FULLTEXT01

HYDROLOGICAL MODELING AS A TOOL
FOR SUSTAINABLE WATER RESOURCES

MANAGEMENT: A CASE STUDY OF THE
AWASH RIVER BASIN
Selome M. Tessema
May 2011
TRITA-LWR.LIC 2056
ISSN 1650-8629
ISRN KTH/LWR/LIC 2056-SE
ISBN 978-91-7501-021-2
Selome M. Tessema TRITA LWR.LIC 2056
© Selome M. Tessema 2011
Licentiate Thesis
Environmental Management and Assessment Research Group
Department of Land and Water Resources Engineering
Royal Institute of Technology (KTH)
SE-100 44 STOCKHOLM, Sweden
Reference to this publication should be written as: Tessema, SM (2011) Hydrological

modeling as a tool for sustainable water resources management: a case study of the Awash
River basin. TRITA LWR.LIC 2056
ii
Hydrological modeling as a tool for sustainable water resources management: case study of the Awash River Basin
A BSTRACT
The growing pressure on the world‘s fresh water resources is enforced by population
growth that leads to conflicts between demands for different purposes. A main concern on
water use is the conflict between the environment and other purposes like hydropower,
irrigation for agriculture and domestic and industry water supply, where total flows are
diverted without releasing water for ecological conservation. As a consequence, some of
the common problems related to water faced by many countries are shortage, quality dete-
rioration and flood impacts. Hence, utilization of integrated water resources management
in a single system, which is built up by river basin, is an optimum way to handle the ques-
tion of water. However, in many areas, when planning for balancing water demands major
gaps exist on baseline knowledge of water resources. In order to bridge these gaps, hydro-
logical models are among the available tools used to acquire adequate understanding of the
characteristics of the river basin. Apart from forecasting and predicting the quantity and
quality of water for decision makers, some models could also help in predicting the impacts
of natural and anthropogenic changes on water resources and also in quantifying the spatial
and temporal availability of the resources. However, main challenges lie in choosing and
utilizing these models for a specific basin and managerial plan. In this study, an analysis of
the different types of models and application of a selected model to characterize the Awash
River basin, located in Ethiopia, is presented. The results from the modeling procedure and
the performance of the model are discussed. The different possible sources of uncertainties
in the modeling process are also discussed. The results indicate dissimilar predictions in
using different methods; hence proper care must be taken in selecting and employing avail-
able methods for a specific watershed prior to presenting the results to decision makers.
Key words: Hydrological characteristics; Distributed hydrological modeling;

Rainfall-runoff correlation; Streamflow prediction; Sustainable water management;
Awash River basin.
iii
iv
A CKNOWLEDGMENTS
I would like to acknowledge all organizations and people who have supported me in one
way or another to produce this thesis. The research was supported by the Swedish
International Development Cooperation Agency (SIDA), Department for Research
Cooperation and StandUp for Energy.
First, I would like to express my sincere respect to my former supervisor Bijan Dargahi for
giving me the opportunity to start my research work and for introducing me to his invalua-
ble expertise. My thanks also go to Anders Wörman for his advice and all the staff in the
division of Hydraulic Engineering, for all the support and the interesting chats in the little
‗fika‘ room. I would then like to express my utmost thanks and gratitude to my main
supervisor Berit Balfors for her kindness and critical thinking, my co-supervisor Ulla
Mörtberg for all the support and her encouragement to make me think out of the box. I
would like to extend my gratitude and thanks to my co-supervisor Shimelis Setegn for
providing me data and the useful comments on the manuscripts. Further, my special thanks
go to Muluneh Mekonnen, Visiting Scientist at National Hydrology Research
Institute, Environment Canada, for his comments on the manuscripts and his continuous
support in every way he can.
I am very grateful to Britt Chow and Aira Saarelainen for all their heartfelt concerns to
make me feel comfortable and their unconditional effort to keep me smile throughout my
stay in KTH. Thank you Jerzy for all your kindness and effort to fix my never ending
computer problems! I would like to extend my thanks to all colleagues at the department of
LWR who have supported me with useful advice and also help me in practical issues. My
special appreciation and thanks goes to those who have been participating in resolving the
ups and downs occurred in my journey and giving me important tips, above all
Joanne Fernlund. Your concern encouraged me to come this far.
Thank you my dear friends here and abroad for your support and encouragement to make
this happen! My appreciation goes to my husband‘s families for their support and the great
times we spent together every time we get the opportunity. My special thanks go to Abune
Elias for his spiritual and fatherly concerns; it feels like gaining back a lost father. My grati-
tude extends to my sister in-law Meseret Mengistu for her concern and support not to
mention her vast knowledge and wisdom she shared with me. My genuine thanks go to my
dearest mom, Ewuye, for her eternal love and for all her miseries in trying to provide the
best of everything for her children. Thanks to my brothers for always being available when
I needed to hear their voices and for all the humorous chat we had. I have no words to
forward to my beloved husband, Misael for sharing the best and the worst part of life with
me and for all he had to go through while I am away. I sincerely admire his amazing
patience particularly to our ‗Terrible-Two‘ little angel, Emanuel who is the great source of
my deep joy.
