton, F., MacGillivray, E., Morton, A., Fraser, D., Tang, P. &
Kam, A. (2010). Early Warning And On-Line Mapping For Flood Events, Geoscience
13
Providing Efficient Decision Support

for Green Operations Management:
An Integrated Perspective
Shaofeng Liu1 and Meili Jiang2
1University of Plymouth, Plymouth
2Hunan Institute of Technology, Hengyang
1UK
2China
1. Introduction
Green operations management (GOM) has emerged to address the environmental and social
issues in operations management, so that the Triple Bottom Line (3BL) sustainability can be
achieved simultaneously (Rao & Holt, 2005; Zhu, Sarkis & Geng, 2005). The concept of GOM
was originally formed in the 1990s. Early research on GOM mainly directed towards
segmented areas of operations management, such as quality management. Over the past
decades, GOM has attracted significant research interests from academia, many issues
remain under-addressed which have hindered the effectiveness of GOM practice (Zhao &
Gu, 2009; Yang et al, 2010), although the needs for and benefits of GOM cannot be
overemphasised for sustainable development (Svensson, 2007). One of the main reasons for
GOM lagging behind quality management advances has been identified as lack of true
integration of environmental and social objectives into business operations, i.e.
environmental management and social values were viewed as narrow corporate legal
functions, primarily concerned with reacting to environmental legislation and social codes
of practice. Subsequently research and managerial actions focused on buffering the
operations function from external forces in order to improve efficiency, reduce cost, and
increase quality (Carter & Rogers, 2008; White & Lee, 2009). Research further reveals that the
root cause behind the company’s isolated approach to the 3BL sustainability is not because
the managers do not appreciate the importance and urgency of addressing them, but lack of
efficient support for the management of the complexity of sustainable decisions, especially
the provision of powerful analysis approach to support effective decision evaluation (Hill,
2001; Taylor & Taylor, 2009; Zhao & Gu, 2009).
This paper proposes an integrated sustainability analysis approach to provide holistic
decision evaluation and support for GOM decision making. There are two key objectives to
explore the integrated approach: (a) to understand the GOM decision support requirements
from a whole life cycle perspective; (b) to address the GOM decision support issue using
multiple decision criteria. Based on a case study in production operations area, the paper
concludes that the integrated sustainability analysis can provide more efficient and effective
support to decision making in GOM.
This chapter is organised as follows. Next section reviews related work on methods and
tools that have been developed to address GOM decision issues, and identifies the gap in
literature. Section 3 proposes an integrated approach for systematic analysis of sustainability
in GOM. Application of the integrated approach to real operations situation is discussed in
Section 4. Then Section 5 reflects on the strengths and limitations of the proposed approach,
and draws conclusions.
2. Related work
Sustainability or sustainable development was first defined by the World Commission on
Environment and Development as the “development that meets the needs of present
generations while not compromising the ability of future generations to meet their needs”
(WCED, 1987). It has then been recognised as one of the greatest challenges facing the world
(Ulhoi, 1995; Wilkinson et al, 2001; Bateman, 2005; Espinosa et al, 2008). For development to
be sustainable, it is essential to integrate environmental, social and economic considerations
into the action of greening operations (i.e. the transformation processes which produce
usable goods and services) (Kelly, 1998; Gauthier, 2005; Lee & Klassen, 2008), as operations
have the greatest environmental and social impacts among all business functions of a
manufacturer (Rao, 2004; Nunes & Bennett, 2010). In the context of sustainable
development, operations have to be understood from a network’s perspective, that is,
operations include not only manufacturing, but also design and supply chain management
activities across products, processes and systems (Liu and Young, 2004; Geldermann et al,
2007). Without proper consideration of inter-relationships and coherent integration between
different operations activities, sustainability objectives cannot be achieved (Sarkis, 2003; Zhu
et al, 2005; Matthews et al, 2006). There has been a series of overlapping endeavours and
research efforts in literature aiming to address the environmental and social issues in
operations management, but earlier efforts have been mostly segmented and uncoordinated.
More recently, GOM has been investigated from a more integrative perspective instead of a
constraint perspective, where environmental management and corporate social
responsibility are viewed as integral components of an enterprise’s operations system (Yang
et al, 2010). It means that the research foci have shifted to the exploration of the coherent
integration of environmental and corporate social responsibility with operations
management through business managers’ proactive decision making (Ding et al, 2009; Liu et
al, 2010). Figure 1 illustrates the key ideas of achieving sustainability objectives through
holistic decision making in a sustainable operations system. Compared with traditional
open operations model, i.e. the input-transformation-output model (Slack et al, 2010), there
are four key features in the sustainable operations system:
1. A sustainable operations system is a closed rather than an open system, i.e. the material
(including waste and used products), energy and information produced by the
operations system should be treated and fed back as inputs to keep the system self-
sustainable;
2. The transformation process not only includes the functions and activities that produce
products and service, but also includes that of environmental and social improvements;
3. Wider stakeholders’ requirements need to be addressed. Apart from customers, other
important stakeholders include environment, community, employee, public etc.
Therefore, the requirements such as discharge from operations process, information
and social benefits have to be properly addressed;
for Green Operations Management: An Integrated Perspective 249
4. The role of suppliers is changing. Apart from the provision of materials, energy,
information and human resources, suppliers are also responsible for recovering used
products and materials.
Holistic decision making

goods &
Process of service
environmental Customers
materials improvement used products
energy Operations discharge

Strategy
Environment
Suppliers information Process of social raw materials
improvement
people
information & Other
Structural Infrastructural
social benefits Stakeholders
decisions decisions
Government
regulations, Public
material
recovery Transformation process goodwill, Employee
information Community
material, energy, information
Fig. 1. Holistic decision-making in a sustainable operations system

Many decisions have to be made for green production operations in relation to design,
manufacture and supply chain management (Sakis, 2003; Galasso et al, 2009). Slack et al
(2010) classified operations strategic decisions into two main categories: structural and
infrastructural. Structural decisions are those which primarily influence design activities,
such as those relating to new product design and supply network design, while
infrastructural decisions are those which influence manufacturing planning and control, and
management improvements, such as inventory management and supplier development.
Miltenburg (2005) defined six production decision areas: production planning and control,
organisation structure and control, human resource, facilities, sourcing, and process
technology. This classification scheme was also adopted by Choudhari et al (2010), and 54
decision types were further identified under the six decision areas with 113 decision
alternative suggested for the decision types. However, the papers did not provide
discussion on how the optimal decision choice could be reached for each decision type.
There has been a broad consensus that decision evaluation holds the key to reaching
optional decision choice especially under complex decision situations (Mehrabad & Anvari,
2010). Decision evaluation enables decision makers to perform scientific analysis, to weigh,
score and rank the alternatives against decision criteria, and to assess the consequences of
each decision alternative, so that optimal decision choices become more transparent to
decision makers (Karsak & Ahiska, 2008; Chituc et al, 2009). In recent years, sustainability
analysis has emerged as an important decision aid to provide efficient and effective decision
evaluation in all aspects of business (Hodgman, 2006). Sustainability analysis is important to
operations research because decision making in GOM often has high complexity, cost and
risk. Getting the decisions right will generate not only considerable economic value, but also
great environmental and social impacts which can sharpen a company’s competitive edge
(Noran, 2010).
Sustainability analysis would be theoretically straightforward if key interacting variables
and boundaries of responsibilities were well understood by decision makers (Mihelcic et al,
2003). Unfortunately, such situations are rare, while benefits from sustainability efforts have
been elusive. Practitioners continue to grapple with how sustainability analysis should be
undertaken, due to the complexities and uncertainties of environmental systems involved
and imperfections of human reasoning (Hertwich et al, 2000; Allenby, 2000). According to
Hall & Vredenburg (2005), innovating for sustainable development is usually ambiguous,
i.e. when it is not possible to identify key parameters or when conflicting pressures are
difficult to reconcile, such ambiguities make traditional risk assessment techniques
unsuitable for GOM. Researchers further argue that sustainability analysis frequently
involves a wide range of stakeholders, many of which are not directly involved with the
company’s production operations activities. Decision makers are thus likely to have
significant difficulties in reaching the right decisions if efficient support is not available.
Powerful systematic analysis methodologies have great potential in guiding the decision
makers to navigate through the complexities and ambiguities (Matos & Hall, 2007). This
section reviews the most influential analysis methodologies that could facilitate efficient and
effective GOM decision making: life cycle assessment and multi-criteria decision analysis.
2.1 Life cycle assessment

Life cycle assessment or analysis (LCA) is regarded as a “cradle to grave” technique that can
support environmentally friendly product design, manufacture and management (Hunkeler
et al, 2003; Verghese & Lewis, 2007; Jose & Jabbour, 2010). It can be used to assess the
environmental aspects and potential impacts associated with a product, process or a system
(Matos & Hall, 2007; Kim et al, 2010). It also allows decision makers to evaluate the type and
quantity of inputs (energy, raw materials, etc.) and outputs (emissions, residues, and other
environmental impacts, etc.) of production operations in order to completely understand the
context involving product design, production, and final disposal (Fuller & Ottman, 2004;
Jose & Jabbour, 2010). Standards for the application of LCA have been set up by ISO
(International Standards Organisation). A four stage LCA process has been defined (ISO
14040, 1997), as illustrated in Figure 2.
Fig. 2. Four key stages of LCA process (ISO 14040)

LCA can be conducted along two different dimensions: Product Life Cycle (PLC) and
Operational Life Cycle (OLC) analysis. A new product progresses through a sequence of
stages from development, introduction to growth, maturity, and decline. This sequence is
known as the PLC (Sarkis, 2003; Bevilacqua et al, 2007; Gunendran and Young, 2010). On the
other hand, OLC includes stages of procurement, production, packaging, distribution, use,
end-of-life disposal and reverse logistics (Nunes & Bennett, 2010). The nature of both of
these analytical tools can generate important insights into environmentally conscious
practices in organisations, and there are close interdependencies between PLC and OLC. For
example, in the PLC introduction phase, procurement is more influential than production for
sustainable practices, whilst in the maturity and decline stages of the PLC, efficient end-of-life
and reverse logistics are more influential than distribution operations. It is also not difficult
to understand that distribution decisions such as facility locations and modes of
transportation will not only influence the forward but also the reverse logistics networks
(Bayazit & Karpak, 2007; Chan et al, 2010). However, it is widely acknowledged that
environmental methods (LCA in general, PLC and OLC analysis in specific) should be
“connected” with social and economic dimensions to help address the 3BL, and that this is
only meaningful if they are applied to support decision making process and are not just a
“disintegrated” aggregation of facts (Matos & Hall, 2007). It is advantageous that PLC and
OLC analysis are conducted to obtain a more holistic picture of the economic and ecological
impacts of production operations (Neto, et al, 2010).
2.2 Multi-criteria decision analysis

There is no doubt that GOM decision making has multiple criteria to meet simultaneously,
i.e. environmental, social and economic performance objectives. GOM decisions can envelop
quantitative, qualitative, tangible and intangible factors. Multi-Criteria Decision Analysis
(MCDA) is a generic approach that can empower decision makers to consider all the
decision criteria and decision factors, resolve the conflicts between them, and arrive at
justified choice. Over the past three decades, several variants of MCDA have been
developed. This section compares four widely used MCDA methods: AHP, ANP, fuzzy set
theory and fuzzy AHP/ANP.
Analytical Hierarchy Process (AHP) was introduced by Saaty (2005) for solving
unstructured problems. Since its introduction, AHP has become one of the most widely
used analysis methods for multi-criteria decision making. AHP requires decision makers to
provide judgements about the relative importance of each criterion and specify a preference
for each decision alternative using each criterion. The output of AHP is a prioritised ranking
of the decision alternatives based on the overall performance expressed by the decision
makers (Lee, 2009). The key techniques to successfully implement AHP include developing
a goal-criteria-alternatives hierarchy, pairwise comparisons of the importance of each
criterion and preference for each decision alterative, and mathematical synthesization to
provide an overall ranking of the decision alternatives. The strength of AHP is that it can
handle situations in which the unique subjective judgements of the individual decision
makers constitute an important part of the decision making process (Anderson et al, 2009).
However, its key drawback is that it does not take into account of the relationships between
different decision factors.
Analytic Network Process (ANP) is the evolution of AHP (Saaty & Vargas, 2006). Given the
limitations of AHP such as sole consideration of one way hierarchical relationships among
decision factors, failure to consider interaction between various factors and “rank reversal”,
ANP has been developed as a more realistic decision method. Many decision problems
cannot be built as hierarchical as in AHP because of dependencies (inner/ outer) and
influences between and within clusters (goals, criteria and alternatives). ANP provides a
more comprehensive framework to deal with decisions without making assumptions about
the independence of elements between different levels and within the same level. In fact,
ANP uses a network without the need to specify levels as in a hierarchy (Sakis, 2003; Dou &
Sakis, 2010) and allows both interaction and feedback within clusters of elements (inner
dependence) and between clusters (outer dependence). Such interaction and feedback best
captures the complex effects of interplay in sustainable production operations decision
making (Gencer & Gurpinar, 2007). Both ANP and AHP derive ratio scale priorities for
elements and clusters of elements by making paired comparisons of elements on a common
property or criterion. ANP disadvantages may arise when the number of decision factors
and respective inter-relationships increase, requiring increasing effort by decision makers.
Saaty and Vargas (2006) suggested the usage of AHP to solve problems of independence
between decision alternatives or criteria, and the usage of ANP to solve problems of
dependence among alternatives or criteria.
Both AHP and ANP share the same drawbacks: (a) with numerous pairwise comparisons,
perfect consistency is difficult to achieve. In fact, some degree of inconsistency can be
expected to exist in almost any set of pairwise comparisons. (b) They can only deal with
definite scales in reality, i.e. decision makers are able to give fixed value judgements to the
relative importance of the pair wise attributes. In fact, decision makers are usually more
confident giving interval judgements rather than fixed value judgements (Kahraman et al,
2010). Furthermore, on some occasions, decision makers may not be able to compare two
attributes at all due to the lack of adequate information. In these cases, a typical AHP/ANP
method will become unsuitable because of the existence of fuzzy or incomplete
comparisons. It is believed that if uncertainty (or fuzziness) of human decision making is not
taken into account, the results can be misleading.
To deal quantitatively with such imprecision or uncertainty, fuzzy set theory is appropriate
(Huang et al, 2009; Kahraman et al, 2010). Fuzzy set theory was designed specifically to
mathematically represent uncertainty and vagueness, and to provide formalised tools for
dealing with the imprecision intrinsic to multi-criteria decision problems (Beskese et al,
2004; Mehrabad & Anvari, 2010). The main benefit of extending crisp analysis methods to
fuzzy technique is in its strength that it can solve real-world problems, which have
imprecision in the variables and parameters measured and processed for the application
(Lee, 2009).
To solve decision problems with uncertainty and vague information where decision makers
cannot give fixed value judgements, whilst also taking advantage of the systematic
weighting system presented by AHP/ANP, many researchers have explored the integration
of AHP/ANP and fuzzy set theory to perform more robust decision analysis. The result is
the emergence of the advanced analytical method - fuzzy AHP/ANP (Huang et al, 2009).
Fuzzy AHP/ANP is considered as an important extension of the conventional AHP/ANP
(Kahraman et al, 2010). A key advantage of the fuzzy AHP/ANP is that it allows decision
makers to flexibly use a large evaluation pool including linguistic terms, fuzzy numbers,
precise numerical values and ranges of numerical values. Hence, it provides the capability
of taking care of more comprehensive evaluations to provide more effective decision
support (Bozbura et al, 2007). Details of the key features, strengths and weaknesses of
different MCDA methods are compared in Table 1.
2.3 Gap between GOM decision requirements and existing research

Separately, both LCA and MCDA are popular analysis technologies for decision making and
sustainable development. The reason why LCA stands out from other eco-efficiency
technologies such as Environmental Accounting, Value Analysis and Eco-indicators, is in its
capability of highlighting environmental issues from a holistic (“cradle to grave”)
perspective. By breaking down the environmental problems into specific issues at different
life cycle stages that can be articulated by operations managers, it helps decision makers to
explicitly capture, code and implement corresponding environmental objectives in their
decision making process. MCDA’s main merit is in its competence in handling complex
decision situations by incorporating multiple decision criteria to resolve conflicting interests
and preferences.
Analysis Selected
Key elements Strengths Weaknesses
methods references
Can handle situations Relationships between
Multi-criteria and in which decision decision factors are not
multi-attributes maker’s subjective considered;
(Saaty, 2005;
hierarchy; judgements constitute inconsistency of the
Anderson et al,
Pair wise comparison; a key part of the pairwise judgements;
AHP 2009)
graphical decision making cannot deal with
representation. process uncertainty and
vagueness
Inconsistency of the
pairwise judgements;
Allows interaction and
Control network with cannot handle situations (Saaty & Vergas,
feedback between
sub-networks of where decision makers 2006; Dou &
ANP different decision
influence can only give interval Sarkis, 2010)
factors
value judgements or
cannot give values at all
Mathematical
representation; handle
Can solve real-world
uncertainty, vagueness (Beskese et al,
decision problems Lack of a systematic
Fuzzy set and imprecision; 2004; Mehrabad &
with imprecision weighting system
theory grouping data with Anvari, 2010)
variables
loosely defined
boundaries.
Fuzzy membership
Fuzzy Combined strengths of
functions together with Time consuming; (Kahraman et al,
AHP/ fuzzy set theory and
priority weights of complexity. 2010)
ANP AHP/ANP
attributes
Table 1. Comparison between different multi-criteria decision analysis methods
GOM decisions need to address the 3BL, which undoubtedly require MCDA methods. In
the meantime, it is critical that environmental and social concerns be addressed right from
the early stage of product and operational life cycles, so that the adverse impact can be
minimised or mitigated. Therefore, GOM decision making requires MCDA and LCA to be
explored in an integrated rather than isolated manner. By considering both LCA and
MCDA technologies together, it could provide decision makers with the vital analysis
tools that enable systematic evaluation for improved decision making capabilities. It
therefore could allow operations managers to take concerted decisions (and actions as the
implementation of the decisions), not only to limit, but also to reverse any long term
environmental damage, and thus ensuring that operations activities are undertaken in a
sustainable manner.
Despite the urgent requirements from GOM for powerful analysis support, there is little
report in the literature discussing the successful integration of both LCA and MCDA
technologies in support of GOM decision making. Next Section of this paper proposes an
integrated approach to fill the gap.
3. An integrated sustainability analysis approach

Based on the understanding of the strengths and limitations of different MCDA and LCA
methods, this paper proposes an integrated sustainability analysis approach for GOM
decision support. The rationality behind the integration approach is that, through
integration, LCA will enhance MCDA with the product and operational life cycle stage
information so that sustainable operations decision can be made from a more holistic view
(through life view), MCDA will enhance LCA with multi-criteria and decision situation
information to help pin down stage-specific decision variables and correlations to decision
goals and alternatives. The proposed approach comprises two key elements: ANP and OLC
analysis. The ANP analysis allows decision makers to address the complexity of decisions
situations in sustainable operations. OLC analysis allows decision makers to address the
environmental issues across different stages of product’s operational life cycle stages.
Seamless integration of the ANP and OLC analysis is enabled through immediate sharing of
consistent information about the GOM decision context and content.
3.1 Performing OLC analysis to understand decision problems from life cycle
perspective
During the OLC process – procurement, production, distribution, use, end-of-life treatment
and reverse chain, different green issues need to be addressed at different stages. Therefore
environmental objectives may be defined in different forms for GOM decision making. For
example, greener material selection at procurement stage, cutting down greenhouse gas
emission at production stage, or reducing energy consumption at the use and distribution
stages, and safe waste management for end-of-life treatment, and product recovery through
reverse logistics. Figure 3 illustrates more comprehensive environmental objectives used at
different OLC stages for green operations decision making.
Minimise use of Process design Reverse plan End-of-Life

Minimise energy
Decision materials -reduce waste -collecting product collection
& other resource
objectives -minimise pollution packaging & & treatment;
consumption;
at different Recycled materials -use min. resource used products hazardous waste
consumer
OLC stage collected safely;
behaviour
Renewable Product design
Reverse plan
resources -modularity
-reuse, recycle
-disassembly
Responsible -recondition
-recycleability
suppliers -remanufacture
Reverse chain Environment

Suppliers
Procurement Production End-of-life

Distribution Use
treatment
All
Stakeholders
Forward chain
Fig. 3. OLC analysis for GOM

In sustainability analysis, decision objectives can be instantiated by using appropriate
indicators (OECD, 1991; Bell & Morse, 1999). An indicator expresses one or more
characteristics that can be empirically observed or calculated. An indicator aims at

identifying those aspects of phenomenon that are considered to be important as for
monitoring and control. Therefore, it is a piece of information that refers to an intrinsic
attribute or to a set of attributes of the phenomenon or associated to other phenomena
closely related to the former. In GOM, indicators are usually described with reference to the
three principal sustainability dimensions. For example, environmental indicators include
greenhouse gas emissions, quantity of wastes, etc. Social indicators can be unemployment
rate and crime level etc. While economic indicators include GDP, inflation rate and so on.
Researchers have recognised that it is the system of indicators rather than individual
indicators that is more significant for GOM (Bottero et al, 2007). Although made up of
individual indicators, the system of indicators can describe and give inter-correlated
information from a logical and functional view. The proposed integrated sustainability
analysis approach in this paper explores a system of indicators.
3.2 Development of decision network models with ANP

In order to address the decision making challenges for GOM from an holistic perspective, it
is extremely important:
1. To identify the relationships between the key components in a sustainable operations
system: operations strategy, operations structural and infrastructural decisions,
environmental issues and social issues. The relationships should be based on the
understanding of the contents of each component. For example, how environmental
issues such as waste management, reduce-reuse-recycle, and pollution control can be
addressed by operations strategies and its structural and infrastructural decisions.
Similarly, how social issues such as staff and customer safety, employment policy,
workplace stress, price manipulation, honesty and transparency of supplier
relationships can be addressed by the operations strategies and decisions;
2. To define operations decision hierarchy/network, i.e. the dependency between
operations strategic decisions, structural decisions and infrastructural decisions. Within
the decision network, if decision on one network node changes, what are the decision
propagation paths along the network and consequences to other decisions? What needs
to be done to manage the decision changes?
To address the above issues and to make sure multiple criteria including environmental and
social objectives from the OLC analysis (Section 3.1) are integrated into the decision making
process, Analytic Network Hierarchy (ANP) technology has been explored. The result of the
process is a GOM analytical model consisting of a control hierarchy, clusters and elements,
as well as interactions between clusters and elements. Six key steps have been undertaken to
develop the GOM analytical model.
Step 1. specify decision goal based on the OLC analysis of the GOM decision problem, and
define decision criteria clusters, sub-criteria and detail criteria.
Step 2. design alternatives for the specific GOM decision problem. It is advised that
adopting alternatives from “good practices” in the filed and using preliminary
elimination increases the quality of decision alternatives.
Step 3. determine the network structure and interactions. The output of this step will be a
control hierarchy, as shown in Figure 4. It should be noted that in the Figure the
influence between the elements are bi-directional, which means that the importance
of the criteria influences the importance of the alternatives, and vice versa.
1. Conventional operations 2. Environmentally friendly

performance criteria operations
Quality Speed Pollution Waste

control minimisation
Cost
Flexibility
Reduce Reuse recycle
Dependability
3. Socially responsible
4. Alternatives
operations
Community Alternative 1
Customer
benef its
saf ety
Alternative 2
Honesty & Employment

Alternative N
transparency policy
Fig. 4. The GOM ANP network control hierarchy

Step 4. create a super-matrix based on the network control model from step 3. The super-
matrix is composed of ratio scale priority vectors derived from pair-wise
comparison matrices. The super-matrix structure is shown as follows:
1 2 3 4
1 w11 w12 w13 w14
2 w21 w22 w23 w24

W=
3 w31 w32 w33 w34
4
w41 w42 w43 0
In the above equation, subscript number 1 shows the criteria cluster belonging to
conventional operations performance objectives; subscript number 2 shows the criteria
cluster belonging to environmental friendly operations; subscript number 3 shows the
criteria cluster belonging to socially responsible operations; subscript number 4 shows the
criteria cluster belonging to alternative operations. In the super-matrix, w11, w12, w13, and so
on represent the sub-matrices. The cluster which has no interactions is shown in the super-
matrix with a value of zero. Those non-zero values mean that there are dependencies
between the clusters. For example, w12 means that cluster 1 depends on cluster 2. Similar to
that in AHP, the 1 – 9 scale system developed by Saaty (Saaty & Vargas, 2006) is used in this
research, and pairwise comparisons are made to create the super-matrix.
Step 5. yield the weighted super-matrix. The un-weighted super-matrix from step 4 must
be stochastic to obtain meaningful limiting results. This step is to transform the un-
weighted into a weighted super-matrix. To do this, firstly the influence of the
clusters on each other is determined, which generates an eigenvector of the
influences. Then the un-weighted super-matrix is multiplied by the priority
weights from the clusters, which yields the weighted super-matrix.
Step 6. stabilise the super-matrix. This step involves multiplying the weighted super-
matrix by itself until the row values convergence to the same value for each column
of the matrix.
By the end of step 6, the limiting priorities of all the alternatives should be computed and
shown in the matrix. The alternative with the highest priority should become transparent
and will become the optimal choice to decision makers.
3.3 The integration of OLC analysis and ANP

In order to provide efficient support for holistic decision making in GOM, decision
requirements need to be met to address multiple criteria across OLC stages. As shown in
Figure 5, information about decision criteria and indicators derived from the OLC analysis
empowers decision makers to target the most important environmental issues from a life
cycle perspective, so that adverse impact of decisions from one stage to another can be
minimised and mitigated. ANP addresses multiple decision criteria by incorporating
decision makers’ preferences and score their judgement within the ANP pair-wise
comparison matrices. The dependence between multiple criteria is taken into account
through the ANP network interactions. By integrating ANP and OLC analysis, efficient
decision support will be provided for holistic decision making to be made to address
multiple decision criteria throughout operational whole life cycle.
Fig. 5. Integration of OLC analysis and ANP

By integrating the information about the interactions among the three main sustainability
dimensions and their cause-effect relations in the OLC analysis and ANP, it allows decision
makers to measure the interconnections and the influences both horizontally (across
multiple decision criteria) and vertically (across life cycle stages), which provides an
integrated approach for the sustainability analysis for holistic decision making in GOM.
Two strategies have been explored for the information integration between OLC analysis
and ANP model development: one is through master data management, and the other is
through meta-data management.
In a GOM decision support system, master data is a very important concept which supports
data integrity and consistence. Master data are persisted, long lived data which need to be
stored centrally. They can be used by all business functional units and at all levels of
organization (Monk & Wagner, 2009). Examples of master data in an operations system are
material master, vendor/supplier master, and customer master records. Additionally,
master data also includes hierarchies of how individual products, customers and accounts
aggregate and form the dimensions which can be analyzed. Master data management is
carried out to ensure that material master, vendor/ supplier master and customer master for
example are consistently stored and maintained, so that all information users, both people
(including decision makers) and computer systems, can access the right information at all
times.
As the demand for business intelligence grows, so do the available data sources, data
warehouses, and business reports that decision makers depend on for business decisions.
While master data management can be used to effectively integrate business data across
business functions, metadata management provides an extra layer of reliability when GOM
decision support systems use multiple data sources. Separate departmental deployments of
business solutions (resulting from being in charge of different OLC stages) have inevitably
created information silos and islands, which makes it significantly more difficult to manage
the information needed to support holistic decision making. This is especially the case
where the data sources change, which will have significant impact on the GOM decisions.
Decision makers will also tend to put various trust in the available information with various
origins. Metadata management can provide powerful support for data traceability and give
decision makers essential assurance of the integrity of information on which their decisions
are based. A generic definition of metadata is “data about data” (Babin & Cheung, 2008). In
green operations management, typical use of metadata has been identified as helping to
provide quick answers to the questions such as: What data do I have? What does it mean?
Where is it? How did it get there? How do I get it? And so on. The answers to these
questions will have a profound impact on the decisions to be made.
4. Application of the integrated approach to case study

This section discusses the application of the proposed integrated sustainability analysis
approach to a case example from Plastics Manufacturing industry.
4.1 The case

Manufacturing industry is, without a doubt, a major contributor to world’s economy (e.g.
GDP growth). At the same time, manufacturing has been in the centre of the root cause for
environmental issues. Along with the wave of business globalisation, more and more social
problems are being unfolded from the manufacturing industry. It is a common
acknowledgement that the quicker to take effective means to tackle the environmental and
social problems caused by manufacturing industry, the better (Nunes & Bennet, 2010). As
one of the fastest developing countries on the planet, China has been branded as the world’s
manufacturing workshop in recent years. Its adverse impact on 3BL sustainability can no
longer be ignored. According to a 1998 report of the World Health Organisation, of the ten
most polluted cities in the world, seven were in China and the situation has not changed
much (Chow, 2010). Therefore, Chinese manufacturing industry can provide perfect cases
for researchers to study the GOM issues. This paper looks at a case from a Chinese Plastics
Manufacturing company.
One of the most influential products from Plastic Manufacturers is plastic bags. Highly
convenient, strong and inexpensive, plastics bags were appealing to business and
consumers as a reliable way to deliver goods from the store to home. However, many issues
associated with the production, use and disposal of plastic bags may not be initially
apparent to most users, but now are recognised extremely important and need to be
addressed urgently. By exploring the integrated OLC and ANP approach with the case
study, this paper aims to help decision makers achieve better understanding of the full
ecological footprint of the products, and to provide efficient decision support in dealing
with the associated negative impacts on environment and social equity.
4.2 Eliciting decision criteria and indicators through OLC

It was recognised that the Plastic Manufacturing company needs to understand plastics bags
life cycle impacts by undertaking streamlined OLC to elicit environmental indicators. The
information can then be used to enlighten operations managers and help them make
informed decisions. From the manufacturer’s viewpoint, planning ahead ensures that any
potential risks to business are anticipated whenever possible. A key benefit is that a
proactive approach is likely to be more scientifically sound than a reactive approach, which
is merely responding to government legislation or consumer concerns.
Figure 6 demonstrates the application of LCA to plastic bags, with energy inputs and
emission output at each key stage of the life cycle. In terms of waste management, three
potential strategies can be implemented: to make recyclable bags, reusable bags and
degradable bags. Specific indicators for environmentally friendly operations obtained from
the OLC process include minimum energy consumption, gas emission, and land and water
pollution. Indicators for socially responsible operations include minimum damage/ threat
to human health, wildlife and tourism etc.
Recycle (recyclable bags)
Gas
Fuel
Electricity
Oil
Reuse (reusable bags)

Energy Energy
Shipping & Use & immediate Waste

Manufacturing
Raw material transportation disposal management
Virgin material
Emission Emission Emission Compost
Discharge Discharge Discharge (degradable bags)
Threat to wildlif e
Pollution to
Pollution to Damage to human health
Air & water
Sulphur land & water
(toxic chemicals, Loss of livestock
CO2) Impact on tourism
Fig. 6. OLC analysis for plastic bags
4.3 Developing the GOM analytical model and results

Plastic grocery bags were first introduced to Chinese supermarkets over 30 years ago.
Today, 80 percent of grocery bags are plastic (FMI, 2008). To address the environmental and
social issues resulting from plastic bags, many crucial decisions that plastic manufacturer
needs to make during the whole life-cycle. Four decision alternatives can be derived from
the OLC analysis:
1. To make recyclable plastic bags. This alternative seems to be taken at rather late stage of
the life cycle, but corresponding considerations are required at early stages such as in
the material selection stage so that recyclable plastic bags can be sorted into proper
categories and processed later.
2. To make reusable plastic bags. This alternative requires appropriate actions to be taken
at early stages of the life cycle. For example, at the material selection and manufacturing
stages, appropriate considerations should be taken so that the reusable bags have the
strengths to be reused for a certain number of times.
3. To make degradable plastic bags, such as those which degrade under micro-organisms,
heat, ultraviolet light, mechanical stress or water.
4. To replace plastic bags with paper bags. For some time, manufacturers were forced to
make a key decision – “plastic or paper“. Research clearly showed that paper shopping
bags make a much larger carbon footprint from production through recycling. For
example, a paper bag requires four times more energy to produce a plastic bag. In the
manufacturing process, paper bags generate 70 percent more air and 50 times more
water pollutants than plastic bags (FMI, 2008).
4.3.1 Development of the network control hierarchy for the Plastic Bags case
The generic ANP models discussed in Section 3 include comprehensive factors of GOM
decision situations, this Section applies the GOM analytical models to a customised situation
of dealing with the Plastic Bags. Therefore, a simplified version of the generic ANP Control
Hierarchy for the case has been developed, as shown in Figure 7. The analytical models
discussed in the case study were developed using Super Decisions® software. As can be seen
from the Figure 7, the GOM network consists of four clusters. Each cluster has one or more
elements that represent the key attributes of the cluster. Elements for Clusters 2, 3 and 4
have derived from the case OLC process (Section 4.2). Connections between the clusters
indicate the influence and dependency. A reflexive relationship on a cluster in the model
(such as the one for Cluster 2) means that there is inter-dependency between the elements in
the same cluster.
4.3.2 Pair-wise comparisons for the Plastic Bags case

In complex decision making using ANP, a series of pair-wise comparison are made to
establish the relative importance of the different clusters and elements with respect to a
certain component of the network, including clusters comparison and elements (of the
clusters) comparison. Both cluster and element comparisons are based on a ratio scoring
system.
In the pair-wise comparison process for the Plastic Bags case, a 9-scale ratio scoring system
suggested by Saaty (2005) has been employed. A judgment or comparison is the numerical
representation of a relationship between two elements that share a common parent. Each
judgement reflects the answers to two questions: which of the two elements is more important
with respect to a higher level criterion, and how strongly, using the 1—9 scale shown in Table 2.
At the level of clusters comparison, there are four 4x4 matrices containing the judgements
made pairwise comparisons established from the survey and discussion with groups of
Fig. 7. GOM network control hierarchy for the Plastic Bags case
Importance Definition Explanation

1 Equal Importance Two activities contribute equally to the objective
Experience and judgment slightly favour one
3 Moderate importance
activity over another
Experience and judgment strongly favour one
5 Strong importance
activity over another
Very strong or demonstrated An activity is favoured very strongly over
7
importance another; its dominance demonstrated in practice
The evidence favouring one activity over another
9 Extreme importance
is of the highest possible order of affirmation
Sometimes one needs to interpolate a
For compromise between the above
2,4,6,8 compromise judgment numerically because there
values
is no good word to describe it.
If activity i has one of the above
nonzero numbers assigned to it A comparison mandated by choosing the smaller
Reciprocals
when compared with activity j, then element as the unit to estimate the larger one as a
of above
j has the reciprocal value when multiple of that unit.
compared with i
If consistency were to be forced by obtaining n
Rationales Ratios arising from the scale
numerical values to span the matrix
When elements are close and nearly
1.1-1.9 For tied activities indistinguishable; moderate is 1.3 and extreme is
1.9.
Table 2. The Fundamental Scale (Saaty, 2005)
experts of sustainability assessment. The pairwise comparison matrices at this level allow
decision makers to evaluate the relationships existing between the different sustainability
aspects, i.e. economic, environmental and social. For example, Table 3 represents the
comparison among the four clusters from the point of view of socially responsible
operations. The priority vectors are calculated and shown in the last column of the Table.
Social Environmental Economic Social Priority

Alternatives
sustainability sustainability sustainability sustainability vector
Alternatives 1 7 5 3 0.740430
Environmental
1/7 1 3 1/3 0.168434
sustainability
Economic
1/5 1/3 1 1/5 0.091136
sustainability
Social
1/3 3 5 1 0
sustainability
Table 3. Pairwise cluster comparison matrix with respect to the social sustainability
When the priority vectors for all four clusters are calculated and aggregated in one table, a
Cluster Matrix is formulated for the Plastic Bags case, as shown in the Figure 8. The Cluster
Matrix is used later to transform super-matrix from unweighted to weighted format.
Fig. 8. Cluster matrix

Once the clusters comparisons are done, it is essential to perform the pairwise comparisons
at more detailed level, i.e. comparisons between elements (of the clusters). The element
comparisons can be done in similar manner as for the cluster comparisons.
4.3.3 Super-matrix formation and global priorities for the Plastic Bags case
The result of all pairwise comparisons is then input for computation to formulate a super-
matrix. In the Plastic Bags case, three different super-matrices have been generated: a Un-
weighted, a Weighted super-matrix and a Limit Super-matrices. Super-matrices are
arranged with the clusters in alphabetical order across the top and down the left side, and
with the elements within each cluster in alphabetical order across the top and down the left
side. An Un-weighted Super-matrix contains the local priorities derived from the pair-wise
comparisons throughout the network. Figure 9 shows part of the Un-weighted matrix for
the Plastic Bags case (because of the space limit, part of the super-matrix is hidden. In the
real software environment, the whole super-matrix can be seen by scrolling the bars on the
interface).
Fig. 9. The Un-weighted Super-matrix for the Plastic Bags case

The Unweighted Supermatrix has to be transformed into the Weighted Supermatrix. The
transformation process involves multiply the Unweighted Supermatrix by the Cluster
Matrix, so that the priorities of the clusters can be taken into account in the decision making
process. The Weighted Supermatrix for the Plastic Bags case in shown in Figure 10.
However, the Weighted Supermatrix is very difficult for decision makers to use because of
the distribution of vector values. This requires conducting a finishing touch in the GOM
analytical model development process by transforming the Weighted Supermatrix into a
Limit Supermatrix. The Limit Supermatrix is obtained by raising the weighted supermatrix
to powers by multiplying itself. When the column of numbers is the same for every column,
the Limit Supermatrix has been reached and multiplication process is halted. The Limit
Supermatrix for the Plastic Bags case is shown is Figure 11. A graphical overview of the
Limit Supermatrix of the case is shown in Figure 12, in which the consequence of all
alternatives is more visualised.
Based on the Limit Super-matrix results and their visual representation shown in the
Figures 11 and 12, recommendation of the decision alternatives can be drawn as follows:
1. Paper bags as a replacement of plastic bags is the least ideal choice, based on evidence
from the super-matrix: a paper bag requires more energy than a plastic bag (in the super-
matrix, alternative 4 Paper bags contributes a lot less to objective 2.1 Minimum energy
consumption); in the manufacturing process, paper bags generates a lot more air and
water pollutions than plastic bags (in the super-matrix, alternative 4 Paper bags
contributes a lot less to objectives 2.2 and 2.4); paper bags also take up more landfill space.
2. Plastic bags recycling is not a preferred choice in terms of achieving economic, the cost,
objective.
3. Making degradable bags is a relatively ideal choice (with an overall value of 0.98 in the
Figure 12).
4. Making reusable bags is the preferred choice because it has the highest overall score
based on the data collected from the company.
Fig. 10. The Weighted Super-matrix for the Plastic Bags case
Fig. 11. The Limit Super-matrix for the Plastic Bags case
Fig. 12. Visualised representations of the global priorities of the four alternatives
5. Discussion and conclusions

The focus of the paper is on an integrated sustainability analysis for the holistic decision
making in GOM. The approach integrates two core elements: an Operational Life Cycle
(OLC) assessment, and Analytic Network Process (ANP) for multi-criteria decision analysis.
At different stages of OLC (procurement, production, distribution, use, and reverse
logistics), GOM has different strategic focus. Understanding the OLC influence on
operations foci, the environmental and social sustainability issues can be better addressed in
the operations decision making process.
The strengths of the integrated approach lies in that information about decision objectives
and indicators derived from the OLC analysis is directly fed into the GOM analytical model
development, which allows decision makers to find the optimal solution to decision
problems to achieve the multiple sustainability criteria. At its highest level, GOM decision
criteria include economic, environmental and social objectives. Within each of the three
areas, more specific criteria have been generated from the OLC analysis. For example,
economic criteria are further broken down to cost reduction, high margin, productivity
improvement, maximum profit etc. Environmental sustainability includes such criteria as
waste minimisation, reduce-reuse-recycle, and pollution control. Social criteria are based on
labour, discrimination, mistreatment, health and safely, working hours, minimum wages
etc. For operations decision making across multi-stages of OLC, there can be a huge number
of decision variables and decision alternatives for each decision problem. Under such
complex decision situations, GOM analytical models allow operations managers to weigh
the importance of each criterion, to rate the satisfaction level of each decision alternative, to
calculate aggregated score for decision choices against criteria, and to predict the
consequence of each alternative (Sarkis, 2003). Therefore managers can confidently and
transparently perform “what-if” and sensitivity analysis for each decision option,
subsequently improve their judgements and make informed decisions. The novelty of the
integrated approach is augmenting the ANP by OLC analysis so that life cycle stage impact
on the green operations decisions is coherently incorporated.
The benefit of the integrated sustainability analysis approach is that it provides a formal,
evidence-based justification for operations decisions that integrates environmental, social
and economic sustainability objectives into operations manager’s proactive decision making
process. Therefore, environmental and social values are not just talked (in words) but also
enacted (in actions).
The evaluation of the integrated sustainability analysis approach has been illustrated
through a decision case from the plastic manufacturing industry. The case study shows that
the approach has great potential in providing scientific evidence to support GOM decision
making under complex situations and with multiple decision criteria.
Limitations of the approach include:
- It is developed for and evaluated in production operations case. Its applicability to
service operations needs further exploration.
- At this stage, the research has not considered feedback of clusters to the decision
support system yet. Further work needs to explore a mechanism and tool to manage the
feedback.
6. Acknowledgement
The work reported in this paper has been undertaken at the University of Plymouth, UK,
and is funded by International Research Networking and Collaboration scheme under grant
number UOP/IRNC/GH102036-102/1 for the project “Green Operations Management -
integration challenges and decision support (also known as GOM project)”
7. References
Allenby, B.R., 2000. Implementing industrial ecology: the AT&T matrix system. Interfaces,
30(3), 42-54.
Anderson, D.R., Sweeney, D.J., Williams, T.A. & Wisniewski, M., 2009. An Introduction to
Management Science: Quantitative Approaches to Decision Making. Cengage Learning
EMEA, London.
Babin, G. & Cheung, W.M., A meta-database supported shell for distributed processing and
system integration, Knowledge-Based Systems 21(2008), 672-680.
Bateman, N., 2005. Sustainability: the elusive element of process improvement. Journal of
Operations Management, 25(3), 261-598276.
Bayazit, O. & Karpak, B., 2007. An analytical network process-based framework for
successful total quality management: an assessment of Turkish manufacturing
industry readiness. International Journal of Production Economics, 105, 79-96.
Bell, S. & Morse, S., 1999. Sustainability indicators: measuring the immeasurable. Earthscan,
London.
Beskese, A., Kahraman, C. & Irani, Z., 2004. Quantification of flexibility in advanced
manufacturing systems using fuzzy concept. International Journal of Production
Economics, 89(1), 45-56.
Bevilacqua, M., Ciarapica, F.E. & Giacchetta, G., 2007. Development of a sustainable product
lifecycle in manufacturing firms: a case study. International Journal of Production
Research, 45(18-19), 4073-4098.
Bottero, M., Mondini, G. & Valle, M., 2007. The use of analytic network process for the
sustainability assessment of an urban transformation project. In proceedings of
International Conference on Whole Life Urban Sustainability and its Assessment.
Glasgow, 2007.
Bozbura, F.T., Beskese, A. & Kahraman, C., 2007. Prioritisation of human capital
measurement indicators using fuzzy AHP. Experts with Applications, 32(4), 1100-
1112.
Carter, C.R. & Rogers, D.S., 2008. A framework of sustainable supply chain management:
moving toward new theory. International Journal of Physical Distribution and Logistics
Management 38(5): 360-387.
Chan, H.K., Yin, S. & Chan, F.T.S., 2010. Implementing just-in-time philosophy to reverse
logistics systems: a review. International Journal of Production Research, 48 (21), 6293–
6313.
Chituc, C.M., Azevedo, A. & Toscano, C., 2009. Collaborative business frameworks
comparison, analysis and selection: an analytic perspective. International Journal of
Production Research, 47 (17), 4855–4885.
Choudhari, S.C., Adil, G.K. & Ananthakumar, U., 2010. Congruence of manufacturing
decision areas in a production system: a research framework. International Journal of
Production Research, 48 (20), 5963–5989.
Chow, G.C., 2010. China’s Environmental Policy: a Critical Survey. CEPS Working Paper
No. 206. April 2010.
Ding, L., Davis, D. & McMahon, C.A., 2009. The integration of lightweight representation
and annotation for collaborative design representation. Research in Engineering
Design, 19(4), 223-234.
Dou, Y. & Sakis, J., 2010. A joint location and outsourcing sustainability analysis for a
strategic off-shoring decision. International Journal of Production Research, 48(15),
567-592.
Espinosa, A., Harnden, R. & Walker, J., 2008. A complexity approach to sustainability –
Stafford Beer revisited. European Journal of Operational Research, 187, 636-651.
FMI, 2008. Plastic Grocery Bags – Challenges and Opportunities. Food Marketing Institute.
September 2008.
Fuller, D.A. & Ottman, J.A., 2004. Moderating unintended pollution: the role of sustainable
product design. Journal of Business Research, 57(11), 1231-1238.
Galasso, F., Merce, C. & Grabot, B., 2009. Decision support framework for supply chain
planning with flexible demand. International Journal of Production Research, 47(2),
455-478.
Gauthier, C., 2005. Measuring corporate social and environmental performance: the
extended life-cycle assessment. Journal of Business Ethics, 59(1-2), 199-206.
Geldermann, J., Treitz, M. & Rentz, O., 2007. Towards sustainable production networks. .
International Journal of Production Research, 45(18-19), 4207-4224.
Gencer, C. & Gurpinar, D., 2007. Analytical network process in supplier selection: a case
study in an electronic firm. Applied Mathematical Modelling, 31, 2475-2486.
Gunendran, A.G. & Young, R.I.M., 2010. Methods for the capture of manufacture best
practice in product lifecycle management. International Journal of Production
Research, 48 (20), 5885–5904.
Hall, J., Vredenburg, H., 2005. Managing the dynamics of stakeholder ambiguity. MIT Sloan
Management Review, 47(1), 11-13.
Hertwich, E., Hammitt, J. & Pease, W., 2000. A theoretical foundation for life-cycle
assessment. Journal of Industrial Ecology 4(1), 13-28.
Hill, M., 2001. Sustainability, greenhouse gas emissions and international operations
management. International Journal of Operations and Production Management 21(12):
1503-1520.
Hodgman, K.L., 2006. The sustainability analysis framework: an effective knowledge

communication tool in a whole of government context. Systems, Cybernetics and
Information, 5(3), 64-69.
Huang, C.T., Yeh, T.M., Lin, W.T. & Lee, B.T., 2009. A fuzzy AHP-based performance
evaluation model for implementing SPC in the Taiwanese LCD industry.
International Journal of Production Research, 48(18), 5163-5183.
Hunkeler, D., Saur, K., Finkbeiner, M., Schmidt, W.-P., Jensen, A.A., Strandorf, H. &
Christiansen, K., 2003. Life Cycle Management. SETAC Publications: Pensacola, FL.
ISO 14040, 1997. Environmental management – life cycle assessment: principles & framework.
Japan Workshop on Industrial Ecology, 1-3 March 1997, Irvine, CA. Washington,
DC: National Academy Press.
Jose, C. & Jabbour, C.J.C., 2010. In the eye of the storm: exploring the introduction of the
environmental issues in the production function in Brazilian companies.
International Journal of Production Research, 48(21), 6315-6339.
Kahraman, C., Beskese, A. & Kaya, I., 2010. Selecting among ERP outsourcing alternatives
using a fuzzy multi-criteria decision making methodology. International Journal of
Production Research, 48(2), 547-566.
Karsak, E.F. & Ahiska, S.S., 2008. Improved common weight MCDM model for technology
selection. International Journal of Production Research, 46(24), 6933-6944.
Kelly, K.L., 1998. A systems approach to identify decisive information for sustainable
development. European Journal of Operational Research, 109, 452-464.
Kim, J., Park, K., Hwang, Y. & Park, I., 2010. Sustainable manufacturing: a case study of the
forklift painting process. International Journal of Production Research, 48(10), 3061-
3078.
Lee, A.H., 2009. A fuzzy AHP evaluation model for buyer-supplier relationship with the
consideration of benefits, performance, costs and risks. International Journal of
Lee, S.Y. & Klassen, R.D., 2008. Drivers and enablers that foster environmental management
capabilities in small- and medium-sized suppliers in supply chains. Production and
Operations Management, 17(6), 573-586.
Liu, S., Duffy, A.H.B., Whitfield, R.I. & Boyle, I.M, 2010. Integration of decision support
systems to improve decision support performance, Knowledge and Information
Systems – an International Journal 22(3), 261-286.
Liu, S. & Young, R.I.M., 2004. Utilizing information and knowledge models to support
global manufacturing co-ordination decisions. International Journal of Computer
Integrated Manufacturing 17(6): 479-492.
Malos, S. & Hall, J., 2007. Integrating sustainable development in the supply chain: the case
of life cycle assessment in oil and gas and agricultural biotechnology. Journal of
Operations Management, 25, 1083-1102.
Matthews, J., Singh, B., Mullineux, G., Medland, A. J (2006). A constraint-based approach to
investigate the 'process flexibility' of food processing equipment. Journal of
Computers and Industrial Engineering. 51(4), 809-820.
Mehrabad, M.S. & Anvari, M., 2010. Provident decision making by considering dynamic and
fuzzy environment for FMS evaluation. International Journal of Production Research,
48(15), 4555-4584.
Mihelcic, J.R., Crittenden, J.C., Small, M.J., Shonnard, D.R., Hokanson, D.R., Zhang, Q.,
Chen, H., Sorby, S.A., James, V.U., Sutherland, J.E. and Schnoor, J.L., 2003.
Sustainability science and engineering: the emergence of a new meta-discipline.
Environmental Science and Technology, 37, 5314-5324.
Miltenburg, J., 2005. Manufacturing strategy – how to formulate and implement a winning plan.
Portland: Productivity Press.
Monk, E. & Wagner, B., 2009. Concepts in Enterprise Resource Planning. Course Technology
Cengage Learning. Boston, USA.
Neto, J.Q.F., Walther, G., Bloemhof, J., van Nunen, J.A.E.E. & Spengler, T., 2010. From
closed-loop to sustainable supply chains: the WEEE case. International Journal of
Noran, O., 2010. A decision support framework for collaborative networks. International
Journal of Production Research, 47(17), 4813-4832.
Nunes, B. & Bennet, D. (2010), Green operations initiatives in the automotive industry: an
environmental reports analysis and benchmarking study. Benchmarking: An
International Journal 17(3): 396-418.
OECD, 1991. Environmental Indicators. OECD, Paris.
Rao, P., 2004. Greening production: a South-East Asian experience. International Journal of
Operations and Production Management, 24(3): 289-320.
Rao, P. & Holt, D., 2005. Do green supply chains lead to competitiveness and economic
performance? International Journal of Operations and Production Management 25(9):
898-916.
Saaty, T.L., 2005. Theory and Applications of the Analytic Network Process: Decision Making With
Benefits, Opportunities, Costs, and Risks. 3 edition. Rws Pubns. ISBN-13: 978-
1888603064.
Saaty, T.L. & Vargas, L.G., 2006. Decision Making with the Analytic Network Process. Springer,
New York.
Sarkis, J., 2003. A strategic decision framework for green supply chain management. Journal
of Clearner Production 11: 397-409.
Slack, N. Chambers, S. & Johnston, R., 2010. Operations Management (6th Edition), FT Prentice
Hall, London.
Svensson, G., 2007. Aspects of sustainable SCM: conceptual framework and empirical
example, Supply Chain Management – An International Journal 12(4): 262-266.
Taylor, A. & Taylor, M., 2009. Operations management research: contemporary themes,
trends and potential future directions. International Journal of Operations and
Production Management 29(12): 1316-1340.
Ulhoi, J.P., 1995.Coporate environmental and resource management: in search of a new
managerial paradigm. European Journal of Operational Research, 80, 2-15.
Verghese, K. & Lewis, H., 2007. Environmental innovation in industrial packaging: a supply
chain approach. International Journal of Production Research, 45(18-19), 4381-4401.
WCED, 1987. Our Common Future. World Commission on Environment and Development.
White, L. & Lee, G.J., 2009. Operational research and sustainable development: tackling the
social dimension. European Journal of Operational Research 193: 683-692.
Wilkinson, A., Hill, M. & Gollan, P., 2001. The sustainability debate. International Journal of
Operations and Production Management, 21(12), 1492-1502.
Part 3
Applications in Agriculture
14
Uncertainty Analysis Using Fuzzy Sets for

Decision Support System
Mohd Najib Mohd Salleh, Nazri Mohd Nawi and Rozaida Ghazali
Universiti Tun Hussein Onn Malaysia, Johor,
Malaysia
1. Introduction
In agricultural domain application, it is becoming increasingly important to preserve
planting material behavior when interact with an environment that is not under its control.
However, the uncertainty always inherent such inaccurate decisions when a present of
incomplete information in sampling data set (Chao, et al., 2005; Latkowski, 2002; Michelini,
et al., 1995). As a result, the proper decision may need to adapt changes in their environment
by adjusting its own behavior. Many different methods for dealing uncertainty have been
developed. This research work proposes incomplete information with fuzzy representation
in objective function for decision modeling. Firstly, we integrate expert knowledge and
planting material data to provide meaningful training data sets. Secondly, fuzzy
representation is used to partition data by taking full advantages of the observed
information to achieve the better performance. Finally, we optimally generalize decision tree
algorithms using decision tree technique to provide simpler and more understandable
models. The output of this intelligent decision system can be highly beneficial to users in
designing effective policies and decision making.
2. Preliminary study
The major problems in decision modeling are missing information of ecological system, such
as weather, fertilizer, land degradation, soil erosion and climate variability during planting
material selection in physiological analysis. This underlying obstacle will return poor results
when the aggregation of all the databases in planting material. If we try to develop a
decision modeling to apply knowledge to user, then we have to integrate historical records
of planting material behavior, expert knowledge’s perception and decision algorithm
learning. From an information or knowledge provider's perspective, the challenge is to gain
an expert user's attention in order to assure that the incomplete information is valued
(Rouse, 2002). One of the more widely applied in representing uncertainty is fuzzy
representation (Michelini et al., 1995; Mendonca et al., 2007). This representation required a
mathematical support to make further treatment for interpretation of missing values as
information, which is compatible with the observed data. We analyze the complexity of
ecological system which have missing values caused by the uncertainty to the user, then
representing the observed data into plausible values and discuss the outcome of an
empirical study for the missing information in induction learning. We have observed the
most commonly used algorithms for decision modeling of missing values (Tokumaru et al.,
2009) and resulted in better understanding and comprehensible rules in planting material
selection.
2.1 The concept of uncertainty

The uncertainty clearly shows the limitations of the currently adopted approach to the
knowledge representation. In particular, it has shown that the result of a measurement
represents incomplete knowledge and this knowledge can be usefully employed only if its
missing value can be somehow estimated and quantified. However, we have to measure the
degree of uncertainty to rank objects for presentation to a user. The approach requires a
suitable mathematical theory for handling incomplete knowledge. The practice refers
mainly to the probability theory to treat missing values. In the normal situation, the
presence of knowledge is recognized, but the exact value is not known, even if it is possible
to locate it within a closed interval. By definition, we always take the same value, although
unknown, within the estimated interval. The problem of how to estimate the missing value
and measure the uncertainty is still open topic in the field, despite much contribution that
have been made in the literature over the years. From information or knowledge provider's
perspective, the challenge is to gain an expert user's attention, which to assure that the
information or knowledge content is valued. The aim of a measurement process is in fact
always to take a decision. Moreover, decisions are sometimes required within the
measurement process itself.
2.2 Related work

This section reviews some approaches representing missing values in information system
when the decision tree is constructed. The development of measurement science has
identified the uncertainty concept to quantify the incomplete information during the
decision making. The measurement of uncertainty is treated in a purely probabilistic way
and has been considered the available mathematical theory capable of handling incomplete
information. In some case, the valuable information is obtained in incomplete decision table.
The possible decomposition of information table can search possible information within the
missing value (Latkowski, 2002). He addressed the degree of similarity between objects on
the attributes can be measured by operator product and the imprecise attribute values are
expressed in a probability distribution. When the results from possible tables are
aggregated, the probabilistic values of decision table can be expressed in a probability
distribution and the extended set of possible members of subset is obtained. The elimination
of irrelevant features sometimes gives a little or no additional information beyond that
subsumed by the remaining features (Koller, et al. 1996). They proposed an efficient
algorithm for feature selection which able to compute an approximation to the optimal
feature selection criterion. (Chao, et al., 2005) provided comprehensive study dealing with
missing values in cost-sensitivity decision trees. He reported that cost-sensitive decision
learning algorithms should utilize only known values, it is desirable to have missing values to
reduce the total cost of test and misclassification and take advantage of missing values for cost
reduction. A conceptual framework of modeling decision making processes with incomplete
knowledge in agriculture is presented (Passam, et al., 2003). He compared with existing
methodologies and claimed the result to be more realistic, since he took into account the
available information without enforcing an artificial accuracy. While Schafer (1997) introduced
Uncertainty Analysis Using Fuzzy Sets for Decision Support System 275
the concept of different levels of suitability for learner biases, the fact that no algorithm biases
can be suitable for every target concept, the idea that there is no universally better algorithm is
fast maturing on the machine learning community. It might do better to map different
algorithms to different groups of problems with practical importance.
Often cases with incomplete descriptions are encountered in which one or more of the
values are missing. This may occur, for example, because these values could not be
measured or because they were believed to be irrelevant during the data collection stage.
Considerable attention and significant progress has been devoted in the last decade toward
acquiring “classification knowledge”, and various methods for automatically inducing
classifiers from data are available. Quinlan (1996) had focused several approaches of
measurement in high levels of continuous values with small cases on a collection of data
sets. A variety of strategies are used in such domain and returned poor result as some
imprecise data is ignored.
The construction of optimal decision trees has been proven to be NP-complete, under
several aspects of optimality and even for simple concepts (Murthy, et al., 1993). Current
inductive learning algorithms use variants of impurity functions like information gain,
gain ratio, gini-index, distance measure to guide the search. Fayyad (Fayyad, 1994)
discussed several deficiencies of impurity measures. He pointed out that impurity measures
are insensitive to inter-class separation and intra-class fragmentation, as well as insensitive
to permutations of the class probability distribution. Furthermore, several authors have
provided evidence that the presence of irrelevant attributes can mislead the impurity
functions towards producing bigger, less comprehensible, more error-prone classifiers.
There is an active debate on whether less greedy heuristics can improve the quality of the
produced trees. Others showed that greedy algorithms can be made to perform arbitrarily
worse than the optimal. On the other hand, Murthy and Salzberg (1993) found that one-level
look-ahead yield larger, less accurate trees on many tasks. (Quinlan, et al., 1995) reported
similar findings and hypothesized that look-ahead can fit the training data but have poor
predictive accuracy. The problem of searching for the optimal sub-tree can be formulated.
The coefficient matrix of the defining constraints satisfies the totally uni-modular property
by solving the integer program (Zhang et al., 2005). He provided a new optimality proof of
this efficient procedure in building and pruning phase. The cost is imposed on the number
of nodes that a tree has. By increasing the unit node cost, a sequence of sub-trees with the
minimal total cost of misclassification cost and node cost can be generated. A separate test
set of data is then used to select from these candidates the best sub-tree with the minimal
misclassification error out of the original overly grown tree. Besides cost-complexity
pruning, a number of other pruning algorithms have been invented during the past decade
such as reduced error pruning, pessimistic error pruning, minimum error pruning, critical
value pruning, etc. It is clear that the aim of the pruning is to find the best sub-tree of the
initially grown tree with the minimum error for the test set. However, the number of sub-
trees of a tree is exponential in the number of its nodes and it is impractical computationally
to search all the sub-trees. The main idea of the cost-complexity pruning is to limit the
search space by introducing selection criterion on the number of nodes. While it is true that
the tree size decides the variance-bias trade-off of the problem, it is questionable to apply an
identical punishment weight on all the nodes. As the importance of different nodes may not
be identical, it is foreseeable that the optimal sub-tree may not be included in the cost-
complexity sub-tree candidates and hence it will not be selected. An interesting question to
ask is whether there is an alternative good algorithm to identify the true optimal sub-tree.
Most of the data mining application reduced to search for the intervals after discretisation
phase. All values that lie within this interval are then mapped to the same value. It is
necessary to ensure that the rules induced are not too specific. Several experiments have
shown that the quality of the decision tree is heavily affected by the maximal number of
discretisation intervals chosen. Recent development of estimation uncertainty using fuzzy
set theory is introduced. The fuzzy practicable interval is used as the estimation parameter
for the uncertainty of measured values. When the distribution of measured values is
unknown, results that are very near to the true values can be obtained in this method. In
particular, the boundaries among classes are not always clearly defined; there are usually
uncertainties in diagnoses based on data. Such uncertainties make the prediction be more
difficult than noise-free data. To avoid such problems, the idea of fuzzy classification is
proposed. The new model of classification trees which integrates the fuzzy classifiers with
decision trees is introduced (Chiang, et.al., 2002 ). The algorithm can work well in
classifying the data with noise. Instead of determining a single class for any given instance,
fuzzy classification predicts the degree of possibility for every class. Classes are considered
vague classes if there is more than one value for the decision attribute. The vague nature of
human perception, which allows the same object to be classified into different classes with
different degrees, is utilized for building fuzzy decision trees (Yuan et al., 1995). The
extension of earlier work of decision tree C4.5 by Quinlan (1996), each path from root to leaf
of a fuzzy decision tree is converted into a rule with a single conclusion. A new method of
fuzzy decision trees called soft decision trees (Olaru et.al., 2003) is presented. This method
combines tree growing and pruning, to determine the structure of the soft decision tree,
with refitting and back-fitting to improve its generalization capabilities. A comparative
study shows that the soft decision trees produced by this method are significantly more
accurate than standard decision trees. Moreover, a global model variance study shows a
much lower variance for soft decision trees than for standard trees as a direct cause of the
improved accuracy. Fuzzy reasoning process allows two or more rules or multi branches
with various certainty degrees to be simultaneously validated with gradual certainty and
the end result will be the outcome of combining several results. Yuan (1995) proposed a
novel criterion based on the measurement of cognitive uncertainty and criterion based on
fuzzy mutual entropy in possibility domain. In these approaches, the continuous attributes
are needed to be partitioned into several fuzzy sets prior to the tree induction, heuristically
based on expert experiences and the data characteristics. The effectiveness of the proposed
soft discretization method has been verified in an industrial application and results showed
that, comparing to the classical decision tree, higher classification accuracy was obtained in
testing. The soft discretization based on fuzzy set theory one inherent disadvantage in these
methods is that the use of sharp (crisp) cut points makes the induced decision trees sensitive
to noise. As opposed to a classical decision tree, the soft discretization based decision tree
associates a set of possibilities to several or all classes for an unknown object. As a result,
even if uncertainties existed in the object, the decision tree would not give a completely
wrong result, but a set of possibility values. Experimental results showed that, by using soft
discretization, better classification accuracy has been obtained in both training and testing
than classical decision tree, which suggest that the robustness of decision trees could be
improved by means of soft discretization.
Basically, the true value is unknown and it could be estimated by an approximation.
Sometimes, we found some of the attributes may be form the similar classes with the
presence of missing values. Rough Sets are efficient and useful tools in the field of
knowledge discovery to generate discriminant and characteristic rules. This method
provides relative reduct that contains enough information to discern objects in one class
from all the other classes. From the relative reduct produced by the quick algorithm, rules
are formed. If a new object is introduced into the data set with the decision value missing,
one could attempt to determine this value by using the previously generated rules.
Although rough set have been used for classification and concept learning tasks (De Jong et
al, 1991; Janikow, 1993; Congdon, 1995), there is rather little work on their utility as a tool to
evolve decision trees. (Jenhani, 2005) proposed an algorithm to generate a decision tree
under uncertainty within the belief function framework.
3. Method and material

