Perera PDF

Ungauged Catchment Hydrology:
The case of Lake Tana Basin
B.Upul Janaka Perera

March, 2009
Ungauged Catchment Hydrology: The case of Lake

Tana Basin
by
B.Upul Janaka Perera
Thesis submitted to the International Institute for Geo-information Science and Earth Observation in
partial fulfilment of the requirements for the degree of Master of Science in Geo-information Science
and Earth Observation, Specialisation: Integrated Watershed Modelling and Management
Thesis Assessment Board

Dr. Ir. M.W.Lubczynski
Dr. Ir. J.Schellekens
Dr. Ing.T.H.M.Rientjes
Dr.A.S.M.Gieske
(Chairman)
(External examiner)
(First supervisor)
(Second supervisor)
WREM Department, ITC, Enschede

Deltaris, Delft
INTERNATIONAL INSTITUTE FOR GEO-INFORMATION SCIENCE AND EARTH OBSERVATION

ENSCHEDE, THE NETHERLANDS
Disclaimer
This document describes work undertaken as part of a programme of study at the International
Institute for Geo-information Science and Earth Observation. All views and opinions expressed
therein remain the sole responsibility of the author, and do not necessarily represent those of
the institute.
I dedicated this thesis to my father,

who is no more with us to see the completion of it
though he anticipated this to turn out.
Abstract
Estimating the water balance of a hydrologic basin often requires an effective procedure to quantify
the contribution of the ungauged catchments. In this study a regionalizing based procedure is followed
to quantify the flows of the ungauged catchments in the Lake Tana basin. Linking model parameters
to physical catchment characteristics is a popular approach that enables the application of a
conceptual model to an ungauged site. The HBV semi-distributed conceptual rainfall-runoff model is
selected to simulate runoff of nine gauged catchments on a daily basis in the period 1994-2003. Some
eight model parameters are selected for model calibration that also are used in the regionalisation
procedure to simulate runoff from ungauged catchments. For nine gauged catchments the optimum
parameter sets are derived through an automatic optimisation procedure based on Monte Carlo
Simulation. By solving the multi-objective calibration problem that measures the different aspect of
the hydrograph: (1) overall water balance (relative volume error), (2) overall shape of the hydrograph
(Nash-Sutcliffe value). In the calibration procedure, for each catchment and for each single objective
function best and worst values are used for rescaling to allow comparison. Thus, model performance
is assessed through the use of rescaled objective functions that serve as criteria for selection of a best
parameter set. In regionalisation, optimal model parameters are related to selected physical catchment
characteristics (PCCs) that are used to estimate parameter values for ungauged catchments. Since
PCCs from ungauged catchments generally differ from those of gauged catchments, also model
parameter values will changed. By establishing the relations between values of HBV model
parameters and PCCs, information was transferred from the gauged catchments to ungauged
catchments. The transferred parameter sets were used to simulate the runoff from the ungauged
catchments of the Lake Tana basin.
Four parameter regionalization methods (multiple regression, spatial proximity, similarity approach
and sub-basin mean) were tested in this study to transfer model parameter values to the ungauged
catchments. To evaluate the reliability of the simulations from the ungauged catchments, a water
balance model of Lake Tana is developed at daily time step in the period 1994-2003. This model used
area-volume and elevation-depth relations based on a bathymetric survey to simulate lake level
fluctuations by calculation of the net inflow by estimating lake areal precipitation, lake areal
evapotranspiration, inflow from gauged catchment, inflow from ungauged catchment and lake
outflow. The water balance model of Lake Tana is evaluated by selected objective functions that are
the Relative Volume Error and Nash-Sutcliffe objective functions as well.
Among nine gauged catchments in Lake Tana basin six of them are satisfactory calibrated under the
criteria of relative volume error between 5% and Nash-Sutcliffe value greater than 0.6 for the period
of 1994-2000. Daily lake level simulation with inflow from ungauged catchment estimated from
regression method shows the best performance with a relative volume error of -2.17% and a NashSutcliffe coefficient of 0.92.
Key words: Regionalization, Water balance, HBV, Monte Carlo Simulation
Acknowledgements
First of all I thank the Load God for his endless mercy, grace and wisdom upon me during all these
days in ITC and in all my life.
I expressed my sincere gratitude to my supervisor Dr.Irr. Tom Rientjes for his valuable support,
guidance, encouragement and critical comments through out the research period. My sincere thanks
also to second supervisor Dr. Ambro Gieske for his encouragement and comments to improve my
research work and also former second supervisor Ing. Remco Dost for his assistance and cooperation
until the last day in ITC.
I acknowledge the debt I owed to the National Water Supply and Drainage Board, for providing me
this opportunity to continue my higher studies.
I wish also to thanks the Government of The Netherlands through the Netherlands Fellowship
Program (NFP), which made this study possible by its financial support.
My appreciation is incomplete, if I do not mention the support and help extended by Lal Muthuwatta
and Alemseged Tamiru Haile by means of constructive criticism, valuable suggestion and
encouragements.
I would like to acknowledge Dr. Ben Maathuis, Ir. Gabrial Parodi and Ir.Arno Van Lieshout for their
support during various part of the study.
I would also like to express my appreciation to all WREM department staffs and ITC community who
helped me directly or indirectly during my study.
I would like to extend my gratitude to the Sri Lankan colleague for their laudable support throughout
my stay in The Netherlands.
Above all, I wish to thanks my wife Treeni, my mother and mother-in law, for their moral support,
without which I will never be able to complete this study. Apologies to my two daughters, Nathari and
Nuveena. I wish you will forgive me when you realized the real reason behind parting you for such a
long period. Dearest thaththa I wish you also forgive me not to being with you at you last hours.
Upul Janaka Perera

[email protected]
iii
Table of contents
1.
Introduction....................................................................................................................................1
1.1. Background .............................................................................................................................1
1.2. Problem definition and importance of the study.....................................................................1
1.3. Scope of the research ..............................................................................................................2
1.4. Previous studies and applications ...........................................................................................3
1.5. Objectives ...............................................................................................................................4
1.6. Research questions..................................................................................................................4
1.7. Outline of the Thesis...............................................................................................................4
2.
Study area and data preparation..................................................................................................5

2.1. Study Area ..............................................................................................................................5
2.2. Climate....................................................................................................................................6
2.2.1.
Precipitation ..........................................................................................................6
2.2.2.
Temperature...........................................................................................................8
2.2.3.
Potential evapotranspiration................................................................................10
2.2.4.
Hydrological data ................................................................................................11
2.3. Delineation of gauged and ungauged catchments.................................................................13
3.
Hydrological modelling and regionalization .............................................................................15

3.1. Hydrological modelling ........................................................................................................15
3.1.1.
Hydrological process ...........................................................................................15
3.1.2.
Classification of hydrological model ..................................................................15
3.1.3.
Model selection ...................................................................................................16
3.2. HBV Model...........................................................................................................................16
3.3. Regionalization .....................................................................................................................17
3.3.1.
Approach of regionalization................................................................................18
3.3.2.
Selection of regionalization method....................................................................18
3.4. Model parameter selection....................................................................................................19
3.5. Selection of physical catchment characteristics ...................................................................19
4.
Methodology .................................................................................................................................25
4.1. Model Calibration.................................................................................................................25
4.2. Approach of calibration ........................................................................................................25
4.3. Objective functions...............................................................................................................26
4.4. Determination of optimum parameter set .............................................................................27
4.5. Establishing the regional model............................................................................................28
4.6. Regression analysis...............................................................................................................29
4.7. Test of significance and strength ..........................................................................................30
5.
Results and discussion .................................................................................................................31

5.1. Results of model calibration .................................................................................................31
5.2. Model validation ...................................................................................................................36
v
5.3. Sensitivity of model parameters........................................................................................... 38

5.4. Effect of land cover changes................................................................................................ 40
5.5. Results regionalization......................................................................................................... 41
5.5.1.
Simple linear regression ..................................................................................... 41
5.5.2.
Multiple linear regression................................................................................... 42
5.5.3.
Validation of regional model.............................................................................. 46
5.5.4.
Spatial proximity ................................................................................................ 47
5.5.5.
Area ratio ............................................................................................................ 47
5.5.6.
Sub-basin mean................................................................................................... 48
6.
Simulating lake water balance................................................................................................... 49

6.1. Water balance ...................................................................................................................... 49
6.2. Water balance terms from observed data............................................................................. 49
6.2.1.
Lake areal rainfall............................................................................................... 49
6.2.2.
Open water evaporation...................................................................................... 50
6.2.3.
Surface water inflow........................................................................................... 52
6.2.4.
Groundwater inflow and outflow ....................................................................... 53
6.2.5.
Lake out-flow ..................................................................................................... 53
6.3. Model development ............................................................................................................. 53
6.4. Results of water balance ...................................................................................................... 54
7.
Conclusion and recommendation .............................................................................................. 57

7.1. Conclusion ........................................................................................................................... 57
7.2. Recommendation ................................................................................................................. 58
References ............................................................................................................................................ 59
List of Acronomys ............................................................................................................................... 63
Appendix A: List of previous studies................................................................................................. 64
Appendix B: Data assimilation........................................................................................................... 65
Appendix C: General procedure in DEM hydro processing........................................................... 70
Appendix D: HBV model setup.......................................................................................................... 71
Appendix E: Testing significance of regression equation................................................................ 77
Appendix F: Albedo calculation from MODIS images.................................................................... 79
Appendix G: Adjusted parameter space ........................................................................................... 83
Appendix H: Weight of Meteorological station for Gauged and Ungauged catchments ............. 84
Appendix I: Result of sensitivity analysis ......................................................................................... 86
Appendix J: The effect of parameter sensitivity .............................................................................. 90
Appendix K: Correlation of catchment characteristics................................................................... 95
Appendix L: Physical catchment characteristics ............................................................................. 96
vi
List of figures
Figure 2-1:
Figure 2-2:
Figure 2-3:
Figure 2-4:
Figure 2-5:
Figure 2-6:
Figure 2-7:
Figure 2-8:
Figure 2-9:
Figure 2-10:
Figure 2-11:
Figure 2-12:
Figure 2-13:
Figure 4-1:
Figure 5-1:
Figure 5-2:
Figure 5-3:
Figure 5-4:
Figure 5-5:
Figure 5-6:
Figure 5-7:
Figure 5-8:
Figure 5-9:
Figure 5-10:
Figure 6-1:
Figure 6-2:
Figure 6-3:
Figure 6-4:
Study Area...................................................................................................................... 5
Meteorological and gauging stations considered within and close to Lake Tana
Basin............................................................................................................................... 6
Annual average rainfall in each meteorological station with its elevation .................... 7
Annual average rainfall distribution of Lake Tana basin............................................... 7
(A) Rainfall-elevation relation in southern part of Lake Tana. (B) Rainfallelevation variation in northern and eastern part of Lake Tana ...................................... 8
Average daily rainfall distribution throughout the year for several rainfall
stations in Lake Tana basin............................................................................................ 8
(A) ADT variation for six different stations over the year during 1994-2003. (B)
ADT variation of the stations with its elevation ............................................................ 9
ADT distribution over the Lake Tana basin .................................................................. 9
Potential evapotranspiration distribution in Lake Tana basin .....................................11
(A) Unexpected river discharge in Gelda river 1998/1999, (B) Double mass
curve for Gelda river with respect to rainfall for the same period...............................12
(A) Suspicious River flow data in Gilgel Abbay River in 1996/1997, (B) Double
mass curve of Gilgel Abey river with respect to rainfall for the same period .............12
Suspicious river flow data in Ribb river with respect to rainfall .................................12
Gauged and ungauged catchments in Lake Tana basin ...............................................13
Variation of average of best 25 NS and |RVE| value against run number ...................28
Scatter plot of decisive scaled criteria against model parameters ...............................32
Left hand side shows the average of best 25 parameters in each 15 calibration
runs and right hand side shows best parameter value in each 15 calibration runs ......34
Model calibration results of Ribb, Gilgel Abbay, Gumara, Megech, Koga (19942000) and Kelti (1997-2000) catchments.....................................................................36
Model validation results of Ribb, Gilgel Abbay, Gumara, Megech, Koga and
Kelti catchment (2001-2003) .......................................................................................37
Sensitivity analysis of Gilgel Abbay, Gumara, Ribb and Kelti catchments ................39
Simulated hydrograph with different forest percentage in Gilgel Abbay
catchment in 1994 ........................................................................................................41
Simulated hydrograph with different forest percentage in Ribb catchment in
1994..............................................................................................................................41
Catchment relation layout in spatial proximity method...............................................47
Annual average runoff with respect to catchment area for gauged catchment from
1994 to 2003.................................................................................................................48
Catchment relation layout based on the area ratio .......................................................48
Water balance component used in the water level simulation model ..........................50
Selected meteorological station in and around Lake Tana to estimate the areal
rainfall from 1994 to 2003 ...........................................................................................51
Average albedo variation during the year ....................................................................52
Schematic view of lake water level simulation model.................................................54
vii
Figure 6-5:
Figure 6-6:
viii
Lake level simulation based on bathymetric relation derived from Pietrangeli

(1990) and Wale (2007) ...............................................................................................55
Comparison of simulated lake level in different regionalization techniques...............55
UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN
List of tables
Table 2-1:
Table 3-1:
Table 3-2:
Table 5-1:
Table 5-2:
Table 5-3:
Table 5-4:
Table 5-5:
Table 5-6:
Table 5-7:
Table 5-8:
Table 5-9:
Table 5-10:
Table 5-11:
Table 5-12:
Table 5-13:
Table 6-1:
Table 6-2:
Major gauged and ungauged catchment delineated in the Lake Tana basin................14
Selected model parameters and their priory range.......................................................19
Selected PCCs ..............................................................................................................20
Calibrated model parameters for gauged catchments (1994-2000) .............................35
Time of concentration for selected gauged catchments in Lake Tana basin (Wale,
2008) ............................................................................................................................35
Model validation results from year 2001 to 2003 ........................................................37
Correlation matrix between model parameters and PCCs for 6 selected catchments;
significant correlation coefficients are in bold text .....................................................42
Statistical characteristics for the regression equation FC ............................................43
Statistical characteristics for the regression equation BETA ......................................43
Statistical characteristics for the regression equation LP ............................................44
Statistical characteristics for the regression equation ALFA ......................................44
Statistical characteristics for the regression equation KF............................................45
Statistical characteristics for the regression equation KS............................................45
Statistical characteristics for the regression equation PERC.......................................45
Statistical characteristics for the regression equation CFLUX....................................46
Validation of the regional model of gauged catchments from 2001 to 2003...............46
Results of NS and RVE for selected bathymetric relation...........................................54
Lake Tana water balance components simulated from 1994 to 2003..........................56
ix
1.
1.1.
Introduction
Background
The most important natural resource required for the survival of all living species is water. Water is
limited, scarce, and also not well distributed in relation to the population needs. Hence proper
management of water resources is essential to satisfy the current demands as well as to maintain
sustainability. Water resources planning and management in the 21st century is becoming difficult due
to the conflicting demands from various stakeholder groups, increasing population, rapid urbanization,
climate change producing shifts in hydrologic cycles. Further variability in runoff in large ungauged
international river basins affects water availability to downstream countries making management of
transboundary water difficult and sometimes causing geopolitical tension. In most international river
basin systems, the runoff variability of upper basins caused by natural or human interference is a
serious concern in water resources management of all riparian countries. The upper Blue Nile basin is
a typical example. The Blue Nile River Basin annually contributes about 60% of runoff to the Nile
River (Conway, 2005) with its 10% areal occupancy. Reliable runoff information from this region is
of great importance for sustainable management of water resources as the Ethiopian Highland is the
major contributor to the Blue Nile River Basin.
1.2.
Problem definition and importance of the study
A better understanding of the hydrological characteristics of different watersheds in the headwaters of

the Nile River is of considerable importance because of the international interest in the utilization of
its water resources, the need to improve and augment development and management activities of these
resources, and the potential for negative impacts of climate change in the future. Conflicting views of
water resource utilization and ownership has been challenging the development of appropriate
management of the Nile River Basin for the countries that depend on its resources including Egypt,
Sudan, and Ethiopia. Egypts agricultural production and domestic requirements depends primarily
(99%) on the waters of the Nile River and use of up to 55.5 billion cubic meters per year (Gheith and
Sultan, 2002). Sudans agriculture also relies heavily on these waters; while Ethiopia, at the Niles
headwaters, is interested in further developing the water resources to meet development needs and
attain water security through equitable sharing among the Nile countries (Arsano and Tamrat,
2005). Only 5% of Ethiopias surface water (0.6% of the Nile Basins water resource) is currently
being utilized by Ethiopia while cyclical droughts cause food shortages and intermittent famine
(Arsano and Tamrat, 2005). There is an increasing demand for irrigation and hydropower
development in Ethiopia to cope with the recurrent drought and its impacts and to increase
agricultural production to cope with increasing population. Currently more than 90% of energy use in
Ethiopia depends on the fuel wood (Kebede et al., 2006). This has resulted in extensive vegetation
degradation which by itself effects the hydrological cycles.
Since there is no better infrastructure for water impounding systems or reliable irrigation schemes to
compensate for prolonged low rain period, the Upper Blue Nile River Basin experiences drought and
1
famine. Further ungauged basins require planning of long term strategies such as hydropower
generation, large-scale irrigation schemes and ecological protection as still many people largely
depend on rain-fed agriculture and small-scale irrigation. Therefore estimation of runoff is important
in gauged and ungauged catchment to understand the temporal and spatial variability of water yield
for the local economies as well as the downstream countries. By scarcity of long term hydrological
and meteorological data on the Upper Blue Nile basin, this study is concentrated only on the Lake
Tana basin which is the source for the Blue Nile River. The basin occupies 10% of the Upper Blue
Nile River basin and contributes 8% of the discharge (Kebede et al., 2006).
1.3.
Scope of the research
Rainfall runoff models are often used to predict stream flow in space and time domain for operational
and scientific investigations. By extrapolating and regionalization it enables us to simulate the
catchment response of catchments for which time-series are not available (Heuvelmans et al., 2006;
Seibert, 1999; Wagener and Wheater, 2006; Xu, 1999). Depending on several aspects the prediction
of this discharge regimes in ungauged and gauged catchments bring along a given degree of
uncertainty. In predicting gauged catchments several aspect can be considered: different model
structures represent the real world differently and cause some degree of uncertainty, specific
information and data required for these models cause uncertainty and the model parameters required
for model calibration also contribute to the uncertainty issue in predicting stream flows. For better
water management these predictions should be done accurately by reducing the uncertainties. Even
though the prediction of discharge is possible and relatively simple in well gauged catchments, the
prediction of discharge in ungauged catchment is complex and has higher degree of uncertainty. In
gauged catchments the values of model parameters can be identified by calibrating the model against
observed discharge. In ungauged catchments the observed data are not available or not sufficient for
model calibration. Hence to predict the model parameters in ungauged catchments will depends on
other sources of information.
The reliable estimation of continuous stream flow time-series in ungauged catchments has remained a
largely unsolved problem so far (Wagener and Wheater, 2006), but significant insight has been gained
in recent years. The establishment of the Prediction in Ungauged Basins initiative by the International
Association of Hydrological Sciences (IAHS) shows that much has still to be done in this area
(Wagener and Wheater, 2006). Several ways of obtaining the required information are addressed in
literature. Analyzing the methods that have been attempted to estimate runoff from ungauged
catchment from the previous studies, regionalization of model parameters, which is the process of
transferring model parameter from comparable catchments to catchment of interest, was the common
approach. It is possible to acquire necessary information in order to reduce model parameter
uncertainty and uncertainty in predictions of discharge regimes in ungauged catchments.
The problem involved with developing even the simplest models lie primarily in data availability.
Most established hydrological models are data intensive, yet the Lake Tana basin has limited rain
gauge coverage, few long term temperature records, few gauged sub-catchments, and very scarce daily
data. The size and complexity of the basin, together with the lack of data, is therefore a severe
constraint to the application of sophisticated hydrological models. Application of a physically based
model is practically not feasible because of their extreme data demand and instead conceptual models
2
are selected which have fewer parameters (Wale, 2008). Hence the HBV model is going to be adopted
for this research with considering the following:
The large number of application of this model, under various physiographic and
climatological conditions, has shown that its structure is very robust and general, in spite of
its relative simplicity (Lindstrm et al., 1997).
The HBV model is a conceptual semi distributed model and a catchment can be partitioned
into sub-basins, elevation zones and land cover types (Krysanova et al., 1999; Menzel and
Brger, 2002; SMHI, 2006). As the Lake Tana basin has mountainous topography this
characteristic of HBV is more relevant to this study.
It has been commonly proposed that parsimonious models are best suited to regionalization
techniques (Crock and Norton, 2000), where HBV model has limited number of sensitive
parameters and able to simulate the dominant hydrological processes.
In the HBV model input data have been kept as simple as possible. This characteristic is
suitable for this study area where data scarcity and data quality are the major problem.
From the Ministry of Water Resources (MoWR) and National Meteorological Agency (NMA),
hydrological and meteorological data has been available for nine well gauged catchments and fifteen
meteorological stations in Lake Tana basin respectively. By establishing relationships through the
well gauged data between conceptual model parameters and physical catchment characteristics,
parameters for the HBV model can be determined in poorly gauged catchments which will lead to a
reduction of parameter uncertainty. Subsequently, this reduction of parameter uncertainty will lead to
reduction of uncertainty in prediction of discharge regimes.
1.4.
Previous studies and applications
A number of methods have been attempted to estimate the runoff from ungauged catchments during
the past three decades. A summary of selected previous studies (see Table-list of previous studies of
regionalization) shows that regionalization of model parameters was the most common approach. It is
clearly indicated that each study used a conceptual hydrological model and there was no proper guide
line for selection of physical catchment characteristics. The most common parameter regionalization
method relates hydrologic model parameters with physical catchment characteristics (PCCs, hereafter)
using regression methods. This approach seems effective in estimating the model parameters of
ungauged basins with limited success (Heuvelmans et al., 2006; Kokkonen et al., 2003; Mwakalila,
2003; Yu and Yang, 2000). Several new methods were also attempted to better regionalize model
parameters. These include the application of spatial proximity (Merz and Blschl, 2004; Parajka et al.,
2005), flow duration curves (Yu and Yang, 2000), basin similarity (Parajka et al., 2005), neural
networks (Bastola et al., 2008; Heuvelmans et al., 2006), and regional calibration methods (Fernandez
et al., 2000; Hundecha and Brdossy, 2004). It is understood from the previous studies that the
accuracy of estimation of runoff from the ungauged catchment cannot be guarantied by any particular
method, but such depends on the quality of data available, specific basin responses, hydrological
model and the regionalization method (Kim and Kaluarachchi, 2008).
1.5.
Objectives
The general objective of this study is to apply a hydrological model and regionalization to simulate
runoff from gauged and ungauged catchment. To achieve the above objectives, the following specific
objectives were addressed.
To make use of computer codes to incorporate automatic calibration in the HBV approach
To apply Monte Carlo Simulation (MCS) for model calibration and parameter analysis
To simulate the water balance of Lake Tana
1.6.