v
vi
TABLE OF CONTENTS
Abstract ............................................................................................................ iii
Acknowledgments .............................................................................................. v
Table of contents ............................................................................................ vii
List of Papers ................................................................................................... ix
Introduction ....................................................................................................... 1
Scope and objectives .................................................................................................................... 1
Organization of the thesis ........................................................................................................... 2
Water Resources Management ..........................................................................2
Integrated Water Resources Management (IWRM)............................................................... 2
Hydrological models .................................................................................................................... 3
Constraints in selecting hydrological model ............................................................................. 4
Challenges in the study area ........................................................................................................ 5
Methods..............................................................................................................7
Literature review ........................................................................................................................... 7
Statistical analysis.......................................................................................................................... 7
Hydrological modeling ................................................................................................................ 8
Modeling Protocol ..........................................................................................................................................................................................................8
Results and Discussion .................................................................................... 12
Results from the statistical analysis .......................................................................................... 12
Hydro-meteorological time series trend analysis ....................................................................................................................................................... 12
Spatial distribution and representativeness of rainfall data...................................................................................................................................... 13
Results from hydrological modeling ........................................................................................ 14
Sensitivity analysis on model structure....................................................................................................................................................................... 14
Sensitivity analysis of parameters............................................................................................................................................................................... 15
Model calibration and validation............................................................................................................................................................................... 16
Model validation at interior points ............................................................................................................................................................................ 19
Water balance of the upper Awash River Basin.................................................................................................................................................... 20
Model prediction performance and uncertainties...................................................................................................................................................... 21
Conclusions ...................................................................................................... 22
References ........................................................................................................ 25
Other references ......................................................................................................................... 28
vii
viii
L IST OF PAPERS
I. Tessema SM, Setegn SG, Mörtberg U. 2011. Hydrological characterization of Awash
River catchments for prediction of stream flow. Submitted to Hydrological Processes,
April 2011.
II. Tessema SM, Setegn SG, Mörtberg U. 2011. Effects of different retention parameter
estimation methods on the prediction of surface runoff using the SCS curve number
method. Manuscript.
ix
x
I NTRODUCTION pollution that further deteriorates the water

resources, which threatens human health
Water is a major issue in life that needs to be and the natural hydrologic scheme of aquatic
protected and nourished. ―Water is not a ecosystems that add to the competition for
commercial product like any other but, water. The impact of its long term deteriora-
rather, a heritage which must be protected, tion can be quantified in the form of social,
defended and treated as such‖, as stated in economic and environmental aspects.
the European Union Water Framework An appropriate way to handle the question
Directive (WFD, EU 2000). The growing of water is to use a single system of water
pressure on the world‘s fresh water management, which is built up by river
resources is mainly due to and enforced by basin, in an integrated way. As defined in the
population growth that leads to conflicts Global Water Partnership (GWP, 2000),
between demands for different purposes. ‗Integrated Water Resources Management
Water is increasingly seen as a resource for (IWRM) is a process which promotes the
renewable energy generation, involving con- co-ordinated development and management
struction of dams and reservoirs for hydro- of water, land and related resources, in order
power, to meet increasing energy demands to maximize the resultant economic and
for enabling development and economic social welfare in an equitable manner with-
growth. At the same time, water is used for out compromising the sustainability of vital
other major welfare issues, such as irrigation ecosystems.‘ Hence, water resources man-
for food security and for household supply. agement should be implemented in a sus-
As a consequence, some of the common tainable way that encompasses balancing
problems related to water faced by many water needs, among other uses, for hydro-
countries are shortage, quality deterioration power and irrigation, flood protection and
and flood impacts, which call for a greater providing environmental flows for biodiver-
awareness and action. One of the United sity and ecosystem services.
Nations Millennium Development Goals
(MDG) adopted by world leaders in the year Hydrological models are among the available
2000 is to halve the proportion of people tools used to forecast and predict the quanti-
without sustainable access to safe drinking ty and quality of water for decision makers
water (UN, 2000). (Chow et al., 1988). Some of these models
could also predict the impacts of natural and
However, unsustainable ways of construct- anthropogenic changes on water resources
ing water infrastructure and utilizing water and also to quantify the spatial and temporal
lead to considerable impacts, altering the availability of the resources. However, the
regime of stream flows, flooding risk to the challenges lie in choosing and utilizing these
downstream and reducing or complete models for a specific basin and managerial
blocking of water to the ecosystem are plan.
among the many (see e.g. Molle et al., 2005).