There are 3300 records of planting material had been collected in physiological analysis.
These records are represented as a table of examples which described by a fixed number of
features along with a predicted label denoting its class. In oil palm industry, the Elite palms
with special characteristic, such as high bunch oil content, slow stem growth and short
leaves were identified and selected from backcross and progeny test compact seeds.
Effective breeding and selection requires a large genetic variation such as current oil palm
breeding populations. Apart from targeting the primary objectives in oil palm breeding,
several traits of interest include improvement of physiological traits (such as bunch index
and vegetation measurement). Thus, it has become apparent to develop not only high
yielding planting materials but also with novel traits. In this research study, we attempt to
mine the oil palm germplasm for yield and novel traits for use in breeding and ultimately
commercialization.
Total Economic Production (TEP) composites of oil yield and kernel yield per palm per year.
Oil yield is derived from a composite of characters and is dependent on a number of
components, the most important being fresh fruit bunch (FFB) and fruit composition, such
as mesocarp to fruit (M/F), shell to fruit (S/F) and oil to bunch (O/B). The oil palm fruit is a
drupe, consisting the mesocarp, shell and kernel. Within the fruit, a major component that
determines a high O/B, besides M/F, is S/F. Shell and mesocarp contents of the fruit is
negatively correlated, a reduction in shell content subsequently increases the mesocarp
content, while the kernel remains unchanged. Among others, a strategy in developing
planting materials for high oil yield is through the selection against shell thickness. Fruit
size is indicated by the mean fruit weight (MFW) in bunch analysis. Generally, MFW is
positively correlated with the oil-related-traits such as mesocarp to fruit (M/F), oil to dry
mesocarp (O/DM), oil to wet mesocarp (O/WM), oil to bunch (O/B) and oil yield (OY),
and negatively related with kernel yields and F/B. The O/B is the product of oil to bunch is
strongly associated with F/B, M/F and O/WM. Beside that, we incorporate vegetation
measurements which provide the height, trunk diameter, frond production and leaf area per
every palm tree.
In decision modeling analysis decision tree structure refers to all the unique values which
some records may contain incomplete information with discrete set of values or continuous
attributes. Figure 1 shows the comparison of complete and incomplete information of
physiological analysis during the initial study.
Significance Attributes involved in factor

Contribution
3
% of Significance
2.5
2 Overall
1.5 Active
1 Missing
0.5
0
d
t
o u rea
ia
nt
es s
p
ss
le ht
el
TF
ui
ro
es
ar
D
g
te
rn
Fr
ro
O
P
A
ei
oc
en
on
ke
To
H
un
nd
af
ip
C
Le
Tr
ll
io
il
he
O
et
Fr
S
P
Attributes
Fig.1. Number of complete and missing attributes in physiological analysis

By integration of expert knowledge and planting material, it seem reasonable to calculate
the number of samples in the set Tj which belong to the same predicted class Cj., assigned as
(1), then the probability of a random sample of the set T belongs to the class Cj . We estimate
an attribute xi provides data in fuzzy set and gives a membership degree as in (2), where p is
the number of specified clusters, k is the number of data points, xi is the i-th data points,
i(xi) is a function that returns the membership of xi in the j-th cluster
n
freq(C i , T )    (C j ). (1)
j 1
p
  j (x j )  1 (2)
j 1
Some data sets can not be separated with clearly defined boundaries due to the data bases
which contain incomplete information, but it is possible to sum the weight of data items
belong to the nearest predicted class.
3.1 Fuzzy representation

Since fuzzy set theory can be used to describe imprecise information as membership degree
in fuzzy clustering, therefore a given item sets are divided into a set of clusters based on
similarity. We present fuzzy C-means based on objective function to quantify the goodness
of cluster models that comprise prototypes and data partition. Fuzzy cluster analysis
assigned membership degrees to deal with data that belong to more than one cluster at the
same time; they determine an optimal classification by minimizing an objective function. A

standard minimized objective function simply expressed as
p 2
n
l    (  k ( xi ))m xi  c k (3)
k 1 i 1
It can be seen that the characteristic functions constitute between ambiguity and certainty to
the closed subsets xi of the boundary with respect to the set interval (Heiko et al., 2004). We
use more robust method of viewing the compatibility measurement for query xi takes the
weighted average of the predicate truth values. The equation show how the weighted
average compatibility index is computed. The average membership approach computes the
mean of the entire membership values.
n n
xi  (  ( i ( pi ))  wi ) /  wi (4)
i 1 i 1
The degrees of membership to which an item sets belongs to the different clusters are
computed from the distances of the data point to the cluster centers. With combination
expert knowledge and most relevant ecological information, the membership degree can be
calculated to determine some plausible data point lies to center of the nearest cluster. An
iterative algorithm is used to solve the classification problem in objective function based
clustering: since the objective function cannot be minimized directly, the nearest cluster and
the membership degrees are alternately optimized. In fuzzy clustering analysis, the
calculation of cluster centre value is given as follow, and each of observed data is assigned
to the centre using Euclidean distance.
c j  (  (  j ( xi ))m xi ) /(  (  j ( xi ))m (5)
3.2 Decision tree construction

We develop decision tree with expert knowledge and additional information in Figure. 2(a)
and 2(b), the generated nodes form dynamic structure that changes when all elements
remain possible to be tested. All available information is applied to derive the fuzzy
maximum estimation which describes the imputation of estimates for missing values in
cluster centers. In decision tree (Quinlan, 1993) the attribute depends on its entropy
computation among the rest of the attributes, it can be simply formulated the uncertain
subset xi from the interval of element xo are dropped out. The entropy can be measured as
the item sets are divided into subset Tj
n freq(C i , Tj ) freq(C i , Tj )
 Tj
log 2
Tj
. (6)
j 1
In the measured value xi, the possible subset is uniquely determined by the characteristic
function. The measurement of the dispersal range of value xi is relative to the true value
xo.
Fig. 2(a). Decision Tree Construction with incomplete information
Fig. 2(b). Decision Tree Construction with expert knowledge and additional information
3.3 Proposed method

During the decision tree construction, the problem of missing values has to be dealt with
both when processing the training examples and classifying a case using a learned tree. In
some cases, an attribute may have associated missing values which are observed in the
training examples and would appear as multiple values or null. When the decision tree is
used to classify unseen cases, it is difficult to encounter a test on an attribute whose value is
unknown. We present a reactive method for generalizing the decision tree into combination
of binary and ternary tree, so called reactive decision tree by removing one or more
inconsistence conditions of the attribute values. The method includes setting a preference
ordering for conflicting attribute value and forming consistent sub trees or the exact tree
leaf, which represents the predicted class at that leaf.
For each attribute, disjoint classes are separated with clearly defined boundary. These
boundaries are critical since small changes close to these points will probably cause
complete changes in classification. However, the imprecise and missing information often
regarded as a difficult problem to define clearly about the class boundaries. This causes
several problems for the selection of a test to partition the training set which require
comparison of tests based on attributes with different numbers of unknown values. The
comparison of the test which had been selected is sometimes in a sensible manner. The
existence of imprecise information in real-world problems, the class boundaries may not be
defined clearly. To obtain lower and upper approximations, it is necessary to briefly
evaluate the qualitative attribute values, a set of evaluation grades may first be supplied
from existing possible values. It provides a complete set of distinct standards for assessing
qualitative attributes. In accomplishing this objective, an important aspect to analyze is the
level of discrimination among different counting of evaluation grades, in other words, the
cardinality of the set used to express the information. The cardinality of the set must be
small enough so as not to impose useless precision on the users and must be rich enough in
order to allow discrimination of the assessments in a limited number of degrees. We are able
to make a decision when we know the exact values of the maximized quantity. However,
the method has difficulties for knowledge discovery at the level of a set of possible values,
although it is suitable for finding knowledge. This is because the number of possible tables
exponentially increases as the number of imprecise attribute value increases. We explore the
relation between optimal feature subset selection and relevance. The motivation for
compound operators is that the feature subsets can be partitioned into strongly relevant,
weakly relevant and irrelevant features (John et al, 1994). The wrapper method (kohavi et
al., 1994) searches for an optimal feature subset tailored to a particular algorithm and a
domain. This approach showed the feature subset selection is done using the induction
algorithm as a black box, which is no knowledge of the algorithm is needed during the
interface. We compare the wrapper approach to induction without feature subset selection
and to Relief (Kira et al., 1992), a filter approach to feature subset selection, and FOCUS
(Dietterich, et al., 1996). The improvement in accuracy also is achieved for some data sets in
Naive-Bayes algorithm. The Maximum Acceptable Error (MAE) provides an improved
estimate of the confidence bounds of concentration estimates. This method accommodates
even strongly nonlinear curve models to obtain the confidence bounds. The method
describes how to define and calculate the minimum and maximum acceptable
concentrations of dose-response curves by locating the concentrations where the size of the
error, defined in terms of the size of the concentration confidence interval, exceeds the
threshold of acceptability determined for the application.
3.4 Decision rule construction

We investigate the possible of rules generated or classifiers during the classification rule
development. As we had known that one of the most simple to understand, readable and
manipulate is the decision tree. Decision trees represent a series of IF......THEN type rules
which are linked together and can be used to predict properties for our observations based
upon the values of various features. We describe approximation boundary of the attribute
value process by ambiguity Ax, impossible Ix, necessary values Nx and given information
attributes Wx. Generally, we derived our attribute with the values as (Ax, Ix, Nx, Wx).
Initially, we assume that our domain of the attribute x is Dx = { a , b , c} and the values of
attribute x is { ,  ,  , Dx } .
We argue more precise as the rule being uncertainty to the expert as information when there
are less information provided which represents decision tree with uncertainty. Decision-tree
algorithms suggest a more focused approach to rule extraction. The rule extraction assumes
that the user is interested in only one attribute as class as a consequent of every rule. This
assumption significantly reduces the search space with the problem of finding association
rules between any database attributes. A set of mutually exclusive and exhaustive if-then
(production) rules can be easily extracted from a decision tree.
3.5 Improved algorithm and methodology

In this framework, the formal method is an approach towards the extraction of highly
similar groups of subsets from a collection of elements described by a set of attributes A.
The paradigm occurred within the attributes considered represent binary features with only
two possible values true and false.
 1 xi  A
GA ( X )  
0 x i  A
We build the functions G A ( X ) constitute between ambiguous and certainty to the closed
subsets Xi of the boundary with respect to the set inclusion. The function builds operators
join and meet to provide the least upper and lower approximation in the set boundary
respectively. From set theory viewpoint, the concepts represent complete maximal wide
range of subsets which is described by the elements. It may be reduced to the discovery of
the closed sets to the test rules. The algorithm used to discover the concept set comes from
the set theory. It generates possible subsets in an iterative manner, starting by the most
specific test value. At each step, new subsets are generated as members coupled, where the
subset is the intersection of the intents of the already existing subsets. We define the border
of a partially constructed subset through an element-wise insertion to the root node. The
overall decision tree construction is supported by a structure that once the algorithm has
finished its work contains the entire three-valued decision tree. Firstly, the set in the root
node is a dynamic structure that changes when all elements remain possible to be tested.
When the attribute depends on its entropy computation among the rest of the attributes, it
can be simply formulated the new border always includes the new element whereas all
elements of the old border that are greater than new elements are dropped out.
 1 x p  U p  N p  { v} 
 
G A ( X )  unknown x U p  { v}  Dp 
 
 0 v  xp 
To consider the universal set X and its power set P(X), let K be an arbitrary index set, it can
proved that a function Nec is a necessity function if and only if it satisfies the following
relationship:
Nec(  Ak )  inf Nec( Ak )

kK
kK
While the function Pos is a possibility function if and only if it satisfies the following
relationship:
Pos(  Ak )  sup Nec( Ak )

kK kK
for any family { Ak |k  K } in P(X).

We have analyzed through examples how to aggregate individual elements by considering
the overall contributions to the agreement. We now present the considered algorithm in a
general and precise way.
1. Decision makers V  { v1 ,........, vm } sort the alternatives of X  { x1 ,.......xm } according to

the linguistic categories of   { 1 ,...... p } . Then, we obtain individual weak orders
R1 ........, Rm which rank the alternatives within the fixed set of linguistic categories.
2. Taking into account the scores s1 ,........s p associated with  1 ,.......,  p , a score is assigned
to each alternative for each alternative for every decision maker: Si ( xu ), i  1,...., m ,
u  1,......, n
3. We aggregate the individual opinions by means of collective scores which are defined
as the average of the individual scores:
1 m
S( xu )   Si ( xu )
m i 1
and we rank the alternatives through the collective weak order R:
xu Rxv  S( xu )  S( x v )
4. We calculate the overall contributions to the agreement for all the decision makers:
w1 ,......, wm .
a. if wi  0 for every i  {1,...., m} , then we obtain the new collective scores by :
1 m
S w ( xu )   wi  Si ( xu )
m i 1
and rank the alternatives by means of the collective weak order R w .
x u R w x v  S w ( xu )  S w ( x v )
b. Otherwise, we eliminate those decision makers whose overall contributions to the

agreement are not positive. We now initiate the decision procedure for the
remaining decision makers V   { vi  V |wi  0}
We apply the method of weighted equivalence classes to information tables containing
missing values. We briefly compare the method where uses of indiscernible classes with the
method of weighted equivalence classes. The proposed method is based on possible values
in the data sets, which consist of precise values, are obtained from an information table.
Each possible subset is dealt with by applying rough sets to information tables containing
precise information, and then the results from the possible tables are aggregated. In other
words, the methods that are already established are applied to each possible table.
Therefore, there is no doubt for correctness of the treatment.
The value selection assumes that the algorithm is interested in only one attribute as class as
a consequent of the entropy estimation. By reducing the data, we generate different subset
from selected condition attribute. This assumption significantly reduces the search space
with the problem of analyzing the ambiguity of attribute values. When each attribute has
associated a numerical score with respect to decision attribute, each alternative obtain a
more focused approach to the selected values of subsets. Then, we obtain a distance between
each individual preference and the collective one through the distance among the individual
Decision
Table
Table
Training Testing
data Split data
Table Table
Classify
Fuzzy
Result
Representation
Generate
Rules OK
?
cut
Fig. 3. Framework to determine rough set and induce desirable values to validate a classifier
and collective scoring vectors. We measure the agreement in each subset of possible values,
and a weight is assigned to each decision makers, the overall contribution to the agreement.
Those decision makers whose overall contribution to the agreement is not positive are
expelled and we re-initiate the process with the opinions of the decision makers which is
positively contribute to the agreement. The sequential process is repeated until it determines
a final subset of decision makers where all of them positively contribute to the agreement.
Then, we apply a weighted procedure where the scores each decision makers indirectly
assigns to the alternatives are multiplied by the weight of the corresponding decision maker,
and we obtain the final ranking of the alternatives. As a result, the mutually exclusive
attribute value can be easily extracted from a decision tree in the form of ambiguity
reduction. Consequently, we assign the examples with missing values of the test attribute to
the ‘yes’, ‘no’ and ‘yes-or-no’ outgoing branches of a tree node.
3.6 New decision tree construction

New extended ID3 algorithm based on T2, T3 calculates optimal decision trees up to depth 3
by using two kinds of decision splits on nodes. Discrete splits are used on discrete attributes,
where the nodes have as many branches as there are possible attribute values. Interval splits
are used on continuous attributes where a node has as many branches as there are intervals.
The number of intervals is restricted to be either at most as many as the number of existing
classes plus one, if all the branches of the decision node lead to leaves, or to be most as 2
otherwise. The attribute value “unknown” is treated as a special attribute value. Each
decision node has an additional branch, which takes care of unknown attribute values. In
fact, this way of treating unknown attributes is reported to perform better than that of C4.5.
In the tree-building phase, at each node, all attributes are examined, in order to select one on
which a split will be performed for the node. When the attribute is discrete , the relative
frequencies of all of its possible values are calculated. For continuous attributes, the same
approaches would be inefficient because of the number of possible values and the resulting
low frequencies of them. For that reason, local discretisation is used. Finally, a split is
performed on an attribute if it results in maximum accuracy. Consequently T3 produces a
tree hierarchy, which determines how important is an attribute in the classification process,
in contrast to C4.5 which uses the gain ration. To carry out this local discretisation of a
continuous attribute, its values have to be partitioned into multiple intervals. The set of
intervals that minimizes the classification error is found by a thorough exploration instead of
heuristically applying recursive binary splitting. The search for these intervals is
computationally expensive, so T3 restricts decision trees to three levels tests, where only the
third level employs binary splits of continuous attributes. T3 does not use a pruning technique.
Instead it uses a parameter called Maximum Acceptable error (MAE). MAE is a positive real
number less than 1, used as a stopping criterion during building the tree. T2 was observed to
use a greedy approach when building the tree, thus further splitting at a node would stop only
if the records already classified in this node, belonged to a single class. However, this greedy
approach is not optimal, because minimizing the error in the leaf nodes does not necessarily
result in minimizing the overall error in the whole tree. In fact, it was proved that a strategy
choosing locally optimal splits necessarily produces sub-optimal trees. It should be noted here
that classification error indicates how many instances of a training set have incorrectly
classified, while generalization error indicates how many instances of a testing set have been
incorrectly classified. Furthermore, even minimizing classification error does not always cause
minimization of the generalization error due to over-fitting.
By introducing MAE, we allow the user to specify the level of purity in the leaves and stop
to further building of the tree at a potential node split. We set MAE to have 4 distinct values,
namely 0.0 to 0.3, meaning that splitting at a node stops even if the error in that node is
equal to or below a threshold of 0 to 30 percent respectively. More precisely, building the
tree would stop at a node in two cases. In the first case, building stops when the maximum
depth is reached, i.e. 3 when T3 is used or 2 when T2 is used. In the second case, building
stops at that node only all the records remaining there to be classified to the same class in a
minimum proportion of 70 to 100 percent. We used eight different version of T3 in our
experiments, with different depth and length with MAE set to 0.1 – 0.3.
We associate a score to each value and aggregate each individual values by means of the
average of the individual scores, providing a collective weak order on the set of alternatives.
Then we assign an index to each value which measures their overall contribution to the
alternatives. Taking into account these indices, we weight individual scores and obtain a
new collective ranking of alternatives. Once the exact value is chosen, a branch relative to
each value of the selected attribute will be created. The data are allocated to a node
according to the value of the selected attribute. This node is declared as a leaf when the gain
ratio values of the remaining attributes do after excluding the opinions of those decision
makers whose overall contributions to the agreement are not positive. The new collective
ranking of alternatives provides the final decision. Since overall contribution to the
agreement indices usually is irrational numbers, it is unlikely that the weighted procedure
provides ties among alternatives. The proposed decision procedure penalizes those
individuals that are far from consensus positions, this fact incentives decision maker to
moderate their opinions. Otherwise, they can be excluded or their opinions can be
underestimated. However, it is worth emphasizing that our proposal only requires a single
judgment to each individual about the alternatives. We can generalize our group decision
procedure by considering different aggregation operators for obtaining the collective scores.
Another generalization consists in measuring distances among individual and collective
scoring vectors by means of different metrics.
4. Experimental result
The presence of the incomplete information in ecological system impacts on classification
evaluation of planting material behavior in physiological analysis, since the semantics are no
longer obvious and uncertainty is introduced. In figure 3, the graph shows the observed
data are more likely to satisfy with additional information. The experiment combines
planting material with expert knowledge and ecological information to generate an ROC
curve. It can be seen that, in most of the data sets, the number of data belonging to the
various categories do not exactly match the results of physiological analysis. This is because
some of the planting material that should be classified in same type A has been misclassified
into type B and vice versa.
In this experiment, we examine gradually the selected records of complete, ranked features
and missing records of real physiological traits. The result, in Table 1 shows that the features in
missing values compare to the others complete records. By adding additional features of
ecological information to physiological trait, it shows less correlation between each feature in
missing values and the others. The possible clusters rearrange their structure and rules are
generated mostly come from combination of planting material and ecological information. As
a result, the selected proper fuzzy values in decision tree construction provides less and simple
rules. It removes some irrelevant features during the selection of subset of training data.
Figure 4 shows some of the rule production which obtained from the experiment. From it,
the program obtained two decision tables for each clinician, one consisting of those with an
‘induce’ response and the other with a ‘Don’t induce’ response (these tables are not shown).
Rather than attempt to create a reduct from indistinguishable rules at this point we
Fig. 3. Comparison of most observed attribute in missing values
Table 1. Summary of comparisons of agrees and disagree classification based od decision

tree algorithms
generated all the possible rules from the decision table and then removed those that
contradicted each other from the 'induce' and 'don't induce' set. By this means we hoped to
prevent the production of spurious rules, while still producing a reasonable number of rules
from a small decision table.
The analysis uses the trait set to produce good rules and is examined to see if any, and the
contradicted rules are removed from the final rule - set. This is done by removing all
unknown or missing values set with the same parameters that have equal or worse values
than the original rule. We extracted the relevant parameters for each of these trait under
their physiological analyse. Those traits induced for reasons other than our indications were
removed from the database. We then calculated the ‘actual’ rate of induction for the relevant
indications. The rules obtained above are then applied to the relevant database.
Experimental results have shown that our proposed method produces relatively small sized
and comprehensible trees with high accuracy in generalisation and classification. It
improves the performance of existing classifier in terms of both generalisation accuracy and
particularly classification accuracy. Our proposed also outperforms C4.5 in terms of tree size
and classification accuracy. However, this method’s generalisation accuracy remains lower
than that of C4.5
Fig. 4. The production rules by proposed method
5. Discussion and summary

This study has focused on domains with relatively high levels of unknown values and small
testing set. It can be seen that the proposed method can be used for small samples and
produces better result with expert knowledge and additional information. We have measured
the degree of uncertainty using membership function and the proposed method is near to the
true values of the uncertainty and the relative errors of the estimation are very small.
However, this method still produces fairly large error after pruning with more missing values.
This shows that the proposed method can only be applied to several conditions. The
uncertainty estimation of measured values can be directly obtained using fuzzy representation
and it is no longer necessary to estimate the standard deviation by mean value. For systems
with small samples and unknown distributions, the proposed method is more suitable. When
the training set is partitioned, ignoring cases with unknown values of the tested attribute leads
to very inferior performance. During classification, attempting to determine the most likely
outcome of a test works well in some domains, but poorly in others. Combining all possible
outcomes is more resilient, giving better overall classification accuracy in these domains.
6. Conclusion
A missing value is often described as an uncertainty in the development of a decision
system. It is especially difficult when there is a genuine disagreement between experts in the
field and also complex and unstated relationships between the variables that are used in the
decision. The field of planting materials is particularly difficult to study because of the wide
variation in environmental factors, and the differing population groups of genotypes serve.
This decision support system allows the breeders or experts to make decisions in a way that
is similar to their normal practice, rather than having to declare their knowledge in a
knowledge engineering sense. Feedback from the experts in their domain should also be
collected to refine the system especially in the evaluation of the decision trees themselves.
In this research work, we points out the extension of the splitting criterion in the missing
values together. In addition to this, the splitting criterion also discards the irrelevant predictor
attributes for each interior node. From the experiments reported, the classification error using
our method was still low as compared to the original complete data sets by including
additional attributes. It showed that discovered rules have a strong predictive power in
missing values; this rule was not captured by C4.5 algorithm. On the other hand, the proposed
method with his splitting criteria allows normally undeclared rules to be discovered. An
added advantage here is that every expert has the opportunity to study the rules and make
decisions on planting materials selection. On the other hand, the rough sets technique produce
coarse splitting criterion and are often used for knowledge discovery from databases. It is also
useful in any situation where a decision table can be constructed. It has an advantage in
producing a set of comprehensible rules. The fact that this technique produces a lower and
upper approximation of the true value, it allows a degree of uncertainty to be represented. The
rules that are generated by the decision tree could be applied to a database of ‘standard’
procedures, or one that reflects their planting materials to obtain standard rules or drawing up
guidelines for the development of an expert system in the oil palm industry.
We have focused on measurement of uncertainty in decision modeling. In this study, a special
treatment of uncertainty is presented using fuzzy representation and clustering analysis
approach in constructing the decision model. The uncertainty is not only due to the lack of
precision in measured features, but is often present in the model itself since the available
features may not sufficient to provide a complete model to the system. The result of the study
shows that uncertainty is reduced and several plausible attributes should be considered
during classification process. This formalization allows us to the better understanding and
flexibility for selecting planting material in acceptance of the classification process.
7. Acknowledgement
The author would like to thank Universiti Tun Hussein Onn Malaysia for funding the
research grant.
8. References
Chao, S., Xing Z.,Ling X. & Li, S., (2005). Missing is Useful: Missing Values in Cost-Sensitive
Decision Trees. IEEE transactions on knowledge and data engineering, vol. 17, no. 12.
Chiang I-Jen, Jane Hsu Y.J., (2002). Fuzzy classification trees for data analysis. Fuzzy Sets
and Systems, International Journal in Information Science and Engineering, Fuzzy Sets
and Systems, Volume 130, Issue 1, Pages 87-99
Congdon, C.B., (1995). A comparison of genetic algorithms and other machine learning
systems on a complex classification task from common disease research, The
University of Michigan.
De Jong, & Spears. M., (1991). “ Learning Concept Classification Rules using Genetic
Algorithms”. Proceedings of the Twelfth International Joint Conference on
Artificial Intelligence .
Dietterich, T., Kearns, M., & Mansour, Y. (1996). Applying the weak learning framework to
understand and improve C4.5. 13th International Conference on Machine Learning, pp.
96–104.
Fayyad, U. M. & Irani, K. B. (1993). Multi-interval discretization of continuous valued
attributes for classification learning. 13th International Joint Conference on Artificial
Intelligence, pp. 1022–1027.
Fayyad, U. M. (1994). Branching on attribute values in decision tree generation. 12th National
Conference on Artificial Intelligence, pp. 601–606.
Heiko, T., Christian, D. & Rudolf, K. (2004). Different Approaches to Fuzzy Clustering of
Incomplete Datasets. International Journal of Approximate Reasoning 35 239-249.
I., Jenhani, Z., Elouedi, N. B., Amor, K., Mellouli. Qualitative Inference in Possibilistic
Option Decision Trees. In Proceedings of ECSQARU'2005. pp.944~955
Janikow., Z. (1993). Fuzzy Processing in Decision Tree. In Proceedings of International
Symposium on Artifical Intelligence.
Janikow C.Z, (1996). Exemplar learning in fuzzy decision trees, in Proceedings of FUZZ-IEEE.
Koller, D. and Sahami, M. (1996). Toward Optimal Feature Selection, 13th International
Conference on Machine Learning, pages 284-292.
Latkowski, R. (2002). Incomplete Data Decomposition for Classification. 3rd International
Conference, Malvern, PA, USA.
Mendonca, F., Vieira, M., & Sousa,.M. (2007). Decision Tree search Methods in Fuzzy
Modeling and Classification, International Journal of Approximate Reasoning.
Michelini, R. & Rossi, G. (1995). Measurement uncertainty: a probabilistic theory for
intensive entities Measurement, 15 143–57.
Murthy, S., Kasif, S., Salzberg, S., and Beigel, R. (1993). OC1: Randomized induction of
oblique decision trees. 11th National Conference on Artificial Intelligence, pp. 322–327.
Olaru, C. & Wehenkel, L. (2003). A complete fuzzy decision tree technique. Fuzzy sets and
systems 138, 221 – 254, Elsevier.
Passam., C., Tocatlidou. A., Mahaman, B. D. & Sideridis., A. B. (2003). Methods for decision
making with insufficient knowledge in agriculture. EFITA Conference, 5-9 July.
Debrecen, Hungary.
Quinlan, J. (1993). Introduction to Decision Tree. Morgan Kaufman Publisher.
Quinlan, J. (1996). Improved use of continuous attributes in c4.5. Journal of Artificial
Intelligence Research, 4:77-90.
Rouse, B. (2002). Need to know - Information, Knowledge and Decision Maker. IEEE Transaction
On Systems, Man and Cybernatics-Part C: Applications and Reviews, Vol 32, No 4.
Schafer, J., (1997). Analysis of Incomplete Multivariate Data. Chapman and Hall, London.UK.
Schafer, J., and Graham, J., (2002). Missing data: Our view of the state of the art.
Physchological Methods 7, 147-177.
Tokumaru, M. & Muranaka, N. (2009). Product-Impression Analysis Using Fuzzy C4.5
Decision Tree. Journal of Advanced Computational Intelligence and Intelligent
Informatics, Vol.13 No 6.
Wang, X. & Borgelt. C. (2004). Information Measures in Fuzzy Decision Tree. Fuzzy Systems,
Proceedings and IEEE International Conference.
Yuan, Y. & Shaw., J. (1995). Induction of fuzzy decision trees. Fuzzy Sets System. 69:125–139.
Zhang, Y. & Huang, C. (2005). Decision Tree Pruning via Integer Programming.
15
A Web-Based Decision Support System for

Surface Irrigation Design
José M. Gonçalves, André P. Muga and Luis Santos Pereira
CEER Biosystems Engineering Research Center,
Institute of Agronomy, Technical University of Lisbon
Portugal
1. Introduction
Surface irrigation systems have the largest share in irrigated agriculture all over the World.
The performance of surface irrigation systems highly depends upon the design process,
which is related to the appropriateness and precision of land leveling, field shape and
dimensions, and inflow discharge. Moreover, the irrigation performance also depends on
farmer operative decisions, mainly in relation to land leveling maintenance, timeliness and
time duration of every irrigation event, and water supply uncertainties (Pereira, 1999;
Pereira et al., 2002).
The design procedures of farm surface irrigation drastically changed in recent years. The
classical ones based upon empirical rules (Criddle et al., 1956; Wilke & Smerdon, 1965). A
quasi-rational methodology taking into consideration the main design factors was
developed by the Soil Conservation Service and based upon intensive field observations
(SCS, 1974, 1979). This methodology was widely applied and adopted with optimization
procedures (Reddy & Clyma 1981a, 1981b). It assumes the soil classification in infiltration
families related with soil texture and obtained from infiltrometer observations (Hart et al.,
1980). For furrows design, an empirical advance curve was applied relating inflow
discharge, slope, length, and infiltration. Other classical methods refer to the volume-
balance models using the continuity equation and empirical information (Walker &
Skogerboe, 1987; Yu & Sing, 1989, 1990; Clemmens, 2007). These type of models also apply
to irrigation management (Latimer & Reddel, 1990; Camacho et al., 1997; Mailhol et al.,
2005).
Numerous mathematical computer models for surface irrigation simulation were
developed. They originated a new age of design methods, with increased quality of
procedures because they allow the quantification of the integrated effect of main irrigation
factors (length, discharge, slope, soil roughness, shape, and infiltration) on performance,
thus, enlarging the solution set with higher precision and effectiveness than the traditionally
empirical methods. Strelkoff & Katopodes (1977) first presented an application of zero-
inertia modeling for border irrigation. Further developments were applied to borders,
basins and furrows (Fangemeier & Strelkoff, 1978; Clemmens, 1979; Elliott et al., 1982), and
were followed by furrow surge flow modeling (Oweis & Walker, 1990). The kinematics
wave and the hydrodynamics model for furrows were later adopted (Walker & Humpherys,
1983; Strelkoff & Souza, 1984). Computer models for design of basin irrigation include
BASCAD (Boonstra & Jurriens, 1978) and BASIN (Clemmens et al., 1993), and for border
irrigation include the BORDER model (Strelkoff et al., 1996). The models SRFR (Strelkoff,
1993) and SIRMOD (Walker, 1998) apply to furrows, basin and border irrigation and adopt
various approaches for solving the continuity and momentum equations. Reviews were
recently produced by Pereira et al. (2006) and Strelkoff & Clemmens (2007).
In addition to hydraulics simulation models, surface irrigation design requires the
application of other type of models such as for irrigation scheduling, land leveling,
distribution systems, and cost and environmental analysis. In practice, it is usually difficult
to manage data for an interactive application of these models in design when they are not
integrated with a common database. The decision support systems (DSS) methodology
provides the framework to explore the synergy between mathematical simulation models,
data and user knowledge through its integration aimed to help the decision-maker to solve
complex problems. The DSS methodology makes handling data of various types easier and
effective, and favors the integration of simulation models and their interactive application. It
provides for a decision-maker learning process and it supports a decision process and
related choices through multicriteria analysis. DSS models are often applied to irrigation
planning and policy analysis (Bazzani, 2005; Riesgo & Gómez-Limón, 2006), as well as to
performance assessment and water demand and delivery simulation (Raju & Duckstein,
2002; Rao et al., 2004; Oad et al., 2006; Raju et al., 2006). However, few applications are
developed for irrigation design (McClymont, 1999; Hornbuckle et al., 2005; Gonçalves et al.,
2009; Pedras et al., 2009).
The variety of aspects influencing irrigation performance (Burt et al., 1997; Pereira and
Trout, 1999; Pereira et al., 2002) makes the design process quite complex, and a multicriteria
analysis approach is then advantageous. Alternative design solutions may be ranked
following various objectives and criteria, such as improving the irrigation performance,
achieving water saving, attaining high water productivity, or maximizing farm incomes.
Using a DSS and multicriteria analysis is helpful to produce appropriate comparisons
among alternative design solutions and to perform a trade-off analysis (Roy & Bouyssou,
1993; Pomerol & Romero, 2000). In this line, the DSS SADREG for surface irrigation design
was developed, tested and applied to different agricultural, economical and environmental
conditions, considering several irrigation methods, equipments and practices. Applications
include the Lower Mondego Valley, Portugal, improving basin irrigation for water savings
and salinization control in the Upper Yellow River Basin, China, improving furrow
irrigation in Fergana Valley, Aral Sea Basin, Uzbekistan, and modernizing furrows and
border irrigation in Euphrates Basin, Syria. The objective of this chapter is to present the
DSS model and its Web application, describing procedures to design and users support.
The development of DSS for Web allows a better application flexibility, improving the user
support for database access, and enlarging the number of users, particularly in the world
areas where the water scarcity demand for a better use of irrigation water. The application
version here presented comprises a Web and a simulation engine module, which includes
the models integrated in SADREG. The Web access of DSS allows an easier transfer of
knowledge and tools to improve the procedures to evaluate and design field irrigation
systems. The DSS location on a server allows data sharing and comparison of results by
different users. Some examples of Web applications in the irrigation domain show its
usefulness (Thysen and Detlefsen, 2006; Car et al., 2007).
The organization of this chapter considers, first, the description of DSS model, with details
about the process of design and selection of surface irrigation alternative; after, in sub-
chapter 3, the explanation of Web methodologies in a DSS context, referring the
A Web-Based Decision Support System for Surface Irrigation Design 293
programming languages and the architecture of Web client and server; and, in sub-chapter
4, the description of the DSS software use and Web interface.
2. DSS model
SADREG is a DSS designed to assist designers and managers in the process of design and
planning improvements in farm surface irrigation systems – furrow, basin and border
irrigation. It includes a database, simulation models, user-friendly interfaces and
multicriteria analysis models.
2.1 DSS architecture

SADREG is comprised of two components: design and selection (Fig. 1). The first
component applies database information and produces a set of design alternatives in
agreement with the user options. These alternatives are characterized by various hydraulic,
economic, and environmental indicators that allow appropriate selection and ranking. The
selection component is based upon multicriteria analysis for ranking the alternatives, thus,
supporting the decision-maker to select the best design solution. The decision-maker
participates in all decision processes through interface dialogue structures that allow
expressing design and management options and priorities.
DSS
Design
Data
component
Design User
Interface
models (designer)
Selection
Alternatives
component
Multicriteria User
Interface
models (manager)
ranked
alternatives
Fig. 1. Conceptual structure of SADREG

The database refers to data of various natures, some constituting input data required for
computations, others are created through successive computations (Fig. 2). Main
components are:
1. General field data, relative to soils and environment (e.g., soil infiltration, soil water
holding capacity, soil salinity, groundwater depth, and climate), crops (planting date,
yields, yield function parameters), equipment characteristics and costs, and operational
costs and benefits;
2. Field data characterizing the field under design: length, width, slopes, surveying data,
and characteristics of the respective water supply;
3. Data created through simulations performed with the models SIRMOD for surface
irrigation simulation (Walker, 1998) and ISAREG for irrigation scheduling simulation
(Pereira et al., 2003);
4. Projects data, referring to data characterizing the design alternatives; and
5. Selection data, referring to the value functions and decision priorities.
Input data
for design
Output simulation
data
Field sizes and

ISAREG results Priorities
topographic data
Soils and
SIRMOD results
Environment Projects Value functions
Water supply Land levelling Groups of

alternatives
Crops Design
alternatives
Equipment
Multicriteria
analysis
Economic, costs
and incomes
Fig. 2. Data base components in relation with respective data sources and uses in the design
process
SADREG includes various computational models and tools (Fig. 3). The simulation of the
surface irrigation systems – furrows, borders, and basins – is performed by the surface
irrigation simulation model SIRMOD (Walker, 1998), which is integrated within SADREG.
The water balance simulation to define the application depths and timings is performed by
the ISAREG model (Pereira et al., 2003), which is linked (loose integration) with SADREG
and explored interactively. Calculations relative to land leveling and farm water distribution
systems are performed through specific built-in tools. These computational tools provide for
the characterization of each design alternative, including a complete set of performance
indicators. The resulting data are later handled by an impact analysis tool, so creating all
data required for multicriteria analysis. The impact analysis tool performs calculations
relative to crop yields and related incomes (benefits), costs, and environmental impacts as
described later. Ranking and selection of alternatives are performed with composite
programming and the ELECTRE II models (Roy & Bouyssou, 1993; Pomerol & Romero,
2000).
The SADREG applications scope comprises: (a) a single field analysis relative to alternative
design options for furrow, basin, or border irrigation, considering several decision variables
such as field slopes, farm water distribution systems and runoff reuse; and (b) an irrigation
sector analysis, when a spatially distributed database relative to the farm systems is
available. In this case, the alternatives are assessed jointly with modernization options
relative to the conveyance and distribution network. It applies to farm land parcels or fields,
of rectangular shape, with a well known geographical location, supplied by a hydrant, a
gate, or other facility at its upstream end.
DESIGN SELECTION
Impact analysis
Integrated models Built-in tools
tool
SIRMOD
Land levelling
Surface irrigation
design
simulation Multicriteria analysis models:
Composite programming
and/or ELECTRE II
ISAREG Distribution
water balance systems
Creation of Ranking and selection of

design alternatives design alternatives
Fig. 3. Model base components

A field is characterized by its length and width, slope and land surface unevenness, soil
infiltration and soil water retention characteristics, both assumed as spatially uniform for
design purposes. Large fields may be subdivided into units, which are field fractions
supplied by a single outlet, all having the same width, length, and slope. In addition,
subunits may be considered when the water available or management constraints impose
that a unit is not fully irrigated simultaneously. In order to design the upstream water
distribution, the field characterization requires definition on which direction, OX or OY, the
distribution system should be located. Hydrants or farm gates supply the water to the field
with known discharge and hydraulic head. In large fields, farm canals or pipes supplied by
those hydrants or gates may deliver the water to the upstream distribution system through
various outlets, equaling the number of units. In small fields, generally only one unit is
considered and the outlet coincides with the hydrant or gate.
A hierarchical approach is used to develop the design alternatives (Fig. 4). Data referring to
field characteristics common to all alternatives are organized in a workspace. Included in the
field workspace are the projects whose data structure is aimed at developing a set of design
alternatives relative to:
1. The crop type (e.g., cereals versus row crops),
2. The irrigation method;
3. The land leveling solution, to be defined in agreement with the irrigation method, field
longitudinal and cross slopes, and the selected upstream distribution side,
4. The water supply conditions that influence, together with the irrigation method, the
options relative to the number and size of units and the outlet’s discharge; and
5. Costs and other financial parameters.
Land parcel under design Database
First level of Second level of

Main field characteristics
irrigation options irrigation options
WORKSPACE (field data) PROJECT GROUP OF

ALTERNATIVES
Soil water retention irrigation method
Soil infiltration land levelling Distribution system
Field length & width crop Tail water management
Slopes water supply conditions
Land surface uneveness Number of units & outlets
DESIGN ALTERNATIVE
Costs
Inflow rate per unit
width or per furrow
Number of subunits
Design simulation generating

Performance Modelling
indicators that characterize
attributes tools
each alternative
Multicriteria analysis:
Ranking of alternatives inside Groups
and/or among Groups of alternatives
Fig. 4. Scheme of the creation of field design alternatives using a multilevel approach for
design and application of multicriteria ranking and selection
The design alternatives are clustered into groups included in a project and relative to:
1. The upstream distribution system, which depends upon the selected irrigation method
and equipment available; and
2. The tail end management system, which also depends upon the irrigation method and
the equipment available.
The alternatives constitute complete design solutions. Within a group, they are differentiated
by the operative parameters: the inflow rate per unit width of land being irrigated or per
furrow, and the number of subunits.
2.2 Irrigation system design

2.2.1 Decision variables
The surface irrigation methods considered are level basin irrigation, with flat or furrowed
soil surface, graded basins, borders, and graded furrows (Table 1). Level furrows are treated
as furrowed level basins. For cereals, furrows of corrugated type may be considered.
The decision variables relative to the irrigation design process are described in Table 2. They
depend upon the irrigation method since it influences the field layout and land leveling, the
water supply and distribution, the tail water management, and the farm irrigation
management.
2.2.2 Land leveling

When starting a project, the user must select the irrigation method, the upstream
distribution side, and carry out a land leveling simulation adopting cross and longitudinal
field slopes appropriate to the considered irrigation method and the actual field slopes. The
land leveling simulation tool computes the cut and fill volumes required to change from the
actual elevations za(x, y) into the target elevations:
z(x,y) = z0 + Sx(x-x0) + Sy(y-y0) (1)

where x0, y0, z0 = coordinates of the centre of gravity of the field (m); and Sx and
Sy = longitudinal and cross slopes (m m-1) along the OX and OY axis, respectively (Dedrick
et al., 2007). The plan position (value of z0) is iteratively changed until the cut to fill ratio
becomes > 1.0 and < 1.2 (Fig. 5). Results include the cut and fill depths and volumes, and
related costs.
Soil surface
Irrigation method Field slopes Field inflow conditions Tail end conditions
condition
Point inflow at one or
Flat or
Level basin Zero in all directions
various locations, or to Diked
furrowed
individual furrows
Point inflow at one or Diked for basins
Graded basin and Flat or Longitudinal slope  0
various locations, or to and open for
borders furrowed and cross slope = 0
individual furrows borders
Longitudinal and Inflow to individual
Graded furrows Furrowed Open or diked
cross slope  0 furrows
Table 1. Irrigation methods
2.2.3 Infiltration and SIRMOD application

The modified Kostiakov equation is applied in SIRMOD model (Walker, 1998) to compute
soil infiltration. For continuous flow, it takes the form:
Start
Select irrigation method
Level basins Borders and Graded furrows

graded basins and borders
Select Select cross and

longitudinal slope longitudinal slopes
Field topographic data
Change field Simulate land

slopes levelling
no cut
1< <1.2 ?
fill
yes
Results
Fig. 5. Flowchart of the application of the land leveling simulation tool.

Z  K  a  f0  (2)
where Z = cumulative infiltration (m3 m-1);  = infiltration time (min), K (m3 m-1 min-a) and a
(dimensionless) are empirically adjusted parameters; and f0 = basic infiltration rate
(m3 min-1 m-1).
For surge-flow, the procedure developed by Walker & Humpherys (1983) is adopted:
 For infiltration on dry soil (first wetting), Eq. (2) is applied;
 For infiltration on wetted soil (third and successive wettings) the parameters a, K and f0
in Eq. (2) are modified into Ks, as, and f0s, thus, producing the surge infiltration equation;
and
 For infiltration during the second wetting, a transition curve is applied.
This transition equation balances the effects represented by the equations for dry and wetted
soil (Walker, 1998; Horst et al., 2007):
Z = [K + (K - Ks) FP] [a + (a - as) FP]+ [f0 + (f0 – f0s) FP]  (3)
where FP (dimensionless) = distance-based factor computed from the advance distances xi-2
and xi-1 relative to the surge cycles i-2 and i-1.
To characterize each field, SADREG includes a set of infiltration data concerning families of
infiltration curves for continuous and surge flow, typical of seasonal irrigation events (first,
second, and later irrigations) under flat soil infiltration conditions. Field observations of
infiltration can be added to this set of infiltration curves and be used for the respective
design case study. When no field infiltration data are available, the user selects the curves to
be used considering the available soil data. To adjust the parameters for furrow irrigation,
the procedure proposed by SCS (1979) and Walker (1989) is applied. It is based on the
average wetted perimeter WP (m) and the adjusting coefficient (Cadj). WP is given by
0.425
 qn 
WP  0.265    0.227 (4)
 S 
 0
where q = furrow inflow discharge (l s-1); n = Manning’s roughness coefficient (s m-1/3); and
So = furrow longitudinal slope (m m-1). The adjusting coefficient Cadj is estimated as
WP
C adj  0.5, if  0.5 , else C adj  WP (5)
FS FS
where FS (m) = furrow spacing. The parameters K and f0 [Eq. (2)] and the surge-flow
infiltration parameters are adjusted as
K adj  C adj  K (6)
f0 adj  C adj  f 0 (7)
SIRMOD model is applied for several input conditions that cover all situations relative to
the user options to create alternatives. The continuous input variables consist of: field length
Decision variables
- upstream distribution side (OX or OY)
Field layout and land - field length (FL) and field width (FW)
levelling - cross slope (SoC)
- longitudinal slope (So)
- number of outlets (No)
- number of units (Nu)
Water supply
conditions - total field supply discharge (QF)
- outlet discharge (Qo) and hydraulic head (Ho)
- time duration of field delivery (tF)
Design - continuous and constant inflow rate for basins

and borders
Farm distribution - continuous or surge flow with automatic or
system manual surge valves for graded furrows,
- lined or unlined canal, or gated pipes, or layflat
tubing (for all methods)
- no tail water runoff (case of basins and diked
furrows)
- open tail end without tail water reuse
Tail water management - open tail end with reuse by pumping to the
upstream end
- open tail end with gravity reuse on downstream
fields
- irrigation timing
Management Irrigation scheduling - required application depths
- application time duration (tap)
Distribution system - inflow rate per unit width or per furrow (qo)
operation - number of sub-units (ns)
Table 2. Design and management decision variables

(FL, m), required application depth (Zreq, mm), field longitudinal slope (So), Manning’s
hydraulics roughness coefficient (n), furrow spacing (FS), and inflow rate per unit width or
per furrow (q, l s-1). The discrete input variables required are: inflow regime (continuous or
surge flow), tail end management (diked, open, or open with reuse), cross section shape (flat
or a given furrow type), and parameters of the infiltration equation.
Any run of SIRMOD produces output data including: application, advance and recession
times; infiltration depths; runoff and percolation volumes; and performance indicators,
mainly those defined in Table 3 (Walker & Skogerboe 1987; Pereira & Trout 1999).
Indicator Definition Units
 Zreq
  100 Z lq  Z req
Application efficiency Ea   D %
 Zlq  100 Z lq  Z req
 D
Zlq
Distribution uniformity DU   100 %
Z avg
Water requirement Zavg (root)

Er   100 %
efficiency Zreq
Zavg
Infiltration efficiency IE   100 %
D
Zrun
Tail water runoff ratio TWR   100 %
D
Zdp
Deep percolation ratio DPR   100 %
D
Zreq - the average water depth (mm) required to refill the root zone in the lower quarter of
the field
D - the average water depth applied to the irrigated area (mm)
Zlq - the average depth of water infiltrated in the lower quarter of the field (mm)
Zavg - the average depth of water infiltrated in the whole irrigated area (mm)
Zavg (root) - the average depth of water infiltrated stored in the root zone (mm)
Zrun - the depth of water that runs off at the tail end of the field (mm)
Zdp - the depth of water that percolates below the root zone (mm)
Table 3. Performance indicators

SIRMOD is used iteratively for a given alternative to search the irrigation parameters
referred to above that comply with the application of the required irrigation water depth
Zreq (mm). This one is computed by running interactively the model ISAREG for the
considered crop, soil, and climate. In SIRMOD simulation, an appropriate inflow rate q (qmin
< q < qmax, l s-1 m-1) is iteratively selected using increments q = 0.1 l s-1. However, it may
happen that the model does not converge and it becomes necessary to have a successful
simulation trying different combinations of internal simulation parameters (Walker, 1998):
time step (1-2.5 minutes), space step (1-25 m, according to the field length), time weight
factor (0.60-0.65), space weight factor (0.50-0.65), and type of hydraulics simulation model
(full hydrodynamic or zero inertia). When an inflow rate q is found a new iteration starts to
compute the time tap (min) required to apply the water depth Zreq. This is performed by
maximizing the water requirement efficiency (Table 3), which should become close to 100%.
Results are then saved in the database to be further applied to other alternatives. The results
for intermediate values for each of the continuous input variables are calculated by
interpolation between those stored in the database instead of running the model several
times.
2.2.4 Generation of alternatives

The procedure to build up and characterize design alternatives for a given field is
performed through the following sequential steps:
1. Selection of the irrigation method: level basin (flat or furrowed), graded basins or
border, or graded furrows; for furrows, selection includes the cross section type and the
furrow spacing;
2. Definition of the field side, along the axis OX or OY, where the upstream distribution
facilities will be located;
3. Land leveling simulation, with selection of the field slopes more appropriate to the
irrigation method, and the actual field topography;
4. Selection of the inflow regime for graded furrows: continuous or surge-flow. The surge-
flow irrigation management follows the one-fourth length rule for the advance phase,
i.e., with increased duration of successive cycles; after the advance is completed, a
continuous flow with half-inflow rate applied to the full unit is adopted. For basin and
borders, only a continuous inflow rate is considered;
5. Definition of the number of outlets and field units;
6. Selection of the water distribution system: gated pipe, lay-flat tubing, earth canal, or
lined canal; when the surge-flow option is selected, the respective control system may
be either manual or automatically operated;
7. Selection of a tail water management option for borders and furrows: diked, i.e., closed
at the tail end, or open. When open, the tail end runoff may be reused by pumping to
the upstream end, reused by gravity in downstream fields, or not reused. Basins are
diked; and
8. Estimation of an appropriate crop irrigation scheduling, thus, the required
application depths and timings. With this purpose, the ISAREG model is used
interactively with SIRMOD for searching the application depths appropriate to the
irrigation method.
2.2.5 Impact evaluation

The cost analysis considers the investment and the operation and maintenance (O&M) costs
(Mjeld et al., 1990; Solomon et al., 2007). The investment cost refers to the farm distribution
system equipment and the initial land leveling. The net present value of the investment cost
is calculated on a yearly basis using a capital recovery factor that is a function of the annual
interest rate and the lifetime of the components (Tab. 4). The annual O&M costs include land
leveling maintenance, distribution system operation and maintenance, and reuse pumping
costs. The database shall include the duration times relative to all irrigation tasks depending
upon the irrigation method and the equipment used. The production costs not referring to
irrigation (e.g. fertilization, cultivation, and harvesting) do not differ among alternatives and
do not interfere on ranking and selection of alternatives. Thus, they are not included in the
cost analysis.
Investment costs (IC): Formulas

  N sub  1  j N LT  
ICva - Present worth cost of one given component
IC va  IC   1      
(€)   j  1  1  iTA  
  
iTA  (1  iTA )N AP
CRF - capital recovery factor (-) CRF 
 1  iTA N AP
1
N comp
AFC - Annual fixed cost (€/year) AFC  CRF   IC va j
j 1
ICLL - initial land leveling cost (€) IC LL  a0  a1  tmac

tmac - time machine operation for land leveling (h) tmac  tuexc  Vexc or tmac  tuarea  A
ICD - distribution system cost (€) IC D  b0  b1 · FW
ICR - reuse system cost (€) IC R  r0  r1 · A
ICT - total investment annual cost (€/year) IC T  IC LL  IC D  IC R
OMC - Operation & maintenance costs:
OMCR - Reuse pumping system cost (€/year) OMC R  r2 · VR
OMCLL - Land leveling maintenance cost (€/year) OMC LL  a 0  a 1 · a 2 · A
OMCD - Distribution system cost (€/year) OMC D  h'0  t Mt
tMinst - labour to install the distribution system
t Minst  h 1 · FW
(h/year)
tMrem - labour to remove the distribution system t Mrem  h 2 · FW
(h/year)
tMope – labour to operate a sub-unit (h) t Mope  h 3 · S U W
tMfree - free man power time in a sub-unit irrigation t Mfree  t ap - t Mope
(min)
t MopU  t MopSU · ns, if t Mfree  t Mm
tMopU - total labour to operate a unit (h)
t MopU  t ap · ns, if t Mfree  t Mmin
Nirrig
tMt - total labour by the distribution system (h/year) t Mt  t Minst   ( N u  t MopU )
j 1
OMC T  OMC LL  OMC D  OMC

OMCT – total O&M annual cost (€/year)
Symbols:
Ncomp=Number of a equipment components
A=field area (ha)
Nirrig=annual number of irrigation events
a0 =leveling machines fixed cost
NLT=equipment component life time (years)
(€/operation)
ns=number of subunit per unit;
a1=hourly cost of the land leveling
Nsub=number of component replacements
machines (€/h)
during NAP
a2=unitary maintenance leveling machine
time (h/ha) Nu=number of units per field

b0=distribution system fixed cost (e.g., r0=reuse system fixed cost (€)
valves) (€) r1=unitary reuse system cost (€/ha)
b1=unitary distribution system (dist.syst.) r2=unitary reuse system variable cost (€/m3)
cost (€/m) SUW=subunit width (m);
FL =field length (m) t uexc=unitary machinery for cut volume
FW =field width (m) (h/m3)
h0=hourly cost of the man power (€/h) tap=application time (min)
h1=unitary man power time to install tMmin=additional time required to other
dist.syst. (h/m) tasks relative to irrigation (min);
h2=unitary man power time to remove tuarea=unitary machinery time to land
dist.syst. (h/m) smoothing (h/ha)
h3=unitary man power time to operate Vexc=land leveling excavation volume (m3)
dist.syst. (h/m) VR=pumping reuse volume (m3/year)
IC= investment costs
iTA=annual interest rate (decimal)
NAP=analysis period (years)
Table 4. Cost analysis equations

A water-yield function is used to estimate the crop yield from the computed total water use
during the irrigation season. A quadratic function (Fig. 6) is adopted to calculate the relative
yield (Y/Yopt) from the relative water application (W/Wopt) (Solomon, 1984):
Y/Yopt  k 0  k 1 · (W/Wopt )  k 2 · (W/Wopt )2  k 3 · (W/Wopt )3  k 4 · (W/Wopt )4

(8)
where W = actual water available for crop use during the irrigation season (mm);
Wopt = seasonal water required for achieving maximum crop yield (mm); Y and Yopt = crop
yields (kg ha-1) corresponding to W and Wopt, respectively; and ki (i = 1, ..., 4) are empirical
coefficients typical for the crop, and environmental and agronomical conditions under
consideration. The decreasing branch of this function is related with soil drainage conditions
and excess water impacts on yields. To consider these effects, the user should adjust to the
local conditions the descending branch of the quadratic function, as indicated in Fig. 6
where 3 types of descending branches are presented.
The environmental attributes considered for selection of the design alternatives are:
1. The total irrigation water use during the crop season;
2. The nonreused runoff volume, that represents a nonconsumed and nonbeneficial
fraction of water use and is an indicator for potential degradation of surface waters;
3. The deep percolation volume, which represents also a nonconsumed and nonbeneficial
fraction of water use and is an indicator for potential degradation of ground waters;
4. The potential land leveling impacts on the soil quality; and
5. The potential for soil erosion.
The total water use, runoff and deep percolation volumes are calculated with SIRMOD for
every single event of the irrigation season. The seasonal values are obtained by summing up
these results. The attribute relative to land leveling impacts is expressed by the average cut
depth because the smaller are the cut depths, the lower are the impacts on soil quality.
0,8
relative yield (Y/Yopt) 0,6
0,4
0,2
0
0 0,5 1 1,5 2 2,5 3 3,5 4
Relative water application (W/Wopt)
Fig. 6. Irrigation water-yield functions. Examples of quadratic equations with different

descending branches for a crop susceptible to excess water in a poorly drained soil (─), for a
susceptible crop in a drained soil (- - -), and for a non-susceptible crop in a well drained soil
(─ ─)
The potential soil erosion attribute is a qualitative index, EI, which takes values 1-9 in an
ascending scale of risk of soil erosion due to the applied irrigation water. EI is an empirical
function of the inflow rate q per unit width or per furrow, the longitudinal slope So, and the
soil type considering the empirical concept of maximum nonerosive inflow rate as proposed
by Criddle et al. (1956). For furrows in a silt-loam soil, EI values relative to single events are
given in Table 5 as a function of the product q So, where q in l s-1 and So in percentage. For
different soil types this scale values should change similarly to the maximum nonerosive
inflow rate. The seasonal EI is the geometric average of the single event values.
q So (l s-1 %) <0.30 0.3-0.5 0.5-0.62 0.62-0.75 0.75-0.87 0.87-1.0 1.0-1.25 1.25-1.5 >1.5
EI 1 2 3 4 5 6 7 8 9
Table 5. Potential erosion index EI for furrows in a silt-loam soil
2.3 Multicriteria analysis

The selection of the best irrigation design alternative is a multiple objective problem whose
rational solution requires multicriteria analysis. This methodology integrates different types
of attributes on a trade-off analysis, allowing the comparison between environmental and
economic criteria (Pomerol & Romero, 2000). Multicriteria analysis supports a better
understanding of the irrigation impacts while enabling us to achieve a satisfactory
compromise between adversative decision-maker objectives.
The process starts with the definition of the design objectives and related attributes (Fig. 7).
These attributes are then transformed into criteria through user-defined value or utility
functions [Eqs. (9) and (10) presented below]. The alternatives and respective criteria are
tabled in a payoff matrix, which synthesize the more relevant data for the selection analysis.
A first screen of the alternatives may be done prior to the application of multicriteria using a
dominance and satisfaction preanalysis.
User definitions Design objectives
Attributes of the design

alternatives
Value functions
Evaluation of alternatives
relative to each criterion
Payoff matrix
Decision-making Dominance and satisfaction

priorities pre-analysis
Multicriteria analysis with

Composite Programming
and/or ELECTRE II
Ranking
Fig. 7. Scheme of application of the multicriteria methodology

A set of design objectives and correspondent attributes are presented in Table 6.
Objectives Alternative Attributes units

Maximizing benefits Yield value €/year
Investment cost €/year
Minimizing costs
Operation and maintenance cost €/year
Minimizing water degradation and Deep percolation m3/year
the non-beneficial water use Runoff m3/year
Average land levelling cutting depth cm
Maximizing soil conservation
Erosion index index
Table 6. Objectives and attributes for design selection.
The decision criteria refer to:
1. Economic criteria relative to the yield value, the initial investment cost, and the
operation and maintenance costs;
2. Environmental criteria relative to the potential degradation of groundwater and surface
waters and the reduction of nonbeneficial water uses relative to tail-end runoff and
deep percolation, potential erosion due to irrigation water flowing over the soil surface,
and soil degradation due to land leveling cuts. Hydraulic criteria are represented in the
environmental criteria through controlling runoff and percolation.
The attributes are scaled according to a measure of utility using value or utility functions,
which are applied to the environmental and economic criteria. This approach enables to
compare variables having different units, which is one of the primary benefits of
multicriteria methodology. With this procedure, the utilities Uj for any criterion j are
normalized into the 0-1 interval (zero for the more adverse and 1 for the most
advantageous result).
The following linear utility function is applied for the economic criteria:
U j (x j )    x j   (9)
where xj = attribute:  = graph slope, negative for costs and positive for benefits; and the
parameter  = utility value Uj(xj) for a null value of the attribute. A logistic utility function
(Fig. 9) is adopted for the environmental criteria:

K bj  x Mj  x j 
 
Uj xj 
e
K aj 
K bj  x Mj  x j 
(10)
e e
where Kaj and Kbj = function parameters; and xMj = maximum attribute value corresponding
to a null utility for the j criterion.
1
0.9
0.8
0.7
0.6
Utility
0.5
0.4
0.3
0.2
0.1
0
0 20 40 60 80 100 120 140 160 180 200
De e p per colation (m m /ye ar)
Fig. 8. Logistic utility function relative to deep percolation where xa = 40 mm and xb= 110
mm, and U(xa) = 0.95 and U(xb) = 0.05
To adjust this function (10) to user preferences, it is necessary to select the attribute values xa
and xb that correspond to a very low and a very significant impact, e.g., such that
U(xa) = 0.95 and U(xb) = 0.05, as shown in Fig. 8. Based on these values, and for a specific
criterion, the parameters Ka and Kb are then calculated as follows (Janssen, 1992):
  1  U ( x ) .e Kb  M  xa  
a
K a  ln   (11)
 U(xa ) 
 U ( x a )   1  U ( xb )  
ln  
  1  U( xa )   U ( xb ) 
Kb   (12)
xb  x a
The dominance preanalysis is a procedure to select the nondominated alternatives. For these
alternatives do not exist any other feasible alternative that could improve the performance
relative to any criterion without decreasing the performance of any other criterion. The
multicriteria selection applies to those nondominated alternatives. The satisfaction
preanalysis screens the alternatives set by selecting the user acceptable ones, i.e., those that
for every criterion perform better than a minimum level required by the decision maker.
To apply the multicriteria methods, the user needs to assign priorities by selecting the
weights j that represent the relative importance of each criterion j as viewed by the decision
maker. These can be directly defined by the decision maker or calculated by the Analytical
Hierarquical Process (AHP) method (Saaty, 1990).
Two multicriteria methods may be applied: the composite programming (Bogardi &
Bardossy, 1983) and the ELECTRE II (Roy & Boussiou, 1993; Roy, 1996). The composite
programming is an aggregative multicriteria method that leads to a unique global criterion.
It is a distance-based technique designed to identify the alternative closest to an ideal
solution using a quasi-distance measure. This method allows the analysis of a
multidimensional selection problem by a partial representation in a two dimensions trade-
off surface.
The distance to the ideal point (Lj = 1-Uj) relative to each alternative ak, is a performance
measure of ak according to the criterion j. The ideal point represents the point on the trade-
off surface where an irrigation design would be placed if the criteria under consideration
were at their best possible level. If this distance is short, this performance is near the
optimum. The composite distance L is computed for each set of N criteria as
1/p
N 
L    j  Lpj  (13)
 j  1 
where Lj = distance to the ideal point relative to criterion j; and p = balancing factor between
criteria. Each composite distance corresponds to a distance-based average, arithmetic or
geometric, respectively when p = 1 or p = 2. The balancing factor p indicates the importance
of the maximal deviations of the criteria and limits the ability of one criterion to be
substituted by another. A high balancing factor gives more importance to large negative
impacts (a larger distance to the ideal point) relative to any criterion, rather than allowing
these impacts to be obscured by the trade-off process.
The ELECTRE II is an outranking method that aims to rank alternatives. It is based on the
dominance relationship for each pair of alternatives, which is calculated from the
concordance and discordance indices. The concordance represents the degree to which an
alternative k is better than another alternative m. A concordance index is then defined as the
sum of weights of the criteria included in the concordance set relative to the criteria for
which the alternative k is at least equally attractive as the alternative m. The discordance
reflects the degree to which an alternative k is worse than alternative m. For each criterion
from a discordance set, that includes the criteria for which alternative k is worse than m, the
differences between the scores of k and m are calculated. The discordance index, defined as
the largest of these differences, reflects the idea that, beyond a certain level, bad
performances on one criterion cannot be compensated by a good performance relative to
another criterion. The decision maker indicates the thresholds that are used to establish a
weak and a strong outranking relationship between each pair of alternatives.
For every project of a given field (Fig. 4), SADREG produces a large set of alternatives as a
result of numerous combinations of design and operation variables. As mentioned before,
these alternatives are clustered by groups relative to different water distribution equipment
and tail end management options. The multicriteria analysis allows the alternative selection
for each group and the ranking and comparison of the groups of a given project. This
analysis plays an important role on automatic management of large amount of data,
screening the alternatives, removing those not satisfactory or dominated, and ranking and
selecting the most adequate according to the user priorities.
3. Web-based DSS methodology

The Web DSS application is based on client-server architecture. It comprises a Web module
and a simulation engine, which includes the simulation models of SADREG (Fig. 9). The
Web module controls the simulation engine, creates the user interface, importing and
showing numerical and graphical data. The architecture of DSS Web Server and Client are
schematized in Fig. 10 and 11. PHP and C# languages are applied to achieve higher system
flexibility, and to minimize client system requirements. The SQLServer is applied for
database management, which allows a simultaneous connection of several users. The server
is established by four component modules, each one responsible for a task (Jia & Zhou,
2005): (1) Communication – the interface with the Web applications using TCP/IP like
transport way; (2) Logic – the control of execution and respective data flow; (3) Simulation –
the computation of simulation models; and (4) Data abstraction – the isolation and
optimization data model and data modifications.
Fig. 9. Abstract system architecture

The application is multiuser, with properly warrant data protection among the several
users. The user input data are provided to server that controls the simulation engine and
return the output data to server that will be accessed by user. The software developed is
based on first-in first-out philosophy, which uses a resources list that is being used and free
up to the end of the simulation (Jia & Zhou, 2005). The communications with the server are
performed in a first step of software development with TPC sockets, to avoid entropy. For
normal system operation, a protection layer based on SSL protocol will be added for a better
guarantee of communications security. The same procedure is adopted with the Web
application that will use HTTPS connections. The protocol used in the communication with
the server, without state and atomic operations, will be published to permit create other
applications based on this service. The user does not interact directly with the simulation
engine to shorten the time required to complete the tasks. When the user requests a
simulation, the order remains in a waiting list until the simulation engine is available to
process it. Then the produced results are recorded in the simulation data base, the user is
notified and the output becomes available on the Web application.
External
Services
CExternalRequest
CListenAndOperate Clients
CExecuteModels CFacadeDataLayer
CListen
CAbstractFactoryPersistence CDBConnection
CSingletonMain
CWindowsFactoryPersistence CDBConnectionODBC
CLinuxFactoryPersistence CDBConnectionSQLSRV Main
Fig. 10. Architecture of DSS Web Server
MVC_Model
DataAcess_Model .....
.....
.....
Login_Model Field_Model
MVC_HTTPRequest cls_ServerComm
cls_FacadeDataLayer
MVC_Controler MVC_ModelReturn .....
.....
.....
cls_Field
MVC_ViewReturn
DataAcess_View .....
.....
.....
Field_View
Login_View
MVC_View
Fig. 11. Architecture of DSS Web Client

This approach confines the model relatively to Web system, allowing its reuse and
maintenance without impact on system. This model could be transformed in a stand-alone
application for off-line use. The database was designed to integrate all DSS information
required, from agronomic, climatic and economic subjects. All records had specialized
information to provide auditoria and controls of permissions, allowing the possibility to
share data and experiences by groups of users. The access control to user data is guaranteed
at level of server. In the server the data abstraction is provided by pattern AbstractFactory
(Fig. 12) to guarantee without impact on models source code a platform change. The
communications with server are made with TCP/IP sockets applying a protocol developed
specifically. The use of norms like SOAP and WebServices will be included, in a higher
layer, like a protection SSL layer, with a guarantee of security, without state and atomic
operations, will be published to permit create other applications based on this service.
SADREG server
Abstract Factory
Facade
CommandParser
Models
Listener
SIRMOD
Fig. 12. Models integration on DSS Web server.