1.7.
Research questions
What are the appropriate regionalization methods that can be used to extend the observed
hydrologic regimes to ungauged basins?
Which model parameters are most sensitive in MCS?
Which physical catchment characteristics are available and suitable to relate to HBV model
parameters?
Can accuracy of inflows from ungauged catchments be defined?
Outline of the Thesis
This thesis contains seven chapters and is organized as follows:

Chapter 1 gives general introduction to the study with emphasis on water scarcity, problem and
importance of the study and objectives of the study. Chapter 2 gives brief description of Lake Tana
basin, data availability and the data quality. In chapter 3 insights is given in hydrological modelling by
expounding the hydrological processes playing a role in rainfall runoff generation and describe the
different aspect concerning the hydrological modelling. Further the approach of regionalization is
described. Chapter 4 describe the methodology adopted in model calibration and regionalisation. In
chapter 5 a detailed discussion is raised against the results of model calibration and regionalisation.
Chapter 6 present the development of water balance model and simulation of lake water level. Chapter
7 finally discusses the conclusion and recommendations by this study.
2. Study area and data preparation

2.1.
Study Area
The Lake Tana basin, one of the important sub basins in Upper Blue Nile river basin, located north
western part of Ethiopia, occupies an area of 15,000 km2 and drains to the Lake Tana which covers
3,060km2. The basin lies approximately between latitude 10.75N and 13.0N and longitude 36.5E and
38.5E. Its altitude varies from 1784m at the Lake level to about 3400m of mountain in east of Lake
Tana at the edge of the Ribb catchment. More than 40 rivers feed the lake. Among them are GilgelAbbay, Ribb, Gumera and Magetch that contribute more than 93% of the inflow. The only surface
outflow of the basin is the Blue Nile River, which comprises 7% of the Blue Nile flow at the EthioSudanese border (Conway, 2000).
Figure 2-1: Study Area
The basin is characterised by Lake Tana which is located in a wide depression of the Ethiopian Plato
and is surrounded by high hilly and mountainous terrain. The outflow of the basin is through a narrow
valley at the south-east direction of the lake.
2.2.
Climate
The climate of the region is tropical highland monsoon with one rainy season between June and
September. The seasonal distribution of rainfall is controlled by the northward and southward
movement of the inter-tropical convergence zone (ITCZ). Moist air masses are driven from the
Atlantic and Indian Oceans during summer (June - September). During the rest of the year the ITCZ
shifts southwards and dry conditions persists in the region between October and May (Kebede et al.,
2006). Even though it is associated with atmospheric circulation, the topography has also an effect on
the local climate. The air temperature shows large diurnal variation but small seasonal changes and is
comparatively uniform throughout the year.
Temperature, precipitation, wind and solar radiation are by far most important meteorological variable
which drives the hydrological processes in a catchment. Precipitation, temperature and potential
evapotranspiration are the main inputs for the HBV model considered in this research and described
in Chapter 3. Great attention has to be paid in preparing and analysing the distribution of these data
within the catchment in rainfall-runoff modelling practice. It is chosen to apply a lumped soil moisture
routine in the HBV model since no detailed information about the catchments is available with respect
to geographical location and elevation. Finally for every catchment average areal precipitation,
temperature and potential evapotranspiration is derived as model input.
2.2.1.
Precipitation
For this study daily precipitation was collected from the National Meteorological Agency (NMA) in
Ethiopia for the period of 1994-2003 of fifteen rain gauge stations within and close to the study area is
shown in Figure 2-2.
Figure 2-2: Meteorological and gauging stations considered within and close to Lake Tana Basin
6
There are missing and suspicious incorrect rainfall data for several stations and used most common
methods described by Bras (1990) and Dingman (1994) such as normal ratio, weighted average and
regression to fill the missing gaps and Double-mass curve analysis to analyse the consistency of
suspected stations are used. The calculated annual average rainfall for each station from the above
method is shown in Figure 2-3 with respect to their elevation. A detailed description about the gap
filling techniques and determination of the areal rainfall over the catchment, accompanying issues and
the used observation stations for averaging is given in Appendix B.
3000.0
3500
2500
2000
1500.0
1500
1000.0
1000
500.0
500
Normal average
Regression
da
r
fir
Ad
az
di
sZ
em
en
D
_T
ab
or
N
_M
ew
ich
a
C
he
w
ah
it
En
G
on
go
ra
G
or
el
g
D
an
o
Ze
g
D
_E
st
if
Ay
ke
l
0
e
0.0
Ad
et
Se
ke
la
D
an
gi
la
En
je
ba
ra
Ba
hi
rD
ar
P (mm)
2000.0
Station Elevation (m,amsl)
3000
2500.0
Normal ratio
Elevation
Figure 2-3: Annual average rainfall in each meteorological station with its elevation
Figure 2-4: Annual average rainfall distribution of Lake Tana basin
These data series were used for calculating the areal rainfall over the catchments. To get insight in the
variability of the precipitation across the catchments, the Annual Average Rainfall (mm/year) over the
period 1994-2003 is shown in Figure 2-4.
The southern part of the Lake Tana basin is wetter than the western and the northern parts (see Figure
2-4). The rainfall-elevation relation over the entire basin show poor agreement. While examining the
rainfall-elevation relationship on the southern part separately shows acceptable agreement and show
great influence by the effect of elevation. But northern and eastern part of Lake Tana showed poor
relation and can not explain such effect clearly. Figure 2-5 depicts this different of rainfall-elevation
relation in two parts of the basin.
A
B
2500.0
2500.0
R2 = 0.8294
R2 = 0.3997
2000.0
P (mm)
P (mm)
2000.0
1500.0
1500.0
1000.0
1000.0
500.0
1500
2000
2500
3000
500.0
1500
3500
1700
1900
2100
2300
2500
2700
2900
3100
Elevation (m,amsl)
Elevation (m,amsl)
Figure 2-5: (A) Rainfall-elevation relation in southern part of Lake Tana. (B) Rainfall-elevation variation in
northern and eastern part of Lake Tana
Analysis of individual stations in the region shows that during the period from June to September it
high rainfall is received and from October to May low rain. Figure 2-6 depicts the average daily
rainfall distribution over the year for selected meteorological stations in the region.
2.2.2.
Temperature
Daily maximum and minimum temperature data was collected from NMA and from the archives for
the period of 1994-2003. Data of six meteorological stations within and close to the study area were
chosen for analysis (see Figure 2-6).
10.0
P (mm)
8.0
6.0
4.0
2.0
0.0
Jan
Feb
Sekela
Mar
Apr
Dangila
May
BahirDar
Jun
Jul
Gorgora
Aug
Gondar
Sep
Oct
D_Tabor
Nov
Dec
N_Mewicha
Figure 2-6: Average daily rainfall distribution throughout the year for several rainfall stations in Lake Tana
basin
8
The missing gaps of each data set were filled using the methods described in section 2.2.1. Areal
temperature for each catchment was estimated by obtaining the weights by inverse distance
interpolation. In order to get insight in the variability present across the catchment, the average daily
temperature (ADT) is shown over the year (see Figure 2-7 [A]).
22.0
20.0
18.0
16.0
14.0
3500
25.0
20.5
20.3
3000
17.6
20.0
16.8
2500
15.8
13.2
2000
15.0
1500
10.0
1000
5.0
500
12.0
Mean daily Temperature
D.
Ta
bo
r
gi
la
ar
irD
G
on
da
r
D.Tabor
ew
ic
ha
Dangila
Gondar
N.
M
BahirDar
N.Mewicha
Da
n
Adet
0.0
Ba
h
Jul Aug Sep Oct Nov Dec
A
de
Jan Feb Mar Apr May Jun
10.0
Mean Temperature (C)
24.0
Elevation (m,amsl)
Mean Temperature ( 0C)
Elevation
Figure 2-7: (A) ADT variation for six different stations over the year during 1994-2003. (B) ADT variation of
the stations with its elevation
During the dry period ADT increases gradually from January to April and start to decline while the
rainy season start. The minimum ADT for the respective station can be observed during July and
August. During this period the basin received higher intensity of rain. Figure 2-7[B] clearly shows
that the temperature of this basin is generally affected by the elevation: lower altitudes have high
temperature and temperature declines while the altitude increases. The overall distribution of average
temperature over the catchment shows lower temperature in southern parts.
Figure 2-8: ADT distribution over the Lake Tana basin
2.2.3.
Potential evapotranspiration
The last necessary input data for the model is potential evapotranspiration (PE). PE data is not found
from the meteorological station or archives. Hence Penman-Montheith method was applied to
estimate PE as this method is widely used in hydrological engineering application. The basic formula
for calculating PE is shown in equation 2-1.
ET0 =
where
900
u2 (es ea )
T + 273
+ (1 + 0.34u 2 )
0.408( Rn G ) +
ETo
Rn
G
T
u2
es
ea
es-ea
[2-1]
reference evapotranspiration [mm day-1]

net radiation at the crop surface [MJ m-2 day-1]
soil heat flux density [MJ m-2 day-1]
mean daily air temperature at 2 m height [C]
wind speed at 2 m height [m s-1]
saturation vapour pressure [kPa]
actual vapour pressure [kPa]
saturation vapour pressure deficit [kPa]
slope vapour pressure curve [kPa C-1]
psychometric constant [kPa C-1]
However for calculation PE, many variables are required such as: relative humidity, temperature,
wind speed, altitude and sunshine hours. In many cases no observation stations are found with all data
sets and if they were, most of the time the quality of the data is insufficient. Many data sets with
missing time series and incorrect data were replaced with new values by the procedures described in
Appendix B. Eventually, average areal PE is calculated for each catchment by finding the weight from
each station using interpolation techniques. A detailed description about determination of the
appropriate dataset of the four variables, accompanying issues and the used observation stations for
averaging is also given in Appendix B.
In order to get insight in the variability across the catchment, the average annual evapotranspiration
(AAE) is shown in Figure 2-9. The minimum average annual PE holds a value of 1291 mm whereas
the maximum holds a value of 1448 mm. The distribution of evapotranspiration over the catchment
shows low values in southern and eastern part and much higher in northern parts. By comparing with
ADT, it shows significant effect for the fluctuation of evapotranspiration.
10
Figure 2-9: Potential evapotranspiration distribution in Lake Tana basin
2.2.4.
Hydrological data
Many sub basins are gauged in Lake Tana basin by hydrometric stations to observe the runoff. The
Lake level is monitored continuously with two level stations. The measuring devise of these stations
are equipped with staff gauges and are recorded manually on daily basis. Some gauged catchments
have several other gauging stations upstream that enable to analyse runoff correlations. Daily river
discharge data series were collected over the catchment from the Ministry of Water Resources
(MoWR) (see Figure 2-2).
According to the Tana Beles Sub-Basin Hydrological Study (SMEC, 2007) and Wale (2008) rating
curves of many rivers in the Lake Tana basin are reliable (Gilgel Abbay, Koga, Megech and Kelti).
However some of them are unreliable as the gauging stations are located middle of the flood plain
(e.g. Ribb and Gumara rivers). One of another major problem is the sediment accumulation that
disturbs the stage-discharge relation ship in many locations.
2.2.4.1.
Analysing hydrological data
The collected river discharge data was screened to identify unreliable and spurious data. Screening
was done graphically by comparison of hydrographs with sub basin hydrographs and double mass
curve analysis with respect to rainfall. Some errors in river discharge data are described here.
In Figure 2-10A daily discharge data of Gelda River shows unexpected increase without any
significant increase in rainfall The analysis of double mass curve clearly indicates the sudden
11
increment of the discharge during year 98/99, violating the well behaved relationship with respect to
the rainfall of previous years (Figure 2-10B).
Rainfall
200
Discharge
0
20
Q (m 3/s)
40
60
100
80
P (mm/day)
150
50
100
0
Cumulative Q (m 3/s)
6000
5000
4000
3000
2000
1000
0
120
Jan-96
Jan-97
Jan-98
Time (days)
Jan-99
2000
4000
6000
8000
Cumulative P (mm/day)
Figure 2-10: (A) Unexpected river discharge in Gelda river 1998/1999, (B) Double mass curve for Gelda river
with respect to rainfall for the same period.
Another type of unrealistic river discharge data can be seen in Gilgel Abbay River during the
recession period at the end of 1996. It shows sudden responses even though the nearby Koga river
does not show such response. Further comparison with basin rainfall also does not show high rainfall
events (Figure 2-11A).
A
Rainfall
Gilgel Abay River
500
Koga River
0
90000
300
40
200
60
100
80
100
Jan-95
Jul-95
Jan-96
Jul-96
Jan-97
Jul-97
Cumulative Q (m3/s)
20
P (mm/day)
Q (m3/s)
80000
400
70000
60000
50000
40000
30000
20000
10000
0
0
2000
4000
6000
8000
Cumulative P (mm/day)
Time (days)
Figure 2-11: (A) Suspicious River flow data in Gilgel Abbay River in 1996/1997, (B) Double mass curve of
Gilgel Abey river with respect to rainfall for the same period
In several years the peak river discharges of Rib River are trimmed and give constant value for a long
period (Figure 2-12). This can be explained for that during extreme flooding the gauging station was
submerged and caused difficulties in getting readings. or the river overflowed. Some of these errors
are detected and explained also by Wale (2008)
Rainfall
200
Discharge
0
20
Q (m 3/s)
40
100
60
80
50
100
0
Jan-94
120
Jan-95
Jan-96
Jan-97
Jan-98
Time (days)
Figure 2-12: Suspicious river flow data in Ribb river with respect to rainfall
12
P (mm/day)
150
2.3.
Delineation of gauged and ungauged catchments
A Digital Elevation Model (DEM) of 90m resolution from Shuttle Radar Topography Mission
(SRTM-version 4) (http://srtm.csi.cgiar.org/) has been used to delineate the gauged and ungauged
catchments by using hydro-processing tools in ILWIS and GIS software. A detailed procedure of
delineation of catchment by ILWIS is available in Appendix C. The extracted 9 gauged catchments
and 10 ungauged catchments are shown in Figure 2-13.
Figure 2-13: Gauged and ungauged catchments in Lake Tana basin
Results from catchment delineation, show that among the nine catchments seven catchments are
partially gauged while catchments in the north-western part are completely ungauged. Nine
catchments are selected as gauged catchments based on the availability of runoff data from 1994 to
2003. Representative areas of the gauged and ungauged catchment are in Table 2-1.
13
Table 2-1: Major gauged and ungauged catchment delineated in the Lake Tana basin
Gauged catchment
Ungauged catchment
Catchment
Area (km2)
Catchment
Area (km2)
Ribb
1407.65
Ungauged Ribb
735.89
Gilgel Abbay
1657.41
Ungauged Gilgel Abbay
2071.95
Gumara
1281.31
Ungauged Gumara
286.63
Megech
531.18
Ungauged Megech
436.78
Koga
297.94
Ungauged Gumero
424.43
Gumero
163.32
Ungauged Garno
364.61
Garno
98.39
Ungauged Gelda
364.37
Gelda
26.13
Ungauged Dema
324.87
Kelti
607.64
Tana west
546.40
Ungauged Gagi Kura
426.80
14
3. Hydrological modelling and regionalization

3.1.
3.1.1.
Hydrological modelling
Hydrological process
At first it is necessary to have insight in hydrological processes before hydrological modelling can be
performed. The formation of the rainfall into stream flow involves many and complex hydrological
processes in upstream area of the catchment. The basis of generating rainfall runoff processes can be
found in the hydrological cycle, which is usually described in terms of precipitation, infiltration,
evaporation, transpiration, surface runoff and groundwater flow. The precipitation over the catchment
is the base of generating runoff. However after infiltration to the soil, the flow path becomes very
unpredictable since the catchment runoff behaviour is closely related to the subsurface physiography,
geometry and geology. Precipitation can be come out of the catchment as runoff through: overland
flow, through flow and groundwater flow. For simulation of these processes a wide variety of
hydrological models as well as applications have been developed over the past decades.
3.1.2.
Classification of hydrological model
Many different types of hydrological models have been developed. Many of these models share
structural similarities because of underlying assumptions, while some of the models are distinctly
different. Therefore, these models are classified according to different criteria. Hydrologists
traditionally proposed two kind of modelling approaches with their strong points and limitations: (1)
physically based and (2) conceptual lumped models. Physically based models consist of formulations
in terms of physical laws expressed in the form deterministic conservation equations for mass,
momentum and energy. The equations are solved numerically by discretizing the hydrological system
into smaller entities on a square or a polygonal mesh. However, accurately modelling of all processes
of the hydrologic cycle becomes very complex, demands an eminent insight in hydrological behaviour
and is very demanding for input data. Due to these properties it is a time-consuming and expensive.
An example of such a model is SHE (Abbott et al., 1986).
As an alternative to physically based distributed models, conceptual lumped models are often used as
robust prognostic tools at catchment scale. The model structures of these models are relatively simple
and often are based on a series of interconnected reservoirs. Further these models are invaluable
instruments for operational water management (e.g. reservoir operation, flood forecasting). The
description of the reservoirs behaviour is kept simple in most structures and their responses are
controlled by parameterizations that are rarely described in terms of physical principles such as
gravity, piezometric heads or hydraulic conductivity and cannot be measured in the field. They must
be estimated using a calibration procedure whereby the model parameters are fined tuned manually or
automatically, by means of optimization algorithms, until the natural system output and the model
output show an acceptable level of agreement. However, once environmental forcing conditions (e.g.,
switching from wet to dry conditions) or catchment characteristics (e.g., land cover pattern) change,
the parameters usually need to be recalibrated (Reggiani and Rientjes, 2005). Because of the fact that
15
the required input and output data are usually easily available, consequently these models are mostly
used in rainfall-runoff modelling. The HBV rainfall-runoff model is an example of a conceptual
model (Bergstrm, 1995).
3.1.3.
Model selection
The above model classifications must be considered when selecting the appropriate hydrological
model in order to contribute to the stated objectives. At first sight, physically-based models are the
most appropriate in modelling rainfall-runoff generation while it is the most elaborate way of
modelling the rainfall-runoff processes. By the rapid improvement of computational power, the
physically based models became practically applicable in the nineties. There has been considerable
discussion regarding the pros and cons of this type of model by Beven (1989). In general it is
concluded that physically based distributed hydrological modelling has clear limitations. Further,
these models suffer from extreme data demand, scale-related problems (e.g. the measurement scales
differ from the process and model scales) and over-parameterization. Therefore, in order to predict
rainfall-runoff in gauged and ungauged catchments these models mostly are not practically applicable.
Conceptual models normally perform at least as well as physically based models in predicting
discharge regimes and furthermore they do not required huge complexity as physically based models
do. Conceptual models on the contrary can be considered as a nice compromise between the need for
simplicity on the one hand and the need for a firm physical basis on the other hand. This is due to the
fact that these models usually are able to capture the dominating hydrological processes at the
appropriate scale with accompanying formulations and therefore are very suitable when used in the
process of regionalization. Furthermore, the HBV model has application in regionalisation (Hundecha
and Brdossy, 2004; Merz and Blschl, 2004; Seibert, 1999) and it demonstrated to be effective. The
improved HBV-96 model is used in this research is described in the next section.
3.2.
HBV Model
The HBV model is a semi-distributed conceptual model and was originally developed at the Swedish
Meteorological and Hydrological Institute (SMHI) (Bergstrm and Forsman, 1973). The general
structure of HBV consists of three model components: (1) snow accumulation and snow melt, (2) the
simulation of soil moisture and runoff, and (3) a response and river routing procedure. For the first
two decades, only minor changes in the basic model structure were made. Experience has shown
however, that the standard version of HBV had some major drawbacks which are outlined in
Lindstrm et al. (1997). Therefore a re-evaluation has been carried out and a new model version has
been developed. The HBV-96 model is the final result of this model revision (Lindstrm et al., 1997).
Henceforward when HBV model is used, it is referred to the HBV-96 model.
The general water balance model is described as:
P E Q =
d
[ SP + SM + UZ + LZ + lakes] + ss
dt
Where
P - precipitation
E - evapotranspiration
Q - runoff
SP - snow pack
16
[3-1]
SM - soil moisture
UZ - upper zone
LZ - lower zone
Lakes - lake volume
ss sink/source
In HBV model applications the water balance equation was simulated as follows. Precipitation over
the catchment is calculated by weighting rain gauge measurements. The areal precipitation is then
distributed over the elevation zones by correcting for altitude with a constant lapse rate.
Evapotranspiration is computed as a function of the soil moisture conditions and potential
evapotranspiration. When the soil moisture exceeds a storage threshold, water would evaporate at the
potential rate. At lower soil moisture values a linear relation between the ratio of evapotranspiration
and potential evapotranspiration and soil moisture is used. The general storage variable is formed by
soil moisture storage and storage in the upper and lower response boxes. Recharge to groundwater is
calculated through a non-linear relation between the ratio of recharge and precipitation and the soil
moisture. The high flow of the catchment is described by the outflow from the upper nonlinear
reservoir, while the base flow is governed from the lower response boxes which receive percolated
water from the upper response box. Runoff from the catchment is given by the sum of the outflow
from the two response boxes.
In this thesis a HBV model code has been developed based on the HBV-96 model version. The
computer language FORTRAN is used and several reasons are on the basis for this decision. When
using FORTRAN, adjustments to the model can be made which are not possible in the regular
interface and such adjustment is the use of MCS for model calibration in this study. Booij (2005)
formerly wrote the HBV model in FORTRAN for the Meuse basin and adjustments were made to the
model based on the selected catchment, methods of routing routine, selected model parameters and
model calibration method. A more profound description of the HBV-96 model is available in
Appendix D.
3.3.
Regionalization
Modelling of catchment runoff processes behaviour with rainfall-runoff models that merely are
approximations of the processes taking place at the catchment scale. This is also the case for the HBV
model. However, values for model parameters have to be established in order to simulate rainfallrunoff generation. Since for the HBV model it is not possible to directly determine the model
parameter values, the parameters are normally estimated through a model calibration process by
fitting the model output to observed discharge data. In many catchments observed discharge data is
not available. Calibration of the model is therefore difficult and prediction of discharge regimes has a
high degree of uncertainty. In order to allow prediction of discharge regimes in ungauged catchments,
a method called regionalization is introduced. This method is used in this study to contribute to the
objectives. There are many definitions of regionalization available in literature, but a generic
definition as stated in Blschl and Sivapalan (1995) is used most often. Regionalisation is the process
of transferring information from comparable catchments to the catchment of interest.
17
3.3.1.
Approach of regionalization
The transfer of catchment information in merely based on some sort of similarity between ungauged
and gauged cachments. A number of methods have been applied to modelling ungauged basin such as
spatial proximity (Merz and Blschl, 2004), physical reasoning and statistical approaches (linking
model parameters to PCCs). The spatial proximity method is based on the principle of catchments that
are close to each other will likely have a similar runoff regime since climate and catchment conditions
will often only vary smoothly in space. So the assumption is made that catchments are highly
comparable with respect to topographic and climatic properties and therefore a particular model
approach and with its parameters that calibrated for gauged catchments can be used to predict the
discharge regime at the ungauged catchment. The most common techniques are nearest neighbour
technique and kriging (Vandewiele and Elias, 1995). Researchers have concluded that geographical
proximity does not guarantee hydrological similarity, and that methods based on physical reasoning
are difficult to realize often due to the difference in scale at which the measurements are made and at
which the model is applied (Bastola et al., 2008).
Another widely used approach to model ungauged catchments is to link MPs to PCCs. This method is
referred to as a regional model method (Fernandez et al., 2000; Heuvelmans et al., 2006; Seibert,
1999; Wagener and Wheater, 2006). To broaden the modeling of ungauged catchment, many
applications of conceptual rainfall runoff (CRR) models are based on functional relationships between
MPs and PCCs. However such approaches are somehow elusive due to the existence of multiple
optima and high interaction between subsets of fitted MPs (Kuczera and Mroczkowski, 1998). Some
improved versions of conventional statistical approaches such as Weighted Regression, Sequential
Regionalization and Regional calibration (Fernandez et al., 2000), have been discussed to overcome
the problem of poor identification of MPs that mostly occur in over-parameterized CRR models.
Wagener and Wheater (2006) discussed the uncertainties and problems of Sequential Regionalization
and Weighted Regression methods for regionalization. However this approach considers the
similarities of catchment characteristics and come up with three steps. First the selected model is
calibrated for a reasonable number of catchments, where sufficiently long and informative
observations of discharge regimes are available. Secondly, most commonly used regression equations
are selected, which predict the model parameter values using one or a combination of PCCs. Each
model parameter is based on a specific equation and thus, a specific regional model is obtained.
Finally, these regional models will be used to estimate the parameter values of the ungauged
catchment (Parajka et al., 2005).
3.3.2.
Selection of regionalization method
The method based on special proximity can only be applied to ungauged catchments located close to a
well gauged catchment. It requires a well gauged catchment in the near surrounding that often is
considered as constrain. Kokkonen et al. (2003) mentioned that a simple transfer of parameters can
outperform regionalization, even if catchments are hydrologically similar. In contrast to this approach,
a regional model can be applied to various catchments even though it requires higher data demand
than the spatial proximity approach. However it is easy to derive these data rather than finding long
term discharge time series. There were two approaches addresses in literature with respect to the
method of similarity of catchment characteristic. One approach predicts the discharge regime with a
priori selected relationship between MPs and PCCs. The major disadvantage of this method is its
18
difficult to identify the firm basis of such relationships in other studies to apply this study with
sufficient confident. Further studies (Hundecha and Brdossy, 2004; Merz and Blschl, 2004; Seibert,
1999) rejected this method due to absence of satisfactory relationships. Hence in this study first the
model was calibrated with respect to observed discharge and established the relationships between
selected PCCs and MPs. Finally the model parameters for ungauged catchments were determined.
3.4.
Model parameter selection
In order to determine the regional model, first model parameters have to be identified by calibrating
the model against observed discharge. The major causes of difficulty in identification of model
parameters are over parameterization and selection of parameters in the calibration. With respect to
the HBV model, several modification have been made to the model structure to reduce the amount of
parameters (Lindstrm et al., 1997), even though Merz and Blschl (2004) mentioned reduction of
over-parameterization is a critical issue. However many studies asses and conclude that HBV model is
parsimonious enough. To establish relationships between PCCs all the parameters should not be used
even though HBV model has more than 30 parameters. Such behaviour will induce extra effort in
establishing statistical relationship. Therefore it is important to determine the most sensitive model
parameters to be considered in the process of regionalization. In HBV model structure these processes
are conceptualized by appointing appropriate model routines such as soil moisture routine which
comprises the Horton overland flow and saturation overland flow, the quick runoff routing which
comprises the macro pore flow and perched subsurface flow and the base flow routing which
comprises the unsaturated subsurface flow and groundwater flow.
All parameters pertaining in a model routine do not affect to the same degree the rainfall-runoff
transformation process. Therefore most sensitive parameters have to be identified in model calibration
and subsequently in regionalization. By this analysis a chance is offered to developed effective
relationships between parameters and PCCs. A number of studies used the HBV model approach and
much experience was gained in demonstrating the most sensitive parameters. Hence, the parameter
spaces from other studies were evaluated. Selected parameters and their spaces are listed in Table 3-1.
Table 3-1: Selected model parameters and their priory range
Name
FC
Description and unit