According to a study made on irrigation in Scope and objectives
Africa (Food and Agriculture Organization The scope of this thesis is to understand and
of the United Nations FAO, 2005), the main characterize the hydrological processes of a
concern on water use is the conflict between specific watershed using statistical and
the environment and agriculture particularly deterministic approaches as a contribution
in lowland rural areas, where total base flows to the scarce information and database sys-
are diverted for irrigation without releasing tem for future management plans. The main
water for ecological conservation. Further, objective of this study was to characterize
excessive utilization of soil and forestry the major hydrological components of the
resources often results in negative impacts upper Awash River basin in Ethiopia under
of water resources especially in the develop- limited data conditions. It is believed that
ing countries where extreme poverty forces the research will contribute to fulfill the
those actions. Human activities induce knowledge gap and to improve the baseline
1
information for planning sustainable water states that ―fresh water is a finite and
management, which aims balancing water vulnerable resource, essential to sustain life,
demands for hydropower, irrigation and development and the environment‖ (ICWE,
ecosystem services as well as flood risk. The 1992). This principle demands a holistic
specific objectives were: to investigate the approach to management, which identifies
applicability of a hydrological model in the all the characteristics of the hydrological
study area for reliable characterization of cycle and its interaction with other natural
hydrological processes of the catchment and resources and ecosystems and considers the
further to assess and discuss the modeling demands placed on the resource together
uncertainties due to input data and other with the threats to it.
sources (Paper I), and to analyze the impacts The base for all IWRM is the hydrological
of model structure with the main focus on cycle. Rain, the main component of the
assessment of the predictions from two dif- hydrological cycle, that falls on a watershed
ferent soil moisture accounting methods for either intercepted by plants, immerse into
specific sub-watersheds in the study area the ground or runs over the surface (Chow,
(Paper II). 1964). The water that runs overland and
Organization of the thesis joins the stream at some point is the surface
runoff while the water that infiltrates into
The first section following this introduction
the ground and makes its way into the
presents the literature review starting with
stream much later is called base flow. How
the global perspective and agreement for
much infiltrates into the ground and how
protection of water and the approaches to
much runs over the surface depends on a
achieve a sustainable water resources man-
number of different factor, including how
agement system. This section also includes a
porous the ground is, how wet the ground
presentation of the theoretical background
was already, how intense the rainfall is, how
in the context of pros and cons of various
arid is the area, how much of the surface is
types of hydrological models and a thorough
covered by vegetation, how high is the slope
description of the general background of the
in the area and how much of the area is
study area. The next section briefly describes
impervious. This interrelationship calls for
the methods used to carry out the study and
an integrated management between surface
the detailed description of the modeling
and groundwater.
procedure. Then the results from and dis-
cussion of the different analyses and model Rivers are always moving and prone to
applications are compiled in the following change and both natural and human-induced
sections. Finally, conclusions from the anal- mechanisms cause rivers to change conti-
ysis in relation to the overall problems and nuously. Natural changes are gradual and
approaches end the thesis. possibly will balance in the long run, while
the human-induced changes may magnify
W ATER R ESOURCE S adverse effects and can imbalance the sys-
M ANAGEMENT tem rapidly. The degradation of the
Mesopotamian marshlands in the Tigris and
In this section some of the principles and
Euphrates river basins are examples of
challenges behind the water resources man-
unsustainable damming and river channeli-
agement, which includes the tools that could
zation during the late 1980s (UNEP, 2001).
alleviate its implementation are presented.
Further, land use change brought by defore-
Integrated Water Resources station and urbanization alter rates of
Management (IWRM) erosion, infiltration and overland flow. High
IWRM is inspired by the principles from the rates of erosion are in turn responsible for
Dublin statement on water and sustainable reducing the capacity of reservoirs of dams
development made in the International Con- constructed for hydropower and irrigation
ference on Water and the Environment and induce flooding to the downstream
(ICWE) in Dublin, 1992. The first principle (Flintan and Tamrat, 2002). Climate change
2
is also induced by land use change, which the hydrological cycle as a system that com-
has impacts on both water quantity and prises its different components as inputs like
quality (IPCC, 2000). The increment of rain- precipitation and outputs like runoff, using a
fall intensity, that causes detachment of soils set of equations that links the inputs and
on degraded and bare land that increase se- outputs (Chow et al., 1988).
diment transport and non-point sources The existing hydrological simulation models
pollutants to the streams; high surface can be grouped according to the runoff gen-
runoff that washes out wastes and garbage, eration process considered in each model.
especially in urban areas with poor drainage When comparing models, stochastic and
systems, to the streams; and the increasing deterministic models are often considered to
carbon dioxide content in the air that affects be at the top level of the classification tree,
the acidity of rain water are some of the in accordance to the way they treat the ran-
examples. The interaction shows that the domness of hydrologic phenomena (e.g.
hydrologic cycle goes through various com- Chow et al., 1988). Stochastic models use
plicated processes using air, soil, vegetation, local hydrometric data to predict flows.
surface and groundwater as a media. Hence, These models allow for some randomness
the integration of land and water manage- that results in different outputs and are
ment is indispensable to account for all the based on analysis of past events, commonly
interactions and in managing the relation- rainfall and river discharge (e.g. Ahmad et al.,
ships between quantity and quality and 2001; Tesfaye et al., 2006). Deterministic
upstream and downstream water interests models generally produce a single output of
(GWP, 2000). runoff for a given rainfall under identical
According to GWP (2000), implementation physical environments. Deterministic mod-
of IWRM would come to a possibility when els can be classified as; lumped models,
helped by management instruments, which where a variable or parameter is assumed to
are tools that enable and help decision- have an average value for the whole catch-
makers to make coherent choices between ment, and distributed models, where all
different alternatives. Among the methods, variables and parameters have different
water resources assessments and develop- values that account for the spatial variation
ment of its knowledge base are necessary for in the catchment.