The interface includes a context online help, allowing a direct support to users. It was
created a simple Web user friendly interface. The user may select the language of the
interface. The system has available also a Web-service interface to integrate with other
systems and a short Web version to be used with portable and limited equipments, such as
PDA for field use.
4. Applying the DSS

The data flow and logic of the execution are represented in the diagram of Fig. 13. To access
the application the user needs to register beforehand asked the system administrator, with
the conditional access by keyword. The interface language is set in the attributes of the user,
although this may change in a straightforward manner by clicking on the button.
4.1 Input data

The user can view the general options of the application, without change them, because
these data are the responsibility of the administrator. These data include, for example, the
technical characteristics of equipment for water supply, the classification of soils (example in
Fig. 14), climatic stations and the curves of infiltration. These data will be accessed during
the timely preparation of projects. In turn, the user data comprises its private information,
which may have access to the group member’s share with which the user has set permission.
The field is the subject of spatial unit design, and its register, inserted in the field workspace,
includes features such as location, soil type, and size and topography data (Fig. 15). A user
can create multiple field workspaces, from a new record or copy of an existing one. Other
user data refers to crops and unit costs, which may well be applied in different fields of the
same user.
User data login
Data Preparation: View general options:

Topography Climatic stations
Water supply Soils
Field
Crops Infiltration curves
Infiltration family Equipments
Economics Technical options
Create a
Project
Land leveling Select design
Design options
solution alternatives
select irrigation
method Surface irrigat.
Select the crop (mod. SIRMOD) Choose
selection
Create Land profile
leveling Create irrigation Create
solution scheduling Alternatives
Multicriteria
select the field Irrigation analysis
slopes scheduling Simulation of
(mod. ISAREG) Alternatives
Ranking of
Land leveling
irrigation Design design
simulation alternatives
scheduling Alternatives
Fig. 13. Data and simulation flowchart
4.2 Projects
Once the user has completed the field data register, he(she) builds projects to create
alternatives. Each project refers to a particular method of irrigation and a land leveling
solution. This solution is obtained by simulation the land leveling, using the field
topography of the parcel and the slopes selected by user. The user will have access to the
simulation results, which included the volumes handled, the spatial distribution of the
depths of excavation and fill and the estimated time of execution of work and its cost.
Interactively, the user can do several simulations to find a solution that he(she) considers
appropriate for the project.
In the next step, the user chooses a crop to be considered in the project within a list of
options. The determination of the irrigation scheduling, i.e., the appropriate application
depths and the irrigation dates along the crop period, is obtained with the ISAREG model.
This calculation requires the application of climatic data from the region where the field is
located, and information about the type of soil and crop characteristics. This process is also
interactive with the user, because there are several possible solutions to these calendars. The
user will then choose the one more appropriate to the project.
Fig. 14. Example of view access to general options data on Web interface (example of soil
type)
Fig. 15. Example of field input data on Web interface

4.3 Design options

The user sets the project options that will be applied in creating alternatives. These options
relate to supply conditions, such as surge or continuous inflow, possible field subdivision in
the longitudinal and cross directions, equipment for water distribution (e.g., gated pipes or
canals), the flow conditions at the downstream end (e.g., open, closed or re-use), and the
constraints in supplying water to the field (flowrate, hydraulic head, frequency). Then the
application runs the model SIRMOD for hydraulic simulation of the process of irrigation or
access to previously recorded results, if they exist in the database, and then proceeds to
create alternatives. The user then accesses the results, refering the projects and groups (Fig.
16), view and analyses the alternatives developed (Fig. 17), and may remove them when
they are unsatisfactory.
Fig. 16. Example of project results view on Web interface
4.4 Selecting alternatives

The next step is the selection of design alternatives through the module using multicriteria
analysis (MCA). For this purpose the user must set the profile to apply for selection (Fig. 18),
which is the set of multicriteria analysis parameters, such as the method, the value functions
associated with the attributes, the decision criteria and corresponding weights defining the
user priorities. The results will include the identification of effective alternatives, i.e., those
that have enough quality to be selected, and the unsatisfactory or dominated, i.e., the ones
that are not considered for the subsequent ranking of alternatives. This process is interactive
with the user, which could generate new rankings when changing the profile selection.
Once this process is ended, the user has information to make the decision, thus choosing the
alternative which he(she) prefers to see implemented, or, the user may seek for another
solution to the problem by creating a new project, thus changing any of the options made.
This iterative process is controlled by the user, resulting in increased number of design
alternatives, which may be all together analyzed and ranked by the MCA module.
Fig. 17. Example of alternative view on Web interface, showing the main results itens
Fig. 18. Example of selection profile definition on Web interface

5. Conclusion
Appropriate design of surface irrigation systems requires a complex manipulation of data,
models, and decisions. The multicriteria approach integrated in a DSS enables us to solve
that complexity while creating and ranking a large number of design alternatives. The DSS
SADREG has shown to be an appropriate tool for (1) generating design alternatives
associated with attributes of technical, economic, and environmental nature; (2) handling
and evaluating a large number of input and output data; (3) evaluating and ranking design
alternatives using multicriteria analysis where criteria are weighted according to the
priorities and perception of the designer and users; and (4) providing an appropriate
dialogue between the designer and the user.
The Web-based DSS allows a simple way to support surface irrigation designers to improve
the project procedures, particularly in the world water scarcity areas where the surface
irrigation requires improvements and the expert technical support is more incipient. With a
simple Web user friendly interface, with several optional languages and an online help, this
tool would contribute to an effective support to enlarge the knowledge of surface irrigation,
its design procedures and field practices.
In conjunction with web application SADREG, tools are available towards the resolution of
specific problems such as land leveling, pipe sizing and economic calculation, as well as
documents on good equipment and irrigation practices. The application maintenance and
user support is the responsibility of the multidisciplinary team that developed the software,
with computer engineers and agronomists, the objective being that of a dynamically adjusts
the system to answer the most frequent difficulties and new challenges more exciting. This
software is available on http://sadreg.safe-net.eu Web site.
6. Acknowledgment
The support of CEER - Biosystems Engineering (POCTI/U0245/2003) is acknowledged.
Studies referred herein are developed in the framework of the research project PTDC/AGR-
AAM/105432/2008 founded by Portuguese Foundation for Science and Technology.
7. References
Bazzani, G. M. (2005). “An integrated decision support system for irrigation and water
policy design: DSIRR.” Environmental Modelling & Software, 20(2), 153-163.
Bogardi, I. & Bardossy, A. (1983). “Application of MCDM to geological exploration,” In: P.
Hansen, ed., Essays and Surveys on Multiple Criteria Decision Making. Springer
Verlag, New York.
Boonstra, J. & Jurriens, M. (1978). BASCAD - A Mathematical Model for Level Basin
Irrigation. ILRI Publication 43, Wageningen.
Burt, C. M.; Clemmens, A. J., Strelkoff, T. S., Solomon, K. H.; Bliesner, R. D.; Hardy, L. A.;
Howell, T. A. & Eisenhauer, D. E. (1997). “Irrigation performance measures:
efficiency and uniformity.” J. Irrig. Drain. Eng., 123, 423-442.
Camacho, E.; Pérez, C.; Roldán, J. & Alcalde, M. (1997). “IPE: Model for management and
control of furrow irrigation in real time.” J. Irrig. Drain. Eng., 123(IR4), 264-269.
Car, N.J.; Christen, E.W.; Hornbuckle, J.W. & Moore, G. 2007. Towards a New Generation of
Irrigation Decision Support Systems – Irrigation Informatics? In: Oxley, L. &
Kulasiri, D. (Eds) MODSIM 2007 International Congress on Modelling and Simulation.

Modelling and Simulation Society of Australia and New Zealand, pp. 74-80. ISBN:
978-0-9758400-4-7, pp135-141.
Clemmens, A. J. (1979). “Verification of the zero-inertia model for border-irrigation
advance.” J. Irrig. Drain. Div, 104(IR3), 339-341.
Clemmens, A. J. (2007). “Simple approach to surface irrigation design: Theory.” e-Journal of
Land and Water, Vol. 1, 1-19 (http://www.sakia.org/ejlw).
Clemmens, A. J., Dedrick, A. R. & Strand, R. (1993).“BASIN 2.0 for the design of level-basin
systems.” In: Allen R. G., ed., Management of Irrigation and Drainage Systems, ASCE,
New York, 875 -882.
Criddle, W. D.; Davis, S.; Pair, C. & Shockley, D. (1956). Methods for Evaluating Irrigation
Systems. USDA SCS Agricultural Handbook 82.
Dedrick, A. R.; Gaddis, R. J.; Clark, A. W. & Moore, A. W. (2007). “Land forming for
irrigation.” In: Hoffman, G. J.; Evans, R. G.; Jensen, M. E.; Martin, D. L. & Elliot, R.
L. eds., Design and Operation of Farm Irrigation Systems (2nd Edition), ASABE, St.
Joseph, MI, 320-346.
Elliott, R. L.; Walker, W. R. & Skogerboe, G. V. (1982). “Zero inertia modeling of furrow
irrigation advance.” J. Irrig. Drain. Eng, 108(IR3), 179-195.
Fangemeier, D. D. & Strelkoff, T. (1978). “Mathematical models and border irrigation
design.” Trans. ASAE, 22(1), 93-99.
Gonçalves, J. M.; Pereira, L. S. (2009). “A decision support system for surface irrigation
design.” J Irrig. Drain. Engng. 135, 343-356
Hart, W. E.; Collins, H. G.; Woodward, G. & Humpherys, A. S. (1980). Design and operation
of gravity or surface systems. In: M. E. Jensen, ed., Design and Operation of Farm
Irrigation Systems. ASAE Monograph No. 3, St. Joseph, 501-580.
Hornbuckle, J. W.; Christen, E. W. & Faulkner, R. D. (2005). “Use of SIRMOD as a quasi real
time surface irrigation decision support system.” In: Zerger, A. & Argent, R. M.
eds., MODSIM 2005 International Congress on Modelling and Simulation.
Modelling and Simulation Society of Australia and New Zealand, 217-223.
Horst, M. G.; Shamutalov, Sh. Sh.; Gonçalves, J. M. & Pereira, L. S. (2007). “Assessing
impacts of surge-flow irrigation on water saving and productivity of cotton.” Agric.
Water Manage. 87, 115-127.
Janssen, R. (1992). Multiobjective Decision Support for Environmental Management. Kluwer
Academic Publisher, Dordrecht.
Jia, W. & Zhou, W. 2005. Distributed Network Systems. Springer.
Latimer, E. A. & Reddel, D. L. (1990). A volume balance model for real time automated
furrow irrigation system. In: Visions of the Future - Technology to enrich our
environment. ASAE, St. Joseph, 13-20.
Mailhol, J. C.; Ruelle, P. & Popova, Z. (2005). Simulation of furrow irrigation practices
(SOFIP): a field-scale modelling of water management and crop yield for furrow
irrigation. Irrig. Sci. 24(1), 37-48.
McClymont, D. (1999). “FIDO v2: Furrow irrigation design optimiser.” In: Multiple
Objective Decision Support Systems Conference, Brisbane
(http://www.ncea.org.au/Irrigation/FIDO/ Project_FIDO.htm).
Mjeld, J.W.; Lacewell, R.D.; Talpaz, H. & Taylor, C.R. (1990). Economics of irrigation
systems. In: Hoffman, G.F.; Howell , T. A. & Solomon, K. H., eds., Management of
Farm Irrigation Systems, ASAE, St. Joseph, MI, 461-493.
Oad, R.; Garcia, L.; Kinzli, K. D. & Patterson, D. (2006). “Decision support systems for
efficient irrigated agriculture.” In: Lorenzini, G. & Brebbia, C. A., eds., WIT
Transactions on Ecology and the Environment, Vol 96, WIT Press, Wessex, 247-256.
Oweis, T. Y. & Walker, W. R. (1990). “Zero-inertia model for surge flow furrow irrigation.”
Irrig. Sci., 11(3), 131-136.
Pedras C.M.G., Pereira L.S., Gonçalves J.M. (2009). “MIRRIG: A decision support system for
design and evaluation of microirrigation systems.” Agric. Water Manage 96, 691-701
Pereira, L. S. (1999). “Higher performances through combined improvements in irrigation
methods and scheduling: A discussion.” Agric. Water Manage. 40(2-3), 153-169.
Pereira, L. S., and Trout, T. J. (1999). “Irrigation methods.” In: van Lier HN, Pereira LS,
Steiner FR, eds., CIGR Handbook of Agricultural Engineering, Land and Water
Engineering, vol. I. ASAE, St. Joseph, MI, 297–379.
Pereira, L. S.; Oweis, T. & Zairi, A. (2002). “Irrigation management under water scarcity.”
Agric. Water Manage., 57, 175-206.
Pereira, L. S.; Teodoro, P. R.; Rodrigues, P. N. & J. L. Teixeira. (2003). “Irrigation scheduling
simulation: the model ISAREG.” In: Rossi, G.; Cancelliere, A.; Pereira, L. S.; Oweis,
T.; Shatanawi, M. & Zairi, A., eds., Tools for Drought Mitigation in Mediterranean
Regions. Kluwer, Dordrecht, 161-180.
Pereira, L. S.; Zairi, A. & Mailhol, J. C. (2006). “Irrigation de Surface.” In: Tiercelin, J. R. &
Vidal, A., eds., Traité d’Irrigation, 2ème edition, Lavoisier, Technique &
Documentation, Paris, 513-549.
Pereira, LS, Cordery, I, Iacovides, I, 2009. Coping with Water Scarcity. Addressing the
Challenges. Springer, Dordrecht, 382 p.
Pomerol, J. C. & Romero, S. B. (2000). Multicriterion Decision in Management: Principles and
Practice. Dordrecht, The Netherlands: Kluwer Academic Publishers.
Raju, K. S. & Duckstein, L. (2002). “Multicriterion analysis for ranking an irrigation system:
an Indian case study.” J. Decision Systems, 11, 499–511.
Raju, K. S.; Kumar, D. N. & Duckstein, L. (2006). “Artificial neural networks and
multicriterion analysis for sustainable irrigation planning,” Computers & Operations
Research, 33(4), 1138–1153.
Rao, N. H.; Brownee, S. M. & Sarma, P. B. (2004). “GIS-based decision support system for
real time water demand estimation in canal irrigation systems.” Current Science, 87,
5-10.
Reddy, J. M. & Clyma, W. (1981a). “Optimal design of furrow irrigation systems.” Trans.
ASAE, 24(3), 617-623.
Reddy, J. M. & Clyma, W. (1981b). “Optimal design of border irrigation system.” J Irrig.
Drain. Eng., 17(IR3), 289-306.
Riesgo, L. & Gómez-Limón, J. A. (2006). “Multi-criteria policy scenario analysis for public
regulation of irrigated agriculture.” Agric. Systems 91, 1–28.
Roy, B. (1996). Multicriteria Methodology for Decision Aiding. Kluwer Academic Publishers,
Dordrecht, The Netherlands.
Roy, B. & Bouyssou, D. (1993). Aide Multicritère: Méthodes et Cas. Economica, Paris.
Saaty, T. L. (1990). ”How to make a decision: the AHP.” Europ. J. Operat. Res., 48(1), 9-26.
SCS (1974). Soil Conservation Service National Engineering Handbook. Section 15. Irrigation,
Chapter 4, Border Irrigation, USDA Soil Conservation Service, Washington.
SCS (1979). Soil Conservation Service National Engineering Handbook, Section 15, Irrigation,
Chap.5, Furrow Irrigation, USDA Soil Conservation Service, Washington.
Solomon, K. H.; El-Gindy, A. M. & Ibatullin, S. R. (2007). “Planning and system selection.”
In: Hoffman, G. J.; Evans, R. G.; Jensen, M. E.; Martin, D. L. & Elliot, R. L., eds.,
Design and Operation of Farm Irrigation Systems (2nd Edition), ASABE, St. Joseph, MI,
57-75.
Solomon, K.H. (1984). “Yield related interpretations of irrigation uniformity and efficiency
measures” Irrig. Sci., 5(3), 161-172.
Strelkoff, T. (1993). SRFR. A Computer Program for Simulating Flow in Surface Irrigation.
Furrows, Basins, Borders. USDA, ARS, USWCL, Phoenix.
Strelkoff, T. & Clemmens, A. J. (2007). “Hydraulics of surface systems.” In: Hoffman, G. J.;
Evans, R. G.; Jensen, M. E.; Martin, D. L. & Elliot, R. L., eds., Design and Operation of
Farm Irrigation Systems (2nd Edition), ASABE, St. Joseph, MI, 436-498.
Strelkoff, T. & Katapodes, N. D. (1977). “Border irrigation with zero inertia.” J. Irrig. Drain.
Div., 103(IR3), 325-342.
Strelkoff, T.; Clemmens, A. J.; Schmidt, B. V. & Slosky, E. J. (1996). BORDER - A Design and
Management Aid for Sloping Border Irrigation Systems. Version 1.0. WCL Report #21.
USDA, ARS, USWCL, Phoenix.
Strelkoff, T. & Souza, F. (1984). “Modelling the effect of depth on furrow infiltration.” J. Irrig.
Drain. Eng., 110 (4), 375-387.
Thysen, I. & N.K. Detlefsen. 2006. Online decision support for irrigation for farmers. Agric.
Water Manage., 86, p.269–276.
Walker, W. R. (1989). Guidelines for Designing and Evaluating Surface Irrigation Systems.
Irrigation and Drainage Paper No. 45, FAO, Rome.
Walker, W. R. 1998. SIRMOD – Surface Irrigation Modeling Software. Utah State University,
Logan.
Walker, W. R. & Humpherys, A. S. (1983). “Kinematic-wave furrow irrigation model.” J.
Irrig. Drain. Div., 109(IR4), 377-392.
Walker, W. R. & Skogerboe, G. V. (1987). Surface Irrigation: Theory and Practice. Prentice-Hall,
Inc., Englewood Cliffs, New Jersey.
Wilke, O. & Smerdon, E. T. (1965). “A solution of the irrigation advance problem.” J. Irrig.
Drain. Div., 91 (IR3), 23-24.
Yu, F. X. & Singh, V. P. (1989). “Analytical model for border irrigation.” J. Irrig. Drain. Eng.,
115(6), 982-999.
Yu, F. X. & Singh, V. P. (1990). “Analytical model for furrow irrigation.” J. Irrig. Drain. Eng.,
116(2), 154-171.
Part 4
Applications in Spatial Management

16
A Decision Support System (FMOTS)

for Location Decision of Taxicab Stands
Ebru Vesile Ocalir, Ozge Yalciner Ercoskun and Rifat Tur
Department of Traffic Planning, Institute of Science and Technology, Gazi University
Department of City and Regional Planning, Faculty of Architecture, Gazi University
Department of Civil Engineering, Faculty of Engineering, Akdeniz University
Turkey
1. Introduction
Taxi systems and their municipal organization are generally problem area in the
metropolitan cities of developing countries. The locations of taxicab stands can be market-
oriented and can cause chaos. A decision support system is needed for the related agencies
to solve such kind of problems. In this chapter, a decision support system for taxicab stands
that can be used in any metropolitan area or municipality is presented. The study attempts
to create a scientific basis for decision makers to evaluate the location choices of taxicab
stands in major cities with the help of GIS and fuzzy logic.
Taxi, which has been worldwide used since the 19th century, is an indispensable component
of urban transport. Compared to other modes of transport, taxi has a relative advantage
with the comfort and convenience that it provides 24 hours a day to its users. However, it is
criticized because of its low occupancy rates and traffic burden it loads on urban streets.
Taxicab stands offer a viable service by providing an identifiable, orderly, efficient, and
quick means to secure a taxi that benefits both drivers and passengers (Giuliani et.al, 2001).
Stands are normally located at high-traffic locations such as airports, hotel driveways,
railway stations, subway stations, bus depots, shopping centers and major street
intersections, where large number of passengers are likely to be found. The choice of
location for taxicab stands depends only on legal permissions. From the legal authorities’
side, there is no evidence of taking some scientific criteria into account while giving the
permission. The entrepreneurs willing to manage a taxicab stand have limitless
opportunities to select any point on urban land. These free market conditions cause debates
on location choices of taxicab stands.
In some cities, taxi companies operate independently and in some other cities, the activity of
taxi fleets is monitored and controlled by a central office, which provides dispatching,
accounting, and human resources services to one or more taxi companies. In both systems,
the optimum organization of taxi companies on urban space is a problem. The taxi company
in the system should provide the best service to the customer, which includes reliability and
minimum waiting time.
As private entrepreneurs, taxi companies tend to locate in some certain parts of the city,
where they believe they can find more passengers. The decision makers, on the other hand,
should find an optimum between taxi companies’ demands and the city’s real needs. They
would rather improve the effectiveness of taxicab stands as a tool to reduce congestion,
while improving convenience for passengers and taxi drivers.
There are two major obstacles confronting the decision makers, who try to evaluate location
choices of taxi companies: The first is about the scale of the analyzed area. Taxicab stands
are the problem of major cities and managing data in this scale requires specific methods. In
this study, GIS based service area analyses helped to define the area, which can be accessed
from taxicab stands within a psychologically accepted time limit (and critical areas that are
out of service range).
The second problem of the decision makers is the lack of a certain measure for this specific
decision. An alternative way of thinking, which allows modeling complex systems using a
higher level of abstraction originating from our knowledge and experience, is necessary.
Fuzzy logic is an organized and mathematical method of handling inherently imprecise
concepts. It has proven to be an excellent choice for many control system applications since
it mimics human control logic. It uses an imprecise but very descriptive language to deal
with input data more like a human operator.
This study appeared as a response to the demands of the municipalities and the taxi
associations in Turkey. The research is focused on both issues: how well the existing taxicab
stands are located and where the most appropriate places for the incoming demand should
be. The pioneering steps of the study have already been presented in a scientific paper
(Ocalir, et. al, 2010). An integrated model of GIS and fuzzy logic for taxicab stand location
decision is built. GIS were used for generating some of the major inputs for the fuzzy logic
model. The location decisions of taxicab stands in big cities, where organizational problems
of taxicab stands slow down a better quality of service, have been examined by this
integrated model and their appropriateness has also been evaluated according to the
selected parameters. With the fuzzy logic application, evaluation of the existing taxicab
stands is done and decision for new taxicab stands is given. The equations are obtained by
artificial neural network (ANN) approach to predict the number of taxicab stands in each
traffic zone.
In the next section, background information for the tools used (the integration of GIS and
fuzzy logic) is provided. It is followed by the introduction of a decision support system, the
Fuzzy Logic Model for Location Decisions of Taxicab Stands (FMOTS), with a case study in
Konya (Turkey). The applicability of the FMOTS model for big cities for the selected case
study Konya (Turkey) with a population of approximately 2000000 will be examined.
The taxicab stand allocating problem came in agenda as taxi use gained importance in recent
years parallel to the development of university and its facilities. Having sprawled to a large
area in a plain, Konya accommodates different sustainable transportation systems. FMOTS
should be applied to this city for understanding supply and demand of taxicab stands.
Advanced GIS techniques on networks, fuzzy logic 3D illustrations and artificial neural
network equations for new demands will be given for taxicab stands in Konya, Turkey. The
paper closes with conclusions.
2. Incorporating fuzzy logic into GIS operations

Defining service areas is an important geographical application (Upchurch et.al, 2004) and
with use of GIS, some studies have been performed. The service area analysis of transit
services by GIS has been performed in O’Neill (1995) and Horner and Murray (2004)’s
researches. In O’Neill’s (1995) study, by defining a transit route's service area, walking
A Decision Support System (FMOTS) for Location Decision of Taxicab Stands 323
distance and travel time were used as acceptable limits for transit users. Another study with
GIS was performed (Walsh et al., 1997) for exploring a variety of healthcare scenarios, where
changes in the supply, demand, and impedance parameters were examined within a spatial
context. They used network analysis for modeling location/allocation to optimize travel
time and integrated measures of supply, demand, and impedance.
Fuzzy systems, describe the relationship between the inputs and the output of a system
using a set of fuzzy IF-THEN set theory (Zadeh, 1965). The classical theory of crisp sets can
describe only the membership or non-membership of an item to a set. Fuzzy logic, on the
other hand, is based on the theory of fuzzy sets, which relates to classes of objects with
unsharp boundaries, in which membership is matter of degree. In this approach, the
classical notion of binary membership in a set has been modified to include partial
membership ranging between 0 and 1.
Fuzzy Logic will be increasingly important for GIS. Fuzzy Logic accounts a lot better to
uncertainty and impreciseness in data as well as to vagueness in decisions and
classifications than Boolean Algorithms do. Many implementations have proved to get
better output data. In the recent years, fuzzy logic has been implemented successfully in
various GIS processes (Steiner, 2001). The most important contributions are in the fields of
classification, measurement, integrated with remote sensing (Rashed, 2008), raster GIS
analysis and experimental scenarios of development (Liu and Phinn, 2003), risk mapping
with GIS (Nobre et al., 2007; Ghayoumian, et.al,2007; Galderisi et.al, 2008), ecological
modeling (Malins and Metternicht, 2006; Strobl, et.al, 2007), qualitative spatial query (Yao
and Thill, 2006), data matching (Meng and Meng, 2007), site selection (Alesheikh et.al, 2008;
Puente et al., 2007) and finally in road network applications (Petrik et al., 2003).
In the scientific literature, there are many studies, which implements fuzzy logic onto some
transport problems. Some of these studies focus on transport networks. Choi and Chung
(2002) for instance, developed an information fusion algorithm based on a voting technique,
fuzzy regression, and Bayesian pooling technique for estimating dynamic link travel time in
congested urban road networks. In another study (Chen et.al, 2008) used fuzzy logic
techniques for determining the satisfaction degrees of routes on road networks. Ghatee and
Hashemi (2009) studied on a traffic assignment model in which the travel costs of links
depend on their congestion. Some other studies are on signalized junctions, which are
closely related with queuing theory: Murat and Baskan (2006) developed a vehicle delay
estimation model especially for the cases of over-saturation or non-uniform conditions at
signalized junctions. Murat and Gedizlioglu (2005) developed a model for isolated
signalized intersections that arranges phase green times (duration) and the other phase
sequences using traffic volumes. In another study, Murat and Gedizlioglu (2007) examined
the vehicle time headways of arriving traffic flows at signalized intersections under low
traffic conditions, with data from signalized intersections.
The model brings some innovation by implementing fuzzy logic to a transport problem for
an area based analysis. Point and linear analyses could not bring solutions to the mentioned
problem. An area-based analysis on traffic zones is required to give transportation planning
decisions about an urban area with respect to some land use data such as employment and
population. An analytic process and fuzzy logic functions have been considered for the
evaluation of model stages. The fuzzy logic permits a more gradual assessment of factors.
The implementation of the whole model requires a big amount of information, so the model
validation has been done only at the traffic zone level.
Increasing awareness about the need of designing and performing new development and
site selection models has made necessary the implementation of many more new factors and
parameters than those presented in traditional location models, which results major
complexity for the decision making processes. That is the reason why in this research a GIS
platform has been used to spatially analyze the service area of taxicab stands, bearing in
mind the hierarchical structure of location factors and considering the fuzzy logic attributes.
The new proposed methodology gathers the necessary tools: GIS software, to organize the
datasets and to apply geo-processing functions on a clear interface; Fuzzy Logic software, to
define and execute the evaluation methodology and to carry out the evaluation process. The
creation of an expert system based on GIS software allows the user to query the system
using different groups of criteria. This makes the planning process for the decision makers
easier.
The fuzzy logic gives the system a type of evaluation closer to the complex reality of urban
planning. The FMOTS is a helpful tool based on a multi-criteria evaluation methodology.
Location attributes are an important source of information for empirical studies in spatial
sciences. Spatial data consist of one or few cross-sections of observations, for which the
absolute location and/or relative positioning is explicitly taken into account on the basis of
different measuring units. Using fuzzy logic operators for modeling spatial potentials brings
more sophisticated results because the information about membership values gives a more
differentiated spatial pattern. A lot of the events, which are analyzed using spatial analysis
techniques, are not crisp in nature. To analyze such events, it is common to commute fuzzy
data into crisp data, which stands for a loss of information (Wanek, 2003).
GIS operations for data interpretation can be viewed as library of maps with many layers.
Each layer is partitioned into zones, where the zones are sets of locations with a common
attribute value. Data interpretation operations available in GIS characterize locations within
zones. The data to be processed by these operations include the zonal values associated with
each location in many layers (Stefanakis et al., 1996). A new class of data with measurement
operations in this field that correspond to the area and length characterizing the traffic zones
is computed. The transformation of these values is accomplished through fuzzification. This
study gives a methodology to quantify taxicab stand complexity using fuzzy. In essence
fuzzy logic opens the door to computers that understand and react to the language and the
behavior of human beings rather than machines (Narayanan et al., 2003).
3. System design
In this study, an expert system based on fuzzy logic with related modelling software and the
GIS software is developed. The model permits us generate digital suitable locations and can
also be used to evaluate existing taxicab stands from a sustainable point of view. It is indeed
a decision support system, which is an integrated model of GIS and fuzzy logic applications.
The design of the proposed model is given in Fig. 1. After the definition of the problem, GIS
application begins. The tabular data, together with geo-referenced data, are processed by
GIS applications. However, some outputs of this phase needs some further process with a
fuzzy logic application. In this fuzzy logic application, in addition to some former tabular
data, the amount of roads which have (AND don’t have) adequate taxi service, are
calculated by network analysis tool of GIS and used as basic inputs.
The location analysis of the taxicab stands in Konya begins with GIS. The first level of a GIS
study is the database creation process, which is necessary for network analysis. Some points
should be noticed before dealing with the networking database, such as the data structure,
data source and the consistency of the database.
START
PROBLEM
DEFINITION
Need to evaluate
taxicab stand locations GEO-REFERENCED
in major cities DATA
Position of the taxi

TABULAR DATA GIS APPLICATION stands
(Site survey by GPS)
Data about taxi stands Prepare thematic maps
Information on Konya
Information on Konya Define service area of taxi road network,
land-use variables stands boundaries of the
townships and
districts
NEW DATA SET location of social
facilities
Taxi stands in each district
Amount of roads that have/don’t

have access to a taxi stand in
three minutes
EVALUATION OF THE EXISTING

FUZZY LOGIC TAXICAB STANDS
APPLICATION
FMOTS DECISION FOR NEW TAXICAB
STANDS
STOP
Fig. 1. The design of the decision support system

The geo-referenced data are grouped as spatial and descriptive data. Spatial data determine
location, form and relationships with other features. The data of the location of taxicab
stands all over Konya and the road network map have been received from a GIS software
provider in Turkey (Basarsoft) as a contribution to our study. The spatial data collected for
this study are: the current road network for the whole city, the district boundaries, locations
of the social facilities and the locations of taxicab stands. Descriptive data are listed on the
database. The database includes traffic directions and average speeds. Traffic zones are
overlaid onto the road network. The data related to the traffic zones contains employment,
area and population information and have been received from Konya Municipality for this
study. These data are current names of the roads, the districts, population of the districts,
traffic zones, car ownerships and the names and the addresses of taxicab stands.
Traffic zones are generally used as planning units for urban transportation planning
processes. Many thematic maps were produced according to these geo-referenced data,
such as population map, the number of taxis in the taxi stands with population, matched
and unmatched taxicab stand locations. Some comparisons among the districts according
to the number of taxicab stands are made. The GPS was calibrated in the most accurate
way. All projections were set to UTM (Universal Transverse Mercator) and WGS84
reference systems, zone 36 referring to Konya. No overlaying and accuracy problems
appeared for overlaying operations with the same projection and reference system in the
metropolitan scale.
The other data about road network include directions as two-ways and one-way and speed
limits. The data were compiled in GIS environment together with population, district
boundaries and urban social facilities. These data were used as the criteria for the site
selection of taxicab stands. The results were matched with the spatial data and overlaid with
districts.
The accessibility and service area analyses were produced by network analysis tool of GIS.
The road network of Konya, including all updated directions, district borders and the
location of taxicab stands are overlaid as thematic maps. A psychological threshold of 3
minutes is accepted for taxi users to have access to any taxi. Service area analysis of 53
taxicab stands was performed for each of the stands in 3 minutes drive time according to the
road network map. With the help of an additional network analysis tool, travel time was
specified in a problem definition dialog. The defined service area and the network according
to the given travel time (3 minutes) were displayed as cosmetic layers in the view. These
cosmetic layers were added onto one by one and then they were turned to a thematic layer
as 3 minute network from the location of taxicab stands (Fig. 2). A geodatabase was built
showing the lengths of roads within and outside the service area. The obtained data were
used as inputs of a fuzzy logic application.
Fig. 2. The road network and three minutes psychological threshold (in red) for the
accessibility of existing taxicab stands
With the fuzzy logic application, evaluation of the existing taxicab stands is done and
decision for new taxicab stands is given. The equations are obtained by artificial neural
network (ANN) approach to predict the number of taxicab stands in each traffic zone.
4. Background information on Konya

The area of interest is Konya, a big to medium size city in Turkey. The population of Konya
by the year 2008 is 1969868 (TUIK, 2008). There are 434000 vehicles of which 185000 are cars
by the year 2009 (TUIK, 2009). Total road network is 1060 km and 53 taxicab stands are
determined in the field study. Land-use data are obtained from the Konya Municipality and
the company Basarsoft for the year 2009. Although there are 213 districts in Konya, 99 of
them, which are close to the central areas, are included in the FMOTS model.
City centre is composed of two adjacent parts, the traditional and the modern centres. The
development is in industrial areas located in the northern and the north eastern parts of the
city. The dense settlements are in the north western parts of Konya. Population in these
districts are expected to increase in the following years. In the northern part of the city, an
important university campus is located. The southern, the south-eastern and the eastern
parts of the city are lower-density settlements .
Bicycle is an important mode of travel in Konya. The flat structure of the city supports
bicycle networks. However, developments in the recent years have brought a more intensive
use of motorised vehicles.
Fig. 3. The selected urban districts (esp. city center) which are included in the study and the
locations of existing taxicab stands
There are no restricted conditions about site selection of taxicab stands in Konya. A fuzzy
model is developed to build up a decision support system for decision makers.
5. The Fuzzy Model of Taxicab Stands (FMOTS)

For the FMOTS operation, 5 parameters, which may influence the optimum number of
taxicab stands, have been defined (Fig.4). The first two of them, employment density
(Emp/km2) and population density (Pop/km2), are related with the urban morphology and
describe the patterns of land use. The third parameter is the car ownership level (PCP),
which is accepted as an indicator of the income level. The fourth and fifth parameters
(ROSA and RWSA), the amounts of roads within and out of taxicab stands’ catchment area,
which are limited to 3 minutes as a psychological threshold, are the supply and the potential
of taxi service respectively. The last two parameters have been determined by GIS. The
proposed model is suggested to be a useful support tool for decision makers for defining the
optimum number of taxicab stands especially in the newly developed regions and zones. It
is assumed that each taxicab stand is available to supply necessary number of taxis when
needed.
Fuzzy Inference
Input Linguistic Output Linguistic
(Fuzzy) (Fuzzy)
Fuzzification
Linguistic (Fuzzy) Levels

Defuzzification
Numerical (Crisp) Level

Input Variables for Fuzzy
Model of Taxicab Stands
(FMOTS)
1. Emp/km2 (employment
density)
2. Pop/km2 (population density)
3. PCP (Private Car per 1000
People)
4. ROSA (Roads Outside Service Output of Output Numeric
Area (km)) (Crisp) Variable
5. RWSA (Roads Within Service
FMOTS
Area (km))
Fig. 4. Prototype FMOTS Model
Triangular membership functions have been used extensively in different applications

because of their simplicity. A triangular membership function can be defined by a triplet
1 , 2 , 3  , as shown in Fig 5. The membership function for a triangular membership
function is defined as:
0, x  1

  x   1 1  x  2
 2  1 
(x)  
 3  x  2  x  3
    
 3 2
0, x  3
 x 
1.0
0 1 2 3 x
Fig. 5. Triangular Membership Function

In FMOTS, each of the input parameters has been defined by three fuzzy linguistic terms:
low, medium and high. The “medium” term was considered to be the mean value of the
data set and, therefore, is a single point value. Based on these concepts of the data
classification, membership functions were determined for employment density (Fig 6-a),
population density (Fig. 6-b) and private car ownership (Fig. 6-c). Applying a similar
method of data classification, membership functions were determined for the amounts of
road out of service area (Fig.6-d) and amount of roads within service area (Fig. 6-e) from GIS
analysis.
A triangular membership function is therefore used (Fig 6 a-e). The output parameter, on
the other hand, is consisted of five linguistic terms: very_low, low, medium, high and
very_high (Fig 6-f).
By creating a rule base for the FMOTS, the available data are used in addition to experts’
knowledge. The experts are specialized in traffic and transport engineering and city and
regional planning. By FMOTS, for each of the 5 input parameters, 3 linguistic terms are
defined, which makes a total number (35 = 243) rules. However, in the model, 189
meaningful rules have been used. In Table 1, a part of the rule base is seen.
Degree of membership, µ
1.0
0.8
0.6
Low Medium High
0.4
0.2
0.0
0 2500 4910
Employment density (emp-km2), people/km2
a)
1.0
0.8
0.6
Low Medium High
0.4
0.2
0.0
0 15000 36000
Population density (pop-km2), people/km2
b)
1.0
0.8
0.6
Low Medium High
0.4
0.2
0.0
0 50 95
Private car ownership (PCP), car/1000 people
c)
1.0
0.8
0.6
Low Medium High
0.4
0.2
0.0
0 25 50
Amount of roads out of service area (ROSA), km
d)
1.0
0.8
0.6
Low Medium High
0.4
0.2
0.0
0 3 5 8 11
Amount of roads within service area (RWSA), km
e)
Very Low Low Medium High Very High
1.0
0.8
0.6
0.4
0.2
0.0
0 1.5 2.5 3.5 5
Number of taxicab stands
f)
Fig. 6. Membership function of input parameters (a-e) and output parameter (f)
If Then
Private
Roads Roads
Car per
Outside Within
1000
Emp/km2 Pop/km2 Service Service FMOTS
Persons
Number Area (km) Area (km)
of the rule (ROSA) (RWSA)
(PCP)
1 low low low low medium very_low
2 low low high low medium medium
3 medium medium medium low medium medium
4 high high medium large low very_high
5 medium medium high high low medium
Table 1. A part of the rule base of the FMOTS
By developing FMOTS, a fuzzy logic expert system software has been used. In the model,
CoM defuzzification mechanism has been managed. No weightings were applied, which
means no rule was emphasized as more important than others. In this phase, a crisp value is
achieved by defuzzification of the fuzzy value, which is defined with respect to the
membership function. In other words, FMOTS defines the necessary number of taxicab
stands for each zone, with respect to the input parameters.
5.1 Findings and comparisons

In this study, some findings are reached. In Fig.7, some main findings of the study are
presented in 3D graphics. As seen in the figure, the parameter combinations have different
influences on FMOTS.
(a)
(b)
(c)
(d)
Fig. 7. Three dimensional representation of some selected variables on FMOTS: (a) RWSA–
ROSA relationship; (b) Emp/km2-Pop/km2 relationship; (c) RWSA-Pop/km2 relationship;
(d) ROSA-Emp/km2 relationship.
In Fig.7-a, with an increase of roads within accessed area and an increase of roads outside
accessed area, an increase in necessary taxicab stands is represented. In the researched area, in
the worst condition (ROSA=50 and RWSA= 0) the number of maximum taxicab stands is 4.
In Fig.7-b, population and employment potentials have a strong influence on the number of
taxicab stands. Increase of both of these inputs increase also the number of necessary taxicab
stands.
In Fig 7-c, increase of population density, together with a decrease in roads within service
area decrease the number of necessary taxicab stands in that zone. A low population density
together with a high level of roads within accessed area decrease the number of necessary
taxicab stands in that zone.
In Fig. 7-d, the relationship between the employment density and the amount of roads
without service area is demonstrated. In the model, if the amount of roads without service
area exceed 25 km, the need for taxicab stands increase in that area. For a population density
of 3000 people per km2 and for maximum amount of roads outside accessed, 4 taxicab
stands are needed.
6. Conclusions
In this study, an integrated model of GIS and fuzzy logic, FMOTS, for taxicab stand location
decision is used. The model is based on fuzzy logic approach, which uses GIS outputs as
inputs. The study brings some innovation by implementing fuzzy logic to a transport
problem for an area based analysis, different from the others that focus on networks and
queuing theory.
The model for location decisions of taxicab stands brings an alternative and flexible way of
thinking in the problem of a complex set of parameters by integrating a GIS study and fuzzy
logic.
FMOTS gives some results to evaluate the necessary number of taxicab stands in traffic zone
level, which are consistent with the observations of the planners.
The scale of the analyzed area requires specific methods to compute the data, such as the
areas which do not have adequate taxi service. GIS based service area analyses helped to
define the area, which can be accessed from taxicab stands within a psychologically
accepted time limit (and critical areas that are out of service range). The obtained data were
useful in fuzzy operations.
FMOTS addresses many of the inherent weaknesses of current systems by implementing: a)
fuzzy set membership as a method for representing the performance of decision alternatives
on evaluation criteria, b) fuzzy methods for both parameter weighting and capturing
geographic preferences, and c) a fuzzy object oriented spatial database for feature storage.
These make it possible to both store and represent query results more precisely. The end
result of all of these enhancements is to provide spatial decision makers with more
information so that their decisions will be more accurate.
The location decisions of taxicab stands in Konya have been examined by this integrated
model and their appropriateness has been evaluated according to the selected parameters.
Despite lack of useful data in some newly developed areas, consistent results could be
achieved in determining the necessary number of taxicab stands for city cells. The
consistency of the model can be increased if the infrastructure of the districts is improved
and some other necessary data are collected. In addition, definition of some other input
parameters would help development of the FMOTS model.
The proposed fuzzy logic model is indeed a decision support system for decision makers, who
wish to alleviate taxicab stand complexity in the short to medium term. Taxi and its control
systems can make use of the vague information derived from the natural environment that in
turn can be fed into expert systems and so provide accurate recommendations to taxi drivers,
the customers, motoring organizations and local authorities.
Broad future research can be suggested with the study. The model can be developed for a
decision support system for determining the necessary number of taxicabs assigned for each
of the taxicab stands in an urban area.
7. Acknowledgement
The authors are grateful to the GIS software company Basarsoft (Turkey) and Konya
Municipality for supply of the necessary data.
8. References
Alesheikh, A. A.,Soltani, M. J.,Nouri, N.,Khalilzadeh, M., Fal 2008. Land assessment for
flood spreading site selection using geospatial information system. International
Journal of Environmental Science and Technology (5) : 4, 455-462.
Chen, M.H., Ishii, H., Wu, C.X., May 2008. Transportation problems on a fuzzy network.
International Journal of Innovative Computing Information and Control (4): 5,
1105-1109.
Choi, K., Chung, Y.S., Jul-Dec 2002. A data fusion algorithm for estimating link travel time,
ITS Journal (7): 3-4, 235-260.
Galderisi, A., Ceudech, A., Pistucci, M., August 2008. A method for na-tech risk assessment
as supporting tool for land use planning mitigation strategies. Natural Hazards
(46): 2 ,221-241.
Ghatee, M., Hashemi, S.M., Apr 2009. Traffic assignment model with fuzzy level of travel
demand: An efficient algorithm based on quasi-Logit formulas. European Journal
of Operational Research (194):2, 432–451.
Ghayoumian, J, Saravi, M.M., Feiznia, S, Nouri, B., Malekian, A, April 2007. Application of
GIS techniques to determine areas most suitable for artificial groundwater recharge
in a coastal aquifer in southern Iran. Journal of Asian Earth Sciences (30):2,364-374.
Giuliani, R.W., Rose, J.B., Weinshall, I., June 2001. Taxi Stands in Times Square and the
Theater District A Technical Memorandum for the Midtown Manhattan Pedestrian
Network Development Project. Department of City Planning & Department of
Transportation, City of New York.
Horner, M.W., Murray, A.T., Sep 2004. Spatial Representation and Scale Impacts In Transit
Service Assessment. Environment and Planning B-Planning & Design 31 (5): 785-797.
Liu, Y.,Phinn, S. R., Nov 2003. Modelling urban development with cellular automata
incorporating fuzzy-set approaches. Computers, Environment and Urban Systems
(27): 6, 637-658.
Malins, D, Metternicht, G., March 2006. Assessing the spatial extent of dryland salinity
through fuzzy modelling. Ecological Modelling (193): 3-4,387-411.
Meng, Z., Meng, L., Sep 2007. An iterative road-matching approach for the integration of
postal data. Computers, Environment and Urban Systems (31):5, 597-615.
Murat Y.S., Baskan,O., Jul 2006. Modeling vehicle delays at signalized junctions: Artificial
neural networks approach. Journal of Scientific & Industrial Research (65): 7, 558-564.
Murat, Y.S., Gedizlioglu, E., Feb 2005. A fuzzy logic multi-phased signal control model for
isolated junctions.Transportation Research Part C-Emerging Technologies (13):1,
19-36.
Murat, Y.S., Gedizlioglu, E., May 2007. Investigation of vehicle time headways in Turkey.
Proceedings of The Institution of Civil Engineers-Transport (160): 2, 73-78.
Narayanan, R., Kumar, K., Udayakumar, R., Subbaraj, P., 2003. Quantification of congestion
using Fuzzy Logic and Network Analysis using GIS. Map India Conference 2003,
New Delhi.
Nobre, R.C.M., Rotunno, O.C., Mansur, W.J., Nobre, M. M. M.,Cosenza, C. A. N., Dec 2007.
Groundwater vulnerability and risk mapping using GIS, modeling and a fuzzy
logic tool. Journal of Contaminant Hydrology (94): 3-4, 277-292.
Ocalir, E.V., Ercoskun, O.Y., Tur, R., 2010. An Integrated Model of GIS and Fuzzy Logic
(FMOTS) for Location Decisions of Taxicab Stands, Expert Systems with
Applications (37):7, 4892-4901.
O’Neill, W.A., 1995. A Comparison of Accuracy and Cost of Attribute Data In Analysis of
Transit Service Areas Using GIS. Journal of Advanced Transportation 29 (3): 299-320.
Petrík, S., Madarász, L., Ádám, N ,Vokorokos,L., 2003. Application of Shortest Path
Algorithm to GIS Using Fuzzy Logic. 4th International Symposium of Hungarian
Researchers on Computational Intelligence, November 13-14, Budapest, Hungary.
Puente, M.C.R., Diego, I.F., Maria, J.JO.S., Hernando, M.A.P., Hernaez, P.F.D.A.,2007. The
Development of a New Methodology Based on GIS and Fuzzy Logic to Locate
Sustainable Industrial Areas. 10th AGILE International Conference on Geographic
Information Science, Aalborg University, Denmark.
Rashed, T., Sep 2008. Remote sensing of within-class change in urban neighborhood
structures. Computers Environment and Urban Systems (32): 5, 343-354.
Stefanakis, E., Vazirgiannis, M., Sellis, T., 1996. Incorporating Fuzzy Logic Methodologies
into GIS Operations, Int. Conf. on Geographical Information Systems in Urban,
Regional and Environmental Planning, Samos, Greece, pp. 61—68.
Steiner, R., 2001. Fuzzy Logic in GIS,
http://www.geog.ubc.ca/courses/geog570/talks_2001/fuzzy.htm.
Strobl, R.O., Robillard, P.D., Debels, P., June 2007. Critical sampling points methodology:
Case studies of geographically diverse watersheds. Environmental Monitoring and
Assessment (129): 1-3,115-131.
TUIK (2008) Address Based Population Census Reults, Ankara.
TUIK (2009) Haber Bülteni, Motorlu Kara Taşıtları İstatistikleri, Sayı 67, Şubat 2009, Ankara.
Upchurch, C., Kuby, M., Zoldak, M., Barranda, A., March 2004.Using GIS to generate
mutually exclusive service areas linking travel on and off a network. Journal of
Transport Geography 12(1) :23-33.
Walsh S.J., Page P.H., Gesler W.M., June 1997. Normative Models and Healthcare Planning:
Network-Based Simulations Within A Geographic Information System
Environment. Health Services Research 32 (2): 243-260.
Wanek, D., 2003. Fuzzy Spatial Analysis Techniques in a Business GIS Environment. ERSA
2003 Congress, University of Jyväskylä – Finland.
Yao, X. Thill, J.C., July 2006. Spatial queries with qualitative locations in spatial information
systems. Computers, Environment and Urban Systems (30):4, 485-502.
Zadeh, L.A., 1965. Fuzzysets. Information and Control, Vol. 8, pp. 338-352.
17
Sensor Network and GeoSimulation: Keystones

for Spatial Decision Support Systems
Nafaâ Jabeur1, Nabil Sahli1 and Hedi Haddad2
1Dhofar University
2 Laval University
1Oman
2Canada
1. Introduction
Natural hazards and man-made disasters are victimizing large numbers of people and
causing significant social and economical losses. By developing efficient Spatial Decision
Support Systems (SDSS), managers will be efficiently assisted in identifying and managing
impending hazards. This goal could not be reached without addressing significant
challenges, including data collection, management, discovery, translation, integration,
visualization, and communication. As an emergent technology, sensor networks have
proven efficiency in providing geoinformation for any decision support system particularly
those aiming to manage hazardous events. Thanks to their spatially distributed nature, these
networks could be largely deployed to collect, analyze, and communicate valuable in-situ
spatial information in a timely fashion. Since some decisions are expected to be taken on-
the-fly, the right data must be collected by the right set of sensors at the right time. In
addition to saving the limited resources of sensor networks, this will speed up the usability
of data especially if this data is provided in the right format. In order to boost the decision
support process, a thorough understanding and use of the semantics of available and
collected heterogeneous data will obviously help to determine what data to use and how
confident one can be in the results ultimately. An appropriate representation of the geo-
information should enhance this process.
Data collected by sensors is often associated with spatial information, which makes it
voluminous and difficult to assimilate by human being. In critical situations, the hazard
manager has to work under pressure. Coping with such collected data is a demanding task
and may increase the risk of human error. In this context, Geosimulation emerges as an
efficient tool. Indeed, mapping the collected data into a simulated environment which takes
into account the spatial dimension may dramatically help the hazard manager to easily
visualize the correlation between data collected by sensors and the geospatial constraints.
In this chapter, we first present fundamental concepts of SDSS and the most important
challenges related to their development. Second, we outline the sensor network technology
as an emergent tool for leveraging SDSS. Third, we present the Geosimulation approach as
another keystone to enhance SDSS. In this part, we summarize the current opportunities,
research challenges, and potential benefits of this technique in SDSS. Finally, for better
efficiency, we propose an encoding that emphasizes the semantics of available data and
tracks events/effects propagation. Based upon conceptual graphs, this encoding will be
used to increase the benefits from sensor networks and geosimulation in SDSS.
2. Spatial decision support systems

Literature about spatial decision support systems (SDSS) is abundant and covering all their
facets is beyond the scope of this chapter. In this section we only introduce the main
characteristics of these systems and some important challenges related to their development.
2.1 Definition and importance of SDSS

The concept of spatial decision support systems has emerged by the end of 1980’s from the
integration of two technologies: decision support systems (DSS) and geographic information
systems (GIS) (Sugumaran and Sugumaran, 2005). Beyond the broad definition that SDSS
are DSS designed to support decision-makers solving complex spatial problems, in the
literature there is no consensus about what SDSS exactly are. Keenan (Keenan, 2003)
attributed the disagreement on the definition of SDSS to the fact that DSS themselves are not
clearly defined. Consequently, in the literature SDSS are often defined by their common
characteristics rather than by what they exactly are. For example, Goel (Goel, 1999) listed the
following common traits characterizing SDSS: they are designed to solve ill-structured
problems, they have user interfaces, they combine models and data and they contain tools
allowing users to explore and evaluate candidate solutions. They are supposed to provide
an iterative problem-solving environment. However, we only develop the following main
characteristics:
Class of problems
Malczewski (Malczewski, 1999) defined a SDSS as “an interactive computer based system
designed to support a user or group of users in achieving a higher effectiveness of decision-
making while solving a semi-structured spatial decision problem”. Spatial decision
problems are often ill- or semi-structured. They are multidimensional, characterised by
uncertainty (which is inherent to spatial data), have decision objectives and factors that can
not be fully and formally specified and do not have a unique evident solution (Densham,
1991; Gao et al., 2004; Ademiluyi and Otun, 2009). Consequently, spatial decision-making
processes are usually based on the “what-if” analysis: the decision-maker defines the
problem, collects and analyses data and evaluates the consequences of several alternatives
(scenarios) in order to select the best one (Ademiluyi and Otun, 2009). Moreover, spatial
decisions require the implication of several stakeholders; they are iterative, interactive and
participative processes (Densham and Goodchild, 1989; Goel, 1999). In addition, several
studies have demonstrated that human beings have several cognitive limits when reasoning
about spatial data (Kahneman and Tversky, 1982). For all of these reasons, spatial decisions
are inherently complex and it is very helpful to have SDSS that support them.
Main components
Decision support systems are often developed based on the “DDM paradigm”, i.e., a DSS
should provide capabilities related to dialog, data and modeling (Sprague and Watson, 1996).
SDSS are complex systems that integrate multidisciplinary technologies, but as an extension
of DSS, they should have architectural components that balance among those three
capabilities (Ademiluyi and Otun, 2009). Although the number and exact description of
Sensor Network and GeoSimulation: Keystones for Spatial Decision Support Systems 339
components mentioned in the SDSS literature differs (see (Sugumaran and Degroote, 2010)
for more details), a SDSS should at least integrate the following capabilities (Densham, 1991;
Densham and Goodchild, 1989; Keenan, 2003):
1. Spatial and non spatial data management capabilities (SDBMS component): include
tools supporting the collection and storage of data, the translation between different
data models and data structures, and retrieving data from storage.
2. Analytical and spatial modeling capabilities (Modeling and analysis component):
provide decision-makers with access to a variety of spatial models in order to help them
in the decision-making process. Examples of models include statistical, mathematical,
multi-criteria evaluation, cellular automata and agent-based models (Sugumaran and
Degroote, 2010).
3. Spatial display and report generation capabilities (User interface component): support
the visualization of the data sets and the output of models using several formats such as
maps, graphics and tables.
Domain specific
Like DSS, SDSS are often domain- and problem specific, although there are some SDSS that
are designed to be generic (Sugumaran and Degroote, 2010). SDSS have been applied to
several domains, such as urban planning, environment and natural resource management,
transportation and business, which led to the use of several spatial modeling techniques and
technologies. Reusing modeling techniques is not always evident, and the development of
SDSS is often complex, expensive and requires the acquisition of specific domain or problem
knowledge.
2.2 SDSS vs. GIS

Although GIS offer spatial data management and analysis capabilities that are attributes of
SDSS, they are usually not considered as SDSS (Sugumaran and Degroote, 2010) The major
reasons are that GIS lack of analytical modeling capabilities and techniques for effective
scenarios evaluation (Segrera et al., 2003). In addition, Malczewski (Malczewski, 1999)
pointed out that GIS do not usually have tools for presenting and evaluating choices with
conflicting multi-criteria and goals. These deficiencies limit the effectiveness of GIS in
supporting complex spatial decision problems (Densham, 1991). SDSS are thus usually
developed based on GIS software coupled with additional modeling and analysis
techniques such as multi-criteria decision analysis (MCDA), artificial intelligence, agent-
based modeling and simulation techniques.
2.3 Challenges for SDSS

Although SDSS have reached the maturity stage and have been successfully implemented in
several domains, there are still several challenges to be addressed. Sugumaran and Degroote
(Sugumaran and Degroote, 2010) identified technical, technological, social and educational
challenges, but in the following we only focus on some important challenges:
1. As we previously mentioned, an important capability of any SDSS is the presentation of
output and scenario alternatives to users. Evers (Evers, 2007) listed the lack of
capabilities for evaluating alternatives as a major reason for the inefficacity of SDSS.
Uran and Janssen (Uran and Janssen, 2003) indicated that an SDSS must allow the user
creating and testing alternatives using intuitive and simple techniques for end users. In
addition, it is very important for the decision-maker to be provided with the relevant
information, and any lack or over-load could potentially limit her ability to make good
informed decisions. Providing the relevant and meaningful information is an important
challenge. Most used techniques are visual (maps, 3D visualisations, etc.) or
quantitative (tables, graphs, etc.). There is currently a lack of qualitative techniques that
allow the decision-maker exploring the consequences of her alternative solutions with a
simple and meaningful language.
2. Developing SDSS for environmental monitoring and crisis situations requires not only
the capability of modeling dynamic spatial phenomena, but also the ability to update
those models with real-time data. Moreover, the system must provide an output that is
relevant and easy to understand to the non-technical decision-maker, because any
misleading or ambiguity can have to catastrophic consequences. However, GIS
traditionally do not handle dynamic spatial models, and in order to remedy this limit,
they have been coupled with modeling techniques from several research fields,
especially cellular automata and agent-based simulations. Although this has led to the
emergence of Dynamic GIS (Albrecht, 2007), dynamic spatial knowledge representation
and modeling is an active research field (Haddad and Moulin, 2010a), and the lack of
standards and efficient representation formalisms makes challenging the development
of SDSS for real-time or near-real time spatial problems.
3. Decision-makers often need to examine various situations simultaneously at different
levels of detail (macro-, meso- and micro- scales of representations). This is an
important issue since the modeled phenomena and observed patterns may be different
from one level of detail to another, and since interferences may arise between
phenomena developing in different levels of detail. The level of detail is very important
and complex problem, because it does not only require modeling the problem’s
dimensions at different spatial and temporal levels of details but also linking these
different levels of detail.
4. Most SDSS were developed independently of one another, and with the development of
internet and web-based SDSS (Sugumaran and Sugumaran, 2005), there is more and
more a need for sharing and accessing spatial data from several distributed and
heterogeneous sources. Interoperability is then an important challenge to be addressed.
The Open Geospatial Consortium (OGC) has been developing different interoperability
standards for web applications, including Web Map Service (WMS) and Web Feature
Service (WFS), and the issue of developing SDSS using these standards has been
recently raised (Zhang, 2010). In addition, semantic interoperability is an active research
field in geographic anthologies and semantic web (Wang et al., 2007), and the lack of
standards makes the development and deployment of ontology-driven SDSS difficult.
In the remaining of this chapter we focus on SDSS for dynamic environmental monitoring
and crisis response and we present an approach based on sensor networks and
geosimulation techniques to address their related challenges.
3. Sensor network-based SDSS

