ADDIS ABABA UNIVERSITY

SCHOOL OF GRADUATE STUDIES

ADDIS ABABA INSTITUTE OF TECHNOLOGY

SCHOOL OF CIVIL AND ENVIRONMENTAL

ENGINEERING
MSc. THESIS
ON
ESTIMATION OF LOW FLOW QUANTILES IN
UNGAUGED RIVER CATCHMENTS

(CASE STUDY OF DEDESSA RIVER BASIN)

FEBURARY 2015
 ADDIS ABABA UNIVERSITY
SCHOOL OF GRADUATE STUDIES
ADDIS ABABA INSTITUTE OF TECHNOLOGY
SCHOOL OF CIVIL AND ENVIRONMENTAL
ENGINEERING

MSc. THESIS

ON

ESTIMATION OF LOW FLOW QUANTILES IN

UNGAUGED RIVER CATCHMENTS
(CASE STUDY OF DEDESSA RIVER BASIN)

BY: TATEK WORKU

ADVISOR: Dr.Ing. DEREJE HAILU

FEBURARY 2015
 ADDIS ABABA UNIVERSITY
ADDIS ABABA INSTITUTE OF TECHNOLOGY
SCHOOL OF CIVIL AND ENVIRONMENTAL
ENGINEERING

ESTIMATION OF LOW FLOW QUANTILES IN

UNGAUGED RIVER CATCHMENTS
(CASE STUDY OF DEDESSA RIVER BASIN)

Thesis Submitted to the School of Graduate Studies in Partial

Fulfillment of the Requirements for the Degree of
Master of Science
In
Hydraulic Engineering
By
Tatek Worku Marga
APPROVED BY BOARD OF EXAMINERS
Dr. Ing. Dereje Hailu ----------------------- -------------------------

(Advisor) Signature Date

Dr. Geremew Sahilu ----------------------- -------------------------

(Internal) Signature Date

Dr. Agizew Nigussei ----------------------- -------------------------

(External) Signature Date

Dr. Isayas G/Yohannes ----------------------- -------------------------

(Chairman) Signature Date

 DECLARATION AND COPY RIGHT

Tatek Worku Marga, declare that this Thesis is my own original work and that it has not been
presented and will not be presented by me to any other university for similar or any other degree
award.

Signature__________________________

It may not be reproduced by any means, in full or in part, except short extracts in fair dealing, for
research or private study critical scholarly review or discourse with an acknowledgement,
without written permission of the directorate of postgraduate studies, on behalf of both the author
and university of Addis Ababa.

I
 DEDICATION
I dedicate this work to my family, my friends and especially to my parents who are the reasons
for who I am!

II
 ACKNOWLEDGEMENT

I would like to express my sincere gratitude to my advisor, Dr.Ing Dereje Hailu, for his
unreserved assistance, constructive and timely comments at all stages of my work and also for
supplying me relevant materials to carry out the research. I should strongly appreciate his
patienceful guidance in a lot of discussions we made on various problems I faced during the
course of the work.
I am very grateful to Addis Ababa Institute of Technology, Department of Civil &
Environmental Engineering for allowing me to take part in the Master Program and for my
employer Oromia Irrigation Development Authority (OIDA) for providing me to learn my MSc.
Program. I am also grateful to Ethiopian Road Authority (ERA) and Ministry of Water, Irrigation
and Energy for their financial support and their hospitality in all my works.
I also acknowledge, Hydrology Department of the Ministry of Water, Irrigation and Energy and
the National Meteorological Service Agency, who supplied me data free of charge.
Finally, I would like to express my warm feeling of appreciation and tank to my friends who
helped me in all stages especially to Kidist Endale, Roba Muhyedin, and all staff member of
Oromia Irrigation Development Authority (OIDA), who were spiritually with me, and gave me
the strength to finalize my duties successfully. Thanks for your encouragement and true
friendship! Your friendship meant a great deal to me.

III
 ABSTRACT
Estimates of low flow statistics are required for a variety of water resource applications. At
gauged river sites, the estimation of low flow statistics requires estimation of annual n-day
minimum stream flows, low flow estimates, as determined by probabilistic modeling of observed
data sequences, are commonly used to describe certain stream flow characteristics.
Unfortunately, reliable low-flow estimates can be difficult to come by, particularly for gauging
sites with short record lengths. The shortness of records leads to uncertainties not only in the
selection of a distribution for modeling purposes but also in the estimation of the parameters of a
chosen model. In flood frequency analysis, the common approach to mitigate some of these
problems is through the regionalization of frequency behavior. The same general approach is
applied here to the case of low-flow estimation, with the general intent of not only improving
low-flow estimates but also illustrating the gains that might be attained in doing so.
In this study low flow regionalization has been performed using three approaches: flow duration
curves, base flow index and low flow frequency analysis. Low flow frequency analysis of
Dedessa river basin have been carried out using 7-day annual minimum flow series extracted
from the average daily stream flows observed at 10 stream flow gauging sites across Dedessa
river basin. L- Moment ratio diagrams were used for identifying and grouping of stations in to
hydrological homogeneous regions and hence the underlying statistical distribution.
Accordingly, the basin was delineated in to two homogeneous regions. The homogeneity of the
delineated regions was tested using the Cv-based and discordance tests. The goodness of fit test
(Z-statistics) and the minimum standard error of estimates have been used for assessing the
suitability of the selected distribution. Consequently, the Wakebay lower boundary distribution is
found to be a suitable candidate for region one and the generalized extreme value distribution for
region two. The method of probability weighted moments was considered as the best parameter
estimation procedure as compared with the method of moments and the maximum likelihood
method.
Flow duration curves of 1, 7, 10, 30 days have been derived for 10 stations. The basins flow
duration curves are characterized in to two homogeneous classes based on the characteristics
shown on the right tail end of the basins flow duration curves. Base flow separation has been

IV
made by the standard method called the institute of hydrology method. Accordingly the base
flow index is calculated and the basin is grouped in to three classes based on their range of BFI.
Regression analyses were conducted on the delineated regions to predict the index low flows
(mean annual 7-day minimum flow) for ungauged catchments using catchment characteristics.
The multiple coefficients of determinations R2 was used as a measure of the ability of the
regression model to describe variations in the dependent variables. The value of R2 was found to
be 0.997 for region one and 0.84 for region two. In addition the standardized low flow frequency
curves which will be used for estimating low flow quantiles at ungauged catchments have been
established for the two delineated regions using the standardized flow data.

V
 ABBREVIATIONS
LFFC Low-flow Frequency Curve‟s
BFI Base-flow index
DEM digital elevation model
FDC Flow duration curve
LMRD L-moment ratio diagram
MAM7 Mean annual 7day minimum flow
MOWIE Ministry of water Irrigation and Energy
MAMF Mean annual minimum flow
OIDA Oromia Irrigation Development Authority
NMSA National Metrological Service Agency
PPCC Probability plot correlation coefficient
PET Potential Evapotranspiration
PWM Probability weighted moments
LFFA Regional low flow frequency analysis
B/t Between
RMSE Root mean square error
DRB Dedessa River basin
AAit Addis Ababa Institute of Technology
WMO World metrological organization

VI
Contents Pages

DECLARATION AND COPY RIGHT........................................................................................... I

DEDICATION ................................................................................................................................ II
ACKNOWLEDGEMENT ............................................................................................................ III
ABSTRACT .................................................................................................................................. IV
ABBREVIATIONS ...................................................................................................................... VI
LIST OF FIGURES ....................................................................................................................... X
LIST OF TABLES ........................................................................................................................ XI
1 INTRODUCTION .................................................................................................................. 1
1.1 Background ...................................................................................................................... 1
1.2 Statement of the problem ................................................................................................. 2
1.3 Objective of the research:- ............................................................................................... 3
1.4 Description of the study area ............................................................................................ 4
1.4.1 Location .................................................................................................................... 4
1.4.2 Climate ...................................................................................................................... 5
1.4.3 Soil ............................................................................................................................ 5
1.4.4 Land Cover................................................................................................................ 6
1.5 Methodology and procedure............................................................................................. 7
1.6 Structure of the thesis ....................................................................................................... 7
1.7 Limitations ....................................................................................................................... 8
2 LITERATURE REVIEW ....................................................................................................... 9
2.1 Introduction ...................................................................................................................... 9
2.2 Definition of low flow .................................................................................................... 10
2.3 Methods of low flow analysis ........................................................................................ 10
2.3.1 Flow duration curve ................................................................................................ 11
2.3.2 Low flow frequency analysis .................................................................................. 12
2.3.3 Definition and history of flow frequency analysis .................................................. 13
2.4 Selection of the most robust low flow estimation method ............................................. 16
2.4.1 Descriptive ability tests ........................................................................................... 16
2.5 Low flow statistics ......................................................................................................... 19
2.5.1 Moments of distributions and their sample estimates ............................................. 20
2.6 Low flow regionalization ............................................................................................... 23
2.6.1 Procedure for regionalization .................................................................................. 24

VII
 2.6.2 Criteria for Low Flow Regionalization ................................................................... 24
2.7 Base flow ........................................................................................................................ 26
2.7.1 Definition and Introduction..................................................................................... 26
2.7.2 Base flow Separation Methods ............................................................................... 26
2.7.3 Standard Method of Base Flow Separation Method ............................................... 27
2.8 Low flow estimation of un-gauged catchments ............................................................. 28
2.9 Previous Study................................................................................................................ 29
3 DATA ANALYSIS FOR DEDESSA RIVER BASIN ......................................................... 31
3.1 Source and availability of data ....................................................................................... 31
3.1.1 Data used ................................................................................................................. 31
3.2 Data Screening ............................................................................................................... 36
3.2.1 Annual minimum series .......................................................................................... 36
3.2.2 Test for independency and stationery ..................................................................... 36
4 LOW FLOW ANALYSIS .................................................................................................... 39
4.1 Flow Duration Curves .................................................................................................... 39
4.1.1 Characteristics of flow duration curves in the low flow regimes ........................... 39
4.2 Base Flow Separation and BFI ....................................................................................... 41
4.2.1 Base Flow Separation ............................................................................................. 41
4.2.2 Base Flow Index ..................................................................................................... 43
4.3 Conclusive remarks ........................................................................................................ 43
5 RESULT OF REGIONALIZATION .................................................................................... 44
5.1 Introduction .................................................................................................................... 44
5.2 Identification regions...................................................................................................... 44
5.3 Homogeneity test of region ............................................................................................ 45
5.3.1 CV and LCv homogeneity test ................................................................................ 45
5.3.2 Discordance Measure Test ...................................................................................... 46
5.4 Delineation of the region ................................................................................................ 47
6 SELECTION AND EVALUUATION OF BEST FIT STATSTICAL PARENT
DISTRIBUTION OF LOW FLOW .............................................................................................. 50
6.1 Statistical distributions ................................................................................................... 50
6.2 Selection of distribution by LMRD ................................................................................ 50
6.3 Selection of frequency distribution ................................................................................ 51
6.4 Result for selection of parameter estimation method ..................................................... 52
6.5 Derivation of regional frequency curve for the homogeneous region ........................ 54
7 PREDICTION OF LOW FLOW FOR UN-GAUGED CATCHMENTS ............................ 55
VIII
 7.1 Regression Model ........................................................................................................... 55
7.2 Result of regression model ............................................................................................. 56
8 SUMMARY, CONCLUSION AND RECOMMENDATION ............................................. 58
8.1 Summary and conclusion ............................................................................................... 58
8.2 Recommendation ............................................................................................................ 59
9 REFERENCE ........................................................................................................................ 61
10 APPENDICES ...................................................................................................................... 63
10.1 APPENDIX A............................................................................................................. 63
10.2 APPENDIX B ............................................................................................................. 65
10.3 APPENDIX C ............................................................................................................. 67
10.4 APPENDIX D............................................................................................................. 69
10.5 APPENDIX E ............................................................................................................. 72