effective water management. In order to Deterministic models can be further classi-
evaluate the resource availability and quality fied as empirical, conceptual and physically-
against the demands, the assessment should based models. Empirical models, which are
address the occurrence in space and time of usually lumped, are based on analyses of
both surface and groundwater. Likewise, parallel input-output time series with no
Irvine et al. (2005) stated that effective explicit account of physical processes. One
implementation of the WFD requires utiliza- of the recent methods in this category is an
tion of mathematical models, to provide a Artificial Neural Network (ANN) model
synthesis of complex natural processes and that has the ability to recursively learn from
to identify the likely response within and the data, also called a data-driven model (e.g.
among domains of natural and anthropogen- Govindaraju and Rao, 2000; Antar et al.,
ic changes. 2006). Conceptual models, which can be
Hydrological models lumped or distributed, are generally com-
posed of mathematical descriptions of the
A hydrological model is an approximation of
processes of catchment response. These
the complex reality using a system concept.
models represent the catchment as inte-
A system is a group of interacting or inter-
grated conceptual components but also
dependent components forming a complex
incorporate some aspects of physical
whole. The overall intent of the hydrologic
processes. Some examples of the conceptual
system analysis is to study the system func-
models are the Stanford Watershed Model
tion and predict its output. The models treat
(SWM, see Crawford and Linsely, 1966),
3
HBV (Bergström and Forsman, 1973), Stochastic models attempt to establish a

TOPMODEL (Beven and Kirkby, 1979), linkage between several phenomena from
the Identification of unit Hydrographs And historical data without internal description
Component flows from Rainfall, Evapora- of the physical processes involved. One of
tion and Streamflow data (IHACRES, the common uses of this type of model is
Jakeman and Hornberger, 1993) and the Soil for forecasting inflows into a reservoir sys-
and Water Assessment Tool (SWAT, Arnold tem. The simulated result could then be used
et al., 1998). In both empirical and concep- for estimating reservoir storage requirements
tual approaches measured hydrological and determining optimal operating polices
variables, usually stream flow data, are for the use of available water (Salas, 1993).
needed for calibration in order to develop Data requirements for this type of models
and parameterize the model. Physically- are relatively low as compared to the physi-
based models are based solidly on an under- cally based models. However, there is a need
standing of the physical processes using of long historical data to simulate a reliable
fundamental hydrodynamic laws. These future natural river flows. The parameteriza-
models are fully distributed and use a finite tion process might be taken as easier since
difference or finite element grid system. the number of parameters to be estimated is
Some of the distributed models currently in relatively small. Still, computationally they
use are the Systeme Hydrologique Europeen might not be necessarily faster than corres-
model (SHE, Abbott et al., 1986), the Insti- ponding deterministic models due to the
tute of Hydrology Distributed Model difficulties of generating good random
(IHDM, Calver and Wood, 1995) and Water numbers with equal likelihood appearance
& Environmental Consultants – Catchment and non-correlated values in a row (Mulligan
(WEC-C, Croton and Barry, 2001). The and Wainwright, 2004). On the other hand,
main driving force in using physically-based since the prediction does not encompass the
models is their applicability in ungauged interaction with the soil or land use
catchments with the assumption that all the processes it is not possible to do analysis on
parameters needed for the model are mea- different scenarios of land and water
surable. Detailed descriptions about the resources management, which is useful to
types of models could be found in e.g. detect negative impacts from anthropogenic
Becker and Serban (1990), Refsgaard (1996), changes.
Jones (1997), Beven (2001), and Mulligan Deterministic models complexity varies
and Wainwright (2004). according to their representation of the spa-
Constraints in selecting hydrological tial variability of a catchment and their
model process description. Empirical models are
comparable with the stochastic models with
Even though the development of models is regard to process description (Refsgaard,
well advanced, choosing an appropriate 1996). Physical processes are not involved in
modeling system for practical applications of the assessment of the mathematical equa-
integrated management is still a challenge. In tions behind this type of models but rather
addition to the purpose of the modeling and are based on analysis of simultaneous input
the types of the analysis to perform, one has and output time series. The common use of
to consider various factors in relation to the this type of model is a rainfall-runoff model-
model structure that includes process ing for individual catchments. Moreover,
description, numerical discretization, effec- they have also been in use in conceptual and
tive parameterization and computational physically-based models as a generator of
efficiency in time and cost. Moreover, an some components, for instance unit hydro-
extensive data requirement in relation to its graph models used to represent the ground
availability is the main constraint in model water system in conceptual models (e.g.
application. Refsgaard, 1996; Minns and Babovic, 1996).