3.1 Survey on sensor network-based SDSS
A SDSS embodies geomatic principles for situating decision-making in space. It often uses a
GIS component to provide spatial analysis functionality (McCarthy et al., 2008). In a variety
of applications, updated spatio-temporal data on current conditions in the environment of
interest is decisive in making appropriate decisions. This data could be collected from the
field using existing techniques, such as sensors, satellites, sonar, and radars.
Sensors are small devices, deployed in an environment for the purpose of collecting data for
specific needs, such as monitoring temperature, pressure, moisture, motion, vibration, or
gas or tracking objects. Since the energy and processing constrained sensors are frequently
prone to failure, robbery, and destruction, they are unable to achieve their tasks
individually. For this reason, they are deployed within an extended infrastructure called a
sensor network. In this infrastructure, the spatially distributed sensors could implement
complex behaviors and reach their common goals through a collaborative effort. Thanks to
this collaboration, they are able to collect important spatio-temporal data for a variety of
applications. This quality motivates the link between SDSS and sensor networks, especially
when data is needed to be collected remotely and in real-time with tools which can operate
in harsh environments.
In the literature, several works have proposed sensor network-based SDSS. In the oil and
gas industry, Supervisory Control & Data Acquisition Systems (SCADA) are using sensors
in the monitoring of the processes within a total operating environment. The sensors are
connected throughout an oil refinery or pipeline network, each providing a continuous flow
of information and data about the operating processes. Chien and his colleagues (Chien et
al., 2010) have deployed an integrated space in-situ sensor network for monitoring volcanic
activity. In addition to ground sensors deployed to the Mount Saint Helens volcano, this
network includes a spacecraft for earth observation. The information collected with sensors
is integrated with data collected with an intelligent network of “spider” instrument
deployed to the surface of the Mount Saint Helens volcano. The resulting data are used by
hybrid manual reporting systems such as the Volcanic Ash Advisory Centers (VAAC) and
Air Force Weather Advisory System (AFWA). Within the same context, Song and his
colleagues (Song et al., 2008) have proposed OASIS. OASIS is a prototype system that aims
at providing scientists and decision-makers with a tool composed of a “smart” ground
sensor network integrated with “smart” space-borne remote sensing assets to enable prompt
assessments of rapidly evolving geophysical events in a volcanic environment. The system
constantly acquires and analyzes both geophysical and system operational data and makes
autonomous decisions and actions to optimize data collection based on scientific priorities
and network capabilities. The data collected by OASIS is also made available to a team of
scientists for interactive analysis in real-time.
O’Brien and his colleagues (O’Brien et al., 2009) have presented a distributed software
architecture enabling decision support in dynamic ad hoc sensor networks, which enables
rapid and robust implementation of an intelligent jobsite. The architecture includes three
layers: a layer for expressive approachable decision support application development, a
layer for expressive data processing, and a layer for efficient sensor communication.
According to the authors, the implemented prototype can speed up application
development by reducing the need for domain developers to have detailed knowledge
about device specifications. It can also increase the reusability of software and protocols
developed at all levels and make the architecture applicable to other industries than
construction.
Wicklet and Potter (Wicklet and Potter, 2009) have presented three steps for information-
gathering, from sensor data to decision support. These steps are: data validation, data
aggregation and abstraction, and information interpretation. Ling and his colleagues (Ling
et al., 2007) have proposed a sparse undersea sensor network decision support system based
on spatial and temporal random field. The system has been presented as suitable for
multiple targets detection and tracking. In this system, an optimization based random field
estimation method has been developed to characterize spatially distributed sensor reports
without making any assumptions on their underlying statistical distributions.
In (James et al., 2008), the authors have proposed an approach that aims at performing
information-centric analysis within a GIS-based decision support environment using expert
knowledge. By shifting towards a more information-centric approach to collect and use
sensor measurements, more advanced analysis techniques can be applied. These techniques
can make use of stored data about not only sensor measurements and what they represent,
but also about how they were collected and the spatial context related to them. In a related
work, Rozic (Rozic, 2006) has proposed REASON, which is a Spatial Decision Support
Framework that uses an ontology-based approach in order to interpret many different types
of data provided by sensor measurements. In such systems, binding and transforming the
sensor data into timely information which is relevant to the problem is a challenging task. In
order to solve this issue, we have proposed in a related work (Jabeur and Haddad, 2009) to
allow sensors to autonomously cooperate and coordinate their efforts while emphasising on
data semantics. By encoding causality relationships about natural phenomena and their
effects in time and space with the formalism of conceptual graphs, we have proposed an
approach that implements a progressive management of hazardous events. This approach
would provide the decision-makers with timely valuable information on hazardous events
of interest, such as floods and volcano eruption.
In (Tamayo et al., 2010), the authors have conducted the design and implementation of a
decision-support system for monitoring crops using wireless sensor networks. The
prototype implemented includes tools that provide real-time information about the crop
status, surrounding environment and potential risks such as pests and diseases.
In (Filippoupolitis and Gelenbe, 2009), the authors have proposed a distributed system that
computes the best evacuation routes in real-time, while a hazard is spreading inside a
building. The system includes a network of decision nodes and sensor nodes, positioned in
specific locations inside the building. The recommendations of the decision nodes are
computed in a distributed manner then communicated to evacuate or rescue people located
in the vicinity.
In order to support rain-fed agriculture in India, Jacques and his colleagues (Jacques et al.,
2007) have proposed COMMON-Sense Net system, which aims at providing farmers with
environment data. This system can provide farmers with valuable data allowing them to
improve the production of semiarid regions in a cluster of villages in Southern Karnataka.
Mastering a ship in heavy seas is always a challenge, especially in night time. The officer of
the watch has to “sense” the weather and waves in order to preserve the cargo and navigate
the ship without undue motions, stress or strains. SeaSense (Nielsen, 2004) is designed to
support the natural human senses in this respect by providing an accurate estimation of the
actual significant wave height, the wave period, and the wave direction. Furthermore, and
based on the measurements and estimations of sea state, SeaSense provides actual
information on the sea-keeping performance of the ship and offers decision support on how
to operate the ship within acceptable limits.
3.2 Opportunities and challenges in sensor network-based SDSS

Any SDSS should provide decision-makers with the right data, at the right time, from the
right spatial location. In addition, the SDSS must allow the decision-makers to have a global
view about what is happening in the application environment. As this environment may be
large-scale, harsh, and likely impossible for human to access, the requested data could only
be collected remotely. There are several existing techniques that can help in achieving this
task, such as radars, satellites, and sensors. Depending on the availability of resources and
the application context, one or more technologies can be deployed.
Sensor networks afford SDSSs with opportunities along five axes (Fig. 1):
1. For which goals data are collected: The sensor network should collect the requested
data depending on the goals of the application. For example, in a flood scenario, sensors
should collect data on the level and movement of water, the conditions of soil slope,
and water pollution. Moreover, sensors can deliver to the SDSS data in several formats
including those that are compatible with widely used standards.
2. When data will be collected: sensors should be activated at the right time to collect and
deliver the requested data to decision-makers through the SDSS.
3. Where data will be collected: sensors should collect data from the right location and
thus decision-makers could have updated information on locations of interest.
4. Who will collect data: sensors should collect the requested data depending on their
current locations. If they are not explicitly appointed by the decision-makers, they
should self-organize in order to identify the subset of sensors that will be in charge of
collecting the requested data.
5. How data will be collected: sensors generally follow sleep/wakeup cycles. They are
programmed to collect data according to predefined schedules. Depending on the
current situation, this schedule has to be adapted in order to collect data at appropriate
times. Moreover, data can be collected through a collaborative effort from several
sensors.
What are the

goals ?
How data
will be When data
collected ? needed?
SDSS
Where data
Who will
to be
collect data ?
collected ?
Fig. 1. Opportunities in sensor network-based SDSS

As sensors are commonly characterized with their limited processing capabilities and short-
battery lives, they are commonly prone to failure and have to save their energy by following
sleep/wakeup cycles. In addition, due to the characteristics of operating environments, the

sensors can be destructed, lost, and even stolen. Consequently, the topology of the sensor
network is frequently changing and thus pulling data from the network could be affected.
This affects the fluent feeding of the SDSS with the requested data. As such, this represents
the main challenge of binding SDSS with sensor networks.
Binding SDSS and sensor networks concerns several issues, including:
1. Communication: as sensors are generally spatially distributed and do not maintain
reliable communication pathways, the exchange of data between the SDSS and the
sensor network should not be straightforward. Thus, data should not be pulled out
appropriately from the network.
2. Data format: the Open Geospatial Consortium’s Sensor Web Enablement (Botts et al.,
2006) working group has created an entire suite of standards related to sensor
networks. In this context, SensorML could be used to describe sensors and sensor
platforms, O&M (an XML schema which complements SensorML) could be used to
describe measurements made by sensors, and SOS (Na and Priest, 2006) can make use
of SensorML and O&M to provide an interactive, useful view of current and historical
conditions at a site. In spite of such standards, some SDSSs may use proprietary data
formats that may cause some incompatibility with sensor data.
3. Synchronization: in some real-time applications, communication between the sensor
network and the SDSS must be in timely fashion. This could not be guaranteed all the
time, particularly due to sensor network constraints.
4. GeoSimulation-based SDSS for crisis response

Hardly a day passes without some form of crisis, disaster or tragedy hitting the headlines.
At the same time, new technologies have made great steps, which, in theory at least, have
been accomplished for the safety of human beings and the environment. Unfortunately,
many technological achievements are responsible for most of the disasters occurring today.
Paradoxically, we need technology to better deal with these disasters when they occur.
When problems are well-structured and relatively static (e.g., resource allocation),
mathematical programming methods can be built into the SDSS. For dynamic situations,
such as crisis management, simulation is a better tool (Fedra and Reitsma, 1990; Zhang and
Grant, 1993). In this section we first make an overview of SDSS used for crisis response with
a particular emphasis on those based on geo-simulation. We then enumerate the main
challenges and opportunities of geosimulation-based SDSS for crisis response.
4.1 Review
SDSSs have been successful in addressing the spatial component of natural and
technological hazards (Fedra, 1997). As a result, several specialized SDSS have been
developed that assist decision-makers with hazards. For instance, regarding wildfires, a
multidisciplinary system has been developed by Keramitsoglou and his colleagues
(Keramitsoglouet al., 2009) to provide rational and quantitative information based on the
site-specific circumstances and the possible consequences. The system's architecture consists
of several distinct supplementary modules of near real-time satellite monitoring and fire
forecast using an integrated framework of satellite Remote Sensing, GIS, and RDBMS
technologies equipped with interactive communication capabilities. The system may handle
multiple fire ignitions and support decisions regarding dispatching of utilities, equipment,
and personnel that would appropriately attack the fire front.
Garbolino and his colleagues (Garbolino et al., 2007) proposed Computer-Aided

Management of Emergency Operations (CAMEO) which is a SDSS for assessing the impact
of a gas release from a hazardous material transportation accident on the population in a
dense urbanized area. The SDSS uses a database that centralizes, in real time, the data on
hazardous material transportation coming from both embarked sensors monitoring the
physical conditions of the material in the tank of the truck, the truck route, its speed etc. and
an installed optical sensor placed in a highway station which detects orange plates of the
hazardous material trucks crossing the highway station. CAMEO is associated to the GIS
platform in order to integrate the databases on the stakes of the territory (population,
housing, companies, etc.) with the maps of the atmospheric dispersion distances of toxics
coming from the simulations computed with CAMEO. The overlapping of the maps of these
gas release distances and the data on the stakes allow the user to identify the exposed (or
impacted) stakes.
Many other SDSS were developed for droughts and agricultural pestilence (MacLean et al.,
2000; Agatsiva and Oroda, 2002), floods, hurricanes, wildfires, and earthquakes (Fulcher et
al., 1995; Radke, 1995; Sanders and Tabuchi, 2000; Jaber et al., 2001; Hodgson and Cutter,
2002; Jensen and Hodgson, 2006). Most of these SDSS were designed to help human
decision-makers to take the right decisions. However, more sophisticated SDSS can also
assist these decision-makers in planning interventions during crisis situations. These SDDS
are called geosimulation-based SDSS as they use geosimulation to support human planners.
To better understand these SDSS, we discuss the necessity of the geosimulation process in
what follows.
Human planning requires the simulation of plans, which is itself based on anticipation and
schematisation (Hoc, 1987). When facing a new situation, a human being is able to detect
similarities with well-known situations and to anticipate events or properties which are not
yet totally identified (Hoc, 1987) thanks to his sense of anticipation. According to Craik
(Craik, 1946), planning is a very refined activity implying a mental formulation of future
states of the world in order to simulate the real world behaviour and then to predict future
events. However, human planning capabilities are limited. In (Kahneman and Tversky,
1982), the authors claim that simulation is difficult for humans, which is proven by Forbus’
experiences (Borbus, 1981) which show that simulation cannot be done by humans, expect in
trivial cases.
Planning becomes more complex when addressing uncertain situations such as during crisis
response interventions. In such cases, accurate predictions about plan executions remain a
hard task for human planners. The AI community proposed solutions such as the
Simulation-Based Planning approach SBP which consists in associating planning and
simulation. Each generated plan is simulated in order to be tested and evaluated. The most
appropriate plan is kept. As we focus on crisis situations, the spatial component is crucial
which makes the problem even more complex.
In order to plan within a large-scale space, a person relies on a sophisticated mechanism
based on cognitive maps which are mental constructs that we use to understand and know
environments and use to make spatial decisions (Kitchin, 1994). Despite its sophistication, it
remains very limited as many errors, mainly metrical, can occur. Sources of these metrical
errors include distortion, distances, and directions. Consequently, and despite his exceptional
cognitive capabilities, a human being has several limitations when trying to plan in a real
large-scale geographic space. An SDSS based on a geosimulation (simulation using GIS
data) approach would help overcome some of these limitations. Simulation is even more
important for emergency planning. Indeed, when based on realistic assumptions, simulation
offers an efficient way of modeling crisis response tasks.
In the domain of forest firefighting for example, we developed a geosimulation-based SDSS
which relies on a four-layer architecture (Sahli and Moulin, 2009; Sahli and Jabeur, 2011). An
adaptation to this architecture is discussed later (see Subsection 4.3). In the air navigation
domain, (Ozan and Kauffmann, 2002) developed a practical tool for tactical and strategic
decision-making in aircraft-based meteorological data collection called TAMDAR
(Tropospheric Airborne Meteorological Data Reporting). This onboard system gathers
meteorological data (using airborne sensors) as aircraft fly in the troposphere and transmit
this data to the National Weather Service. Collected data will help decision-makers to
process different operational alternatives (by conducting various what-if analyses using a
GIS-based user interface) and to determine the best strategy for system operation. The SDSS
is composed of a customized simulation-optimization engine, a data utility estimator, a GIS-
based analysis layer, and a user interface. They can also conduct various what-if analyses
effectively by using GIS-based user interface.
In the domain of civilian evacuation, and as evacuees move, a specialized SDSS can
continuously track evacuees using their current spatial location coordinate. In this context,
Nisha de Silva designed a prototype of a geosimulation-based SDSS named CEMPS
(Configurable Evacuation Management and Planning Simulator) (Nisha de Silva, 2000). The
aim was to produce an interactive planning tool to produce simple scenarios wherein
emergency planners are able to watch a simulation proceed and provide limited interaction
to obtain information on the progress of the evacuation. CEMPS integrates a dynamic and
interactive evacuation simulation model with a GIS which defines the terrain, road network,
and related geographical elements such as the hazard source and shelters, as well as the
population to be evacuated.
4.2 Challenges and opportunities for GeoSimulation-based SDSS

In the following, we briefly enumerate the main challenges that geosimulation-based SDSS
is facing as well as the opportunities that it represents.
1. Simulation models are difficult to integrate in the SDSS: In many crisis domains, experts
have done a lot of effort on model development. The difficulty in developing the DSS is
not then a lack of available simulation models but rather making these models available
to decision-makers (Muller et al., 2003). These simulation models are usually used in
research labs without being integrated into the decision-making process. Muller and his
colleagues (Muller et al., 2003) explained this gap by the following reasons: data
requirements are usually only attained in a research setting; models are complex and
assumptions are not well understood by managers; and deriving model input
parameters is extremely time consuming and difficult.
2. Multi-disciplinary problem: To integrate simulation models in SDSS, one needs
expertise in database management system, geographic information systems, computer
operating systems, remote sensing and Internet searching for data gathering, graphics,
as well as specific domain knowledge. Unfortunately, only few
professionals/researchers have all these skills. Cross disciplinary research groups have
to be set and collaborate more in order to achieve better geosimulation-based SDSS.
3. Granularity: The accuracy of predicting entities (resources and actors within the
environment) behavior depends on whether the SDSS simulator uses a micro, meso, or
macro modeling approach. The choice among the three levels of granularity depends on
the trade-offs that must be made in order to maintain realistic computing power when
processing large amounts of data.
4. Validation of simulated models: A simulator is intended to mimic a real-world system
using actual data associated with that system. It uses assumptions that generally
simplify the complex behavior of the real-world. When the SDSS simulates a highly
dynamic environment as during crisis response, it is a big challenge to validate the
simulation models in use.
5. Integrating data reported by citizen: Many governments are implementing large-scale
distributed SDSS to incorporate data from a variety of emergency agencies in order to
produce real-time “common operational pictures” of crisis situations. This goal remains
a very hard task. In this context, social networks have the potential to transform
emergency response by providing SDSS with data collected by citizens. Individuals are
often in the best position to produce immediate, empirical and real-time observance of
events simply because they are there. Pictures or videos taken by citizens from a cell
phone can provide invaluable information. Integrating these data and media generated
by citizens and other non-state actors with the main SDSS can enhance governmental
response to crisis. This will not only imply reviewing the command and control
strategies and the communication infrastructure, but also opening new research
avenues on how to collect, filter, and integrate these new data into the SDSS.
5. Including a full Emergency Management Cycle: The emergency management
community has long recognized that society’s response to disaster events evolve
through time. The succession of emergency response stages after a disaster event,
known as the Emergency Management Cycle, includes a phase of response/rescue
operations, a later phase of recovery/reconstruction, and a stage focused on mitigation
and monitoring for future events (Cutter, 1993). Most of existing SDSSs only focus on
one stage. It would be a considerable improvement if SDSS can address all stages
involved in emergency response.
6. Web-based SDSS: As with other applications deployed via the web, the Internet based
SDSS provides advantages over traditional desktop applications. First, the application is
centrally located, simplifying distribution and maintenance. In addition, the Internet
based approach increases the user base by reducing costs of access to users. However,
Internet standards have some limitations for use in spatial applications, but new
software and plugins continue to be developed. Current applications offer map display,
but frequently fall short of providing comprehensive GIS functionality. Future
developments offer the possibility of a distributed SDSS that could connect with
datasets held at distant locations on the Internet.
4.3 Combining sensors and GeoSimulation-based SDSS

When facing a natural disaster or a crisis situation, decision-makers need to monitor the
situation and then plan interventions (evacuation, rescue, etc.) while taking into account the
constraints of the real word (real large-scale, dynamic, and uncertain environment).Sensors
are of course one of the possible means of monitoring the environment. However, a ‘deploy
and ignore’ approach of sensors is not appropriate in this context. In remote areas, where
human interventions cannot be provided appropriately, sensors must have reasonable
intelligence and autonomy to reconnoiter their surroundings, respond to changing
environmental conditions, cope with frequently changing communication pathways and

network topology, and carry out automated diagnosis and recovery.
Static sensors are used to monitor the environment and send sensed data to the SDSS, which
may use this data in different ways. For example it can map the data (usually after
treatment) on the geo-referenced map (Sahli and Moulin, 2009). A human user can visualise
the data before taking any decision. The sensed data can also be used to feed a simulated
model (within the SDSS) of the event occurring in the dynamic environment. We have
already followed this approach for the forest fire problem in (Sahli and Moulin, 2009) and
the train derailment problem (caused by rock falls) in (Sahli et al., 2008).
In this section we are going to focus on another type of sensors which is very useful during
crisis response but much more difficult to manage. These are mobile sensors (whether
autonomously or embedded in moving vehicles). They offer more flexibilities and
opportunities for data acquisition and tracking events of interest in a highly dynamic
environment. However, connectivity and coverage among the network components have to
be maintained. To this end, it is important to prevent mobile nodes to move freely,
anywhere and anytime, creating communication holes. Holes can also result from the
absence of communication connectivity (sleep, robbery, or shutdown of some sensors) as
well as from heavy network activities. The goal of maintaining a WSN without
communication holes cannot be always guaranteed due to simultaneous movements,
sleep/shutdown of some nodes, or jamming factors.
In the particular cases of natural disaster (forest fire, oil slick, flood, etc.) or crisis situations,
mobile sensors have to deal with the constraints of the real world (real large-scale
environment, dynamism, and uncertainty). The "real large-scale environment" constraint
implies that the geographical aspects of the terrain could affect the sensors move. For
example, moving a sensor in a forest in fire is not always possible as movements depend on
the vegetation and the natural obstacles in the surrounding geographic space. The
"uncertainty" and "dynamism" constraints imply that the plan of sensors relocation could
fail at any time. For instance, a sensor which was moving to a safe place during a forest fire,
could realize that fire has changed its direction and thus his destination is no longer safe.
Under the constraints cited above, the real environment is not necessary the best place to
plan sensors relocation for the following reasons:
1. The relocation process needs the collaboration of several sensors which implies a high
volume of exchanged messages between the physical sensors. Given the nature of the
environment (natural disaster) and the commonly limited resources of sensors, such
communication should be avoided.
2. The movement of each sensor strongly depends on different parameters, namely, the
spatial characteristics of the surrounding space (elevation, slope, etc.), the evolution of
the situation (e.g., fire progression in a forest), and the position of the other
resources/sensors. These parameters are partially or totally unknown for individual
sensors.
3. Individual sensors have only a partial vision of the situation. Each sensor is only aware
of its neighborhood. Thus it lacks of the global vision of the situation, which is in this
particular case necessary to find a global relocation strategy.
The artificial intelligence community has proposed solutions to similar problems through
the Simulation-Based Planning approach (SBP) (Lee and Fishwick, 1994) which consists of
associating planning and simulation. Each generated plan is simulated in order to be tested
and assessed. The most appropriate plan is kept. However, when applied to real world
problems, the current SBP approach does not suggest an efficient interaction between the
real world and the simulated environment. In addition, monolithic planning turned out to
be ineffective (Desjardins et al., 1999). Agent-based planning appeared then as a good
alternative. Combining both techniques (simulation-based planning and agent-based
planning) may then help to solve such problems. The SDSS which supports this idea should
be able to build a synchronous parallel between the real world and the simulated environment
(which is mainly the geo-referenced map) of the SDSS.
The sensor relocation application can be thought of as a layered architecture as illustrated in
Fig. 2. This design philosophy is inspired by the layered simulation model we proposed in
(Sahli and Moulin, 2009). The real sensors are of course located in the real world. Each of
these sensors has a representative in the simulated environment. This representative is a
software agent. Sensor relocation is thus planned by these agents in the SDSS and then
communicated to the real sensors. The SDSS should include four layers as shown in Fig 2. A
full description of these layers can be found in (Sahli and Jabeur, 2011).
Real Environment RE SDSS

Effective relocation Relocation planning
Resources (mobile sensors) Communication Agents
Environment dynamics Sensors Simulated Model
Real spatial data GIS Simulation platform
Fig. 2. Four-layer architecture of geosimulation based SDSS for sensors relocation

As the relocation planning takes place in the SDSS, the SDSS needs to be continuously
updated by the real environment to maintain the coherency and synchronization between
the two environments. In this context, the communication between the physical sensor and
its agent representative is crucial. As for communication within the SDSS, heavier message
traffic is to be exchanged between different representatives in order to: (i) inform each other
about corresponding sensors status; (ii) check the status of others' sensors; (iii) coordinate
relocation. Coordinating the relocation is not an individual decision. Indeed, a sensor
having the intension to move must inform and get the approval of the surrounding sensors
especially that the movement must be for the sake of achieving a common goal. Besides, all
moves should be aligned with a strategic action which may be dictated by a human
decision-maker.
An SDSS supporting these four layers of Fig. 2 offers the following advantages: (i) Most of
the interaction messages between agents (sensors' representatives) are taking place in the
SDSS (locally). Even a high traffic of exchanged messages will not really affect the
performance of the relocation process; (ii) Since the relocation planning will be conducted
by the representative agents in the SDSS, each agent will have access to all relevant data
(spatial data of the surrounding space from GIS, the last evolution of the situation from the
simulated environment, and the position of the other resources and sensors from the other
co-existing agents). More details about the internal architecture of these agents can be found
in (Sahli and Moulin, 2009); (iii) As all agents are in the SDSS, a human decision-maker can
have a global idea about the current situation. Therefore, he/she can take part in the
relocation planning process by dictating strategic actions.
Moreover, planning in a highly dynamic environment using a simulation-based approach

implies dealing with another problem: when should the simulation process be launched in
the SDSS and how the simulation process should be coupled with the effective relocation?
Early planning applications were based on the assumption that a planner’s environment is
stable and that it is possible to create a complete plan before executing it (Russel and
Norvig, 1995). Such an assumption is inappropriate when considering dynamically
changing environments. Instead, it is necessary to adopt a continual planning (CP) approach
in which the activities of planning and execution are interleaved and in which re-planning is
triggered by changes in the environment or when the executed actions abort (Ambros-
Ingerson and Steel, 1988). Nevertheless, deciding when and how to interleave planning and
execution is one of the most complex problems of CP. In this context, we have already
proposed a dedicated planning approach, named Anticipated Continual Planning (ACP) in
(Sahli and Moulin, 2009). Our ACP is based on the Continual Planning paradigm but
enhanced by a preventive re-planning step (anticipating and periodic) and by a corrective
one (following an unexpected event). In both cases, a global plan is elaborated in parallel
and by both software agents and human decision-makers. Our approach enhances the
interleaving process of the classical CP (agents are not forced to stop the planning process in
order to wait for the execution), increases the quality of the plan (the plan is periodically
updated and eventual problems are anticipated and taken into account before having
repercussion on the plan), and gives more guarantees about reaching a final solution (since a
global plan is always available and in which human decision-makers have taken part).
5. Towards better efficiency in sensor network-based SDSS

As we previously mentioned, geospatial semantics is currently an active research topic
which would resolve the problem of interoperability between different SDSS using
heterogeneous spatial data sources and formats. Geospatial semantics is also very valuable
for improving the efficiency of SDSS as this would help in providing the decision-makers
with the right data, in the right format, at the right time. In this section we argue this idea
and illustrate it with two of our previous works on sensor network-based SDSS and
geosimulation-based SDSS.
5.1 Semantic aspects for better efficiency in SDSS

Maximizing the benefits from a sensor network-based SDSS would require to feed the SDSS
with the right data, in the right format, from the right locations, with the right sensors, at the
right time. This could be reached by a thorough examination of data semantics. Data
semantics could help in delimiting the sensing areas and sensing resources (Jabeur and
Haddad, 2009; Jabeur et al., 2008). In addition to preserving the sensor network resources,
this would speed up getting data from the network.
Once data is pulled out from the sensor network, we can maximize the benefits from the
semantics of geo-information by finding effective methods to: (1) encode data for different
purposes, and (2) transform this data from one representation to another. The use of the
appropriate data representation at the right time would improve the analysis of the current
situation, pose the right queries to the network, increase data sharing, and reduce the
consumption of the limited resources (Jabeur and Haddad, 2009).
Several approaches can be used for data representation in sensor network while increasing
the benefits from the semantics of data. Semantic networks (Lehmann, 1992) and Frames
(Minsky, 1974) are examples of these approaches. The use of the XML and GML (Cox, 2006)
languages provides a standard and common ways of cataloguing and exchanging sensor
network information. The Open Geospatial Consortium (OGC) proposes the Observations
and Measurements (O&M) standard (Cox, 2006) that defines a conceptual schema and XML
encoding for observations and for features involved in sampling when making observations.
OGC also proposes the Sensor Observation Service (SOS) (Na and Priest, 2006) standard that
defines a web service interface for the discovery and retrieval of real time or archived data
produced by all kinds of sensors. By using GML, O&M documents can be generated in SOS
software. In this case, measurement records are stored to a database with a schema based on
the O&M specification. When this database is queried, the SOS server pulls the relevant
records from the database and automatically generates the O&M document in XML format
based on a template that is stored on the server (McCarthy, 2007).
5.2 Encoding data using conceptual graphs

Knowledge about the semantic of hazardous events and their behaviours are extremely
valuable for developing SDSS that can predict situations evolution and prevent dangerous
consequences before they occur. For example, rising water levels indicate that flooding may
be occurring, and the sudden motion of a slope can signify a slope failure. Due to complex
chain reactions of concurrent events, there is an urgent need to automate the sensor network
activities while preventing the waste of its limited resources. The representation of available
data in the appropriate format and the encoding of expert knowledge in a machine-usable
form serve to reinforce any decision support that a monitoring system may draw.
In (Jabeur and Haddad, 2009) authors proposed an approach that takes benefit from the
semantics of data on natural phenomena conceptualized as spatio-temporal events and
formalised using conceptual graphs (CGs). Based on Peirce’s existential graphs and
semantic networks of artificial intelligence, CGs were introduced by Sowa (Sowa, 1984) and
provide an extensible system of logic to capture, represent and reason with the semantic of
real-world knowledge. The approach proposed in (Jabeur and Haddad, 2009) uses causal
knowledge about hazardous events and their effects in time and space to implement a
progressive hazard monitoring. Hazardous events (events of interest, such as flood and rain
fall) are explicitly conceptualised as static states and dynamic changes (punctual and
durative) which are delimited in space and time (Haddad and Moulin, 2010a). The effects of
a given hazardous event are formalised using causality relationships that define semantic,
spatial and temporal causal constraints between cause and effect situations. These
constraints are derived from the fact that human recognition of causal relations is based
upon recognition of precedence and contiguity between the cause and the effect (Kitamura
et al., 1997). Knowledge about hazardous events, their effects and risk levels were
formalised using CGs and used to improve the efficiency of the sensor-network. When an
event of interest is triggered by the sensors, the system first identifies what are its possible
effects, their associated level risk, and where and when they may occur and then reserves
the relevant sensor network resources in order to collect data in right place and time (Jabeur
and Haddad, 2009).
In a related work (Mekni and Haddad, 2010) authors used semantic knowledge about events
and their spatio-temporal effects to develop an integrated knowledge-based multi-agent
geo-simulation framework for intelligent sensor web deployment. Figure 3 illustrates the
main components of the framework.
Fig. 3. Architecture of the knowledge-based multi-agent geo-simulation framework for

intelligent sensor web deployment (Mekni and Haddad, 2010)
Multi-Agent Geo-Simulation is used to simulate the behaviour of a sensor network in a
dynamic virtual geographic environment. Sensors are modeled as intelligent agents
embedded in an informed virtual space where dynamic phenomena can occur. Sensor
agents have reasoning capabilities allowing them to reason about the virtual space and to
react to its dynamic phenomena. They have perception ranges and can be active or sleeping.
Spatio-Temporal knowledge encoded in CGs is used for two main purposes. First, it is used
during the geosimulation to support sensor agents reasoning capabilities. In the beginning
of the simulation only some sensor agents are active. When an active sensor agent detects
the occurrence of a situation of interest (such as a storm) in its perception range, it asks the
knowledge base for its consequences, delimits when and where they may occur and then
wakes up the relevant sleeping sensors within certain distance. Once the situation of interest
is out the perception range of the sensor agent it switches to the sleeping mode. Second, the
Spatio-temporal knowledge is used to analyse the results of the geo-simulation and to offer
decision support to users (evaluation of different sensor deployment plans).
6. Conclusion
When considering any DSS, it is very important to know the degree of involvement of
human experts in the process of taking decisions. This issue is more crucial during crisis
response, especially when decisions are to be taken in real-time. As the spatial component
arises, the trust of decision-makers in SDSS needs a longer discussion. In this chapter, we
discussed briefly the problem from the cognitive point of view. A longer investigation
supported by a real example can be found in (Sahli and Moulin, 2006). The cognitive aspect
helps in understanding the limits of SDSS and pinpointing the issues where human
decision-makers could not be overlooked.
When dealing with geographic reasoning, which is typically based on incomplete information,
a human planner is able to draw quite accurate conclusions by cleverly completing the
information or applying certain default rules based on common sense. Despite these cognitive
capabilities, the planner has limitations when it comes to simulate complex events, particularly
in a real large-scale geographic space. This task could be achieved properly by an SDSS. While
the SDSS “knows” better the spatial environment as it uses GIS data, it does not have a refined
sense of anticipation and judgment as a human expert does. Even if a plan generated by the
SDSS seems to be well-grounded, it may not be feasible in reality or may go against certain
doctrines. Human experts can propose to adjust/change the SDSS' recommendations
according to their own experience and anticipation sense. For these reasons, we draw the
conclusion that geosimulation based SDSS complements human planning skills when
addressing complex problems such as crisis situations. It is worthwhile to investigate this
degree of complementarities and come up with some general guidelines that could be
customized depending on the application domain.
7. Acknowledgement
The research leading to these results has received Research Project Grant Funding from the
Research Council of the Sultanate of Oman Research Grant Agreement No [ORG DU ICT 10 003]
8. References
Agatsiva, J., & Oroda, A. (2002). Remote sensing and GIS in the development of a decision
support system for sustainable management of the drylands of eastern Africa: A
case of the Kenyan drylands, International Archives of the Photogrammetry, Remote
Sensing and Spatial Information Sciences, 34:42–49.
Albrecht, J. (2007). Dynamic GIS, in The Handbook of Geographic Information Science, ed. J.
Wilson and S. Fotheringham, London: Blackwell, pp. 436–446.
Ambros-Ingerson J.A. Steel, S. (1988). Integrating planning, execution and monitoring.” In Proc.
of the 7th (US) National Conference on Artificial Intelligence (AAAI 88), St Paul, MN,
USA. Americain Association for Artificial Intelligence, Morgan Kaufmann, pp. 83-88
Botts, M., Percivall, G., Reed, C., Davidson, J. (2006). OGC Sensor Web Enablement:
Overview and High Level Architecture. Open Geospatial Consortium Whitepaper.
Open Geospatial Consortium.
Cárdenas Tamayo, R.A; Lugo Ibarra, M.G. and Garcia Macias, J.A. (2010): Better crop
management with decision support systems based on wireless sensor networks,
Electrical Engineering Computing Science and Automatic Control (CCE), 2010 7th
International Conference on, 2010, 412 – 417, Tuxtla Gutierrez
Chien, S.; Tran, D.; Doubleday, J.; Davies, A.; Kedar, S.; WEbb, F.; Rabideau, G.; Mandl, D.;
Frye, S.; Song, W.; Shirazi, B. and Lahusen, R. (2010). A Multi-agent Space, In-situ
Volcano Sensorweb, International Symposium on Space Artificial Intelligence,
Robotics, and Automation for Space (i-SAIRAS 2010). Sapporo, Japan. August 2010
Cox, S. (2006). Observations and measurements. OpenGIS Discussion Paper. Open
Geospatial Consortium (2006)
Craik, K.J.W. (1943). The Nature of Explanation. London: Cambridge. University Press, 1943.
Cutter, S. (1993). Living with Risk: The Geography of Technological Hazards, London: Edward Arnold.
Densham, P. J., Goodchild, M. F. (1989). Spatial Decision Support System: A Research
Agenda. Proceedings of GIS/LIS ’89, Florida, pp. 707-716.
Densham, P.J. (1991). Spatial decision support systems. In: D.J. Maguire, M.F. Goodchild,
D.W. Rhind (eds.) Geographical information systems: principles and applications,
pp. 403-412. London: Longman.
Desjardins, M.; Durfee, E.; Ortiz, C. and Wolverton, M. (1999). A survey of research in
distributed continual planning. AI Mag. vol. 4, 1999, pp. 13-22.
Evers, M. (2007). Requirements for Decision Support in Integrated River Basin Management,
Journal for Management of Environmental Quality, Vol. 19(1), 2007
Fedra, K. & Reitsma, R.F. (1990). Decision Support and Geographical Information Systems.
In Geographical Information Systems for Urban and Regional Planning, Research Report:
RR-90-009. Laxenburg, Austria: ACA, International Institute for Applied Systems
Analysis (IIASA).
Fedra, K. (1997). Integrated risk assessment and management: Overview and state-of-the-
art, Journal of Hazardous Materials, 61:5–22.
Filippoupolitis, A. & Gelenbe, E. (2009). A distributed decision support system for building
evacuation. In Proceedings of the 2nd conference on Human System Interactions
(HSI'09). IEEE Press, Piscataway, NJ, USA, 320-327
Forbus, K.D. (1981). A Study of Qualitative and Geometric Knowledge in Reasoning about
Motion, MIT Artificial Intelligence Laboratory, Research Report AI-TR-615, 1981.
Fulcher, C.; Prato, T.; Vance, S.; Zhou, Y. and Barnett, C. (1995). Flood impact decision
support system for St. Charles, Missouri, USA, 1995 ESRI International User
Conference, 22–26 May, Palm Springs, California, pp. 11.
Gao, S., D. Sundaram, and J. Paynter. (2004). Flexible support for spatial decision making.
Paper presented at the 37th Annual Hawaii International Conference on System
Sciences, Honolulu, Hawaii.
Garbolino, V.; Sacile, R.; Olampi, S.; Bersani, C.; Alexandre, N.; Trasforini, E.; Benza, M. and
Giglio, D. (2007). A Spatial Decision Support System prototype for assessing road
HAZMAT accident impacts on the population in a dense urban area: a case study
of the city of Nice, French Riviera, Proceedings of the eight International
Conference on Chemical & Process Engineering, Ischia, 24-27 June 2007.
Goel, R. K. 1999. Suggested framework (along with prototype) for realizing spatial decision
support systems (SDSS). Paper presented at Map India 1999 Natural Resources
Information System Conference, New Delhi, India.
Haddad, H. & Moulin, B. (2007). Using Cognitive Archetypes and Conceptual Graphs to
Model Dynamic Phenomena in Spatial Environments, In: ICCS 2007, LNAI 4604, U.
Priss; S. Polovina & R. Hill (Eds.), pp. 69–82, Springer-Verlag, Berlin Heidelberg
Haddad, H. & Moulin, B. (2010a). A Framework to Support Qualitative Reasoning about
COAs in a Dynamic Spatial Environment. International Journal of Experimental &
Theoretical Artificial Intelligence , Vol. 22(4), pp. 341 – 380, Taylor & Francis
Haddad, H. & Moulin, B. (2010b). Multi-Agent Geo-Simulation in Support to Qualitative
Spatio-Temporal Reasoning, Modeling Simulation and Optimization - Focus on
Applications, Shkelzen Cakaj (Ed.), pp. 159-184, ISBN: 978-953-307-055-1, INTECH
Hoc J. M. (1987). Cognitive Psychology of Planning. Academic press, 1987.
Jaber, A.; Guarnieri, F. and Wybo, J.L. (2001). Intelligent software agents for forest fire
prevention and fighting, Safety Science, 39, pp. 3–7.
Jabeur, N. & Haddad, H. (2009). Using Causality Relationships for a Progressive
Management of Hazardous Phenomena with Sensor Networks. In proceedings of
the International Conference on Computational Science and Applications (ICCSA
2009), June 29-July 2, 2009, Yongin, Korea, pp: 1-16
Jabeur, N.; McCarthy, J.D. and Graniero, P. (2008). Improving Wireless Sensor Network
Efficiency and Adaptability through an SOS Server Agent. In proceeding of the 1st
IEEE International workshop on the Applications of Digital Information and Web
Technologies (ICADIWT 2008). 4-6 August, Ostrava, Czech Republic, pp: 409-414.
Jensen, J.R.; Hodgson, M.E.; M. G-Q Quijano, J. Im.(2009). A Remote Sensing and
GIS-assisted Spatial Decision Support System for Hazardous Waste
Site Monitoring, Photogrammetric Engineering and Remote Sensing, 75(2):

169-177.
Kahneman, D. & Tversky, A. (1982). The simulation heuristic. In Kahneman, Slovic &
Tversky (Eds.) Judgement under uncertainty: Heuristics and biases. New York:
Cambridge University Press, 1982.
Keenan, P. B. (2003). Spatial Decision Support Systems, in M. Mora, G. Forgionne, and J. N.
D. Gupta (Eds.), Decision Making Support Systems: Achievements and challenges for the
New Decade: Idea Group, pp. 28-39.
Keramitsoglou, I.; Kiranoudis, C. T.; Sarimveis, H. and Sifakis, N. (2004). A multidisciplinary
decision support system for forest fire crisis management. Environmental
management, vol 33, No 2, pp 212-225
Kitamura, Y.; Ikeda, M. and Mizoguchi, R. (1997). A Causal Time Ontology for Qualitative
Reasoning. In: The Fifteenth International Joint Conference on Artificial
Intelligence IJCAI 1997, pp. 501–506 (1997)
Kitchin, R. M. (1994). Cognitive Maps: What Are They and Why Study Them?, Journal of
Environmental Psychology, 1994, pp. 14:119.
Lee, J.J. & Fishwick, P.A. (1994). Real-Time Simulation-Based Planning for Computer
Generated Force Simulation, Simulation, vol. 63, n° 5, 1994, pp. 299-315.
Lehmann, F. (1992). Semantic networks. Computers and Mathematics with Application, 23
(2-5), 1-50, (1992)
Ling, B.; Zeifman, M.; Traweek, M. and Wettergren, T. (2007). A Sparse Undersea Sensor
Network Decision Support System Based on Spatial and Temporal Random Field,
Unattended Ground, Sea, and Air Sensor Technologies and Applications IX. Edited
by Carapezza, Edward M.. Proceedings of the SPIE, Volume 6562, pp. 65620P (2007).
MacLean, D.A.; MacKinnon, W.E.; Porter, K.B.; Beaton, K.P.; Cormier, G. and Morehouse, S.
(2000). Use of forest inventory and monitoring data in the spruce budworm
decision support system, Computers and Electronics in Agriculture, 28:101–118.
Malczewski, J. 1999. GIS and multicriteria decision analysis. New York: John Wiley & Sons, Inc.
McCarthy, J.D. (2007). Representing spatial and domain knowledge within a spatial decision
support framework. M.Sc. Thesis, University of Windsor (2007)
McCarthy, J.D.; Graniero, P.A. and Rozic, S.M. (2008). An integrated GIS-expert system
framework for live hazard monitoring and detection. Sensors 8: 830-846.
Mekni, M. and Haddad, H. (2010). A Knowledge-Based Multi-Agent Geo-Simulation
Framework: Application to Intelligent Sensor Web Deployment, Proceedings of the
fourth International Conference on Sensor Technologies and Applications, SENSORCOMM
2010, July 18-25, 2010 – Venice / Mestre, Italy, pp.329-335, ISBN: 978-0-7695-4096-2
Miller, R.; Guertin, D.P. and Heilman, P. (2003). An Internet-based Spatial Decision Support
System for Rangeland Watershed Management.” First Interagency Conference on
Research in the Watersheds: October 27-30, 2003. U.S. Dept. of Agriculture,
Agricultural Research Service, p. 725-730.
Minsky, M. (1974). A framework for representing knowledge. Cambridge, MA, USA:
Massachusetts Institute of Technology, (1974)
Na, A., Priest, M. (2006). OpenGIS Sensor Observation Service Implementation Specification.
Open Geospatial Consortium
Nisha de Silva, F. (2000). Challenges in Designing Spatial Decision Support Systems for
Evacuation Planning, Natural Hazards Research paper, 2000
O’Brien, W.; Julien, C.; Kabadayi, S; Luo, X. and Hammer, J. (2009). An Architecture for
Decision Support in ad hoc Sensor Netwoks, Information Technology in Construction,
Vol. 14, 2009, pg. 309
Ozan, Erol & Kauffmann, Paul (2002). A Simulation-based Spatial Decision Support System
for a New Airborne Weather Data Acquisition System, 2002 Industrial Engineering
Research Conference, May 19-22, 2002, Orlando, Florida
Radke, J. (1995). A spatial decision support system for urban/wildland interface fire
hazards, 1995 ESRI International User Conference, 22–26 May, Palm Springs,
California, pp. 14.
Rozic, S. M. (2006). Representing spatial and domain knowledge within a spatial decision
support framework. M.Sc. Thesis, University of Windsor (2006)
Russel, S. & Norvig P. (1995). Artificial Intelligence, a Modern Approach, Prentice Hall, 1995.
Sahli, N. & Jabeur, N. (2011). Agent-Based Approach to Plan Sensors Relocation in a Virtual
Geographic Environment” in Proceedings of NTMS 2011, Paris, France, 7-10
February 2011.
Sahli, N. & Moulin, B. (2006). Agent-based geo-simulation to support human planning and
spatial cognition (extended version). J.S. Sichman and L. Antunes (Eds.): MABS
2005, Springer-Verlag Berlin Heidelberg LNAI 3891, pp. 115 – 132, 2006.
Sahli, N. & Moulin, B. (2009). EKEMAS, an agent-based geo-simulation framework to
support continual planning in the real-word” [Quick Edit] Applied Intelligence,
Vol. 31, No. 2. (1 October 2009), pp. 188-209.
Sahli, N.; Mekni, M. and Moulin, B. (2008). A Multi-Geosimulation Approach for the
Identification of Risky Areas for Trains, Workshop Agents in Traffic and
Transportation, part of the International Conference AAMAS 2008, Portugal, 2008.
Sanders, R., & Tabuchi, S. (2000). Decision support system for flood risk analysis for the
River Thames, United Kingdom, Photogrammetric Engineering & Remote Sensing,
66(11), pp. 1185–1193.
Segrera, S.; Ponce-Hernandez, R. and Arcia, J. (2003). Evolution of decision support system
architectures: Applications for land planning and management in Cuba. Journal of
Computer Science & Technology 3(1), pp. 40–46.
Sowa, J.F. (1984). Conceptual Structures: Information Processing in Mind and Machine.
Addison-Wesley, Massachusetts (1984)
Sprague, R. H. and H. J. Watson (1996). Decision support for management. Upper Saddle River,
N.J.: Prentice Hall
Sugumaran, R. & Degroote, J. (2010). Spatial Decision Support Systems: Principles and
Practices, CRC Press; 1 edition
Sugumaran, V., and Sugumaran, R. 2005. Web-based Spatial Decision Support Systems
(WebSDSS): Evolution, Architecture, and Challenges, Third Annual SIGDSS Pre-
ICIS Workshop, Designing Complex Decision Support: Discovery and Presentation of
Information and Knowledge, Las Vegas, Nevada (USA).
Uran, O. & Janssen, R. (2003). Why are spatial decision support systems not used? Some
experiences from the Netherlands. Computers, Environment and Urban Systems 27(5),
pp. 511–526
Wang, Y, Gong, J. and Wu, X. (2007). Geospatial semantic interoperability based on
ontology, GEO-SPATIAL INFORMATION SCIENCE, Vol 10(3), pp. 204-207
Wicklet, G. & Potter, S. (2009). Information-gathering: From sensor data to decision support
in three simple steps, Intelligent Decision Technologies, Vol. 3(1), 2009, pp. 3-17
Zhang, C. (2010). Develop a Spatial Decision Support System based on Service-Oriented
Architecture, Decision Support Systems, Chiang S. Jao (Ed.), ISBN 978-953-7619-
64-0, pp. 406, January 2010, INTECH, Croatia
Zhang, M. & Grant, P.W. (1993). Pages 107-114 in Proceedings of the Thirteenth Annual ESRI User
Conference, May 1993. Redlands, California: Environmental Systems Research Institute.
Part 5
Applications in Risk and Crisis Management

18
Emerging Applications of Decision Support

Systems (DSS) in Crisis Management
Gabriela Prelipcean1 and Mircea Boscoianu2
1Stefan cel Mare University of Suceava
2Henri Coanda Air Force Academy of Brasov
Romania
1. Introduction
The recent financial crisis, the growth of the frequency of extreme risk events are
motivations for a new analysis of the mechanisms and processes that implies high risk and
high uncertainty. The new dynamics and the severe impact of the extreme events (natural
hazards, terrorism, technological accidents, and economic/financial crises), but also the
complexity of interventions have motivated the scientific community to find new efficient
solutions for decision making process dedicated for crisis management, especially for
solving the following aspects: time urgency, the complexity of event, the volatility/ rapidly
changing of event/ decision conditions, the chaotic surrounding environment ingredient,
the human behaviour in critical situations (emotional stress), the consequences for decision
failure, poor data, frequent interruption during the decision making process.
Extreme risk events are no longer only high impact - low probability events. Indeed, the
literature demonstrates that these types of events are more frequent and also their impact is
also more critical. The governments are responsible not only for the post crisis management
of extreme events. In this case, better knowledge could offer opportunities to reduce the
impact.
Emerging aspects regarding the integration of the concept of hybrid DSS in the treatment of
extreme risk management will be presented. The efficiency of using intelligent ingredients
in decision support systems (DSS) is linked to the human limits but also on the dynamics of
huge data in the context of imprecision and uncertainty. Different applications for the use of
hybrid DSS in flood and drought risk management, asymmetric risk and crisis management
exercises will be also presented.
Many extreme risk events are associated with modern human environment and in this case
the spectrum of risks is changed with a difference between the perceived possibility and
reality (Renn, 1992). Extreme risk is expressed by the potential for the realization of
unwanted, adverse consequences to human life, health, property, or the environment. In
Kaplan, Garrick (1981) is proposed the risk triplet (S-scenario, P-likelihood, D-possible
consequence), a framework that responds to the nature of disaster events that can occur,
how likely is a particular event, what are the consequences.
Decision makers in extreme risk environments should respond in a new different manner
because modern crisis management requires urgent developments toward better, more
elaborated and appropriate means for extreme risks. Extreme risk management needs to
ensure a better interoperability of different emergency services (police, fire chief, health
sector, civil protection) to provide the appropriate information (of course, after data fusion
and data filtering) at the right place in the critical moments. The new global environment is
very complex and the dynamics of changing is difficult to understand. Decision making in
critical or special situations is very complex because the systems are complex, the dynamics
is difficult to understand, and adaptability is essential. Even the technologies to cope with
the crisis and high risk events have developed considerably there are some underlying
problems that complicate high risk prevention and multiple crisis response: an inadequate
communication between different actors and different levels; the relative inadequate data
fusion, selection, filtering and standardization impacted information database; the difficulty
to update information about the development of the extreme risk (victims damages, rescue
team technologies, in the case of natural/man-made hazards, or specific information in the
case of financial crushes and crises); the access to existing databases and action plans it is
relatively slow.
The interest is to develop an integrated framework capable to support emergency decisions
via a better understanding of the dynamics of extreme events and a better detection and
management of new risk situation. In this case, the focus is on the characteristics of
genericity and adaptability of the framework in order to build a flexible, adaptive, robust
and efficient system that does not depend on a particular case and it is easy to be extended
in a creative manner.
In an uncertain and highly dynamic environment this type of applications could offer a
robust but adaptive support for all decision making factors. Based on this type of
applications is possible to build a generalized framework capable to support different types
of decisions, not only in economy and finance, but also in military and law enforcement
applications, in critical periods, or in high risk events.
2. An introduction in extreme risk modelling

First step in management of extreme risk events is to identify the most critical endpoints
assessed in terms of the possible impacts on humans, communities, and environment. Then,
it is necessary to specify the particular nature of each risk that could affect the endpoints in
both terms of the likelihood and the degree of damage. The exploration of possible external
impacts which might occur is based on the following aspects related to the risk triplet
R=R(Si, Pi, Di): the impacts/damage position outside/inside our control area; the scale of
these impacts and their degree of irreversibility; the trans-boundary characteristics of these
impacts; the potential threat to human ethics/morality.
The typical risk scenarios require an integrated perspective in selecting the adequate mix of
risk reduction measures such as increased natural hazards damage potential due to urban
development and agglomeration; increased systemic risk due to cascaded facilities or
organizations; a severe decline in preparedness of local communities that wait for the
government action (via public warning systems) and a low ability to minimize losses; a
critical environmental degradation.
An extreme risk framework could be build considering the following processes: the problem
formulation (potential risk events examination); risk assessment (provides an objective and
integrated judgment in terms of scientific evaluations regarding extreme risk events
identification, pathways, exposure assessment, exposure-damage response assessment); risk
management and first subjective decision making process to select regulatory measures
Emerging Applications of Decision Support Systems (DSS) in Crisis Management 361
from alternative options. There exist conceptual/functional separations between risk

assessment and risk management. The explicit representations of risk scenarios can provide
a better assessment and permits regulatory decisions under volatility and uncertainty. In the
literature is underlined the critical role of risk communication and stakeholders
participation in extreme risk analysis.
Another branch of the literature is focused on the relationships between hazards and
damages in terms of risk or vulnerability (Wisner, 2004). In the “pressure and release” (PAR)
model the aim is to understand risk in more realistic terms of vulnerability in the field of
disaster sciences based on the disaster risk model of Alexander (1993) expressed by:
Risk = Hazard x Vulnerability (1)

In Smith (2001), the risk is modelled in a similar way:
Risk = Hazard (Probability) x Loss (Damage)/Preparedness (Resilience) (2)

In Wisner (2004) "risk" is viewed as "hazard time’s vulnerability" or a unit measure of the
possible damages under an exposure probability to the hazards. The overwhelming
tendency in both theory and practice is to view extreme risk management as a holistic
process involving prevention, planning, acute response, recovery, and learning (Comfort,
1988; Nudell, Antokol, 1988; Coombs, 1999; Fink, 2002; Regester, Larkin, 2002; Curtin,
Hayman and Husein, 2005). Preventing all extreme threats from materializing is not only
implausible; it is simply impossible (Wildavsky, 1988). Prevention requires that one knows
the source and dynamics of threats, but the literature shows this is impossible (Reason, 1997;
Pauchant, Mitroff, 1992; Anheier, 1999; Gauld, Goldfinch, 2006). The decision maker will
take into account the following operational steps: size up the event situation; activate the
DSS; add the real time information and select the type of event scenario; adapt the base
event scenario; simulate and evaluate the final scenario and the recommended decision;
implement decision.
In principle, an integrated framework for extreme risk analysis should include the following
four steps: scenario formulation - collection and analysis of data related to hazards in terms
of their possible origins, pathways, and mitigation; extreme risk assessment – the list of
potential extreme events together with their exposure or vulnerability; extreme risk
management - with development of mitigation measures and procedures based on the
output from the risk characterization; communication by using a dedicated platform to
enable a better understanding of the rationale behind the categories of risk assessment.
In the modern literature are presented a lot of applications, procedures and activities
capable to anticipate, prepare for, prevent, reduce different types of risks/losses associated
to different type of crises, but there are only few integrated frameworks to deal directly with
the extreme event. In this case, the decision makers need a huge technical assistance to
support decision making process before, during, and after extreme events.
All these applications are based on a huge quantity of data, a dynamic or real time selection
between alternatives and the implementation of the final solution should be define with
precision, but also with adaptability and creativity. Other aspects are related to the
possibilities to integrate and to mix different types of models, methods and
techniques/decision tools and to understand the limits of acceptability. DSS is one of the
most efficient technologies in the treatment of extreme risk and crisis, because it offers
adaptability, robustness and is easy to use in a modular-adaptive framework. Soft
computing techniques could be used together withy DSS/ IDSS not only as a mathematical
ingredient, because its efficiency is given by a better capability of selection, a higher speed of
analysis, and also on adding the advantages of adaptability, flexibility, and modularity.
3. Emerging techniques to understand extreme risk management (ERM)

Extreme risks are high impact, high uncertainty, and low probability emergency events with
high negative impact/losses due to uncertainty spread to different levels; in the case of
financial events appear also the contagion phenomenon. Extreme events demand immediate
action because of serious threats to the environment; contagion brings a new dynamics in
recent financial events. The urgent need for action is to respond hints that the time interval is
very short. Managers should act in time, but with surgical precision because crisis represents
critical turning points with decisive changes. Extreme risk and multiple crises could be
describes as the manifestation of an unexpected risk that develops very quickly in to an
emergency/disaster situation. An emergency/disaster situation contains also surprise and
response elements. Emergency response represents the synthesis of knowledge based on
experience, procedures and activities to anticipate, prepare for, prevent, reduce or overcome
all the risks associated with extreme events. The interest is to offer framework easy to use, but
efficient in reducing the negative consequences. A complex system supports a variety of key
decision across a wide range of decision making and integrates two processes: how to
recognize and size up a situation and how to evaluate the course of action (CoA).
The steps supported by an intelligent system are: size up the situation using the embedded
information; search a knowledge base and recognize the first type of similar situations;
diagnose the historical experience against the problem; adapt and modify the solution of the
historical data base to work in the new context; rehearse the solution to verify that it is likely
to realistically work; the implement of decision.
The literature reveals that decision makers do not normally use the classic rational choice
techniques, but they rely on experiences; it is not the intention to find the mathematical best
way, but to find a first good enough solution implemented and move to the next step (this
natural style is called recognition-primed decision making, RPDM). Modern decision
makers face specific gaps: some decision makers are afraid to apply RPDM based on past
experiences, but others do not have a personal knowledge base of past experiences enough
to cover unfamiliar extreme events. A simple framework capable to rapidly support the
decision maker with experience for the past or from others scenarios and to guide the
decision maker to rapid near optimal decision, is useful for small scale events, but an
intelligent tool is needed for complex systems.
The decision support system (DSS) concept, launched before PCs, was focused on the use of
interactive computing in unstructured/quasi-structured decision making activities. Sprague
(1980) argued this limitation for solving only unstructured problems and proposed an
extension that includes every system. Druzdel and Flynn (2002) define DSS as an interactive
system based on computer, capable to assist the user in selection activities; in this concept
DSS provide data based management and refined the conventional access to information
and the capability of find function via a support in building rational models. McNurlin and
Sprague (2004) described DSS as “computer based systems that help decision-makers
confront ill-structured problems through direct interaction with data and analysis models”.
The decision support system (DSS) concept, related to an interactive, flexible, and adaptable
system developed for an intelligent decision making support, offers a better information
management for a better coordination of activities. DSS are very efficient instruments in
complex situations/complicated environments, where decision makers need a robust
support to analyse multiple sources of knowledge (Martinsons, Davison, 2007). DSS was
used with success in management and the evolution was always linked to the dynamics of
informatics systems, databases and expert systems. The new IT application has decisive
influence DSS, with application that spread in all domains of activities. Modern DSS are able
to support operational capability and strategic decision making in complex systems. A
decision making framework for extreme risk should consider a structure based on
quantifying decision variables decoupled in their own control systems. These variables are
combined through the practical conditions offered by knowledge based infrastructure to
include all decision scenarios. Because each decision scenarios affect the system in a
different way, these effects can be also used to rank decisions and, in this way, to offer a
better adaptability of the global framework. The main objectives of DSS are related to a
better adaptability of decision making and the build of a preliminary study for decision
making in technical case where is not possible an efficient planning of such activities.
The characteristic elements of a dedicated DSS are: addressing unusual problems and assist
in improving the quality of decisions to resolve them; DSS is a productivity enhancement
tool for decision-making activity of the expert who holds an active control system;
construction and development cause an evolutionary DSS - makers, system developers that
are influencing each other in a process that does not end at a specified time; DSS has a data
integrator and is adapted to the particularities of the individual application and user,
limited to just one method or information technology; DSS may have several stages of
completion, from the core of system to application systems; DSS can be addressed according
to its stage of development and more users (decision makers, analysts, and developers of
tools).
DSS is a technology capable to improve the ability of managers in decision making to better
understand the risks and the dynamics of the extreme events within the constraints of
cognitive, time, and budget/liquidity limits. In principle, DSS design is based on the
following three capabilities: a sophisticated database management tool with real time access
to internal and external data, information, knowledge; powerful modelling capabilities
accessed by the portfolio of models used by the management; friendly, but powerful
interface designs that enable interactive queries, reporting and interface graphical user.
Since a crisis begins with events, first step in decision making process is to categorize/ rank
the events and propose a first selection of the proper models. Let a knowledge base with N
different cases corresponding to N different events, each having its own model, and a
situation vector filled with environmental and costs parameters. Based on situation vectors,
some cases are selected as possible candidate events that are associated to the portfolio of
models; the models get the necessary information from the corresponding database (online
data entered in the infrastructure through human user or automated systems and offline
data collected for corresponding similar situations/ events). Models are equipped with
controllers that compute the performances of decision variables. It results scenarios for
solving the problem with different performances (time, costs). The interest is to find the
optimal solution that offers a few numbers of ranked possible scenarios, with their effect
degree and the decision maker can select one of them. It is useful to introduce a system
capable to gets the decision maker idea about the scenarios, to refine the decision for the
future same cases and to allocate the higher degrees.
The purposes of a DSS (Fig.1) is to offer a better support for decision makers, allowing better
intelligence design, or selection via a better understanding of the whole problem and its
dynamics, a better managing knowledge via efficient solutions for non structured decisions.
The main characteristics of a special purpose DSS are the following:
- a structured knowledge that describes specific aspects of the decision makers
environment, how to accomplish different tasks;
- it incorporates the ability to acquire and maintain a “complete” knowledge and its
dynamics;
- it incorporates the ability to select any desired subset of stored knowledge and to
present the global knowledge in different customized ways/reports;
- it offers a direct interaction with the user (decision maker) with adaptability and
creative flexibility to the situation and users.
Other Web,
computer- communication
based systems
Data
bases
Data Model External

management management models
Intelligent knowledge
based ingredients
User interfaces
Crisis decision maker
Organizational Decision implementation

knowledge based
Decision
evaluation/performances
Fig. 1. The decision making process in crisis management