IX
 LIST OF FIGURES
FIGURE 1:1 MAP OF ETHIOPIAN RIVER BASIN WITH THE STUDY AREA .....................................................4
FIGURE 1:2 DEDESSA MAIN RIVER SYSTEM & SUB BASIN .............................................................................6
FIGURE 3:1 HYDROLOGICAL GAUGING STATIONS WITHIN THE BASIN ................................................... 32
FIGURE 3:2 METROLOGICAL STATIONS WITHIN THE BASIN ....................................................................... 33
FIGURE 3:3 DEM DATA OF THE BASIN ............................................................................................................... 35
FIGURE 3:4 DATA COMPARISON BEFORE AND AFTER FILLING .................................................................. 36
FIGURE 4:1 FLOW DURATION CURVES OF THE WHOLE STATIONS ............................................................ 39
FIGURE 4:2 GROUP ONE FDC‟S ............................................................................................................................. 40
FIGURE 4:3 GROUP TWO FDC‟S ............................................................................................................................ 40
FIGURE 4:4 TYPICAL HYDROGRAPH AND BASE FLOW HYDROGRAPH ..................................................... 41
FIGURE 4:5 TYPICAL HYDROGRAPH .................................................................................................................. 42
FIGURE 4:6 SEPARATED BASE FLOW HYDROGRAPH ..................................................................................... 42
FIGURE 5:1 LCS VS. LCK MOMENT RATIO DIAGRAM FOR 10 STANDARDIZED STATION ..................... 45
FIGURE 6:1 REGIONAL AVERAGE OF LMRD FOR THE REGIONS OF DRB .................................................. 50
FIGURE 6:2 DEVELOPED LOW FLOW FREQUENCY CURVES OF THE DIFFERENT REGION ................... 54
FIGURE 7:1 COMPARISON OF ACTUAL & ESTIMATED MEAN ANNUAL LOW FLOW FOR REGION ONE
........................................................................................................................................................................... 57
FIGURE 7:2 COMPARISON OF ACTUAL & ESTIMATED MEAN ANNUAL LOW FLOW FOR REGION TWO
........................................................................................................................................................................... 57

X
 LIST OF TABLES

TABLE 3.1 HYDROLOGICAL GAUGING STATIONS WITHIN THE BASIN ................................................... 31

TABLE 3.2 RANGE OF MISSING DATA IN % FROM THE RECORD LENGTH ................................................ 32
TABLE 3.3 RESULT OF DATA INDEPENDENCE & STATIONARITY USING W-W TEST FOR DEDESSA
RIVER BASIN ................................................................................................................................................... 37
TABLE 3.4 RESULT OF DATA RANDOMNESS USING LAG ONE SERIAL CORRECTION COEFFICIENT
TEST FOR DIDESSA RIVER BASIN .............................................................................................................. 38
TABLE 4.1 STATIONS GROUPED BASED ON THE RANGES OF THEIR BFI .................................................. 43
TABLE 5.1 RESULTS OF CV BASED HOMOGENEITY TEST FOR DEDESSA RIVER BASIN ...................... 45
TABLE 5.2 RESULT OF DISCORDANT MEASURE TEST FOR THE DEDESSA RIVER BASIN ..................... 46
TABLE 6.1SELECTED CANDIDATE DISTRIBUTIONS FOR THE REGIONS IN DARB .................................. 51
TABLE 6.2 CANDIDATE DISTRIBUTION WITH ADOPTED METHODS OF DISTRIBUTION ...................... 52
TABLE 6.3 GOODNESS OF FIT MEASURE FOR THE CANDIDATE DISTRIBUTION FOR EACH REGION 52
TABLE 6.4 STANDARD ERROR OF ESTIMATE FOR DIFFERENT PROCEDURES AND RETURN PERIOD
........................................................................................................................................................................... 53
TABLE 6.5 SELECTED DISTRIBUTION AND THEIR METHOD OF PARAMETER ESTIMATION FOR THE
DIFFERENT REGIONS .................................................................................................................................... 53
TABLE 6.6 RECOMMENDED METHOD AND PROCEDURE METHOD FOR THE DIFFERENT REGION .... 54
TABLE 7.1 DERIVED REGRESSION EQUATION FOR THE DIFFERENT REGION ......................................... 56
TABLE 10.1 TABLE HYDROLOGICAL GAGGING STATION USED FOR THE STUDY AND THEIR
CONDITION ..................................................................................................................................................... 63
TABLE 10.2 SUMMARY OF SITE STATISTICS BY CV MOMENTS BY CV MOMENTS RESULT SHEET-1.
LFFA-AS COMPUTER PROGRAM ................................................................................................................ 65
TABLE 10.3 MOMENTS AND L_ MOMENT RATIOS FOR DIDESSA RIVER BASIN, ETHIOPIA; RESULT
SHEET-2 LFFA-AS COMPUTER PROGRAMTABLE B.2 L-MOMENTS AND L_ MOMENT RATIOS . 66
TABLE 10.4 MATHEMATICAL EXPRESSIONS OF DISTRIBUTIONS ............................................................... 67
TABLE 10.5 RESULT OF REGRESSION MODEL FOR ESTIMATION OF ANNUAL MINIMUM FLOW FROM
CATCHMENTS CHARACTERISTICS ........................................................................................................... 69
TABLE 10.6 ANNUAL TOTAL FLOW, BASE FLOW AND BASE FLOW INDEX ............................................. 72

XI
 1 INTRODUCTION

1.1 Background

Sustainable water resources planning and management requires data to enable quantification of
water quality and quantity. Information is required on the rates of transfers and storage of water
within a catchment. Lack of adequate hydrological data introduces uncertainty in both the design
and management of water resources systems. As a result of low conversion of rainfall to runoff a
precarious balance between available water resources and water demand. The rapid population
growth characteristic of the country, which is increasing water demand for domestic,
agricultural, and industrial purposes, is causing water scarcity. The magnitude of this scarcity
and its variation in both space and time are largely unknown because of lack of hydrological
data. Catchment degradation in its various forms continues without effective control measures
due partly to uncertainty regarding the adverse effects on water resources. This uncertainty again
arises from lack of adequate hydrological data that should enable quantification of effects of
specific land use practices on quality and quantity of water resources. In addition, floods and
droughts occur with frequencies and magnitudes because of lack of relevant hydrological data.
These cause annually major social, economic and environmental tragedies.

There are many possible definition of low flow, in this study low flow is the simple lowest mean
flow over durations ranging from one to D days in a year (FREND 1989).
Low flows of a stream vary randomly. Also, the durations of low flows are important. A
prolonged period of low flows is the drought. Thus, any analysis of low flows must consider both
the magnitude and duration. The duration can be taken into account by averaging the flows over
a period of time (1 day, 7 days, 10 days, 30 days, etc.) The most critical low flows on record
must be examined for meaningful regulatory decisions. Because the historical record is only a
sample from an ensemble of possible occurrences, the analysis of low flows must be based on
probability theory. Frequency analysis provides a relationship between events and their
probabilities of occurrences.

1
In general, this relationship is determined in the form of a theoretical probability distribution
(Loganathan et al., 1985).

The estimation of flow regimes at un-gauged sites may be achieved by transfer of statistics
derived from gauged catchments using regionalization procedures. The term regionalization in
hydrology refers to grouping catchments in to homogenous regions.
A homogenous region is one which has sites with similar low flow characteristics and therefore
has the same standardized frequency distribution form and parameters. Low flow frequency
analysis is used to estimate low flow quintile (Q) at the interest site or project location. The rarity
of the low flow event being estimated is expressed by its return period (T) (Mkhandi et al.,
1997).

Regional low flow frequency analysis is one of the practical means providing low flow
information at sites with little or no local data. In the regional approach, available low flow data
series from hydrological homogenous region are pooled in dimensionless form and a frequency
distribution is fitted to combined data. However, the main problem of in low flow frequency
analysis is the determination of probability distribution that can provide a curve that defines the
regional average relation between standardized flow magnitudes (QT) and return period (T)
(Raymond, 2001).

1.2 Statement of the problem

Due to increased food insecurity and recent food price hikes in the country, it is envisaged that
there would be huge demand for irrigated agriculture by the farmers themselves and as a matter
of policy priority by the government at all level. This will undoubtedly create huge demand on
the water resources particularly during the lean flow season. The extra demand on the river may
create undesirable environmental as well as upstream downstream conflict. Therefore, for
informed decision making and appropriate policy recommendations on the minimum
environmental and downstream requirements, estimation of low flow is required that enables
generation of flows at any required reach of the river system, low flow estimates are vital for
planning water supplies, water quality management issuing and renewing of waste disposal

2
permits, hydropower, and the impact prolonged drought on aquatic ecosystems. Now a day‟s
extreme low flow events are more diligently analyzed in the emerging field of Eco-Hydrology.
Estimation of low flow is also important for small scale irrigation projects that contribute
significantly to poverty alleviation by means of increased crop production and generation of rural
employment (Adane, 2003).So analysis of low flows would provide an accurate understanding of
the demand that may safely be placed on a stream flow.

There is any study haven‟t been done concerning to low flow characteristic of the dedessa river
basin so far. Also the volume and duration of dependable base flow is not quantified. Most of all
the availability and quality of information is not adequate so further development of any water
resource project within the basin are difficult and unreliable unless the low flow characteristics is
well known.

Therefore, this study will try to analyze and characterize low flow of the un-gauged river
catchments of Dedessa river basin and provide the necessary information about the low flow of
the basin.

1.3 Objective of the research:-

The main objective is to identify and assess the suitability of statistical methods and flow
duration curves analysis to estimate low flow characteristics of un-gauged catchments of dedessa
river basin.
The specific objectives are:-
 To assess the potential of using hydrological homogenous regions as the basis for
estimating low flow characteristics of un-gauged catchments.
 Develop daily, weekly, decadal, monthly flow duration curves.
 Analysis of low flow frequencies.
 Developing base flow index
 Developing regression model

3
1.4 Description of the study area

1.4.1 Location

The study area, Dedessa sub basin is situated in the south-west part of the Abbay Basin and it is
located between 360 02‟ and 360 46‟ east longitude, and between 70 43‟ 55‟‟ and 80 13‟ 53‟‟
North latitude. as shown in Fig 1.1 and the largest tributary of the Blue Nine River in terms of
volume of water contributing roughly a quarter of the flow as measured at the Sudan border.
Draining an area of nearly 34,000km2, the Dedessa river basin originates in the Mt. venin and
Mt. wache ranges, flowing in an easterly direction for about 75Kms, and then turning rather
sharply to the north until it reaches the Blue Nile River. The major tributaries the Dedessa River
are the Wama entering from the ester the Dabana from the west and the Anger from the east

Figure 1:1 Map of Ethiopian River basin with the study area

4
1.4.2 Climate

The climate of Dedessa river basin is dominated by two factors: its near-equatorial location and
an altitude ranging from 626 meters above sea level to more than 3000meters.
The annual rainfall in the basin ranges between 1535 to 2214 mm, average being 1835mm. the
area is characterized by a humid tropical climate of heavy rainfall in its upper highlands. Almost
70% of the total annual rainfall is received during kiremt. The maximum and minimum monthly
temperature varies between 22.40 C-30.90 C and 8.40 C- 14.40 C respectively.
The rainfall in the dedessa river basin is uni-modal type, and most of the rainfall is concentrated
in the June to September period with virtual drought from November through February, Annual
totals from less than 1500mm to more than 2000mm. the five wettest months and the dry season
four months cover 63% and 7% of the total annual rainfall respectively.

1.4.3 Soil

The 2007 soil survey/study and the 2012 soil survey have identified nine major soil units,
classified based on the World Reference Base for Soils Resources (ISSS- ISRIC-FAO, 1998).
These soil units are Vertisols, Cambisols, Luvisols and Nitisols, as identified by the 2007 survey
and Major portion of the soils in the catchment area of Dedessa river basin are vertisols. Occur
on flat and almost flat lands, and developed on basic rocks and alluvial deposits. The Vertisols
of have clay fraction throughout the profile and the clay content ranges from 24 to 75%. The
nature of clay of the Vertisols is extremely hard when dry and very sticky and very plastic when
wet. These soils crack widely to more than 15 cm width during the dry season. However, the
cracks close up again on rewetting and as a result of this expansion and contraction the soil
forms grooved shiny pedfaces, known as slickensides and wedge shaped structure at depth of
20-100 cm.

The seasonal characteristic changes of Vertisols causes the development of micro-reliefe, known
as gilgai, comprising low mounds and hollows. The Vertisols are difficult to till and require
greater skill in management for better production than other soils.

5
/
Figure 1:2 Dedessa Main River System & Sub Basin

1.4.4 Land Cover

The dominant land cover of the dedessa river basin is savanna woodland. The flatter areas along
the Dedessa River are cover by tall grass while the sloping lands and hilly areas mostly covered
by woodland vegetation. In addition, there are riparian forests covering narrow strips of land on
banks of Dedessa River and its tributaries. Furthermore, the recently established settlement areas
and cultivated lands comprise a significant proportion of the land cover.
Land Form
The dedessa river basin are divided in to thee physiographic regions. These are the dedessa plane
ranges between 1270-1400m, upland or chain of ridges ranges between 1400-2500m and plateau
greater than 2500m.
6
The catchment characterized by mountainous highly rugged and dissected topography with deep
slops. The lowest part of the catchment is characterized by valley floor with flat to gentle slopes.
The dedessa plane is found to the Dedessa River. The chains of ridges are rugged and dissected
by the tributaries of Dedessa Rivers. The plateau is found on the top of the basaltic ridges. The
dedessa plane is covered by alluvial soils. They are mainly silt to sandy reddish brown and black
cotton soils are products of the breakdown of granitoids and basalts. While the plateaus are being
made of tertiary basalts the uplands are made up of Precambrian granitoids and tertiary volcanic
rocks. Pediments are gently slope erosion surface.