The data requirements and computational
4
needs are usually low in comparison to con- explanations of the catchment in compari-
ceptual and physically-based models. Gener- son to the other models (Refsgaard, 1996).
ally, empirical models have high predictive The common use of this type of model is
power. However, their low explanatory for prediction of the effects of natural and
depth makes them specific to the given con- anthropogenic changes in the catchments on
ditions and hard to generalize for application stream flow, for instance surface and ground
on other catchments with similar physical water development for hydropower or irri-
characteristics (Mulligan and Wainwright, gation. However, the major constraint is the
2004). extensive data requirements, since all para-
Conceptual lumped models use semi- meters that correspond to each modeling
empirical equations with physical basis un- unit are assumed to be measurable from the
derneath to describe the hydrological field. Practically, it is impossible to measure
processes, mainly due to their behavior of such amount of data for different reasons,
using the average value of parameters and for instance the time and cost of measure-
variables for the whole catchment ment and the inadequacy of the available
(Refsgaard, 1996). This type of models measurement techniques. These models tend
represents water flows in the catchment as to have good explanatory depth but low
individual storages corresponding to the predictive power (Mulligan and Wainwright,
different components that are interlinked in 2004). Hence, the models often need to be
a row. Data requirements are low in compar- calibrated to get good agreement with
ison to the fully distributed models, which observed data. Moreover, the availability of
makes them suitable to use in most cases. computational efficient machine to run these
This type of model has a number of parame- models is another constraint to consider.
ters that should be calibrated using observed Semi-distributed models use multiple
data for a specific application (Sorooshian, lumped units in a catchment either as sub-
1991). The computational efficiency could basins or hydrological response units
vary according to their complexity, for (HRU), with a perception that the response
instance the number of free parameters to of the grouped cells ought to be similar.
be calibrated. The common use of this type Their hydrological process description is
of model is for the simulation of rainfall- based on the conceptual type. This type of
runoff process for flow forecasting. They model can compromise between the distri-
also can give an indication of the effects of buted and lumped models in their represen-
land use changes in qualitative and quantita- tation of spatial variability and data
tive ways with a minimum amount of data requirement (Merritt et al, 2003). In addition,
(Merritt et al, 2003). So in the case of mini- these models can offer computational and
mum requirements of detail description of parametric efficiency in the case of inte-
the area, this type of model might be a good grated modeling of the units (Irvine et al.,
choice. However, their limited representa- 2005).
tion of the spatial variability of the natural
Challenges in the study area
system might restrict their applicability for
integrated management purpose. Ethiopia is one of the countries that can be
considered to have abundant water
Physically-based distributed models calculate
resources, available for hydropower and
the flows of water and energy directly from
irrigation, at the same time as having signifi-
the fundamental physical equations with the
cant problems related to water resources,
law of conservation as a foundation. This
such as flooding, drought and depletion of
type of model represents the spatial variabili-
ecosystem services. Despite the fact of the
ty of nature by splitting the catchment into
substantial hydropower potential, it has one
either square cells or triangular irregular
of the lowest levels of per capita electrical
networks. Apart from the detailed descrip-
consumption in the world (Flintan and
tion of the hydrological processes, the
Tamrat, 2002). The country plans for major
models provide potentially more truthful
5
Figure 1. Location map of the study area.
investments in dams for hydropower pur- basin is estimated to be around 14 million as

poses, which at the same time are expected derived from the 1994 population and hous-
to provide water for irrigation and flood ing census (Ministry of Water and Energy,
control (Ministry of Water and Energy, 2010). According to Taddese et al. (2003),
2010). On the other hand, the country still there are three functional dams in the basin
needs to work hard to fulfill the MDG that for hydropower generation and irrigation
focuses on sustainable access to safe drink- development. Furthermore, the majority
ing water. According to the MDG reports (48 to 70%) of the existing large-scale
(2010), currently the country lies on less irrigated agriculture of the country is located
than 50% coverage based on the indicator along this river (Flintan and Tamrat, 2002;
that stands for proportion of population Achamyeleh, 2003).
using an improved drinking water source. The major problems that restraints the de-
The Awash River basin is one of the signifi- velopment in the Awash River basin are a
cant spots, from development aspect, from recurrent large scale flooding and drought.
the total of eight basins in the country. The The impacts of flooding so far include both
basin is located in the Rift Valley (Figure 1) the tragic human life‘s loss and extensive
and is at a leading position in the develop- property damages (Emergency appeal, 2006).
ment practice of the country, particularly in According to Desalegn (2006), the occur-
irrigation. The Awash River originates at an rence of drought is, on average, once every
elevation of about 3000 m a.s.l. in the central two years and causes a great deal of damage
part of Ethiopia and joins Lake Abe at an to crops and livestock. Water management
elevation of about 250 m a.s.l. after flowing problems and investments in inappropriate
for a total of 1200 km. The basin covers a development projects hinder the prevention
total area of about 113,000 km2 and drains in and mitigation of the impacts. The major
six regional states that include the capital causes of the disasters include population
city, Addis Abeba. The population of the growth, wetland degradation, soil erosion
6
and sedimentation. The impacts from the metals (like As, Cr, Fe, Mn and Se), nitrate,
population growth result in deforestation phosphate, E.coli, and coliform are the
and overgrazing that in turn increase the soil common ones.