Maximum soil moisture content [L]
BETA
LP
ALFA
KF
KS
PERC
CFLUX
Parameter in soil routine [-]

Limit for potential evapotranspiration [LT-1]
Response box parameter [-]
Recession coefficient upper zone [T-1]
Recession coefficient lower zone [T-1]
Percolation from upper to lower [LT-1]
Maximum value of capillary flow [LT-1]
3.5.
Prior range
100-800
1-4
0.1-1
0.1-3
0.0005-0.15
0.0005-0.15
0.1-2.5
0.0005 2
Default value
Use a value for
the region
1
1
0.9
0.09
0.01
0.5
0.5
Selection of physical catchment characteristics
The next challenge is to derive PCCs in order to establish statistically significant relationship between
the model parameters and PCCs. A reasonable number of well gauged catchments with good quality
19
data of climate, hydrology, physiography, lithology and geology should be available to derive PCCs.
However prior to selection of PCCs for regionalization, evaluation has to be done as there may be
inter-dependency or inter-correlation between different PCCs. Therefore a preliminary list of PCCs
should be composed based on the available data and the physical meaning of the model parameters.
Then statistical analysis should be performed to identify the highly correlated PCCs. Correlation
analysis and principle component analysis are the generally used methods to asses the correlations
(Heuvelmans et al., 2006). Another more pragmatic approach is to evaluate those PCCs used in other
studies and select them based on hydrological insight. Hence inter-correlation of the PCCs is also
assessed inherently.
Several studies were evaluated in order to select appropriate PCCs. The detailed list of all the studies
is shown in the Appendix A. Eventually the PCCs were evaluated and the selected ones are show in
Table 3-2 with the condition of availability of PCCs.
Table 3-2: Selected PCCs
Group
Geography and
physiography
Land use
Geology and soil
Climate
Parameter
AREA
LFP
MDEM
HI
AVGSLOPE
SHAPE
CI
EL
DD
CROPD
CROPM
GL
URBAN
FOREST
LEP
NIT
VER
LUV
SAAR
PWET
PDRY
PET
PCC and Unit

Catchment area [km2]
Longest flow path [km]
DEM mean [m]
Hypsometric integral [-]
Average slope of catchment [%]
Catchment shape [-]
Circularity index [-]
Elongation ratio [-]
Drainage Density[m/km2]
Cultivated Dominantly[%]
Cultivated Moderately[%]
Grassland[%]
Urban[%]
Forest[%]
Leptosol area [%]
Nitosol area [%]
Vertisol area [%]
Luvisol area [%]
Standard annual average rainfall [mm]
Mean precipitation wet season (Jun. to Sep.) [mm]
Mean precipitation dry season (Oct. to May) [mm]
Mean annual evapotranspiration [mm]
A brief description of each group of the PCC is given below.

1. Geography and physiography
Catchment area:
Area of the catchment is easily derived for each of the catchments. The amount of water reaching the
outlet of the catchment mainly depends on its area. The model parameter BETA was reasonably
20
related to area according to the studies by Booij (2005) and Seibert(1999). In the study of Wale
(2008) the area was related to BETA, FC and LP with high significance level.
Longest flow path:
The longest flow path is one of the outputs in catchment delineation processes in ILWIS. This
indirectly is an indication of time for water to reach the gauging station. Wale (2008) established a
reasonable relation between FC and longest flow path.
DEM mean:
Mean elevation is one of the frequently used PCC and is obtained as an output in the processes of
delineation of catchment using SRTM DEM. Deckers (2006) mentioned that the elevation is highly
correlated to slope. Thus, necessary attention should be made when establishing a relation between a
certain PCC and elevation as well as slope.
Hypsometric integral:
This indicates the distribution of elevation across the catchment and simply calculated as:
HI =
H mean H min
H max H min
where:
[3-2]
Hmean average altitude of the basin above mean sea level [m]
Hmax maximum altitude of the basin above mean sea level [m]
Hmin minimum altitude of the basin above mean sea level [m]
Average slope:
Slope is one dominant factor that controls the water flow velocity where a high slope result in high
velocities that reduce the travel time of water to reach the catchment outlet. A percentage slope map
was calculated using the function in ILWIS and categorised according to the FAO slope classes.
Class a: Level to undulation, dominant slope ranging from 0 to 8 percent
Class b: Rolling to hilly, dominant slope ranging from 8 to 30 percent
Class c: Steeply dissected to mountainous, dominant slope over 30 percent
Catchment shape:
Catchment shape is not a widely used PCC. This PCC is however expected to affect the hydrological
processes at the catchment scale (Deckers, 2006). It is determined through formulae [3-3].
SHAPE =
H max H min
[3-3]
AREA
21
Circularity index:
The circularity index is calculated as the ratio of perimeter square to the catchment area.
CI =
P2
A
[3-4]
where P and A are perimeter [km] and area [km2] of the catchment respectively.
Elongation ratio:
Elongation ration represent how the shape of the basin deviates from a circle. It is calculated by
dividing the diameter of a circle with the same area as the catchment, by the length of the catchment.
EL =
Dc
L
[3-5]
where Dc: is the diameter of the circle with the same area as the catchment, L: is the maximum length
of the catchment along a line basically parallel to the main stem.
2. Land use
The characteristic land cover is one of the most used PCCs when establishing a regional model. This
includes land cover type such as forest, grassland, crops and urban etc. It is a general concept that
deforestation increases the soil erosion, since it changes the soil properties and infiltration rates. The
land cover map, which has been updated by Landsat ETM+ and evaluated based on field data, used
here was collected from ITC archives.
3. Geology and soil
Soil:
The soil map of the major soil groups of the catchment classified as per the FAO soil group was used
for this study. This map was collected from the GIS department of EMWR. The dominant soils are
described below.
Leptosols are extremely gravelly and stony. They can be seen over continuous rock in very shallow
layers. These are azonal soils and common in mountainous regions,
Luvisols are soils that have higher clay content in the subsoil than in the topsoil as a result of
pedogenetic processes (especially clay migration) leading to an argic subsoil horizon. These soils have
high-active clays throughout the argic horizon. Further these soils have a medium to high storage
capacity for water and nutrients and are well aerated.
Nitisols are deep, well-drained, red, tropical soils with diffuse horizon boundaries and a subsurface
horizon with more than 30 percent clay and moderate to strong angular block structure elements that
easily fall apart into characteristic shiny, polyhedric (nutty) elements. These soils have relatively high
content of weathering minerals and surface soil may contain several percent of organic matter, in
particularly under forest or tree crops.
Vertisols are churning, heavy clay soils with high proportion of swelling clays. These soils form deep
wide cracks from the surface downward when they dry out.
All the above descriptions are taken from the FAO soil groups (WRB, 2007).
22
4. Climate
Standard annual average rainfall
The most commonly used characteristic is the standard annual average rainfall (SAAR) with respect
to the climatic PCCs. For this characteristic the data are frequently available. However this is not
commonly applied in other studies but in Deckers (2006) it proved to be a good indication for climatic
variability.
Mean precipitation wet season and dry season
It is observed that there are two clear seasons for precipitation in the region, with high rainfall during
June to September and low rainfall during October to May. Hence average daily rainfall in the wet
season and the dry season was selected separately as climate PCCs.
Average annual evapotranspiration
The average annual evapotranspiration also has significant distribution over the catchment and varied
from 1290 to 1450 mm/year. Kim (2008) has used this characteristics for the upper Blue Nile river
basin and got reasonable relation with model parameters used in his study. In this study average
annual evapotranspiration was used as a PCC.
23
4.
Methodology
The approach of regionalization was applied to estimate the flow of ungauged catchments. First, the
HBV model was calibrated against observed discharge to determine well performing parameter sets of
gauged catchments. Next, a relationship was made between the model parameters (MPs) and physical
catchment characteristics (PCCs) to establish the so called regional model that serves to estimate
model parameters for ungauged catchments. Then the HBV model was used to simulate the
discharged for ungauged catchments. Finally, the water level simulation model of Lake Tana was
developed to increase our knowledge of the water balance of the Lake with the simulated ungauged
catchments runoff. In the following subsection a description of the procedure is given.
4.1.
Model Calibration
Any hydrological model must be proven for its reliability, accuracy and predictive ability. At the
initial run the model probably will not give satisfactory result as the input data do not reflect the real
world with enough accuracy. Hence every hydrological model requires adjustment of the model
parameter values, hydrological influences and stresses in order to tune the model. The reliability of
the model can be improved by fine-tuning termed model calibration or simply calibration. The
procedure of adjusting the model input parameters is necessary to match the model output with
measured field data for the selected period and situation entered to the model (Rientjes, 2007).
4.2.
Approach of calibration
The process of model calibration in order to identify the optimum model parameter set is done either
by manually or by computer-based automatic procedures. In manual calibration, the user adjusts the
parameters interactively in successive model simulations. As this approach mainly depends on the
users experience, only intelligent steps will be made through the parameter space that will be an
advantage. Some previous studies proved that this method demonstrated good model performance
(Lidn and Harlin, 2000). However manual calibration is subjective and the parameter derived may be
prone to be bias due to involvement of users experience. Also the process is very time-consuming
and it does not have clear point at which the calibration process can be said to be completed. To
overcome these disadvantages, automatic calibration has been pursued. In automatic calibration model
parameters are adjusted automatically according to a specific search algorithm or optimization
algorithm that considers numerical performance measures for the goodness of fit. Further, automatic
calibration is fast with respect to manual calibration and the confidence of the model simulation can
be explicitly stated (Madsen, 2000).
In this study calibration of the HBV model was done using Monte Carlo Simulation (MCS) to select a
well performing optimum parameter set based on the objective functions, since the objective of this
study was to establish a robust regional model rather than to optimally calibrate the catchment models.
The MCS is a technique where through numerous model simulations with randomly generated model
parameter sets, an optimum value for the objective function(s) is sought (Booij et al., 2007). For all
25
selected parameters a parameter space was determined through defining lower and upper boundary
values (Table 3-1).
4.3.
Objective functions
In model calibration the model parameters have to be adjusted until the observed natural system
output and the simulated model output show an acceptable level of agreement. This goodness of fit is
always evaluated through an objective function which is selected based on several criteria. These
criteria should be selected properly to evaluate different aspects of the hydrograph. Some criteria that
were considered in selecting the objective function in this study are as follows, see (Madsen, 2000):
1. A good agreement between the average simulated and observed catchment runoff volume (i.e.
a good water balance)
2. A good overall agreement of the shape of the hydrograph
Based on these criteria relative volume error (RVE) and the Nash-Sutcliffe (NS) coefficient are used
as objective functions to evaluate model performance.
Relative volume error
The relative volume error (RVE) criterion is defined as:
(Qsim ,i Qobs,i )
100%
RVE = i =1 n
Qobs,i
i =1
[4-1]
where Qsim: is the simulated flow, Qobs: is the observed flow, i: is the time step, n: is the total number
of simulation time steps of the calibration period. RVE ranges between to + where with a zero
value implies no difference between the simulated and observed discharge. The relative volume error
between +5% and -5% indicate that the model performed well while relative volume error between
+5% and +10% or between -5% and -10% indicate that model perform reasonably. However, at the
same time the distribution of the discharge throughout the calibration period can be completely
wrong. Therefore, this criterion should always be used in combination with another criterion that
considers the overall shape agreement.
Nash-Sutcliffe coefficient
The Nash-Sutcliffe coefficient (NS) is used to evaluate the overall agreement of the shape of the
simulated and observed hydrograph. NS measures the efficiency of the model by relating the goodness
of fit of the simulated data to the variance of the measured data. NS coefficient is defined according to
the following equation:
26
NS = 1
(Q
i =1
n
sim ,i
Qobs ,i ) 2
(Qobs,i Qobs ) 2
[4-2]
i =1
where Qsim: is the simulated flow, Qobs: is the observed flow, Qobs : is the mean of the observed flow,
i: is the time step, n: is the total number of time steps used during calibration. NS can range between
and 1and a value of 1 corresponds to a perfect match of model discharge to the observed discharge.
The NS value goes to 0 indicate that the model predictions are as accurate as the mean of the observed
data. When the value is less than 0 (between and 0) the observed mean is a better predictor than
the model. This criterion is used frequently in various studies and when values range between 0.6 and
0.8, the model performed reasonably. The model is said to perform very good when the values is
between 0.8 and 0.9 and it perform extremely well when the value is between 0.9 and 1 (Deckers,
2006).
4.4.
Determination of optimum parameter set
The MCS requires a feasible parameter space to determine the values for the objective function while
the HBV model simulates the discharge. Hence as discussed in section 3.4.1 sensible parameters and
the feasible parameter spaces were determined for MCS in model calibration. In MCS there are huge
numbers of parameter sets generated randomly and it is not know for which parameter values the
model performs best. In order to evaluate which parameter set performs best over the two selected
criteria, the value of each criterion was scaled over their own range determined through the calibration
runs. The NS value was scaled based on its minimum and maximum value and scaling was done
through equation [4-3].
C ' NS ,i ,n =
C NS ,i ,n min(C NS ,i ,ntot )
max(C NS ,i ,ntot ) min(C NS ,i ,ntot )
[4-3]
where C: value for the criterion, NS,i: NS value for specific catchment, n: calibration run number,
ntot: total calibration runs.
Since RVE varies between and + and performance is best at a value of 0, positive values as well
as negative values may occur. For RVE scaling was determined through equation [4-4].
C ' RVE ,i ,n =
C RVE ,i ,n max C RVE ,i ,ntot

min C RVE ,i ,ntot max C RVE ,i ,ntot
[4-4]
where C: value for the criterion, RVE,i: NS value for specific catchment, n: calibration run number,
ntot: total calibration runs.
After having scaled NS values and RVE values, for each calibration run the lowest value among these
two were selected. This is shown in equation [4-5].
27
C 'i ,n = min{C ' NS ,i ,n , C ' RVE ,i ,n }
[4-5]
where C: scaled value of the criteria, i: specific catchment, n: calibration run number
Finally, the optimum parameter set for each catchment is determined by selecting the highest values
of all selected minimum values as determined through equation [4-5]. The best performed parameter
set was selected by equation [4-6] considering these two criteria.
Ci = max{min(C 'i ,ntot )}
[4-6]
where C: scaled value of the criteria, i: specific catchment, ntot: total calibration run number.
0.85
2.4
0.84
2.0
0.83
1.6
|RVE|
NS
After determine the feasible parameter range, it needs to determine the so called optimum run number.
In MCS, at the optimum run number it is assumed that its examine the entire range of parameter
values. To determine the optimum run number the model was run for Gilgel Abbay catchment by
increasing the run number starting from 500 runs up to 80,000. In the mean time the average of NS
and |RVE| of the best 25 parameter sets was calculated in the decisive scale. The results are plot
against the run number (see Figure 4-1).
0.82
1.2
0.81
0.8
0.80
0.4
0.79
0.0
20000
40000
60000
80000
Run #
20000
40000
60000
80000
Run #
Figure 4-1: Variation of average of best 25 NS and |RVE| value against run number
The figure shows while increasing the run number the NS and |RVE| value come to a stable value. It
can be concluded that the parameter space is well examined when the run number reach around
50,000 runs. Harlin and Kung (1992) also state that when the mean of the selected criteria shows
stability implies the parameter space is well examined. Thus, for each calibration run number is fixed
to 60,000 and selected the calibration period from 01-01-1994 to 31-12-2000.
4.5.
Establishing the regional model
The relation between HBV model parameters and PCC allows us to understand and perhaps
quantitatively predict how a change in physical properties of a catchment will affect its hydrological
response (Mwakalila, 2003). In order to set up a regional model to predict the model parameters in
ungauged catchments, a statistically significant and hydrologically meaning full relation should be
established between PCCs and optimized MPs. After determining the MPs through model calibration
28
and selection of physical catchment characteristics as explained in section 3.4, a method for
establishing the relationship is applied. The most commonly used approach is through regression
analysis (Bastola et al., 2008; Booij et al., 2007; Heuvelmans et al., 2006; Kim and Kaluarachchi,
2008; Mwakalila, 2003; Parajka et al., 2005; Xu, 2003) and this approach is used in this study as well.
A problem in this study is the limited number of available gauged catchments. In principle increasing
the number of catchment will increase the reliability and the efficiency of the regional model. It is
noted that in this study gauged catchments with RVE in between +5% and -5% and NS value greater
than 0.6 were use to establish the regional model.
4.6.
Regression analysis
A regression analysis is the application of a statistical procedure for determining a relationship

between variables (Haan, 2002). In this procedure one variable is expressed as a function of another
variable. The variable to be determined is termed the dependent variable while others are called the
independent variables. In this study MPs were taken as dependent variable and PCCs were taken as
independent variables. With respect to regionalization two type of regression methods are applied.
One is simple regression method, which tries to determine a relationship between one independent
variable and a dependent variable. The other type is called multiple regression method, which tries to
assess multiple independent variables to explain the dependant variable.
Linear regression analysis
In this study linear regression is performed with respect to each model parameter. To optimize this
simple relationship between MPs and PCCs the significance and strength of the relation was tested.
First the correlation coefficient (r) was tested and in addition a t-test [eq. 4-7] was performed.
tcor =
r n2
1 r 2
[4-7]
where, tcor: t value of the correlation, r: correlation coefficient, n: sample size.