1.5 Methodology and procedure

Generally the study comprises the following methods and procedures.

 Collection of relevant data.
 Hydrological date
 Meteorological data
 Topographic and digitized map of the basin.
 Checking of adequacy consistency and independency of the data.
 Computing of statistical parameters of selected stations within the basin.
 Regionalization of the basin into homogenous regions based on statistical value using Arc
View GIS.
 Selection of frequency distribution for the delineated regions.
 Derivation of frequency curves using the standardized flow data for all regions.
 Regional homogeneity test describing the delineated region.
 Developing Regression model for estimation of annual minimum flow for each regions.

1.6 Structure of the thesis

This thesis is composed of eight chapters. Chapter 1 gives a brief general back ground and
introduction on low flows and a detailed description of the study area It also outlined the
methodology and procedures used in this study.

7
Chapter 2 describes methods of low flow analysis, historical background of low flow frequency
analysis, flow duration curve techniques, parameter estimation methods and criteria to choose a
frequency distribution. The chapter goes on describing low flow regionalization, low flow
estimation of un-gauged catchments, base flow definition and method of base flow separation.
Chapter 3 deals with source and availability of data, screening of low flow data and computation
of different statistical parameters.
Chapter 4 presents the results of FDC‟s and base flow groupings.
Chapter 5 gives result of regionalization that is identification of homogeneous region and test of
homogeneity of region, delineation of the basin into homogeneous low flow regions is described
,the methods used ,description of the delineated regions is also given.
Chapter 6 outlines the identification of the underlying statistical distribution of low flows,
selection and evaluation of best fit statistical parent distribution of low flow, the use of L-
moment diagram approach is explained and derivation of regional low flow frequency curves.
Chapter 7 deals with regional regression models of low flow and the result is presented in this
chapter.
Chapter 8 gives a summary, conclusions and recommendations of the work.

1.7 Limitations

World Meteorological Organization recommended as minimum record length of data of 30

years‟ duration or more against which for estimation and prediction of low flow analysis. Owing
to data constraint, this study considered an average of 10 years and above as a baseline period.
The limitation of short record data is that the data are unlikely to capture the natural variability of
the catchment, also insufficient number and distribution of gauging station in the Dedessa river
basin difficult to regionalize the catchment in to additional homogeneous region/group.

8
 2 LITERATURE REVIEW

2.1 Introduction

There are three primary scenarios when low flow information is required. The first is when a
water resources scheme is being developed. This normally requires a pre-feasibility study to
determine whether the objectives of the proposal can be achieved. For scheme with huge capital
investment, this is followed by much more detail design work, including estimate of the
frequency of low flow. The second is during operational phase of a water resource scheme once
it is constructed, and including decision how to manage the scheme on the day by day basis. For
example it may be necessary to determine how much water can be diverted from river for
hydropower purpose without infringing legal abstraction conditions. These may be depends on
the time of the year or on the river discharge at the downstream point. In many countries, such
operational constraints are not formalized and the operator must be aware of, and sympathetic to,
the needs of many downstream water users. For example, water could be needed by households,
or agriculture, electricity production, navigation, industrial abstraction, tourism, the dilution of
industry or domestic effluent, maintaining ecosystem for food production, or maintaining natural
biodiversity of the river. The third scenario is when it is necessary to make operational decision
today based on estimates of future river flows which looks days, weeks, or sometimes months
ahead. These forecasting increasing the efficiency of water uses and are economic importance in
terms of reducing the operational cost of water resource scheme.
Historically, most applications of low flow information have been in design and operation of
schemes for specific water sector such as public water supply, irrigation, energy, navigation and
industry. It is now recognized that there is a need to provide a long-term baseline monitoring and
analysis low flow to support integrated river basin management. This provides a framework for
environmental agencies to make decisions regarding the catchment wide development of water
resource.

9
2.2 Definition of low flow

A discussion of low-flow hydrology and related issues should ideally begin with a definition of
what „low flow‟ really is. This term could mean different things to different interest groups. To
many it may be considered as the actual flows in a river occurring during the dry season of the
year, others may be concerned with the length of time and the conditions occurring between
flood events. Yet others may be interested in the effects of changes in the total flow regime of a
river on sustainable water yield and riparian ecology. The latter may perceive „low flows‟ not
only as discharges occurring during a dry season, but as a reduction in various aspects of the
overall flow regime.
International glossary of hydrology (WMO, 1974) defines low flow as „flow of water in a stream
during prolonged dry weather‟. This definition does not make a clear distinction between low
flows and droughts. A low flow is a seasonal phenomenon, and an integral component of a flow
regime of any river. Drought, on the other hand, is a natural event resulting from a less than
normal precipitation for an extended period of time. But here in this study low flow is the
simplest lowest mean flow over durations ranging from one day to D days in a year (FREND,
1989).

2.3 Methods of low flow analysis

Low-flow regime of a river can be analyzed in a variety of ways dependent on the type of data
initially available and the type of output information required. Consequently there exist a variety
of low-flow measures. The term „low-flow measure‟ used here, refers to the different methods
that have been developed for analyzing, often in graphic form, the low-flow regime of a river.
The most common means of quantifying low-flow information of streams is with, statistical
estimates of the magnitude and frequency of occurrence. Methods using regression analysis to
relate low-flow frequency characteristics and selected basin characteristics could provide
significant benefits for managers responsible for protecting surface-water quality and allocating
surface-water supplies. Low-flow frequency characteristics with different consecutive-day
averaging periods and recurrence intervals provide quantitative information that can be used in
the management of a variety of additional water-quality and supply problems

10
2.3.1 Flow duration curve

The cumulative frequency distribution of daily mean flow shows the percentage of time during
which specified discharge are equaled or exceeded during the period of record. The relationship
is normally referred to as the flow duration curve and although it does not convey any
information about the sequencing properties of flows, it is one of the most informative methods
of displaying the complete range of river discharges from low to flood flows. The curve is most
conveniently derived from daily discharge data by assigning daily flow values to class intervals
and counting the number of days within each class interval. The proportion of the total number
of days above the lower limit of any class interval is then calculated and plotted against the lower
limit of the interval. Sometimes in plotting the flow duration curve the discharge axis is
standardized by the mean flow. This facilitates comparison between catchments. Because it
reduces differences in the location of the flow duration curve which are caused by differences in
the mean annual runoff.
FDC may also be obtained by applying a D day moving average to the original hydrograph. This
enables the portion of D day periods when the average discharge over D days is greater than a
given value to be estimated. The low flow indices derived from the flow duration curve would be
Q 95(1), Q95 (10) Q70 (1), Q70 (10) etc….This is determined based on countries adoption and
the purpose of the indices.
A flow duration curve presents the relationship between the magnitude and frequency of daily,
weekly, monthly or any other interval of stream flow. It shows the percentage of time during
which specified discharges are equaled or exceeded during the period of record. Generally a
normal probability scale is used for the frequency axis and a logarithmic scale for the discharge
axis. Because of the assumed lognormal distribution of daily discharges, these transformations
assist in linear zing the plot. A one-day flow duration curve can be derived manually. The
method involves assigning daily flows for the complete period of record to class intervals and
counting the number of occurrences within each interval. The total number of occurrences above
the lower limit of each class interval is expressed as a percentage of the total number of days in
the record and this percentage is plotted against the lower limit of the interval to give the
percentage of time flow is equaled or exceeded. Flow duration curves of other durations can be
obtained by applying a D-day moving average to the original hydrograph

11
2.3.2 Low flow frequency analysis

Unlike the FDCs, which shows the proportion of time during which a flow is exceeded, a Low-
flow Frequency Curve‟s (LFFC) shows the proportion of years when a flow is exceeded or
equivalently the average interval in years („return period‟ or „recurrence interval‟) that the river
falls below a given discharge.
LFFC is normally constructed on the basis of a series of annual flow minima (daily or monthly
minimum discharges), which are extracted from the available original continuous flow series
(one value from every year of record). Similar to FDCs, LFFCs may be constructed using the
flow minima series of different averaging intervals. In the case of daily data, the minima of 1, 3,
7,10, 15, 30, 60, 90, 120, 150, 180 or 183, 273 and 284 may be analyzed FREND (1989).
The independence of extracted annual flow minima averaged over long intervals (i.e. over 7
days) must be ensured. The average of the annual series of minimum 7-day average flows is
known as Mean Annual 7-day Minimum flow (MAM7) Gustard (1992).
The 7-day period covered by MAM7 eliminates the day-to-day variations in the artificial
component of the river flow. Also, an analysis based on a time series of 7-day average flows is
less sensitive to measurement errors. At the same time, in the majority of cases there is no big
difference between 1-day and 7-day low
flows. For seasonal low-flow frequency analysis, minimum flow values are selected from the
season of interest (winter or summer low flows). FREND (1989) have examined Weibull,
Gumbel, Pearson Type III, and log- normal probability distributions for fitting the sequences of
annual minimum flows and methods for estimating distribution parameters in different regions
and for minima of different averaging intervals.
Low flow frequency analysis is one of the practical means of providing low flow information. In
the low flow frequency analysis, available low flow data series are fitted to a frequency
distribution. However, the main problem of in low frequency analysis is the determination of
probability distribution that can provide a curve that defines the regional average relation
between standardized flow magnitudes (QT) and return period (T) Raymond (2001).
The slope of LFFC may be considered as a low-flow index. It is represented by the difference
between two flow values (normalized by the catchment area), one from high and another from
low probability domains. The larger this slope is, the more variable is the flow regime of the
river and vice versa. The issue, which often arises in low-flow frequency analysis, is that the

12
observed stream flow time series (and, consequently, low-flow time series) often contain zero
flow values. In arid climates stream flow may naturally frequently fall to zero. Similar situation
arises in cold regions, where the streams may be completely frozen in winter. On the other hand,
the recorded zero flows are often those which are below the measuring limit of a stream flow
gauge (censored flows). Zero values may not be ignored in statistical analysis of low-flow series.
Distributions fitted to series with zero flows will result in a positive probability of negative
stream flows, unless the distribution is explicitly constrained to have a lower bound of zero. Such
results are physically meaningless. Also, constraining some distributions to have a zero lower-
bound may reduce their flexibility. Fitting a probability distribution for low-flow series
containing zero values may be performed using conditional probability adjustment procedure
suggested by Haan (1977). An extra parameter describes the probability of encountering a zero
flow in a sample, while a continuous distribution is established for non-zero flows. The results
are then adjusted to the full sample. Low-flow frequency indices are widely used in drought
studies, design of water supply systems, estimation of safe surface water withdrawals,
classification of Streams‟ potential for waste dilution (assimilative capacity), regulating waste
disposal to streams, maintenance of certain in-stream discharges, etc. Numerous specific indices
may be obtained from a LFFC. These indices are similar to characteristics derived from FDC,
but unlike FDC, time sequencing of discharges used in the analysis is not disturbed. The most
widely used indices are 7-day 10- year low flow (7Q10) and 7-day 2-year low flow (7Q2), which
are defined as the lowest average flows that occur for a consecutive 7-day period at the
recurrence intervals of 10 and 2 years, respectively.

2.3.3 Definition and history of flow frequency analysis

A flow frequency curve shows the portion of years when a flow is exceeded, or equivalently the
average interval in years that a river falls below a given discharge.
Application of techniques on flow frequency analysis goes back to 1914 when Fuller (1914)
derived empirical frequency curves from annual maximum flood peaks by graphical method.
From here further developments were made with analytical methods to get different types of
distributions such as the skewed Pearson Type III distribution, Foster (1924) and Log-Normal
distribution by Hazen (1924).In low flow frequency analysis, Gumbel (1941) introduced the
extreme value theory to hydrology. The statistical method was boosted further by the
13
development of an extreme value probability paper by Powell (1943), which he used to construct
flood frequency curves. In the development of low flow frequency curves, Matalas (1963)
investigated the fitting of four theoretical distributions to low flow data.
Riggs (1968) prepared a very useful review of frequency curves and later produced a manual
that describes in detail the procedures for defining and evaluating low flow frequency curves.