DSS-ERM should offer an intensive level of use in decision making process, both in the case
of a crisis and before/after the crisis. This ability is based on the dynamic analysis of the
current situation and „similar” past situation. The output of using this instrument could be
represented by direct effects (a better decision capability, better efficiency, better objectivity
in decision making process, less errors and ambiguities in communication, a good stimulus
to adopt excellence and the new style of work, a better use of creativity and innovation) and
indirect effects (creation of new skills and new jobs, more efficient, better competitive, better
adaptability of the structure to critical situations).
4. Principles to construct an integrated DSS for extreme risk/ crisis

management (DSS - ER/CM)
The system should be built and implemented in a flexible and evolutionary philosophy. It
should be considered good knowledge of different applications that implies extreme risks, a
good identification of the task linked to the capabilities of decision makers, a capability to
refresh data/knowledge bases, a good knowledge and selection of adequate methods for the
design. The basic aspects regarding the building of DSS-ER/ CM architecture is presented in
Fig. 2. In this case, the system is modular and should be evolved. The flexibility will add
good capability to modify in an efficient manner and the better adaptability to the treatment
of new risks/events.
Decision makers that deal with crisis situations need intelligent instruments like software
tools capable to deal with the questions related to the global perspective thinking, the
response to unpredictability of human reaction under stress, and the impact of
communication systems failure. In crisis management, the interest is to integrate different
types of risk, the interactions between these risks at different levels to provide a dynamic
complete picture that take into account all possible hazards and the phases of related
planning including mitigation, preparedness, response, recovery after extreme event with
the aim to express a structured solution. Automation, networking, systems integration and
intelligent decision support improve the performance of complex decision, such typical for
crisis management. The use of vague concepts is important in the context of uncertainty/
imprecise information and artificial methods (knowledge bases, fuzzy logic, multi-agent
systems, natural language, genetic algorithms, and neural networks) could develop
emerging capabilities that mimic human characteristics (approximate reasoning, intuition,
and just plain common-sense). The main features of a dedicated DSS for crisis management
are: real time data, efficient response, a user friendly interface to support decision makers
that work in difficult conditions, a good quality of information, data recovery capabilities.
In the concept of Intelligent Decision Support Systems (IDSS) (Gadomski, 1998; Guerlain,
Brown, 2000; Turban, 2004) are developed effective smart systems for problem solving and
decision-making (Turban, 2004; Dahr, Stein, 1997) that deal with complex, imprecise and
non-structured situations. IDSS are dynamic because they develop and implement more
effective and productive support systems. The need for IDSS arrives from the growing need
for relevant/effective DSS to deal with a dynamic, uncertain, complex management
environment, the need to build context-tailored, not general purpose systems, the increased
acceptance that intelligent technologies can improve decision quality and work
productivity. In order to obtain an optimum solution, this type of DSS should be real-time,
distributed, robust, and fault tolerant. The assistance process is provided by taking into
account the elements or subsystems of high uncertain/ vague information in a changing
dynamic and stochastic scenarios. The main generic capabilities are related to the
independence from specific database management systems, the data fusion capability, the
level of integration of resources management, the accuracy of the graphical interface, the
way to detailed in time information, the use of post-action verification procedures and an
efficient access to the technical documentation and history.
The critical need for the integration of intelligent ingredients in DSS is linked to the human
decision makers limits: cognitive, economic, time and competitive demands. Both data
management and knowledge management (facts and inference rules used for reasoning) are
very important. The decision maker do needs a robust framework capable to capture
imprecision, uncertainty, learn from the data/information and continuously optimize the
solution by providing interpretable decision rules (Sousa, 2007).
Computing; Communication; Maps;

Graphic; Reports; Display/ Visualization
DSS Tool Estimate extreme event

Model and data Assessment parameters:
base selection - Type of extreme events
- Magnitude
- Location and area
- Global recognized picture
Suggest new
(synthesis)
instruments and tools
Monitoring, survey
Measurements
Studying knowledge
base and database
events
Fig. 2. Basic steps in building a DSS-ER/CM architecture

Fuzzy logic (FL)/fuzzy inference systems (FIS) applied to DSS provide formal methodology
to capture valid patterns of reasoning about uncertainty. Artificial neural networks (ANN)
use the capabilities of rule extraction from a trained network positions. Evolutionary
Computing (EC) is a powerful global optimization instrument based on the simulation of
evolution by iterative generation and alteration processes operating on a set of candidate
solutions (Abraham, 2002). Due to the complementarities of these instruments (ANN, FIS,
EC), the interest is to integrate various type of intelligent instruments to form a synergic
integrated framework system.
ANN is defined by the architecture, the connection strength between neurons, node
properties, learning rules and the objective function of the ANN represents the complete
status of the network. After initialization, the learning process adapts the weights to changes
for handling imprecise information. The training patterns can be thought of as a set of input
pattern - output pattern pairs. The generalization represents the capability of the network to
provide a correct matching in the form of output data for a set of previously unseen input
data. In the Conjugate Gradient Algorithm (CGA) the search is proceeds along the conjugate
gradient direction to determine the step size, which optimize the performance function
along the line. The next search direction is determined so that it is conjugate to the previous
search direction. The interest is to combine the new steepest descent direction with the
previous search direction.
FIS are based on the concepts of fuzzy set theory, fuzzy if-then rules, and fuzzy reasoning.
The architecture of FIS contains a rule base, which contains a selection of fuzzy rules; a
database, which defines the membership functions used in the fuzzy rule; a reasoning
mechanism, which performs the inference procedure. FIS employ different inference
models: Mamdani-Assilian, with the rule consequence defined by fuzzy sets, Takagi-
Sugeno, with an inference in which the conclusion of a fuzzy rule is constituted by a
weighted linear combination of the crisp inputs.
EAs are population-based adaptive methods for optimization, based on the principles of
natural selection and survival-of-the-fittest. This dynamics of populations could be
expressed by:
x[t +1] = s(v(x[t])) (3)

Fuzzy controllers use a model of the expert knowledge that specifies the most important
properties of the process. Adaptation of FIS using evolutionary computations has been
widely explored in the literature (Abraham, 2000).
The contingency management tool (CMT) is a distributed system based on fuzzy logic;
knowledge based systems and distributed systems concepts that provide decision support
for a global picture of critical situations. The advantages of using an intelligent decision
support tool for contingency management, as the CMT, are: reduction of the global risk, an
overall increase of efficiency and reliability, a global view of the recognized picture based on
a better quality of information, DSS for human resources in real-time in critical situations,
DSS for training and experimentation. The use of soft computing techniques is welcome
because it results a better capability of selection, a higher speed of analysis, and also on
adding the advantages of adaptability, flexibility, modularity.
Fuzzy set theory (Zadeh, 1965) is a generalization of the conventional set theory that
provides a strict mathematical framework to deal with the uncertainty inherent to
phenomena whose information is vague/ imprecise and allows its study with some
precision and accuracy. Fuzzy logic allows expressing knowledge with linguistic concepts
(Ross, 2004; Zimmermann, 1996) and provides a good way to express imprecision that is
inherent in the vagueness of such concepts (Jackson, 1999; Ross, 2004). Expert Systems
(Turban, 2004) proceeds knowledge intensive tasks to perform inference for determining a
priority list of which subsystems should be fixed and in what sequence. In crisis
management it is difficult to decide which systems should be fixed first. Expert Systems
allow different reasoning processes such as the fuzzy multi-criteria. The advantage of
using cooperation supported by communication networks enables the separation of data,
the transparent access, the sharing of computational power, decentralized decision
processes, thereby enabling the increase of robustness, redundancy and efficient resource
usage (Sousa, 2006).
There are a lot of efficient applications of CMT: DSS for management of actions in crisis
context, DSS for management of equipment repair priorities under disaster/emergency
situations, DSS for in time advice on the selection of resources for increase reliability and
prevention of failures/ incidents, human resource training for contingency situations. Any
critical facility subject to extreme events is a possible candidate for using the CMT. The
application for risk management responds to a hierarchical and distributed decision making
process that offers a specific infrastructure (libraries, knowledge bases, databases, and
inference engine) and a set of generic templates that reduce the time to market a customized
application. CMT collects and compiles input information on the status of the subsystems
and the dynamics of risks (Sousa, 2006).
The analysis of the performances of a DSS is very important because it offers the possibility to
compare, to select, to update, to improve the knowledge (models, database), but also the time
to response and the capability to adapt to rapidly changing conditions. The main performance
parameters for an integrated DSS-ER/CM are focused on the following aspects:
- employ a naturalistic decision model (allow decision maker to characterize event size
up information as a starting point; support user in recognizing a similar problem;
support user in analyzing recognized case; support the user in customized the selected
analogous problem to better fit the real time conditions; support the user in
implementing the final decision as an efficient CoA);
- employ case based reasoning to take advantages of expertise and historical database
experiences (implement an intelligent capability; implement a knowledge base of
observation; allow flexible/ adaptable observation knowledge base modules; allow
revision of observation);
- trigger recognition with partial characterization of an event is limited (allow fuzzy
input; recognize an event with limited information; allow a mixture of known/
unknown information; fusion capabilities);
- use system parameters that end user is likely to have available (operate on a standard
laptop/ PDA; the use of an Internet connection);
- establish a central network based capability for training and experimenting (establish an
web based, game oriented version for multiple applications; implement a training mode;
implement a capability to improve the performances to predict decisions from each web
based training session; capture web based gaming statistics; implement a capability to
improve the performance to predict decisions from each web based gaming session).
In Fig. 3 are presented a simplified view regarding the performances of a dedicated DSS for
crisis management.
For a better efficiency, DSS - ER/CM should operate on a standard IT infrastructure
(windows based PDA/ laptop, Internet connection) and will employ a user friendly GUI
(Graphical User Interface). Once the application is downloaded (with known expertise
condition), direct human interface with the system will be limited to interaction during
performance. Output data are provided in typical reports (both for training and
experimentation) designed by the user, according to the user preferences.
Maintenance will be provided through automatically updates accessible for different type of
end users. Real time diagnostics should be performed by experts to ensure the properly
operation, but no support equipment will be required (updates ensure that the performance
is optimal via a permanent adaptation).
5. Applications regarding the use of DSS in extreme risk management

5.1 Applications of DSS in flood risk management
Flood risk is the most frequent extreme risk event especially in the last decades. Modelling
flood dynamics could be expressed by a warning system as a technical way to reduce flood
risks, but also to manage adverse situations. The potential benefit of a flood forecast
depends on the accuracy/precision of data/ models, the efficiency of users to build decision
based on the large spectrum of information for monitoring and warning, the efficiency of
operation for protection and evacuation. The situation could change the thinking of risk
manager and real time information is not sufficient because a huge quantity of data should
be fused, filtered and proceeded.
Extreme events scenario:

- natural hazard
- assimetric extreme
events
- economic and
financial crisis
Models of damage, disruption, and

restoration/recovery
Type 1 Type 2 ........ Type n
Economic Human
Economic Human and
data Environment
model ands ocial social
data
model data
Economic Social impacts

impact
DSS for selection

Performance the model of
objectives resilience
Resilience
outcomes
Fig. 3. The performances of dedicated DSS-ER/CM

The classic literature on prediction of flooded areas was focused mainly on deterministic
models. A new branch of literature uses soft computing (especially artificial neural networks
- ANN) to simulate the critical level or the dynamics of flood in a real-time monitoring
framework. ANN is an empirical modelling instrument that has an ability to identify highly
complex relationships input - output in a black box type model which does not require
detailed information about the system. Nevertheless it learns about the relationship between
input parameters and controlled and uncontrolled variables by studying previously
recorded data. The advantages of ANN lie in its ability to handle huge quantity of data
from nonlinear dynamical systems and, especially, when relationships are not fully
understood.
The advantage of DSS in modelling flood risk management is given by its interactive nature,
flexibility in approach and also on the capability to provide graphic interfaces. It also allows
the identification of the critical training/testing of the model by simulating different input
data. The use of DSS is efficient in the estimation of the critical level of flood warning. Flood
warning system (FWS) based on DSS can immediately inform the people living downstream
to take precautions and evacuation/rescue before the event. The FWS developed by the
Khetchaturat and Kanbei is based on the following steps: the selection of areas that flood
frequently occurred; prototype system design and data modelling flood phenomenon; the
development of an early warning network between the web server and users in the area via
the Internet; the development of DSS for flooding. In this framework, the authors had
obtained good results.
DSS-ANN for flood crisis management is represented by a parallel, distributed processing
structure composed of three layers: an input layer for data collection, an output layer used
to produce an adequate response to entry, and intermediate layers capable to act as a
collection of features detected. Network topology consists of a set of neurons connected by
links and organized in layers. Each node in a layer receives and processes weighted input
from the previous layer and sends its output to nodes in the next layer through links. For
each link is assigned a weight that represents an estimator of the force connection. Weighted
summation of inputs is converted to output according to the transfer function.
DSS-ANN for flood warning is efficient in the monitoring process of a flood risk zone,
meaning that the orderly evacuation can occur before the actual onset of flooding and
require long term sustainability. Integrating human knowledge with modelling tools lead to
the development of DSS-ANN to assist decision makers during the different phases of flood
management. DSS-ANN is developed as a planning tool to address both virtual and
specialists, but also all those who have different positions in the management of flood risk
management and also contributes to the selection of the options to reduce damages (using
an expert system approach), forecasting floods (using ANN), and modelling operations
according to the impact of flooding.
5.2 Applications of DSS in drought risk management

Drought is another extreme event with a major impact on economic, social, and
environment. In this case, management is only a reactive one. DSS tools should address
multiple risks related and, also, build the global picture of vulnerable areas to effectively
manage this type of events. In the literature are presented different frameworks based on
DSS instruments to help farmers to assess this type of risk. In the classic architecture built on
four levels (knowledge, information, data, and presentation), experts use the knowledge
base as a tool to express the interpretations of the problem and permits the interaction with
users in a simple way. Knowledge base used to develop specific knowledge domain
includes sequential techniques, data mining, modelling and simulation techniques, and
other innovative knowledge-based models.
DSS based on geo-spatial information provides real-time drought assessment, in an easily
accessible online interface. The algorithms based on data mining permits an effective
identification of the correlation between global climate events and local drought conditions.
Methods based on knowledge discovery in DSS are based on association rules that identify
relationships between the influenced parameters. Association rules are easy to understand;
experts in the field can use this joint decision making process. This framework enables each
level to make application for any lower level using an open standard interface. For example,
the database knowledge permits the construction of a drought index. If data base are
limited, data layer can retrieve data from relational databases to distributed space and
standard queries on the data layer to give a response to the request higher.
Web-based DSS is another interesting instrument for drought risk management. It provides
information for farmers, experts, end users, capable to improve the efficiency by allowing
resources to better manage this type of risks.
5.3 Applications of DSS in asymmetric extreme risk management

Asymmetric threats refer to weapons/tactics used to circumvent the technological
superiority and include the use of surprise or the use of weapons/tactics in an
unplanned/unexpected manner. Asymmetry is present on each side: open information on
the potential targets/victims, versus highly secretive, compartmentalized behaviour of
terrorists. Asymmetric attacks exploit vulnerabilities outside of the accepted laws arm
conflicts. Terrorism is the main asymmetric threat and represents a calculated use of
violence to destroy human life and critical infrastructure/key resources or/and inoculate
fear for intimidation societies. The specific characteristics of terrorist threat are related to the
key role of uncertainty, limited deterrence, and the private - public good nature for
providing security. A key feature is represented by the generalized, diffused nature of it
(where, when and how terrorist may strike) also linked to the degree of uncertainty. If the
authorities had advance knowledge of the timing and location of a future asymmetric attack,
actually thwarting it would be a relatively minor affair, involving the deployment of limited
police/military power. Antiterrorism represents the defensive measures to reduce the
vulnerability of people and property, while counter-terrorism represents the offensive
response to prevent, deter, and interdict terrorist tasks.
Operational trends are continuously changing. The technological evolution increase the
knowledge of the operational theatre and the speed of response, but a huge load of
uncertain data and information generated by environment still remains a problem. DSS is a
computerized system capable to complement human decision in the context of human
limitations. Situation awareness (SA) is essential for decision makers and the improvement
of decisional process enhances also the performance of SA via the concept of the right
information, at the right place, at the right time (Endsley, 2005). In the literature, there are
presented solutions on contextualized user centric task oriented knowledge services (web
and portal techniques) and the progress on visualization and human computer interaction
devices. A new architecture developed along with information management concept and
typical mechanism for special forces to effectively interoperate with IT between joint
environments and with different other government departments. The critical issues
(common data and service access, information exchange/ sharing, infrastructure security,
privacy/ confidentiality) should be also integrated in this architecture.
MUSKETEER (MUlti environment decision Support and Knowledge Exploitation TErrorism
Emergency Responses) is a complex program capable to identify, refine, develop, and
integrate decision support and knowledge exploitation tools and demonstrate the efficiency
of improving, the forces ability to respond to the asymmetric attacks. The interest is to build
a framework to support the creation maintenance and sharing SA (common operational
picture – COP, and to assist complex problem solving and decision making at different
levels and different locations).
5.4 Applications for crisis management exercises (DSS-CME)

The objectives of CME are: to enhance the efficiency and operability of structure in crisis
situations; to enhance the capability of all actors involved and their confidence to act
adequately in crisis situations; team-building – partners should trust one another in terms of
communications and valuable contacts are facilitated and consolidated; capabilities to gain
additional awareness of the necessity for collaboration; it permits the identification of
vulnerabilities of strategies; CME reveal where and when it is necessary to work together; it
permits the identification of vulnerabilities in interdependencies of critical infrastructures
across sectors; training for design new structures concepts and measures of performances.
The content and form of exercises depend on the type of extreme risks involved, the level of
exercise, and the types of actors involved.
a. Discussions based exercises address possible mechanism, processes, plans and policies
for any given DSS-CME on a theoretical level. Procedures and possible solutions are
analyzed and different strategies are compared. Even if these exercises are less about
testing existing procedures, then developing new responses to crises they imply the
introducing a new innovative topics.
b. Action oriented exercises are focused on realistic scenarios and represents an important
experience for actors, because it permits the identification the deficiencies. This type of
CME provides an interesting way for intensive training at very low resources.
c. Target oriented levels of CME:
- in implementation level mechanism are clearly organized; it is easy to be
implemented in a real world situations;
- in tactical exercises the priorities are coordination, collaboration, and decision
making, including joint-exercises;
- the strategic exercises is focused on the interaction between the organizations
involved and the complex decision associated.
CME are efficient because they permit an important gain of knowledge with very low
allocated resources. In the case of formation of decision makers the exercises begin with
simple basic, low level of complexity exercises with the interest to test the efficiency of
communication channels. Next step is to add value through more complex exercises,
including networking and joint-exercises.
6. Conclusion
There is an emerging interest for different applications of DSS in critical decision making for
extreme risk and crisis management. The nature of DSS tools have changed significantly and
are equipped now with a variety of intelligent tools such as graphics, visual interactive
modeling, artificial intelligence techniques, fuzzy sets, and genetic algorithms that adds new
capabilities, well adapted for extreme risk and crisis management. The focus is to
demonstrate the efficiency of using DSS in an extended list of applications, including all the
phases in the management of natural hazards, terrorism, technological accidents, but also
financial crises and multiple crises. All these problems should be treated in a
multidisciplinary, modular and scalable framework, flexible, adaptive, and robust. This
framework should be also design in a friendly manner, so that users can input new
data/task easily. DSS-ER/CM is flexible so that new risks and impact functions can be easily
incorporated by users and it should incorporate also financial-economic modules. In this
philosophy, DSS-ER/CM could improve decision making process in extreme risk

management with impact on the global efficiency in multiple crises management.
DSS-ERM/CM supports real time decisions based on a high number of parameters and
criteria, where knowledge is expressed using vague and uncertain concepts, difficult to
asses for human knowledge. Fuzzy logic, artificial neural networks, genetic algorithms are
appropriate ingredients capable to support the objective to represent human knowledge. An
innovative solution is the integration of the contingency management tool (CMT), a
powerful knowledge-based system tool. To cope with the need of decentralised decision-
making, it requires the development of an inherently distributed system. DSS mixed with
CMT system ingredient enables the following features: the support decision under critical
situations reducing the risk of questionable decisions; risk reduction through preventive
actions; increased level of response supported by better training; it overcomes critical cases
when experts are not available; give the information availability for an effective support to
decision makers with a global perspective, in real time. The use of CMT in extreme risk
management responds efficiently to a hierarchical and distributed decision making process
that offers a specific infrastructure (libraries, knowledge bases, databases, and inference
engine) and a set of generic templates capable to reduce the time to market a customized
application. In this philosophy, DSS-ERM could improve decision making process in
extreme risk management with impact on the global efficiency in multiple crises
management.
In this chapter are presented the aspects regarding the integration of the concept of hybrid
DSS in the treatment of extreme risk management. The critical need for a more efficient use
of intelligent ingredients in DSS is linked to the human decision makers limits: cognitive,
economic, time and competitive demands. Both data management and knowledge
management (facts and inference rules used for reasoning) are very important and decision
makers need a better framework, capable to capture imprecision or uncertainty. More than
that they learn from the data/information and optimize the solution in order to provide
interpretable decision rules. In chapter four some applications for the use of hybrid DSS in
flood and drought, asymmetric extreme risk management together with the treatment of
crisis management exercises (CME) are also presented. Future work should be more focused
on the valuation of these types of frameworks in a way that permits better networking
capabilities, costs reduction and the minimizing of the duration for periodic updates.
7. Acknowledgment
This work is supported partially by the National Centre for Programme Management,
Romania.
8. References
Armstrong, J, Collopy, F.K. (1993). Causal forces: structuring knowledge for time series
extrapolation. Journal of Forecasting, No.12, pp. 103-115
Abraham, A. (2001). Neuro-Fuzzy Systems: State-of-the-Art Modeling Techniques, In:
Connectionist Models of Neurons, Learning Processes, and Artificial Intelligence, J. Mira
and A. Prieto (Eds.), 269-276, Springer-Verlag Germany, Granada, Spain
Abraham, A. (2002). Optimization of Evolutionary Neural Networks Using Hybrid Learning
Algorithms, Proceedings of the IEEE International Joint Conference on Neural Networks
(IJCNN'02), 2002 IEEE World Congress on Computational Intelligence, vol.3, pp. 2797-
2802, Hawaii, IEEE Press
Alexander, D. (1993). Natural Disaster, UCL Press
Anheier, H. (ed.) (1999). When Things go Wrong: Failures,Bankruptcies, and Breakdowns in
Organizations, Thousand Oaks, CA: Sage
Comfort, L.K. (ed.) (1988). Managing Disaster: Strategies and Policy Perspectives, London: Duke
Press
Coombs, W.T. (1998). An analytic framework for crisis situations: Better responses from a
better understanding of the situation, Journal of Public Relations Research, Vol. 10, pp.
179–193
Curtin, T.; Hayman, D. & Husein, N. (2005). Managing Crisis: A Practical Guide, Basingstoke:
Palgrave Macmillan
Dahr, V. & Stein, R. (1997). Intelligent Decision Support Methods – The Science of Knowledge
Work, Prentice Hall
Druzdel, M. J. & Flynn, R. R. (2002). Decision Support Systems. (A. Kent, Ed.) Encyclopedia of
Library and Information Science
Endsley, M.R. (2004). Situation awareness: Progress and directions. In: A cognitive approach to
situation awareness: Theory, measurement and application, S. Banbury & S. Tremblay
(Eds.), 317–341, Aldershot, UK: Ashgate Publishing
Fink, E. (2002). Changes of Problem Representation, Springer Verlag, Berlin, New York
Gadomski, A. M.; Bologna, S.; Di Constanzo, G.; Perini, A., & Schaerf, M. (2001). Towards
intelligent decision support systems for emergency managers: the IDA approach,
International Journal Risk Assessment and Management, vol.1, no.3-4
Gadomski, A. M.; Balducelli, C.; Bologna, S. & DiCostanzo, G. (1998). Integrated Parellel
Bottom-up and Top-down Approach to the Development of Agent-based
Intelligent DSSs for Emergency Management. Proceedings of the International
Emergency Management Society Conf. TIEMS'98: Disaster and Emergency Management
Gadomski, A. M.; Bologna, S.; Di Costanzo, G.; Perini, A. & Schaerf, M. (2001). Towards
intelligent decision support systems for emergency managers: the IDA approach.
International Journal Risk Assessment and Management, Vol.2, No. 3,4
Gauld, R. & Goldfinch, S. (2006). Dangerous Enthusiasms: E-government, Computer Failure and
Information System Development. Dunedin: Otago University Press
Guerlain, S.; Brown, D. & Mastrangelo, C. (2000). Intelligent decision support systems,
Proceedings of the IEEE Conference on Systems, Man, and Cybernetics, Nashville, USA
Holsaplle, C. W. & Whinston, A. B. (1996). Decision Support Systems: A Knowledge-based
Approach, West Publishing Company, Minneapolis/St Paul
Jaques, T. (2000). Developments in the use of process models for effective issue
management, Asia-Pacific Public Relations Journal, Vol.2, pp. 125–132
Janssen, T. L. (1989). Network expert diagnostic system for real-time control, Proceedings of
the second international conference on industrial and engineering applications of artificial
intelligence and expert systems, Vol. 1 (IEA/AUIE ‚89)
Kaplan, S. & Garrick, B. (1981). On the quantitative definition of risk, Risk Analysis, Vol.1,
no.1, pp.11 -28
Kebair, F., Serin, F., & Bertelle, C. (2007). Agent-Based Perception of an Environment in an
Emergency Situation, Proceedings of the International Conference of Computational
intelligence and Intelligent Systems (ICCIIS), World Congress of Engineering (WCE),
London, UK, pp. 49-54
Klein, M.R. & Methlie, L.B. (1995). Knowledge-based decision support systems with Applications
in Business, 2nd Edition. England: John Wiley & Sons
Lam, J. & Litwin, M.J. (2002). Where's risk? EWRM knows! The RMA Journal, November 2002
Lee, K.M. & Kim, W. X. (1995). "Integration of human knowledge and machine knowledge by using
fuzzy post adjustment: its performance in stock market timing prediction". Expert Systems,
12 (4), 331-338
Mamdani E. H., & Assilian S. (1975). An experiment in Linguistic Synthesis with a Fuzzy
Logic Controller, International Journal of Man-Machine Studies, Vol. 7, No.1, pp. 1-13
Marques, M. S.; Ribeiro, R.A. & Marques, A.G. (2007). A fuzzy decision support system for
equipment repair under battle conditions. Fuzzy Sets and Systems, Journal of
Decision Systems, Vol. 16, No. 2
Martinsons, M. G. & Davison, R. M. (2007). Strategic decision making and support systems:
Comparing American. Japanese and Chinese Management Decision Support Systems,
Vol.43, pp. 284–300.
McNurlin, B. C., & Sprague, R. H. (2004). Information systems management in practice (6th ed.).
Englewood Cliffs, NJ: Prentice Hall.
Mendelsohn L.B. (1995). Artificial Intelligence in Capital Markets, Chapter 5: Global Trading
Utilizing Neural Networks: A Synergistic Approach, Virtual Trading, Probus Publishing
Company, Chicago, Illinois
Nudell, M. & Antokol, N. (1988). The Handbook for Effective Crisis Management, Lexington:
Lexington Books
Pan J.; Kitagawa H.; Faloutsos C. & Hamamoto M. (2004). AutoSplit: Fast and Scalable
Discovery of Hidden Variables in Stream and Multimedia Database,. Proceedings of
the Eighth Pacific-Asia Conference on Knowledge Discovery and Data Mining (PAKDD
2004), Australia, May 26-28
Pauchant, T.C. & Mitroff, I.I. (1992). Transforming the Crisisprone Organization. Preventing
Individual, Organizational and Environmental Tragedies, Jossey Bass Publishers, San
Francisco
Pearson, P. (2006). Data representation layer in a Multi-Agent DSS, IOS Press, Vol.2, No.2, pp.
223-235
Pearson, C. & Clair, J. (1998). Reframing Crisis Management, Academy of Management Review,
Vol. 23, No. 1, pp. 59–76
Prelipcean G. & Boscoianu M. (2008). Computational framework for assessing decisions in
energy investments based on a mix between real options analysis and artificial
neural networks, Proceedings of the 9th WSEAS International Conference on
Mathematics & Computers in Business and Economics (MCBE’80), pp. 179-184
Prelipcean G.; Popoviciu N. & Boscoianu M. (2008). The role of predictability of financial
series in emerging market applications, Proceedings of the 9th WSEAS International
Conference on Mathematics & Computers in Business and Economics (MCBE’80), pp.
203-208
Prelipcean, G. & Boscoianu, M. (2009). Patents, Academic Globalization and the Life Cycle of
Innovation, Balkan Region Conference on Engineering and Business Education &
International Conference on Engineering and Business Education,15-17 October
2009, Lucian Blaga University of Sibiu, Romania, pp. 149-152
Prelipcean, G. (2010). Emerging Microfinance Solutions for Response and Recovery in the Case of
Natural Hazards, Proceedings of the International Conference on Risk Management,
Assessment and Mitigation, RIMA10, Plenary Lecture, WSEAS Press, ISBN 978-960-
474-182-3, ISSN 1790-2769, pp.220-226
Prelipcean, G. & Boscoianu, M. (2010). Innovation, Technological Change and Labor saving in the
Aftermath of the Global Crisis, Balkan Region Conference on Engineering and
Business Education & International Conference on Engineering and Business
Education, 15-17 October 2009, Lucian Blaga University of Sibiu, Romania, pp.169-
173
Prelipcean, G.; Boscoianu, M. & Moisescu, F. (2010). New Ideas on the Artificial Intelligence
Support in Military Applications, Conference Information: 9th WSEAS International
Conference on Artificial Intelligence, Knowledge Engineering and Data Bases,
February 20-22, 2010, University of Cambridge, pp. 34-39
Regester, M. & Larkin, J. (2002). Risk issues and crisis management—A casebook of best practice
(2nd ed.), Kogan Page, London
Renn, O. (1992). Concept of risk: A classification, In: Social Theories of Risk, Praeger Publisher
Schwarz, H.; Wagner, R. & Mitschang, R.R. (2001). Improving the Processing of Decision
Support Queries: The Case for a DSS Optimizer, Proceedings of (IDEAS '01)
Sousa, P.; Pimentao, J. & Ribeiro, R.A. (2006). Intelligent decision support tool for
priorotizing equipement repairs in critical/disaster situations, Proceedings of the
International Conference on Creativity and Innovation in Decision Making and Decision
Support (CIDMDS 2006)
Sprague Jr., R.H. & Carlson, E.D. (1982). Building Effective Decision Support Systems, Prentice-
Hall, Englewood Cliffs New Jersey
Sugeno, M. (1985). Industrial Applications of Fuzzy Control, Elsevier Science PubCo.
Takagi, T. & Sugeno, M. (1983). Derivation of fuzzy control rules from human operator's
control actions, Proceedings of the IFAC Symposium on Fuzzy Information, Knowledge
representation and decision analysis, Marseilles, France, pp. 55-60
Toevs, A.; Zizka, R.; Callender, W. & Matsakh, E. (2003). Negotiating the Risk Mosaic: A
Study on Enterprise-Wide Risk Measurement and Management in Financial
Services. The RMA Journal, March 2003
Turban E. & Aronson J.E. (2004). Decision Support Systems and Intelligent Systems, Prentice
Hall, Upper Saddle River, NJ, 1998
Turban, E.; Aronson, J.E. & Liang, T.P. (2004). Decision Support Systems and Intelligent
Systems. Vol. 7, ed. 2004, Prentice-Hall
Wildavsky, A. (1988). Searching for Safety, New Brunswick, NJ: Transaction Press
Wisner, B.; Blaikie, P.; Cannon, T. & Davis, I. (2004). At Risk: Natural Hazards, People's
Vulnerability and Disasters (2nd edition), Routledge UK
Zadeh, L. (19650. Fuzzy sets, Information and Control, Vol.8, pp. 338–353
Zhang, C. (1992). Cooperation under uncertainty in distributed expert systems, Artificial
Intelligence, Vol.56, pp. 21-69
Zimmermann, H. (1987). Fuzzy Sets, Decision Making, and Expert Systems, Kluwer Academic
Publishers, Boston, Dordrecht, Lancaster
19
Risk Analysis and Crisis Scenario Evaluation

in Critical Infrastructures Protection
Vittorio Rosato1, Vincenzo Artale1, Giovanna Pisacane1,
Gianmaria Sannino1, Maria Vittoria Struglia1,
Aberto Tofani1 and Eddy Pascucci2
1ENEA, Energetic and Environmental Modelling Unit,
Casaccia Research Centre, Roma
2Ylichron S.r.l., Roma
Italy
1. Introduction
Critical Infrastructures (CI) are technological systems (encompassing telecommunication and
electrical networks, gas and water pipelines, roads and railways) at the heart of citizen’s life.
CI protection, issued to guarantee their physical integrity and the continuity of the services
they deliver (at the highest possible Quality of Service), is one of the major concern of public
authorities and of private operators, whose economic results strictly depend on the way they
are able to accomplish this task .
Critical Infrastructure Protection (CIP) is thus a major issue of nations as the impact of
CIs malfunctioning or, even, their outage might have dramatic and costly consequences for
humans and human activities (1; 2). EU has recently issued a directive to member states
in order to increase the level of protection to their CIs which, in a EU-wide scale, should
be considered as unique, trans-national bodies, as they do not end at national borders but
constitute an unique, large system covering all the EU area (3).
Activities on CI protection attempt to encompass all possible causes of faults in complex
networks: from those produced by deliberate human attacks to those occurring in normal
operation conditions up to those resulting from dramatic events of geological or meteorologic
origin. Although much effort has been devoted in realizing new strategies to reduce the risks
of occurrence of events leading to the fault of CI elements, a further technological activity is
related to the study of possible strategies to be used for predicting and mitigating the effects
produced by CI crisis scenarios. To this aim, it is evident that a detailed knowledge of what is
going to happen might enormously help in preparing healing or mitigation strategies in due
time, thus reducing the overall impact of crises, both in social and economic terms.
CIP issues are difficult to be analyzed as one must consider the presence of interdependence
effects among different CIs. A service reduction (or a complete outage) on the electrical
system, for instance, has strong repercussions on other infrastructures which are (more or
less) tightly related to the electrical system. In an electrical outage case, for instance, also
vehicular traffic might have consequences as petrol pumps need electrical power to deliver
petrol; pay tolls do need electrical current to establish credit card transactions. As such, also
378
vehicular traffic on motorways might strongly perceive the effects (after a certain latency
time) of an outage on the electrical system. This is a less subtle interdependence than that
present for CI which are more directly related to the electrical power delivery, such as railway
traffic; nevertheless, all these effects must be taken into account when healing and mitigation
strategies must be envisaged for the solution of a crisis event (4).
This work reports of a new strategy aimed at realizing tools for the prediction of the onset
of crisis scenarios and for a fast prediction of their consequences and impacts on a set
of interdependent infrastructures, in terms of reduction of the services dispatched by the
infrastructures and the impact that services unavailability might have on population. All
that in order to provide a new generation of Decision Support Systems which can support CIs
operators in performing preparedness actions and to optimizing mitigation effects.
The present chapter is composed of 3 sections: in the first, the general layout of the system
is proposed, where each task of the system is described. This section contains the general
description of the risk analysis and the tools which are used to make quantitative evaluations.
The second section will encompass a general description of the meteo-climate simulation
models which provide an accurate evaluation of the precipitation level expected on short-
and medium-long period. The last section will be entirely devoted to the description of the
impact evaluation of crisis scenarios.
2. General description of the DSS layout

The chapter will report on the main ideas at the origin of a new class of Decision Support
System (DSS) which attempts to combine data (of several types and sources), dynamic data
(from field sensors), dynamical predictions (weather, climate) in order to produce a dynamical
risk assessment of CIs and a subsequent evaluation of the impact that predicted crises
scenarios might have of technological infrastructures, services, population.
MIMESIS (Multi Infrastructure Map for the Evaluation of the Impact of Crisis Scenarios) is an
example of this new DSS concept, realized to evaluate the risk to which CIs present on a given
area are exposed, and to study physical interdependencies among different networks.
The tool can be used by CIs holders, Regional and National Agency for Civil Protection, Land
Control Agencies.
MIMESIS is composed of the following components:
1. a geo-database designed to store, query, and manipulate geographic information and
spatial data which contains stored land e CIs data;
2. a "static" analysis tool which, combining topological information with land data, is able to
evaluate a static risk exposition index to each constitutive element of all CIs;
3. a "dynamic" analysis tool which, combining information from land sensor networks,
satellite images, weather forecast, is able to evaluate a dynamic risk through the acquisition
of dynamic real-time data;
4. the crisis scenario generator that inputs data to the DSS which evaluates the impact of
faults at the topological and functional level of the CIs by using CI federated simulators;
5. a GIS user interface which allows to view crisis scenario and its evolution.
The general layout of the MIMESIS DSS is reported in Figure1
MIMESIS’s goal is to predict crisis scenarios, to evaluate their impact in order to achieve
prompt response and to help mitigating their effects.
Risk Analysis
Risk Analysis and Crisisand Crisis
Scenario Scenario
Evaluation Evaluation
in Critical Infrastructures in Critical Infrastructures Protection
Protection 3793
Fig. 1. General workflow of the MIMESIS DSS.
2.1 Geo-database
A major problem that risk assessments and mitigation strategies must cope with is the lack of
a centralized data repository allowing comprehensive risk analysis and risk prediction of CI
in a given region. Without a mean to consider, on the same ground, all CIs and their mutual
interdependencies, any efficient way to predict and mitigate crisis scenarios could be realized.
Impact mitigation strategies should, in fact, consider the different responses of the different
CIs, their perturbation in relation to the crisis, the different latency times for perturbation
spreading, the different timings in the healing actions.
To consider all these issues, the first necessary action is to constitute a control room where
data of all CIs should be made available. For this reason, the geo-referenced database
plays a central role in the MIMESIS tool. It contains a large set of data of different kinds:
(1) regional nodes of critical infrastructures, such as electrical stations, power generators,
high-to-medium-to-low voltage electrical transformers, telecommunication switches and
primary cabins, railways and roads with the specific access points, gas and water pipelines
and their specific active points; (2) geographic and elevation maps of the region with the
highest possible accuracy; (3) position, flow rates, hydrographic models of all water basins
(rivers, natural or artificial lakes, hydroelectric reservoirs etc.); (4) landslide propensity of
the different areas according to historical repositories, where dates and landslides types are
recorded; (5) geo-seismic data provided by the soil geo-seismic national agency, supported by
real-time data provided by in-situ accelerometers (where available); (6) geo-seismic data on
geological faults ; (7) social (cities, population densities), administrative (counties, districts)
and economical (industrial areas classified in terms of energetic consumptions, produced
GDP, types of resources the area is dependent on, etc.) data of the region; (8) agricolture’s
380
maps, fisheries etc. (9) traffic data on motorways and major urban roads (Origin-Destination
matrices, if available) (10) railways data with passengers and goods traffic; (11) any other data
related to other infrastructures (water, gas-oil pipelines, wherever present in the territory).
Such a huge database should be provided in a GIS format, allowing a precise geo-referenced
of each constitutive network’s elements.
2.2 Static risk analysis

MIMESIS workflow starts with a periodic (yearly) evaluation of the "static" risk to whom each
constitutive element of CI is submitted because of its geographical position. It is possible, in
fact, to determine a number of risk indices, each related to a specific risk threat, which could
be evaluated on the base of historical, geographical, geo-seismic data. In all cases, the risk
function Ri (r ) of the site r related to the risk agency i, say Ri (r ), can be expressed as follows:

Ri (r ) = Pi (r ) I (r ) Vi (r )dr (1)
where Vi (r ) is a suitable, normalized function estimating the weight of the specific agent i
in resulting a threat for the infrastructures (seismicity, presence of water basins, historical
propensity of the terrain to landsliding etc.) integrated over a suitable area surrounding the
CI element; Pi (r ) represents the sensitivity of the specific CI element to the threat i (for a given
CI element located in the point r, the value Pi (r ) might be larger if i is the seismic threat rather
than the flood threat caused by the nearby presence of water basins); I (r ) (which is essentially
independent on the threat i) is the sum of the impacts that the absence of the CI element in
r produces upon failure in its network and in the other CI networks which are functionally
related to that.
We have distinguished a number of agents which could produce risk evidences for the CI
elements. Among them:
• geo-seismic risk; each CI element is evaluated as a function of its position in the seismicity
map. In our database, we have the update italian seismic which is periodically updated by
the National Institute of Geophysics and Volcanology (INGV) (5).
• landslide risk; each CI element is evaluated as a function of its position in the italian
inventory of landslides (resulting from the ISPRA project IFFI (6) ).
• water basins proximity risk; each CI element is evaluated as a function of its position with
respect to the regional water basin. Integration in Eq.(1) is performed over a circle of a
radius which could be varied (from few hundreds meters up to a few kilometers, in the
proximity of rivers with large discharge.
The impact value can be estimated as being the reduction of the Quality of Service (QoS) of the
specific service provided by the network containing the faulted element. However MIMESIS
attempts to evaluate also the economic and social impact that the QoS loss implies.
2.3 Dynamic risk analysis

Analogously to "static" risk analysis, a time-dependent assessment of the risks to which
CI elements are exposed can be performed by predicting, at the short- and medium-time
scales, the amount of precipitations expected in a given area. Severe rainfalls are, in
fact, meteorological events which can produce flooding of water basins and/or determine
Risk Analysis
Scenario Scenario
Protection 3815
landslides in prone grounds. The MIMESIS system daily produces an high-resolution weather
forecast with precipitation mamaps of 1 square kilometre. The numerical models involved
in weather and climatological predictions will be will be described in section 3. Then, static
analysis should be updated by evaluating the risk values upon the application of precipitation
data. In fact, the value Ri (r ) of Eq. 1 could be recalculated by modifying the function V (r ).
Let consider V (r ) as the current flow of a river. Upon consistent precipitations, the river
flow could increase and the integration of the function V (r ) over a given area could produce
an higher, over-threshold value for the risk value Ri (r ). For the landslide risk threat, the
V (r ) function could measure the extent of the correlation (captured from historical data)
precipitation abundance/landslide propensity. When precipitations are abundant, a large
landslide probability could be triggered in a specific area (that comprised in the integration of
V (r ) in Eq. 1) and the risk function Ri (r ) consequently increase.
2.4 GIS database and the interface

The GIS database is designed to store, query, and manipulate geographic information and
spatial data. Inside a geodatabase, the feature classes are organized as datasets, a collection of
data presented in tabular form.
Data are stored together in a relational database where geospatial data are defined by vectorial
representation. In the vector based model, the basic units of spatial information are points,
lines and polygons, each of these is made up only as a series of one or more coordinate points.
The database contains also the topological information of a network which is necessary for
the topological algorithm analysis network developed in MIMESIS.
The application is based on Intergraph’s GeoMedia Object Model, which supplies the
function for representation and analysis of GIS data. The interface presents a layer-based
representation shown in Figure 2
In the layer-based approach, the spatial data are presented in a set of thematic maps,
called layers, which denote some given theme such as electrical network, telecommunication
network, rivers, railways, roads etc. Each layer is a geo-referenced data set and associated
with a table of the geodatabase containing also the attributes of geographic data. MIMESIS
proposes a dynamic GIS representation adding space-temporal information to represent the
evolution of a scenario. Indeed, the interface allows to control the DSS and to show the results.
As the DSS evaluates the impact of faults at the topological and functional level of the CIs by
using federated simulators, the interface shows the consequences of crisis scenarios enabling
to detect hidden interdependencies between the different infrastructure.
3. Meteo and climatological forecasts

3.1 Weather and climate numerical models
Weather forecast and climate prediction must be considered as a key component of the new
class of Decision Support System described in this work of Decision Support System.
Weather forecast provides detailed and reliable short-term spatial analysis of mesoscale
weather events, while climate predictions provide long-term climate change indicators
at a lower spatial resolution. Weather forecast will be used in the MIMESIS DSS for the
assessment of the so called dynamic risk, while climate predictions are more suitable for the
generation of crisis scenario at a larger time scale and for assessing the impact of climate for
designing new interventions and extension of the existing infrastructures.
382
Fig. 2. MIMESIS Graphical User Interface.
Both numerical weather forecast and climate prediction involve the use of mathematical
models of the atmosphere, represented by a set of discretized dynamical equations and by
local deterministic parameterizations of physical processes occurring at sub-grid scales. Such
processes are supposed to be in statistical equilibrium with the resolved-scale flow and are,
therefore, treated as to their mean impact on the resolved scales.
Climate, seasonal and weather models are basically identical in structure, independent of the
specific time scale for which they have been developed, and often share the dynamical core
and physical parameterizations. Differences only lie in the resolution at which models are run,
which may imply specific tuning of the sub-grid representation, although the mathematical
core of the two approaches is conceptually distinct.
Climate is, by definition, the statistical mean state of the Earth system with its associated
variability. Therefore, numerical simulation of climate, as performed by General Circulation
Models (GCMs), is a boundary condition problem, and changes in the system equilibrium
derive from slow changes in boundary forcing (such as the sea surface temperature, the
solar constant or the greenhouse gas concentration). On the other hand, Numerical Weather
Prediction models (NWPs) are used to predict the weather in the short (1-3 days) and medium
(4-10 days) range and depend crucially on the initial conditions. For instance, small errors in
the sea surface temperature or small imbalances in the radiative transfer have a small impact
on a NWP model but can dramatically impair GCM results.
To partly overcome this problem, coupled Atmosphere-Ocean models (AOGCMs) have been
developed. In order to allow an adequate description of the system phase space the GCM
Risk Analysis
Scenario Scenario
Protection 3837
simulation runs would last tens of years. The consequent computational cost limits the
spatial resolution of climate simulations, so that local features and extreme events, which
are crucial to good weather predictions are, by necessity, embedded in sub-grid process
parameterizations.
A similar restriction holds for global Weather Prediction Models (WPMs) that are currently
run at different meteorological centers around the world, whose prediction skill is enhanced
by performing several model forecasts starting from different perturbations of the initial
conditions (ensemble forecasting), thus severely increasing computational requirements.
Future high resolution projections of both climate and weather rely on three classes
of regionalization techniques from larger scale simulations: high-resolution "time-slice"
Atmosphere GCM (AGCM) experiments (8), nested regional numerical models (9), and
statistical downscaling (10), each presenting its own advantages and disadvantages.
At present, dynamical downscaling by nested limited area models is the most widely adopted
method for regional scenario production, its reliability being possibly limited by unavoidable
propagation of systematic errors in the driving global fields, neglecting of feedbacks from the
local to the global scales and numerical noise generation at the boundaries. This technique,
however, possesses an unquestionable inherent capacity to fully address the problem of
weather prediction and climate change at finer resolutions than those allowed by general
circulation models, as it allows local coupling among different components of the Earth
system at a reasonable computational cost.
3.2 Impacts and risks prediction

Recently there has been a growing concern for the quality and reliability of both weather and
climate predictions, which are expected to improve the planning and management of several
economic, social and security sectors (i.e. agriculture, energy production, tourism, transport).
Either on long or on short time scales, our ability to predict atmospheric phenomena
clearly has a direct impact on sensitive issues, such as water resource management and
hydro-geological risk assessment. However, these issues intrinsically depend on a variety of
additional phenomena, e.g. processes determining soil quality, vegetation type and extension,
and water demand and distribution. Therefore, resource availability and risk control are
the integrated result of natural events, socio-economical interactions and political decision,
and only interdisciplinary strategies can tackle the definition of priorities and means of
intervention.
Following this approach, the UK Met Office has addressed, in conjunction with key energy
players, the problem of responding to the challenge of climate change in the areas of energy
transmission and distribution, network planning, energy trading and forecasting (11). The
main findings of the project regarded network design standards, including changes in risk
profiles for critical elements such as transformers, cables and conductors, the reduction in
thermal plant output, the wind power potential and the vulnerability of infrastructures to
extreme events such as snow and windstorms. The project also assessed changes in energy
demand (gas and electricity). On the long time scale the analysis provided energy companies
with guidelines on how to deal with climate projections and their related uncertainties.
This approach is also advisable for the rest of Europe, recognized as a hot spot in the last
Intergovernmental Panel on Climate Change (IPCC) report (12), (13), which further urged
the scientific community to undertake the difficult, but inescapable, task of supporting
governments and authorities for a responsible environmental and economic planning. The
IPCC report states that nearly all European regions are expected to be negatively affected by
384
future impacts of climate change and these will pose challenges to many economic sectors.
Climate change is expected to magnify regional differences in Europe’s natural resources
and assets. Negative impacts will include increased risk of inland flash floods, and more
frequent coastal flooding and increased erosion (due to storminess and sea-level rise). In
Southern Europe, climate change is projected to worsen conditions (high temperatures and
drought) in a region already vulnerable to climate variability, and to reduce water availability,
hydropower potential, summer tourism and, in general, crop productivity. It is also projected
to increase health risks due to heat waves and the frequency of wildfires (14).
Such dramatic changes are attributed to the anthropogenic warming arising from augmented
carbon dioxide emissions, which have a discernible influence on many physical and biological
systems, as documented in data since 1970 and projected by numerical models. Carbon
dioxide and the other greenhouse gases affect the atmospheric absorption properties of
longwave radiation, thus changing the radiation balance.
An immediate impact of this altered energy balance is the warming of the lower troposphere
(an increase of the global temperature of about 0.6◦ C has been observed over approximately
the last 50 years) that, in turn, affects the atmospheric hydrological cycle. Although extremely
relevant as to their effects on human activities, hydrological processes are still poorly
modeled, and projections are affected by severe uncertainties. Climate models can hardly
represent the occurrence probability and the duration of extreme rainfall or drought events,
even in today’s climate conditions (15), so that governmental authorities now explicitly
demand innovative science-based approaches to evaluate the complexity of environmental
phenomena.
3.3 Hydrological impacts

From the point of view of weather forecast on shorter time scales (from daily to seasonal),
a reliable representation of the hydrological cycle is also essential for the early warning
of extreme events and the evaluation of their short-term impacts. Heavy rain may cause
considerable compaction and erosion of the soil by its force of impact, sealing the soil surface
and channeling the water to run off the surface carrying away the topsoil with it.
Considering hilly and mountainous areas, heavy rain may produce enormous erosion by
mudflow generation, while rainwater running off hard impervious surfaces or waterlogged
soil may cause local flooding (16). At the other extreme, the early prediction of drought events
deriving from extreme rainfall deficits would also greatly benefit from accurate forecasts of
the relevant hydrological variables. As a matter of fact, drought is, unlike aridity, a temporary
phenomenon characterized by high spatial variability (17), whose representation could be
effectively approached with high resolution regional models.
In order to extend their ability to describe the hydrological cycle, river routing modules
are currently being incorporated in atmospheric models in order to link the meteorological
forcing (in particular rainfall) to the hydrological response of a catchment. Together with
land modules (which mimic the interactions between the atmosphere and the biosphere)
they represent an alternative to the rainfall-runoff models which, in recent years, have been
employed in a wide range of applications to assess impacts of weather, climate or land-use
change on the hydrological cycle (18), (19). As atmospheric models are routinely run in either
meteorological or climatological applications, such an extension is likely to be a feasible and
economic way to help hydrologists to derive quantitative figures about the impacts of the
observed or expected environmental changes. It should be stressed again, however, that the
Risk Analysis
Scenario Scenario
Protection 3859
impact studies, which follow up model projections, are definitely in need of complex systems
capable of crossing information from different disciplines and of managing huge amounts of
data.
The uncertainty involved in this type of impact assessment limits the value of the results and
great care should be taken in evaluating model skill in predicting the driving meteorological
variables. Precipitations, the main atmospheric driver of hydrological catchment response,
are unfortunately still a critical output of model diagnostics (20). Although the complexity
of cloud parameterizations is always increasing, this is no guarantee of improved accuracy,
and better representation of clouds within NWP models has been the focus of recent research
(21),(22), (23), (24). Numerical models explicitly resolve cloud and precipitation processes
associated with fronts (large scale precipitation), while parametrizing small scale phenomena
by means of the large-scale variables given at the model’s grid points. The most important
parameters are humidity, temperature and vertical motion. The vertical velocity determines
the condensation rate and, therefore, the supply of liquid water content. Temperature also
controls the liquid water content, via the determination of the saturation threshold. Moreover,
the temperature distribution within a cloud is also important in determining the type of
precipitation - rain or snow. The complexity of the parameterization of cloud processes is
limited by the associated numerical integration time (25). Model spatial resolution is crucial
for a reliable treatment of condensation processes, as vertical motion of air masses is often
forced by orography, whose representation therefore needs to be as accurate as possible.
Again, regional models, due to their higher spatial resolution and reduced computational
costs, seem to be the most appropriate tool for downscaling precipitation fields, at the same
time preserving the complexity of convection parameterization. However, the reliability of
precipitation forecasts provided by state-of-the art meteorological models also depends on
their ability to reproduce the sub-grid rain rate fluctuations which are not explicitly resolved.
In particular, the assessment of precipitation effects on the hydrological scales requires an
accurate characterization of the scaling properties of precipitation, which is essential for
assessing the hydrological risk in small basins, where there is a need to forecast watershed
streamflows of a few hundred square kilometres or less, characterized by a concentration time
of a few hours or less.
At these smaller space time scales, and specifically in very small catchments and in urban
areas, rainfall intensity presents larger fluctuations, and therefore higher values, than at the
scale of meteorological models (26). In order to allow a finer representation of land surface
heterogeneity than that allowed by nominal grid resolution, mosaic-type schemes are being
investigated, which account for topographical corrections in the sub-grid temperature and
pressure fields and downscale precipitation via a simple statistical approach. Such schemes
allow simple implementation of space dependent probability distribution functions that may
result from ongoing research on statistical downscaling of precipitation (27). As already
mentioned, a stochastic approach has also been successful in improving precipitation forecast
reliability as to its large scale statistical properties. In the last decade, ensemble prediction
systems have substantially enhanced our ability to predict precipitation and its associated
uncertainty (28). It has been shown that such systems are superior to single deterministic
forecasts for a time range up to two weeks, as they account for errors in the initial conditions,
in the model parameterizations and in the equation discretization that might cause the
flow-dependent growth of uncertainty during a forecast (29), (30), (31). At the same time,
multi-model ensemble approaches have been indicated as a feasible way to account for model
errors in seasonal and long-term climate studies (see Figure 3).
386
Fig. 3. Climatological precipitation field representing the period 1958-1998. The map has
been produced analyzing the output of the 9 models involved in the EU ENSAMBLE project.
It has been proved that, under the assumption that simulation errors in different models are
independent, the ensemble mean outperforms individual ensemble members (29), (32), (33).
By sampling modeling uncertainties, ensembles of GCMs should provide an improved basis
for probabilistic projections compared to an ensemble of single model realizations, as the latter
only samples different initial conditions, i.e. a limited portion of a specific model phase space.
Ensemble predictions are therefore increasingly being used as the drivers of impact forecasting
systems (34), (35), (20), thus reinforcing the already pressing demand for complex numerical
systems that allow rapid inter-comparison between model realizations and multivariate data
analysis.
4. Evaluation of impacts of crisis scenarios on CIs

4.1 Crisis scenario generator and impact evaluation
Once the risk assessment of the different CI elements is concluded and, after that, the state of
risk of one (or more) elements of one (or more) infrastructures have taken over a pre-defined
threshold, the system produces an alarm and defines a "crisis scenario". If one defines the
generic k-th given infrastructure the set Gk (Ω) of elements (nodes and arcs, together with the
network which connects them), we define a "crisis scenario" a new set Gk (Ω ) where Ω is the
set of constitutive elements of the infrastructure k without the elements (nodes and/or arcs)
which have been supposed to be lost (or damaged) because the above threshold risk value.
MIMESIS first performs a topological analysis in order to see whether the network of the
Risk Analysis
Scenario Scenario
Protection 387
11
infrastructure is still topologically connected and determines which are the new topological
"critical" points of the network. Topological analysis of the network is carried out through the
evaluation of the following quantities (7):
• nodes and links centrality indices (Betweenneess centrality, Information centrality);
• network’s diameter, min paths, min cuts;
• spectral analysis of the Adjacency and Laplacian matrices associated to the network.
Topological analysis is a first mean to assess the integrity of the network. The presence of
disconnected components of the graph can be easily seen by evaluating the eigenvalues of
the Laplacian matrix associated to the network. If the graph G(associated to the network)
has n vanishing eigenvalues, the graph has n different disconnected components. Centrality
measures tend to identify which are the most relevant elements (nodes, arcs) of the network.
Node i Information centrality, for instance, determines which is the increase of the min paths
among all the other nodes if node i is lost (when network i is lost, the min paths connecting all
other nodes originally passing through i should be re-evaluated and they will produce new
min paths larger than the original ones). As far Information Centrality is concerned, larger is
the Information Centrality, larger is the importance of the node to provide "good connections"
among all the others.
After a first assessment of the perturbed network upon topological analysis, the MIMESIS
tool performs the most relevant action: the estimate of the reduction of the Quality of Services
produced by the perturbation occurred to the network due to the faults in one or more
of its elements. This task is accomplished by using "domain" or "federated" simulators of
CIs. For domain simulators we intend commercial or open source simulators of specific
infrastructures: electrical (such as Powerworld, E-Agora, Sincal etc.), telecommunications
(NS2), railways (OpenTrack) etc. Federating domain simulator is one of the major
achievements granted by a strong collaboration among the european CIP scientific community
(e.g. the projects IRRIIS (36) and DIESIS (37)). For federated simulators we intend a new
class of simulators which "couple" two or more domain simulators through some specific
synchronization mechanisms and some interdependency rules which allows to describe how,
and to what extent, a CI determines the functioning of another. MIMESIS integrates the
outcome of one of the most successful EU FP7 projects, the DIESIS project (37) which has
attempted to design a general model which could allow to integrate more domain simulators
in an unique framework. The key role in the DIESIS mode is played by the ontology model
(KBS) which is able, to an abstract level, to describe a generic Critical Infrastructure and its
links with other Infrastructures. A generic element of a network from this abstract space can
be subsequently "mapped" into the real space of a specific Critical Infrastructure (electrical,
telecommunication or others) by adapting the generic elements to the specific case. The
ontology model allows to avoid the problem of directly connecting systems which have
different structures, different constitutive elements, different functioning laws: Integration
is firstly performed in a "meta-space" (the abstract space of Critical Infrastructures) (38) and
then mapped into the spaces of the single infrastructures. A brief sketch of the KBS approach
is outlined in the following section.
4.2 The DIESIS ontological approach for CI simulators integration