The following hypothesis was tested. The null hypothesis H0 and the specific hypothesis H1 are:
H0: The correlation between the PCC and model parameters is zero, = 0.
H1: The correlation between the PCC and MP is not zero, 0
If tcor > tcr the null-hypothesis is rejected (MP are associated with PCC in the population)
In order to determine the critical value tcr, the degree of freedom df and , a number between 0 and 1
to specify the confident level have to be determined. The default value for is 0.05 for 95% confident
interval. It must be note that the choice of level is largely subjective. In this study a significant level of
= 0.1 was chosen to a two-tail test with n-2 degree of freedom. Using this information, for tcr a value
of 2.132 was found (critical value from t distribution table). In order to determine at what value of r
the hypothesis is rejected the test statistic was solved. Result shows a value of 0.72, thus r greater than
0.72 or smaller than -0.72 results in a significant relationship.
29
Multiple regression analysis

The second method applied was based on the multiple regression analysis to optimize the linear
relation with forward selection or backward removal method. Multiple regression is used to predict
MPs from several independent PCCs. The forward entry approach the initially established regression
model, which incorporates the significant best PCC will be extended by forcing a second independent
variable in the regional model. This step will be accepted if the entry statistic (i.e. significance level,
) of both independent variables is not exceeded. The statistical tools are used to choose the most
significant independent variable to be added. Further some more steps are executed, until the last
added independent variable does not significantly contribute to the regression model. This equation is
selected for further use.
The other method is backward elimination method that is applied in the event if all available PCCs
exceed the entry statistic, i.e. at the first step no independent variable could be added when applying
the forward entry method. This will terminate the regression analysis and no regression model would
be established. But in the backward removal method all expected PCCs are forced into the model.
Then the statistical tools will assess the removal statistic (i.e. significance level, ) and evaluate the
eligibility to remove independent variables from the model. This will continue until the remaining
independent variables do not exceed the exit statistics. In this way the multiple regression model is
determined.
The r2, the coefficient of determination, is not the only measure to rely on but also the significance of
the relation has to be tested. In order to test the significance, the multiple regression equations can be
tested by a test of significance of individual coefficients or by a test of overall significance.
4.7.
Test of significance and strength
A hypothesis test has to be applied in order to determine if the regression equation is significant. It is
possible to test several hypotheses. But for each test, assumptions have to be made and it is assumed
that the error terms, , are not correlated and normally distributed. Further they have an average of
zero and a constant variance. In this study these assumptions were made and two hypothesis tests were
executed in order to test the significant of the regression equation. Those are null-hypothesis and
specific hypothesis. Further the strength of the determined regression equation is also tested by r2
called coefficient of determination. A detailed description of determination of significance of the
regression equation with hypothesis test and the strength is described in Appendix E.
30
5. Results and discussion

5.1.
Results of model calibration
Since it is difficult to estimate values of model parameters from the field measurements, all the model
ranging from parsimonious lumped to complex distributed physically based need to be calibrated. In
order to select an optimum parameter set, in this study MCS as calibration method was used. The
HBV model was calibrated against the observed discharge by using the predefined parameter ranges
as explained in section 3.4.1. However at the initial calibration, several catchments out of the total of
nine catchments resulted in poor model performance with respect to the RVE. By screening the
observed discharge data several unreliable and spurious data are identified for respective catchments.
The methods described in Appendix B were adopted to correct those data points and corrected data
has been used for calibration.
After calibrating the model for the period from 1994 to 2000 best performing parameter set is derived
for each catchment with respect to the water balance and overall shape agreement of the observed
discharge using RVE and NS respectively as objective functions (see equation [4-1] and [4-2]).
Further by narrowing the parameter space some parameters are well identifiable. It can be observed
that model is most sensitive in small ranges of parameter space for particular model parameters while
model perform insufficient outside this range. To get an insight of well identifiable parameters a
scatter plot of decisive scaled values (see equation [4-5]) against a priori selected parameter space of
each MPs were made and Figure 5-1 shows selected eight parameters for catchment Gilgel Abbay.
According to the results except for ALFA, others are not well identifiable since the model performed
equally well for each parameter value within the a priori parameter space. An uneven distribution can
be seen from the scatter plot of FC, thus it is also defined as poorly identifiable and the parameter
space remained unchanged for further calibration. In the case of ALFA the scatter plot shows that that
model performed well for lower values. Hence the parameter space of ALFA can be narrowed.
It is observed that for other gauged catchments the scatter plots show the same coherence and in order
to assess all catchments at the same way the following approach is chosen to narrow the parameter
spaces:
For each catchment the best 6,000 out of 60,000 values of the decisive scaled values are
derived. This for the reason that it is assumed that no good model performance will be
realized when using a parameter set belonging to a poorer decisive scaled value.
Hereafter, for every model parameter the respective parameter values are determined from
which the minimum and maximum values of all 6,000 best runs are derived.
These minimum and maximum values are used as a new boundary for the adjusted parameter
space.
31
1
FC[mm]
0.8
0.8
0.7
0.7
0.6
0.5
0.4
0.3
0.4
0.3
0.2
0.1
0
200
400
600
800
1000
LP[-]
0.9
ALFA[-]
0.9
0.8
0.8
0.7
Decisive scale
Decisive scale
0.5
0.1
0
0.6
0.5
0.4
0.3
0.7
0.6
0.5
0.4
0.3
0.2
0.2
0.1
0.1
0
0
0.2
0.4
0.6
0.8
1.2
0.5
1.5
1
KF[1/day]
0.9
KS[1/day]
0.9
0.8
0.8
0.7
0.7
Decisive scale
Decisive scale
0.6
0.2
0
0.6
0.5
0.4
0.3
0.6
0.5
0.4
0.3
0.2
0.2
0.1
0.1
0
0
0
0.05
0.1
0.15
0.2
Decisive scale
0.6
0.5
0.4
0.3
0.1
0.15
0.2
CFLUX[mm/day]
0.9
0.8
0.8
0.7
0.05
PERC[mm/day]
0.9
Decisive scale
BETA[-]
0.9
Decisive scale
Decisive scale
0.9
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.2
0.1
0
0
0.5
1.5
2.5
0.5
1.5
2.5
Figure 5-1: Scatter plot of decisive scaled criteria against model parameters
The above procedure is adopted for every catchment to narrow the parameter space. The posterior
parameter spaces for each catchment are shown in Table G-2 in Appendix G. The result showed that if
there is no major change in the parameter space, any change of parameter space will not help to find
the best performing parameter set. The a priori parameter spaces were maintained for further use.
It showed that when the run number is 60,000, the entire parameter space was defined. But it can be
observed that in different calibration run different well performed parameter sets can be found. To
reduce this parameter uncertainty, the procedure was repeated 15 times and an optimum set for each
run was defined. This was through a procedure of selecting and averaging over the 25 best performing
32
sets for each run. In the remaining of the procedure, to remove the outliers the variance over the 25
values were defined and parameter values that lies outside the standard deviation were denied.
Graphical results of this procedure depicted in Figure 5-2 indicate that parameters are not unique but
somehow converge to an optimal value. Finally the optimum parameter set was selected again taking
the average of these 15 parameter values.
Average-FC
500
500
400
400
300
200
6
9
Number of runs
12
15
Average-BETA
4.0
3.5
3.5
3.0
3.0
2.5
2.5
2.0
12
15
12
15
2.0
1.5
1.5
1.0
1.0
0.5
0.5
0.0
6
9
Number of runs
Best-BETA
4.0
BETA
BETA
200
0.0
0
6
9
Number of runs
12
15
Average-LP
1.4
1.2
1.0
1.0
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
0.0
6
9
Number of runs
Best-LP
1.4
1.2
LP
LP
300
100
100
0.0
0
6
9
Number of runs
12
15
Average-ALFA
0.6
0.5
0.5
0.4
0.4
0.3
12
15
0.3
0.2
0.2
0.1
0.1
0.0
6
9
Number of runs
Best-ALFA
0.6
ALFA
ALFA
Best-FC
600
FC (mm)
FC (mm)
600
0.0
0
6
9
Number of runs
12
15
6
9
Number of runs
12
15
33
Average-KF
0.25
0.25
0.20
0.20
0.15
0.10
0.05
0.05
0.00
0
12
15
0.25
0.25
0.20
0.20
0.15
12
0.10
0.05
0.05
0.00
0
6
9
Number of runs
12
15
Average-PERC
2.00
1.80
PERC
1.20
1.00
0.80
0.60
0.40
0.20
0.00
0
6
9
Number of runs
12
CFLUX
0
6
9
Number of runs
12
15
12
15
Best-PERC
6
9
Number of runs
12
15
12
15
Best-CFLUX
2.00
1.80
1.60
0.20
0.00
6
9
Number of runs
1.20
1.00
0.80
0.60
0.40
0.20
0.00
15
Average-CFLUX
2.00
1.80
1.60
1.40
1.20
1.00
0.80
0.60
0.40
2.00
1.80
1.60
1.40
1.60
1.40
1.40
1.20
1.00
0.80
0.60
0.40
0.20
0.00
0
6
9
Number of runs
Figure 5-2: Left hand side shows the average of best 25 parameters in each 15 calibration runs and right hand
side shows best parameter value in each 15 calibration runs
This procedure was applied for all the catchments and optimum parameter sets were established.
Finally, the model was run and checked weather the NS and RVE values are in acceptable range. The
model calibration results shown in Table 5-1 indicate that the model performance of Ribb, Gilgel
Abbay, Gumara, Megech, Koga and Kelti catchments are satisfactory with RVE within 5% and NS
greater than 0.6. In this case for Ribb and Kelti catchments rainfall was corrected by 20% and 18%
respectively.
34
15
0.15
0.10
0.00
6
9
Number of runs
Best-KS
0.30
KS
KS
6
9
Number of runs
Average-KS
0.30
PERC
0.15
0.10
0.00
CFLUX
Best-KF
0.30
KF
KF
0.30
Table 5-1: Calibrated model parameters for gauged catchments (1994-2000)

Gilgel
Ribb
Abbay
Gumara Megech Koga
Kelti
Gumero
Garno
Gelda
309.03
434.39
349.86
193.04
730.05
196.62
469.23
221.25
141.14
LP
1.23
0.73
2.08
0.63
1.31
0.87
1.56
0.71
1.34
0.42
1.60
0.62
1.10
0.26
2.58
0.23
1.20
0.86
ALFA
0.31
0.24
0.25
0.29
0.41
0.28
1.08
0.27
0.51
KF
KS
0.07
0.10
0.08
0.09
0.03
0.07
0.03
0.09
0.07
0.05
0.03
0.10
0.03
0.13
0.10
0.11
0.003
0.15
PERC
1.09
1.02
1.44
1.47
1.63
1.53
2.32
1.61
1.41
CFLUX
0.60
1.09
0.72
0.79
0.74
0.83
0.39
1.35
1.00
NS
0.78
0.85
0.72
0.61
0.67
0.66
0.16
0.33
0.41
RVE%
-1.61
-0.35
-2.44
2.91
-0.06
-2.00
0.01
0.00
-0.06
FC
BETA
The result of the calibration was not satisfactory for catchments with a relatively small area such as
Gumero (163.32km2), Garno (98.39km2) and Gelda (26.13km2). Hence these model parameters were
ignored in establishing the regional model. Wale (2008) mentioned that the time of concentration,
which is define as the length of time takes for water to travel from hydrologically most remote point
to the outlet is relatively small and it is difficult to capture the quick runoff on daily time steps for
these catchments. Time of concentration is measured according to the following equation and the
basin time of concentration is shown in Table 5-2.
L Lc
Tc = 0.7
S
0.38
[5-1]
where, Tc: is the time of concentration [hr], Lc: is the distance from the outlet to the centre of the
catchment [km], L: is the length of the main stream [km], S: is the slope of the maximum flow
distance path (Dingman, 2002).
Table 5-2: Time of concentration for selected gauged catchments in Lake Tana basin (Wale, 2008)
Gilgel
Catchment
Ribb
Gumara Megech Koga
Kelti
Gumero Garno
Abbay
Tc [hr]
32.63
28.64
30.15
19.49
16.38
24.29
9.56
8.27
Gelda
4.66
Further it was mentioned that those gauging stations are not placed at the catchment outlet, but at
some location upstream that has easy road access. As such it is assumed that rainfall-runoff time
series of those catchments can not be considered trustworthy. Figure 5-3 shows the model calibration
results of accepted catchments.
35
20
200
40
150
60
100
80
50
100
120
20
300
40
60
200
80
100
100
120
Gum ara
0
40
200
60
80
100
100
120
Megech
100
0
20
Q (m 3/s)
80
40
60
60
40
80
20
100
120
Koga 100
0
20
Q (m3/s)
80
40
60
60
40
80
20
100
0
Apr-94
120
Apr-95
Apr-96
Apr-97
Apr-98
Apr-99
Apr-00
Kelti 200
0
20
Q (m 3/s)
150
40
60
100
80
50
100
0
Apr-97
120
Apr-98
Apr-99
Apr-00
Time (days)
P
Qobs
Qsim
Figure 5-3: Model calibration results of Ribb, Gilgel Abbay, Gumara, Megech, Koga (1994-2000) and Kelti
(1997-2000) catchments
Model validation
Representing the real world system by a model approach may not be accurate, since the real world is
too complex. Models therefore are uncertain and models cannot be stated reliable when only one field
situation is simulated. As such, it may occur that under different hydrologic stress conditions the
model doesnt accurately represent the real world behaviors despite the fact that optimal and
calibrated model parameters are used (Rientjes, 2007).
36
P (mm)
20
300
P (mm)
Q (m 3/s)
400
5.2.
P (mm)
400
P (mm)
Q (m 3/s)
Gilgel Abbay
P (mm)
250
P (mm)
Q (m 3/s)
Ribb 300
Ribb 200
60
50
80
100
20
300
40
200
60
100
80
100
20
300
40
200
60
100
80
100
Megech
200
0
20
Q (m 3/s)
150
40
100
60
50
80
100
0
Koga 100
Q (m 3/s)
P (mm)
400
P (mm)
Q (m 3/s)
Gumara
500
80
20
60
40
40
60
20
80
100
Kelti 120
100
Q (m 3/s)
P (mm)
400
P (mm)
Q (m 3/s)
Gilgel500
Abbay
20
80
40
60
60
40
P (mm)
Q (m 3/s)
40
100
P (mm)
20
150
80
20
0
100
Apr-01
Oct-01
Apr-02
Oct-02
Apr-03
Oct-03
Time (day)
Pl
Qobs
Qsim
Figure 5-4: Model validation results of Ribb, Gilgel Abbay, Gumara, Megech, Koga and Kelti catchment (20012003)
The calibrated model is then run for the validation using the time series of meteorological variables
that were not used for calibration. The model validation period runs from 2001 to 2003 (see Figure
5-4). Table 5-2 shows the model validation results from 2001 to 2003.
Table 5-3: Model validation results from year 2001 to 2003
Ribb
Gilgel Abbay
Gumara
Megech
Koga
Kelti
NS
0.87
0.85
0.79
0.51
0.65
0.67
RVE
3.55
-2.32
-9.87
2.87
-9.83
-5.30
37
5.3.
Sensitivity of model parameters
To get a clear understanding of the model behaviour with respect to the model outcome (i.e. model
hydrograph) a sensitivity analysis is performed. Each of the eight model parameters contribute to
conceptualizing the rainfall-runoff processes which all together result in simulating the hydrograph.
Therefore if we change one of the model parameters the hydrograph will change. But every parameter
does not contribute the equal amount of change to the hydrograph as described in section 3.4. When
trying to establish relationships between model parameters and PCCs, it is most effective to
investigate the most sensitive model parameters. Furthermore, it is useful to understand the influence
of change in model parameter values on the hydrograph when evaluating the relationships. Figure 5-5
shows the sensitivity analysis for the Gilgel Abbay, Ribb, Gumara and Koga catchments.
Gilgel Abbay catchment
25
0.9
RVE
NS
20
0.85
15
10
0.8
5
0
0.75
-5
-10
0.7
-15
-20
0.65
-125
-100
-75
FC
-50
BETA
-25
LP
25
ALFA
KF
50
KS
75
PERC
100
125
-125
-100
FC
CFLUX
-75
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
75
PERC
100
125
CFLUX
Gumara catchment
40
0.85
NS
RVE
0.75
30
0.65
20
0.55
10
0.45
0.35
-10
0.25
-20
0.15
-125
-100
-75
FC
-50
BETA
-25
LP
-30
0
ALFA
25
KF
50
KS
75
PERC
100
125
-125
CFLUX
-100
FC
-75
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
75
PERC
100
125
CFLUX
Ribb catchment
0.8
50
RVE
NS
40
0.75
30
20
0.7
10
0
0.65
-10
-20
0.6
-30
-40
0.55
-125
-100
FC
38
-75
BETA
-50
-25
LP
0
ALFA
25
KF
50
KS
75
PERC
100
CFLUX
125
-125
-100
FC
-75
BETA
-50
-25
LP
0
ALFA
25
KF
50
KS
75
PERC
100
CFLUX
125
Kelti catchment
80
0.75
NS
RVE
0.65
60
0.55
40
0.45
20
0.35
0.25
-20
-40
0.15
-125
-100
FC
-75
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
75
PERC
100
CFLUX
125
-125
-100
FC
-75
BETA
-50
-25
LP
0
ALFA
25
KF
50
KS
75
PERC
100
CFLUX
Figure 5-5: Sensitivity analysis of Gilgel Abbay, Gumara, Ribb and Kelti catchments
With respect to the NS value, among the 8 model parameters FC, BETA and KF act as the most
sensitive parameter and show non-linear model performance to these parameters. LP and ALFA show
relatively high sensitivity and also the non-linear behaviour to the model performance. KS, PERC and
CFLUX act as less sensitive parameters and show less response to the model performance. The
changes to FC, BETA and LP largely effect the volume of discharge and show higher variation in
RVE, but not much effect from the other parameters. Further, the change of most of the parameter
values resulted linear variation for RVE performance indicator.
Even though the relation between NS value with respect to the percentage deviation of parameter
values show parabolic shape, it is not symmetrical. This can be explained by careful observation and
visual interpretation of the hydrograph produced for each deviation of sensitive parameters (see
Figures in Appendix J). For each model parameter an indication is given for which part of the
hydrograph a specific model parameter is most sensitive.
It can be observed that small values of FC result in more responsive model behaviour. This is due to
the reduction of storage capacity. This will result in a rapid drop of actual evaporation in dry season
as water is quickly release from the system (see Figure J-1). Thus more water is released from the
system and resulted in positive values for RVE. When the FC increases, the storage capacity will
increase and the model increases the actual evaporation. Such causes delays in runoff response time
and also reduces the discharge. This gives negative values to the RVE. Further the model does not
behave the same way while the FC decreases or increases. When FC decreases, the model response
very quickly and show significant changes on the hydrographs that results in large changes in NS and
RVE values. But not such response is observed when FC increases.
Small values of BETA result in more infiltration from the soil moisture state to the quick runoff
reservoir. In this case the potential to generate more discharge is higher (see Figure J-2). Hence it
shows rapid variation of RVE and NS values while it reduces. But this higher response is not visible
when increasing the value of BETA. It responses smoothly and the changes in NS and RVE are
relatively small.
It is observed that lower values of LP can result in lower discharges; due to the fact that LP is a limit
where above the evapotranspiration reaches its potential value. A high soil moisture storage the
39
125
evapotranspiration reaches its potential value and in total more precipitation will evaporate (see
Figure J-3). This will give negative values for RVE. When the value increases, the actual evaporation
reduces and the model releases more water to the system. But in this case also the reduction and
increase of the LP value does not result in similar model response. It shows a different gradient for
RVE when decreasing and increasing the LP value. In the decrement side of LP the gradient is low
and in the increment side of the LP the gradient shows some higher value.
When changing the ALFA this affects the model performance with respect to NS value, but such is not
clearly observed for RVE. When reducing the ALFA value, the model reduces the fast flow (see
Figure J-4) and when ALFA increases it produces quick flow faster. This affects the shape of the
hydrograph and also affects the NS value.
In changing KF, the NS value shows asymmetrical parabolic variation and RVE shows linear
variation. Even though the variation of KF affects the NS value considerably, it does not affect much
to the RVE. When reducing KF, the fast flow does not response well and also affects the hydrograph
(see Figure J-5) and peaks are not well simulated. But when increasing the KF, it responses well and
peaks are nicely matched. The model response differently when reducing and increasing the KF value
and such causes the unsymmetrical curve for the NS values.
5.4.
Effect of land cover changes
In order to assess the sensitivity of the response of the catchments to changes in the land use, different
land use scenarios were generated and the model was run using the same meteorological input used in
the calibration and validation periods while optimised parameters were keep unchanged. Two
hypothetical land use scenarios were established. The first scenario was generated by reducing the
forest percentage to 10% as uniformly distributed in each catchment to investigate the impact of
deforestation on the runoff generation. The second scenario was used to investigate the effect of
intensified afforestation and it was generated by assuming that the 90% of the catchment was forested.
By visual inspection of simulated hydrographs that clearly indicate that strong deforestation has an
effect of increasing the peak flow to a considerable extent. (see Figure 5-6 and 5-7). Although the
peak flow was increased consistently for all rainfall events, the extent to which it was increased was
higher in the wet season than in the dry season.
Intensified forestation will increase the canopy interception and part of the precipitation will not
contribute to runoff. Further it increases the transpiration, water holding capacity and increases delay
of the quick runoff and overland flow. This will cause the reduction of high runoff.
40
0
400
20
Q (m 3/s)
60
200
P (mm)
40
300
80
100
100
120
Apr-94
Jul-94
Oct-94
Time (days)
Forest 90%
Actual Forest(21%)
Forest 10%
Figure 5-6: Simulated hydrograph with different forest percentage in Gilgel Abbay catchment in 1994
250
20
200
40
60
P (mm)
Q (m 3/s)
150
100
80
50
100
120
Apr-94
Jul-94
Oct-94
Time (days)
Forest 90%
Actual Forest(12%)
Forest 10%
Figure 5-7: Simulated hydrograph with different forest percentage in Ribb catchment in 1994
5.5.
5.5.1.
Results regionalization
Simple linear regression
The correlation coefficient was established between PCCs and MPs in order to determine the
significance of each relationship (Table). When the correlation coefficient lies outside critical values
of -0.72 to 0.72, the corresponding correlation is significant. Thus, the null hypothesis is rejected. In
Table 5-4 significant correlation coefficients are presented in bold text.
41
Table 5-4: Correlation matrix between model parameters and PCCs for 6 selected catchments; significant
correlation coefficients are in bold text
FC
BETA
LP
ALFA
KF
KS
PERC
CFLUX
[mm]
[-]
[-]
[-]
[1/d]
[1/d]
[mm/d]
[mm/d]
AREA
-0.18
-0.01
0.49
0.13
0.62
0.07
-0.77
-0.82
LFP
-0.17
-0.15
0.53
-0.51
0.25
0.52
-0.16
-0.74
MDEM
0.04
-0.58
0.33
-0.64
-0.11
0.66
-0.51
-0.51
HI
-0.03
-0.44
-0.44
-0.07
-0.81
0.77
-0.76
0.77
AVGSLOPE
-0.30
SHAPE
CI
0.65
EL
DD
CROPD
CROPM
0.78
0.54
-0.74
-0.39
0.47
-0.48
0.39
-0.45
-0.29
0.37
0.92
-0.04
-0.46
-0.52
-0.90
-0.37
-0.75
0.12
0.58
-0.31
-0.42
-0.55
0.43
0.28
-0.83
-0.37
-0.59
-0.32
0.59
-0.01
-0.48
0.57
-0.66
0.23
0.64
-0.35
0.30
-0.04
0.03
0.35
-0.52
0.10
-0.22
-0.83
0.20
-0.16
0.06
0.77
-0.71
0.77
-0.36
0.27
0.69
-0.55
GL
-0.18
-0.54
0.36
-0.19
0.08
0.61
-0.42
-0.67
URBAN
FOREST
-0.53
0.67
-0.19
-0.61
0.42
-0.60
-0.71
0.47
-0.72
0.20
0.59
-0.09
-0.05
0.23
-0.13
-0.63
LEP
-0.50
-0.36
0.20
-0.23
-0.59
0.57
0.14
-0.44
NIT
0.26
-0.31
-0.56
0.26
-0.49
-0.16
0.69
-0.26
VER
LUV
0.65
0.04
0.07
0.40
-0.73
0.30
0.81
0.15
0.18
0.37
-0.88
-0.55
0.71
-0.08
0.09
0.38
SAAR
0.45
0.52
-0.31
0.12
0.75
-0.29
-0.42
0.62
PWET
PDRY
0.45
0.41
0.39
0.69
-0.21
-0.45
0.07
0.20
0.73
0.71
-0.24
-0.36
-0.46
-0.31
PET
-0.13
-0.11
-0.23
-0.09
-0.59
0.13
0.43
5.5.2.
0.49
0.81
-0.05
Multiple linear regression
It is assumed that the use of multiple PCCs will give better relation than the use of only one.
Therefore relations between PCCs and MPs were assessed through multiple linear regression analysis.
This was done by the forward entry method and backward removal method as described in section 4.6.
FC
FC[mm] corresponds to the maximum basin-wide water holding capacity of the soil. The value of FC
can be estimated based on soil type and the rooting depth of the predominant vegetation and can
further be refined in the calibration process (Hundecha and Brdossy, 2004). The small FC values
imply shallow hydrologically active soil depths, which may be realistic given that bare rock covers a
substantial portion of the catchment areas in these regions (Merz and Blschl, 2004). Zhang and
Lindstrom (1997) showed that FC is correlated with climate, geological, geographical and
hydrological conditions of the catchment. Merz and Blschl (2004) revealed that FC has a positive
correlation with porous aquifers and tends to increase the storage capacity of the catchment. Seibert
(1999) found that there is a strong relation in between FC and lake percentage. Wale et al. (2008)
found that FC has a negative relation with catchment area, longest flow path and average elevation of
the catchment.
42
In this study FC showed significant positive correlation with CI and negative correlation with HI and
DD. The highest correlation is with HI. The forward entry method was executed with HI as the initial
variable. There was no other significant variable that improve the strength of the relation and the
procedure was terminated. The regression equation was determined with only HI with R2 of 66.3%.
The statistical characteristics are shown in Table 5-5.
Table 5-5: Statistical characteristics for the regression equation FC
FC = 0 + 1 HI
0
1
Coefficient
3520.82
-6651.21
p-value
0.0351
0.0487
tcal
3.1317
-2.8032
Std error
1124.26
2372.70
R2
66.3%
BETA
The HBV model covers a wide range of soil conditions in the runoff generation function with
empirical parameters, BETA and earlier described FC. BETA describes how the runoff coefficient
increases as the soil moisture approaches its limit of water holding capacity. BETA is thus more an
index of heterogeneity than of soil properties in the basin. A BETA-value of zero implies that the basin
is entirely lacking in water-holding capacity in the soil, whereas a high BETA-value indicates such
homogeneous conditions that the whole basin may be regarded as buckets that overflow
simultaneously when their field capacity is reached (Bergstrm and Graham, 1998).
The study done by Seibert (1999) showed that BETA and catchment area has positive correlation.
Hundecha and Brdossy (2004) showed that BETA correlated with soil type and land used. Merz and
Blschl (2004) found that BETA is negatively related to elevation and topographic slope. In the study
of Deckers (2006) significant relation could not be established, and a backward elimination method
was used to establish two regressions equations with Arable, URBAN and URBAN and SAAR . In the
study of Wale (2008) BETA has significant positive correlation with catchment area.
In this study, BETA was negatively correlated to SHAPE and positively correlated to CROPD. As the
highest relation is with SHAPE, the forward entry method was executed including SHAPE as the
initial variable. The results of forward entry method showed that BETA is correlated with SHAPE and
HI with R2 of 96.02%. The statistical characteristics are shown in Table 5-6.
Table 5-6: Statistical characteristics for the regression equation BETA
BETA = 0 + 1 SHAPE + 2 HI
0
1
2
Coefficient
p-value
tcal
Std error
R2
7.551
0.0100
5.85
1.2918
96.02%
-8.544
0.0429
-3.39
2.5233
-0.036
0.0034
-8.50
0.0043
LP
The next HBV model parameter in the soil moisture routine is LP that defines the minimum soil
moisture at which the full potential evaporation takes place from the soil water. By a soil moisture
below LP the actual evaporation reduces linearly to zero until the soil drains completely. In the study
43
by Hundecha and Brdossy (2004) LP showed negative correlation with soil type. In the results of
Wale (2008) LP has significant relation with catchment area. In this study LP has significant positive
correlation with HI and negative correlation with VER. The forward entry method was executed with
highly correlated HI as the initial variable. This result showed that LP was correlated to HI and LUV
with R2 of 91.1%. The statistical characteristics are shown in Table 5-7.
Table 5-7: Statistical characteristics for the regression equation LP
LP = 0 + 1 HI + 2 LUV
0
1
2
Coefficient
p-value
tcal
Std error
R2
-2.2435
0.0258
-4.13
0.5432
91.1%
5.8697
0.0133
5.27
1.1141
0.0027
0.0471
3.26
0.0008
ALFA
In the response routine of HBV model, ALFA is the measure of the non-linearity in the upper
reservoir. Booij (2005) indicated that small catchments with steep hills and low permeable soils
generally result in more non-linear behavior in the fast flow mechanisms than large sub-basins with
flat terrain and high permeable soils. Hundecha and Brdossy (2004) established a positive relation
with soil type and land use. But in the study of Wale (2008) and Deckers (2006) any significant
relation with ALFA was not found and backward elimination was performed. Wale (2008) established
two regression equations with bare and hilly percentage and logarithm of average altitude and bare
land percentage. Deckers (2006) established a relation with elevation, hypsometric integral and
permeability percentage. In this study ALFA has significant positive correlation with VER and
negative correlation with AREA, HI and AVGSLOPE. Therefore optimisation of the linear relation
with forward entry method was executed initiating with variable AREA. After adding the catchment
characteristic URBAN the R2 increased up to 95.1% and this regression equation was accepted. The
statistical characteristics are shown in Table 5-8.
Table 5-8: Statistical characteristics for the regression equation ALFA
ALFA = 0 + 1 AREA + 2 URBAN
0
1
2
Coefficient
p-value
tcal
Std error
R2
0.45233
0.0003
18.63
0.02428
95.1%
-0.00009
0.0251
-4.17
0.00002
-0.73650
0.0341
-3.71
0.19865
KF
KF is the recession coefficient of the upper response reservoir in the quick runoff routine of the HBV
model. The recession coefficient is determined using ALFA and two additional parameters hq [mm/d]
and khq [d-1] representing respectively a high flow rate and a recession coefficient at a corresponding
reservoir volume [mm] (see Appendix D). In this study KF showed significant simple linear relations
with URBAN (-0.72), SAAR (0.75) and PWET (0.73). The forward entry method was executed by
taking SAAR as the initial variable. By adding other variables to the equation the strength was not
improved and the simple relation with R2 of 56.35% was accepted. The statistical characteristics are
shown in Table 5-9.
44
Table 5-9: Statistical characteristics for the regression equation KF
KF = 0 + 1 SAAR
0
1
Coefficient
p-value
tcal
Std error
R2
-0.06555
0.3095
-1.16
0.05636
56.35%
0.00009
0.0855
2.27
0.00004
KS
KS is the recession coefficient for the base flow routine in the HBV model and represents the slope of
linear behaviour of slow-flow of the catchment with respect to excess water acquired from the quick
runoff routine. In the study of Seibert (1999) KS is highly correlated with percentage of forest and
lake. KS is sensitive to soil type, land use, size and slope of the catchment according to the study done
by Hundecha (2004). In the study of Wale (2008) for this area any significant relation between KS and
catchment characteristics could not be established. In this study KS has significant simple linear
relation with HI (0.77), AVGSLOPE (0.92) and VER (-0.88). The forward entry method was executed
by adding AVGSLOPE, which is the highest correlated variable, as initial variable. The strength of
the equation could not be improved by addition other variables and the simple linear relation was
accepted with R2 of 85.25%. The statistical characteristics are shown in Table 5-10.
Table 5-10: Statistical characteristics for the regression equation KS
KS = 0 + 1 AVGSLOPE
0
1
Coefficient
p-value
tcal
Std error
R2
0.0187
0.2093
1.49
0.0125
85.25%
0.0018
0.0086
4.81
0.0004
PERC
In the study of Merz and Blschl (2004) it was shown that PERC is negatively related to the river
network density and it was suggested that in catchments with few streams a larger portion of water
penetrates deep into the subsurface than is the case for catchments with a large river network density.
According to Hundecha (2004) percolation is sensitive to soil type. In the study of Wale (2008) a
significant simple relation with percentage of Luvisols was established. In this study PERC has
significant negative relation with AREA (-0.82), LFP (-0.74) and DD (-0.83). By showing the highest
negative simple correlation with DD, that agreed to the arguments made by Merz and Blschl (2004).
The forward entry method was executed by adding DD as the first variable and after including SAAR,
R2 increased up to 89.9%. The statistical characteristics are shown in Table 5-11.
Table 5-11: Statistical characteristics for the regression equation PERC
PERC = 0 + 1 DD + 2 SAAR
0
1
2
Coefficient
p-value
tcal
Std error
R2
7.4926
0.0088
6.11
1.2266
89.9%
-0.0128
0.0192
-4.61
0.0028
-0.0005
0.0864
-2.52
0.0002
45
CFLUX
The parameter CFLUX, which also belongs to the soil water routine, describes the maximum capillary
flow from the upper response box to the soil moisture zone. In this study CFLUX has significant
simple negative correlation with SHAPE (-0.83) and positive correlation with PDRY (0.81). Therefore
optimisation of the linear relation with the forward entry method was executed with SHAPE as the
initial variable. The results of the stepwise forward entry regression showed that CFLUX is correlated
with SHAPE, PDRY and PET with R2 of 99.8%. The statistical characteristics are shown in Table
5-12.
Table 5-12: Statistical characteristics for the regression equation CFLUX
CFLUX = 0 + 1 SHAPE + 2 PDRY + 3 PET