2.3.3.1Statistical distributions

Several statistical distributions are available for use to date but the problem arises in choosing
the right one to fit the observed annual minimum flow data. Some of the distributions used for
estimating low flow quantiles with a particular probability of exceedance or non exceedance are:
1. Normal distribution (N)
2. Log-Normal (LN2, LN3) (Hazen, 1914)
3. Exponential distribution (EXP)
4. Gamma distribution (G2)(Moran, 1957)
5. Pearson Type III (P3) (Foster, 1924)
6. Log Pearson Type 3 (LP3) (USWRC, 1967)
7. General Extreme Value (GEV) (Jenkinson, 1955)
8. Gumbel (EV1) (Gumbel, 1941)
9. Wake by (WAK4. WAK5) (Houghton, 1977)
10. Weibull (Wu & Goodridge, 1959)
11. Generalized Pareto distribution (Gpar)
12. Log - Logistics (LLG) (Ahmed et al, 1988)
2.3.2.1.2 Parameter estimation methods
Selection of a distribution is followed by estimation of parameters. Since the Parameters are to
be estimated from the sample data, the estimates are subject to sampling errors. The three most
popular parameter estimation methods used are:

14
 1. Method of Moments (MOM)

This method is one of the most commonly and simply used method for estimating the parameters
of a probability distribution. However, it is well known that some moment estimates are biased,
especially for the cases of short sample series or higher order moments (Wallis 1974).
It is confirmed theoretically (Kirby, 1974) that coefficient of variation (CV) and skewness (Cs)
in random samples are bounded and that the bounds are a function of sample size. Much research
work has been done in deriving factors or functions by which the bias can be corrected.
2. Maximum Likelihood (ML)

In this method the parameter estimates are determined by maximizing the sample log likelihood
function. The unknown parameters may be obtained by setting each of the partial derivatives
with respect to each parameter equal to zero and solving the resulting equations simultaneously.
These equations unfortunately do not often take a simple closed form and the numerical solutions
have to be used. Even though the maximum likelihood estimator is asymptotically unbiased,
sufficient and consistent, sometimes there is a failure to obtain proper solutions due to the
complexity of log likelihood functions, particularly when sample size is small or when the
distribution has more than two parameters. This method is statistically the most efficient one in
large samples.
3. Probability Weighted Moments (PWM)

Probability weighted moments (PWM) were introduced by Greenwood et al. (1979) and further
analyzed by Hosking (1986). PWMs were proposed for the purpose of derivation of expressions
for the parameter of distributions whose inverse forms x = x(F) can be explicitly defined, such as
Extreme Value type 1, General Extreme Value, Log Logistic and Wakeby distributions. The use
of PWMs was later extended to be used to estimate P3 parameters (Song and Ding, 1988) and
other distributions whose inverse forms cannot be explicitly defined (Hosking, 1986).
The PWM method has recently come to be regarded as the best method by research hydrologists
(Hosking, 1986). This procedure leads to relatively unbiased quantile estimates due to the fact
that the PWMs being linear functions of the data, suffer less from the effects of sampling
variability than do the conventional moments, because the latter involve higher powers of the
data.

15
A study by Gunasekara (1990) shows that regional estimates by PWMs yield quintile estimates
with low variability and low bias.

2.3.3.2 Quantile Estimation

After a distribution or a number of distributions are selected to fit the data, their parameters must
be estimated. The estimated parameters are used to calculate quantile estimates (QT) for different
return periods.
This is achieved by using the distribution function in which the parameter of the distribution are
replaced by their estimates and the relationship between return periods (T) and the probability of
exceedance (F) is expressed as

F=1/T …………………………………………………………………….2.1
Where, F=F (QT) is the probability of having a low flow of magnitude QT or larger.
Thus the problem reduces to evaluating QT for a given value F.Chow (1964) proposed a general
form for calculating QT as

QT =U1‟ +KT* µ 2 …………………………………………………….2.2

Where KT= frequency factor which is a function of return period & parameter of distribution.
U1‟, µ2 = moments of distribution which can be calculated by using the estimated parameters.

2.4 Selection of the most robust low flow estimation method

Descriptive ability tests were applied for evaluating the proposed statistical distribution
procedures and come up with the most robust low flow frequency model for each region in the
basin. In addition to descriptive ability test selection of the most efficient method which gives the
smallest standard error of estimate of different statistical distribution are used.

2.4.1 Descriptive ability tests

The indicators that are used in selecting one procedure from another from descriptive ability
point of view vary from the traditional chi – square and kolmogorov – simronov tests to the
recently introduced L- moment ratio diagrams.

16
The objective of all indicators is to assist the hydrologist in the choice of a distribution that best
describes a set of sample data.
Some recent techniques are introduced that can be grouped under descriptive ability tests. One of
them is goodness of fit, here; this method is discussed as a methodology in discriminating the
best statistical distribution.
Goodness of fit measure
For the already identified regions, goodness – of – fit measure (Z) helps to test whether a given
distribution fits the data acceptably or not. The common approach is to choose a distribution
from a number of distributions that fits the data best. This measure is based on L-moments of the
at-site data. If the observed sites do form a homogenous region
(a) L- moments of the sites are well summarized by regional average, indicating that the scatter
of the individual sites‟ moments about regional average represents no more than sampling
variability and
(b) It can be assumed that the region‟s population L – moments are likely to be close to the
average of the sample L – moments of the observed data.
In most cases the distribution being tested will have location and scale parameters which can be
chosen to match the regional average mean and L – CV. The goodness of fit will therefore be
judged by how well the L – skewness and L – kurtosis of the fitted distribution match the
regional average L – skew ness and L – kurtosis of the observed data.
For a chosen distribution, the goodness of fit measure Zdist is defined as
  4dist  t4   4 
Z dist
   ………………………………………..2.3
  4 

Where t 4 is the average L – kurtosis value computed from the data of a given region,  4

denotes the standard deviation of t 4 , which can be obtained by repeated simulation of a

homogenous region with the distribution under test and sites having low flow record lengths the
same as those of the observed data.
 4 is the bias of t 4 and  4dist is the average L – kurtosis value computed from simulation for a
fitted distribution.

17
For short record lengths (n  20) and large population L skewness T3  0.4 , a bias

correction factor for t4 is required. Hence in a goodness of fit test the fitted L – kurtosis  4dist is

compared not with the regional average t 4 itself but with the bias corrected version t 4 -  4 .
The steps involved in a goodness of fit test are:
Step 1. Consider a distribution.
Step 2. Fit each distribution to the group average L – moments
t , and t3 .Denote the L – kurtosis of the fitted distribution by  4dist
Where,
N


i 1
ni t i 
t  N
………………………………………………….2.4
n
i 1
i

N
i 
n
i  1
i tr
tr  N
r  3 , 4 , ....... …………………….....2.5
n
i  1
i

 L2 
t i   L – Moment ration at site i  t i   
 L1 

L2 and L1 are as defined in chapter 2 sections 2.5.1.2

N = number of sites in a region
ni = record length at site i.

Step 3. Fit the considered distribution to the group average L – moments.

Step 4. Simulate a large number Nsim1 (here 500) of regions from the fitted
distribution. For the mth simulated region calculate the regional
m 
average L – skewness t3 and L – kurtosis t4m  .

Step 5. Calculated the bias and standard deviation of t 4 using the following
expressions.

18
  t 
N sim 1
4  N 1
sim1 4
m
 t4 ………………………………………………….2.6
m  1

and
1
  N sim 1 4 m  2
 4  N sim1  11  
  t  t4 
2
 N sim1   ……………………………..2.7
2
4
  m 1 

Small values of Zdist imply that the considered distribution can be accepted as the true underlying
frequency distribution for the region. According, to Hosking and Wallis (1993) if Z dist  1.64

, the distribution is acceptable.

2.) Standard error of estimates
The standard error of estimate accounts for the error due to small samples, but not the error due
to the choice of inappropriate distribution. The standard error of estimate depends in general on
the method of parameter estimation. Consequently, each method gives a different standard error
of estimate; the most efficient method is that which gives the smallest standard error of estimate
(Rao.A.R, 2000).

2.5 Low flow statistics

Preference to L-moment statistics (Lcv, Lcs, Lck) over conventional moments is given because
of the fact that conventional moments exhibit substantial bias and variance for the small samples
encountered in hydrological applications. L- moment method is a powerful and efficient method
to compute any statistical parameters, Also it cannot be influenced with the presence of outliers
(Roa &Hammed, 2000).
 The statistical parameters computed includes
 Mean
 Standard deviation
 Coefficients of Skewness
 Coefficients of Variations and
 Coefficients of Kurtosis

19
2.5.1 Moments of distributions and their sample estimates

2.5.1.1Conventional moments
Moments about the origin or about the mean are used to characterize probability distribution
.Moments about the origin are the expected values of power of random variable .For a
distribution with a probability density function f(x), the rth moment about the origin is given by
(Roa., 2000)

 r =  x r f x  dx ,  1 =  = mean………………………….2.8


The central moment‟s  r are computed by Eq.2.8


r   x

   )r f (x) dx, 1  0 ……………………………………..................2.9

It can be easily proved that the relationship between  r and  r is given by Eqs 2.8 and 2.9
(Kendall and Stewart, 1967).
r
r
r      r -j   r - 1 )j ………………………………………………………2.10
j0  j
r
r
 r =    r  j  r - 1 )j ………………………………………………………2.11
j  0  j

Sample moments mr` and mr` on the other hand are calculated by using Eqs.1.7.and 1.8
n
1
mr` =
n

i 1
xir m 1 = x = sample mean ………………………. …………2.12

r
1 n
mr   xi  x  , m1  0 ……………………………………………………….2.13
ni 1

The relationship as in Eqs.2.8 and 2.9 hold for sample moments as well. These sample moments
are often biased and may be corrected (Cunnane, 1989). For example, some of the corrected
central moments are given below
n

 x  x  …………………………………………………………2.14
1
mˆ 2 
2

n 1
i
i 1

20
 n
n
x  x  ………………………………………………2.15
n  1 n  2 
mˆ 3 
3
i
i 1

n2 n
mˆ 4   x  x  ……………………………………2.16
4

n 1 n  2 n  3 i 1
i

However, in small samples the bias may be larger than can be properly corrected by using simple
expressions in n. The conventional moment ratios are defined as below.
Moments of distributions and their sample Estimates

Coefficient of variation Cv = 2 /  `i ………………………………………….2.17

1 2

3
Coefficients of skewness Cs = 3 /  2 2 ………………………………………2.18

Coefficient of kurtosis Ck =  4 /  2 2 ……………………………………..2.19

2.5.1.2L-moments

Hosking (1990) introduced the L-moments, which are linear functions PWMs. The L-moments
are more convenient than PWMs because they can be directly interpreted as measure of scale and
shape of probability distribution. In this respect they are analogous to conventional moments.
L-moments have the following advantage (Cunane,1989).
Compared to conventional moments, L-moment can characterize wide range of distribution
Sample estimates of L-moments are so robust that they are not affected by the presence of outlier
in the data set
They are less subjected to bias in estimation
L-moments yield more accurate estimates of the parameters of a fitted distribution.