Within the DIESIS project, a framework for CIs simulators integration has been proposed
that allows the separation of the scenario representation (Scenario Definition Phase) from the
388
simulation framework (Federated Simulation Phase). The main idea is to develop a Knowledge
Base System (KBS) based on ontologies and rules providing the semantic basis for the federated
simulation framework (40), (41).
In particular, a federation of simulators can be considered as a System of Systems where

each simulator represents an independent entity with its own behavior and purpose. The
super-system considers the interaction of these entities and puts in evidence an emergent
behavior that does not reside in any component system (42). Therefore, in a federated
simulation, the stand-alone simulators must be linked together so that an understandable and
meaningful information exchange could be performed. This requires that simulators could
interact and cooperate. The proposed ontological approach allows both an uniform modeling
of heterogeneous infrastructures and the easy representation of inter-domain dependencies.
The DIESIS KBS establishes a common formalism, for scenario and domain knowledge
experts, to represent the main aspects, elements and properties of the considered domains
and their interconnections. The KBS is based on the Ontology Web Language (OWL) and
the ontologies are defined through proper specifications of classes, properties and individuals
(instances of classes). The individuals represent the physical/logical entities that form the
universe of a specific domain. For instance, a specific electrical load is an individual of the
electrical ontology. The OWL allows to group individuals in classes that define the properties
shared by all their individuals. The properties can be used either to specify relationships
between individuals and data type values (Datatype Properties) or to describe relationships
among individuals (Object Properties). Then, if we denote with Pr the set of properties we
have
Pr = DP ∪ OP (2)
with
DP = { p | p Datatype Property} (3)
and
OP = { p | p Object Property} (4)
The KBS architecture includes the following ontology definitions:
• World ONTology (WONT);
• Infrastructure domain ONTologies (IONTs);
• Federation ONTology (FONT).
The WONT is a general template providing the basic structures and rules to define CI
domains. In particular, the WONT allows the definition of CI domain elements (through
the WONT class Component), their physical and logical interconnections (through the WONT
object property isConnected) and the dependencies among different CI domains (through the
WONT object property dependsON).
The IONTs inherit the basic template from the WONT and represent the specific knowledge
of the critical infrastructure domains. For instance, the electric IONT class Load (that models
the electric load element) and the telecommunication IONT class Switch (that models the
telecommunication switch element) are subclasses of the WONT class Component. In addition,
the railway IONT property isLinked that models the connection between two railway stations
is a sub-property of the WONT property isConnected; similarly, the telecommunication IONT
property transmits that models the connection between a transmitter and a receiver is another
Risk Analysis
Scenario Scenario
Protection 389
13
sub-property of the WONT property isConnected. Analogously, all dependencies among the
considered CI domains are modeled through ad-hoc designed sub-properties of the WONT
property dependsON.
In the following, given a CI domain Xi , Ci indicates the set of all components of Xi and Pri
indicates the set of properties related to the components of Xi . Then, a generic IONT can be
represented as IONTi = {Ci , Pri }.
Once the IONT has been defined to model a particular domain, it is possible to create
individuals (instances of IONT classes) to represent actual network domains (for example
the electrical power network or the telecommunication network of a specific city district).
Similarly to the IONT definitions, we indicate with Ci∗ the set of the all instantiated
components belonging to the domain Xi and with Pri∗ the set of instantiated properties related
to Xi . Then, the IONT instance IONTi∗ can be expressed as IONTi∗ = {Ci∗ , Pri∗ }.
The FONT includes all IONTs of the domains involved in the considered scenario (e.g.
electrical, telecommunication, railway domains). The FONT properties (sub-properties of the
WONT property dependsON) allow to model dependencies among components of different
domains (e.g. the FONT property feedsTelco models the electrical supply of telecommunication
nodes). The sets of the FONT properties and of the FONT instantiated properties are defined
respectively as:
FPr = {sp( a, b) | sp sub-property of dependsON, a ∈ Ci , b ∈ Cj , i = j} (5)

and
FPr ∗ = {sp( a, b) | sp sub-property of dependsON, a ∈ Ci∗ , b ∈ Cj∗ , i = j} (6)
The Figure 4 summarizes the proposed KBS architecture
The FPr ∗ allows the definition of FONT rules that express the semantic dependency. The
FONT rules, expressed using the Semantic Web Rule Language (SWRL), have been actually
translated into their JAVA counterpart and implemented through "if-then-else" constructs
embedded within the Federation Managers which incapsulate the simulators of each CI
domain ((39)).
The DIESIS distributed simulation framework using the KBS allows the components (involved
in the defined FONT properties) to exchange the functioning status values.
5. Conclusions
CIP is a major concern of modern nations. EU has issued a Directive (3) to increase awareness
of this duty as CIs are become transnational bodies whose care must be a shared concern.
Whether markets liberalization has produced, at least in principle, benefits for the consumers,
it has de facto imposed a deep revision of the governance strategies of the major national
infrastructures. In many countries, there has been a (sometimes sudden) fragmentation of the
ownership and the management of relevant parts of infrastructures (see, for instance, those
of gas and oil distribution, telecommunications, electrical transmission and distribution,
motorways and railways) which has strongly weakened the centralized governance model
which has been substituted with a model of "diffused" governance of the infrastructures.
Many different industrial players autonomously own and manage parts of the infrastructures,
leading its global control more complex. The lack of information sharing among operators
of different parts of the infrastructures, due in some cases to industrial competition reasons,
reduces the technical control of the whole infrastructure and, even more, reduces the
390
Fig. 4. KBS architecture.
possibility of a control of the "systems of systems" defined by the entanglement of all

interconnected CIs.
The new generation of analysis and risk prediction of CIs, whose the system MIMESIS would
be a prototype, is intended to overcome the above mentioned drawbacks imposed by modern
market policies and the increased CIs interdependency. The MIMESIS system has its core in
weather and climate predictions and by new hydrological models. In a future perspective
of rapid and (sometimes) strong climatic changes and the occurrence of a large fraction of
extreme events, prediction capabilities will be a winning asset against the occurrence of large
and uncontrolled outages of one or more CIs.
Far from being usable only for assessing risk prediction related to meteo-climatic events,
MIMESIS could also be used for "reverse engineering" actions related to CIP against external
malicious attacks. MIMESIS could, in fact, be used to estimate the most vulnerable points of
the networks, not only in relation to their single topology and functions, but also with respect
to the role they play in the wider context within the cited "system of systems". An attentive
"off-line" use of the tool through the insertion of random faults could reveal which types of
faults produces largest perturbations to the systems, leading to suggestions in the network’s
Security area.
Risk Analysis and Crisis Scenario Evaluation in Critical Infrastructures Protection 391
Other than being a prediction tool for external events, MIMESIS could also be used to
correctly design new branches of existing infrastructures, by allowing the ex ante evaluation
of their impact on the Quality of Services of all the system of interdependent infrastructures.
We do believe that data availability by CI owners, their integration with other types of data
(geophysical, economic, administrative etc.), the use of advanced numerical simulation tools
for weather and climatic predictions, the use of CI simulators (both single domain or
"federated" simulators) could represent an invaluable clue to realize a new generation of
tools for increasing protection and enhancing security of CIs.
6. References
[1] UCTE "Interim Report of the Investigation Committe of the 28 September 2003 Blackout
in Italy", 17 October, 2003.
[2] "Final Report on the August 14, 2003 Blackout in the United States and Canada: Causes
and Recommendations", U.S. - Canada Power System Outage Task Force, April
2004.
[3] EU Council Directive 2008/114/EC on the identification and designation of European
critical infrastructures and the assessment of the need to improve their protection
(December 8, 2008).
[4] Rosato V., Tiriticco F., Issacharoff L., De Porcellinis S., Setola R. (2008), Modelling
interdepedent infrastructures using interacting dynamical models, Int. J. Crit.
Infrastr. Vol. 4 n.1/2 (63,79)
doi:10.1504/IJCIS.2008.016092
[5] http://esse1-gis.mi.ingv.it/s1_en.php
[6] IFFI stands for Inventory of Landsliding Phenomena in Italy
(http://www.apat.gov.it/site/en-GB/Projects/IFFI_Project/default.html
[7] Boccaletti, S. et al, (2006). Complex Networks: Structure and Dynamics Phys. Reports,
Vol.424, No.4-5, (February 2006) (175-308),
doi:10.1016/j.physrep.2005.10.009
[8] Cubasch, U et al (1995). Regional climate changes as simulated in time-slice experiments.
Clim Change, 31: 273-304,
doi: 10.1007/BF01095150
[9] Giorgi, F et al (1999). Regional climate modeling revisited. An introduction to the special
issue J Geophys Res 104: 6335-6352.
doi:10.1029/98JD02072
[10] Wilby, RL et al (1998). Statistical downscaling of general circulation model output: a
comparison of methods. Water Resources Research 34: 2995-3008.
doi:10.1029/98WR02577
[11] Buontempo C. et al., Multiscale projections of weather and climate at UK Met Office, in
"Management of Weather and Climate Risk in the Energy Industry" Proceedings of
the NATO Advanced Research Workshop on Weather/Climate Risk Management
for the Energy Sector, Santa Maria di Leuca, Italy, 6-10 October 2008, Troccoli A.
ed., NATO Science for Peace and Security Series C: Environmental Security,
Springer Science+Business Media B.V. 2010, ISBN: 978-90-481-3690-2.
[12] Solomon, S., D. Qin, M. Manning, Z. Chen, M. Marquis, K.B. Averyt, M. Tignor and
H.L. Miller (eds.), (2007). Contribution of Working Group I to the Fourth
Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge
University Press, Cambridge, United Kingdom and New York, NY, USA, 996 pp.
[13] Giorgi, F (2006). Climate change hot-spots Geophys. Res. Lett. 33, L08707,
doi:10.1029/ 2006GL025734
[14] Adger, W.N. et al. (2007). Summary for policy-makers. In Parry, M.L. Canziani, O.F.,
Palutikof, J.P., Hanson, C.E., van der Linden P.J., (eds.) Climate Change 2007:
Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the
Fourth Assessment Report of the Intergovernmental Panel on Climate Change. pp.
7-22. Cambridge University Press: Cambridge.
[15] A. Bronstert (2004). Rainfall-runoff modelling for assessing impacts of climate and land-
use change, Hydrol. Process. 18, 567–570
doi: 10.1002/hyp.5500
[16] T. J. Toy, G. R. Foster, K. G. Renard (2002). Soil erosion: processes, prediction,
measurement, and control, John Wiley & Sons Inc., New York
[17] Dracup, JA et al. (1980). On the definition of droughts Wat. Resour. Res., 16 (2), 297-302,
doi:10.1029/WR016i002p00297
[18] Bronstert, A et al (2002). Effects of climate and land-use change on storm runoff
generation: present knowledge and modelling capabilities. Hydrological Processes
16(2): 509–529,
doi: 10.1002/hyp.326
[19] Niehoff, D et al. (2002). Land-use impacts on storm-runoff generation: scenarios of land-
use change and simulation of hydrological response in a meso-scale catchment in
SW-Germany J. of Hydrology 267(1–2): 80–93,
doi: 10.1016/S0022-1694(02)00142-7
[20] Pappenberger, F. et al. (2005). Cascading model uncertainty from medium range
weather forecasts (10 days) through a rainfall-runoff model to flood inundation
predictions within the European Flood Forecasting System (EFFS). Hydrol. Earth
Syst. Sci., 9, 381–393,
doi:10.5194/hess-9-381-2005
[21] Wilson, D. et al. (1999). A microphysically based precipitation scheme for the UK
Meteorological Office Unified Model. Q. J. R. Meteorol. Soc., 125, 1607–1636.
[22] Tompkins, A. M. (2002). A prognostic parameterization for the subgrid-scale variability
of water vapour and clouds in large-scale models and its use to diagnose cloud
cover J. Atmos. Sci., 59, 1917–1942.
[23] Jakob, C. (2003). An improved strategy for the evaluation of cloud parameterizations in
GCMs Bull. Amer. Meteorol. Soc., 84, 1387–1401,
doi: 10.1175/BAMS-84-10-1387
[24] Illingworth, A et al. (2007). Cloudnet-Continuous evaluation of cloud profiles in seven
operational models using ground-based observations. Bull. Am. Met. Soc, 88, 883-
898 ,
doi:10.1175/BAMS-88-6-883
[25] Tiedke, M. (1993). Representation of Clouds in Large-Scale Models Mon. Wea. Rew., 121,
3040-3061,
Risk Analysis and Crisis Scenario Evaluation in Critical Infrastructures Protection 393
doi: 10.1175/1520-0493(1993)121
[26] Deidda, R. et al. (2006). Space-time multifractality of remotely sensed rainfall fields ,J. of
Hydrology, 322, 2–13,
doi:10.1016/j.jhydrol.2005.02.036
[27] Palmer, TN et al. (2005). Representing model uncertainty in weather and climate
prediction Annu. Rev. Earth Planet. Sci., 33:163–93,
doi: 10.1146/annurev.earth.33.092203.122552
[28] Palmer, TN (2002). The economic value of ensemble forecasts as a tool for risk
assessment: from days to decades. Q. J. R. Meteorol. Soc. 128:747–774,
doi: 10.1256/0035900021643593
[29] Palmer, TN et al. (2004). Development of a european multimodel ensemble system for
seasonal-to-interannual prediction (demeter), Bull. Am. Met. soc, 85,853-872,
doi: 10.1175/bams-85-6-853
[30] Palmer, TN (2000). Predicting uncertainty in forecasts of weather and climate Rep. Prog.
Phys., 63 (2), 71–116,
doi: 10.1088/0034-4885/63/2/201
[31] Zhu, YJ et al. (2002). The economic value of ensemble-based weather forecasts Bull.
Amer. Meteor. Soc., 83, 73–83.
[32] Hagedorn, R. et al. (2005). The Rationale Behind the Success of Multi-model Ensembles
in Seasonal Forecasting - I. Basic Concept Tellus, 57A, 219-233.
[33] Lambert, S. J. and G. J. Boer (2001). CMIP1 evaluation and intercomparison of coupled
climate models Climate Dynamics, 17, 83-106,
doi: 10.1007/PL00013736
[34] de Roo, A. et al. (2003). Development of a European Flood Forecasting System Int. J.
River Basin Manage, 1, 49–59.
[35] Gouweleeuw, B. et al. (2005). Flood forecasting using probabilistic weather predictions.
Hydrol. Earth Syst. Sci., 9, 365–380.
[36] http://www.irriis.org
[37] http://www.diesis-project.eu/
[38] Usov, A. Beyel, C., Rome E., Beyer U., Castorini E., Palazzari P., Tofani A., The DIESIS
approach to semantically interoperable federated critical infrastructure simulation
Second International Conference on Advances in System Simulation (SIMUL), 22-27
August 2010, Nice, France. Los Alamitos, Calif. [u.a.]: IEEE Computer Society, 2010,
pp. 121-128 ISBN 9780769541426.
[39] Tofani, A. Castorini, E. Palazzari, P. Usov, A. Beyel, C. Rome, E. Servillo, P. An
ontological approach to simulate critical infrastructures, Journal of Computational
Science, 1 (4): 221-228, 2010
doi:10.1016/j.jocs.2010.08.001
[40] T. Rathnam, C.J.J. Paredis Developing Federation Object Models using Ontologies,
Proceedings of the 36th Winter Simulation Conference, 2004.
doi: http://doi.ieeecomputersociety.org/10.1109/WSC.2004.1371429
[41] R.K. McNally, S.W. Lee, D. Yavagal, W.N. Xiang Learning the critical infrastructure
interdependencies through an ontology-based information system, Environment and
Planning B: Planning and Design, 34, pp. 1103-1124, 2007
doi:10.1.1.132.5215
Part 6
Miscellaneous Case Studies

20
Applications of Decision Support System in

Aviation Maintenance
Peng Zhang, Shi-Wei Zhao, Bin Tan, Li-Ming Yu and Ke-Qiang Hua
Civil Aviation University of China,Tianjin
China
1. Introduction
The basic purpose of aviation maintenance is to prevent the decreased levels of original
aircraft airworthiness, safety and reliability design at the lowest cost, which is of great
significance to the running of air transport enterprises. According to statistics, aviation
maintenance accounts for approximately 20% to 30% of direct operating costs, not including
indirect costs caused due to maintenance, such as flight delays, material and spare parts,
corporate image.
In recent years, as the rapid development of civil aviation industry, a large number of new
technologies have been used in airplanes, which increase the complexity of the civil aviation
maintenance, especially the application of digital information technology. In response to
modern aviation maintenance of the new situation, so as to adapt to the future development
trend of digital maintenance and improve the efficiency and quality of modern aircraft
maintenance, establishing the necessary modern air transport aircraft maintenance decision
support system becomes increasingly urgent. For this reason, this paper in-depth analysis
the problem of how to build the aviation Maintenance Decision Support System (MDSS).
2. The components of aviation maintenance decision support system

With modern communication technology, fault diagnosis technology and modern control
technology, aviation maintenance decision support system can achieve real-time data
exchange between air and ground, and it is able to provide effective decision support for
rapid fault diagnosis and digital maintenance. This system should be composed of
information acquisition and processing, aircraft maintenance support, material
management, maintenance management, maintenance event evaluation, etc. As shown in
Figure 1.
2.1 Information acquisition and processing

Aviation maintenance decision support system needs a variety of real-time data and
historical data for decision support judgments, and all these data are collected by
information acquisition and processing module.
2.1.1 Real-time data acquisition and processing

With the large number of information technology applied in civil aircraft, modern aircrafts
are equipped with central maintenance computer, aircraft condition monitoring system and
other real-time aircraft state parameters acquisition and monitor systems, which make it
possible to monitor aircraft state parameters during the whole flight. Specifically, the
collected aircraft state parameters are inputted to the aircraft online fault diagnosis system,
which diagnoses the fault of aircraft by the established online fault diagnosis model. The
result should be real-time transmitted to the ground system through the air-ground data
link (such as Aircraft Communications Addressing and Reporting System(ACARS)), so as to
provide decision support for ground maintenance and achieve rapid repair of aircraft
failure, thus improving efficiency in the use of aircraft.
Aviation Maintenance
Information Aviation Maintenance
Aviation Maintenance Air Material Management Maintenance Event
Acquisition Engineering
Technique Supporting and Control Evaluation
and Processing Management
Real-time Data Acquisition and Processing
Historical Data Collection and Processing
Air Material Inventory Control Strategy

Fault Methods and Models Evaluation

Fault Diagnose Methods and Models
Technology Electronic Documents
Interactive Aviation Maintenance
Maintenance Task Distribution

Maintenance Decision-Making
Maintenance Task Coordiating

Maintenance Plan Making
Air Material Loss Model
Execution
Evlation
Aviation Maintenance Decision Support System
Fig. 1. Aviation maintenance decision system diagram
2.1.2 Historical data acquisition and processing

The process of aviation maintenance will accumulate a large amount of historical data.
These data can be divided into three types, as shown in Figure 2.
Fig. 2. Historical data acquisition and processing diagram

1. State parameters information of the flight is usually closely linked to aircraft failure
recorded by Quick Accesses Recorder (QAR), through the in-depth analysis of recorded
flight status parameters, we can learn the condition of airplane in flight. Using the
causal relationship between the system failure symptoms and the cause of the
malfunction, we can create online aircraft fault diagnosis model to provide decision
support for online aircraft fault diagnosis, and based on which establish a on-board
fault diagnosis system.
Applications of Decision Support System in Aviation Maintenance 399
2. Aircraft maintenance records are the depiction of the overall evolution of the aircraft and
systems. Aircraft maintenance record data can be used to obtain deep relationship
between the system failure symptoms and the cause of the malfunction, it becomes
particularly useful when comes to learn the relationship between system and
environment, and the interaction chain relationships between system and system, that
makes it possible to conduct more comprehensive and in-depth fault diagnosis on the
basis of online fault diagnosis, so as to provide firm decision support for the ground
maintenance.
3. Air material is the material of aviation maintenance, in order to achieve the complexity
capabilities and high-performance of modern aircraft, there are many types of aircraft
systems, while different types of air material have different price, different loss of
regularity and different storage conditions (the required temperature, humidity, storage
time). Thus the improper selection of air material control strategy may lead to
deficiency or excess of air material, on the other hand, if make good use of air material
historical consumption data, then we can understand the status of air material loss and
establish an appropriate procurement strategy to maintain a reasonable level of air
material reserve, to improve aviation material utilization and the efficiency of
enterprises.
2.2 Aviation maintenance technical support

Aviation maintenance technical support modules are based on the collected and processed
data and the existing maintenance resources. Specifically, it can be divided into two aspects:
2.2.1 The fault diagnosis model and method

Fault diagnosis model is divided into on-line and off-line diagnosis model. On-line
diagnosis model receives history aircraft state parameters from data acquisition and
processing module, through data acquisition and processing module for the aircraft state
parameters of the process history data, using more sophisticated expert system, fuzzy
identification and other traditional fault diagnosis, established to identify different systems
and components of the on-line fault diagnosis model based on the case of actual operation,
can be used time of diagnosis, diagnostic accuracy and other parameters on-line fault
diagnosis model to evaluate the results based on evaluation of diagnostic model of the
online database to be updated. In addition, based on economic considerations, it should also
be established on-line diagnosis based on Minimum Equipment List (MEL), Maintenance
Steering Group 3 (MSG-3), etc. to develop appropriate law judge to determine the "serious
failure" of the fault code list immediately through the ACARS system, then it will be
transferred to the fault code list issued ground, or for storage, such as the plane landed
using WirelessFidelity /Personal Digital Assistant (WiFi/PDA) and other technical means of
transmission fault code list, reducing the flow of ground data link and improve the efficient
use of ground data link.
Off-line fault diagnosis model based on information acquisition and processing module of
the aircraft maintenance record information obtained through in-depth analysis of the
system fault laws, particularly between the system and the environment, systems and cross-
linking between the laws of the system, the establishment of fault symptoms, and the causal
relationship between the fault types, the use of artificial neural network, Petri net fault
diagnosis technology of modern aircraft comprehensive diagnostic data. Diagnosis entered
into the aviation maintenance technical documentation for the interactive system, direct
access to a viable maintenance program.
In addition, according to the system with historical fault data can also use trend analysis,
curve fitting, time series and other methods for different types of faults predict the trend
predicted results using the fault diagnosis can be assessed for failure diagnostic support.
Aircraft Real-time Status

Information
Online Diagnose
Online Fault
Diagnose Model System End
Library
N Maintenance Reasonable
Unreasonable Fault Practise
Y Offline Fault Evalution
Online Fault
Diagnose Model Add to Fault Code List Diagnose
Reasonable
Evaluation Result
Reasonable
Offline Fault Diagnose
Fault Offline Fault
System
Classification Diagnose Model
End Based on Library
Not
MEL/MSG-3 serious
Serious
Storage Fault
Transmission Code Lists
Fault Code
List By
ACARS Fault Code Wifi PDA
Lists
Transmission
On Ground
Fig. 3. Diagram of air-ground fault diagnosis
2.2.2 Maintenance technical documentation

The aviation maintenance technical documentation is another important aspect of aviation
maintenance technical support, In particular, maintenance of aviation interactive electronic
documents. Modern aircraft is an integrated body of advanced manufacturing technology,
process technology, and information technology. It produces and uses lots of technical
information in the design, production, use and maintenance process, for example, data from
the manufacturer including Aircraft Maintenance Manual(AMM), Fault Reporting Manual
(FRM), Fault Isolation Manual (FIM), Wiring Diagram Manual (WDM), System Scheme
Manual (SSM), Illustrated Parts Catalogs (IPC), Master Minimum Equipment List (MMEL),
etc; technical data from time to time update including Service Bulletin (SB), Service Letter
(SL), etc; council documents including Federal Aviation Regulations (FAR, CCAR),Advisory
Circular (AC), Airworthiness Directives (AD), etc. According to their own conditions,
Airlines refer to the manufacturer and the requirements of the council to develop various
documents, which are the important tools and resources to support the maintenance, for
example, Minimum Equipment List (MEL), Operations Specifications (OS), Technology
Policy and Procedures Manual (TPPM), Check Manual (TM), Outline Reliability Manual
(RPM). Traditionally, many of these technical documents use paper for the storage medium,
or in part, have been digitized, but a large number of technical documents have problems of
queries difficult, not easy to carry and preservation and difficult to update.
Interactive aviation maintenance technical manual integrates advantages of technology of
application computer, multimedia, database and network, organize and manage complex
content operating manuals and maintenance information according to relevant standards,
implement accurate performance of required information of maintenance staff with
optimized way. The purpose of it is to accelerate the progress of aviation maintenance,
enhance the effectiveness of maintenance.
Fig. 4. The type of maintenance technology documents
2.3 Management and control of air materiel
Fig. 5. Management and control of air materiel modelling

Air material inventory, which is the highest cost in aviation maintenance process, is an
important component of aviation maintenance process for the complexity of modern aircraft
components, variety of aircraft types and materials, and the use of different rules of air
material, different storage requirements and flight safety. Therefore, control of air material
has great influence on the efficiency of enterprises. Through we use historical information
for information acquisition and processing module to get air materiel, it can gain the
regularity of loss and buy of air materials using techniques of statistical analysis and data
mining, such as the loss regularity of repairable air materials and unrepairable air materials,
the maintained regularity of maintained air materials, etc., then through the rational control
strategy, it establishes the corresponding loss mode, implements optimization of air material
inventory and improve the quality of maintenance.
In addition, air material inventory control should consider the method of shared regional
aviation maintenance units, especially with high value, and there may be more serious
problems with lack of critical flight equipment, such as aircraft engines, etc. Through
modeling and analysis of historical data, it can construct a reasonable model for regional air
material sharing, and then contribute to enhance the efficiency of the industry for the whole
region.
2.3.1 Aviation maintenance management

Aviation maintenance management is divided into maintenance program development,
allocation of maintenance tasks, and maintenance tasks scheduling. Under the premise of
security, the main goal of aviation maintenance is economic. A fast, highly reliable
maintenance can increase the number of aircraft flight hours, thereby enhancing economic
efficiency. Aviation Maintenance Management Module manages and controls the
maintenance project based on the summarization of maintenance tasks from technical
support module.
Maintenance plan is the core of maintenance tasks. It determines the implementation of
maintenance methods based on the level of maintenance tasks, the existing maintenance
resources (manpower, equipment, facilities, material, etc.) and the aircraft status.
Maintenance plan includes many aspects, such as carrying out step by step or in parallel
ways, the name of maintenance, repair content, maintenance workers, and materials
required for the maintenance, etc. Maintenance plan makes reasonable arrangements for the
future state of repair to ensure the smooth implementation of aviation maintenance.
Concrete repair work is completed by mechanical, electronic, structural, accessories
maintenance workshop. The allocation of maintenance tasks are to maintain a reasonable
distribution of work to the appropriate workshop, to enable the repair shops carrying out
the maintenance work step by step or in parallel way and ensure the smooth progress of
maintenance projects.
In the actual maintenance projects, maintenance programs will inevitably be adjusted,
because the schedule might be unreasonable, or something’s comes up as emergency or
accident. Maintenance scheduling is to use reasonable scheduling strategy and appropriate
decompose or combine maintenance tasks or just change the sequence to ensure the smooth
implementation of maintenance works in accordance with the progress of the plan.
2.3.2 Maintenance results evaluation

Aviation maintenance decision support system is an open system. To achieve an objective
and reasonable assessment, there should be a reasonable maintenance event parameters
evaluation system, thus achieving optimal and update maintenance decision support
system. Evaluation of maintenance events should contain at least three aspects: (1)
Applicability of on-line fault diagnosis model. On-line fault diagnosis system is an on-board
system, so its diagnosis should be real time and needs clear failure description while its
relative diagnosis accuracy is not high. (2) The applicability of off-line fault diagnosis model.
Off-line fault diagnosis system has sufficient resources and time to judge the on-line
diagnosis system’s results in detail, therefore, off-line fault diagnosis system is not critical on
the diagnosis time, but the accuracy of diagnosis must be high, thus reasonable maintenance
tasks can be determined, therefore, the effectively maintenance can be implemented and
maintenance costs will be reduced. (3) Reasonable air materiel control model. Air material
involves in various aspects, such as mechanical components, electronic components, and
there are repair materials and maintenance consumables. Reasonable air materiel
configuration not only meets the maintenance requirements, but also reduces the cost from
procurement, storage, transportation, etc. Otherwise the airline may have a very negative
impact on operations.
3. Theory and examples of decision support system of aviation maintenance

The decision support system of aviation maintenance, which has the properties of discrete
event dynamic systems and continuous time dynamic system, is a typical hybrid system.
The system of acquisition and processing of real-time data should contain functions of the
acquisition of parameters information, filtering, transformation and so on, and it should use
theory of continuous variable dynamic systems to construct, while acquisition and
processing of history data mostly use relevant theories of statistics and analysis. It contains a
lot of people work in the system of management and control of air materials, such as
maintenance process, procurement process, etc., and has obvious properties of discrete
event dynamic system, but through reasonable conversion, it can use a variety of modern
control methods of minimum variance control, optimal control, stochastic control, adaptive
control and robust control to implement. Aviation maintenance technical documents may
pay more attention to technology and theory of multimedia, database/data warehouse.
Simultaneously, because systems using discrete event dynamic system property can not be
expressed using the traditional problem of parsing, it can use intelligence theory to solve,
for example, fault diagnosis can use theories of neural network, Petri nets, fuzzy reasoning,
agent, and so on.
With the development of theory and technology of computer simulation, it can use more
approaches for the research of the model and method of aircraft fault diagnosis, the model
of air materiel use and the air material inventory control, decision support system of
aviation maintenance, and use full-simulation analysis of the system to establish a rational
and efficient decision support system for aviation maintenance. The following is a certain
type of aviation turbine jet engine fault diagnosis model of aircraft tire building process and
inventory control process as an example.
3.1 Aircraft fault diagnose based on adaptive FPN

Aero-engine, which has structural complexity, possesses a number of components, works in
the high temperature and harsh environment, so it may be fault anytime. When there is
fault, because of the number of similar fault, it makes difficulties for aero-engine
troubleshooting. For example, a certain type of air carrier turbojet has the fault of fuel
consumption of large slip, it can use adaptive Fuzzy Petri Nets (FPN) to establish fault
diagnosis model to position the cause of such fault. The process of model setting up as
follows:
(1) The definition of model input variables and output variables, the model input variables
is the logic state of fault symptoms, defining as
X   x1 x2 x3 x4 x5 x6 x7  ,
The meaning as shown in table 1; model output variables is degree of confidence of fault
causes, defining as
Y   y1 y2 y3 y4 y5  ,
The meaning is as shown in table 2.
Variable name x1 x2 x3 x4
Oil in nozzle at Oil at compressor Oil at White smoking
Symptom of
the bottom of entrance, fore tank airplane at aft pressure
Phenomenon
turbine bottom skin tail cone reducer
Variable Name x5 x6 x7
Oil beads at Ejector tube
Symptom of Flight in the air, oil
centrifugal emitting purple
Phenomenon consumption heavy
ventilator’s vent smoke
Table 1. Definition of fault symptom and input variables
Variable
y1 y2 y3 y4 y5
Name
Oil pump’s Centrifugal In bearing Three class

One-way valve of
Fault one-way valve cavity, scavenging
radiator interface not
Cause valve not membrane pressure too pump
sealing
sealing rupture high inefficient
Table 2. Definition of fault cause and output variables

(2) Initializing the model parameters, membership parameters of input fault symptoms and
fault cause, this is given by the expert system and repair factory’s statistical data,
respectively set to
D0   d01 d02 d03 d04 d05 d06 d07  and D0   d11 d12 d13 d14 d15 d16 d17  ,
as shown in table 3 and table 4.

(3) According to the fuzzy inference rules of FPN, it defines network algorithms on each
floor, uses the maximum degree of membership and the principle of membership grade
threshold (γ>0.8) and combines with the definition of fuzzy algorithms to construct the fault
diagnosis model FPN.
(4) According to the actual maintenance condition, amend membership between fault
symptom and fault causes in initial model parameters, and the amendment formula is as
follows:
times of i symptom which belong to j reason

d1i  { } (1)
the total times of i symptom
The parameters mean i  1, 2, ,7; j  1, 2, , 5 . It can make FPN amend model parameters
online in accordance with fault occurrence.
Fault Symptom xi Fault Cause yi

Membership Grade d0i
x1 x2 x3 x4 x5 x6 x7 y1 y2 y3 y4 y5
1 0 0 0 0 0 0 d01 0.65 0.35 0 0 0
0 1 0 0 0 0 0 d02 0.30 0.50 0 0.20 0
0 0 1 0 0 0 0 d03 0 0.20 0.75 0 0.05
0 0 0 1 0 0 0 d04 0.30 0 0 0.56 0.14
0 0 0 0 1 0 0 d05 0.20 0 0.50 0.30 0
0 0 0 0 0 1 0 d06 0.47 0 0.28 0 0.25
0 0 0 0 0 0 1 d07 0.10 0.30 0 0 0.60
Table 3. The membership grade relationship statistics between fault symptoms and fault
causes in the repair factory
Fault Symptom xi Fault Cause yi

Membership Grade d0i
x1 x2 x3 x4 x5 x6 x7 y1 y2 y3 y4 y5
1 0 0 0 0 0 0 d11 0.80 0.20 0 0 0
0 1 0 0 0 0 0 d12 0.20 0.65 0 0.15 0
0 0 1 0 0 0 0 d13 0 0.20 0.80 0 0
0 0 0 1 0 0 0 d14 0.25 0 0 0.70 0.05
0 0 0 0 1 0 0 d15 0.12 0 0.20 0.68 0
0 0 0 0 0 1 0 d16 0.22 0 0.66 0 0.12
0 0 0 0 0 0 1 d17 0.22 0.10 0.10 0 0.58
Table 4. Expert database of membership grade relations between fault symptoms and fault
causes
According to the above flow, adaptive FPN fault diagnosis model is set up as shown in
Figure 6. Using MATLAB and 18 fault symptom state vectors, simulation is carried out,
simulation results and the actual cause confirmed in maintenance as shown in Table 5, the
diagnostic accuracy rate is 100%. This shows that the reliability of model diagnosis is
relatively high. On the other hand, item 7 and 10, there are some deviation between output
of fault model diagnosis and result of actual maintenance, because take method of threshold
membership grade to eliminate fuzzy, diagnostic efficiency little decreases, but at the same
time to avoid impact of random factors, which could result in the deviation of diagnostic
results.
d 01 d11 d 02 d12 d 03 d13 d14 d 04 d15 d 05 d16 d 06 d17 d 07

t001 001 t011 011 t002 002 t012 012 t003 003 t013 013 t014 014 t004 004 t015 015 t005 005 t016 016t006 006 t017 017 t007 007
d1 x1 d2 x2 d3 x3 d4 x4 d5 x5 d6 x6 d7 x7
t11 11 t12 12 t13 13 t14 14 t15 15 t16 16 t17 17
x21 x22 x23 x24 x25 x26 x27
t21 21 t22 22 t23 23 t24 24 t25 25 t26 26 t27 27
x31 x32 x33 x34 x35 x36 x37 x38 x39 x40 x41 x42 x43 x44 x45 x46 x47 x48 x49 x50
t31 31 t32 32 t33 33 t34 34 t35 35
y1 y2 y3 y4 y5
t53 53 t54 54
t41 41
t52 t55  t56
55
52 56
t51 t57
51 57
Fig. 6. Fault diagnosis model of adaptive FPN
Serial Number 1 2 3 4 5 6
Fault Symptom
[0 1 0 1 0 0 0] [1 0 0 0 1 0 0] [0 1 1 0 0 0 1] [1 0 0 1 0 0 0] [1 0 1 0 0 0 0] [1 0 0 0 0 0 1]
Matrix X
Diagnosis Cause y4 y4 y2 y1 y3 y1
Actual Cause y4 y4 y2 y1 y3 y1
Serial Number 7 8 9 10 11 12
Fault Symptom
[1 1 0 0 0 0 0] [0 0 1 0 0 1 0] [1 0 1 0 0 0 1] [0 1 0 1 0 1 0] [1 0 0 0 0 1 0] [0 0 1 0 1 0 1]
Matrix X
Diagnosis Cause y1 , y2 y3 y1 , y3 y1 , y4 y1 y3
Actual Cause y2 y3 y1 , y3 y4 y1 y3
Serial Number 13 14 15 16 17 18
Fault Symptom
[0 0 1 1 0 1 0] [1 0 0 0 1 1 0] [0 0 0 1 0 0 1] [0 0 0 1 0 1 0] [0 1 0 1 1 0 0] [0 0 1 01 0 1]
Matrix X
Diagnosis Cause y3 y1 , y3 y5 y4 y4 y3
Actual Cause y3 y1 , y3 y5 y4 y4 y3
Table 5. Comparison table of model diagnosis result and actual cause

Through the theory of adaptive FPN, it analyzes the cause of turbojet engine heavy oil
consumption, sets up a heavy oil consumption fault symptoms - fault cause diagnostic
model, and gives an intuitive expression of complex relationship between the fault cause
and fault symptoms. It has reality significance to the aviation maintenance factory fast
locating fault cause, improving maintenance efficiency and reducing maintenance costs of
enterprises. And it also provides a new idea for the further development of the aviation
maintenance decision-making support system.
3.2 Tire inventory management and control model

Suppose that the same type of aircrafts in a fleet of airline “A” consume and replace tires
everyday, amount of which is X1 (k) ( k is a date serial number ),and X1 (k) is the random
process. Considering the regularity of flights, we can think that there is no relationship
between loss of statistical laws of tires and dates from the engineering point of macro view.
At the same time, it is considered that the amount of tire loss of any day does not affect the
amount of tire loss of any day in the future, that is, X1 (k) and X1 (k+1) (h ≠ 0) are
independent.
Based on above given conditions, obviously, tires required changing everyday is followed
Poisson distribution, so it is a Poisson process. For controlling air material tires inventory
needs new tires supplement at a certain period, or refresh used tires. Set the period is M
days, we should consider the statistical character of process XM (k) = X1 (k) + X1 (k+1)+ ……
+ X1 (k + M -1). X1 (k) , X1 (k+1) , …… , X1 (k + M -1) is independent of each other, followed
the same distribution, and has limited mathematical expectation and variance. According to
central limit theorem, when M tends to infinity (actually, as long as M is enough large) XM
(k) obeys the normal distribution. It is that XM (k) is followed normal random process.
It should be noted that the different fleets, different aircrafts have different X1 (k). As
different airlines with different route, taking off and landing frequency of flights are quite
different every day from twice a day to dozen a day, and different aircrafts may have a huge
difference in tire wear, therefore, the tire’s loss properties statistical of different fleet are
different.
After the acquisition of air material inventory and maintenance data from a number of
airlines, it uses Statistical Product and Service Solutions (SPSS), MATLAB and other
software for data statistical processing. A case of airline “A” is analyzed. We use the tire loss
data of a fleet which includes ten same aircrafts from March 1, 2008 to November 1, 2008.
After data analysis, there are tables as follows, Table 6 is about the data, and Table 7 is about
the results of fitting analysis with normal process, even process, Poisson process, and
exponential process.
Analysis below shows:
1. From Table 7, tires loss process is a random process and has a very high fitting degree
with Poisson process.
2. As can be seen from Table 7, with the increase in computing cycles, the fitting degree
with the normal process is increasing.
3. From the autocorrelation analysis, we learn that when the delay is 0, the correlation
coefficient is 1, the relevant degree is less than 0.2 under the rest delay.
We can get the conclusion from below analysis as follows:
1. Taking the date serial number as the independent variable, tires loss time-series can be
handled as a Poisson process.
2. Tire loss data correlation is weak between different days. Thus from an engineering
point of view, the loss of tire number of different dates are independent.
3. With the increase of computing cycles, random distribution tends to normal
distribution. Application of engineering control theory to solve the problem of optimal
inventory seems necessary.
Date Quantity 3 Days 7 Days 10 Days 20Days

3.1 1
3.2 1 2
3.3 0
3.4 0 2
3.5 0 0
3
3.6 0
3.7 0
3.8 1 1
3.9 0
3.10 0
3.11 1 1 2
3.12 0
3.13 0
3.14 0 2
……
3.15 2 ……
4
3.16 0
3.17 1 1
3.18 0 3
3.19 0
3.20 0 0
3.21 0
3.22 1
3.23 0 1
3.24 0
3.25 0 2 2
3.26 0 1
3.27 1
3.28 0
0
3.29 0 4
Table 6. The fleet tires loss situation (2008.3.1—2008.11.1)
Excessive tire inventory will increase storage costs, administrative costs, regular test costs
and renovation costs, etc. On the other hand, the lack of tire inventories will cause difficulty
in timely maintenance and affect the flight delays, and cause economic and social loss, thus,
the optimal inventory levels need to be determined.
Normal Process Uniform Process Poisson Process Index Process

Period
Fitting Degree Fitting Degree Fitting Degree Fitting Degree
1 Day 0.000 0.000 0.981 0.000
3 Days 0.001 0.000 1.000 0.000
7 Days 0.351 0.000 0.999 0.001
10 Days 0.862 0.147 1.000 0.027
20 Days 0.781 0.180 0.930 0.011
Table 7. Fitting degree of tires loss data compared to different stochastic process
Figure 7 to Figure 10 shows the correlation analysis result of data
Fig. 7. Auto-correlation at period of 1 day Fig. 8. Auto-correlation at period of 3 days
Fig. 9. Auto-correlation at period of 7 days Fig. 10. Auto-correlation at period of 10 days

Considering the long period of tire repair, we take tire inventory as a normal distribution
process. Table 8 is the assumptions and parameters of the optimal inventory analysis.
Symbol Descriptions of volume and parameter

N0 Average inventory
s Stock volume variance
G1 Inventory costs（100 ¥ / a tire）
G2 The loss of delay for the lack of tires（10000 ¥ / a tire）
k Number of tires
F Total cost
Table 8. Analysis of assumptions and parameters of the optimal inventory
Because the event of tires shortage is a small probability event, it sets inventory costs for
N0G1. For convenience of calculating, it is considered that the number of lack tires is no
more than 5, and corresponding loss of the lack of tires is as follows:
5   k  N 0 2 5  k  N 0 2
1  k 
G2  k  G2 
2
e 2 e 2 2 (2)
k 1 2 k 1 2
Then total cost is:
5  k  N 0 2
k 
F  N 0G1  G2  e 2 2 (3)
k 1 2
As can be seen from the equation, when the average inventory increases, the first increase,
the second reduction, when the average inventory reduction, the first decrease, while the
second increase. When stocks Mean Square Error (MSE) is constant, it can calculate the
optimal value of the average stock. Table 9 gives the MSE of different stocks and optimal
average inventory.
σ 1 2 3 4 5 6 7 8 9
N0 2-3 4-5 6-7 9-10 11-12 13-14 14-15 16-17 18-19
Table 9. Optimal average inventory for the MSE of different stocks

Analysis above shows:
(1) Obviously, with the increase of the variance of inventory, optimal average inventory is
straight up and inventory cost increases rapidly; (2) For different aircraft types, parameters
has large difference, and inventory cost is not the same, so optimal inventory levels will be
relatively large different; (3) It can be seen from qualitative analysis of the theory, when the
period calculated increases, average inventory corresponding increases, then the MSE of
stocks also increases, nonlinear relationship between the two, but not independent.
Therefore, it uses the method of searching to solve the optimal value. Through theoretical
analysis and experimental data analysis, it shows that the process of tire loss can be seen as
independent Poisson process in engineering. When the period calculated (in fact, tire repair
period) is large, it is considered that this process is independent normal random process. In
addition, through the optimal analysis and calculation of optimal average inventory for the
MSE of different stocks, it can be seen that if it uses the method of reducing the MSE of stock
to control inventory, it can reduce the total cost.
4. Prospects and challenges

In recent years, based on widely using of the maintenance theory of on-condition, condition
monitoring, etc., it proposes maintenance concepts, such as digital maintenance, proactive
maintenance and so on, which supported by strong monitoring and diagnosis techniques to
improve the efficiency of maintenance greatly, so there must be strong decision support
system to construct modern aviation maintenance system.
MDSS can provide effective decision support for aviation maintenance. It also improves
maintenance efficiency and controls maintenance costs. However, the aviation MDSS still
needs further research on the following topics:
1. Because modern aircraft has many systems and complex structure, many flight
parameters need to monitor. How to build efficient database and research criteria for
judging the effectiveness of real-time data should conduct further research.
2. Because real-time flight status data downloaded relate to cost problems, how to
establish the fault information download criteria should conduct further research.
3. The system has a property of hybrid system, and there are many ways for DEDS and
CVDS modeling, but theory and method of modeling the hybrid system is still needed
to be improved, so how to organic combine CVDS system with DEDS system and
modeling the hybrid system should conduct further research.
4. MDSS is closely related with the airline operating environment, as there are obvious
differences between airlines, how to select appropriate parameters, and how to
overcome the bad applicability should conduct further research.
5. References
Zhang Yalin. & He Yizheng. (2007). Maintenance Information Management System. Avionics
Technology, Vol.38, No.2, (June 2007), pp. 41-45
Qi Yanjie.; Lv Zhigang. & Song Bifeng. (2008). Design for Maintenance on Modern Airplane.
Journal of Civil Aviation University of China, Vol.26, No.5, (October 2008), pp. 5-9
Qi Yanjie.; Lv Zhigang. & Song Bifeng. (2010). Ultimate Service Life：New Concept for
Aircraft Life Management. Journal of Civil Aviation University of China, Vol.28, No.2,
(April 2010), pp. 22-28
Harry A. Kinnison,Ph.D. (2004). Aviation Maintenance Management, The McGraw Hill
Companies, Inc, ISBN 0-07-142251-X, New York, USA
Sun Li. (2008). Remote Management of Real-Time Airplane Data. Aviation Maintenance &
Engineering, (January 2008)
Tan Xuehua. & Wang Huawei. (2008). Research on Frame of Plane’s Maintenance
Programme Optimization. Computer Technology And Development, Vol.18, No.11
(Nov. 2008), pp. 183-186
Zhu Rui. & Xu Chunsheng.(2005). An Approach for Aircraft Long-Range Real-time

Monitoring. Aviation Maintenance&Engineering,(March 2005),pp.53-55
Zhang Peng. (2004). A Study of Aircraft Health Status Evaluation Based on Flight Data
Trend Monitor. Journal of Air Force Engineering University(Natural Science
Edition),(June 2004), Vol.5, No.3(June 2004), pp. 8-10
Zhu Yanbo.; Zhang Xiaolin. & Tan Xijing. (2006). An Electronic Manual Analysis Method for
Aircraft Maintenance Decision Supporting System. Maintenance&Engineering,(April
2006),pp.28-30
Chang Fei.; Han Qing. & Shang Bolin.(2008). Application of FBG on Flight Vehicle structural
Heath Monitoring Systems. Science Technology and Engineering, Vol.8, No.10(May
2008), pp. 2641-2646
Sun Haidong. & Lv Jian. (2009). On Critical Technologies for Health Management Expert
System of Avionic System. Electronic Optics & Control, Vol.16, No.7(May 2009), pp.
40-44
Fu Keya. & Li Benwei. (2006). Fuzzy Comprehensive Decision Method of Aero engine
Maintenance. Aircraft engine, Vol.33, No.4(2007), pp.55-58
Ma Xiaofeng.(2009) Intelligent Decision Support Systems Design and Analysis of Aeronautic
Equipment Maintenance. Control and Automation Publication Group, Vol.25,
No.7(2009), pp. 25-27
Wang Yu. & Wang Xiaofeng. (2001). Ship Maintenance Decision Support System Based on
Data Base. Computer Aided Engineering, No.4 (Dec. 2001), pp. 6-14
Zhang Tao.; Guo Bo. & Tan Yuejin.(2004). Research on a Mission oriented Maintenance
Resources Deployment Decision Support System. Acta Armamentarii, Vol.26,
No.5(2005), pp. 716-720
Cao Yu.; Liu Yile. & Xu Zongchang.(2003). Realization of Visualization for Maintenance
Intelligence. Computer engineering and design, Vol.24, No.5(May 2003), pp.50-52
Sun Ning. Airplane Health Management in Air China B747-400 fleet operation support.
Maintenance, Vol.114(June 2010), pp.65-67
Zhu Rui. The Research for the Application of Data Mining in Aircraft Health Management.
Information Technology, (April 2007), pp.36-38
Zhao Ruiyun. (2005). Concept and Application of Central Maintenance System. Avionics
Technology, Vol.36, No.2(June 2005), pp. 14-18
Ye Yonglin.; Yang Xueqiang.; Huang Jun. & Liu Xing. (2008) Model Base Design in
Equipment Maintain ，Assistant Decision, Vol.22, No.4(Aug 2008), pp. 92-94
Cao Zhiqiang.; Li Zhao. & Han Hongwei. (2008) Research on Equipment Repairing
Intelligent Decision Support System. Radio Engineering, Vol.38, No.12(2008), pp. 5-7
Hua Keqiang. & Qu Wenfang.(2010). Aviation Supplies Inventory Data Analysis and the
Study of the Optimal Inventory. Aviation Maintenance & Engineering, (April
2010),pp.68-70
LV Changling. & Xie Bangchang. (2009). Data Mining Method & Application, China Renmin
University Press, ISBN978-7-300-09970-5, Beijing, China
Zhang Peng.; Zhao Shiwei. & Wang Yake.(2009). The Research of Aircraft Fault Diagnosis
Based on Adaptive FPN. Chinese control and decision conference, ISBN 978-1-4244-
2723-9, China, June, 2009
21
Evaluating the Power Consumption in

Carbonate Rock Sawing Process by
Using FDAHP and TOPSIS Techniques
Reza Mikaeil1, Mohammad Ataei1 and Reza Yousefi2
1Faculty of Mining, Petroleum & Geophysics, Shahrood University of Technology, Shahrood
2School of mechanical engineering, Sharif University of Technology, Tehran
Iran
1. Introduction
The prediction of rock sawability is important in the cost estimation and the planning of the
stone plants. An accurate estimation of rock sawability helps to make the planning of the
rock sawing projects more efficient. Rock sawability depends on non-controlled parameters
related to rock characteristics and controlled parameters related to properties of cutting
tools and equipment. In the same working conditions, the sawing process and its results are
strongly affected by mineralogical and mechanical properties of rock.
Up to now, many studies have been done on the relations between sawability and rock
characteristics in stone processing. Norling (1971) correlated sawability with petrographic
properties and concluded that grain size was more relevant to sawability than the quartz
content. Burgess (1978) proposed a regression model for sawability, which was based on
mineralogical composition, hardness, grain size and abrasion resistance. Vaya and Vikram
(1983) found a fairly good correlation between the Brinell hardness test and diamond
sawing rates. However, the variables involved were so many that they believed no
mathematical solution would be possible. They also considered the Specific Energy (SE)
concept, in conjunction with mineralogy, to give a better understanding of the sawing
responses of various rock types. Ertingshausen (1985) investigated the power requirements
during cutting of Colombo Red granite in up-cutting and down-cutting modes. He found
out that the required power was less for the up- cutting mode when the cutting depth was
below 20–25 mm. For deeper cuts, however, the power consumption was less for the down-
cutting mode. Wright and Cassapi (1985) tried to correlate the petrographic analysis and
physical properties with sawing results. The research indicated cutting forces to have the
closest correlation. Birle et al. (1986) presented similar work in 1986, but again considered
only blade life as the criterion on which a ranking system should be based. Hausberger
(1989) concluded that an actual sawing test was the most reliable method for determining
the machinability of a rock type. He observed that the higher proportion of minerals with
well defined cleavage planes helps the cutting to be easier. Jennings and Wright (1989) gave
an overall assessment of the major factors which affect saw blade performance. They found
out that hard materials usually require a smaller size diamond than do softer stones because
the load per particle is not sufficiently high and greater clearance is required for swarf.
Conversely, if large diamond grits are used on hard materials, the penetration of the
diamond is limited, and normally either excessive grit pull-out will occur or large wear flats
will appear on the diamond particles. Unver (1996) developed empirical equations for the
estimation of specific wear and cutting force in the sawing of granites. He used mean quartz
grain size, NCB cone indenter hardness number, and mean plagioclase grain size in his
equations. Clausen et al. (1996) carried out a study on the acoustic emission during single
diamond scratching of granite and suggested that acoustic emission could be used in
sawability classification of natural stones. They also concluded that the cutting process is
affected by the properties and frequency of minerals, grain size and degree of interlocking.
Tonshoff and Asche (1997) discussed the macroscopic and microscopic methods of
investigating saw blade segment wear. Luo (1997) investigated the worn surfaces of
diamond segments in circular saws for the sawing of hard and relatively soft granites. He
found out that for the sawing of hard granite, the worn particles were mainly of the macro-
fractured crystal and/or pull-out hole type. Ceylanoglu and Gorgulu (1997) correlated
specific cutting energy and slab production with rock properties and found good
correlations between them. Webb and Jackson (1998) showed that a good correlation could
be obtained between saw blade wear performance and the ratio of normal to tangential
cutting forces during the cutting of granite. Xu (1999) investigated the friction characteristics
of the sawing process of granites with diamond segmented saw blade. The results of the
experimental studies indicated that most of the sawing energy is expended by friction of
sliding between diamonds and granites. Xipeng et al. (2001) found that about 30 percent of
the sawing energy might be due to the interaction of the swarf with the applied fluid and
bond matrix. Most of the energy for sawing and grinding is attributed to ductile ploughing.
Brook (2002) developed a new index test, called Brook hardness, which has been specifically
developed for sliding diamond indenters. The consumed energy is predictable from this
new index test. Konstanty (2002) presented a theoretical model of natural stone sawing by
means of diamond-impregnated tools. In the model, the chip formation and removal process
are quantified with the intention of assisting both the toolmaker and the stonemason in
optimising the tool composition and sawing process parameters, respectively.
Li et al. (2002) proposed a new machining method applicable to granite materials to achieve
improved cost effectiveness. They emphasized the importance of the tribological
interactions that occur at the interface between the diamond tool surface and the workpiece.
Accordingly, they proposed that the energy expended by friction and mechanical load on
the diamond crystal should be balanced to optimize the saw blade performance. Xu et al.
(2003) conducted an experimental study on the sawing of two kinds of granites with a
diamond segmented saw blade. The results of their study indicated that the wear of
diamond grits could also be related to the high temperatures generated at individual cutting
points, and the pop-out of diamonds from the matrix could be attributed to the heat
conducted to saw blade segments. Ilio and Togna (2003) proposed a theoretical model for
the interpretation of saw blade wear process. The model is based on the experimentally
determined matrix characteristics and grain characteristics. The model indicates that a
suitable matrix material must not only provide the necessary grain support in the segment,
but it should also wear at an appropriate rate in order to maintain constant efficiency in
cutting. Eyuboglu et al. (2003) investigated the relationship between blade wear and the
sawability of andesitic rocks. In their study, a multiple linear regression analysis was carried
out to derive a prediction equation of the blade wear rate. They showed that the wear rate of
Evaluating the Power Consumption in Carbonate Rock
Sawing Process by Using FDAHP and TOPSIS Techniques 415
andesite could be predicted from the statistical model by using a number of stone
properties. The model indicated the Shore scleroscope hardness as the most important rock
property affecting wear rate. Xu et al. (2003) carried out an experimental study to investigate
the characteristics of the force ratio in the sawing of granites with a diamond segmented
blade. In the experiments, in order to determine the tangential and the normal force
components, horizontal and vertical force components and the consumed power were
measured. It was found out that the force components and their ratios did not differ much
for different granites, in spite of the big differences in sawing difficulty. Gunaydin et al.
(2004) investigated the correlations between sawability and different brittleness using
regression analysis. They concluded that sawability of carbonate rocks can be predicted
from the rock brittleness, which is half the product of compressive strength and tensile
strength. Ersoy et al. (2004, 2005) experimentally studied the performance and wear
characteristics of circular diamond saws in cutting different types of rocks. They derived a
statistical predictive model for the saw blade wear where specific cutting energy, silica
content, bending strength, and Schmidt rebound hardness were the input parameters of the
model. An experimental study was carried out by Xipeng and Yiging (2005) to evaluate the
sawing performance of Ti–Cr coated diamonds. The sawing performances of the specimens
were evaluated in terms of their wear performances during the sawing of granite. It was
concluded that the wear performance of the specimens with coated diamonds were
improved, as compared with uncoated diamonds. Delgado et al. (2005) experimentally
studied the relationship between the sawability of granite and its micro-hardness. In their
study, sawing rate was chosen as the sawability criterion, and the micro-hardness of granite
was calculated from mineral Vickers micro-hardness. Experimental results indicated that the
use of Vickers hardness microindentor could provide more precise information in sawability
studies. Mikaeil et al. (2008a and 2008b) developed a new statistical model to predicting the
production rate of carbonate rocks based on uniaxial compressive strength and equal quartz
content. Additional, they investigated the sawability of some important Iranian stone.
Yousefi et al. (2010) studied the factors affecting on the sawability of the ornamental stone.
Especially, among the previous studies some researchers have developed a number of
classification systems for ranking the sawability of rocks. Wei et al. (2003) evaluated and
classified the sawability of granites by means of the fuzzy ranking system. In their study,
wear performance of the blade and the cutting force were used as the sawability criteria.
They concluded that with the fuzzy ranking system, by using only the tested petrographic
and mechanical properties, a convenient selection of a suitable saw blade could be made for
a new granite type. Similarly, Tutmez et al. (2007) developed a new fuzzy classification of
carbonate rocks based on rock characteristics such as uniaxial compressive strength, tensile
strength, Schmidt hammer value, point load strength, impact strength, Los Angeles abrasion
loss and P-wave velocity. By this fuzzy approach, marbles used by factories were ranked
three linguistic qualitative categories: excellent, good and poor. Kahraman et al. (2007)
developed a quality classification of building stones from P-wave velocity and its
application to stone cutting with gang saws. They concluded that the quality classification
and estimation of slab production efficiency of the building stones can be made by
ultrasonic measurements.
The performance of any stone factory is affected by the complex interaction of numerous
factors. These factors that affect the production cost can be classified as energy, labour,
water, diamond saw and polishing pads, filling material and packing. Among the above
factors, energy is one of the most important factors. In this chapter, it was aimed to develop
a new hierarchy model for evaluating and ranking the power consumption of carbonate
rock in sawing process. By this model, carbonate rocks were ranked with the respect to its
power consumption. This model can be used for cost analysis and project planning as a
decision making index. To make a right decision on power consumption of carbonate rock,
all known criteria related to the problem should be analyzed. Although an increasing in the
number of related criteria makes the problem more complicated and more difficult to reach
a solution, this may also increase the correctness of the decision made because of those
criteria. Due to the arising complexity in the decision process, many conventional methods
are able to consider limited criteria and may be generally deficient. Therefore, it is clearly
seen that assessing all of the known criteria connected to the power consumption by
combining the decision making process is extremely significant.
The major aim of this chapter is to compare the many different factors in the power
consumption of the carbonate rock. The comparison has been performed with the
combination of the Analytic Hierarchy Process (AHP) and Fuzzy Delphi method and also
the use of TOPSIS method. The analysis is one of the multi-criteria techniques that provide
useful support in the choice among several alternatives with different objectives and criteria.
FDAHP method has been used in determining the weights of the criteria by decision makers
and then ranking the power consumption of the rocks has been determined by TOPSIS
method. The study was supported by results that were obtained from a questionnaire
carried out to know the opinions of the experts in this subject.
This chapter is organized as follows; in the second section, a brief review is done on concept
of the fuzzy sets and fuzzy numbers. In the third section FDAHP method is illustrated. This
section is included the methodology of FDAHP method. Fourth section also surveys TOPSIS
method. In the fifth section, after explanation of effective parameters on power
consumption, the FDAHP method is applied for determination of the weights of the criteria
given by experts. Then the ranking the power consumption of carbonate rocks is carried out
by TOPSIS method. Eventually, in sixth and seventh sections, results of the application are
reviewed. These sections discuses and concludes the paper. According to the authors’
knowledge, ranking the power consumption using the FDAHP-TOPSIS is a unique research.
2. Fuzzy sets and fuzzy numbers

To deal with vagueness of human thought, Zadeh (1965) first introduced the fuzzy set
theory, which was oriented to the rationality of uncertainty due to imprecision or
vagueness. A major contribution of fuzzy set theory is its capability of representing vague
data. The theory also allows mathematical operators and programming to apply to the fuzzy
domain. A fuzzy set is a class of objects with a continuum of grades of membership. Such a
set is characterized by a membership (characteristic) function, which assigns to each object a
grade of membership ranging between zero and one. With different daily decision making
problems of diverse intensity, the results can be misleading if the fuzziness of human
decision making is not taken into account (Tsaur et al., 2002) Fuzzy sets theory providing a
more widely frame than classic sets theory, has been contributing to capability of reflecting
real world (Ertugrul & Tus, 2007).
Fuzzy sets and fuzzy logic are powerful mathematical tools for modeling: uncertain systems
in industry, nature and humanity; and facilitators for common-sense reasoning in decision
making in the absence of complete and precise information. Their role is significant when
applied to complex phenomena not easily described by traditional mathematical methods,

especially when the goal is to find a good approximate solution (Bojadziev & Bojadziev,
1998). Fuzzy set theory is a better means for modeling imprecision arising from mental
phenomena which are neither random nor stochastic. Human beings are heavily involved in
the process of decision analysis. A rational approach toward decision making should take
into account human subjectivity, rather than employing only objective probability measures.
This attitude, towards imprecision of human behavior led to study of a new decision
analysis filed fuzzy decision making (Lai & Hwang, 1996). A tilde ‘~’ will be placed above a
symbol if the symbol represents a fuzzy set. A triangular fuzzy number (TFN), M  is shown
in Fig. 1. A TFN is denoted simply as (l m , m u) or (l , m , u) . The parameters l , m and u ,
respectively, denote the smallest possible value, the most promising value, and the largest
possible value that describe a fuzzy event.