Coefficient
p-value
tcal
Std error
R2
-0.2184
0.2689
-1.52
0.1441
96.27%
-0.0082
0.0021
-21.86
0.0004
0.3867
0.0019
22.63
0.0171
0.0007
0.0184
7.28
0.0001
0
1
2
3
5.5.3.
Validation of the regional model
The regional model is established to predict the discharge from the ungauged catchment. Prior to use,
this regional model is assessed by comparing the predicted and observed discharges from the gauged
test catchments. It is not possible to carry out a formal validation process as too limited numbers of
gauged catchments are available in this study. Therefore validation is done with simulated gauged
catchment for the period of 2001 to 2003. Wale (2008) used similar procedure to validate the regional
model in his study.
The established regional model was used to estimate the model parameters of gauged catchment using
their PCCs. Next the discharge was simulated based on the estimated parameter and the model
performance with respect to NS and RVE was evaluated. Table 5-13 shows the parameters derived
from the regional model and the model performances.
Table 5-13: Validation of the regional model of gauged catchments from 2001 to 2003
FC
BETA
LP
ALFA
KF
KS
PERC
CFLUX
NS
[-]
RVE
[%]
Ribb
298.60
1.167
0.699
0.288
0.055
0.098
1.100
0.615
0.85
-1.32
Gilgel Abbay
332.93
1.989
0.720
0.247
0.086
0.084
1.127
1.099
0.83
0.13
Gumara
306.65
1.482
0.827
0.281
0.057
0.079
1.399
0.712
0.75
-22.77
Megech
201.20
1.535
0.700
0.286
0.031
0.085
1.518
0.787
0.54
13.34
Koga
659.31
1.324
0.410
0.425
0.068
0.061
1.626
0.739
0.65
-1.12
Kelti
437.07
2.454
0.724
0.393
0.072
0.054
1.175
1.062
0.53
-42.03
46
5.5.4.
Spatial proximity
The regression method is the most widely used regionalization technique but alternative methods are
in use. The choice of catchments from which information is to be transferred is usually based on some
sort of similarity measure, i.e. one tends to choose those catchments that are most similar to the site of
interest. One common similarity measure is spatial proximity, based on the rationale that catchments
that are close to each other will have a similar runoff regime as climate and catchment conditions will
only vary smoothly in space (Merz and Blschl, 2004). Vandewiele and Elias (1995) used a similar
approach to estimate parameters of monthly water balance model for 75 catchments from neighboring
catchments. In this approach the complete set of model parameters is usually transferred from one or
more gauged catchments to ungauged catchments, while in the regression the parameters are
regionalized independently from each other. In this study parameter values derived for gauged
catchment in upstream areas were transferred to downstream areas of ungauged catchment. Also
catchments which are not gauged or failed to be simulated were given parameters from a neighboring
catchment. Figure 5-8 shows the catchment relation layout based on the spatial proximity method.
Figure 5-8: Catchment relation layout in spatial proximity method
5.5.5.
Area ratio
This method considers only catchment areas by assuming that the catchment area is the dominant
factor that controls the volume of water as produced by rainfall. The simulated annual average runoff
in gauged catchments showed correlation with R2 72.5% for catchment area (Figure 5-9).
47
25000
y = 10.799x - 3004.2
R2 = 0.7252
Q (m 3/s)
20000
15000
10000
5000
0
0
200
400
600
800
1000
1200
1400
1600
1800
Area (km )
Figure 5-9: Annual average runoff with respect to catchment area for gauged catchment from 1994 to 2003
Hence parameter sets of gauged catchments were transferred to ungauged catchments of comparable
area. Figure 5-10 shows the catchment relation layout based on the area ratio.
Figure 5-10: Catchment relation layout based on the area ratio
5.5.6.
Sub-basin mean
The sub-basin mean represents the arithmetic mean (Kim and Kaluarachchi, 2008) of calibrated
parameter set of 6 catchments to simulate the flow from ungauged catchment.
48
6. Simulating lake water balance

6.1.
Water balance
A water balance often leads to an understanding of a hydrological system. Several studies were done
to study the lake water level simulation and water balance of Lake Tana (Gieske et al., 2008; Kebede
et al., 2006; Wale et al., 2008). Dingman (1994) refers to a water balance as the amount of a
conservative quantity entering a control volume during a defined period minus the amount of quantity
leaving the control volume during the same period equals the change in the amount of the quantity
stored in the control volume during the same time period. The simple water balance equation is
formulated as follows:
S
= Inflow Outflow
T
[6-1]
General water balance equation of a lake can be written as:
S
= (P + Qgauged + Qungauged + GWin ) (E + Qout + GWout ) + ss
T
[6-2]
where, P: is lake areal rainfall, Qgauged: is surface water inflow from gauged catchment, Qungauged: is
surface water inflow from ungauged catchment, GWin: is subsurface water inflow, E: is open water
evaporation from the lake surface, Qout: is surface water outflow, GWout: is subsurface outflow, ss: is
sink source term.
6.2.
Water balance terms from observed data
According to the water balance equation there are many processes involved to the lake water balance.
Precipitation, inflow from gauged and ungauged catchment and subsurface inflow contribute to the
total inflow to the lake. Total outflow from the lake consist of open water evaporation, river outflow
from the lake and groundwater recharge. Figure 6-1 depicted the water balance component used in the
water level simulation model.
6.2.1.
Lake areal rainfall
Seven meteorological stations within and around the lake were selected to calculate the areal rainfall
from 1994 to 2003 (see Figure 6-2). Inverse distance weighting functions were used to interpolate the
daily observation of these station to be converted to obtain areal coverage. A detailed description of
interpolation of rainfall is in Appendix B.
49
Q (m 3/s)
800
Qgauged
600
400
200
0
Q (m 3/s)
800
Qungauged
600
400
200
0
80
P (mm)
60
40
20
0
10
ET
ETp (mm)
8
6
4
2
0
Q (m 3/s)
800
Qout
600
400
200
0
Apr-94
Apr-95
Apr-96
Apr-97
Apr-98
Apr-99
Apr-00
Apr-01
Apr-02
Apr-03
Figure 6-1: Water balance component used in the water level simulation model
6.2.2.
Open water evaporation
The main factors influencing evaporation from a open water surface are the supply of heat for
vaporization and the process of vapour transportation away from the evaporative surface. The solar
radiation, wind velocity and the gradient of specific humidity in the air above the open water surface
are the major influencing factors. Even though the evaporation is one of the major component of
water balance of a lake, it is difficult to estimate. Further direct measurement techniques are not
recommended for routine hydrologic engineering applications because they imply a time consuming
procedure requiring expensive equipment in order to obtain precise and carefully designed
experiments (John, 2006).
In this study the Penman combination approach based on Penman-Monteith method is applied which
is widely used as a standard method in hydrological engineering applications to estimate potential
evaporation from open water under varying location and climatic condition. The Penman combination
approach is formulated as follows:
Ep =
Rn +
Ea
+
+
[6-3]
where Ep: is potential evaporation that occurs from the free water evaporation [mm day-1], Rn: is net
radiation exchange for the free water surface [mm day-1], : is slope of saturation vapour pressure
curve at air temperature [kPa 0C-1] : is latent heat of vaporization [MJ kg-1], : is psychrometric
50
constant [kPa 0C-1], Ea: is the drying power of the air given as a daily rate (mm/day) (Vallet-Coulomb
et al., 2001).
Figure 6-2: Selected meteorological station in and around Lake Tana to estimate the areal rainfall from 1994 to
2003
Ea = f (u )(ew e)
[6-4]
where (ew e) is the saturation deficit, different between the saturated (ew) and actual (e) vapour
pressure (mbars), f(u) is Penmans function of wind given by:
f (u ) = 0.26(1 + 0.54U 2 )
[6-5]
where U2 is the wind speed measured at 2m height (Vallet-Coulomb et al., 2001).

Penmans equation does not include heat exchange with the ground, and the deep energy loss G (from
the water surface into the bottom sediments) and advection term have been ignored here. This
assumption is acceptable for monthly or daily estimations in practical hydrological applications (John,
2006). To estimate the net radiation, Rn the following simplified equations are used.
Rn = Rs (1 ) RL
[6-6]
where, Rs: is short wave radiation, : is surface Albedo, RL: is long wave radiation
51
Rs = 0.25 + 0.5 Ra
N
[6-7]
where, N: is maximum possible duration of sunshine hours [hour], n: is actual duration of sunshine
[hour], Ra: is extraterrestrial radiation [MJ m-2 day-1].
Generally in calculating surface evapotranspiration the value of albedo, which represents the fraction
of incoming radiation that is reflected back to the atmosphere, is taken as 0.23 (FAO-56). However
river water is much more absorbent and less reflective. This will result in more solar radiation
absorbed by the water that as such will warm and evaporate the water. The Albedo can be estimated
from a number of satellite sensors and Wale (2008) used landsat ETM+ images to estimate the albedo
of the lake within the range of 0.05 to 0.062.
In this study Terra MODIS Level 1 product was acquired for the year 2000 and 2002 from the
LAADS Web (http://ladsweb.nascom.nasa.gov/data/search.html) and used Surface Energy Balance
System (SEBS) in ILWIS to estimate the albedo. A detailed description of the procedure of estimation
of albedo is annexed in Appendix F. Figure 6-3 shows the variation of average albedo during the year
lies between 0.08 and 0.16. Other meteorological data was used from the Bahir Dar meteorological
station to estimate the open water evaporation by equation [6-3]. The results shows annual average
evaporation over the Lake Tana is 1563mm. If the value of albedo is used as 0.06 through out the year
the evaporation will be 1670mm/year.
0.18
0.16
0.14
Albedo
0.12
0.1
0.08
0.06
0.04
0.02
0
1
31
61
91
121
151
181
211
241
271
301
331
361
Time (days)
Figure 6-3: Average albedo variation during the year
6.2.3.
Surface water inflow
The water coming through rivers, streams and over land flows are considered as surface water inflow
to the Lake. There are four major gauged rivers dominating surface inflow contributing more than
93% of the total inflow (Kebede et al., 2006; Wale, 2008). The flow data from the gauged catchments
with reliable continuous daily flow data are considered directly to the simulation of the water balance
and related lake level. The inflow from the ungauged catchment is considered separately for each
regionalisation method.
52
6.2.4.
Groundwater inflow and outflow
In the vicinity of the Lake Tana there is no groundwater monitoring data, hence the groundwater flow
from the lake and towards the lake is uncertain. However, since the lake is located in a wide
depression of plato, it seems to have slight groundwater flow towards the lake. In the report of SMEC
(2007) it is mentioned that some 80m thick clay layer underlies the lake floor that prevents inflow or
outflow through the lake bottom. By considering the above fact the groundwater component of the
water balance is assumed to be negligible and is ignored in this study.
6.2.5.
Lake out-flow
Daily water levels of the Abbay station at Bahir Dar were obtained from a database of the Hydrology
Department of MoWR and cover the period 1994 to 2003. The daily outflow discharge data is also
obtained from the same way for the same period.
6.3.
Model development
The above water balance components are integrated according to the equation [6-2] and Lake level is
simulated by area-volume and elevation-volume relationship determined by Wale (2008) and
Pietangeli (1990) referred by SMEC (2007). A computer code was developed to simulate the water
level of the Lake by calculating the water balance component in terms of volume. The initial volume
of the lake was fixed according to the observed lake level and the initial lake area was calculated
accordingly. This area is used to calculate the lake areal precipitation and open water evaporation as
both of them are defined as a function of lake surface area.
V Lake(i ) = VLake (i 1) + S i
[6-8]
VLake(i) Lake volume at day i

VLake(i-1) Lake volume at day i-1
Si Change in storage at day i
S
= P + Q gauged + Qungauged E Qout
T
[6-9]
Following equations (6-10 and 6-11) shows the bathymetric relation derived in the previous studies.
Polynomial fitted bathymetry by Pietrangeli (1990) reads:
E = 1.08 10 9 V 2 + 3.88 10 4 V + 1775.58

A = 6.2 10 8 V 2 + 1.72 10 2 V + 2516.3
[6-10]
Polynomial fitted bathymetry by Wale (2008) reads:
E = 1.21 10 13 V 3 1.02 10 8 V 2 + 6.2 10 4 V + 1774.63