As Hosking (1986) defined the L-moments of a real value random variable x in terms of
probability weighted moments (PWMs) as:
r 1
L r = (- 1) r 1 P
k 0
r 1, k Mk ………………………………………………….2.20

Where
r r  k 
   
Pr ,k   1
r k
    ………………………………………………2.21
k   k 
   

21
 L r = the r th L- moment
k 
k  
j

M 10k =   1   M1 j0 …………………………………………………………….2.22
j 0  j
 
Hosking (1986),gave the unbiased estimators of M10k and M 1j0 as
 N  i  N  1
1 N
    
Mˆ 10k 
N
    /   xi k  0 ,1, 2 ..... N 1 ………………….2.23
i 1
  k   k 
   

  i  1  N  1
1 N
    
Mˆ 1 j0 
N
    /   xi j  0 ,1, 2 , ... N  1, …....................2.24
i 1
  j   j 
   
where i = rank of observed flow data in ascending order.
The first few moments are:-
L1 = M100
L2 = M100 - 2* M101
L3 = M100 - 6* M101 + 6* M101
L4= M100 - 12* M101- 30* M102 - 20* M103
Like the conventional moments, L- moments can be used to specify and summarize probability
distributions. In particular L1, the first L- moment, is the mean of a statistical distribution and is
identical to the first conventional moment, and L2 is a linear measure of spread or dispersion
analogous to standard deviation. L-moment ratios, which are analogous to conventional moment
ratios, are defined by Hosking (1990), Roa & Hamed, 2000) as:

t = L2 ……………………………………………………………………………2.25
L1

tr = Lr 3 ………………………………………………………………………..2.26
L2 r 

Where L 1 = measure of location

r = measure of scale and dispersion (LCV)
r3 = measure of skewness (LC k)
r4 = measure of kurtosis (LC k )

22
2.6 Low flow regionalization

Regionalization refers to grouping catchments into homogenous regions. A homogenous region

is one which has sites/stations with similar low flow characteristics and therefore has the same
standardized frequency distribution form and parameters. Moreover such a region must be
geographically continuous.
Several studies have shown that, delineation of regions in the past has often relied on
physiographic, political or administrative boundaries (Getachew, 1996).The resulting regions
were assumed to be homogeneous in terms of hydrologic response.
This assumption actually is not true as it may have very different relief and stations within the
same geographic region, which have high correlation that will cause some bias in the
regionalization (Wiltshire, 1985: Cunnane, 1989, Roa & Hamed, 2000).The importance of
homogeneity has been demonstrated by Hosking et al (1985), Wiltshire (1986).
Regionalization can be done based on geographic proximity, physiographic and climatic
characteristics of the catchments. (Gebeyehu, 1989) Further advancement in the field of flood
frequency analysis has led to a better approach other than geographic proximity. Recently,
researchers made their division of homogeneity by analyzing the statistical characteristics of
flow data of different stations within the basin (Roa & Hamed, 2000).
Raymond (2001) tried to regionalize Tanzania In to homogeneous region and he grouped
Tanzania in to eight homogeneous region based up on the low flow statistics of the gauged
station.
In regionalization, a regional frequency curve is derived. Regional frequency curves show the
portion of years when a flow is exceeded, or equivalently the average interval in years that a
river falls below a given discharge for those region.
The curve is derived from annual minimum discharge of seven days extracted from daily flow
time series. In response to the practical requirements for estimating the flood discharge of
extreme events having return periods several times greater than the length of hydrological
records is used but for low flow usually return period in the order of fifty years or less is used. In
frequency analysis of low flow, two values are associated. The first value is the duration of the
low flow used in the generation of the sample and the second is the desired return period. For
example, if duration of seven days is used along with a ten-year return period, the calculated
discharge (Q), will be the 7-day, 10-year low flow.

23
No matter how the regionalization is done, checking the homogeneity of stations (which made a
homogeneous region) by some homogeneity test is a must.

2.6.1 Procedure for regionalization

The grouping of homogeneous regions is done with the aim of achieving regional low flow
frequency relationships with lower standard errors than those of combined area (Kite, 1977).
The step-by-step procedure is as follows
1. Statistical values of each station under analysis (Lcs, Lck) for the flow is computed
2. LMRD of each station is drawn
3. Moment ratio diagram of each station together with the LMRD of the distribution are plotted.
4. Station that will lie in the same region are identified from the graph
This is done on the basis that station from the same homogenous region will be distributed along
the same distribution of LMRD. (Dalrymple, 1990, URL)

2.6.2 Criteria for Low Flow Regionalization

2.6.2.1 Homogeneity
Regardless of the way regions are formed they have to satisfy the homogeneity criteria (Zrinji &
Burn, 1996). Each of the regions should have two basic properties: dissimilarity from other
regions and homogeneity of low flow characteristics within the region. Hosking et al (1985),
Wiltshire (1986) and Lettenmaier et al. (1987) have demonstrated the importance of regional
homogeneity. There are several tests available to examine regional homogeneity in terms of the
hydrologic response of the stations in a region.
Another way of testing is by the use of the likelihood ratio test of whether the frequency of an
event can be represented by a single regional frequency relationship (Acreman and Sinclair,
1986). Wiltshire (1986) suggested the use of the R-statistic, which is based on similarity of the
geometry.
Later, Vogel and Kroll (1989) proposed the test, which is based on the probability plot
correlation coefficient (PPCC). Hosking and Wallis (1991) suggested a homogeneity test, which
is based on L-moments.

24
Their test statistic is a standardized measure of the inter site variance of the Lcv. Tests used for
determining the degree of homogeneity of regions are given below.
 Cv and L-cv Homogeneity test
 Discordance measure test

2.6.2.1.1 CV based test

A simple test based on the variability in at site Cv values is used .The main statistic considered is
the coefficient of variation of the Cv of the individual sites within the region (CC). In a region of
th
m stations and nj years of records at the j station, having a standardized low flow discharges
(qii,i= 1,2,3,…. n j) the coefficient of variation is defined as.
 nj q ji  q j 2 
 
 j  1 n j  1 

cv j  …………………………………………………………………2.27
qj

The weighted regional coefficient of variation of all the Cvj (CC) is defined as
1 m
cv  cv   2
CC 
cv

j 1 m 1
………………………………………………………………..2.28

The procedure consists of setting up repeated trial grouping of sites until the best possible set is
obtained. Sites with approximately close cvj values and lying within the same geographical
proximity are first grouped together in one region and the internal homogeneity of the regions
expressed numerically in terms of the flow statistics CC determined.
Depending on the value of cv, sites were transferred from one region to the other until the right
balance was achieved. The criterion used to check for regional homogeneity was based on the
value of CC. For homogeneous regions a value of 0.3 or less was allowed.

2.6.2.1.2 Discordance Measure Test

The discordance measure is intended to identify those sites that are grossly discordant with the
group as a whole. The discordance measure D estimates how far a given site i form the center of
the group. (Roa & Hamed, 2000)
If Ui  i 
= t i  , t3 , t 4 i 
 r
is the vector containing the t, t3 and t4 values for site (i) ,then the

group average for NS sites within the region is given by

25
 NS
1
U 
NS

i 1
Ui ………………………………………………………………………..2.29

The Sample covariance matrix is given by

 U  U  U  U
NS
S  NS  1
1
i i
T
…………………………………………2.30
i 1

The discordance measure is defined by

Di 
1
U i  U
T
S 1 U i  U …………………………………...............2.31
3
A site (i) is declared to be unusual if Di is large. A suitable criterion to classify a station as
discordant is that is that Di should be greater than or equal to 3.

2.7 Base flow

2.7.1 Definition and Introduction

Base flow is generally considered to be that component of flow which originates from stored
sources within the catchments, principally groundwater aquifers, as distinct from that flow
component which is rapidly transmitted to streams following rainfall events.
Base flow component tend to increase in wet season due to the rise of the depth of saturated
subsurface flow. The base flow component of river flow is commonly expressed as a proportion
of the total river flow, termed the Base Flow Index (BFI). A catchment‟s with BFI approaching 1
is highly dominated by base flow, while a catchment‟s with BFI approaching 0 receives little
base flow contribution.

2.7.2 Base flow Separation Methods

In order to determine the contribution from overland flow in a watershed to the streams in the
watershed or the contribution of flow from the stored sources it is necessary to separate out the
base flow from stream gage data. The base-flow index is the total volume of base flow divided
by the total volume of runoff for a period (Wahl, 1995).
Hydrologists are often asked to separate runoff into the components that result either from direct
runoff or from ground-water discharge. However, doing so is necessarily an inexact science.

26
The processes that produce flow are not simple, and many factors influence the source, timing,
and magnitude of stream discharges.
Basically there are two methods which are available for separations of base flow from a
hydrograph .These are graphical method and deterministic method.
In graphical method the simple way to make a base flow separation is to draw a straight line
from the point of rise to an arbitrary point on the lower portion of the recession segment of the
hydrograph. This arbitrary point may be so chosen that the base flow separation line should not
be too long and, on the other hand, the base flow should not rise too high. This point may also be
chosen as the beginning point of the fitted part of a ground water recession curve being matched
to the recession segment. According to Linsley, Kohler, and Paulhus, the point may be
approximately taken as A0.2 days after the time of peak flow, where A is the drainage area in
square miles. Another simple procedure is to extend the approach segment forward to a point
directly below, or a little beyond, the peak, and then to connect this point to the arbitrary point on
the recession segment with a straight line. But doing so for the whole hydrograph is tiresome and
needs drawings of line to each hydrographs also it is not possible directly calculate the base flow,
hence for the calculation of base flow you have to adopt an area calculation methods.
To make the base-flow separation process less tedious and more consistent deterministic
procedure was proposed in 1980 by the British Institute of Hydrology called the standard method
or institute of hydrology method.
The method combines a local minimums analysis with a recession-slope test ,with this method
we can estimates the annual base-flow volume of rivers and streams and computes an annual
base-flow index (BFI, the ratio of base flow to total flow volume for a given year) for multiple
years of data at one or more gage sites. (Swan, 1983) .So this method is adopted in this study.

2.7.3 Standard Method of Base Flow Separation Method

The standard method called the Institute of Hydrology Method begins by partitioning the year
into N-day periods and determining the minimum flow within each period. These minimum
flows are the potential turning points on the base-flow hydrograph. (If the year is not evenly
divisible by N, any remaining days will be included in the last period of the year). The default
value for N is 5 days. When several days within one period are tied for the lowest flow, the

27
earliest day will be considered the minimum, except during the last N-day period of each year,
when the minimum will be considered to occur on the latest day.
To determine the turning points on the base-flow hydrograph, the collection of N-day minimum
flows is processed using a turning-point test.
Institute of Hydrology (Standard Method) turning-point test:
Given three adjacent N-day minimum flows, Q0, Q1, and Q2:
Q1 is a turning point IF:
Q1* f<= Q0 and Q1* f<= Q2
Where f is a turning-point test factor that must be greater than 0 and should be less than 1.

2.8 Low flow estimation of un-gauged catchments

Planning and designing hydraulic structures need adequate hydrological information of the
specific area or the region at large. This includes observed stream flows at many sites. But many
sites do not have adequate number of gauging station or that are recently established. There may
not be gauging stations in the catchments at all. In such cases, transfer of required information
from gauged sites to un-gauged sites becomes very essential.
Generated low flow in catchments, among other things, are dependent on the catchments
characteristics. If an interest site has no record of flow, then a low flow, such as the seven day
mean annual minimum flow, Q 7, can be estimated from a relation between Q7 and measurable
catchments characteristics, which has been calibrated from gauged catchments in the region.
The physical characteristics of catchments may be grouped under the following headings
(Cunnane, 1985):
 Size and shape
 Density and distribution system
 Overland and channel slope
 Base flow Index
 Soils and geology
 Rainfall and climate
The selection of catchments characteristics depends on judgment of the likely predictive success
of variables, based on cause and effect or input and output relationship. The catchments

28
physiographic characteristics must be simple enough so that they can be measured from maps
easily and their accuracy should not be highly dependent on the scale of the map (Sine. 2004).
A number of measurable indices can be used to represent these characteristics. The most
commonly used indices are Area, Rainfall, slope, and base flow index. The correlation between
each and combinations of them with the index low flow will be established and then each
correlation is evaluated using any of the available parameters.
In the case of multiple regressions, the degree of correlation of a dependent variable to many
independent variables is measured by any of the following parameters: the standard deviation of
residual, multiple correlation coefficients, and the coefficient of multiple determinations, the
partial correlation coefficients, and the beta coefficients (Yevjevich, 1972).
The first step in regression analysis is the choice of the form of the predictive model. Most
previous studies have used regressions of the form:
Q  aAb Bc C d …………………………………………………2.32

Where Q = mean annual low flow

A = basin area characteristics
B, C… = catchments characteristics
a, b , c , d …. = regression parameters to be estimated.
The model indicated above is not linear but to make the treatment easier logarithmic
transformation is applied to get a linear equation (Chow, 1964). Least square techniques are
applied to determine the parameters and then coefficient of determination is used to choose
significant parameters.

2.9 Previous Study

Little or no research has been done on the same topic of low flow analysis in the Dedessa river
basin; rather two authors (Gebeyehu.A.1989 and Sine.A.2004) have done Regional Flood
Frequency Analysis that can have a relation with this research paper in some degree.
(Gebeyehu.A.1989) has done Regional frequency analysis of the North –western highlands of
Ethiopian‟s basin. (Such as Abbay basin, part of Gibe-Omo and Awash basins together).

29
For regionalizing and delineation of the basin into homogeneous region (Gebeyehu.A.1989)
adopted geographical location and climatic similarity. But (Sine.A.2004) used statistics of the
observed flood data the one which is recommended in recent days.

The other research was done by Tegenu zerfu (M.sc. thesis) with a title of Low flow frequency
analysis for Abbay River basin. The main objective of the study was to delineate the Abbay river
basin into hydrological homogeneous regions which would form the basic units to form and
develop frequency curves for each region. Low flow frequency analysis was done for the Abbay
river basin using 7-day annual minimum flow series and six homogeneous regions were
delineated using a simple test based on the variability of at-site values of Cv. In this research
statistics derived from observed low flow data‟s from different stations are used for regionalizing
and delineation of the basin into homogeneous region.

In addition to the above papers Abbay River Basin Integrated Development Master Plan Project
was done by BCEOM covering major parameters i.e. hydrology, water resources, natural
resources, land resources, demography, socio-economy and infrastructure in the basin.