Fig. 1. A triangular fuzzy number, M
Each TFN has linear representations on its left and right side such that its membership
function can be defined as
 0, x  l,
 ( x  l ) (m  l ), l  x  m,
 )  
(x M (1)
(u  x ) (u  m), m  x  u,
 0, x  u.
A fuzzy number can always be given by its corresponding left and right representation of
each degree of membership:
  ( M l( y ) , M r ( y ) )  (l  (m  l )y , u  (m  u)y ),
M y  [0,1], (2)
Where l( y ) and r ( y ) denote the left side representation and the right side representation of
a fuzzy number, respectively. Many ranking methods for fuzzy numbers have been
developed in the literature. These methods may give different ranking results and most
methods are tedious in graphic manipulation requiring complex mathematical calculation.
The algebraic operations with fuzzy numbers have been explained by Kahraman (2001) and
Kahraman et al. (2002).
3. Fuzzy Delphi Analytic Hierarchy Process

The Analytic Hierarchy Process (AHP) is an approach that is suitable for dealing with
complex systems related to making a choice from among several alternatives and which
provides a comparison of the considered options, firstly proposed by Saaty (1980). The AHP
is based on the subdivision of the problem in a hierarchical form. In fact, the AHP helps
organize the rational analysis of the problem by dividing it into its single parts; the analysis
then supplies an aid to the decision makers who, making several pair-wise comparisons, can
appreciate the influence of the considered elements in the hierarchical structure; the AHP
can also give a preference list of the considered alternative solutions (Bentivegna et al., 1994;
Roscelli, 1990; Saaty, 1980; Saaty & Vargas, 1990).
The AHP is a tool that can be used for analyzing different kinds of social, political, economic
and technological problems, and it uses both qualitative and quantitative variables. The
fundamental principle of the analysis is the possibility of connecting information, based on
knowledge, to make decisions or previsions; the knowledge can be taken from experience or
derived from the application of other tools. Among the different contexts in which the AHP
can be applied, mention can be made of the creation of a list of priorities, the choice of the
best policy, the optimal allocation of resources, the prevision of results and temporal
dependencies, the assessment of risks and planning (Saaty & Vargas, 1990). Although the
AHP is to capture the expert’s knowledge, the traditional AHP still cannot really reflect the
human thinking style (Kahraman et al., 2003).
The traditional AHP method is problematic in that it uses an exact value to express the
decision maker’s opinion in a comparison of alternatives (Wang & Chen, 2007). And AHP
method is often criticized due to its use of unbalanced scale of judgments and its inability to
adequately handle the inherent uncertainty and imprecision in the pair-wise comparison
process (Deng, 1999). To overcome all these shortcomings, FDAHP was developed for
solving the hierarchical problems. Decision makers usually find that it is more confident to
give interval judgments than fixed value judgments. This is because usually he/she is
unable to explicit his/her preference to explicit about the fuzzy nature of the comparison
process.
Delphi method is a technique for structuring an effective group communication process by
providing feedback of contributions of information and assessment of group judgments to
enable individuals to re-evaluate their judgments. Since its development in the 1960s at
Rand Corporation, Delphi method has been widely used in various fields (Liu and Chen,
2007a, Liu and Chen, 2007b, Hoseinie et al. 2009, Cheng and Tang, 2009, Cheng et al. 2009).
On the other hand, Delphi Method use crisp number and mean to become the evaluation
criteria, these shortcomings might distort the experts’ opinion. In order to deal with the
fuzziness of human participants’ judgments in traditional Delphi method, Ishikawa et al.
(Ishikawa et al. 1993) posited fuzzy set theory proposed by Zadeh (Zadeh, 1965) into the
Delphi method to improve time-consuming problems such as the convergence of experts’
options presented by Hwang and Lin (Hwang and Lin, 1987). The FDM is a methodology in
which subjective data of experts are transformed into quasi-objective data using the
statistical analysis and fuzzy operations. The main advantages of FDM (Kaufmann and
Gupta, 1988) are that it can reduce the numbers of surveys to save time and cost and it also
includes the individual attributes of all experts. This paper proposes the use of FDAHP for
determining the weights of the main criteria.
3.1 Methodology of FDAHP

Calculate the relative fuzzy weights of the decision elements using the following three steps
based on the FDM and aggregate the relative fuzzy weights to obtain scores for the decision
alternation.
(1) Compute the triangular fuzzy numbers (TFNs) ãij as defined in Eq. (3). In this work, the
TFNs (shown as Fig. 2) that represent the pessimistic, moderate and optimistic estimate are
used to represent the opinions of experts for each activity time.
Fig. 2. The membership function of the Fuzzy Delphi Method.
aij  ( ij ,  ij ,  ij ) (3)
 ij  Min( ijk ), k  1,..., n (4)
n 1
 ij  (  ijk ) n , k  1,..., n (5)
k 1
 ij  Max( ijk ), k  1,..., n (6)
Where, αij≤δij≤γij,  ij ,  ij ,  ij  [1 / 9,1]  [1,9] and αij,δij,γij are obtained from Eq. (4) to Eq. (6).
αij indicates the lower bound and γij indicates the upper bound. βijk indicates the relative
intensity of importance of expert k between activities i and j. n is the number of experts in
consisting of a group.
(2) Following outlined above, we obtained a fuzzy positive reciprocal matrix A 
  [ a ], a  a  1, i , j  1, 2,..., n
A ij ij ji
Or
 
 (1,1,1) ( 12 ,  12 ,  12 ) ( 13 ,  13 ,  13 ) 
 
  ( 1
A 1 1 ( 23 ,  23 ,  23 ) (7)
 12 ,  12 ,  12 ) (1,1,1)
 
( 1 ,1 ,1 ) (1 ,1 ,1 ) (1,1,1) 
  13  13  13  23  23  23 
(3) Calculate the relative fuzzy weights of the evaluation factors.
  Z  (Z  ...  Z )1

Z i  [ a ij  ...  a in ]1/n , W (8)
i i i n
Where a1  a 2  ( 1   2 ,  1   2 ,  1   2 ) ; the symbol  here denotes the multiplication of

fuzzy numbers and the symbol  here denotes the addition of fuzzy numbers. W  is a row
i

vector in consist of a fuzzy weight of the ith factor. Wi  (1 , 2 ,..., n ) i=1, 2,…n, and Wi is a
fuzzy weight of the ith factor.
4. TOPSIS method
TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) is one of the useful
MADM techniques to manage real-world problems (Yoon & Hwang, 1985). TOPSIS method
was firstly proposed by Hwang & Yoon (1981). According to this technique, the best
alternative would be the one that is nearest to the positive ideal solution and farthest from
the negative ideal solution (Benitez, et al., 2007). The positive ideal solution is a solution that
maximizes the benefit criteria and minimizes the cost criteria, whereas the negative ideal
solution maximizes the cost criteria and minimizes the benefit criteria (Wang & Elhag, 2006).
In short, the positive ideal solution is composed of all best values attainable of criteria,
whereas the negative ideal solution consists of all worst values attainable of criteria (Wang,
2008). In this paper TOPSIS method is used for determining the final ranking of the
sawability of rocks. TOPSIS method is performed in the following steps:
Step 1. Decision matrix is normalized via Eq. (9):
wij
rij  j  1, 2, 3,..., J i  1, 2, 3,..., n (9)
J
 wij2
i 1
Step 2. Weighted normalized decision matrix is formed:
vij  wi  rij , j  1, 2, 3,..., J i  1, 2, 3,..., n (10)

Step 3. Positive ideal solution (PIS) and negative ideal solution (NIS) are determined:
 
A  v1 , v2 , ... , vi , ... , vn Maximum Values (11)
 
A  v1 , v2 , ... , vi , ... , vn Minimum Values (12)
Step 4. The distance of each alternative from PIS and NIS are calculated:
n
  vij  v*j 
2
d *j  (13)
i 1
n
  vij  vj 
2
d j  (14)
i 1
Step 5. The closeness coefficient of each alternative is calculated:
dj
CC j  (15)
dj*  dj
Step 6. By comparing CCj values, the ranking of alternatives are determined.
5. Application of FDAHP-TOPSIS method to multi-criteria comparison of

sawability
The purpose of this paper was to ranking the power consumption of rock in sawing process,
with the help of effective factors. Firstly, a comprehensive questionnaire including main
criteria of effective factor is designed to understand and quantify the affecting factors in the
process. Then, five decision makers from different areas evaluated the importance of these
factors with the help of the mentioned questionnaire. FDAHP is utilized for determining the
weights of main criteria and finally, TOPSIS approach is employed for ranking. By this way,
the ranking of carbonate sawability according to their overall efficiency is obtained.
Carbonate rock sawability depends on non-controlled parameters related to rock
characteristics and controlled parameters related to properties of cutting tools and
equipment. In the same working conditions, the sawing process and its results are strongly
affected by mineralogical and mechanical properties of rock. The mineralogical and
mechanical properties of rock which related to rock sawability are shown in Fig. 3.
Texture
Grain Size & Shape

Textural
Matrix Type & Cementation
Characteristics
Density
Weathering
Mohs Hardness
Hardness
Schmidt Hammer
Uniaxial Compressive Strength

Parameters
related to rock Mechanical
Tensile Strength
sawability Characteristics
Abrasiveness Equal Quartz Content
Young’s Modules
Environmental and Structural

Characteristics
Fig. 3. Important characteristics influencing the rock sawability

5.1 Determination of criteria’s weights

Because different groups have varying objectives and expectations, they judge on rock
sawability from different perspectives. So, affecting criteria have different level of
significance for different users. For this reason, five decision makers are selected from
different areas and these decision makers evaluate the criteria. FDAHP is proposed to take
the decision makers subjective judgments into consideration and to reduce the uncertainty
and vagueness in the decision process.
Decision makers from different backgrounds may define different weight vectors. They
usually cause not only the imprecise evaluation but also serious persecution during decision
process. For this reason, we proposed a group decision based on FDAHP to improve pair-
wise comparison. Firstly each decision maker (Di), individually carry out pair-wise
comparison by using Saaty’s (1980) 1–9 scale (Table 2).
Comparison index score

Extremely Preferred 9
Very strongly Preferred 7
Strongly Preferred 5
Moderately Preferred 3
Equal 1
Intermediate values between the two adjacent judgments 2,4,6,8
Table 2. Pair-wise comparison scale (Saaty, 1980)
One of these pair-wise comparisons is shown here as example:
c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 c11 c12

c1
 1 3 5 1 1 1 5 5 3 1/3 1 3 
c2 
1/3 1 3 1/3 1/3 1/3 3 3 1 1/5 1/3 1 
c3 
1 / 5 1/3 1 1/5 1/5 1/5 1 1 1 / 3 1 / 9 1 / 5 1 / 3
c4  
1 3 5 1 1 1 5 5 3 1/3 1 3 
c 5 
1 3 5 1 1 1 5 5 3 1/3 1 3 
c6  
 1 3 5 1 1 1 5 5 3 1/3 1 3 
D1  c7 
1/5 1/3 1 1/5 1/5 1/5 1 1 1 / 3 1 / 9 1 / 5 1 / 3
c8  
1 / 5 1/3 1 1/5 1/5 1/5 1 1 1 / 3 1 / 9 1 / 5 1 / 3
c9  
1/3 1 3 1/3 1/3 1/3 3 3 1 1/5 1/3 1 
c10 
 3 7 9 3 3 3 9 9 5 1 3 5 
c11  
 1 3 5 1 1 1 5 5 3 1/3 1 3 
c12 
1 / 3 1 3 1/3 1/3 1/3 3 3 1 1/5 1/3 1 
The weighting factors for each criterion were presented in the following steps:
1. Compute the triangular fuzzy numbers (TFNs)
aij  ( ij ,  ij ,  ij )
According Eq. (4)- Eq. (6)
 ij  Min( ijk ), k  1,..., n

2. Fuzzy pair-wise comparison matrix A 
By this way, decision makers’ pair-wise comparison values are transformed into triangular
fuzzy numbers as in Table 3.
C1 C2 C3 C4 C5 C6
C1 (1, 1, 1) (0.7, 1, 1.4) (0.7, 1.2, 2.3) (0.6, 0.7, 1) (0.6, 0.7, 1) (0.6, 0.8, 1)
C2 (0.7, 1, 1.4) (1, 1, 1) (0.7, 1.2, 1.7) (0.7, 0.7, 0.8) (0.6, 0.7, 1) (0.7, 0.8, 1)
C3 (0.4, 08, 1.4) (0.6 0.8, 1.4) (1, 1, 1) (0.4, 0.6, 1) (0.3, 0.6, 0.8) (0.4, 0.6, 1)
C4 (1, 1.4, 1.8) (1.3, 1.4, 1.4) (1, 1.7, 2.3) (1, 1, 1) (0.8, 1, 1.3) (1, 1.1, 1.3)
C5 (1, 1.4, 1.8) (1, 1.4, 1.8) (1.3, 1.8, 3) (0.8, 1.1, 1.3) (1, 1, 1) (0.8, 1.1, 1.3)
C6 (1, 1.3, 1.8) (1, 1.3, 1.4) (1, 1.6, 2.3) (0.8, 1, 1) (0.8, 0.9, 1.3) (1, 1, 1)
C7 (0.4, 0.7, 1.4) (0.6, 0.7, 1) (0.4, 0.8, 1) (0.4, 0.5, 0.8) (0.3, 0.5, 0.8) (0.4, 0.5, 1)
C8 (0.4, 0.7, 1) (0.6, 0.7, 1) (0.7, 0.9, 1) (0.4, 0.5, 0.7) (0.3, 0.5, 0.7) (0.4, 0.6, 0.7)
C9 (0.4, 0.8, 1.4) (0.4, 0.8, 1.4) (0.6, 1, 1.7) (0.3, 0.6, 0.8) (0.3, 0.6, 0.8) (0.3, 0.6, 1)
C10 (1.3, 1.5, 1.8) (1, 1.5, 1.8) (1, 1.8, 3) (0.8, 1.1, 1. 3) (0.8, 1.1, 1.3) (1, 1.2, 1.3)
C11 (1, 1.1, 1.4) (0.7, 1.1, 1.4) (0.7, 1.4, 2.3) (0.6, 0.9 1) (0.6, 0.8, 1) (0.7, 0.9, 1)
C12 (0.7, 1.1, 1.4) (0.7, 1.1, 1.4) (1, 1.3, 2.3) (0.6, 0.8, 1) (0.7, 0.8, 0.8) (0.6, 0.8, 1)
C7 C8 C9 C10 C11 C12

(0.7, 1.5, 2.3) (1, 1.4, 2.3) (0.7, 1.2, 2.3) (0.6, 0.7, 0.8) (0.7, 0.9, 1) (0.7, 0.9, 1.4)
(1, 1.5, 1.7) (1, 1.4, 1.7) (0.7, 1.2, 2.3) (0.6, 0.7, 1) (0.7, 0.9, 1.4) (0.7, 0.9, 1.4)
(1, 1.2, 2.3) (1, 1.1, 1.4) (0.6, 1, 1.7) (0.3, 0.5, 1) (0.4, 0.7, 1.4) (0.4, 0.8, 1)
(1.3, 2, 2.3) (1.4, 1.9, 2.3) (1, 1.7, 3) (0.8, 0.9, 1.3) (1, 1.2, 1.8) (1, 1.3, 1.8)
(1.3, 2.1, 3) (1.4, 2.1, 3) (1.3, 1.7, 3) (0.8, 1, 1.3) (1, 1.2, 1.8) (1.3, 1.3, 1.4)
(1, 1.9, 2.3) (1.4, 1.8, 2.3) (1, 1.6, 3) (0.8, 0.9, 1) (1, 1.1, 1.4) (0.4, 0.6, 1)
(1, 1, 1) (0.6, 1, 1.4) (0.4, 0.8, 1.7) (0.3, 0.5, 1) (0.4, 0.6, 1.4) (0.4, 0.6, 1)
(0.7, 1, 1.7) (1, 1, 1) (0.6, 0.9, 1.7) (0.3, 0.5, 0.7) (0.4, 0.6, 1) (0.4, 0.7, 1)
(0.6, 1.2, 2.3) (0.6, 1.1, 1.7) (1, 1, 1) (0.3, 0.5, 1) (0.4, 0.7, 1.4) (0.4, 0.8, 1)
(1, 2.2, 3) (1.4, 2.1, 3) (1, 1.8, 3) (1, 1, 1) (1.3, 1.3, 1.4) (1, 1.4, 1.8)
(0.7, 1.7, 2.3) (1, 1.6, 2.3) (0.7, 1.4, 2.3) (0.7, 0.8, 0.8) (1, 1, 1) (0.7, 1.1, 1.4)
(1, 1.6, 2.3) (1, 1.5, 2.3) (1, 1.3, 2.3) (0.6, 0.7, 1) (0.7, 0.9, 1.4) (1, 1, 1)
Table 3. Fuzzy pair-wise comparison matrix
3. Calculate the relative fuzzy weights of the evaluation factors.
Z 1   a11  a12  ...a 112 

1/12
 [0.6948,0.968,1.3738]
Z 2   a 21  a 22  ...a 212 
1/12
 [0.7451,0.968,1.2989]
Z 3   a 31  a 32  ...a 312 
1/12
 [0.5373,0.7861,1.2268]
Z 4   a 41  a 42  ...a 412 
1/12
 [1.0284,1.3098,1.7181]
Z 5   a 51  a 52  ...a 512 
1/12
 [1.0502,1.3773,1.8295]
Z 6   a 61  a 62  ...a 612 
1/12
 [0.9,1.1823,1.5363]
Z 7   a71  a72  ...a712 

1/12
 [0.4665,0.6661,1.0885]
Z 8   a81  a 82  ...a 812 

1/12
 [0.508,0.6897,0.9733]
Z 9   a 91  a92  ...a 912 

1/12
 [0.4601,0.7858,1.2189]
Z 10   a 101  a102  ...a 1012 

1/12
 [1.0284,1.4483,1.8295]
Z 11   a 111  a 112  ...a1112 

1/12
 [0.7451,1.1074,1.4128]
Z 12   a 121  a 122  ...a1212 

1/12
 [0.7717,1.0353,1.4128]
 Z i  [8.9356,12.327,16.919]
  Z  (Z  Z  Z )1  [0.0411,0.0785,0.1537]
W1 1 1 2 3
  Z  (Z  Z  Z )1  [0.044,0.0785,0.1454]

W2 2 1 2 3
  Z  (Z  Z  Z )1  [0.0318,0.064,0.1373]

W3 3 1 2 3
  Z  (Z  Z  Z )1  [0.0608,0.1063,0.1923]

W4 4 1 2 3
  Z  (Z  Z  Z )1  [0.0621,0.1117,0.2047]

W5 5 1 2 3
  Z  (Z  Z  Z )1  [0.0532,0.0959,0.1719]

W6 6 1 2 3
  Z  (Z  Z  Z )1  [0.0276,0.054,0.1218]

W7 7 1 2 3
  Z  (Z  Z  Z )1  [0.03,0.056,0.1089]

W8 8 1 2 3
  Z  (Z  Z  Z )1  [0.0272,0.0637,0.1364]

W9 9 1 2 3
  Z  (Z  Z  Z )1  [0.0608,0.1175,0.2047]

W10 10 1 2 3
  Z  (Z  Z  Z )1  [0.044,0.0898,0.1581]

W11 11 1 2 3
  Z  (Z  Z  Z )1  [0.0456,0.084,0.1581]

W12 12 1 2 3
The final weights of each parameter are calculated as follow:

W1  ( i  1  j )1/3  0.07928 , W2=0.0796, W3=0.0655, W4=0.1076, W5=0.1125, W6=0.0958,
3
W7=0.05675, W8=0.0569, W9=0.0619, W10=0.1136, W11=0.0857, W12=0.0847. Mentioned

priority weights have indicated for each criterion in table 4.
Criteria Global weights

Uniaxial Compressive Strength 0.1136
Hardness 0.1125
Equal Quartz Content 0.1076
Abrasiveness 0.0958
Tensile Strength 0.0857
Young’s Modules 0.0847
Grain Size & Shape 0.0796
Texture 0.0793
Matrix Type & Cementation 0.0655
Schmidt Hammer Rebound 0.0619
Density 0.0569
Weathering 0.0568
Table 4. Priority weights for criteria
5.2 Ranking the sawability of carbonate rock

In attempting to present a ranking system for assessing rock sawability, using all mentioned
parameters is difficult from a practical point of view. In this ranking system three following
rules have been considered: (a) the number of parameters used should be small, (b)
equivalent parameters should be avoided, and (c) parameters should be considered within
certain groups. Considering these rules, the parameters which have been chosen for
assessing the rock sawability are listed as follows:
a. Uniaxial Compressive Strength (UCS)
b. Schmiazek F-abrasivity factor (SF-a)
c. Mohs Hardness (MH)
d. Young’s Modulus (YM)
5.3 Uniaxial Compressive Strength (UCS)

Uniaxial compressive strength is one of the most important engineering properties of rocks.
Rock material strength is used as an important parameter in many rock mass classification
systems. Using this parameter in classification is necessary because strength of rock material
constitutes the strength limit of rock mass (Bieniawski, 1989). Factors that influence the UCS
of rocks are the constitutive minerals and their spatial positions, weathering or alteration
rate, micro-cracks and internal fractures, density and porosity (Hoseinie et al. 2009).
Therefore, uniaxial compressive strength test can be considered as representative of rock
strength, density, weathering, texture and matrix type. Thus, the summation of the weights
of five parameters (texture, weathering, density, matrix type and UCS) is considered as
weight of UCS. In total, the weight of UCS is about 0.372.
5.4 Schmiazek F-abrasivity factor (SF-a)

Abrasiveness influences the tool wear and sawing rate seriously. Abrasiveness is mainly
affected by various factors such as mineral composition, the hardness of mineral
constituents and grain characteristics such as size, shape and angularity (Ersoy and Waller,
1995). Schimazek’s F-abrasiveness factor is depend on mineralogical and mechanical
properties and has good ability for evaluation of rock abrasivity. Therefore, this index has
been selected for using in ranking system. F-abrasivity factor is defined as
EQC  Gs  BTS
F (16)
100
Where F is the Schimazek’s wear factor (N/mm), EQC is the equivalent quartz content
percentage, Gs is the median grain size (mm), and BTS is the in direct Brazilian tensile
strength. Regarding the rock parameters which are used in questionnaires, summation of
the weights of abrasiveness, grain size, tensile strength and equivalent quartz content is
considered as weight of Schimazek’s F-abrasiveness factor. In total the weight of this factor
is 0.3687.
5.5 Mohs Hardness (MH)

Hardness can be interpreted as the rock’s resistance to penetration. The factors that affect
rock hardness are the hardness of the constitutive minerals, cohesion forces, homogeneity,
and the water content of rock (Hoseinie et al. 2009). Thus, hardness is a good index of all
above given parameters of rock material. Considering the importance of hardness in rock
sawing, hardness, after Schmiazek F-abrasivity factor, is considered the most relevant
property of rock material. Regarding the questionnaires, summation of the weights of Mohs
hardness and Schmidt hammer rebound value was considered as total weight of mean
Mohs hardness. In total the weight of this factor is 0.1745.
5.6 Young’s Modulus (YM)

According to rock behaviour during the fracture process, especially in sawing, the way that
rocks reach the failure point has a great influence on sawability. The best scale for rock
elasticity is Young’s modulus. Based on ISRM suggested methods (ISRM, 1981), the tangent
Young’s modulus at a stress level equal to 50% of the ultimate uniaxial compressive strength
is used in this ranking system. Regarding the questionnaires, the weight of this factor is
about 0.0847 in total. According to FDAHP results the final weights of major parameters are
shown in fig. 4.
5.7 Laboratory tests

For laboratory tests, some rock blocks were collected from the studied factories. An attempt
was made to collect rock samples that were big enough to obtain all of the test specimens of
each rock type from the same piece. Each block sample was inspected for macroscopic
defects so that it would provide test specimens free from fractures, partings or alteration
zones. Then, test samples were prepared from these block samples and standard tests have
been completed to measure the above-mentioned parameters following the suggested
procedures by the ISRM standards (ISRM, 1981). The results of laboratory studies are listed
in table 5 and used in next stage.
Power consumption ranking
UCS w1 SF-a w2 MH w3 YM w4
0.372 0.3687 0.1745 0.0847
Mt&C 0.0655 Gs & S 0.0796 H 0.1125 E&P 0.0847
T 0.0793 EQC 0.1076 SHR 0.0620
W 0.0568 A 0.0958
D 0.0569 BTS 0.0857
UCS 0.1136
Fig. 4. The final weights of major parameters in power consumption ranking system
UCS BTS EQC Gs SF-a YM MH

Rock sample
MPa MPa % mm N/mm GPa n
1 MHAR(Marble) 71.5 6.8 3.6 0.55 0.135 32.5 3.5
2 MANA(Marble) 74.5 7.1 3.4 0.45 0.109 33.6 3.2
3 TGH(Travertine) 53 4.3 2.8 1.01 0.122 20.7 2.9
4 THAJ(Travertine) 61.5 5.6 2.6 0.85 0.124 21 2.9
5 TDAR(Travertine) 63 5.4 2.7 0.87 0.127 23.5 2.95
6 MSAL(Marble) 68 6.3 3.2 0.52 0.105 31.6 3.1
7 MHAF(Marble) 74.5 7.2 4 0.6 0.173 35.5 3.6
Table 5. The result of laboratory studies
After determining the weights of the criteria with FDAHP method and laboratory studies,
ranking the sawability of carbonate rocks is performed by TOPSIS method. Firstly, the
amount of each criterion is filled in decision matrix for each criterion.
Decision matrix is obtained with respect to important rock properties (Table. 6). Decision
matrix is normalized via Eq. (9) (Table. 7). Then, weighted normalized matrix is formed by
multiplying each value with their weights (Table. 8). Positive and negative ideal solutions
are determined by taking the maximum and minimum values for each criterion:
A   0.1547,0.1863,0.0393,0.0748
A   0.1101,0.1130,0.0229,0.0602
Then the distance of each method from PIS (positive ideal solution) and NIS (negative ideal
solution) with respect to each criterion are calculated with the help of Eqs. (13) and (14).
Then closeness coefficient of each rock is calculated by using Eq. (15) and the ranking of the
rocks are determined according to these values.
UCS SF-a YM MH
C1: C2: C3: C4:
1 71.5 0.135 32.5 3.5
2 74.5 0.109 33.6 3.2
3 53 0.122 20.7 2.9
4 61.5 0.124 21 2.9
5 63 0.127 23.5 2.95
6 73 0.105 31.6 3.1
7 74.5 0.173 35.5 3.6
Table 6. Decision matrix
UCS SF-a YM MH
C1: C2: C3: C4:
1 0.3991 0.3937 0.4243 0.4166
2 0.4158 0.3176 0.4387 0.3809
3 0.2958 0.3556 0.2703 0.3452
4 0.3432 0.3619 0.2742 0.3452
5 0.3516 0.3709 0.3068 0.3511
6 0.4074 0.3065 0.4126 0.3690
7 0.4158 0.5052 0.4635 0.4285
Table 7. Normalized decision matrix
UCS SF-a YM MH
C1: C2: C3: C4:
1 0.1485 0.1452 0.0360 0.0727
2 0.1547 0.1171 0.0372 0.0665
3 0.1101 0.1311 0.0229 0.0602
4 0.1277 0.1334 0.0232 0.0602
5 0.1308 0.1368 0.0260 0.0613
6 0.1516 0.1130 0.0350 0.0644
7 0.1547 0.1863 0.0393 0.0748
Table 8. Weighted normalized matrix
The power consumption ranking of carbonate rocks are also shown in Table 9 in the
descending order of priority.
Rank Carbonate rock dj* dj- CCj

1 MHAF(Marble) 0 0.0886 1
2 MHAR(Marble) 0.0418 0.0532 0.5602
3 MANA(Marble) 0.0697 0.0475 0.4050
4 MSAL(Marble) 0.0742 0.0434 0.3693
5 TDAR(Travertine) 0.0582 0.0317 0.3528
6 THAJ(Travertine) 0.0632 0.0270 0.2992
7 TGH(Travertine) 0.0743 0.0181 0.1958
Table 9. Rankings of the sawability of carbonate rocks according to CCj values
6. Discussion and validation of the new ranking

A new hierarchical model is developed here to evaluate and ranking the sawability (power
consumption) of carbonate rock with using of effective criteria and considering of decision
makers’ judgments. The proposed approach is based on the combination of Fuzzy Delphi
and Analytic Hierarchy Process (FDAHP) methods. Technique for Order Preference by
Similarity to Ideal Solution (TOPSIS) is also used in this study. FDAHP was used for
determining the weights of the criteria according to decision makers then rankings of
carbonate rocks were determined by TOPSIS. The proposed method was applied for Iranian
ornamental stone to evaluation the power consumption in rock sawing process. The
sawability ranking results of tested carbonate rocks are shown in section 5.
For validation of applied ranking system, experimental procedure was carried out. For this
purpose, a fully-instrumented laboratory cutting rig was used. The rig was based on a
commercially available machine and was capable of simulating realistic cutting conditions.
It consists of three major sub-systems, a cutting unit, instrumentation and a personal
computer. Sawing tests were performed on a small side-cutting machine, with a maximum
spindle motor power of 7.5 kW. Cutting parameters such as feed rate, depth of cut, and
peripheral speed control in the monitoring system. The variation of ampere was measured
with a digital ampere-meter. The circular diamond saw blade used in the present tests had a
diameter of 410 mm and a steel core of thickness 2.7 mm, 28 pieces of diamond impregnated
segments (size 40×10×3 mm) were brazed to the periphery of circular steel core with a
standard narrow radial slot. The grit sizes of the diamond were approximately 30/40 US
mesh at 25 and 30 concentrations. This blade is applied for travertine, limestone and marble
types of the stones, which are non-abrasive and medium hard. During the sawing trials,
water was used as the flushing and cooling medium and the peripheral speed and depth of
cut were maintained at constant 1770 rpm and 35 mm. Each rock was sawn at particular
feed rate (200, 300 and 400 cm/min). During the sawing trials, the ampere and power
consumption were monitored and calculated. The monitored ampere and calculated power
consumption are listed in Table 10. According to Tables 9 and 10, the first rock in ranking
(MHAF) has a maximum value of power consumption among other rock samples. It means
that the new developed ranking is correct.
Test 1 Test 2 Test 3

Fr=200 Fr=300 Fr=400
Rock sample
(cm/min) (cm/min) (cm/min)
I(A) P(W) I(A) P(W) I(A) P(W)
1 MHAF(Marble) 12 2280 17.5 4370 28.6 8588
2 MHAR(Marble) 11.5 2090 16.4 3952 22 6080
3 MSAL(Marble) 11.5 2090 15.7 3686 20.2 5396
4 MANA(Marble) 11.2 1976 16.5 3990 22.2 6156
5 THAJ(Travertine) 11.2 1976 15.4 3572 19 4940
6 TDAR(Travertine) 10.6 1748 15.6 3648 17.2 4256
7 TGH(Travertine) 9.5 1330 12 2280 18 4560
Table 10. Ampere and calculated power consumption in sawing trials
The calculated power consumption of each carbonate rock in Figure 5 shows that the new
ranking method for carbonate rock is reasonable and acceptable for evaluating them.
Test 1 Test 2 Test 3
10000
8000
Power Consumption (W)
6000
4000
2000
0
MHAF MHAR MANA MSAL TDAR THAJ TGH
Fig. 5. The power consumption of carbonate rocks in sawing trials
7. Conclusion
In this chapter, a decision support system was developed for ranking the power
consumption of carbonate rocks. This system designed to eliminate the difficulties in taking
into consideration many decision criteria simultaneously in the rock sawing process and to
guide the decision makers for ranking the power consumption of carbonate rocks. In this
study, FDAHP and TOPSIS methods was used to determine the power consumption degree
of the carbonate rocks. FDAHP is utilized for determining the weights of the criteria and
TOPSIS method is used for determining the ranking of the power consumption of carbonate
rocks. During this research a fully-instrumented laboratory sawing rig at different feed rate
for two groups of carbonate rocks were carried out. The power consumptions were used to
verify the result of applied approach for ranking them by sawability criteria. The
experimental results confirm the new ranking results precisely.
This new ranking method may be used for evaluating the power consumption of carbonate
rocks at any stone factory with different carbonate rock. Some factors such as uniaxial
compressive strength, Schmiazek F-abrasivity, mohs hardness and young's modulus must
be obtained for the best power consumption ranking.
8. Acknowledgment
The authors would like to say thank to educational workshop centre of Sharif University of
Technology and central laboratory of Shahrood University of Technology for providing the
facilities in order to perform an original research.
9. References
Ataei, M., (2005). Multi-criteria selection for alumina-cement plant location in East-
Azerbaijan province of Iran. The Journal of the South African Institute of Mining and
Metallurgy, 105(7), pp. 507–514.
Atici U, Ersoy A., (2009). Correlation of specific energy of cutting saws and drilling bits with
rock brittleness and destruction energy. journal of materials processing technology, 209,
pp. 2602–2612.
Ayag, Z., Ozdemir, R. G., (2006). A fuzzy AHP approach to evaluating machine tool
alternatives. Journal of Intelligent Manufacturing, 17, pp. 179–190.
Basligil, H., (2005). The fuzzy analytic hierarchy process for software selection problems.
Journal of Engineering and Natural Sciences, 2, pp. 24–33.
Bieniawski ZT (1989). Engineering rock mass classifications. New York: Wiley.
Benitez, J. M., Martin, J. C., Roman, C., (2007). Using fuzzy number for measuring quality of
service in the hotel industry. Tourism Management, 28(2), pp. 544–555.
Bentivegna, V., Mondini, G., Nati Poltri, F., Pii, R., (1994). Complex evaluation methods. An
operative synthesis on Multicriteria techniques. In: Proceedings of the IV
International Conference on Engineering Management, Melbourne.
Birle J. D., Ratterman E., (1986). An approximate ranking of the sawability of hard building
stones based on laboratory tests, Dimensional Stone Magazine, July-August, pp. 3-29.
Bojadziev, G., Bojadziev, M., (1998). Fuzzy sets fuzzy logic applications. Singapore: World
Scientific Publishing.
Bozdag, C. E., Kahraman, C., Ruan, D., (2003). Fuzzy group decision making for selection
among computer integrated manufacturing systems. Computer in Industry, 51, pp. 13–
29.
Brook, B. (2002). Principles of diamond tool technology for sawing rock. Int. J. Rock Mech.
Min. Sci. 39, pp 41-58.
Buckley, J. J., (1985). Fuzzy hierarchical analysis. Fuzzy Sets and Systems, 17, pp. 233–247.
Burgess R. B., (1978). Circular sawing granite with diamond saw blades. In: Proceedings of
the Fifth Industrial Diamond Seminar, pp. 3-10.
Buyuksagis I.S, (2007). Effect of cutting mode on the sawability of granites using segmented
circular diamond sawblade. Journal of Materials Processing Technology, 183, pp. 399–
406.
Ceylanoglu, A. Gorgulu, K. (1997). The performance measurement results of stone cutting
machines and their relations with some material properties. In: V. Strakos, V. Kebo,
L. Farana, L. Smutny, eds. Proceedings of the Sixth International Symposium on

Mine Planning and Equipment Selection. Rotterdam: Balkema, pp. 393-8.
Chang, D. Y., (1996). Applications of the extent analysis method on fuzzy AHP. European
Journal of Operational Research, 95, pp. 649–655.
Chang, Y. H., Cheng, C. H., Wang, T. C., (2003). Performance evaluation of international
airports in the region of east Asia. In Proceedings of Eastern Asia Society for
transportation studies, Vol. 4, pp. 213–230.
Chan, F. T. S., Kumar, N., (2007). Global supplier development considering risk factors using
fuzzy extended AHP-based approach. Omega International Journal of Management
Science, 35, pp. 417–431.
Chen, C. T., Lin, C. T., Huang, S. F., (2006). A fuzzy approach for supplier evaluation and
selection in supply chain management. International Journal of Production Economics,
102, pp. 289–301.
Chen Chao-Shi, Liu Ya-Ching, (2007). A methodology for evaluation and classification of
rock mass quality on tunnel engineering. Tunnelling and Underground Space
Technology. 22, pp. 377–387.
Cheng Jao-Hong, Lee Chih-Ming, Tang Chih-Huei. (2009). An Application of Fuzzy Delphi
and Fuzzy AHP for Multi-criteria Evaluation on Bicycle Industry Supply Chains.
WSEAS TRANSACTIONS on SYSTEMS and CONTROL. Issue 1, Volume 4.
Cheng Jao-Hong, Lee Chih-Ming, Tang Chih-Huei. (2009). An Application of Fuzzy Delphi
and Fuzzy AHP on Evaluating Wafer Supplier in Semiconductor Industry. WSEAS
TRANSACTIONS on INFORMATION SCIENCE and APPLICATIONS. Issue 5,
Volume 6.
Chou, T. Y., Liang, G. S., (2001). Application of a fuzzy multi-criteria decision making model for
shipping company performance evaluation. Maritime Policy & Management, 28(4), pp. 375–
392.
Clausen R, Wang CY, Meding M. (1996). Characteristics of acoustic emission during single
diamond scratching of granite. Industrial Diamond Review, 3, pp. 96–9.
Delgado N.S, Rodriguez R, Rio A, Sarria I.D, Calleja L, Argandona V.G.R. (2005). The
influence of microhardness on the sawability of Pink Porrino granite (Spain).
International Journal of Rock Mechanics & Mining Sciences,42, pp. 161–6.
Deng, H., (1999). Multicriteria analysis with fuzzy pair-wise comparison. International
Journal of Approximate Reasoning, 21, pp. 215–231.
Enea, M., Piazza, T., (2004). Project selection by constrained fuzzy AHP.Fuzzy Optimization
and Decision Making, (Vol. 3). New York: Kluwer Academic Publishers, pp. 39–62.
Ersoy A, Waller MD (1995). Textural characterization of rocks. J Eng Geol; 39 (3– 4), pp. 123–
36.
Ersoy A, Atici U, (2004). Performance characteristics of circular diamond saws in cutting
different types of rocks. Diamond and Related Materials, 13, pp. 22–37.
Ersoy A., Buyuksagis S., Atici U., (2005). Wear characteristics of circular diamond saws in
the cutting of different hard and abrasive rocks, Wear 258, pp. 1422–1436.
Ertingshausen, W. (1985). Wear processes in sawing hard stone, Industrial Diamond
Review. 5, pp.254–258.
Ertugrul, I., Karakasoglu, N., (2006). The fuzzy analytic hierarchy process for supplier
selection and an application in a textile company. In Proceedings of 5th international
symposium on intelligent manufacturing systems, pp. 195–207.
Ertugrul, I., Tus, A., (2007). Interactive fuzzy linear programming and an application sample
at a textile firm. Fuzzy Optimization and Decision Making, 6, pp. 29-49.
Ertugrul, I., Karakasoglu, N., (2009). Performance evaluation of Turkish cement firms with fuzzy
analytic hierarchy process and TOPSIS methods, Expert Systems with Applications, 36 (1),
pp. 702–715.
Eyuboglu A.S, Ozcelik Y, Kulaksiz S, Engin I.C, (2003). Statistical and microscopic
investigation of disc segment wear related to sawing Ankara andesites, International
Journal of Rock Mechanics & Mining Sciences, 40, pp. 405–414.
Fener M, Kahraman S, & Ozder M.O; (2007). Performance Prediction of Circular Diamond
Saws from Mechanical Rock Properties in Cutting Carbonate Rocks, International
Journal of Rock Mechanics & Rock Engineering, 40 (5), pp. 505–517.
Gu, X., Zhu, Q., (2006). Fuzzy multi-attribute decision making method based on eigenvector
of fuzzy attribute evaluation space. Decision Support Systems, 41, pp. 400–410.
Gunaydin O., Kahraman S., Fener M., (2004). Sawability prediction of carbonate rocks from
brittleness indexes. The Journal of the South African Institute of Mining and Metallurgy,
104, pp. 239-244.
Haq, A. N., Kannan, G., (2006). Fuzzy analytical hierarchy process for evaluating and
selecting a vendor in a supply chain model. International Journal of Advanced
Manufacturing, 29, pp. 826–835.
Hartman H. L., (1992). SME Mining Engineering Handbook. Society for Mining, Metallurgy,
and Exploration, Inc. Littleton, Colorado, 2nd Edition, Vol. 1.
Hartman, H. L., Mutmansky, J. M., (2002). Introductory Mining Engineering. John Wiley &
Sons, New Jersey.
Hoseinie S.H., Aghababaei H., Pourrahimian Y. (2008). Development of a new classification
system for assessing of rock mass drillability index (RDi). International Journal of
Rock Mechanics & Mining Sciences, 45, pp. 1–10.
Hoseinie S.H., Ataei M., Osanloo M. (2009).A new classification system for evaluating rock
penetrability. International Journal of Rock Mechanics & Mining Sciences. 46, pp. 1329–
1340.
Hsieh, T. Y., Lu, S. T., Tzeng, G. H., (2004). Fuzzy MCDM approach for planning and design
tenders selection in public office buildings. International Journal of Project
Management, 22, pp. 573–584.
Hausberger P. (1989). Causes of the different behaviour of rocks when machined with
diamond tools. Industrial Diamond Review, 3, pp. 1-25.
Hwang, C. L., Yoon, K., (1981). Multiple attributes decision making methods and
applications. Berlin: Springer.
Hwang, C.L., Lin, M.L., (1987). Group Decision Making under Multiple Criteria, Springer-
Verlag, Berlin Heidelberg, New York.
Ilio, A.D. Togna, A. (2003). A theoretical wear model for diamond tools in stone cutting, Int.
J. Mach. Tools Manuf. 43, pp. 1171–1177.
International Society for Rock Mechanics, (1981). Rock characterisation, testing and
monitoring: ISRM suggested methods. Oxford: Pergamon.
Ishikawa, A., Amagasa, T., Tamizawa, G., Totsuta, R., Mieno, H., 1993. The Max-Min Delphi
Method and Fuzzy Delphi Method via Fuzzy Integration, Fuzzy Sets and Systems, Vol.
55.
Jennings M., Wright D., (1989). Guidelines for sawing stone, Industrial Diamond Review, 49, pp.
70–75.
Kahraman, C., (2001). Capital budgeting techniques using discounted fuzzy cash flows. In:
Ruan, D., Kacprzyk, J., Fedrizzi, M. (Eds.), Soft Computing for Risk Evaluation and
Management: Applications in Technology, Environment and Finance, Physica-Verlag,
Heidelberg, pp. 375–396.
Kahraman, C., Ruan, D., Tolga, E., (2002). Capital budgeting techniques using discounted
fuzzy versus probabilistic cash flows. Information Sciences, 42 (1–4), pp. 57–76.
Kahraman, C., Ruan, D., & Dogan, I., (2003). Fuzzy group decision making for facility
location selection. Information Sciences, 157, pp. 135–153.
Kahraman, C., Cebeci, U., Ulukan, Z., (2003). Multi-criteria supplier selection using fuzzy
AHP. Logistics Information Management, 16 (6), pp. 382–394.
Kahraman, C., Cebeci, U., Ruan, D., (2004). Multi-attribute comparison of catering service
companies using fuzzy AHP: The case of Turkey. International Journal of Production
Economics, 87, pp. 171–184.
Kahraman S, Fener M, Gunaydin O. (2004). Predicting the sawability of carbonate rocks
using multiple curvilinear regression analysis. International Journal of Rock Mechanics
& Mining Sciences, 41; pp. 1123–1131.
Kahraman S, Altun H, Tezekici B.S., Fener M, (2005). Sawability prediction of carbonate
rocks from shear strength parameters using artificial neural networks. International
Journal of Rock Mechanics & Mining Sciences, 43(1), pp 157–164.
Kahraman S., Ulker U., Delibalta S., (2007).A quality classification of building stones from P-
wave velocity and its application to stone cutting with gang saws. The Journal of the
South African Institute of Mining and Metallurgy, 107, pp 427-430.
Kaufmann, A., Gupta, M.M., (1988). Fuzzy Mathematical Models in Engineering and
Management Science. North-Holland, Amsterdam.
Konstanty, J. (2002). Theoretical analysis of stone sawing with diamonds. J. Mater. Process
Technol. 123, pp. 146-54.
Lai, Y. J., Hwang, C. L., (1996). Fuzzy multiple objective decision making. Berlin: Springer.
Leung, L., Cao, D., (2000). Theory and methodology in consistency and ranking of
alternatives in fuzzy AHP. European Journal of Operational Research, 124, pp. 102–113.
Li, Y. Huang, H. Shen, J.Y. Xu, X.P. Gao, Y.S. (2002). Cost-effective machining of granite by
reducing tribological interactions, J. Mater. Process. Technol. 129, pp. 389–394.
Liu YC, Chen CS (2007). A new approach for application of rock mass classification on rock
slope stability assessment. Engineering Geology.;89, pp.129–43.
Luo, S.Y. (1997). Investigation of the worn surfaces of diamond sawblades in sawing granite,
J. Mater. Process. Technol. 70, pp. 1–8.
Mikhailov, L., Tsvetinov, P., (2004). Evaluation of services using a fuzzy analytic hierarchy
process. Applied Soft Computing, 5, pp. 23–33.
Mikaeil R, Ataei M, Hoseinie S.H. (2008). Predicting the production rate of diamond wire
saws in carbonate rocks cutting, Industrial Diamond Review, (IDR). 3, pp. 28-34.
Mikaeil, .R, Zare Naghadehi, .M, Sereshki, .F, (2009). Multifactorial Fuzzy Approach to the
Penetrability Classification of TBM in Hard Rock Conditions. Tunnelling and
Underground Space Technology, Vol. 24, pp. 500–505.
Mikaeil, .R, Zare Naghadehi, .M, Ataei, .M, KhaloKakaie, R, (2009). A Decision Support
System Using Fuzzy Analytical Hierarchy Process (FAHP) and TOPSIS Approaches
for Selection of the Optimum Underground Mining Method. Archives of Mining

Sciences, Vol. 54, No 2, pp. 341–368.
Norling, R. G. (1971). Mechanical properties and the composition of some Swedish natural
stone types and their effect on cutting results. In: Presentation at the Conference on
Diamond in the Construction and Stone Industry, in German.
Ozcelik Y, Polat E, Bayram F, Ay A.M. (2004). Investigation of The Effects of Textural
Properties on Marble Cutting with Diamond Wire. International Journal of Rock
Mechanics & Mining Sciences, Sci. Vol. 41, No. 3.
Özçelik Y. (2007). The effect of marble textural characteristics on the sawing efficiency of
diamond segmented frame saws. Industrial Diamond Review, 2, pp. 65-70.
Roscelli, R. (1990). Misurare nell_incertezza. Celid, Torino.
Saaty, T. L. (1980). The analytic hierarchy process. New York: McGraw- Hill.
Saaty, T.L., Vargas, L.G. (1990). The Analytic Hierarchy Process series. University of Pittsburg.
Tang, Y., & Beynon, M. J. (2005). Application and development of a fuzzy analytic hierarchy process
within a capital investment study. Journal of Economics and Management, 1(2), pp. 207–230.
Tang, L. L., Kuo, Y. C., Lee E. S. (2005). A multi-objective model for Taiwan notebook
computer distribution problem. In Y.-C. Lan (Ed.), Global integrated supply chain
systems, (pp. 171). Hershey, PA, USA: Idea Group Publishing.
Tolga, E., Demircan, M., Kahraman, C. (2005). Operating system selection using fuzzy
replacement analysis and analytic hierarchy process. International Journal of
Production Economics, 97, pp. 89–117.
Tonshoff, H.K., Asche, J. (1997). Wear of metal-bond diamond tools in the machining of
stone, Industrial Diamond Review, 1, pp. 7–13.
Tsaur, S. H., Chang, T. Y., Yen, C. H. (2002). The evaluation of airline service quality by
fuzzy MCDM. Tourism Management, 23, pp. 107–115.
Tutmez B., Kahraman S., Gunaydin O. (2007). Multifactorial fuzzy approach to the sawability
classification of building stones. Construction and Building Materials , 21, pp. 1672–1679.
Tuysuz, F., Kahraman, C. (2006). Project risk evaluation using a fuzzy analytic hierarchy
process: an application to information technology projects. International Journal of
Intelligent Systems, 21, pp. 229–284.
Unver B. (1996). A statistical method for practical assessment of sawability of rocks. In: M.
Barla, ed. Eurock ’96. Rotterdam: Balkema, pp. 59-65.
Van Laarhoven, P. J. M., Pedrcyz, W. (1983). A fuzzy extension of Saaty’s priority theory.
Fuzzy Sets and Systems, 11, pp. 229–241.
Vaya and Vikram. (1983). Mineralogical parameters in granite sawing, technical paper,
University of Jodhpur, pp. 7-79.
Wang, M. Y., Elhag, T. M. S. (2006). Fuzzy TOPSIS method based on alpha level sets with an
application to bridge risk assessment. Expert Systems with Applications, 31, pp. 309–
319.
Wang, T. C., Chen, Y. H. (2007). Applying consistent fuzzy preference relations to partnership
selection. Omega, the International Journal of Management Science, 35, pp. 384–388.
Wang, Y. J., (2008). Applying FMCDM to evaluate financial performance of domestic airlines
in Taiwan. Expert Systems with Applications, 34, pp. 1837–1845.
Webb, S.W. Jackson, W.E. (1998). Analysis of blade forces and wear in diamond stone
cutting, ASME J. Manuf. Sci. Eng. 120, pp. 84–92.
Wei X, Wang C.Y, Zhou Z.H. (2003). Study on the fuzzy ranking of granite sawability.
Journal of Materials Processing Technology, 139, pp. 277–80.
Wright and Cassapi. (1985). Factors influencing stone sawability, Industrial Diamond Review, 2, pp. 84-
87.
Xipeng, X. Yiging, Y. (2005). Sawing performance of diamond with alloy coatings, Surf.
Coat. Technol. 198, pp. 459–463.
Xu, X. (1999). Friction studies on the process in circular sawing of granites, Tribol. Lett. 7, pp.
221–227.
Xu, X. Li, Y. Malkin, S. (2001). Forces and energy in circular sawing and grinding of granite,
J. Manuf. Sci. Eng. 123, pp. 13–22.
Xu, X.P. Li, Y. Zeng, W.Y. Li, L.B. (2002). Quantitative analysis of the loads acting on the
abrasive grits in the diamond sawing of granites, J. Mater. Process. Technol. 129, pp.
50–55.
Xu, X. Li, Y. Yu, Y. (2003). Force ratio in the circular sawing of granites with a diamond
segmented blade, J. Mater. Process. Technol. 139, pp. 281–285.
Yoon, K., Hwang, C. L. (1985). Manufacturing plant location analysis by multiple attribute
decision making: Part II. Multi-plant strategy and plant relocation. International
Journal of Production Research, 23(2), pp. 361–370.
Zadeh, L. A., (1965). Fuzzy sets. Information and Control, 8, pp. 338–353.
Zare Naghadehi, .M, Mikaeil .R, Ataei, M. (2009). The Application of Fuzzy Analytic
Hierarchy Process (FAHP) Approach to Selection of Optimum Underground
Mining Method for Jajarm Bauxite Mine, Iran. Expert Systems with Applications, 36,
pp. 8218–8226.
Zhu, K., Jing, Y., Chang, D. (1999). A discussion on extent analysis method and applications
of fuzzy AHP. European Journal of Operational Research, 116, pp. 450–456.
22
A Supporting Decisions Platform

for the Design and Optimization
of a Storage Industrial System
Riccardo Manzini, Riccardo Accorsi, Laura Pattitoni and Alberto Regattieri
Department of Industrial Mechanical plants, Bologna University
Italy
1. Introduction
In a recent survey the consulting company AT Kearney (ELA/AT Kearney survey 2004)
states that there are more than 900,000 warehouse facilities worldwide from retail to service
parts distribution centers, including state-of-art, professionally managed warehouses, as
well as company stockrooms and self-store facilities. Warehouses frequently involve large
expenses such as investments for land and facility equipments (storage and handling
activities), costs connected to labour intensive activities and to information systems.
Lambert et al. (1998) identify the following missions:
 Achieve transportation economies (e.g. combine shipment, full-container load).
 Achieve production economies (e.g. make-to-stock production policy).
 Take advantage of quantity purchase discounts and forward buys.
 Maintain a source of supply.
 Support the firm’s customer service policies.
 Meet changing market conditions and again uncertainties (e.g. seasonality, demand
fluctuations, competition).
 Overcome the time and space differences that exist between producers and customers.
 Accomplish least total cost logistics commensurate with a desired level of customer
service.
 Support the just-in-time programs of suppliers and customers.
 Provide customers with a mix of products instead of a single product on each order (i.e.
consolidation).
 Provide temporary storage of material to be disposed or recycled (i.e. reverse logistics).
 Provide a buffer location for trans-shipments (i.e. direct delivery, cross-docking).
Bartholdi and Hackman (2003) conversely recognize three main uses:
 Better matching the supply with customer demands
Nowadays there is a move to smaller lot-sizes, point-of-use delivery, high level of order and
product customization, and cycle time reductions. In distribution logistics, in order to serve
customers, companies tend to accept late orders while providing rapid and timely delivery
within tight time windows. Consequently the time available for order picking becomes
shorter.
 Consolidating products
The reason to consolidate products is to better fill the carrier to capacity and to amortize
fixed costs due to transportation. These costs are extremely high when the transportation
mode is ship, plane or train. As a consequence a distributor may consolidate shipments from
vendors into larger shipments for downstream customers by an intermediate warehouse.
 Providing Value-added processing
Pricing, labelling and light assembly are simple examples of value added processing. In
particular the assembly process is due for a manufacturing company adopting the
postponement policy. According to this policy products are configured as close to customers
as possible.
As a result warehousing systems are necessary and play a significant role in the companies’
logistics success.
2. Classification and notation

A classification of warehouse design and operation planning problems is illustrated in
Figure 1 (Gu et al., 2007).
A more detailed description of each problem category previously identified is given in Table
1. This paper will focus mostly on both warehouse design issues and the operation planning
problems.
Table 1 reports a large number of problems whose literature presents many studies, models
and supporting decision methods and tools. A limited number of studies present integrated
approaches to face simultaneously a few of these problems which are significantly
correlated. The performance of the generic operation usually depends on design decisions
(see Table 1). As a consequence the authors of this chapter decides to develop, test and
apply an original DSS based on an integrated approach to best design and manage a
warehousing system. It takes inspiration from literature models and algorithms developed
during last two decades.
Main operations and functional areas within a general warehousing system are: receiving,
transfer and put away, order picking/selection, accumulation/sorting, cross-docking, and
shipping.
Fig. 2. Typical warehouse operations (Inspired by: Tompkins et al., 2003) show the flows of
product and identifies the typical storage areas and relative logistic movements.
In particular, the receiving activity includes the unloading of products from the transport
carrier, updating the inventory record, inspection to find if there is any quantity or quality
inconsistency. The transfer and put away involves the transfer of incoming products to
storage locations. It may also include repackaging (e.g. full pallets to cases, standardized
containers), and physical movements (from the receiving docks to different functional areas,
between these areas, from these areas to the shipping docks). The order picking/selection
involves the process of obtaining a right amount of the right products for a set of customer
orders. It is the major activity in most warehouses. The accumulation/sorting of picked orders
into individual (customer) orders is a necessary activity if the orders have been picked in
batches. The cross-docking activity is performed when the received products are transferred
directly to the shipping docks (short stays or services may be required but no order picking
is needed). The storage function is the physical containment of products while they are
awaiting customer demands. The form of storage will depend on the size, quantity of the
products stored, and the handling characteristic of products or their product carriers
(Tompkins et al., 2003).
A Supporting Decisions Platform for the Design and Optimization of a Storage Industrial System 439
Fig. 1. Framework for warehouse design and operation problems, (Gu et al., 2007).
2.1 Order picking systems

Order picking (OP) can be defined as the retrieval of items from their warehouse locations in
order to satisfy demands from internal or external customers (Petersen, 1999). In order
picking systems (OPSs) incoming items are received and stored in (large-volume) unit pallet
loads while customers order small volumes (less than unit loads) of different products as
simply shown in Figure 3. Typically, hundreds of customer orders, each made of many
requests (orderlines), have to be processed in a distribution warehousing system per day.
Design and operation problems Decisions

• Material flow
• Department identification
Overall structure • Relative location of departments
• Size of the warehouse
W arehouse design
Sizing and dimensioning • Size and dimension of departments

• Pallet block-stacking pattern (for pallet storage)
• Aisle orientation
• Number, length, and width of aisles
Department layout • Door locations
• Level of automation
• Storage equipment selection
• Material handling equipment selection (order picking,
Equipment selection sorting)
• Storage strategy selection (e.g., random vs. dedicated)
Operation strategy • Order picking method selection
• Truck-dock assignment
• Order-truck assignment
Receiving and shipping • Truck dispatch schedule
W arehouse operation
• Assignment of items to different warehouse departments

Storage SKU-department assignment • Space allocation
• Assignment of SKUs to zones
Zoning • Assignment of pickers to zones
• Storage location assignment
Storage location assignment • Specification of storage classes (for class-based storage)

Order • Batch size
picking Batching • Order-batch assignment
• Routing and sequencing of order picking tours
Routing and sequencing • Dwell point selection (for AS/RS)
Sorting • Order-lane assignment
Table 1. Description of warehouse design (Gu et al. 2007)
Even thought there have been various attempts to automate the picking process, automatic
systems are rarely found in practice. Order picking, like many other material handling
activities, still is a repetitive and labour-intensive activity. Order picking systems, which
involve human operators can be generally organized in two ways, namely as a part-to-picker
system in which the requested products are delivered automatically to a person at an
input/output (I/O) point, or as a picker-to-parts system in which the order picker travels to
storage locations in order to bring together the required products. Figure 4 gives a
comprehensive classification of OPSs (De Koster 2004).
Fig. 2. Typical warehouse operations (Inspired by: Tompkins et al., 2003)
Fig. 3. From stock keeping units (skus) to customer orders

It can be distinguished two types of picker-to-parts systems: low-level systems and high-level
systems. In low-level OPSs the picker picks requested items from storage racks or bins. Due
to the labour intensity, low level systems often are called manual OPSs. Some other order
picking systems have high storage racks; order pickers travel to the pick locations on board
of a stacker or order-pick truck, or a crane. The crane mechanically stops in front of the
correct pick location and waits for the order picker to execute the pick. This type of system is
called high-level or man-aboard system. Parts-to-picker systems include automated storage and
retrieval systems (AS/RS), using mostly aisle-bound cranes that retrieve one or more unit
loads (e.g. of bins: mini-load system, or pallets) and carry the loads to a pick station (i.e. I/O
point). At this station the order picker picks the right quantity requested by the customer
order, after which the residual stock quantity is stored again. This type of system is also
called unit-load OPS. The automated crane can work under different functional modes:
single, dual and multiple command cycles. The single-command cycle means that either a load
is moved from the I/O point to a rack location or from a rack location to the I/O point. In
the dual-command mode, first a load is moved from the I/O point to the rack location and
next another load is retrieved from the rack. In multiple command cycles, the S/R machines
have more than one shuttle and can pick up several loads in one cycle, at the I/O point or
retrieve them from rack locations.
Fig. 4. Classification of order-picking systems (based on De Koster 2004)

Manual-pick picker-to-parts systems are the most common (De Koster, 2004). The basic
variants include picking by article (batch picking) or pick by order (discrete picking). In the
case of picking by article, multiple customer orders (the batch) are picked at the same time
by an order picker. Many in-between variants exist, such as picking multiple orders
followed by immediate sorting (on the pick cart) by the order picker (sortwhile-pick), or the
sorting takes place after the pick process has finished (pick-and-sort).
Another basic variant is zoning, which means that a logical storage area (this might be a
pallet storage area, but also the entire warehouse) is split in multiple parts, each with
different order pickers. Depending on the picking strategy, zoning may be further classified
into two types: progressive zoning and synchronized zoning. Under the progressive (or
sequential) zoning strategy, each batch (possibly of one order) is processed only in one zone
at a time; at any particular point in time each zone processes a batch that is dissimilar from
the others. Hence, the batch is finished only after it sequentially visits all the zones
containing its line items. Under the synchronized zoning strategy, all zone pickers can work
on the same batch at the same time.
3. Conceptual framework and DSS for warehousing systems

Figure 5 illustrates a conceptual framework for the design, control and optimization of an
industrial storage system. This framework is the result of the integration of different models
and supporting decision methods & tools by the adoption of a systematic multi-step
approach. The proposed approach involves several decisions which rarely are faced
simultaneously by the decision maker. As a consequence he/she has to accept local optima
and sub optimizations.
Main decisions deal with the determination of (1) the system type, e.g. automatic or manual
warehousing system, parts-to-picker or picker-to-parts, unit-load or less than unit-load,
forward-reserve or forward only, etc.; (2) the best storage capacity of the system in terms of
number of pallet locations for each sku; (3) the structure of the system, i.e. the layout and
configuration of the system in terms of racks, bins, aisles, etc.; (4) the allocation of product
volumes to the storage area in agreement with the whole capacity defined by (1) and in
presence/absence of a reserve area; (5) the assignment of products to the storage area; (6) the
evaluation of the performance of the adopted system configuration by the simulation of
vehicles’ routes.
A brief and not exhaustive classification of storage systems types has been introduced in
previously illustrated Figure 4 (as proposed by De Koster 2004). The generic form of the
proposed DSS is made of active tables for data entry, reports, graphs and tables of results,
etc. A "Quick report" section reports all necessary information for the user: for example it is
possible to show the sequence of picking in an order according to a given picking list and to
collect a set of performance indices. Next subsections illustrate main data entry forms and
decision steps for the design of a storage system.
This chapter adopts the following terms many times: fast-pick, reserve, bulk, sku, etc.. Which
is the difference between the fast-pick area and the reserve one? The fast-pick area is a site inside the
warehouse where the most popular skus are stored in rather small amounts so that a large
part of daily picking operations can be carried out in a relatively small area with fast
searching process and short travelled routes. The items most frequently requested by
customers are grouped in this storage area, which is often located in an easily accessible area
so that the time of picking and handling is minimized. The location of the items in the fast
pick area is better than any other in the warehouse and related operations, e.g. stocking,
travelling, searching, picking, and restocking, are faster.
3.1 Storage capacity evaluation

The proposed DSS adopts two alternative approaches for the so-called stock capacity
evaluation (step 2): historical inventory (2/HI) based approach and demand profile (2/DP)
based approach. 2/HI identifies a storage capacity of a warehousing system in agreement
with a set of historical, e.g. monthly, stock levels and a specified risk of stockout, as a measure
of probability a generic stockout occurs in a period of time, such a year. The generic value of
inventory level usually refers to the global storage quantity (volume) level of products
including both the level for picking and the level for reserve: companies rarely collect
historical data on their storage quantities and very rarely they distinguish the storage levels
in fast pick area from corresponding levels in bulk area.
The proposed platform supports the determination of the storage level constructing a non
parametric class based frequency analysis and/or a non parametric continuous frequency based
analysis. These are non parametric statistical analyses because they do not identify statistical
distributions, e.g. Normal, Lognormal, Weibull, etc., as a result of best fitting available data
(Manzini et. al 2010). These analyses are the basis for estimating percentiles of historical
variation of inventory, i.e. the expected risk of stockout adopting a specific level of storage
capacity and assuming future movements of products into/out of the system similar to the
historical ones.
The so-called class based analysis of historical storage quantities generates a histogram of
frequency values of storage levels collected in the adopted historical period of time by the
preliminary definition of a number of histogram classes of values. The histogram of
cumulative values identifies the probability of "stockin", i.e. the probability of the
complementary event of the stockout (the probability that stockout would not occur).
The continuous frequency based analysis generates a similar set of graphs without the
preliminary definition of a number of classes of values (historical measures).
The so called DP approach identifies the best level of storage capacity by the analysis of
historical demand profiles. Given a period of time, e.g. one year, and a set of values of
demand quantities for each product within this period, DP quantifies the expected demand
during an assumed subperiod of time t, called time supply (e.g. 3 weeks). As a consequence
this approach assumes to store an equal time supply of each sku. This is a frequently adopted
strategy in industrial applications and is widely discussed by Bartholdy and Hackman
(2003): the equal time strategy - EQT. By this strategy the storage system should supply the
expected demand orders for the period of time t without stockouts. Obviously this depends
on the adopted fulfilment system, which relates with inventory management decisions
significantly correlated to the storage/warehousing decisions object of this chapter.
The output of a storage capacity evaluation is the storage volume of products in the fast pick
area (adopting a forward-reserve configuration system) and the whole storage capacity
(including the bulk area when forward-reserve configuration is adopted). In presence of fast
picking, it usually refers to the lowest level of storage: the so-called 0 level.
This capacity is usually expressed in terms of cubic meter, litres, number of pallets, cases,
carton, pieces of products, etc.
Figure 6 presents the form of the proposed DSS for data entry and evaluation of historical
storage levels given an admissible risk of stockout. This figure also shows a curve “risk of
stockout” as a result of the statistical analysis of historical observations: this is the stockout
probability plot. Obviously given a greater storage capacity this risk decreases.
3.2 Structure of the system: layout & configuration

This section deals with the determination of the layout and configuration of the storage
system as the result of warehouse type (see previous discussion and classification) selection
including the existence/absence of the forward-reserve strategy, picking at low/high levels,
etc.; pallet/unit load dimensions; racks and shelves dimensions; adopted vehicles for
material handling. It is important to underline that the layout of the storage area
significantly depends on the width of the aisles which have to host different kinds of
vehicles:
 vehicles for the pallet-loading/put-away process, which usually involves unit loads;
 vehicles for restocking (generally unit loads) in presence of a forward-reserve system;
 vehicles for picking (unit loads and less than unit loads) at low/high levels.
Figure 7 and 8 present the forms for data entry of unit load parameters and warehouse setting
respectively. All vehicles are characterized in terms of routing strategies, distinguishing
traversal from return (Manzini et al. 2006). It is possible to distinguish the vehicle parameters
for put-away, restocking and picking. The shape factor of a storage system is the ratio
between the frontal length and longitudinal length of system layout: this value can be
optimized in presence of pallet-loading of unit loads, given the location of the I/O depot
area and adopting a shared, i.e. randomized, storage allocation of products. In any different
hypotheses, e.g. in presence of less than unit load picking activities, there is not an optimal
value of this ratio and the user can arbitrary choose it.
Fig. 5. Conceptual framework

Fig. 6. Historical storage levels analysis and capacity evaluation.
Fig. 7. Data entry, unit load parameters.

Fig. 8. Data entry, warehouse parameters.