A = 7.93 10 11V 3 5.81 10 6 V 2 + 1.65 10 1V + 1147.51
[6-11]
The schematic view of the water balance simulation model is shown in Figure 6-4.
53
Figure 6-4: Schematic view of lake water level simulation model
6.4.
Results of water balance
Based on the above mentioned procedure the water level of the Lake Tana was simulated based on the
bathymetric relation and the river discharges that estimated from ungauged catchment by
regionalization. Figure 6-5 shows comparison of lake level simulation based on Pietrangeli (1990) and
Wale (2008). Results are shown in Table 6-1.
Table 6-1: Results of NS and RVE for selected bathymetric relation
NS
RVE (%)
Wale (2008)
0.92
-2.17
Pietrangeli (1990)
0.60
-3.21
54
These results shows that the bathymetric relation derived by Wale (2008) give good results that as
such used to evaluate the reliability of different regionalization techniques.
1789.5
1789
Lake level (m,amsl)
1788.5
1788
1787.5
1787
1786.5
1786
1785.5
1785
Apr-94
Apr-95
Apr-96
Apr-97
Apr-98
Apr-99
Apr-00
Apr-01
Apr-02
Time (days)
Observed
Pietrangeli
Abeyou
Figure 6-5: Lake level simulation based on bathymetric relation derived from Pietrangeli (1990) and Wale
(2008)
One of the uncertain components of the Lake Tana water balance is the discharge from ungauged
catchments. The discharges of ungauged catchments are estimated using four different regionalization
techniques that are used to simulate the water level fluctuation using the result of Wales interpolated
bathymetry (see Figure 6-6). Among the four regionalization techniques, regression method gave the
best results and the worst was by sub-basin mean method.
1793
1792
Lake level (m,amsl)
1791
1790
1789
1788
1787
1786
1785
Apr-94
Apr-95
Apr-96
Apr-97
Apr-98
Apr-99
Apr-00
Apr-01
Apr-02
Time (days)
Observed
Regression
Spatial priximity
Area ratio
Sub-basin mean
Figure 6-6: Comparison of simulated lake level in different regionalization techniques
Except for the regression method, other regionalization techniques show large deviation with
observed lake level and simulated, that gradually increase over the simulated period. Hence the result
from the regression method is used to calculate the Lake Tana water balance component and results
are shown in Table 6-2. It shows that the balance closure term is as large as 85mm. This error of the
water balance is 2.7% of the total lake inflow and the lake relative volume error is 2.17%. These
errors may be due to uncertainty in lake-groundwater interaction, uncertainty in estimating open water
evaporation and lake areal rainfall, runoff from gauged and ungauged catchments.
55
Table 6-2: Lake Tana water balance components simulated from 1994 to 2003
Water balance components
mm/year
MCM/year
Lake areal rainfall
+1347
+4104
Gauged river inflow
+1254
+3821
Ungauged river inflow
+527
+1605
Lake evaporation
-1563
-4762
River outflow
-1480
-4508
Closure term
+85
+260
56
7. Conclusion and recommendation

7.1.
Conclusion
The main objective of this study is predicting discharges at the ungauged catchments as accurate as
possible. Therefore, the research has been executed using classical approaches of regionalization
techniques with the following steps:
First best performing model parameter values were identified by calibrating the HBV model
against observed discharge in daily time steps for gauged catchments.
Secondly select regionalization techniques were applied to transfer model parameter to
ungauged catchment by
o Establishing relationship between calibrated model parameters with climatic and
physiographic data: jointly call the regional model
o Transferring model parameters based on the spatial similarity of the catchment
o Transferring model parameter based on the similarity of area.
o Taking overall basin average of model parameters.
Next the model parameter values at the ungauged catchments were estimated.
Finally daily water level fluctuations of Lake Tana were simulated in order to get insight
which regionalization method performs best to predict discharges from ungauged catchments.
To determine the optimum parameter set, an approach was introduced which comprehends application
of Monte Carlo simulations using a composed multiple objective function (MOF). In order to assess
the model performance, two single objective functions (SOFs) were incorporated in the MOF with
each evaluating a particular aspect of the hydrograph. The two selected SOFs are the commonly used
Nash-Sutcliffe coefficient (NS) and the Relative Volume Error (RVE).
A review of previous studies was carried out to select appropriate model parameters for calibration
since in this study it is tried to establish a robust regional model which addresses all aspects of the
hydrograph. As such efficient and effective model parameters have to be selected. Based on the
gained information, in total 8 model parameters are selected which are: FC, BETA, ALFA, LP, KF, KS,
PERC and CFLUX. Based on the sensitivity analysis on this study KS, PERC and CFLUX show not
much effect to the model performance with respect to both NS and RVE performance indicators. FC,
BETA and LP show significant effect to the model with respect to both performance indicators. Hence
in case of establishing regional model more attention was paid to the sensitive parameters.
Among the 17 river gauging stations in Lake Tana basin, 9 of them have continuous river discharge
data for 10 years from 1994 to 2003 and have areal occupancy of 40.2% of the basin. Out of these
nine gauged rivers, only six of them have representative river flow data that can be simulated with a
reasonable performance of NS value greater than 0.6 and RVE within 5%. These catchments cover
38.3% of Lake Tana basin.
57
Four regionalization methods were used to estimate the river discharge from ungauged catchments.
Those are regression method, spatial proximity method, catchment area ratio method and sub-basin
mean method. Through application of single linear regression analysis and subsequently multiple
linear regression analysis, for each model parameter a significant and relatively strong relationship
was established. Simulation of lake level shows satisfactory results for discharge predicted by
regression method with 2.17% relative volume error and 0.92 NS value. The results of water balance
simulation show gauged catchment river inflow of 1254mm/year, ungauged catchment river inflow of
527mm/year, lake areal rainfall of 1347mm/year, lake open water evaporation of 1563mm/year and
lake outflow of 1480mm/year. The closure term is obtained as 85mm/year and it is 2.7% of the total
annual lake inflow from rainfall and river inflow.
It is observed that effect of land used changes has considerable effect to the peak runoff and it is
concluded that intensive forest in a catchment largely affects reduction of flooding event.
7.2.
Recommendations
To further enhance the result of the regionalization and the subsequent lake level simulation the
following recommendations are formulated.
58
It is observed that the parameters PERC and CFLUX in this study do not show significant
effect to the model performance and it can be ignored when applying the HBV model to
another regionalization study.
The various causes of rainfall in Ethiopia lead to a wide range of rainfall distributions in
space and time. Available rainfall stations however are not well distributed to represent these
rainfall events. As such the use of remote sensing data to estimate areal rainfall should be
further explored.
Remote sensing is often claimed to be effective for calibration purposes in hydrological

modelling and more specific in stream flow modelling. In this respect to make use of observed
soil moisture by remote sensing for model calibration purposes the HBV computer code can
easily be adjusted.
A study of more advanced automatic model calibration techniques is recommended.
References
Abbott, M.B., Bathurst, J.C., Cunge, J.A., O'Connell, P.E. and Rasmussen, J., 1986. An introduction
to the European Hydrological System - Systeme Hydrologique Europeen, "SHE", 1: History
and philosophy of a physically-based, distributed modelling system. Journal of Hydrology,
87(1-2): 45-59.
Abeyou, W., 2008. Hydrological balance of Lake Tana upper Blue Nile basin, Ethiopia, ITC,
Enschede, 94 pp.
Arsano, Y. and Tamrat, I., 2005. Ethiopia and the Eastern Nile Basin. Aquatic Sciences 67(1): 15-27.
Bastola, S., Ishidaira, H. and Takeuchi, K., 2008. Regionalisation of hydrological model parameters
under parameter uncertainty: A case study involving TOPMODEL and basins across the
globe. Journal of Hydrology, 357(3-4): 188-206.
Bergstrm, S., 1995. The HBV model. In: V.P.Singh (Ed.)Computer Models of Watershed Hydrology.
Water Resources Publications, Highlands Ranch, CO.
Bergstrm, S. and Forsman, A., 1973. Development of a conceptual deterministic rainfall - runoff
model. In: Nordic hydrology, 4(1973), pp. 147-170.
Bergstrm, S. and Graham, L.P., 1998. On the scale problem in hydrological modeling. Journal of
Hydrology, 211(1-4): 253-265.
Beven, K., 1989. Changing ideas in hydrology -- The case of physically-based models. Journal of
Hydrology, 105(1-2): 157-172.
Booij, M.J., 2005. Impact of climate change on river flooding assessed with different spatial model
resolutions. Journal of Hydrology, 303: 176-198.
Booij, M.J., Deckers, D.L.E.H., Rientjes, T.H.M. and Krol, M.S., 2007. Regionalization for
uncertainty reduction in flows in ungauged basins. In: Quantification and reduction of
predictive uncertainty for sustainable water resources management : proceedings of
symposium HS2004 at IUGG2007, Perugia, 7-13 July 2007. / ed. by E. Boegh ...[et.al]
Wallingford : IAHS, 2007. ISBN 978-1-90150278-09-1(IAHS Publication ; 313) pp. 329-337.
Bras, R.L., 1990. An Introduction to Hydrologic Science. Addison-Wesley Publishing Company, 643
pp.
Conway, D., 2000. The Climate and Hydrology of the Upper Blue Nile River. The Geographical
Journal, 166: 49-62.
Conway, D., 2005. From headwater tributaries to international river: Observing and adapting to
climate variability and change in the Nile basin. Global Environmental Change Part A, 15(2):
99-114.
Crock, B.F.W. and Norton, J.P., 2000. Regionalisation of Rainfall-Runoff Models. Center for
Resorces and Environmental Studies.
Deckers, D.L.E.H., 2006. Predicting discharge at ungauged catchments, University of Twente,
Enschede, 143 pp.
Dingman, S.L., 1994. Physical hydrology. Prentice Hall, Englewood Cliffs, NJ.
Dingman, S.L., 2002. Physical hydrology Upper Saddle River. Prentice Hall.
FAO-56, FAO Irrigation and Drainage Paper-No.56.
59
Fernandez, W., Vogel, R.M. and Sankarasubramanian, A., 2000. Regional calibration of a watershed
model. Hydrological Sciences Journal, 45(5): 689-707.
Gheith, H. and Sultan, M., 2002. Construction of a hydrologic model for estimating Wadi runoff and
groundwater recharge in the Eastern Desert, Egypt. Journal of Hydrology, 263(1-4): 36-55.
Gieske, A.S.M., Abeyou, W.W., Getachew, H.A., Alemseged, T.H. and Rientjes, T.H.M., 2008. NonLinear parameterization of lake tana's flow system. Proceedings of the Workshop on
"Hydrology and Ecology of the Nile River Basin under Extreame Conditions": 127-144.
Heuvelmans, G., Muys, B. and Feyen, J., 2006. Regionalisation of the parameters of a hydrological
model: Comparison of linear regression models with artificial neural nets. Journal of
Hydrology, 319(1-4): 245-265.
Hundecha, Y. and Brdossy, A., 2004. Modeling of the effect of land use changes on the runoff
generation of a river basin through parameter regionalization of a watershed model. Journal of
Hydrology, 292(1-4): 281-295.
John, D.V., 2006. Simple version for the Penman evaporation equation using routine weather data.
Journal of Hydrology, 331: 690-702.
Kebede, S., Travi, Y., Alemayehu, T. and Marc, V., 2006. Water balance of Lake Tana and its
sensitivity to fluctuations in rainfall, Blue Nile basin, Ethiopia. Journal of Hydrology, 316(14): 233-247.
Kim, U. and Kaluarachchi, J.J., 2008. Application of Parameter Estimation and Regionalization
Methodologies to Ungauged Basins of the Upper Blue Nile River Basin, Ethiopia. Journal of
Hydrology (Under review).
Kokkonen, T.S., Jakeman, A.J., Young, P.C. and Koivusalo, H.J., 2003. Predicting daily flows in
ungauged catchments: model regionalization from catchment descriptors at the Coweeta
Hydrologic Laboratory, North Carolina. Hydrological Processes, 17: 2219-2238.
Krysanova, V., Bronstert, A. and Mller-Wohlfeil, D.-I., 1999. Modelling river discharge for large
drainage basins: from lumped to distributed approach. Hydrological Sciences Journal, 44:
313-331
Kuczera, G. and Mroczkowski, M., 1998. Assessment of hydrological parameter uncertainty and the
worth of multiresponce data. Water Resources Research, 34: 1481-1489.
Lidn, R. and Harlin, J., 2000. Analysis of conceptual rainfall-runoff modelling performance in
different climates. Journal of Hydrology, 238(3-4): 231-247.
Lindstrm, G., Johansson, B., Persson, M., Gardelin, M. and Bergstrm, S., 1997. Development and
test of the distributed HBV-96 hydrological model. Journal of Hydrology, 201(1-4): 272-288.
Madsen, H., 2000. Automatic calibration of a conceptual rainfall-runoff model using multiple
objectives. Journal of Hydrology, 235(3-4): 276-288.
Menzel, L. and Brger, G., 2002. Climate change scenarios and runoff response in the Mulde
catchment (Southern Elbe, Germany). Journal of Hydrology, 267(1-2): 53-64.
Merz, R. and Blschl, G., 2004. Regionalisation of catchment model parameters. Journal of
Hydrology, 287(1-4): 95-123.
Mwakalila, S., 2003. Estimation of stream flows of ungauged catchments for river basin management.
Physics and Chemistry of the Earth, Parts A/B/C, 28(20-27): 935-942.
Parajka, J., Merz, r. and Blschl, G., 2005. A comparision of regionalization methods for catchment
model parameters. Hydrological Earth System Science, 9: 157-171.
Reggiani, P. and Rientjes, T.H.M., 2005. Flux parameterization in the representative elementary
watershed approzch: Application to a natural basin. Water Resources Research, 41(W04013).
60
Rientjes, T.H.M., 2007. Modelling in hydrology. ITC Enschede, The Netherlands.

Seibert, J., 1999. Regionalization of parameter for conceptual rainfall-runoff model. Agricultural and
Forest Meteorology(98-99): 279-293.
SMEC, 2007. Hydrological study of the Tana-Beles sub-basin, part 1.
SMHI, 2006. Integrated Hydrological Modelling System Manual, Version 5.1.
Vallet-Coulomb, C., Legesse, D., Gasse, F., Travi, Y. and Chernet, T., 2001. Lake evaporation
estimates in tropical Africa (Lake Ziway, Ethiopia). Journal of Hydrology, 245(1-4): 1-18.
Vandewiele, G.L. and Elias, A., 1995. Monthly water balance of ungauged catchments obtained by
geographical regionalization. Journal of Hydrology, 170(1-4): 277-291.
Wagener, T. and Wheater, H.S., 2006. Parameter estimation and regionalization for continuous
rainfall-runoff models including uncertainty. Journal of Hydrology, 320(1-2): 132-154.
Wale, A., 2008. Hydrological balance of Lake Tana upper Blue Nile basin, Ethiopia, ITC, Enschede,
94 pp.
Wale, A., Rientjes, T.H.M., Dost, R.J.J. and Gieske, A., 2008. Hydrological balance of Lake Tana,
Upper Blue Nile basin, Ethiopia. Proceedings of the Workshop on "Hydrology and Ecology of
the Nile River Basin under Extreame Conditions": 159-180.
WRB, I.W.G., 2007. World Reference Base for Soil Resources 2006, 2nd Edition. World Soil
Resources Reports No. 103, FAO, Rome.
Xingnan Zhang, G.L., 1997. Development of an automatic calibration scheme for the HBV
hydrological model. Hydrological Processes, 11(12): 1671-1682.
Xu, C.Y., 1999. Estimation of Parameter of a Conceptua; Water Balance Model for Ungauged
catchments. Water Resources Management, 13: 353-368.
Xu, C.Y., 2003. Testing the transferability of regression equations derived from small sub-catchment
to a large area in cetral Sweden. Hydrological Earth System Science, 7(3): 317-324.
Yu, P.S. and Yang, T.S., 2000. Using synthetic flow duration curves for rainfall-runoff model
calibration at ungauged site. Hydrological Processes, 14: 117-133.
61
List of Acronyms
AAE
ADT
ALFA
ALFA
BETA
DEM
FAO
F-test
HBV
IAHS
ID
ITCZ
KF
KS
LP
MCS
MOF
MoWR
MP
NMA
NS
PCC
PE
PERC
PUB
RVE
SMHI
SOF
SRTM
T-test
Annual average evapotranspiration

Average daily temperature
Parameter defining the non-linearity of the quick runoff reservoir in the HBV model
Response box parameter
Parameter in soil moisture routing in the HBV model
Digital elevation model
Food and agriculture organization
Value which indicates the significance level of the regression equation
Hydrologiska Byrns Vattenbalansavdelning (Hydrological Bureau Water balance
section)
International Association of Hydrological Sciences
Inverse distance
Inter-Tropical Convergence Zone
Recession coefficient upper zone
Recession coefficient lower zone
Limit for potential evapotranspiration
Monte Carlo Simulation
Multi objective function
Ministry of water resources
Model parameter
National Meteorological Agency
Nash-Sutcliffe coefficient
Physical catchment characteristics
Potential evapotranspiration
Percolation from upper to lower zone
Prediction in ungauged basin
Relative volume error
Swedish meteorological and hydrological institute
Single objective function
Shuttle Radar Topography Mission
Statistical hypothesis test in which the test statistics has a students t distribution if the
null hypothesis is true
63
Appendix A: List of previous studies

Table A-1: A list of previous studies related to parameter estimation in ungauged basins, after Kim (2008)
PCCb
6, 10, 11
2, 6, 8
1, 3, 4, 5, 6
7, 9, 10
3, 10
--7, 9, 10
9, 11
1
1, 9, 10, 11
2, 5, 7
1, 3, 6, 7, 9, 10
9, 10
Method
Regression
Regression
Regression
Regression
Regression
Proximity, Transfer
Regression
Regression, Regional calibration
Flow duration curve
Regression
Mean, Regression, Transfer
Regression
Regression
2004 HBV
1, 2, 7, 9, 10
Merz and Blschl
2004 HBV
3, 5, 7, 9
Parajka et al.
Heuvelmans et al.
Wagener and
Wheater
Brdossy
Boughton and Chiew
2005 HBV
2006 SWAT
5, 6, 7, 9, 10, 11
6, 7, 9, 10, 11
Optimization of transfer function

Mean, Preset, Regression,
Transfer, Proximity
Mean, Proximity, Regression,
Similarity
Regression, Neural networks
2006 RRMT
2007 HBV
2007 AWBM
3, 7, 9
1, 2, 7, 9, 10
3, 4, 5, 10, 11
Kim and
Kaluarachchi
Two layer WB 1, 2, 3, 4, 5, 6
2008 model
7, 8, 9, 10
1, 2, 3, 5, 6, 7,
2008 TOPMODEL 11
Study
Jarboe and Haan
Magette et al.
Lall and Olds
Hughes
Servat and Dezetter
Vandewiele and Elias
Xu
Fernandez et al.
Yu and Yang
Yokoo et al.
Kokkonen et al.
Mwakalila
Xu
Hundecha and
Brdossy
Bastola
a
Year
1974
1976
1987
1989
1993
1995
1999
2000
2000
2001
2003
2003
2003
Modela
WYM
KWM
Mass Balance
OSE2
CREC, GR3
MWB
MWB
Abcd
HBV
TANK
IHACRES
DBM
NOPEX
Various regressions
Transfer
Regression
Global mean, Sub-basin mean,
Multiple regression,
Regional calibration, Aggregate
calibration, Volume fraction,
Lumped calibration
Multiple linear regression
Neural network
See the reference for further details

PCCs selected for the regional relationships in the references. 1=Drainage area, 2=Basin shape,
3=Precipitation, 4=Temperature, 5=Basin elevation, 6=Channel density or length, 7=Basin slope,
8=Channel slope, 9=Soil type, 10=Land use, 11=Geology.
b
64
Appendix B: Data assimilation

Precipitation, evapotranspiration and temperature are three input variables for the HBV model. These
data are gathered from National Meteorological Agency (NMA) in Ethiopia and ITC archives. The
observation stations and the observed durations are shown in Table B-1.
Table B-1: Meteorological stations and the available time series of observed data
Adet
Sekela
Dangila
BahirDar
Zege
D_Estifanos
Delgi
Aykel
Gorgora
Gondar
Enfiraz
AddisZemen
D_Tabor
N_Mewicha
Rainfall
1994-2003
1994-2003
1994-2003
1994-2003
1995-2003
1994-2003
1994-2003
1994-2003
1994-2003
1994-2003
1994-2002
1997-2006
1994-2003
1994-2003
Ayehu
Yetmen
Enjebara
1994-2003
1994-2003
1994-2006
Wind Speed
[2m]
1994-2003
RH
Sunshine
1994-2003 1994-2003
Tmax
1994-2003
Tmin
1994-2003
1994-2003 1995-2003
1994-2003 1994-2003
1994-2003
1994-2003
1995-2003
1998-2003
1994-2003
1994-2003
1995-2003
1998-2003
1994-2003
1994-2003
2001-2003
1994-2003
1994-2003
2001-2002
2004-2006
2004-2006
1994-2003
1994-2003
2004-2006
2004-2006
1994-2003
1994-2003
2004-2006
1995-2002
2001-2006
1998-2003
2001-2006
1998-2003
1994-2003 1994-1997
1994-2003 1995-2003
1994-1995,
1997,
2002-2003 2002-2003
1995-2003
Among these observation, data series were prepared for the input variable as per the following
procedures since those data sets include many missing and incorrect information.
B.1-Precipitation
Among these stations 16 stations were selected according to the availability of observed data for the
period of 1994-2003. Each data series was consisting of missing data for one day to several weeks.
Following three methods were used to calculate the missing values for each data series.
65
B-1.1 Station-average method

In this approach, P0 is computed as the simple average of the values at the nearby gauges.
p1
p0 =
1
pg
G g =1
where. p0: missing gauged reading,

p1-g: surrounding gauged stations
p2
[B-1]
p0
pG
p3
p4
Figure B - 1 Map showing gauge location
B-1.2 Normal-ratio method

In this approach the value was estimated at the gauged with missing data by weighting the
observations at the G gauges by their respective annual average values, Pg
p0 =
1 G P0
. pg
G g =1 Pg
[B-2]
where, p0: missing gauged reading, P0: the annual average precipitation at the gauge with missing
value, Pg: the annual average precipitation
B-1.3 Regression
Linear regression relation was established with the average rainfall Vs rainfall of the missing station.
p0 = 0 + 1 p
[B-3]
where. p0: missing gauged reading, 0, 1: regression constant calculated by conventional least-squares
regression method, p : average rainfall of surrounding stations.
B-1.4 Rainfall interpolation
Areal rainfall over the catchments is estimated by inverse distance weighting method. The rainfall
intensity at a point P(x,y) out of the rain gauged net work is inversely proportional to distance. The
power parameter m controls how the weighting factor reduces as the distance from the reference point
increases.
n
p=
d
i =1
n
m
i
Ps
m
i =1 d i
[B-4]
where, p : is areal average rainfall, Ps: is rainfall measured at sub region, di: is distance of station
from the region centre, m: is distance weight, n: is number of meteorological stations.
66
30
35
25
30
25
20
Temp.Dangila (0C)
Temp.Debre Tabor (0C)
B-2 Evapotranspiration and mean temperature