30
 3 DATA ANALYSIS FOR DEDESSA RIVER BASIN

3.1 Source and availability of data

To achieve the goal of the research various data have been collected from different agencies and
individuals, which include time series data, topographical data and digitize map of the study
area.

3.1.1 Data used

Flow data
The Dedessa river basin has eighteen (18) hydrological gauging stations. Out of which only the
12 stations data are available and collected from Minister of Water, Irrigation and Energy
(MoWIE). The stations are as summarized in table 3.1 and only Eleven (11) gauging stations
with a minimum of 10 years record length and above were used for the analysis and those
stations have below 10 years record length rejected. Details of the stations are attached in the
annex part (see annex A. Table 10.1)
The stations consist of daily flow series. They were used to produce different duration of low
flows, flow duration curves, base flow, base flow index and also used for low flow frequency
analysis for the whole region.
Table 3.1 Hydrological gauging stations within the basin
NO STATION BEGINNING LAST LENGTH
NUMBER NAME OF STATION YEAR OF YEAR OF OF
RECORD RECORD RECORD
1 114001 Dedessa @ Arjo 1980 2003 24
2 114002 Anger @ Nekemte 1994 2004 10
3 114003 Sifa @ Nekemte 1962 1967 6
4 114005 Dabana @ Abasina 1962 1984 23
5 114007 Anger @ Gute 1982 2004 23
6 114008 Yebu @ Yebu 1979 2005 28

31
7 114009 Urgessa @ Gembe 1980 2003 24
8 114010 Tato @ Gutie 1996 2009 14
9 114012 Indires @ Sire 1981 2006 26
10 114013 Dabana @ Bunobe 1984 2005 22
11 114019 Temssa @ Agaro 1989 2006 18
12 101006 Uke @ Uke 1980 2005 26

Figure 3:1 Hydrological Gauging Stations within the Basin

Table 3.2 Range of missing data in % from the record length

Range of missing data in % No of stations

0-2 6
2-4 3
4- 6 3
>6 1

32
 Rainfall data
Average annual and average monthly rainfalls of the basin were provided by the National
Metrological Service (NMS) stations. These data were used to calculate catchment‟s average
annual and monthly time series of rainfall in turn the output is used for regression model.

Figure 3:2 Metrological Stations within the Basin

33
Figure 3:3 Isohytal Map of Basin

34
Topographic map
Topographic map of the basin with a scale of 1:250,000 is collected from Ethiopian Mapping
Agency to be used as basic input for regionalizing the basin in to homogeneous region and also
used to identify the correct name of the rivers within the basin.
Digitized map
Digital Elevation model DEM of the basin with a scale of 90x90m is collected from Ministry of
Water, Irrigation and Energy of GIS Department. This data is a basic input for developing digital
elevation model (DEM), flow accumulation, flow direction and also used in specifying the exact
location of gauging station .All of which are necessary for regionalizing the basin in to
homogeneous region.

Figure 3:4 DEM Data of the Basin

35
3.2 Data Screening

3.2.1 Annual minimum series

The calendar year is used in defining an annual minimum value. Only stations without zero flow
in their series are analyzed. The7day annual minimum flow series are used in frequency analysis,
as it is usually necessary to nullify the effects of minor river regulations.
In abstracting annual minimum series from daily data, missing data was filled by using the
following procedures (Teganu Zerfu (2009) low flow analysis and regionalization of Blue Neli
River Basin, M.sc. thesis report Arba Minch University.)
(a) Missing periods of up to 15 days during the dry season were completed by interpolation.
(b) Missing data of up to 30 consecutive days in the dry & wet season were completed by
seasonal mean value.
(c) Any calendar year containing periods of missing data and did not fall within any of the above
categories was rejected for further analysis. A typical filled station graph is displayed in fig 3.1.

Figure 3:3 Data comparison before and after filling

3.2.2 Test for independency and stationery

One basic assumption in the use of statistical distribution in hydrological frequency studies is
that the sample data are random variables without any serial correlation. As many hydrological
sequences exhibit a departure from randomness. One of the means of measuring the extent to

36
which a minimum in one year is dependent on the value of the previous is the lag one serial
correlation coefficient denoted by r1. It describes the strength of the relationship between a
value in a series and that preceding it by one time interval. For a strict random series the value of
r1 must differ from zero only by sampling variation and for sequences showing strong
persistence it is close to one.
So the following test is used to detect the randomness and the serial correlation of the data.
1. W-W test (Wald-wolfowitz test)
2. Lag –one serial correlation coefficient test

Out of the analyzed 11 stations one station were found to be dependent using both W-W test and
serial correlation coefficient test. The rest of them were not found to be dependent using either of
the randomness tests. The summarized result and criteria for both tests are show in table 3.3 and
table 3.4 respectively.
Table 3.3 Result of data Independence & Stationarity using W-W test for Dedessa River
basin

==========================================
Station Name Statistic
mean Critical Test Statistic Remark
---------------------------------------------------------------------------------
Dedessa 1.916 1.960 Independent
Anger 1.597 1.960 Independent
Dabana 0.728 1.960 Independent
Uke 1.743 1.960 Independent
Anger 1.538 1.960 Independent
Yebu 1.653 1.960 Independent
Urgessa 1.764 1.960 Independent
Tato 1. 659 1.960 Independent
Indires 0.771 1.960 Independent
Dabana 0.709 1.960 Independent
Temssa 2.942 1.960 dependent

37
Table 3.4 Result of data randomness using lag one serial Correction coefficient test for Didessa
River basin

=========================================
Station L.1-Correlation Upper Lower
Name Coefficient limit limit Remark
----------------------------------------------------------------------------------------------
Dedessa 0.31148 0.35623 -0.4437 Random
Anger 0.34338 0.48800 -0.6880 Random
Dabana 0.02900 0.36980 -0.4537 Random
Uke 0.25385 0.34408 -0.4241 Random
Anger 0.27023 0.36281 -0.4537 Random
Yebu 0.28090 0.41538 -0.3385 Random
Urgessa 0.19519 0.35623 -0.4432 Random
Tato 0.26640 0.44540 -0.5992 Random
Indires 0.07898 0.34408 -0.4241 Random
Dabana 0.07603 0.36980 -0.4650 Random
Temssa 0.60686 0.36281 -0.4537 Correlated

In any time series data outliers may or may not exist. These outliers may come due to personal
error during recording and inadequacy of measuring devices or due to extreme condition of
natural phenomena. Unless the source of the outliers is clearly identified, it is difficult to remove
outliers completely from analysis. Outliers can be excluded from the estimation procedure only if
it is certain that annual minimum flows can be adequately modeled by a single distribution form
(Cunnane, 1989).
So outliers test is not done in this study, but to avoid the effect of outliers an efficient method of
parameter estimation like PWM was used. Even if outliers are retained in analysis, have only
small effect if an efficient method of parameter estimation ML or PWM is used (Cunnane,
1989).

38
 4 LOW FLOW ANALYSIS

4.1 Flow Duration Curves

4.1.1 Characteristics of flow duration curves in the low flow regimes

In this study flow duration curves of 1, 7, 10, 30 day have been derived for 10 stations. From the
flow duration curves starting from the 80th percentile one day discharge, Q80 (1) is used to
generalize catchments .None of the stations have Q80 (1) equal to zero. Based on the
characteristics shown in the right tailed end portion of flow duration curves the basin flow
duration curves are characterized in to two homogeneous classes. This is done by visual
inspection. The whole and grouped flow duration curves are shown below in fig 4.1, fig 4.2 and
fig 4.3

Figure 4:1 Flow Duration Curves of the Whole Stations

Four stations are grouped in group one FDC‟s, station 114005,114001, 114002 and 114007. The
stations do not lie in the same region that is categorized in section 5.3.1 and 5.3.2 up on their low
flow statistics. Two stations namely 114001 &114005 lie in region two.

39
 Figure 4:2 Group one FDC’s

In group two FDC‟s six stations are available, station 114010, 114009, 114008, 101006,114012
and 114013. Based on the regions formed on their low flow statistics station 101006 lie in
region One.

Figure 4:3 Group two FDC’s

Note: - for this thesis grouping of station using flow duration curves simple for establishment of
different durations and represents a measure of hydrological response which embraces the full
regime from flood flows to low flows for each station when, comparing with grouping stations
using statistical method (it show only the low flow regime) & due to this grouping of station
using FDC‟s done.

40
4.2 Base Flow Separation and BFI

4.2.1 Base Flow Separation

In order to determine the contribution from overland flow in a watershed to the streams in the
watershed or the contribution of flow from the stored sources it is necessary to separate out the
base flow from stream gage data. The standard method called the Institute of Hydrology Method
is used for separation of base flow from main hydrograph.
Adopting the prescribed method makes the base-flow separation process less tedious and more
consistent.
For all the ten stations base flow hydrograph is separated and they are attached in annex E.For
illustration typical separated base flow hydrograph is presented in Fig 4.6, 4.7 &
4.8.respectively.

Figure 4:4 Typical Hydrograph and Base flow hydrograph

41
 Figure 4:5 Typical Hydrograph

Figure 4:6 Separated Base Flow Hydrograph

42
4.2.2 Base Flow Index

The base-flow index is the total volume of base flow divided by the total volume of runoff for a
period (Wahl, 1995). The base flow component of river flow is commonly expressed as a
proportion of the total river flow, termed the Base Flow Index (BFI). A catchment‟s with BFI
approaching 1 is highly dominated by base flow, while a catchments with BFI approaching 0
receives little base flow contribution.
From the main hydrograph, base flow hydrograph and base flow index is calculated. The
obtained result that is BFI is used as one of the input parameter for the regression analysis.
Grouped of stations within a range of BFI is shown in table 4.1.
Table 4.1 Stations Grouped based on the ranges of their BFI
Range of Station Base flow
BFI Number index
114008 0.234
114009 0.262
114010 0.346
114012 0.247
114013 0.350
0-0.35 101006 0.127
114001 0.430
114007 0.514
0.4-0.65 114002 0.641
>0.7 114005 0.725

4.3 Conclusive remarks

Those stations grouped in group one FDC‟s have a BFI of 0.43-0.725 which means 43% to
72.5% of the flow contributes from delayed storage and the flow mostly dominated by base flow
.FDC‟s grouped in group two have a BFI of 0-0.35 which means the flow is mostly dominated
by surface runoff.

43
 5 RESULT OF REGIONALIZATION

5.1 Introduction

Before regionalization is done conventional moments, L-moments and L-moment ratios have to
be computed. The computed values of conventional moments, L-moments and L-moment ratios
are given on Appendix B, Tables B.1 and B.2.
For the computation of the above parameters using L-moments a Matlab routine program is used
.These parameters are, later on are important and basic parameters for regionalization, selection
of distribution and various tests. Detail computed statistical values are given on Appendix B,
Tables B.1 and B.2.
On the basis of regionalization procedures shown in section 5.3 identification of homogeneous
region and tests for their homogeneity is done on the following section.

5.2 Identification regions

The Lcs vs. Lck of standardized flow of the stations has been plotted with L-moment ratio
diagram (LMRD) of various distribution function see fig 5.1 and stations which lie in the same
distribution are preliminary grouped as homogeneous region. This is done on the basis that
station from the same homogenous region will be distributed along the same distribution of
LMRD (Dalrymple, 1990, URL).

44
 Figure 5:1 Lcs vs. Lck moment Ratio Diagram for 10 Standardized Station

5.3 Homogeneity test of region

5.3.1 CV and LCv homogeneity test

For the Cv and Lcv homogeneous tests a Matlab routine program used .As to the result obtained
from the program all stations grouped preliminary as homogeneous region satisfies the Cv and
Lcv homogeneity test criteria.
Table 5.1 Results of CV based homogeneity test for Dedessa river basin

Region name cc value Conclusion

Conv.cv. based L-moment Cv-based

method method
Region one 0.275 0.267 Homogeneous *
Region two 0.166 0.102 Homogeneous
*Note: - The region declared to be homogenous if CC < 0.30 according to homogeneity test
procedure stated in section 2.6.2.1

45
5.3.2 Discordance Measure Test

All of stations grouped preliminary as homogeneous region in each region satisfies the
discordance test criteria which are stated in literature review part and a suitable criterion to
classify a station as discordant is that Di should be greater than or equal to 3.