erosion in the basin. The severe and uninhi- Hence, these challenges need integrated
bited soil erosion results in the siltation of water resources management that could
the major reservoirs. Koka dam is one of the alleviate addressing of the specific demands
important reservoirs affected by the sedi- from each stakeholder. The looking for the
mentation. It has been in use starting from optimum solution should basically start with
1960 for hydropower generation, irrigation proper characterizing of the basin with the
developments in the downstream and also as help of the available and appropriate tools
flood attenuation during the heavy rain sea- like hydrological models.
sons. According to Achamyeleh (2003), the
designers of the dam underestimated the M ETHODS
severity of sedimentation on the life of the In this section the methods used to conduct
reservoir. The reduction in the storage the research are described briefly. Further
capacity emphasizes spillage requirements in description is presented in the appended
flood situations and worsening water short- papers.
ages during dry periods (Piguet, 2001).
Wetlands are being converted to agricultural Literature review
lands and reservoirs that result in degrada- A literature review was conducted on IWRM
tion and destruction of the natural ecosys- and different types of hydrological models.
tems of the basin (Flintan and Tamrat, The main emphasis was given to present the
2002). In addition, the neglected role of the various aspects of the models in utilizing
wetlands in providing dry season grazing them as tools to implement IWRM and also
and the constructions that prevent the the constraints in selecting a particular
flooding of the grazing lands creates con- model for a specific location.
flicts among the pastoralists (Flintan and
Tamrat, 2002; Taddese et al., 2003).
Statistical analysis
Moreover, water shortage and displacement The application of standard statistical tools
from the grazing lands adds to the conflicts for the analyses of the available hydro-
among different ethnic pastoralists and also meteorological data was one of the
with the large-scale irrigation organizations approaches used in this study. The analysis
and commercial farmers (Piguet, 2001, was made to get an insight of the correlation
2002). These further result in tampering and between the available climate and stream
destroying of constructions of dykes and flow measurements at the headwaters of the
drainages to induce flooding to the dry Awash River basin. The spatial distribution
grazing lands (Achamyeleh, 2003). and representativeness of rainfall data in
relation to observed stream flow data was
The water quality deterioration is another
assessed using a correlation coefficient
important factor that stresses the shortage of
method. The analysis was based on the daily
water in this basin. The water quality issue in
data from the available rainfall gauges and
the river and its tributaries that includes lake
stream flow measurements used in the study.
Koka has been discussed by many (e.g.
Additional data sets of daily precipitation
Alemayehu, 2001; Zinabu and Pearce, 2003;
was calculated, based on area weighed prin-
Kebbede, 2004; Prabu, 2009). The studies so
ciple, for some of the sub-watersheds that
far agreed that the most probable sources of
contains more than one sub-basin. The
the pollutants are untreated effluents from
reason for the use of these additional data-
factories and industries in the surrounding
sets was to relate the analysis with the simu-
cities and also direct discharge of domestic
lated results from the model application.
wastes (both solid and sewage water).
Among the identified pollutants, with high Since the model used the values from the
nearest station for each sub-basin, in the
risk for human and animal health, heavy
case of a watershed composed of two or
7
more sub-basins the value would be area- data collected for this study includes; digital
weighed. In addition, the annual flow dura- elevation models (DEM) with 30 m posting
tion curve was also analyzed at the outlet to interval were downloaded from the official
characterize the ability of the watershed to website of Earth Remote Sensing Data
provide flows of various magnitudes. The Analysis Center of Japan (ERSDAC, 2009),
shape of a flow duration curve is also helpful climate data (daily rainfall and temperature)
in evaluating the stream and watershed from National Meteorological Agency of
characteristics depending on the time scale Ethiopia, stream flow measurements of five
used in preparing the curve. gauging stations from Ministry of Water and
Energy of Ethiopia, land use and soil maps
Hydrological modeling
including their physical and chemical prop-
Characterization of the specified watershed erties for parameter assigning (Chekol, 2006)
was continued by adapting a hydrological and previous studies for comparison. This
model under limited data conditions. The step helps to identify the outlet of the
application of the model involved model watershed to be analyzed (that would deter-
setup, sensitivity analysis, calibration, and mine the system boundaries) and gives em-
uncertainty analysis. The performance of the phasis on the available parameters including
model was evaluated by comparing the their measurement scale (that would deter-
simulated flow hydrograph with the mine the appropriate numerical discretiza-
observed hydrograph visually and also using tion). Based on the availability of the data,
three model goodness-of-fit statistics the present study focuses on the uppermost
methods. The modeling protocol, as part of the Awash River basin, which is lo-
proposed in Anderson and Woessner (1992) cated between latitude 9° 18‘ N and 8° 17‘ N
and later modified in Refsgaard (1996), was and longitude 37° 57‘ E and 39° 4‘ E. The
adapted for this study. A detailed description area coverage of the watershed is about
of the modeling process follows in the 7630 km2 with an outlet at the Hombole
coming section. gauging station. Some of the main streams
Modeling Protocol that construct the Upper Awash River Basin
The procedure in the protocol (Figure 2) are the Holeta, Berga, Akaki and Melka
starts by clearly defining the purpose of the Kuntire streams. Table 1 summarizes the
application or the modeling effort. As characteristics of the sub-watersheds. The
previously mentioned, the goal of this study location and spatial distribution of the
is to characterize different components of stations are shown in Figure 1.