In order to calculate potential evapotranspiration Penman-Monteith formula was used with variables:
maximum and minimum temperature, relative humidity, wind speed and sunshine hours. Among the
available meteorological stations 6 stations were selected according to the data availability for the
duration of 1994-2003. There were large amount of missing data exist for these four variables. The
method described in section A-1.3 was used to establish a regression model for each variable, thus to
predict the missing value accordingly. Several combinations were tested to get the highest correlation
and that equation was used to calculate the missing values. Figure A-2 showing the relationship
between average temperature of surrounding stations and missing station. This experiment was done
for the other variables as well and regression model was established in order to calculate the missing
values. The outliers were detected by potting the residuals with respect to the mean of the data set.
Again these outliers were corrected according to the regression equation.
15
y = 1.0013x - 2.7183
R2 = 0.876
10
20
15
10
y = 1.0395x - 0.8125
R2 = 0.8925
0
0
10
15
20
25
30
35
10
Temp. Average (0C)
35
40
30
35
20
25
30
35
30
Temp.BahirDar (0C)
25
Temp.Adet (0C)
15
Temp.Average (0C)
20
15
y = 1.0507x + 0.0227
R2 = 0.8446
10
25
20
15
y = 0.9444x + 3.9415
R2 = 0.8694
10
5
0
0
10
15
20
25
30
35
Temp.Average (0C)
10
15
20
25
30
35
Temp.Average (0C)
Figure B - 2 Correlation between average temperature and station temperature
Mean temperature is calculated by the following equation:
Tmean =
1
(Tmax + Tmin )
2
[B-4]
B-3 Observed discharge data

An important variable for the HBV model is observed discharge data. These discharges are used to
calibrate the model parameters. These data sets also consist of missing observations. The following
procedures were applied to fill certain gaps of the observed discharge data.
B-3.1 Interpolating discharge gaps for short duration
Unlike rainfall, stream flow shows strong serial correlation; the value on one day is closely related to
the value on the previous and following days especially during periods of low flow or recession.
67
When the gaps in the records are shorts (less than 2 days), during the period of low flows, a linear
interpolation technique was used to fill gaps between the last value before the gap and the first value
after it. During this operation rainfall values were also examined to check weather there are any
significant change occurred during these days. Further the hydrograph of that station was inspected
graphically with the neighbouring station to ensure that there are no discontinuities in the flow
sequence over the gap.
B-3.2 Interpolating gaps during recession
During the period of recession the flow merely depends on surface and sub-surface storage rather than
rainfall. On this period the flow exhibits a pattern of exponential decay curve. When the gaps exist
during the long recession periods, interpolation was done between the logarithmically transformed
points before and after the gaps.
The slope of the logarithmically transformed flow recession, (also called a reaction factor) from the
last value before the gap Qt0 at time t0 to the first value after the gap Qt1 at time t1 is shown in
equation [B-5].
ln Qto ln Qt1
t1 t0
[B-5]
The reservoir coefficient k is defined as k=1/

Hence discharge Qt at time t within the gap is calculated by equation [B-6]
Qt = Qto exp(
t t0
)
k
[B-6]
B-3.3 Detecting unreliable and spurious data

Screening was done in order to identify the unreliable and spurious data in the observed discharge
time series. The visually identifiable unreliable and spurious data sets were described in section 2.2.4
in Chapter 2. Further analysis is described here to identify the outliers and unreliable individual data
points during the high rainfall period.
First the incremental or decremental differences of precipitation (P) and observed discharge (Q) is
calculated for each consecutive time steps according to the equation [B-5] and [B-6]. Then the ratio
P
Q
was plotted against the time for the entire duration (1994-2003) (Figure B-3 and Figure B-4).
P = Pt Pt 1
[B-5]
Q = Qt Qt 1
[B-6]
68
The majority of value for ratio
P
lies within certain interval (this interval may change from basin
Q
to basin) commonly the ratios are close to zero, but some points are appeared as outliers. These points
are selected and observed in the time series with respect to the rainfall events and adjusted
accordingly.
3000.00
|P|/Q
2000.00
1000.00
0.00
-1000.00
-2000.00
-3000.00
1/1/94
5/16/95
9/27/96
2/9/98
6/24/99
11/5/00
3/20/02
8/2/03
Time (days)
Figure B -3: Showing the |P|/Q ration during 1994-2003 for Ribb catchment
500.00
400.00
300.00
|P|/Q
200.00
100.00
0.00
-100.00
-200.00
-300.00
-400.00
-500.00
1/1/94
5/16/95
9/27/96
2/9/98
6/24/99
11/5/00
3/20/02
8/2/03
Time (days)
450
50
400
45
350
40
35
300
30
250
25
200
20
150
15
100
10
50
0
1-Jan-03
P (mm)
Q (m3/s)
Figure B -4: Showing the |P|/Q ratio during 1994-2003 for Gilgel Abbay catchment
1-Feb-03
1-Mar-03
1-Apr-03
1-May-03
1-Jun-03
1-Jul-03
1-Aug-03
1-Sep-03
1-Oct-03
0
1-Nov-03
Time (days)
Rainfall
Q-Observed
Q-Corrected
Figure B -5: Corrected outliers in Gilgel Abbay catchment in the year 2003
69
Appendix C: General procedure in DEM hydro processing
70
Appendix D: HBV model setup

The HBV hydrological model has a long history and the model has found applications in more than 50
countries. Its first application dates back to the early 1970s. Originally the HBV model is developed at
the Swedish Meteorological and Hydrological Institute (SMHI) for runoff simulation and hydrological
forecasting, but the scope of applications has increased steadily. For instance it has also been used for
studies about the effects of climate change. The model has also been subjected to modifications over
time, so more specific situations could be addressed.
Experience has shown however, that the standard version of HBV had some major drawbacks which
are outlined in Lindstrm et al. (1997). Therefore a re-evaluation has been carried out and a new
model version has been developed. The HBV-96 model is the final result of this model revision
(Lindstrm et al., 1997).
However, the HBV model is not used in the interface released by the SMHI. The computer language
FORTRAN and MATHLAB is used and several reasons underpin this decision. When using computer
codes, adjustments to the model can be made which are not possible in the regular interface.
Adjustments which benefits this study such as the method used for calibrating the model and
evaluating the performance of the model. Monte Carlo Simulation (MCS) was embedded in the
computer code in order to perform automatic calibration. Previously written computer code by M.J.
Booij was adjusted to suit the catchment topology and relevant analysis as the catchment are
calibrated individually and no routing routine is imbedded since the catchment are not connected.
The model consists of 6 modules, which are:
1. Precipitation accounting, representing rainfall, snow accumulation and melt;
2. Soil moisture routine, representing actual evapotranspiration;
3. Quick runoff routine, representing quick flow;
4. Base flow routine, representing slow flow;
5. Transformation function, representing slow flow delay and attenuation;
6. Routing routine, representing flow through river reaches.
In HBV model runoff is generated using three meteorological forcing such as precipitation,
temperature and potential evapotranspiration. The models basis is referred to catchment that by
themselves can be divided in a number of sub-catchments. The model is semi-distribute, since
differences can be made between areas with different altitudes and geographical zones in terms of
forest or field.
D-1 Precipitation accounting
Precipitation is one of the main inputs for the HBV model out of the three main variables in
simulating runoff processes. Other two variables are mean temperature and potential
evapotranspiration. Daily time step of data series were used for all the input variables in this study,
but if desirable even smaller time step can also be set.
71
Precipitation
Evapotranspiration
Snow
Precipitation
Accounting
Snowmelt
Ea = (SM/LP.FC)Ep [D-6]
Direct runoff
Soil moisture
routing
Infiltration
IN = P DR
[D-4]
Seepage
[D-8]
Quick flow reservoir
Quick flow
Q0
Base flow reservoir
Q0 = Kf . UZ(1+ALFA)
[D-9]
Q1 = Ks . LZ
[D-11]
Base flow
Q1
MAXBAS
LAG/DAMP
Sub basin streams
72
[D-2]
Soil moisture storage
Capillary rise
Routing
routine
Refreezing:
Pr = CFR.CFMAX (TT - T )
Water
[D-3]
Cf = CFLUX(FC-SM/FC)
Transformation
function
[D-1]
Refreezing
DR = max{(SM + P FC),0}
Base flow
routing
Snow melt:
Ps = CFMAX (T - TT )
Figure D-1: Schematisation of HBV model structure
The temperature is used for calculations of snow accumulation and melt, but when desired it can be
used to adjust the potential evaporation (Lindstrm et al, 1997; IHMS, 1997). In order to define
precipitation as rainfall or snow, a threshold value is used, TT [0C]. When temperature, T [0C],
becomes smaller than the threshold value, rainfall devolves to snow. Interaction between these two
components takes place through snowmelt and refreezing, respectively shown in equation [D-1] and
[D-2].
Ps = CFMAX (T TT )
[D-1]
Pr = CFR CFMAX (TT T )
[D-2]
-1 0
where, CFMAX: melting factor [mm d
-1
C ], CFR: refreezing factor [-]
D-2 Soil moisture routine

One of the main parts in the HBV approach is the soil moisture routine that controls the runoff
formation. Three main output components are generated from this routine. Those are direct runoff,
indirect runoff and actual evapotranspiration. For each sub-catchment the runoff is generated
independently as they have specific parameter values for soil moisture accounting procedure and
response function.
Direct runoff
The unsaturated top soil layer is represented by the soil moisture reservoir and the volume of the soil
moisture, SM[mm] in the catchment is computed using the precipitation, P[mm/d] as an input which
is supplied by the precipitation accounting procedure. Until the maximum soil moisture storage,
FC[mm], is not exceeded, the precipitation infiltrates into the soil moisture reservoir. If the maximum
soil moisture value is exceeded then precipitation will directly contribute to the runoff, DR[mm/d] as
shown in equation [D-3]:
DR = max{(SM + P FC ),0}
[D-3]
The volume of infiltrating water, IN[mm/d] is calculated from the output of equation [D-3] is shown
by the equation [D-4]
IN = P DR
[D-4]
Indirect runoff
The infiltrated water generally separated into two components. It can be replenished the soil moisture
state or it will seep through the soil layer, which is parameterized by recharge, R[mm/d]. This indirect
runoff, R through the soil layer is determined by the amount of infiltrated water, IN and the soil
moisture content, SM through a power relationship with parameter BETA[-]. This relation is shown in
equation [D-5].
73
SM
R = IN
FC
BETA
[D-5]
This relation between the parameters state that when there in no infiltration occurs, no indirect
discharge is generated. Further while increasing the soil moisture content the indirect discharge is
increase and when its come up to field capacity all the infiltration will contribute to the recharge.
Evapotranspiration
In the soil moisture routine actual evapotranspiration, Ea[mm/d] is related to the measured potential
evapotranspiration, soil moisture content and parameter LP which is a soil moisture limit lies between
0 and 1, beyond that value the actual evapotranspiration occurs at potential rate. This relation is
shown in equation [D-6] and [D-7].
SM
Ea =
E p
LP FC
Ea = E p
if
SM < (LP.FC)
[D-6]
if
SM (LP.FC)
[D-7]
D-3 Quick runoff routing

The runoff generation routing is represented as a response function which transform excess water
from the soil moisture zone to runoff. In the model structure upper non-linear and lower linear
reservoirs are included to represent this response functions. The quick flow and the slow flow are
represented by these reservoirs. The upper non-linear reservoir is managed by the quick runoff
routine. Generally three components can be distinguished in this reservoir. Those are percolation to
the slow flow reservoir, capillary rise back to the soil moisture reservoir and quick runoff.
Percolation
The direct runoff, DR and indirect discharge, R will entered together to the quick runoff reservoir
from which a specific amount transfer to the underlying base flow runoff reservoir is called
percolation, PERC[mm/d]. Percolation will only occur when there is excess water available in quick
runoff reservoir.
Capillary rise
The next component in the quick runoff routine is capillary flow, Cf through which water return again
to soil moisture routine due to capillary action in unsaturated zone. This capillary flow depends on the
amount of water stored in soil moisture reservoir. The maximum value for capillary flow,
CFLUX[mm/d] is a parameter in this model and determines the limitation for capillary flow. The
capillary flow mainly depends on the soil moisture deficit (FC-SM). There will be no capillary rise if
there is no soil moisture deficit. Otherwise, a fraction of the CFLUX will flow capillary upward. This
is shown in the equation [D-8].
FC SM
C f = CFLUX .
FC
74
[D-8]
Quick runoff
The last component is quick runoff, Q0[mm/d] which will occur when the recharge from the soil
moisture routine is higher than PERC and Cf allows and when there is excess water available in the
quick runoff reservoir. The Q0 is determined by the equation [D-9].
Q0 = K f UZ (1+ ALFA)
[D-9]
Where UZ[mm]: is the storage in the quick runoff reservoir, ALFA[-]: is a measurement of the nonlinearity of the reservoir, Kf[1/d]: is recession coefficient. The recession coefficient is formulated
according to the following equation:
Kf =
khq (1+ ALFA)

hq ALFA
[D-10]
where, parameter hq[mm/d] and khq[1/d] represent respectively a high flow rate and a recession
coefficient at a corresponding reservoir [mm].
D-4 Base flow routine
The base flow routine, which represents the slow flow from a catchment, transforms recharge water
from the quick runoff routine. This is represented by equation [D-11].
Q1 = K s LZ
[D-11]
where, Ks[1/d]: is the recession coefficient, LZ[mm]: is the water level in the reservoir.
D-5 Transformation function
The runoff generated from the response routine (Q = Q0 + Q1) is routed through a transformation
function in order to smooth the shape of the hydrograph at the outlet of the sub-basin. The
transformation function is a simple filter technique with a triangular distribution of the weights, as
shown in Figure D-2. The generated runoff of one time step is redistributed over the following days
using one free parameter (MAXBAS). A value of higher than 1 for MAXBAS will redistribute the
runoff over longer period of time. As a result, this will lead to a delay in the sub-catchments
discharge.
75
Figure D-2: Example of transformation function with MAXBAS = 5 (Seibert, 2002)
D-6 Routing routine

The routing function also combines runoff discharge for separate sub-catchment. If there is an inflow
of water from a sub-catchment this will be added to the local runoff computed by the model. The
inflow from other sub-catchments is assumed to flow through a river channel from the outlet of the
upstream sub-basin to the outlet of the current sub-basin where the local runoff is added. If there are
inflows from several other sub-catchmets, each of them is supposed to flow through its own river
channel to the outlet.
76
Appendix E: Testing significance of regression equation

E-1 Null-hypothesis
In null-hypothesis, f-test is used to assess the significance of the total regression equation.
The general multiple regression equation is shown in equation [E-1].
Y ' = 0 + 1 X 1 + 2 X 2 + 3 X 3 + ... p X p +
[E-1]
where Y : the estimated dependent variable by the regression equation, 0 : the intercept, 1-p : the
regression weights, : error term.
H0,general: 1 = 2 = 3 = = p = 0
Ha,general: at least one regression weight is not equal to zero
H0,general is rejected when
F > F1-,p-1,n-p
F=
where,
MS reg
MS res
, F1-,p-1,n-p : can be determine from the F-table.
The mean sums of squared for regression and residual are calculated by dividing the sums of squares
through the degree of freedom.
MS reg =
SS reg
MS res =
SS res
n p
p 1
[E-2]
[E-3]
The equations for calculating the sum of squares for regression and residuals are shown respectively
by [E-4] and [E-5], n: number of observations, p: number of regression coefficient
n
SS reg = (Y j 'Y ) 2
[E-4]
j
n
SS res = (Y j Y ' j ) 2
[E-5]
where, SSreg: sum of squares for regression, SSres: sun of squares for residuals, Yj: estimated value by
the regression equation belong to catchment j, Yj: observed value belongs to catchment j, Y : average
of all observed values. These statistical characteristics are presented in an ANOVA (analysis of
variance) table.
77
In determining F1-,p-1,n-p for F-table it depends on which is applied significance level and the degree
of freedom, df regarding the residual and the regression equation. The number of independent
variables attributed in the regression equation will affect this value.
E-2 Specific hypothesis
In specific hypothesis t-test, based on the t-distribution is used to test each of the independent
variable, weather they significantly contribute to the total regression equation. A t-distribution occurs
when the number of observations is smaller than 30. In this study these conditions are satisfied and
the specific hypothesis can be stated as follows:
H0,specific: 1 = 0
H0,specific: 2 = 0
H0,specific: 3 = 0
.
.
H0,specific: p = 0
If tc >
SE
H0,specific is rejected for the specific regression coefficients:
where, tc: critical t-value determined from t-table (Davis,2002) depending on the degree of freedom, df
and chosen significance level, , SE: standard error which can be taken from the ANOVA table.
E-3 Strength
With respect to the data set, the goodness of fit of the regression equation is measured by r2, called
coefficient of determination is shown in equation [E-6].
r2 =
SS reg
SS tot
[E-6]
where, SSreg: sum of squared for regression [E-4], SStot, the total sum of squares is taken from the
ANOVA table and calculated from equation [E-7].
n
SStot = (Y j Y ) 2
[E-7]
where, SStot: total sum of squares for regression, Yj: observed value belongs to catchment j, Y :
average of all observed values.
When r2 is approaching a value of 1, the regression line is a good estimator of the data. Then the
regression equation explains 100% of the total variance present in the data set.
78
Appendix F: Albedo calculation from MODIS images

Surface Energy Balance System (SEBS) in ILWIS 3.4 Open source provides a set of routine for biographical parameter extraction with the use of satellite earth observation data. These routine derived
the evaporation fraction, net radiation and soil heat flux parameters etc, in combination with
meteorological information as inputs. This routine was used in this study to derive the albedo of the
lake in order to calculate the open water evaporation in water balance study.
F-1 Image acquiring
Terra or Aqua MODIS images can be acquired from Level 1 and Atmosphere Archive and
Distribution system (LAADS Web) products search website. (http://ladsweb.nascom.nasa.gov/data/
search.html). In this study MODIS Level 1B Calibrated Radiances 250m, 500m and 1km cloud free
products and MOD03 Geolocation 1km products for the same day were selected and acquired for the
year 2000 and 2002 over the study area. Terra products were selected by concerning the image
capturing time and the position with respect to the location of the study area.
F-2 MODIS Level-1B data conversion
Here, ENVI software was used to process the MODIS level-1 products, which were received as HDF
file format, into the GeoTIFF format ready for ILWIS importing. First select the desired band to
process. By default all the bands in the input file are selected. But in this case we wish to extract band
1 to 7 and Band 31 and 32. After that the Geolocation file was selected and process in order to convert
the satellite and solar zenith and azimuth angle bands and height band into GeoTIFF format. Next use
the GDAL file-import tool in ILWIS to import the GeoTIFF files to ILWIS. While processing in
ENVI, the raw data will converted to reflectance in this case and no need to do pre processing in
ILWIS again.
Note: If one uses ModisSwath Tool to process the level-1 product file, the pre-processing routine
should be applied to convert the raw to radiance or reflectance. The required calibration coefficient
(scale and offset) provided in the HDF header file can be read with the help of HDFView software.
F-3 Atmospheric correction
SMAC tool in SEBS routine in ILWIS is used to correct the atmospheric effect for visible and near
visible bands of MODIS images. Figure F-1 shows the required input data for atmospheric effect
correction in SMAC. Derivation of atmospheric correction factors are described bellow.
Optical thickness:
The option button can be checked on atmospheric effect correction dialog box if the map of optical
thickness (depth) value is available. Otherwise a value can be entered that will be applied to all pixels.
The Level 2.0 quality assured Aerosol Optical Thickness (AOT) data can be obtained from the
AERONET (AErosol RObotic NETwork) (http://aeronet.gsfc.nasa.gov/new_web/aerosols.html) web
site by selecting AOT in AERONET data type. First nearest site to the study area should be selected
on the map and then choose the image acquisition date. In this study no site is situated close to the
Lake Tana basin, hence a average value was obtained from the surrounding stations (Nairobi, Kibale,
DMN_Maine_Soroa and Bodele). Further the records are not available for the particular dates of
image acquired on 2000 and the 2002. Therefore average yearly value was computed. In the SMAC
79
algorithm a value of AOT is used at wave length of 550nm. But in the records AOT is not provided
for a wavelength of 550nm. By plotting the given AOT value (see Figure F-2) against wavelength and
by polynomial fitting (see Figure 3) required AOT for wavelength of 550nm can be calculated. The
unit of the value is in m, and the value range is between 0.05 and 0.8. The estimate value was 0.31
m for this study.
Figure F-1: Input data for atmospheric effect correction in SMAC tool
Water vapour content:

The option button can be checked if one wishes to use a map of water pressure content value.
Otherwise one value can be given and that will be applied to all the pixels. These values also can be
obtained from the above described AERONET web site by selecting water vapour in AERONET
data type and choosing the image acquisition date. In this study the average value was computed by
selecting surrounding site as there is no nearest site to the study area. The unit of the value is grams
cm-2, and the value range is between 0 and 6. The estimated value for this study is 2.3 grams cm-2.
Ozone content:
The option button can be checked if one wishes to use a map containing ozone content value.
Otherwise one value can be provided that will be applied to entire image. This value can be obtained
by using data from the Ozone Monitoring Instrument (OMI) on the Aura spacecraft at NASA Total
Ozone Mapping Spectrometer web site just entering the latitude and longitude and date.
(http://jwocky.gsfc.nasa.gov/teacher/ozone_overhead_v8.html).
80
Figure F-2: Average aerosol optical thickness value for year 2006 in Nairobi
0.3
0.25
y = 24.914x -0.805
R2 = 0.9864
AOT
0.2
0.15
0.1
0.05
0
0
200
400
600
800
1000
1200
Wave length [nm]

2006
Power (2006)
Figure F-3: The plotting of the AOT measured at different wavelength and polynomial fitting
Surface pressure:
The option button can be checked if one wishes to use a pressure map. Otherwise one value can be
provided that will be applied to entire image. The pressure can be calculated according to the
following equation which is employed in FAO 56.
293 0.0065 z
P = 101.3
293
5.26
where P: is the atmospheric pressure [kPa], z: is the elevation above sea level [m]
81
Solar zenith and azimuth angle map and sensor zenith and azimuth angle map can be acquired with the
respective MODIS images. With the use of above factors atmospheric correction can be done for band
1 to 7.
F-4 Land surface albedo computation
The procedure for land surface albedo calculation is as followed:
Select Operations, SEBS Tools, Pre-processing, Land surface albedo computation. (see Figure
F-4)
Select sensor type MODIS
Provide the input bands (the atmospheric corrected map in the previous operation)
Execute the operation by giving the proper output file name.
The albedo over the Lake Tana was calculated for year 200 and 2002 by this procedure (see Figure F5)
Figure F-4: Input data for land surface albedo computation