Table 5.2 Result of discordant measure test for the Dedessa river basin

REGION ONE

S.No Station Number Discordant Measure Remark

(Di) > or = 3

1 114002 0.370 Homogeneous

2 114007 0.442 Homogeneous

3 101006 0.614 Homogeneous

REGION TWO

S.No Station Number Discordant Measure Remark

(Di) > or 3

1 114001 0.160 Homogeneous

2 114005 0.573 Homogeneous

3 114013 0.355 Homogeneous

4 114008 0.061 Homogeneous

5 114009 0.100 Homogeneous

6 114010 0.100 Homogeneous

7 114012 0.219 Homogeneous

46
5.4 Delineation of the region

The tool used in delineation of homogenous regions where GIS (Arc-view 9.2) software. On the
digitized map of the basins, all stations under analysis were identified according to their
geographical location (latitude and longitude). For each stations the statistical values (LCs , LCk)
which were computed in section 5.2 where given. In addition, the preliminary identification of
the region using LMRD mentioned above where identified on separate field to geo-code with
different color. It is assumed that (LCS, LCK) values of one station varies linearly with (LCS,
LCK) values of the neighboring stations. So from GIS screen, the distance between one station
and its neighboring station was determined and (LCS, LCK) values where interpolated to fix the
boundary between two stations of different regions. The procedures followed in the delineation
of the boundary of the region were as follows:
I. To compute the (LCs, LCk) value of each station
II. To identify the location of stations along the distributions of LMRD
III. To identify the group based on step II
IV. To interpolate between LCs and LCk Values of two stations of different groups to fix two
boundaries, one from the LCs and the other from LCk values.
V. The boundary of the region is fixed between the midway of the two boundaries found in
step IV.

According to the result obtained above the basin is regionalized in to two homogeneous regions

47
Figure 5:2 Map of the Established Homogeneous Region of Dedessa River Basin

48
Name of the region General characteristics

Region One Area 7,901.5km2,

Elevation ranges from 845 to 3000m, Rain fall varies from
1400 to 1800mm, slope ranges 5 to 15% and in some areas
less than 5% land use of the area is dominantly cultivated
and moderately cultivated.
Region Two Area 19,610 km2,
Elevation ranges from 626 to 3041m, rain fall from 1200
to 1800mm, slope of the area varies from 5 to 10 %, land
use of the area wood land , and cultivated

Table 5.3 Summary of land features, catchment characteristics.

49
 6 SELECTION AND EVALUUATION OF BEST FIT STATSTICAL
PARENT DISTRIBUTION OF LOW FLOW

6.1 Statistical distributions

The distributions used in developing models for estimating low flow quintiles with a particular
probability of exceedance or non exceedance and the mathematical expression of the statistical
distribution is shown in appendix C.

6.2 Selection of distribution by LMRD

According to Hosking (1990) the L moment ratio diagrams are based on relationship between the
L moment ratios. A diagram based on Lcs verses Lck, can be used to identify appropriate
distributions. The best parent distribution is the one that the average value of the point (Lcs, Lck)
of all stations with in the region gets close to one of the drawn LMRD of the parent distribution.
The result obtained is as shown on the graph below.

Figure 6:1 Regional Average of LMRD for the Regions of DRB

50
Accordingly, the most possible underlying candidate distributions are summarized on the table
below.

Table 6.1Selected candidate distributions for the regions in DARB

Average regional Selected Candidate

Region Name L-moments (t,t3,t4) Distribution
Region one (0.243,0.319,0.094) Wakeby Lower boundary
Region two (0.267,0.190,0.162) GEV /lognormal
So in the following section test for acceptability of the candidate distribution is done.

6.3 Selection of frequency distribution

Several statistical distributions are available for use to date but the problem arises in choosing
the right one to fit the observed annual minimum flow data. The result of fitting a distribution
also depends on the method of parameter estimation. There is no distribution that is universally
accepted to fit low flows (Cunnane 1986) .The three parameter estimation methods used for
evaluation are
Method of moment (MOM)
Method of maximum likelihood (ML)
Method of probability weighted moment (PWM)
For the candidate distributions which are chosen in the previous section the following method of
estimation procedure were applied these are:
GEV (MOM, ML, PWM)
L Normal (MOM, ML, PWM)
Wake by (WAK4. WAK5) (PWM)

51
 Table 6.2 Candidate distribution with adopted methods of distribution

Candidate Method of estimation

distribution MOM ML PWM
GEV * * *
L Normal * * *
Wake by - - *

For the Wakeby distribution, MOM and ML method is not used because moment estimates of
parameters of wakeby distribution are not easily obtained.

6.4 Result for selection of parameter estimation method

1. Goodness of fit measure

The following results are obtained for goodness of fit

Table 6.3 Goodness of fit measure for the candidate distribution for each region

Region name Candidate Goodness of fit Remark

distribution measure (Zdist)
Region one Wakeby 0.518 Desirable
Region Two Log normal 0.182 Desirable
GEV 0.105 Desirable

Standard error of estimates

Here the standard errors of estimate of the candidate distribution for different return period were
computed to select the best statistical distribution and method of parameter estimation
procedures.
Approximate standard error of estimate as well as parameter variances and covariance may be
estimated by simulation on this particular case it is left out (Roa , 2000). The result is shown in
table 6.4.

52
 Table 6.4 standard error of estimate for different procedures and return period

Return Region One Region Two

period(T)
Wekabay GEV Lognormal
MOM ML PWM MOM ML PWM MOM ML PWM
2 - - 0.042 0.489 0.652 0.348 0.465 0.451 0.587
5 - - 0.122 0.598 0.749 0.357 0.591 0.529 0.597
10 - - 0.157 0.678 0.759 0.475 0.686 0.597 0.687
15 - - 0.346 0.779 0.812 0.478 0.691 0.673 0.759
25 - - 0.459 0.798 0.859 0.675 0.796 0.795 0.761
50 - - 0.589 0.878 0.910 0.759 0.854 0.852 0.819

Based up on the Goodness of fit and minimum standard error of estimates the Wakeby lower
boundary is the best distribution to model the annual low flow series for region one. Although
both log normal and GEV distribution satisfied the goodness of fit test criteria for region Two,
the GEV distribution fits the lower part of the data better than the log Normal distribution.
Hence, the GEV distribution is selected for modeling the annual low flow series of region Two
in Dedessa river basin for both regions PWM selected as best parameter estimated method.

Table 6.5 selected distribution and their method of parameter estimation for the different regions

Region name Distribution and method of parameter estimation

One Wakeby/PWM
Two GEV/PWM

53
6.5 Derivation of regional frequency curve for the
homogeneous region

Based on the result obtained in the above section the following distribution are found to be the
best procedures for describing the annual minimum low flow and predicting acceptable low flow
estimates for the delineated region.

Table 6.6 recommended method and procedure method for the different region

Region name Distribution and method of parameter estimation

One Wakeby/PWM
Two GEV/PWM

For developing the frequency curves of the two regions a computer program (excel) which could
calculate the standardized flow, non standardized flow and the parameter of the selected
distribution is used.

Figure 6:2 Developed Low Flow Frequency Curves of the Different Region

54
 7 PREDICTION OF LOW FLOW FOR UN-GAUGED CATCHMENTS

7.1 Regression Model

A number of physical characteristics exist for a catchments and a number of numerical measures
or indices have also been described in explaining and developing models that relate low flow and
catchments characteristics. As was mentioned in chapter 2, there are a number of measurable
physical characteristics of catchments that might have important relationship with the low flow.
The availability of data representing them is very important in developing a model that relates
index low flow and catchments characteristics is given by
Q  c Aa S s R r BFI t F f ………………………………………………….7.1

Where Q = index low flow, is the mean annual minimum flow.

A = the catchments area in square kilometers.
R = the mean annual rain fall in millimeters.
F = shape factor (L2/A)
BFI = base flow index
S = Slope
And c, a, r, s, t and f are the regression parameters.
To make the analysis simpler, equation (7.1) is transformed to a linear form using logarithmic
transformation
Log Q = log C + a log A + s log S +r log R + t log BFI +f Log F……. 7.2
Then, the estimation will be performed on the parameters, namely, a, r, s, t and log c using
multiple linear regression technique. Then the estimation was performed on the parameters.
An important consideration in multiple regression analysis is to check whether the data are
adequately described or not. In this study the multiple coefficient of determination, R2, is used as
a measure of the ability of the regression model to describe variations in the dependent variable.
The closer R2 is to one, the better the regression model fits the data.

55
7.2 Result of regression model

Table 7.1 Derived regression equation for the different region in Dedessa river basin for
prediction of mean annual low flow (MALF)
Table 7.1 Derived regression equation for the different region

Region name Regression Equation R2

0.735 5.699
One 0.997
Q = 6.98 A BFI

0.9772 0.9193 -0.5534

Two 0.840
Q = 0.00089 A BFI S

Where A=area in Km2

BFI =base flow index
S=slope in %
For all regions, the mean annual minimum flows are computed by the developed regression
equation and compared with the actual annual minimum flows in the form of graph for each
station of all regions. The graphs are shown in figure below.
The developed regressions equations have high coefficient of determination (almost 100%
correlated) for region one. This may be due to unreliability or the number of the collected data.
So, for realistic estimation of mean annual low flow, the input data has to be collected, processed
and predicted properly. Inaccuracy is suspected in the input data particularly the areas of the
catchment at the gauging stations. This needs collection of additional data and further analysis.

56
Figure 7:1 Comparison of Actual & Estimated Mean Annual low Flow for Region One

Figure 7:2 Comparison of Actual & Estimated Mean Annual low Flow for Region Two

57
 8 SUMMARY, CONCLUSION AND RECOMMENDATION

8.1 Summary and conclusion

This study demonstrates the index low flow method of regional low flow frequency analysis. The
main objective of the study was to delineate the Dedessa river basin into hydrological
homogeneous regions which would form the basic units to form and develop frequency curves
for each region.
The flow data which is collected for the analysis were checked for their consistency and missing
data are filled by interpolation and seasonal mean value. The flow data was also tested for its
independency and stationerity and those station which shows serial auto correlation or
dependency behavior were discarded for further analysis.
Low flow frequency analysis was done for the Dedessa river basin using 7-day annual minimum
flow series. Two homogeneous regions were delineated using a simple test based on the
variability of at-site values of Cv. The delineated regions were tested for homogeneity using the
Hosking and Wallis test. In assessing the suitability of a distribution selected for a particular
region, the goodness of fit measure statistic (z) and minimum standard error of estimate were
used. The Wakeby distribution provides a good fit to low flows in region one, while GEV
distribution fits well in region Two. The method of probability weighted moments is considered
as the best parameter estimation procedure compares with the method of moments and maximum
likelihood.
Regression analysis has been applied to develop regression models to predict mean annual
minimum flow from un-gauged catchments using catchments characteristics.
L- Moment ratio diagram provides a practical means to group stations in different region and
also identify the underlying distribution for a given region.
So L- moment ratio diagram was used to group stations to same region and to identify the
underlying statistical distribution.
Flow duration curves of different durations are established. The flow duration curve represents a
measure of hydrological response which embraces the full regime from flood flows to low flows
.The flow regime of DRB is thus described by studying flow duration characteristics and base

58
flow contribution. Accordingly two groups of FDC‟s are established based on their pattern
shown from the 80th percentile and the value greater than of it.
According to the results obtained from the three approaches used for low flow regionalization the
author of this paper recommend that analysis and characterization of homogeneous region have
to be done independently for each approach.
Regression analyses have been applied to develop regression models to predict mean annual low
flow from ungauged catchment using catchment characteristics. The results of the analysis show
that the area is the dominant factor, which is the common independent variable for all regions.
During the analysis, for region one the coefficient of determination was not reliable when all
variables of all station were correlated with the mean low flow at a time. In addition to this, the
rainfall data is not correlated with the mean low flow satisfactorily for all regions. This is
because the rainfall data, which was collected from NMA, is not representative of the whole
basin. Therefore the rainfall data couldn‟t appear in the regression equation. In this regard further
investigation is required by determining the appropriate areal rainfall distribution of the whole
basin.

8.2 Recommendation

 Homogeneity test that consider not only higher moments of low flow but also some
important basin characteristics should be investigated.
 Delineation of hydrological homogeneous regions based on statistical parameter of
gauged sites could be one of the possible alternative methods of regionalization.
 Different photographic and climatic parameters like elevation, soil type, geology and
hydrogeology indexes, etc of each region should be determined so that a better regression
model will be resulted and the estimates of the mean annual, minimum flow will be
accurately estimated.
 Low flow frequency analysis of stations which have zero flows have to be studied
further.
 It is advisable to extend this approach of LFFA for other parts of Ethiopian river basins to
establish the homogeneous regions of the country so that problems related of absence of

59
sufficient discharge data for water resources project planning and design could be
reduced.