the hydrological processes and to set up a Model selection is where the mathematical
model for future use of assessment of the model will be either selected from the exist-
effects from different management plans on ing codes or modified by adding compo-
stream flows. This step will help to identify nents to an existing one or developed from
which type of modeling system is suitable to scratch. Since the scope of this study was to
solve or get an insight to the defined apply the models for the specified purpose
problem. the model selection was made from the
The second step is to build the conceptual existing hydrological models that have been
model of the system based on the specific successfully verified by others in different
purpose. Identification of the system boun- places. The selection was then based on the
daries, available data collection from the combined analysis of the previous steps and
representative authorities and literature the accessibility of the models from the cost
reviews in addition to the field visits of the and user friendliness aspects. For this study
study area are all in this category. In addition the semi-distributed model known as Soil
to the defined purpose, the complexity of and Water Assessment Tool (SWAT) was
the model to be chosen depends on the selected.
analysis of the available data. The available
8
Define Purpose Figure 2. The proce-

dure for hydrological
model application.
Field data Conceptual model Modified after
Anderson and
Woessner (1992) and
Refsgaard (1996).
Model selection
Model construction
Performance criteria
Sensitivity analysis and

Calibration
Validation
Prediction
Presentation of results
Postaudit
SWAT is developed to examine the influ- (HRU), which is a lumped land area within a
ence of topographic, land use, soil and sub-watershed comprised of unique land
climatic conditions on stream flow and cover, soil, slope, and management combi-
sediment yield. It has also been used to pre- nations. The hydrological components (like
dict the potential for hydropower (e.g. Kusre surface flow, lateral flow, groundwater flow
et al., 2010) as well as the impact of land and evapotranspiration) are calculated for
management practices on water, sediment each HRU through the water balance. The
and agricultural chemical yields (Neitsch et cumulative total over a sub-watershed then
al, 2005). It is a freely available model that gives the hydrological balance and main
can be utilized either from the source code stream flow for that sub-watershed. The
or from the Geographic Information System overall watershed hydrologic balance includ-
(GIS) interfaces, which simplifies the inte- ing stream flow at the outlet of the whole
gration of various spatial environmental data watershed is then calculated from the con-
and the use of bulk data. It is a continuous tribution of the upstream sub-watersheds.
time model and allows for a simulation of The hydrology part of the model is sepa-
different physical processes in a watershed. rated into two: the land phase and the
The spatial unit for rainfall-runoff calcula- routing phase of the hydrologic cycle.
tions is the Hydrologic Response Unit
9
Table 1. Summary of the physical characteristics of the five sub–watersheds and

watersheds used for the study.
Characteristics Holeta Berga Akaki Melka Kuntire Hombole
2
Size (km ) 131.1 216.8 884.3 4501.0 7626.3
Dominant slope (%) /
>15/ 42.6 >15/ 36.5 8-15/33.4 2-5 / 30.8 2-5/ 28.9
Areal coverage (%)
Average elevation
2544 2652 2371 2048 2018
(m a.s.l.)
Dominant soil type / Luvisols / Luvisols / Vertisols /
Vertisols / 2422 Vertisols / 2422
depth (mm) 1800 1800 2422
Durum Durum
Dominant land use Corn Eragrostis Teff Eragrostis Teff
Wheat Wheat
The land phase of the hydrologic cycle con- profile is calculated as the difference
trols the amount of water that goes to the between the amount of rainfall and the
main channel from each sub-watershed amount of surface runoff. The rate and
while the routing phase controls the move- velocity of flow is defined by Manning‘s
ment of water through the channel network equation. Two variations of the kinematic
of the watershed to the outlet. wave model are incorporated in SWAT to
In this study, the calculation of surface route the water through the channel net-
runoff was made using the modified Soil work; variable storage is the primary method
Conservation Service (SCS) curve number and Muskingum River routing is the other
procedure (USDA-SCS, 1972). This method option. Detailed description of the model
uses two equations for runoff, of which the could be found in Neitsch et al. (2005).
first one relates runoff to rainfall and reten- Model construction is the setup of the collected
tion parameter while the second equation data to a platform where the modeling task
relates the retention parameter to the curve takes place. In this study, the setup started
number (Paper II). The retention parameter with the delineation of the watershed from
is the maximum potential difference the DEM with the help of the GIS interface.
between rainfall and runoff starting at the This was followed by the definition of the
time the storm begins and used in the daily stream network, setting of the different sub-
curve number calculation. The model incor- basins outlets that correspond to the stream
porates two different options for calculating flow data and discretization of the watershed
the retention parameter. The initial and into HRU's based on the land use, soil and
default setup is to allow the retention para- slope data. This step also involved setting
meter to vary with soil profile water content the initial conditions (like antecedent soil
(here after called SM), which is characterized moisture content) and preliminary selection
by the field capacity, the wilting point and of parameter values from the input data.
the saturation water content. The second Performance criteria should be established in
method is to allow it to vary with the accu- order to evaluate the model output in com-
mulated plant evapotranspiration (here after parison to the observed data during the
called PT), which minimizes the impact of calibration and validation process. The
depending on soil storage and gives empha- criteria should encompass the desired accu-
sis on antecedent climate. racy for the specific purpose in accordance
For the calculation of evapotranspiration, with the realistic condition of the input data.