0.2
0.18
0.16
Albedo
0.14
0.12
0.1
0.08
0.06
0.04
0.02
0
0
50
100
150
200
250
Time (days)
Average
2000
2002
Figure F-5: Albedo variation over the Lake Tana
82
300
350
Recession coefficient upper zone [T-1]
Recession coefficient lower zone [T-1]
KS
Percolation from upper to lower [LT ]
-1
0.0 - 2
0.1 - 2.5
0.0005 - 0.15
100 - 800
1-4
0.1 - 1
0.1 - 3
0.0005 - 0.15
Prior range
Min
Max
Gilgel Abay
Max
Gumara
Min
Max
Megech
Min
Max
Koga
Min
Max
Kelti
Min
Max
Gumero
Min
Max
Garno
Min
Max
Gelda
Min
Table G-2: Posterior parameter spaces for respective catchment for selected calibration parameter
100.0 798.8 135.5 799.9 107.1 799.3 100.0 796.3 101.2 798.9 100.3 799.9 152.6 798.7 104.6 800.0 100.1 695.0
1.000 3.994 1.005 3.998 1.001 3.997 1.004 3.995 1.001 4.000 1.003 3.995
1.00
4.00
1.01
4.00
1.00
4.00
0.100 0.998 0.103 0.994 0.102 1.000 0.101 1.000
0.10
1.00
0.10
1.00
0.10
1.00
0.10
1.00
0.10
1.00
0.101 2.307 0.101 1.964 0.100 2.991 0.106 3.000
0.10
2.99
0.11
3.00
0.10
2.54
0.10
2.97
0.10
1.81
0.0005 0.1499 0.0005 0.1498 0.0005 0.1498 0.0005 0.1500 0.0006 0.1498 0.0008 0.1499 0.0005 0.1499 0.0006 0.1500 0.0006 0.1498
0.0008 0.1500 0.0007 0.1498 0.0009 0.1500 0.0005 0.1496 0.0008 0.1500 0.0014 0.1500 0.0005 0.1499 0.0009 0.1499 0.0012 0.1500
0.102 2.498 0.101 2.498 0.104 2.500 0.104 2.498 0.106 2.498 0.103 2.498 0.101 2.499 0.106 2.499 0.100 2.500
0.003 1.999 0.000 1.997 0.002 1.999 0.001 1.999 0.003 1.997 0.001 2.000 0.006 1.996 0.002 2.000 0.001 1.993
Max
Ribb
Min
0.1
0.5
0.01
200
1
1
0.9
0.09
Default value
FC
BETA
LP
ALFA
KF
KS
PERC
CFLUX
Parameter
Table G-1: Selected model parameter and initial parameter space of each parameter
CFLUX Maximum value of capillary flow [LT-1]
PERC
Maximum soil moisture content [L]

Parameter in soil routine [-]
FC
BETA
LP
ALFA
KF
Limit for potential evapotranspiration [LT-1]

Response box parameter [-]
Description and unit
Name
Appendix G: Adjusted parameter space
83
Appendix H: Weight of Meteorological station for Gauged and

Ungauged catchments
Table H-1: Weights of rainfall stations by inverse distance squired interpolation for gauged catchments
Addis
Zemen
Adet
Bahir Dar
Chewahite
Debre
Tabor
Dangila
Enfiraz
Gondar
Sekela
Zege
0.16
0.17
Gorgora
0.55
0.41
0.29
0.15
0.27
0.38
0.15
0.44
0.18
0.85
0.94
0.06
0.06
0.94
0.27
0.12
Gumara
Enjebara
Gelda
Koga
Gumero
Garno
Kelti
Megech
G.Abay
0.24
Ribb
0.43
0.30
0.88
0.76
Table H-2: Weights of rainfall stations by inverse distance squired interpolation for ungauged catchments
Addis
Zemen
Bahir Dar
Chewahite
Deke
Istifanos
Debre Tabor
Delgi
Enfranz
Gondar
0.2
Zege
0.41
0.27
Gorgora
Ug_Ribb
Ug_G.Abay
0.39
0.21
0.52
Ug_Garno
1.0
0.37
Ug_Gumara
0.68
Ug_Gelda
0.32
0.29
0.34
Ug_Gumero
0.26
0.76
Ug_Dema
0.23
Ug_Megech
0.45
0.29
0.58
0.19
0.24
Ug_Tanawest
1.0
Ug_Gabikura
0.82
0.18
84
Table H-3: Weights of ET stations by inverse distance interpolation for gauged catchments
Adet
Gelda
Koga
0.45
Bahir Dar
Debre
Tabor
0.65
0.35
0.28
Dangila
Gondar
0.27
Gumero
0.25
0.75
Garno
0.4
0.6
1.00
Kelti
1.00
Megech
G.Abay
Nefase
Mewicha
0.40
Gumara
0.60
0.17
Ribb
0.65
0.17
0.80
0.20
Table H-4: Weights of ET stations by inverse distance interpolation for ungauged catchments
Adet
Ungauged Ribb
Ungauged Gilgel.Abbay
0.27
Bahir Dar
Debre
Tabor
0.4
0.6
0.43
Ungauged Garno
Dangila
Gondar
0.3
0.47
Ungauged Gelda
0.78
0.22
Ungauged Gumara
0.63
0.37
0.53
Ungauged Gumero
1.0
Ungauged Megech
1.0
Ungauged Dema
1.0
Ungauged Tanawest
Ungauged Gabikura
0.49
0.51
1.0
85
Appendix I: Result of sensitivity analysis

0.9
NS
0.85
0.8
0.75
0.7
0.65
-125
-100
-75
-50
FC
BETA
-25
LP
0
ALFA
25
KF
50
KS
PERC
75
100
125
100
125
CFLUX
Figure I-1: Ssensitivity of NS value in Gilgel Abbay catchment

0.850
NS
0.750
0.650
0.550
0.450
0.350
0.250
0.150
-125
-100
-75
FC
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
PERC
Figure I-2: Sensitivity of NS value in Gumara catchment
86
75
CFLUX
0.8
NS
0.75
0.7
0.65
0.6
0.55
-125
-100
-75
FC
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
75
PERC
100
125
CFLUX
Figure I-3: Sensitivity of NS value in Ribb catchment

0.75
NS
0.65
0.55
0.45
0.35
0.25
0.15
-125
-100
-75
FC
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
PERC
75
100
125
CFLUX
Figure I-4: Sensitivity of NS value in Kelti catchment
87
25
RVE
20
15
10
-5
-10
-15
-20
-125
-100
-75
-50
FC
BETA
-25
LP
25
ALFA
KF
50
KS
75
PERC
100
125
CFLUX
Figure I-5: Sensitivity of RVE value in Gilgel abbay catchment

40.00
RVE
30.00
20.00
10.00
0.00
-10.00
-20.00
-30.00
-125
-100
-75
FC
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
PERC
Figure I-6: Sensitivity of RVE value in Gumara catchment
88
75
CFLUX
100
125
50
RVE
40
30
20
10
-10
-20
-30
-40
-125
-100
-75
FC
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
PERC
75
100
125
100
125
CFLUX
Figure I-6: Sensitivity of RVE value in Ribb catchment

70
RVE
60
50
40
30
20
10
-10
-20
-30
-125
-100
-75
FC
-50
BETA
-25
LP
0
ALFA
25
KF
50
KS
PERC
75
CFLUX
Figure I-6: Sensitivity of RVE value in Kelti catchment
89
Appendix J: The effect of parameter sensitivity

-77%
400
-77%8
0
30
60
90
120
0
20
300
40
200
60
100
0
-31%
400
30
60
90
120
0
20
40
200
60
0%
400
60
P (mm)
30
200
90
100
0
120
15%
400
100
0% 8
300
80
0
20
40
4
60
2
80
100
15%8
30
60
90
120
0
20
300
40
200
60
100
0
61%
400
300
80
100
61%8
30
60
90
0
20
40
200
60
100
0
120
84%
400
300
80
100
84%8
30
60
90
120
0
20
40
200
100
0
Apr-94
Jul-94
Oct-94
Jan-95
Apr-95
Jul-95
Oct-95
60
Apr-94
80
100
Jul-94
Oct-94
Jan-95
Apr-95
Jul-95
Oct-95
Time (days)
Time (days)
Qobs
Qsim
ETp
ETa
Figure J-1: Sensitivity of FC for Gilgel Abbay catchment. Left hand side shows variation of simulated
hydrograph at each deviation of FC. Right hand side shows the variation of Actual evaporation at each
deviation of FC.
90
P (mm)
ETp and ETa (mm)
100
Q (m 3/s)
100
-31%8
300
80
-76%
0
400
-76% 25
20
300
20
40
15
60
200
10
80
100
100
120
-66%
0
-66% 25
0
400
20
300
20
40
15
60
200
10
80
100
100
120
-52%
0
400
-52% 25
20
300
20
40
15
60
200
10
80
100
0
120
-28%
0
-28% 25
0
400
20
40
60
200
80
100
20
R (mm)
300
P (mm)
Q (m 3/s)
100
120
0%
0
400
20
300
40
60
200
10
5
100
15
0
0%
25
20
15
10
80
100
100
120
20%
0
400
20
300
40
5
0
20% 25
20
15
60
200
10
80
100
100
120
44%
0
400
20
300
40
5
0
44% 25
20
15
60
200
80
100
100
120
68%
0
400
300
5
0
68% 25
20
20
40
15
60
200
10
10
80
100
100
120
Apr-94
Jul-94
Oct-94
Jan-95
Apr-95
Qobs
Oct-95
0
Apr-94
Jul-94
Oct-94
Jan-95
Apr-95
Jul-95
Oct-95
Time (days)
Time (days)
P
Jul-95
Qsim
Recharge
Figure J-2: Sensitivity of BETA for Gilgel Abbay catchment. Left hand side shows variation of simulated
hydrograph at each deviation of BETA. Right hand side shows the variation of recharge at each deviation of
BETA.
91
-84%
-84% 8
0
400
20
300
20
40
40
4
60
200
60
80
100
80
100
120
-52%
20
300
100
-52% 8
0
400
0
20
40
40
4
60
200
60
80
100
80
100
120
-20%
100
-20% 8
0
400
20
0
20
40
300
40
4
60
200
60
80
100
80
100
0%
400
40
60
200
80
100
100
P (mm)
20
300
0
400
0
20
40
4
60
2
80
100
26% 8
20
300
100
120
0
26%
0
20
40
40
4
60
200
60
80
100
80
100
120
59%
0
400
20
300
100
59% 8
0
20
40
40
4
60
200
60
80
100
80
100
0
100
120
75%
0
400
75% 8
20
300
0
20
40
40
4
60
200
60
80
100
100
120
96%
96%
80
100
0
400
20
300
20
40
40
60
200
80
100
4
60
2
80
100
120
Apr-94
Jul-94
Oct-94
Jan-95
Apr-95
Jul-95
Oct-95
0
Apr-94
100
Jul-94
Oct-94
Jan-95
Qobs
Jul-95
Oct-95
Time (days )
Time (days)
P
Apr-95
Qsim
ETp
ETa
Figure J-3: Sensitivity of LP for Gilgel Abbay catchment. Left hand side shows variation of simulated
hydrograph at each deviation of LP. Right hand side shows the variation of Actual evaporation at each
deviation of LP.
92
P (mm)
120
0
ETp and ETa (mm)
0
0%
Q (m 3/s)
-88% 20
-88%
400
20
15
300
40
10
60
200
80
100
100
120
-59%
-59% 20
0
400
20
300
15
40
200
100
60
10
80
100
120
-38%
-38% 20
0
400
20
300
15
40
10
60
200
80
100
100
0
120
-17%
0
400
20
40
60
200
80
100
15
Qf,Qs (mm)
300
P (mm)
Q (m 3/s)
-17% 20
10
5
100
0
120
0%
0%
20
0
400
20
300
15
40
10
60
200
80
100
100
120
24%
5
0
24% 20
0
400
20
300
15
40
10
60
200
80
100
100
120
65%
5
0
65% 20
0
400
20
300
40
15
10
60
200
80
100
100
120
96%
5
0
96% 20
0
400
20
300
40
10
60
200
80
100
100
0
Apr-94
15
120
Jul-94
Oct-94
Jan-95
Apr-95
Jul-95
Oct-95
5
0
Apr-94
Jul-94
Oct-94
Jan-95
Apr-95
Jul-95
Oct-95
Time (days)
Time (days)
P
Qobs
Qsim
Fast flow
Slow flow
Figure J-4: Sensitivity of ALFA for Gilgel Abbay catchment. Left hand side shows variation of simulated
hydrograph at each deviation of ALFA. Right hand side shows the variation of fast flow and slow flow at each
deviation of ALFA.
93
-87%
-87% 20
0
400
20
300
40
200
60
15
10
80
100
100
120
5
0
-74% 20
-74%
0
400
20
300
15
40
60
200
10
80
100
100
120
0
-61% 20
-61%
0
400
20
300
15
40
60
200
10
80
100
100
120
0
-36%
-36% 20
0
40
60
200
80
100
100
120
Qf,Qs (m m )
20
300
P (mm)
Q (m 3/s)
400
15
10
5
0
0%
0
400
20
300
40
60
200
0%
20
15
10
80
100
100
120
0
29%
0
400
29% 20
20
300
40
60
200
15
10
80
100
100
120
54%
0
400
5
0
54% 20
20
300
40
60
200
15
10
80
100
100
120
80%
0
400
5
0
80% 20
20
300
40
60
200
80
100
15
10
5
100
120
Apr-94
Jul-94
Oct-94
Jan-95
Apr-95
Qobs
Oct-95
0
Apr-94
Jul-94
Oct-94
Jan-95
Apr-95
Jul-95
Oct-95
Time (days)
Time (days)
P
Jul-95
Qsim
Fast flow
Slow flow
Figure J-4: Sensitivity of KF for Gilgel Abbay catchment. Left hand side shows variation of simulated
hydrograph at each deviation of KF. Right hand side shows the variation of fast flow and slow flow at each
deviation of KF.
94
0.28
0.48
HI
-0.53 -0.66 -0.90
0.60
0.48
0.22
-0.22
-0.20 -0.28
-0.72 -0.84 -0.42 -0.10 -0.11
-0.79 -0.68 -0.67 -0.88 -0.85
0.18
0.35
0.45
0.14
-0.48 -0.73 -0.37
CROPM
GL
URBAN
FOREST
LEP
NIT
VER
LUV
SAAR
PWET
PDRY
PET
0.11
-0.49 -0.27 -0.16
0.03
0.12
0.04
-0.32
0.27
0.39
1.00
1.00
-0.39
0.58
0.45
1.00
-0.04 -0.68
-0.21 -0.66
0.10
0.02
-0.92 -0.66
-0.91 -0.55
-0.33 -0.45
0.68
-0.20
0.76
0.87
-0.14 -0.50 -0.54 -0.12 -0.78 -0.30
0.26
0.12
-0.05 -0.11 -0.61 -0.24 -0.83 -0.73
0.40
0.24
-0.08
0.06
0.34
0.14
0.22
0.19
1.00
1.00
0.93
0.98
1.00
0.84
1.00
1.00
-0.78 -0.64 -0.77 -0.34
0.46
0.66
0.61
1.00
PET
0.35
-0.19
0.31
0.79
-0.51
-0.29 -0.54 -0.65
-0.21 -0.41
0.19
0.46
1.00
PDRY
-0.01 -0.28 -0.48 -0.49 -0.39 -0.17
0.16
0.01
-0.51 -0.36 -0.25 -0.05 -0.03
0.45
-0.15
0.00
1.00
PWET
0.46
0.00
-0.13 -0.36 -0.54 -0.37
-0.82
0.31
0.23
0.06
0.03
0.54
VER
0.27
-0.62
0.21
-0.29 -0.62
0.06
1.00
NIT
0.40
0.13
0.87
0.25
-0.75
SAAR
0.31
0.54
0.03
-0.15
0.62
-0.04 -0.95
LUV
0.33
0.65
-0.62
0.65
0.41
0.66
LEP
0.06
0.28
0.90
0.66
0.57
FOREST
0.04
0.21
0.23
0.01
URBAN
-0.06
0.79
0.87
1.00
GL
0.74
1.00
CROPM
0.61
1.00
-0.44
-0.61 -0.73 -0.10
0.57
CROPD
0.03
0.62
0.23
CROPD
0.44
DD
0.68
0.62
1.00
DD
0.72
-0.37
0.00
-0.16
SHAPE
0.42
0.57
AVGSLOPE 0.67
0.83
AVGSLOPE
SHAPE
1.00
HI
0.35
1.00
MDEM
0.84
0.90
0.80
LFP
1.00
LFP
MDEM
1.00
AREA
AREA
Appendix K: Correlation of catchment characteristics
95
Gauged
Catchments
Ribb
G.Abay
Gumara
Megech
Koga
Kelti
Ungauged
Catchments
Ribb
G.Abay
Gumara
Megech
Gumero
Garno
Gelda
Dema
Tana West
Gabi Kura
96
CROPM
Grass land [%]

CROPD
Forest [%]
DD
Elongation ratio [-]

EL
Circularity index [-]

CI
Catchment shape [-]

SHAPE
Average slope of
catchment [%]
AVGSLOPE
Hypsometric integral [-] HI
DEM mean [m]
Catchment area [km2]
MDEM
Longest flow path [km] LFP
AREA
0.46
0.41
0.48
0.46
0.47
0.46
0.47
0.49
0.49
0.48
Crop [%]
2264
2166
1920
2234
2230
2338
2093
2142
2038
2003
GL
24.97
79.96
17.87
36.09
44.74
37.64
42.81
45.87
12.81
26.21
Bare land [%]
736
2072
287
437
588
463
391
325
546
427
URBAN
24.91
18.98
16.67
18.04
27.87
36.06
17.95
15.96
21.39
16.09
31.15
19.93
16.60
46.75
46.55
63.42
34.42
40.94
22.03
21.88
40.55
38.69
39.22
37.20
30.49
24.58
35.42
42.55
52.46
35.39
1.36
1.21
1.25
1.75
1.03
0.85
1.27
2.26
4.69
4.80
461.27
389.33
380.00
457.42
395.47
348.10
355.90
397.78
365.77
354.63
73.71
76.60
80.46
97.91
71.85
72.90
60.73
97.84
66.57
95.85
16.06
19.20
19.49
2.06
28.14
7.91
38.93
2.14
33.16
4.13
8.74
0.04
0.03
0.00
0.00
18.62
0.00
0.00
0.00
0.00
0.02
0.12
0.02
0.03
0.01
0.04
0.33
0.02
0.02
0.02
1.47
4.03
0.00
0.00
0.00
0.54
0.00
0.00
0.25
0.00
9.86
22.15
0.00
18.88
48.65
64.86
0.46
22.64
0.00
1.22
0.00
3.13
0.00
0.00
3.65
9.51
0.00
0.00
0.00
0.00
13.37
13.26
23.00
47.66
35.17
19.20
1.44
68.57
19.24
64.38
23.94
53.53
54.73
13.69
10.58
1.40
97.52
6.01
34.20
6.99
36.84
55.86
87.21
5.03
47.61
91.55
Woody savannah [%]

0.00
1.95
3.27
3.34
14.67
8.25
FOREST
0.30
0.51
0.00
9.25
7.69
0.00
Build up area [%]

38.58
0.00
9.23
81.59
4.11
0.20
LEP
4.96
0.00
0.45
0.00
6.11
0.00
Leptosol area [%]
0.06
0.08
0.08
0.16
0.00
0.01
NIT
12.79
0.00
3.86
0.07
0.03
0.00
Nitosol area [%]
30.66
25.66
31.34
10.23
24.25
0.00
1210
1486
1368
1162
1120
999
1414
1131
819
992
1395
1750
1415
1117
1542
1585
Standard annual average

SAAR
rainfall [mm]
51.53
74.26
64.27
89.54
69.61
99.99
VER
444.38
428.28
420.27
422.69
397.48
431.05
Vertisol area [%]
1.55
1.08
1.50
1.24
1.75
1.39
LUV
38.49
30.07
34.00
23.78
46.68
33.48
Luvisol area [%]
61.73
40.33
53.36
48.16
70.16
31.24
8.15
9.86
9.13
7.16
7.39
6.96
9.57
7.04
5.33
6.52
9.16
10.61
9.21
7.09
9.55
9.74
0.84
1.12
0.98
0.96
0.89
0.64
1.02
0.87
0.67
0.79
1.10
1.84
1.16
1.00
1.51
1.60
PWET
44.59
36.76
33.68
37.28
23.48
20.00
Mean precipitation wet

season (Jun. to Sep.)
[mm]
0.48
0.48
0.48
0.50
0.43
0.46
PDRY
2915
2676
2717
2415
2429
2229
Mean precipitation dry

season (Oct. to May)
[mm]
97.67
81.58
84.28
43.92
47.64
62.48
1311
1352
1386
1359
1340
1498
1498
1498
1444
1498
1300
1332
1307
1442
1349
1305
Mean annual
PET
evapotranspiration [mm]
1408
1657
1281
531
298
608
Appendix L: Physical catchment characteristics

Perera PDF

Uploaded by

Copyright:

Available Formats

Perera PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Perera PDF

Uploaded by

Copyright:

Available Formats

Ungauged Catchment Hydrology:

The case of Lake Tana Basin

B.Upul Janaka Perera

Ungauged Catchment Hydrology: The case of Lake

Thesis Assessment Board

WREM Department, ITC, Enschede

INTERNATIONAL INSTITUTE FOR GEO-INFORMATION SCIENCE AND EARTH OBSERVATION

I dedicated this thesis to my father,

Upul Janaka Perera

Study area and data preparation..................................................................................................5

Hydrological modelling and regionalization .............................................................................15

Results and discussion .................................................................................................................31

5.3. Sensitivity of model parameters........................................................................................... 38

Simulating lake water balance................................................................................................... 49

Conclusion and recommendation .............................................................................................. 57

Lake level simulation based on bathymetric relation derived from Pietrangeli

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Problem definition and importance of the study

A better understanding of the hydrological characteristics of different watersheds in the headwaters of

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Scope of the research

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Previous studies and applications

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Outline of the Thesis

This thesis contains seven chapters and is organized as follows:

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

2. Study area and data preparation

Figure 2-1: Study Area

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Station Elevation (m,amsl)

Figure 2-4: Annual average rainfall distribution of Lake Tana basin

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Mean daily Temperature

Jul Aug Sep Oct Nov Dec

Jan Feb Mar Apr May Jun

Mean Temperature (C)

Mean Temperature ( 0C)

Figure 2-8: ADT distribution over the Lake Tana basin

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

reference evapotranspiration [mm day-1]

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Figure 2-9: Potential evapotranspiration distribution in Lake Tana basin

Analysing hydrological data

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Gilgel Abay River

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Delineation of gauged and ungauged catchments

Figure 2-13: Gauged and ungauged catchments in Lake Tana basin

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

Ungauged Gilgel Abbay

Ungauged Gagi Kura

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN

3. Hydrological modelling and regionalization

Classification of hydrological model

UNGAUGED CATCHMENT HYDROLOGY: THE CASE OF LAKE TANA BASIN