60
 9 REFERENCE
Abebe A. (2003) “Ethiopian Journal of water Science and Technology”
Admas, J.C.,Brainerd,W.S.& Goldberg,C.H.(1992) Programmers Guide to Fortran
90.McGraw Hill, Singapore
A.Gustard et al (1989) Flow Regimes from Experimental and Network Data, volume 1
Hydrological Studies ,Institute of Hydrology ,Wallingford.
Baipai,A.C.,Calus,I.M.and Fairley,J.A.(1978).Statistical method for Engineers and Scientists
Brogan, L., Cunnane, C. (2006). Low flows and low flow distributions for Ireland. Irish
National Hydrology Seminar (2006) pp. 85-92.
Chow,V.T.,Maidment ,D.R.and Mays,L.W.(1988).Applied Hydrology. McGraw Hill Book
Company, USA
Cunnane, C. (1989) Statistical Distribution for Flood Frequency Analysis. World Metrological
Organization Operational Hydrology Report No.33.
French Engineering Consultants and ISL Consulting Engineers (1990) Abbay River
Integrated Development Master Plan Project, Main Report Volume 1.
Gebeyehu , A. (1989) Regional Flood frequency Analysis, PHD thesis report
G.L. Giese and M.A. Franklin,(1996) Water-Resources Investigations Report 96-4308
Magnitude and Frequency of Low Flows in the Suwannee River Water management District,
Florida
Gubareva, T., van Nooijen, R., Kolechkina, A., et al., Extreme Flood Estimation by Different
Probability Distribution Laws in Different Climatic Conditions, Geophysical Research Abstracts,
2008,
G. V. Loganathan et al (1985) Methods of Analyzing In stream Flows, Department of Civil
Engineering Virginia Polytechnic Institute and State University Bulletin 148.
Haddad, K., Rahman, A. (2012), Regional flood frequency analysis in eastern Australia:
Bayesian GLS regression-based methods within fixed region and ROI framework – Quantile
Regression vs. Parameter Regression Technique, J. of Hyd. 430 – 43 1, 142-161.
Hamed,K.H., and Rao,A.R.(2000). Flood Frequency Analysis.CRC press LLC,Florida

61
Hosking, J. R. M., Wallis, J. R. (1997), Regional frequency analysis: an approach based on L-
moments. Cambridge University Press, Cambridge, U.K.
Hosking,J.R.M.and Wallis,j.R.(1993). Some Statistics Useful in Regional Frequency Analysis
.Water Resource Research, Volume 29,No.2. New York
Kozhevnikova, I.A., Distribution Parameter Estimation for Small Samples by L_moments,
Melesew,G.D.(1996). Regional Flood Frequency for Namibia and Zimbwabe, M.Sc thesis
report. Dar e salaam.
Raymond J. (2001) Development of Low Flow Prediction Models for Southern Africa PHD,
thesis report, Dar e salaam.
Sine, A. (2004) Regional Flood Frequency analysis, M.Sc. thesis report, Arba Minch University
Swan, D. R. and R. Condie. (1983) "Computation of the Base Flow Index," Water Resources
Branch Inland Water Directorate, Environment Canada
S.H.Mkhandi,R.mngodo and Kachroo (1997) Analysis of Flow Regimes in Tanzania
USWRC,United State Water Resource Council, (1976) .guide lines for Determining Flood
Frequency ,Bulletin No. 17 of the hydrology Committee, Washington, D.C.
Wu, Y.C., 2005. Establishing acceptance regions for goodness-of-fittest by stochastic
simulation. Master Thesis. Department of Bio-environmental Systems Engineering,National
Taiwan University (in Chinese).

62
 10 APPENDICES

10.1 APPENDIX A

Table 10.1 Table Hydrological gagging station used for the study and their condition

mean
catch_ area in no of year of flow Yield
Station _NO km^2 data m^3/S (Lt/Km^2) LATITUDE LONGITUDE %_OF_MISSI Period of recoded
114001 9981.0 24 100.58 10.08 8.683 36.417 2.59 1980-2003
114002 202 11 65.10 322.28 8.794 36.597 2.11 1994-2004
114003*** 6 1962-1967
114005 2881.0 23 58.70 20.37 1.83 1962-1984
114007 3742.0 23 22.26 5.95 9.500 36.583 1.20 1982-2004
114008 47.0 28 0.18 3.94 7.800 36.700 1.15 1979-2005
114009 19.0 24 0.768 40.24 7.833 36.650 3.50 1980-2003
114010 42.0 14 0.750 17.86 3.40 1996-2009
114012 49.0 26 1.21 24.64 9.033 36.850 3.89 1981-2006
114013 47.0 22 11.353 241.56 8.400 36.283 3.97 1984-2005
114019** 47.5 14 9.317 36.517 1.59 1989-2006
101006 52.5 26 1.27 24.29 1.90 1980-2005

63
NOTE
*Stations not satisfy first condition data filling (stations having Zero flows)
** Stations failed to satisfy independency and stationerity
***Stations failed to satisfy minimum no of data

64
10.2 APPENDIX B

Table 10.2 Summary of site statistics by CV moments by CV moments Result sheet-1. LFFA-AS computer program

--------------------------------------------------------

Mean flow and Conventional Moment ratios for

Dedessa river basin,
=============================================
Station Area Mean flow
Name km^2 N m^3/s CV Cs Ck
-------------------------------------------------------------------------------------------------
114001 9981.0 24 3.118 0.310 1.052 1.299
114002 202.0 11 0.729 0.711 0.921 1.284
114005 2881.0 23 2.880 0.148 0.787 1.675
114007 3742.0 23 1.794 0.240 0.931 1.109
114008 47.0 28 0.019 0.100 0.236 1.122
114009 19.0 24 0.141 0.040 0.649 0.815
114010 42.0 14 0.101 0.030 0.144 0.645
114012 47.0 26 0.096 0.040 0.035 0.459
114013 47.5 22 0.160 0.210 0.399 0.987
101006 52.5 26 0.070 0.100 0.972 1.597

65
Table 10.3 Moments and L_ Moment ratios for Didessa river basin, Ethiopia; Result sheet-
2 LFFA-AS computer programTable B.2 L-moments and L_ Moment ratios

==================================================
Station Unbiased
Name N t t3 t4
-------------------------------------------------------------------------
114001 24 0.698 0.606 0.336
114002 11 0.651 0.400 -0.055
114005 23 0.307 0.198 0.178
114007 23 0.408 0.369 0.438
114008 28 0.326 0.348 0.317
114009 24 0.152 0.083 -0.086
114010 14 0.188 0.223 0.269
114012 26 0.255 0.038 0.013
114013 22 0.409 0.229 0.136
101006 26 0.624 0.482 0.134

66
10.3 APPENDIX C

Table 10.4 MATHEMATICAL EXPRESSIONS OF DISTRIBUTIONS

Distribution Distribution function Variant and parameter

Name F(x) ranges
Normal distribution 1 
1
( x )2
2 2
F(x) = e
(N)  2   x  
 and  are parameters
Two parameter
Lognormal F(x) = 1  1  log x  b 2  0< x
exp    
2 ax  2  a  
distribution (LN2)
Three parameter
Lognormal F(x) = 1  1  log( x  m)  b 2  m<x
exp   
2 a( x  m)  2  a  
distribution (LN3)
Exponential m<x
distribution (EXP) 1  x  m
F(x) = exp   
a  a  (i.e. P-III with b = 1)
Two parameter 0  x if a > 0
Gamma ( x ) b 1 x  0 if a < 0
a x
F(x) = exp(  )
distribution(Gam2) / a / (b) a (i.e. P-III with m = 0)

Pearson-III  x  m
b 1 m  x if a > 0
 
distribution (P-III)  a   x  m x  m if a < 0
F(x) = exp  
/ a / (b)  a 

Log Pearson-III  z c

b 1 cz
distribution (LP-III)   ec  x   a>0
 a   z  c
F(x) = exp     z  c
x / a / (b)  a  0  x  ec a<0
If x P-III and z = log x

67
Generalized  >0
Extreme Value 
  x  u  k 
1
 

distribution (GEV) F(x) = exp  1  k  u +  x   if k < 0
  k
   
 
  
  < x ≤ u+ if k >0
k

  x u  
Extreme value F(x) = exp  e    
I     x  
Extreme value  >0
Type1 distribution
(EV1)
Five parameters x = m+  [1-(1-F)b]-c[1-(1-F)-d Note
Wake by It is analytically defined
distribution where F = F(x) only in inverse form
(WAK 5)
Four parameters
 
Wake by
 
 
distribution(WAK 4) x =  1  (1  F )   1  (1  F )  


Generalized
Pareto distribution

x =   1  (1  F ) k
k

(GPar) 1
 k  k
F = F(x) = 1  1  ( x   )
  
Log-Logistic 1 x > a, c >0, b >0

  x  a  c 
 1

distribution (LLg) F(x) = 1    
 b  
 
 
Generalized    
1

 <0,  +
1
  x      ≤ x <
logistic F(x) = 1  1      ,   0 
distribution (GLg)    
 
   >0,-  < x <  +


1
  x   
= 1  exp   ,  0 -< x < 
    

68
10.4 APPENDIX D

Table 10.5 Result of Regression model for estimation of annual minimum flow from
catchments characteristics
REGION ONE

MULTIPLE REGRESSION OF Z2 ON 2 INDEPENDENT VARIABLES

CHECKLIST OF THE DATA USED HERE: Where Z1 =area in Km^2

Z1 Z2 Z3 Z2 = flow in m^3/s
1 3742.00 1.020 0.247 Z3 =base flow index
2 52.00 0.001 0.127
3 202.00 0.167 0.262

THE COVARIANCE MATRIX

Z1 5.10
Z2 0.08 1.753
Z3 -0.20 0.29 0.08

THE CORRELATION MATRIX

Z1 1.000
Z2 0.040 1.000
Z3 0.889 0.690 1.000

THE FITTED REGRESSION EQUATION:

0.735 5.6998
Z2 = 6.98 Z1 Z3

THE MEASURES OF MULTIPLE CORRELATIVE ASSOCIATIONS:

THE MULTIPLE CORRELATION COEFFICIENTS: 0.951
CORRECTED FOR 1 DEGREES OF FREEDOM: 0.964
THE COEFFICIENT OF MULTIPLE DETERMINATIONS: 0.997
CORRECTED FOR 1 DEGREES OF FREEDOM: 0.893

THE STANDARDISED REGRESSION (BETA) COEFFICIENTS:

69
 B2 = 6.980
B1 = 0.735
B3 = 5.6998
REGION Two

MULTIPLE REGRESSION OF Z2 ON 3 INDEPENDENT VARIABLES

CHECKLIST OF THE DATA USED HERE: Where z1=area Km2

Z1 Z2 Z3 Z4 Z2= flow in m^3/s
1 9981 1.52 0.234 1.29 Z3=base flow index
2 2881 2.049 0.725 1.67 Z4= slope in %
3 47.0 0.001 0.350 6.09
4 19.0 0.014 0.346 6.60
5 42.0 0.024 0.43 1.07
6 49.0 0.006 0.514 2.0
7 47.0 0.012 0.641 2.46

THE COVARIANCE MATRIX

Z1 3.14
Z2 1.21 2.98
Z3 0.29 0.51 0.10
Z4 0.32 -0.35 0.02 0.25

THE CORRELATION MATRIX

Z1 1.000
Z2 0.424 1.000
Z3 0.639 0.817 1.000
Z4 0.261 -0.301 0.109 1.000

THE FITTED REGRESSION EQUATION:

0.9772 0.9193 -0.5534

Z2 = 0.00089 Z1 Z3 Z4

THE MEASURES OF MULTIPLE CORRELATIVE ASSOCIATIONS:

THE MULTIPLE CORRELATION COEFFICIENTS: 0.807
CORRECTED FOR 3 DEGREES OF FREEDOM: 0.859
THE COEFFICIENT OF MULTIPLE DETERMINATIONS: 0.844
CORRECTED FOR 3 DEGREES OF FREEDOM: 0.804

70
THE STANDARDISED REGRESSION (BETA) COEFFICIENTS:
B2 =0.00089
B1 = 0.9772
B3 =0.9193
B4 = -0.5534

71
10.5 APPENDIX E

Base flow hydrograph and BFI of the analyzed station

Table 10.6 Annual total flow, base flow and base flow index

Station No Total flow ,m^3/s per Base flow,m^3/s per Base flow index
year year
114001 36709.05 8589.92 0.234
114002 23756.97 6242.24 0.262
114005 21429.16 15536.14 0.725
114007 8123.71 2006.56 0.247
114008 65.11 23.00 0.350
114009 280.32 96.99 0.346
114010 275.27 118.36 0.430
114012 440.62 226.48 0.514
114013 4144.52 2656.64 0.641
101006 465.54 59.12 0.127

72
73
74
75
76
77