Many other parameters are not described in this brief illustration of the proposed DSS.
Figure 9 shows an exemplifying set of reports collecting the results as output of layout &
configuration design. They are grouped in different sections: historical stock, BM
required, warehouse sizing, etc. A few exemplifying results are: number of historical
observations, storage volume available, number of levels, number of bays, number of
aisles, etc.
Fig. 9. Warehouse sizing reports
3.3 Items allocation

This sections deal with the application of the so called allocation strategy, i.e. the
determination of the fraction of storage volume for each sku that is a product (also called
item). In particular the manager is interested in the following critical question: which is the
best amount of space to assign to any skus? This question refers to the fast pick area in presence
of a forward-reserve storage system and adopting a dedicated storage, which adopts fixed
storage locations for the generic sku.
Bartholdi and Hackman (2003) discuss this issue in order to reduce the number of restocks
in a fast pick (forward) and bulk storage (reserve) picking system. The fast pick area cannot
contain the right volume of each item required to satisfy the total customer demand in a
specific supply period of time e.g. a month or a year. Consequently, it is necessary to ensure
replenishment of picked goods from a bulk storage area, known as reserve area.
Therefore, consideration has to be given to an appropriate choice between the space
allocated for an item in the fast pick area and its restock frequency (Bartholdi and Hackman,
2003). They discuss three different allocation strategies for calculating the volume to be
assigned to each sku assuming it incompressible, continuously divisible fluid. The models
proposed for determining the sku level of stock are based on the following notation:
 let fi be the rate of material flow through the warehouse for the skui;
 let the physical volume of available storage be normalized to one. vi represents the
fraction of space allocated to skui so that:
∑ =1 (1)
Three different levels of stock for skui are defined as follows:
i. Equal Space Strategy (EQS). This strategy identifies the same amount of space for each
sku. The fraction of storage volume to be dedicated to the skui under EQS is:
= (2)
ii. Equal Time Strategy (EQT). In this strategy each sku i is replenished an equal number of
times according to the demand quantities during the period of time considered. Let K
be the common number of restocks during a planning period so that:
= (3)
From equations (1) and (3):
=∑ (4)
The fraction of storage volume to be dedicated to the skui under EQT is:
= =∑ (5)
iii. Optimal Strategy (OPT). Bartholdi and Hackman (2003) demonstrate that this strategy
minimizes the number of restocks from the reserve area. The fraction of available space
devoted to skui is:
=∑ (6)
A critical issue supported by the what-if multi-scenario analysis, which can be effectively
conducted by the proposed DSS, is the best determination of the fraction of storage volume
for the generic sku as the result of the minimization of pickers travelling time and distance
in a forward – reserve picker to part order picking system.
Equations (2), (5), and (6) are fractions of fast pick volume to be assigned to the generic item
i. As a consequence it is necessary to preliminary know the level of storage to be assigned to
the fast pick area in order to properly defined each dedicated storage size. A company
usually traces and knows the historical picking orders with a high level of detail (date of
order, pickers, picked skus and quantities, visited locations, pickers id, restocking
movements, etc.) thanks to traceability tools and devices (barcode, RFID, etc.), but it rarely
has "photographs" of the storage systems, i.e. historical values of storage levels. In presence
of forward-reserve systems and a few (or many) photographs of storage levels, the generic
inventory level, made available by the warehouse management system, does not distinguish
the contribution due to fast pick area and that stored in bulk area as discussed in section 3.1.
The proposed DSS supports the user to define the whole level of storage for fast picking, the
level of storage for bulk area and those to be assigned to each sku in both areas.
Items allocations affect system performance in all main activities previously discussed:
pallet-loading (L), restocking/replenishment (R) and picking (P). This is one of the main
significant contributions of the proposed DSS to knowledge and warehousing system
optimization. The following questions are still open: Which is the effect of storage allocation to
travel time and distances due to L, R and P activities? Given a level of storage assigned to the fast
pick area, is the OPT the best allocation storage for travel distance and time minimization?
We know that the OPT rule, equation (6), supports the reduction of the number of restocks
(R) but the logistic cost of material handling is also due to (I) and (P) activities. The multi-
scenarios what-if analysis supported by the proposed DSS helps the user to find the most
profitable problem setting which significantly varies for different applications.
Figure 10 presents exemplifying results of the item allocation for an industrial application.
This is a low level forward-reserve OPS: for each sku the number of products in fast pick
and reserve areas is determined.
Fig. 10. Allocation of products, fast-pick area.
3.4 Skus assignment

The storage assignment problem deals with the assignment of products to storage locations in
order to identify which is the best location for the generic product (Cahn, 1948). Decisions
on storage assignment affect both time and costs due to I, R and P activities. The assignment
problem has been formalized by Frazelle and Sharp (1989) and classified as a Non –
Polynomial (NP) hard problem.
A list of typical indices adopted to rank the skus for the assignment of storage locations
follows:
 Popularity (P). This is defined as the number of times an item belongs to an order in a
given set of picking orders which refer to a period of time T:
, =∑ (7)

where
1
=
0 ℎ
(xij) product-order incidence matrix.
 Cube per Order Index (COI) can be defined as the ratio of volume storage to inventory for
the generic sku to the average number of occurrences of sku in the order picking list for
a given period of time (Haskett, 1963). Given an sku i, COI is defined as the ratio of the
volume of the stocks to the value of its popularity in the period of interest T. Formally:
,
, =∑ (8)

where
vi,T average storage level of sku i in time period T.
 Order Closing Index (OC). Order Completion (OC) assignment is based on the OC
principle introduced by Bartholdi and Hackman (2003). Bindi (2010) introduced the
Order Completion rule based on an index called OC index that evaluates the probability
of a generic item being part of the completion of an order, composed of multiple
orderlines of different products (items). The OC index is the sum of the fractions of
orders the generic item performs. For a generic sku and a time period T, OC is defined
as follows:
, =∑ , (9)
where
xij
f ij ,T  m ( j)
(10)
order j in
period T
 x ij
z1
m(j) number of orderlines for picking order j.

According to the previous hypotheses the OC index for a certain item can assume the
following special values:
 Minimum value = 1/ Total Number of Items, when the item belongs to all the customer
orders.
 Maximum value = number of orders, when the item belongs to all customers orders and
there are no other items.
 Turn Index (T). Given an sku i, it is defined as the ratio of the picked volume during a
specific period of time T to the average stock stored in T. The index can be written as:
 pij
j
Ti ,T  (11)
order j in vi ,T
period T
where
pij picked volume of product i in the order j. The unit of

measurement is the same as vi,T.
The literature presents several storage assignment policies that can be classified in one of the
following main categories (Van der Berg and Zijm 1999, Manzini et al. 2006 and 2007):
 Randomized Storage
This policy provides for skus randomly assigned to the first available space in the
warehouse. The random storage policy is widely adopted in the warehousing industry
because it is easy to use, often requires less space than other storage methods, and uses
all the picking aisles intensively.
 Dedicated Storage
This policy reserves specific locations for each sku within the warehouse. It requires
more space in the pick area for storage but allows the pickers to memorize fixed
locations of skus producing time labour saving. The choice of dedicated location to
assign a generic item follows one of the following rules:
 class based storage rule. This rule defines several classes as groups of skus located in
storage areas more or less favorable to satisfying particular criteria. Frazelle (2002)
punctually states the two most frequently used criteria used to assign a class of
products to storage locations are popularity and the cube per order index (COI) as
defined by Haskett (1963).
 ranked index based rules. They are based on the ascending or descending values of
one of the previously introduced indices e.g. P, COI, OC, or T defined for each sku.
The P-based assignment rule considers a list of items sorted by decreasing value of
popularity and assigns the highest of them to the nearest location from the depot
area (I/O point) i.e. the most favourable location. The COI-based assignment rule
considers a list of items sorted by increasing value of COI and assigns the lowest of
them to the most favourable location. The OC-based assignment rule arranges
items in a similar way to the P-based rule: it considers a list of items sorted by
decreasing value of OC index and assigns the highest of them to the most
favourable locations. The Turn-based assignment rule assigns items in the same
way as for the previous OC rule but uses Turn index instead of OC.
 correlated storage policy. This policy locates items with a high degree of correlation
close to each other, which is usually based on the frequency of being in different
picking orders. The allocation of products within a storage area can be based on
different types of correlation existing between products. Once the correlation has
been calculated for all pairs of products, the couples with the highest value of
correlation are stored together. For example, customers may usually order a certain
item together with another. These products might reasonably have high correlation
and it may be useful to locate them close together within the system to reduce the
travelling distance and time during the picking activity. In order to group
products, the statistical correlation between them should be known or at least be
predictable, as described by Frazelle and Sharp (1989), and by Brynzér and
Johansson (1996).
The proposed tool assigns the location to the generic sku by the Cartesian product, as the
direct product of two different sets. The first set (RANKsku)i is made of the list of skus
ordered in agreement with the application of a ranking criterion (see Figure 10), e.g. the
popularity measure based rule or a similarity based & clustering rule (Bindi et al. 2009). The
second set is made of available locations ordered in agreement with a priority criterion of
locations assignment, e.g. the shortest time to visit the location from the I/O depot area. As
a consequence most critical sku are assigned to the nearest available locations. Obviously
the generic sku can be assigned to multiple locations in presence of more than one load
stored in the system.
This assignment procedure refers to the products quantities subject to picking, e.g. located
in the so called fast picking area: this is the so called low level OPS. In high level systems all
locations at different levels can be assignable in agreement with the adopted ranking
procedure. Both types of warehouses are supported by the proposed DSS. Figure 10 shows
the form for setting the assignment of products within the system.
Fig. 10. Storage assignment setting, ranking criterion and ranking index.
Correlated storage assignment rules are supported by the DSS in agreement with the
systematic procedure proposed and applied by Bindi et al. (2009 and 2010). Figure 11
exemplifies the result of the assignment of products within the fast pick area by the use of
different colours, one for each sku. Similarly the result of the assignment of products to the
higher levels can be shown as illustrated in Figure 12.
3.5 Performance evaluation

This section deals with the evaluation and analysis of the performance of the systems in
terms of monetary and not monetary costs. Examples of the second set of costs: meters and
hours spent in travelling by the pickers in a period of time, e.g. one day, a year, etc.; number
of vehicles and pickers; percentage level of use for each logistic resource.
Given a system layout, an allocation of skus to fast pick area, the assignment of products,
and the performance of adopted vehicles, it is possible to simulate the routes travelled by
pickers and restockers to satisfy (L), (R) and (P) activities.
The analysis can be conducted comparing different operating scenarios as the result of
different settings (system layout & configuration, storage allocation, storage assignment,
etc.) as previously illustrated, and it is supported by a visual animation and a set of graphs
and summary tables as illustrated in Figure 13. The adopted parameters of the what-if
setting are: routing strategy, ranking index, allocation strategy, and the depot location.
Fig. 11. Storage assignment results, fast-pick area.

Fig. 12. Storage assignment results, reserve area.
Fig. 13. Multi-scenario what-if analysis.

Figure 14 shows the form for the visual simulation of the picking orders in forward-reserve
OPS. This simulation also quantifies the costs due to restocking. Similarly it is possible to
simulate the behaviour of the system including pallet loading (L) activities, and/or in
presence of AS/RS (adopting Chebyshev metric), and/or in presence of correlated storage
assignment.
Fig. 14. Visual animation and simulation run.
4. Case study
The proposed DSS has been applied to a low level picker to part OPS for spare parts of
heavy equipment and complex machinery in a popular manufacturing company operating
worldwide. The total number of items stored and handled is 185,000 but this is continuously
growing due to new business acquisitions and above all to engineering changes to address
new requirements for pollution control and reduction.
The subject of the analysis is the picking activities concerning medium-sized parts weighing
less than 50 pounds per piece. These parts are stored in light racks corresponding to about
89,000 square feet of stocking area. This area contains more than 3,000 different items.
The horizon time for the analysis embraces the order profile data during four historical
months. The number of order picking lines is 37,000 that correspond to 6,760 different
customer orders. The picking list presents an average of 86 orders fulfilled per day with the
average depth varying around 6 items per order.
The result of the design of the order picking system is a 58,400 square foot picking area (350
feet x 170 feet). Table 2 demonstrates that OPT strategy significantly reduces the number of
restocks for the historical period of analysis in agreement with Bartholdi and Hackman
(2010). The reduction is about 55% compared to EQS, and about of 62% compared to EQT,
thus confirming the effectiveness of OPT strategy.
Table 3 reports the values of traveled distances and aisles crossed in retrieving operations
i.e. excluding restocking for different assignment rules and allocation strategies. Table 3
demonstrates that COI and P assignment rules reduce picking activities and cost the most.
In particular, the best performance is obtained by adopting the COI assignment rule and the
EQS allocation strategy, quite different from the OPT strategy which minimizes the number
of restocks (see Table 2).
EQS EQT OPT

Total Restocks 3,650 4,269 1,635
% Reduction 55.2% 61.7%
Table 2. Restocks with different allocation strategies
Allocation strategies
EQS EQT OPT
Traveled Aisles Traveled Aisles Traveled Aisles
Assignment rules
distance crossed distance crossed distance crossed
COI 6,314,459 33,579 6,025,585 33,659 6,706,537 34,482
OC 6,536,697 33,922 8,047,296 36,210 7,241,533 35,424
P 6,379,887 33,713 7,254,318 35,270 6,869,774 34,655
T 8,015,507 35,766 8,155,378 36,191 8,717,042 36,497
Table 3. What-if analysis results. Traveled distance [feet] and aisle crossed [visits] during a
picking period of 4 months
Figure shows where the most frequently visited skus are located in the fast pick area: the
size of circles is proportional to the popularity value respectively according to the return
and the traversal strategies.
5. Conclusions and further research

This chapter presents an original DSS for the design, management and optimization of a
warehousing system. The large amount of decisions is usually faced separately as
demonstrated by the literature proposing sub-optimal models and supporting decision
methods. The proposed DSS is the result of the integration of different decisions, models
and tools by the adoption of a systematic and interactive procedure. It supports the design
of the system configuration, the allocation of skus, and their assignment to storage location,
the vehicle routing and sequencing within the system. The evaluation of the performance is
supported by the dynamic construction of vehicle routes to satisfy material handling needs
collected in a period of time, named observation period.
Further research is expected on the following topics of interest:
 3D computer aided design - CAD of the mechanical structure of the system as a result
of the best system configuration
 System validation and analysis of vehicle congestions by the execution of a dynamic

and visual evaluation of system performance. A similar analysis can be conducted by
the adoption of visual interactive simulation commercial tool, e.g. AutoMod TM
simulation software. The development of ad hoc tools for a similar analysis conducted
on warehousing systems is achieved.
Fig. 15. Storage assignment in return strategy with I/O located at (x,y)=(170,0)
6. Acknowledgment
The authors would like to thank Prof. John J. Bartholdi of the Georgia Institute of
Technology who gave us several useful suggestions to improve the research.
7. References
Bartholdi, J. and Hackman, S.T., 2003. Warehouse & distribution science,
<http://www2.isye.gatech.edu/people/faculty/John_Bartholdi/wh/book/editio
ns/history.html> (date of access: January 2010).
Bindi, F., Manzini, R., Pareschi, A., Regattieri, A., 2009, Similarity-based storage allocation
rules in an order picking system. An application to the food service industry.
International Journal of Logistics Research and Applications. Vol.12(4), pp.233-247.
Bindi, F., Ferrari, E., Manzini, R., Pareschi, A., 2010, Correlated Storage Assignment and
Isotime Mapping for Profiling Sku, XVI International Working Seminar on production
Economics, Innsbruck (Austria),Volume 4, pp 27-41. Edited by W.Grubbström and
Hans H. Hinterhuber.
Bindi, F., 2010. Advanced Models & Tools for Inbound & Outbound Logistics in Supply
Chain, PhD thesis in industrial engineering of University of Padova, 81-86.
Brynzér, H. and Johansson, M.I., 1996. Storage location assignment: using the product
structure to reduce order picking times. International Journal of Production Economics,
46, 595-603.
Cahn, A.S., 1948. The warehouse problem, Bulletin of the American Mathematical Society, 54,
1073.
De Koster, R. and T. Le-Duc, 2004. Single-command travel time estimation and optimal rack
design for a 3-dimensional compact AS/RS, Progress in Material Handling Research:
2004, 2005, 49-66
Ela/ATKearney, 2004, Excellence in Logistic.
Frazelle, E. H., Sharp, G. P., 1989. Correlated assignment strategy can improve any order
picking operation. Industrial Engineering, 21, 33-37.
Frazelle, E.H., 2002. World-class Warehousing and Material Handling. McGraw Hill, New York.
Haskett, J.L., 1963, Cube per order index – a key to warehouse stock location, Transportation
and Distribution Management, 3, 27-31.
Gu, J., Goetschalckx, M., McGinnis, L.M., 2007. Research on warehouse operation: A
comprehensive review, European Journal of Operational Research, 177(1), 1-21.
Lambert, D.M., Stock, J.R. and Ellram, L.M., 1998. Fundamentals of logistics management.
Singapore: McGraw-Hill.
Manzini, R., 2006. Correlated storage assignment in an order picking system, International
Journal of Industrial Engineering: Theory Applications and Practice, v 13, n 4, 384-394.
Manzini, R., Gamberi, M., Persona, A., Regattieri, A., 2007. Design of a class based storage
picker to product order picking system, International Journal of Advanced
Manufacturing Technology, v 32, n 7-8, 811-821, April 2007.
Manzini, R., Regattieri, A., Pham, H., Ferrari, E., 2010, Maintenance for industrial systems.
Springer-Verlag, London.
Petersen, C.G., 1999. The impact of routing policies on warehousing efficiency, International
Journal of Operations and Production Management, 17, 1098-1111.
Tompkins, J.A., White, J.A., Bozer, Y.A., Frazelle, E.H. and Tanchoco, J.M.A., 2003. Facilities
Planning. NJ: John Wiley & Sons.
Van den Berg, J.P., Zijm, W.H.M., 1999, Models for warehouse management: classification
and examples, International Journal of Production Economics, 59, 519-528.
23
Challenges in Climate-Driven Decision

Support Systems in Forestry
Oleg Panferov1, Bernd Ahrends2, Robert S. Nuske2,
Jan C. Thiele1 and Martin Jansen1
1Georg-August University of Goettingen
2Northwest German Forest Research Station
Germany
1. Introduction
The history of Decision Support Systems in forestry is quite long as well as the list of created
systems and reviews summarizing their merits and flaws. It is generally recognized that a
modern decision support system (DSS) should address simultaneously as many economical
and ecological issues as possible without becoming overly complex and still remain
understandable for users (Reynolds et al., 2008). The ongoing global change including the
climate change sets new boundary conditions for decision makers in the forestry sector. The
changing growth conditions (Albert & Schmidt, 2010) and expected increasing number of
weather extremes like storms force forest owners to make decisions on how to replace the
damaged stands and/or how to mitigate the damages. This decision making process
requires adequate information on the future climate as well as on complex climate-forest
interactions which could be provided by an appropriate climate-driven decision support
tool. Both the damage factors and the forest management (e.g. harvesting) result in changes
of the structure of forest stands. The structural changes result in immediate changes of
albedo and roughness of land surface as well as of microclimatological conditions within the
stand and on the soil surface. The consequences are manifold. The changed stand density
and leaf area index trigger energy and water balance changes which in turn increase or
decrease the vulnerability of the remaining stand to abiotic and biotic damage factors like
droughts or insect attacks. A change of the microclimatic conditions might strengthen the
forest against drought, but at the same time reduce its resistance to windthrow. The sign
and extent of vulnerability changes depend on complex interactions of the effective climatic
agents, above- and belowground forest structure, and soil. There are many DSS that are
capable of assessing one or several risk factors; however there are few that are able to assess
the additional increase or decrease of risks triggered by modification of forest structure
resulting from previous damage or forest management activities. Disregarding these effects
will inevitably lead user to either under- or overestimation of the potential damages. The
question arises whether these additional risks are significant enough to be considered in a
DSS.
In this chapter we present a new DSS developed according to the above mentioned
requirements and capable to provide decision support taking into account economical and
ecological considerations under the conditions of changing climate - the Decision Support
System – Forest and Climate Change. We then use the modules of that system to investigate
the contribution of additional forest damage risks caused by structure changes into the
general vulnerability of forest stand under changing climate.
2. DSS WuK
The Decision Support System – Forest and Climate Change (DSS-WuK) was developed to
assist forest managers in Germany selecting an appropriate tree species under changing
climate, considering future risks (Jansen et al., 2008, Thiele et al., 2009). It operates with a
daily time step and combines traditional statistical approaches with process-based models.
2.1 Input
The system is driven by SRES-climate scenarios and hindcast data modelled by coupled
ocean - atmospheric general circulation model ECHAM5-MPIOM developed by the Max-
Planck-Institute for Meteorology (Roeckner et al. 2006). Additional input data are a digital
elevation model (Jarvis et al., 2008) and its derivatives: aspect and slope, soil data from the
national forest soil map (Wald-BÜK, Richter et al., 2007) at a resolution of 1:1 Mio. and
atmospheric nitrogen deposition estimated using a modified version of the extrapolation
model MAKEDEP (Alveteg et al., 1998, Ahrends et al., 2010a) with data from Gauger et al.
(2008). The user of the DSS provides additional input on stand characteristics (tree species,
age, yield class) on the forest of interest.
2.2 Model structure

Besides the climate-dependent forest growth the system estimates also biotic (insect attacks)
as well as abiotic risks (wind damage and drought mortality). At present the system is able
to evaluate the site sensitive growth and risks for the five most important tree species in
German forestry (Norway spruce, Scots pine, European beech, Sessile oak, Douglas fir) of any age
and state growing on any of the 124 soils typical for Germany. Thus, the economic outcome
estimated by the DSS-WuK is based on complex interactions of many factors, e.g. climate,
soils, tree species, stand characteristics and relief. The results are averaged and summed up
over climatological periods of 30 years as defined by the World Meteorological
Organisation. The period of 1971-2000 was chosen as the “present state” and replaced the
old standard reference period of 1961-1990. To characterize the future development of
forests the system provides information aggregated for the three periods: 2011-2040, 2041-
2070 and 2071-2100.
The system is web-based and freely accessible. To tackle the usual requirements of
transparency, focus on target audience, and usability we involved the potential end users
into the development process during all phases. As a result the system took a two-stage
form:
 The first stage presents pre-processed data covering entire Germany. The report
includes growth and yield data, the various risks, and an economic evaluation. The user
can choose a tree species, soil conditions at the spatial resolution of the employed
regional climate model (e.g. 0.2°x0.2° of CLM, Datastream 3). The report of the first
stage is, therefore, based on rather coarse input data and enables the user to decide
whether a more detailed analysis (second stage) is needed. The obvious advantage of a
system based on pre-processed data is the very short response time. The disadvantages,
however, are the coarse spatial resolution and the fixed stand characteristics for the
Challenges in Climate-Driven Decision Support Systems in Forestry 461
different tree species, since the first stage does not include stand growth. To avoid the
underestimation of risks the stands are chosen roughly at their rotation age, as these
stands have the highest vulnerability to wind and insect damage.
 If a user is interested in a detailed description of changes in growth and yield data or
the various risks, she/he can obtain a more sophisticated analysis by starting the
second stage of the system. In this part of the DSS-WuK the on-demand simulations for
a user-defined stand with downscaled climate data (1 km resolution) are carried out.
Because of the relatively long response time, the system sends a report as a pdf file via
email. In the second stage, the stand does not have a static structure any more but
grows climate-dependent using an individual-tree growth model. The stand
characteristics are updated every 10 years and submitted to the risk modules which
carry out their calculations with a new structure with a daily time step (Fig. 1). The
effect of growing stands on the risks is evaluated, meaning the stand characteristics
determine the effect of damage factors, i.e. extreme weather events. The risks in the
system are defined as a percentage of area loss of a fully stocked stand.
Fig. 1. A simplified flow chart of the core model process in DSS-WuK.

Both stages use the same core model (Fig. 1) and all calculations are carried out with daily
resolution. All models get an update of their input data after each 10 year period. The risk
model for drought mortality is coupled to the soil water model. The soil water itself, is
influenced inter alia by the current stand characteristics. The soil water and climate
conditions influence the output of the wind damage and insect attack risk models as well as
the stand growth model. In the next iteration, the soil water model gets the stand
characteristics actualized by the stand growth model. The results are finally aggregated to
30 year climatological periods mentioned above.
However, working at a level of forest stand the model is not able to consider the effect of the
risks on stand growth, structure and microclimate, yet (Fig. 1). The estimated damage risks
are not taken into account in calculations of stand growth and are executed (conceptually) in
parallel. The system, thus, ignores the fact that a tree, which was, for example, thrown by
wind, cannot be damaged by insects or drought later on. Overcoming this weakness is one
of the challenges in forest decision support systems development. In the following, we will
present the importance of this point by performing numerical experiments with a reduced-
complexity, directly-interacting part of the simulation system DSS-WuK.
3. Setup of the experiment: risk assessment and damage factors interactions

under current and future climate
The main purpose of the article is to demonstrate the interaction of factors influencing the
forest vulnerability to abiotic risks and the feedback of forest damage to forest
vulnerability under real and projected future climatic conditions. Results for a real forest
stand under present climate are presented first, in order to show how the 30-years
aggregated risk probabilities are formed. A detailed analysis of risk probabilities and
interaction of different factors leading to or inhibiting the forest damage was performed
on example of the real spruce stand research forest site “Otterbach”, Solling, Central
Germany. The meteorological data measured at the station of German Weather Service
nearest to the forest stand with more than 30-years series of meteorological measurements
were used as the basis for the investigation. The numerical experiments and analysis were
carried out for generated idealized boreal forest stands under the conditions of climate
scenarios C20 (reference) and SRES A1B.
To demonstrate the abilities of the DSS tool to evaluate the dynamics of abiotic risks and
their variability for various climatic conditions the idealized forest stands growing in
maritime and continental boreal zones of Finland and Russia, respectively, were generated.
The hypothesis that similar forest damage can trigger different feedbacks in different
climatic zones and under different development of climate conditions will be evaluated. The
mechanism how the forest structure changes caused by forest damage or forest management
activities are practically taken into account in coupled modules of DSS-WuK is described
below.
3.1 Feedbacks of forest structure changes

Generally any changes in forest structure influence a forest ecosystem in many ways and
perform a feedback to forest microclimate and stand vulnerability. In ecosystem and Soil-
Vegetation-Atmosphere-Transfer process-based models the interaction of factors and the
feedbacks can be summarized as shown in Fig. 2. The damage or thinning event reduces
stand density and, thus, the Leaf Area Index (LAI) which is one of the most important
parameters in forest ecosystem models. The decrease of leaf area leads immediately to the
decreasing of rain and light interception and, therefore, to the enhanced supply of light and
water under the forest canopy. This in turn results in increased evaporation from soil
surface under the canopy, but decrease of transpiration due to reduced LAI. Thus, the
interplay of all these factors may lead to the increase or decrease of soil moisture depending
on particular site conditions: latitude, meteorological conditions, soil type, tree species,
initial LAI, extent of damage/management etc. These changed conditions may increase or
decrease forest vulnerability and the probability of various consequent damages. If, for
example, the resulting effect of windthrow would be an increase of soil moisture in a stand
it would, on the one hand lead to the stand destabilisation against next storm, but at the
same time – reduce the probability of drought stress (Ahrends et al., 2009).
Fig. 2. Impacts of changing forest structure on stand vulnerability resulting from triggered
interaction of ecosystem processes.
Therefore, to quantify the contribution of additional damage caused by forest structure
changes the experiments with each of the chosen forest sites were carried out twice as
shown in Fig. 3:
 with static structure, i.e. the climate caused structure damages were registered and
summed up for the 30-years climatic periods, but the actual structure of modelled
stands remained unchanged.
 with dynamic structure, i.e. the part of stand was removed according to the damage
degree and the calculations proceed with the changed structure until the next damage
event and so on for the rest of particular 30-years period.
3.2 Site description, tree species, and soil conditions

3.2.1 Otterbach, Solling
A 85-year-old pure spruce stand in Otterbach, located in the central part of Germany about
60 km North-West of Göttingen in the Solling highland, at 51.77° N and 09.45° E and about
300 m above sea level, was selected to study the possible effects of windthrow events on
forest vulnerability to following windthrow damage risk – windthrow feedback. The
characteristics of the investigated spruce stand are given in Table 1.
Fig. 3. The scheme of calculations with damage-independent structure (upper panel) and
with forest structure updated (LAI and stand density reduced) after each damage event
during 30-years period (lower panel).
Parameter Unit Value Source

age years 85 Olchev et al., 2009
tree density trees ha-1 350 Olchev et al., 2009
average tree height m 34.0 Olchev et al., 2009
diameter at breast height, dbh cm 36.0 Olchev et al., 2009
max. leaf conductance cm s−1 0.53 Federer et al., 2003
max. leaf area index m2 m−2 4.6 Olchev et al., 2009
relative winter LAI - 0.8 Hammel & Kennel, 2001
Fine root length m m-2 3100 Federer et al., 2003
Critical leaf water potential MPa -2.0 Czajkowski et al., 2009
Albedo - 0.14 Federer et al., 2003
Albedo with snow - 0.14 Federer et al., 2003
Table 1. Structural characteristics and model parameters of the forest in Otterbach (central
Germany).
To describe the soil water budget a characteristic soil profile for this stand was selected
(Fröhlich, 2009). According to FAO (1990) the soil type is a Dystric Cambisol derived from
Triassic sandstone covered by loess. The soil hydraulic parameters are given in Tab. 2.
According to Fröhlich (2009) free drainage is accepted at the lower boundary condition at
2 m depth.
Depth Stone Ψu θu θs b Ku fi
[%] kPa - - - mm day-1 -
+5-0 (FF) 0 -6.3 0.384 0.848 5.23 2.3 0.92
0-5 2 -6.3 0.615 0.621 6.4 4.98 0.92
5-15 3 -6.3 0.615 0.621 6.4 4.98 0.92
15-20 3 -6.3 0.508 0.517 6.12 4.98 0.92
20-40 3 -6.3 0.508 0.517 6.12 4.4 0.92
40-60 30 -6.3 0.472 0.482 8.7 4.4 0.92
60-70 30 -6.3 0.472 0.482 8.7 1.76 0.92
70-95 30 -6.3 0.371 0.377 26.5 1.76 0.97
95-200 30 -6.3 0.371 0.377 26.5 1.28 0.97
Table 2. Soil hydraulic at saturation (subscript ‘s’) and at the upper limit of available water
(subscript ‘u’). Note that ψ is matrix potential, K is hydraulic conductivity, θ is volumetric
water fraction, b is the pore-size distribution parameter, fi is the Clapp-Hornberger
inflection point, Ku is the hydraulic conductivity at the upper limit of available water and FF
is the forest floor. The stone content was taken from Feisthauer (2010).
The climate conditions at the research site “Otterbach” in Solling was characterized by
means of long-term measurements at the meteorological station “Silberborn” (51.76°N,
9.56°E, 440 m a.s.l.) of German weather Service located near “Otterbach” site. The
measurements time series cover the period from 1976 to 2008. To account for 140 m altitude
difference between “Otterbach” site and “Silberborn” station the station’s temperature was
corrected using constant gradient of 0.0065°C m-1.
3.2.2 Finland and Russia

The changes of forest risks in 21st century in dependence on changing climate were projected
for two idealized forest stands located in two variations of boreal climate zone: a maritime
one (Finland) and a continental one (Central Russia). The location for Finnish site was
approximated at the centre of ECHAM5 grid cell (62.48°N, 28.125°E) and the location of
Russian site – at 53.16°N, 91.875°E. To separate the effects of climate and damage feedbacks
from the effects of other factors – soils, topography, vegetation composition etc. both Finnish
and Russian stands were modelled identically: monocultural, mean tree height of 16 m,
mean diameter at breast height (dbh) of 20 cm, and a stand density of 860 trees ha-1. The
Podzol soil type was assumed as typical for boreal zone according to FAO (2003): Map of
World Soil Resources. The soil characteristics are given in FAO (2003) and therefore do not
need to be repeated.
3.3 Modelling methods

The climate, wind and drought damage modules of DSS-WuK were used to investigate the
contribution of additional structure-caused risks to the vulnerability of forest stands. The
forest structure dynamics leads to both short- (days, months) and long-term (years, climatic
periods) changes. To investigate the dependence of forest risk factors and of additional
damage on short- and long-term climate variations we used the SRES climate scenario A1B
(IPCC, 2000) modelled by a coupled ECHAM5-MPIOM general circulation model for AR4 of
IPCC as climatological input.
3.3.1 Leaf area index

For modelling the leaf area index dynamics the LAI model by Ahrends et al. (2010b),
applied previously at the Solling site (Sogachev et al., 2010, Panferov et al., 2009, 2010,
Ahrends et al., 2009) is used. The model was calibrated to the maximum LAI value given by
Olchev et al. (2009). After a windthrow event the damaged trees are ‘removed’ from the
stand; accordingly the stand density and leaf area index, decrease. The calculations with
other modules continued from the time point of damage with the new values of structural
characteristics. The changes in structure result in a modified stand’s water balance, which
might enhance or inhibit the next windthrow event, thus creating positive or negative
feedback.
3.3.2 Soil water

To simulate the water balance the 1D-SVAT model BROOK90 (version 4.4e) is used
(Federer, 1995; Federer et al., 2003). It is a detailed, process-oriented model which can be
used to investigate the potential effects of changes in leaf area index. For each soil horizon
the parameters of the water retention curve and the hydraulic conductivity Brooks & Corey
(1966) function were fitted through the van Genuchten parameter (van Genuchten, 1980)
given by Fröhlich (2009) for Otterbach site and by Hammel & Kennel (2001) for the forest
floor. The parameter of the water retention curve for Finland and Russia were deduced from
soil texture classes (Clapp & Hornberger, 1978). The architecture of root systems is mainly
influenced by the parent material, soil type, bulk density, chemical soil conditions, and
depth of ground water, tree species and age. However, in each particular situation the
information on rooting depth and the root distribution within the soil profile is a main
source of uncertainty. For estimation of the effective rooting depth the rules from
Czajkowski et al. (2009) are applied. The relative root density is modelled as a function of
soil depth (Ahrends et al., 2010b).
3.3.3 Wind load

The model of atmospheric boundary-layer SCADIS (SCAlar, DIStribution, Sogachev et al.,
2002) was used to calculate the wind load on trees. The model implements a two-equation
closure approach. The model constants were modified by Sogachev and Panferov (2006) to
enhance the applicability of approach for description of air flow over heterogeneous
landscape coved by forest vegetation. The modification was shown to perform well for
different vegetation canopies. In present study the model was driven by measured
meteorological parameters as well as by climate model data. Other input parameters are
forest structure: height, stand density, Leaf Area Density and aerodynamic drag coefficient
Cd. Model equations and details about numerical schemes and boundary conditions can be
found in Sogachev et al. (2002). It should be noted however, that the model is capable to
describes separately the mean and gust loads on trees which allows more exact evaluation of
possible wind damage cause and of high risk areas within the stand (Panferov and
Sogachev, 2008).
The estimation of critical loads leading to overturning or breakage of trees is described in
details in Panferov et al. (2009, 2010). Here we would like to show the modification of
critical loads caused by changes soil moisture, LAI and rooting depth, which in turn were
triggered by changes in forest structure.
It was assumed that the critical values of wind load derived from tree pulling experiments
were obtained under some “normal” conditions characterising certain soil moisture and
temperature, rooting depth and applied on a standard leaf area typical for a given tree
species. The variations of these characteristics increase or decrease critical wind load
threshold.
Influence of rooting depth (RD) is taken into account as the linear factor describing the ratio
of actual tree anchorage (RDact) to “reference” value (RDref) for given tree species, age and
state fRD= RDact/(RDref). As an indicator of soil moisture provided by soil water module the
Relative Extractable Water (REW) (Granier et al., 2007) was adapted. The effect of soil
moisture on windthrow is expressed as a linear function of REW deviations relatively to its
reference value: REWref = 0.6 REWt,s, for the certain combination of tree species (subscript
“t”) and soil type (subscript “s”). The critical wind load (CWL) for actual soil moisture and
forest structure is, therefore, derived from its reference value: CWLact= CWLref/REWact* fRD.
The system distinguishes between wet, dry (REWref < 0.6 REWt,s) and frozen (Tsoil < 0°C)
conditions. Under dry and frozen conditions the probability of overturning is assumed zero
and only a damage caused by stem breakage is possible.
Thus, when the actual wind load calculated from climate and wind modules exceeds the
CWL (either for overturning or for stem breakage) modified by outcome of soil and LAI
modules, the next wind damage event occurs. The stand density and LAI are reduced,
which in turn leads to changes in stand microclimate, soil water regime and tree anchorage
as described above, and consequently to further modification of CWL. The new CWL can be
either higher or lower then the one before the damage event, so the stand vulnerability is
either increased or decreased. Simultaneously the risk of drought stress is changed as well.
4. Results
4.1 Reference period, Otterbach
4.1.1 Climate and water budget components
As mentioned above the system was run at first for the real spruce stand at research site
“Otterbach” under real climate measured during the period from 1976 to 2008.
Fig. 4 demonstrates the time courses of air temperature and maximal wind speed at the
research site. A slight trend towards higher temperatures can be observed in the 32-year
series. The warming trend generally results in a destabilization of forest stands in terms of
wind damage as the number of days with frozen soil decreases in winter - e.g. the anchorage
is reduced during the period of winter storms (Peltola et al., 1999). The wind curve shows
several years with very strong storm events: 1983, 1990, 1993, and 1994. The Fig. 5 shows the
course of precipitation sums and of the major water balance components calculated by the
soil water module of the system. Except for 1994 the storm years were not accompanied by
unusually high annual precipitation sums.
4.1.2 Windthrow
The calculations of wind damage with unchanged structure, i.e. without feedback clearly
shows that despite the importance of soil, relief and vegetation factors the driving force
defining the temporal pattern of damage is the wind (Fig. 6). The extensive windthrow
events coincide with the strong windstorm years listed above, with the largest event in 1993
– the year with the strongest storm. The effect of other factors, however, is clearly visible in
the non-linear relationship between storm strength and damage amount. For instance the
damage in 1990 is lower than in 1994 and 1995, although the storm was stronger. The
possible reason could be deduced from Fig. 5 where it is shown that 1990 was also drier than
both 1994 and 1995. The drier soil provided better anchorage for trees.
12 70
Tmean
Maximal Windspeed (m s-1)

10 Vmax
60
Air temperature (°C)
50
8
40
6
30
4
20
2 10
0 0
1976 1980 1984 1988 1992 1996 2000 2004 2008
Years
Fig. 4. Mean annual air temperature and maximal windspeed measured at the
meteorological station of German Weather Service near research station “Otterbach” for the
simulation period from 1976 to 2008.
1800
1600
]
-1
1400
water flux [mm year
1200
1000
800
600
400
200
0
1976
1978
1980
1982
1984
1986
1988
1990
1992
1994
1996
1998
2000
2002
2004
2006
open field precipitation throughfall transpiration soil water outflow
Fig. 5. The main water budget components for the simulation period from 1976 to 2007.
This effect could be observed even better when the feedbacks of the windthrow damage are
taken into account (Fig. 6, lower panel). The reduction of stand density and LAI resulting
from first strong storm in 1990 leads to considerable destabilization of stand and increase of
its vulnerability, so that the strongest storm in 1993 produced even higher damage. This, in
its turn results in more than 5% damage increase during following storms in 1994-1995.
Thus, the decreasing resilience of stands and decreasing of CWL produce a cascade of wind
damage events at the end of the period. Though each event in itself is not very extensive, the
constantly decreasing CWL sum up the small damages to the significant annual sums. The
constant small windthrows are also observed in reality at the Otterbach research site. This
cascade is not observed in modeling results when the feedbacks are not considered (Fig. 6,
upper panel).
50
45 Storm 24 January 1993
(Vmax > 50 m s-1)
daily windthrow risk [%]
40
35
30
25 Several hurricanes from Daria to
20 Vivian (Vmax > 40 m s-1)
15
10
5
0
1.1.1975 1.1.1980 1.1.1985 1.1.1990 1.1.1995 1.1.2000 1.1.2005
50 Storm 24 January 1993

45 (Vmax > 50 m s-1)
daily windthrow risk [%]
40
35
30
Several hurricanes from Daria to
25
Vivian (Vmax > 40 m s-1)
20 “Anatol” 3 December
15 1999 (Vmax > 30 m s-1)
10
5
0
1.1.1975 1.1.1980 1.1.1985 1.1.1990 1.1.1995 1.1.2000 1.1.2005
Fig. 6. Daily windthrow risk without (upper panel) and with feedback (lower panel) for the
simulation period.
4.1.3 Interaction of risk factors

To analyse the causes and consequences of these additional risks caused by damaged
forest structure, the strong storm and forest damage event of 25.01.1993 is considered here
in details (Fig. 7).
The upper panel shows the stand LAI reduced after storm from 3.7 to 2.4 m2 m-2. The
reduced LAI resulted in slightly decreased transpiration during following summer
comparing to the simulation without feedback (Fig. 7, middle panel). Another effect of
decreased LAI is the higher soil water content in “feedback simulation” which is the
combined result of lower transpiration and higher throughfall (lower interception due to
fewer trees).
5 25
daily windthrow risk

LAI [m 2 m -2] 4 20
3 15
[%]
2 10
1 5
0 0
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700
days after winthrow event
LAI without FB LAI with FB risk without FB risk with FB
7 25
transpiration [mm d -1]

6
20
5
4 15
[%]
3 10
2
5
1
0 0
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700
T without FB T with FB risk without FB risk with FB
1.2 25
1
relative soil water
20
0.8
content [-]
15
[%]
0.6
10
0.4
0.2 5
0 0
0 50 100 150 200 250 300 350 400 450 500 550 600 650 700
RW without FB RW with FB risk without FB risk with FB
Fig. 7. Effect of static and dynamic (with feedback) forest structure on water budget
components after strong wind storm at 25.01.1993.
However, towards the next winter (1994) the temperature limited transpiration and
increased precipitation result in highest soil water content both for feedback (reduced LAI)
and without feedback (normal LAI) calculations. Therefore, the 5% higher damage
produced by the next storm (1994) in calculation with feedback is caused only to very small
degree by reduced anchorage in soil, but mainly due to destabilisation of forest resulted
from reduced stand density. Another consequence of this destabilisation is the occurrence of
several small windthrows in the feedback calculations, which are definitely the product of
reduced CWL and therefore not observed in calculation without feedback. The latter one
registers only the damage of two main storms in 1994.
The strong events in 1994 caused further decrease of LAI in the feedback version which lead
to even stronger difference in transpiration in summer 1995 (Fig. 7, middle panel, days 450-
650) and, therefore, significantly higher soil moisture (Fig. 7, lower panel, days 450-650).
Combined with reduced stand density the total effect was such a strong reduction of CWL
that it led to the untypical windthrow event in late summer around the day 600 from initial
windthrow event in 1993. The damage risks from feedback are twice higher then without
feed under the presented soil and climate conditions.
4.2 Climate projection SRES A1B and damage risks in Finland and Russia
4.2.1 Projected climate
The experiment was limited to the two variations of boreal climate: the more continental one
in Russia and the maritime one in southern Finland. The Norway spruce and Scots pine
stands were chosen as typical for boreal zone. According to the same criteria a podzol was
taken as soil type for the experiment. The calculations were carried out for a period of 120
years (1981 – 2100) with daily time step.
The analysis of the projected climate data shows a considerable increase of both summer
and winter temperatures for boreal climate zone (Fig. 8). The precipitation in Finland
increases noticeably during the 21st century, while the increase is due to higher summer and
winter precipitation. In Russia the projected increase of precipitation is very weak, whereas
only the winter precipitation shows the consistent rise (Fig. 9).
Both in Finland and in Russia the maximal windspeed shows similar pattern during 21st
century: strong increase from present conditions to the period 2011-2040 and almost
constant values (or even slight decrease) towards 2100 (Fig. 10). For both sites the winter
winds are stronger then in summer, while in Finland the difference is more pronounced.
The period 2041-2070 has the highest number of strong storms (not shown here) and period
2071-2100 with second highest number.
4.2.2 Wind damage and feedbacks

All trends of climate variables projected by ECHAM5-MPIOM as SRES A1B scenario
indicate an increase of vulnerability of maritime and continental boreal forests. The
increased temperature leads to reduced number of days with frozen soil, decreasing the
anchorage of trees during autumn and spring storms. Higher precipitation result in higher
soil moisture also leading to reduced tree anchorage. Stronger winds and higher number of
storms on the other hand increase the wind load on forest. However, it is expected that pine
trees will be less vulnerable to the wind damage than spruce.
If simulations are performed with the given setup the system indeed projects higher
windthrow risks for boreal forests during 21st century. Figure 11 shows the windthrow risks
in 21st century for Finland and Russia aggregated over 30-year periods. The upper panel
demonstrates the nonlinearity of damage dependence on combination of influencing

climatic factors. The projected damage risks are generally higher for Finland due to higher
windspeed and precipitation. The differences in temporal patterns of Russian and Finnish
Finland Russia
10 10
Air temperature[°C]
8 Tmean Tmean
Air temperature [°C]

8
6 30-years-mean 6
30-years-mean
4 4
2 2
0 0
-2 -2
1980 1990 2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100 1980 1990 2000 2010 2020 2030 2040 2050 2060 2070 2080 2090 2100
Years Years
Finland Russia
14
Winter 14 Winter
Air temperature [°C]
10 Summer
10 Summer
6
Ta [°C] 6
2
2
-2 -2
-6 -6
-10 -10
1981-2010 2011-2040 2041-2070 2071-2100 1981-2010 2011-2040 2041-2070 2071-2100
Climatic Periods Climatic Periods
Fig. 8. Changes of mean annual air temperature (upper panels) and mean seasonal
temperatures for 30-years climatic periods (lower panel) in 21st century in Russia and
Finland according to SRES A1B (ECHAM5-MPIOM).
Finland Russia
1300 1300
1200 Precipitation
Precipitation [mm year-1]
1200
Precipitation [mm year -1]
Precipitation
1100 30-years-mean 1100 30-years-mean
1000 1000
900 900
800 800
700 700
600 600
500 500
400 400
1981 1991 2001 2011 2021 2031 2041 2051 2061 2071 2081 2091 1981 1991 2001 2011 2021 2031 2041 2051 2061 2071 2081 2091
Years Years
Finland Russia
1000 1000 1000 1000
Winter Summer Annual
800 Winter Summer Annual 800

Precipitation (mm 6month -1)
800 800
Precipitation (mm 6month )
Precipitation [mm year-1]

-1
Precipitation (mm year-1)
600 600 600 600
400 400 400 400
200 200 200 200
0 0 0 0
1981-2010 2011-2040 2041-2070 2071-2100 1981-2010 2011-2040 2041-2070 2071-2100
Climatic Periods Climatic Periods
Fig. 9. Changes of annual precipitation sum (upper panels) and mean seasonal precipitation
sums for 30-years climatic periods (lower panel) in 21st century in Russia and Finland
according to SRES A1B (ECHAM5-MPIOM).
Finland
6.00
Winter
5.00
Summer
Vmax [m s-1]
4.00
3.00
2.00
1.00
0.00
1981-2010 2011-2040 2041-2070 2071-2100
Climatic Periods
Russia
4.00
Winter
3.00 Summer
Vmax [m s-1]
2.00
1.00
0.00
1981-2010 2011-2040 2041-2070 2071-2100
Climatic Periods
Fig. 10. Changes of seasonal-averaged daily maximal windspeed (Vmax) for 30-years
climatic periods (lower panel) in 21st century in Russia and Finland according to SRES A1B.
sites are not easily explained. In Russia the maximum of windthrow falls on the period of
2011-2040 where the maximal increase of mean windspeed is observed. The maximal
increase of windspeed in Finland is also observed in 2011-2040, however the maximal
damage risk is projected for 2041-2070. The main reason is the higher number of extreme
storms events during 2041-2070 though the mean Vmax is almost as high as in 2011-2040.
Additionally the joint effect of increased precipitation and temperature which are
considerably higher than in 2011-2040 contributes to the difference. The following decrease
of wind damage toward 2100 in Finland is caused by reduction of windstorms comparing to
previous period. Although the storms number during the period 2071-2100 is similar to the
reference period the slightly higher mean Vmax and considerably higher temperature and
precipitation result in twice higher damage during 2071-2100 comparing to P0. The analysed
dynamics of wind damage demonstrates that to capture such joint effects of several climatic
parameters the system has to use coupled process-based approach.
Finland Russia
10 10
spruce spruce
8 8
windthrow [%]
windthrow [%]
pine pine
6 6
4 4
2 2
0 0
P0 P1 P2 P3 P0 P1 P2 P3
Finland Russia
8 8
spruce spruce
feedback [%]
feedback [%]
6 6
pine pine
4 4
2 2
0 0
P0 P1 P2 P3 P0 P1 P2 P3
Fig. 11. Changes of windthrow risks and feedback contribution to wind damage aggregated
for 30-years climatic periods (upper panel) in 21st century in Russia and Finland according
to SRES A1B. Periods P0: 1981-2010, P1: 2011-2040, P2: 2041-2070, P3: 2071-2100.
The lower panel of figure 10 shows the contribution of feedbacks to initial damage, i.e. the
additional damage produced in model when the vegetation damaged by initial windthrow
event is “removed” from the modelled stand as described in previous sections. Comparing
the upper and lower panels as well as Finnish site to the Russian one, one can see again the
non-linear dependences of damage risks on combinations of effecting factors. Generally the
magnitude of feedbacks is proportional to the magnitude of initial damage. Therefore the
feedback contribution in Finland is higher than in Russia. However, the complexity of
influencing agents might produce a complex response which would be higher or lower than
considering only one factor. For instance the damages in P0: 1981-2010, P1: 2011-2040
(Finland) are almost of the same magnitude – for pine even lower in 2011-2040 - however
due to the higher windspeed in 2011-2040 the same or lower initial damage produces more
extensive consequences. The same is true for Russian site when compare the periods P2:
2041-2070 and P3: 2071-2100. Generally the contribution of feedbacks to initial damage in
modelled stands is almost negligible comparing to real conditions presented in section 4.1.
The main reason is the coarse resolution of Global Circulation Model (ECHAM5-MPIOM)
which is not really suitable for evaluation of wind damage event at stand level. Due to wide
area averaging it underestimates considerably the small-scale variations of windspeed
caused by local surface heterogeneities and therefore – wind gustiness and Vmax. In present
study it was implemented for purely demonstration purposes. For any practical use of DSS
with climate projections, the downscaled scenarios should be implemented.
5. Conclusions
The additional damage caused by forest structural changes resulting from previous damage
or forest management activities might contribute considerably to the projected forest risks
and it is advisable to take it into account in a decision support process. It is emphasized that
the degree of contribution as well as a sign of feedback in each particular case will strongly
depend on the local or regional combination of climatic and soil conditions with tree species,
age and structure. The DSS-user should be aware that one kind of damage might strengthen
(windthrow -> windthrow) or weaken (windthrow -> drought) the contribution of other
damage factors. Although, the projected total amount of damaged forest might even remain
the same the losses would be redistributed between damage factors, which are the
extremely important information for the forest owner looking for a decision support.
Therefore, the adequate evaluation of projected risks in forestry requires the coupled
modelling capable to consider the combined effect of risk factors.
The DSS-user should be also aware that although the system is quite advanced it has certain
limitations. We would like to list the main limitations here. In present form is applicable in
Germany only as the empirical functions are derived there. The systems works with
monospecies stands only. The user should also take into account that the climate projections
are not forecasts and the skill of estimation of future risks is limited by uncertainties of
climate scenarios.
6. Acknowledgment
The development of DSS-WuK was supported by the German Ministry for Education and
Research (Program klimazwei) we gratefully acknowledge this support.
7. References
Ahrends, B.; Jansen, M. & Panferov, O. (2009). Effects of windthrow on drought in different
forest ecosystems under climate change conditions. Ber. Meteor. Insti. Univ. Freiburg,
Vol. 19, pp. 303-309. ISSN 1435-618X
Ahrends, B.; Meesenburg, H., Döring, C. & Jansen M. (2010a). A spatio-temporal modelling
approach for assessment of management effects in forest catchments. Status and
Perspectives of Hydrology in Small Basins, IAHS Publ. Vol. 336, pp 32-37. ISBN 978-
1-907161-08-7
Ahrends, B.;Penne, C. & Panferov, O. (2010b): Impact of target diameter harvesting on
spatial and temporal pattern of drought risk in forest ecosystems under climate
change conditions. The Open Geography Journal, Vol. 3, pp. 91-102. ISSN 1874-9232
Albert, M. & Schmidt, M. (2010). Climate-sensitive modelling of site-productivity
relationships for Norway spruce (Picea abies (L.) Karst.) and common beech (Fagus
sylvatica L.), Forest Ecology and Management, Vol. 259, No. 4, (February 2010), pp.
739-749, ISSN 0378-1127.
Alveteg, M.; Walse, C. & Warfvinge, P. (1998). Reconstructing historic atmospheric
deposition and nutrient uptake from present day values using MAKEDEP. Water,
Air, and Soil Pollution, Vol. 104, No. 3-4, (March 1998), pp. 269-283, ISSN 0049-6979.
Brooks, R. H. & Corey A. T. (1966). Properties of porous media affecting fluid flow. J.
Irrigation and Drainage Div., Proc. Am. Soc. Civil Eng. (IR2), Vol. 92, pp. 61-87. ISSN
0097-417X
Clapp, R. B. & Hornberger, G. M. (1978). Empirical equations for some soil hydraulic
properties. Water Resources Research, Vol. 14, No. 4. Pp. 601-603. ISSN 0043-1397
Czajkowski, T.; Ahrends B. & Bolte A. (2009). Critical limits of soil water availability (CL-
SWA) in forest trees - an approach based on plant water status. vTI agriculture and
forest research, Vol. 59, No. 2. pp. 87-93. ISSN 0458-6859
FAO (1990). Soil Map of the world. Vol 1: Legend. World soil resources report, Rome. pp 119.
ISSN 92-5-103022-7
FAO (2003). http://www.fao.org/ag/agl/agll/wrb/soilres.stm
Federer, C. A. (1995). BROOK90: A simulation model for evaporation, soil water and stream
flow, Version 3.1. Computer Freeware and Documentation. USDA Forest Service, PO
Box 640, Durham NH 03825, USA.
Federer, C. A.; Vörösmarty C. & Feketa, B. (2003). Sensitivity of annual evaporation to soil
and root properties in two models of contrasting complexity. J. Hydrometeorology,
Vol. 4, pp. 1276-1290. ISSN 1525-755X
Feisthauer, S. (2010). Modellierte Auswirkungen von Kleinkahlschlägen und Zielstärken-
nutzung auf den Wasserhaushalt zweier Fichtenreinbestände mit LWF-BROOK90.
Georg-August University, Master's Thesis, Göttingen. pp. 118, unpublished
Fröhlich, D. (2009). Raumzeitliche Dynamik der Parameter des Energie-, Wasser- und
Spurengashaushalts nach Kleinkahlschlag. PhD, Berichte der Fakultät für Forstwissen-
schaften und Waldökologie, Georg-August Universität, Göttingen, pp. 116
Gauger, T.; Haenel, H.-D., Rösemann, C., Nagel, H.-D. , Becker, R., Kraft, P., Schlutow, A.,
Schütze, G. , Weigelt-Kirchner, R. & Anshelm, F. (2008). Nationale Umsetzung
UNECE-Luftreinhaltekonvention (Wirkung). Abschlussbericht zum UFOPLAN-
Vorhaben FKZ 204 63 252. Im Auftrag des Umweltbundesamtes, gefördert vom
Bundesministerium f. Umwelt, Naturschutz und Reaktorsicherheit. Dessau-
Rosslau.
Granier, A.; Reichstein, M., N. Breda, I. A. Janssens, E. Falge, P. Ciais, T. Grunwald, M.
Aubinet, P. Berbigier, C. Bernhofer, N. Buchmann, O. Facini, G. Grassi, B. Heinesch,
H. Ilvesniemi, P. Keronen, A. Knohl, B. Koster, F. Lagergren, A. Lindroth, B.
Longdoz, D. Loustau, J. Mateus, L. Montagnani, C. Nys, E. Moors, D. Papale, M.
Peiffer, K. Pilegaard, G. Pita, J. Pumpanen, S. Rambal, C. Rebmann, A. Rodrigues,
G. Seufert, J. Tenhunen, T. Vesala & Wang O. (2007). Evidence for soil water control
on carbon and water dynamics in European forests during the extremely dry year:
2003. Agr. and Forest Meteorology, Vol. 143, No 1-2. pp. 123-145. ISSN 0168-1923
Hammel, K. & Kennel, M. (2001). Charakterisierung und Analyse der Wasserverfügbarkeit
und des Wasserhaushalts von Waldstandorten in Bayern mit dem
Simulationsmodell BROOK90. Forstliche Forschungsberichte München, 185. Heinrich
Frank. München. pp 148. ISBN 0174-1810
IPCC (2000) - Nakicenovic N. and Swart R. (Eds.) Special Report Emission Scenarios,
Cambridge University Press, UK. pp 570, ISBN: 92-9169-113-5
Jansen, M.; Döring, C., Ahrends, B., Bolte, A., Czajkowski, T., Panferov, O., Albert, M.,
Spellmann, H., Nagel, J., Lemme, H, Habermann M, Staupendahl K, Möhring B,
Böcher M, Storch S, Krott M, Nuske R., Thiele, J. C., Nieschulze, J., Saborowski, J. &
Beese, F. (2008). Anpassungsstrategien für eine nachhaltige Waldbewirtschaftung
unter sich wandelnden Klimabedingungen – Entwicklung eines
Entscheidungsunterstützungssystems "Wald und Klimawandel" (DSS-WuK).
Forstarchiv Vol. 79, No. 4, (July/August 2008), pp. 131-142, ISSN 0300-4112
Jarvis, A.; Reuter, H.I., Nelson, A., Guevara, E. (2008). Hole-filled SRTM for the globe
Version 4, available from the CGIAR-CSI SRTM 90m Database
http://srtm.csi.cgiar.org.
Olchev, A.; Radler, K., Panferov, O., Sogachev A., & Gravenhorst G. (2009). Application of a
three-dimensional model for assessing effects of small clear-cuttings on radiation
and soil temperature, Ecological Modelling, Vol. 220, pp. 3046-3056. ISSN 0304-3800
Panferov, O. & Sogachev, A. (2008). Influence of gap size on wind damage variables in a
forest, Agric. For. Meteorol., Vol. 148, pp 1869–1881, ISSN: 0168-1923
Panferov, O.; Doering, C., Rauch, E., Sogachev, A., & Ahrends B. (2009). Feedbacks of
windthrow for Norway spruce and Scots pine stands under changing climate.
Environ. Res. Lett., Vol. 4, doi:10.1088/1748-9326/4/4/045019. ISSN 1748-9326
Panferov, O.; Sogachev, A. & Ahrends, B. (2010). Changes of forest stands vulnerability to
future wind damage resulting from different management methods. The Open
Geography Journal, Vol. 3, pp. 80-90. ISSN 1874-9232
Peltola, H.; Kellomäki, S., Väisänen, H. (1999). Model computations on the impact of climatic
change on the windthrow risk of trees. Climate Change, Vol. 41, No. 1, pp. 17-36.
ISSN 1573-1480
Reynolds, K.M.; Twery, M.; Lexer, M.J.; Vacik, H.; Ray, D.; Shao, G. & Borges, J.G. (2008).
Decision Support Systems in Forest Management. In: Handbook on Decision Support
Systems 2, Frada Burstein, F. & Holsapple, C.W., pp. 499-533, Springer, ISBN 978-3-
540-48715-9, Berlin Heidelberg.
Roeckner, E.; Brokopf, R.; Esch, M.; Giorgetta, M.; Hagemann, S.; Kornblueh, L.; Manzini,
E.; Schlese, U. & Schulzweida, U. (2006). Sensitivity of simulated climate to
horizontal and vertical resolution in the ECHAM5 atmosphere model, Journal of
Climate, Vol. 19, No. 16, (August 2006), pp. 3771-3791, ISSN 0894-8755.
Richter, A.; Adler, G.H., Fahrak, M., Eckelmann, W. (2007). Erläuterungen zur
nutzungsdifferenzierten Bodenübersichtskarte der Bundesrepublik Deutschland im
Maßstab 1:1000000 (BÜK 1000 N, Version 2.3). Hannover. ISBN 978-3-00-022328-0
Sogachev, A.; Menzhulin, G., Heimann, M., Lloyd, J. (2002). A simple three dimensional
canopy - planetary boundary layer simulation model for scalar concentrations and
fluxes. Tellus B, Vol. 54, pp. 784–819. ISSN 0280-6509
Sogachev, A. and Panferov, O. (2006). Modification of two-equation models to account for
plant drag. Boundary Layer Meteorology. Vol. 121. pp 229-266. ISSN 0006-8314
Sogachev, A., Panferov, O., Ahrends, B., Doering, C., and Jørgensen, H. E., (2010). Influence
of abrupt forest structure changes on CO2 fluxes and flux footprints, Agricultural
and Forest Meteorology, doi:10.1016/j.agrformet.2010.10.010, ISSN: 0168-1923

Efficient Decision Support Systems Pract

Uploaded by

Copyright:

Available Formats

Efficient Decision Support Systems Pract

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Efficient Decision Support Systems Pract

Uploaded by

Copyright:

Available Formats

EFFICIENT DECISION

Copyright © 2011 InTech

Publishing Process Manager Petra Zobic

First published August, 2011

A free online edition of this book is available at www.intechopen.com

Efficient Decision Support Systems – Practice and Challenges in Multidisciplinary Domains,

Part 1 Applications in Business 1

Chapter 1 Application of Decision Support System

Chapter 2 Intelligent Agent Technology

Chapter 3 Modeling Stock Analysts Decision Making:

Chapter 4 Business Intelligence – CDMB – Implementing

Chapter 5 Decision Support Systems Application

Chapter 6 Intelligence Decision Support Systems in E-commerce 109

Chapter 7 Quick Response in

Chapter 8 Collaboration in Decision Making:

Chapter 9 A Lean Balanced Scorecard Using the Delphi Process:

Part 2 Applications in Water Resource Management 185

Chapter 10 Linking a Developed Decision Support System

Chapter 11 Estimating the Impact on Water Quality under Alternate

Chapter 12 Flood Progression Modelling and Impact Analysis 227

Chapter 13 Providing Efficient Decision Support for Green

Part 3 Applications in Agriculture 271

Chapter 14 Uncertainty Analysis Using Fuzzy Sets

Chapter 15 A Web-Based Decision Support System

Part 4 Applications in Spatial Management 319

Chapter 16 A Decision Support System (FMOTS)

Chapter 17 Sensor Network and GeoSimulation:

Part 5 Applications in Risk and Crisis Management 357

Chapter 18 Emerging Applications of Decision Support Systems (DSS)

Chapter 19 Risk Analysis and Crisis Scenario Evaluation

Part 6 Miscellaneous Case Studies 395

Chapter 20 Applications of Decision Support System

Chapter 21 Evaluating the Power Consumption

Chapter 22 A Supporting Decisions Platform for the Design

Chapter 23 Challenges in Climate-Driven

Book Volume 2 extends the applications of decision support systems (DSSs) to

Section 1, including Chapter 1 through 9, illustrates several applications of intelligent

and its adoption in banking industry. Chapter 2 focuses on the improvement of

occurrence of uncertainty factors. Chapter 18 illustrates the DSS for industrial

Application of Decision Support System in

1.1 Research motivations

1.2 Statement of problems

organizational problems (53%) or an inability to access the most relevant information

2.1 Introduction of Customer Relationship Management

2.2 Measurement of success with CRM

2.2.1 Process efficiency

2.2.2 Product quality

2.3 CRM technologies and success with CRM

2.3.1 Developing information technologies in the banking industry

2.3.2 Customization of CRM functions/modules

2.3.3 Selecting the CRM vendors

2.3.4 Conducting DSS (Decision Support System)

CT6 Non-availability of appropriate CRM(ERP) packages.

Table 1. A Summary of Variables Included in the Present Study

3.1 The research question and the hypothesis

3.2 Research design

3.2.1 Research population and samples

3.3 Statistical methods

Fig. 1. A Summary of Statistical Methods

Estimate S.E. C.R. P