Hargreaves method (Hargreaves et al., 1985) In this study, the performance of the model
that only needs air temperature data was was evaluated by comparing the simulated
used. Since the curve number method is stream flow hydrograph with the observed
used to calculate the surface runoff, the hydrograph visually and also using
amount of water that infiltrates to the soil three model goodness-of-fit measures;
10
Nash-Sutcliff Efficiency (NSE), coefficient with the observed stream flow and also to
of determination (R2) and percent bias minimize the potential range that bound the
(PBIAS), which are widely used in hydrology value of each parameter. Then the automatic
(Moriasi et al., 2007). The performance of calibration and uncertainty analysis method,
the model can be explained based on the Parameter Solution (ParaSol, van Griensven
recommended ranges: NSE values greater and Meixner, 2007), which is incorporated in
than 0.75 are considered to show good SWAT, was used. The Objective Function
model efficiency, NSE between 0.36 and (OF) used in this study was the Sum of the
0.75 shows satisfactory performance and Squares of the residuals (SSQ) with an
values less than 0.36 are considered to be objective of minimizing the differences
unsatisfactory (e.g. Motovilov et al., 1999; between the simulated and measured time
Moriasi et al., 2007; Van Liew et al., 2007). series. Since SSQ has a high priority of
Similarly, PBIAS values greater than and minimizing the large differences and often
equal to twenty five percent (±25%) are tend to force the model to underestimate the
considered satisfactory. peak flows for a result of lower OF values
Sensitivity analysis is the step where the uncer- (e.g. Eckhardt and Arnold, 2001), further
tainties of the modeling process, either due tuning of parameters was made manually by
to model structure or the estimated parame- visual comparison of the simulated and
ter values, could be evaluated. It also helps observed hydrographs. In Paper II, the
to prioritize parameters to include in the manual calibration was done only for the
calibration process, particularly for models two most sensitive parameters, selected
with considerable amount of parameters. In based on the previous experience (Paper I),
this study, the sensitivity analysis was per- and all the parameter space was used. The
formed to select the most sensitive parame- calibration was based on the first three years
ters, out of the total of 27 flow parameters of data (1996-1998).
that are included in SWAT, for calibration. Validation is the step where the capabilities
A sensitive parameter, in this study, was one of the calibrated model in simulating accept-
that changed the model outputs of the able results could be confirmed. In this
stream flow significantly per unit change in study, the validation of the model was per-
its value. For this purpose, the automated formed to test if the calibrated parameter set
Latin Hypercube One-factor-At-a-Time would behave consistently for the watershed
(LH-OAT, van Griensven et al. 2006) global using different observed datasets in another
sensitivity analysis procedure was used. The period than the calibration (1999-2000). In
structure of the model was adjusted interac- this case, the stream flow dataset was the
tively to evaluate the different methods (one only observed data used for model valida-
at a time) that were incorporated in the tion. In Paper I, the performance of the
model, which concerned the estimation of model was also evaluated based on the
retention parameters (SM and PT) and stream flow prediction at four interior points
channel routing (variable storage and from a parent watershed calibration. The
Muskingum). The analysis was based on the other three interior points at tributary rivers,
model performance on simulating a realistic Holeta, Berga and Melka Kuntire were used
stream flow that best represented the to evaluate the model performance based on
observed value. the parent watershed (Hombole) calibration,
Calibration is the process of finding the in a similar way as discussed in Reed et al.
optimum set of parameters that would help (2004). In order to comprehend the ability
the model to reproduce the observed data of the model in simulating for un-gauged
within the desired accuracy. The calibration watersheds the discharge data from the inte-
of the selected sensitive parameters was un- rior points were not used for the calibration
dertaken in different steps. In Paper I, process, but left for validation. In the case of
manual calibration (trial-and-error) was the Akaki sub-watershed, the study evaluated
first step to reach at some level of agreement two scenarios; the calibrated result from
11
using its own discharge data and as a valida- (Paper I). The following two sub-sections
tion for Hombole calibration. briefly describe the outcome.
Prediction is the step where the response of Hydro-meteorological time series trend
the watershed to future changes would be analysis
quantified by applying the calibrated model. Air temperature was analyzed based on the
The calibrated parameter values and the available ten years (1996-2006) data from
catchment conditions would be kept two stations (Addis Abeba Observation and
unchanged while checking the effects of Debre Zeit). The months with the maximum
different scenarios or estimated changes in and minimum average temperature occur-
the watershed. The uncertainty at this step rences were May and December respectively,
includes both the uncertainties from the for both stations. However, the variation
calibrated parameters and from the esti- between different months was not signifi-
mated future spatial and temporal changes. cant as is indicated by the rather low values
Since the scope of the current study is to of the standard deviation (on average 1.1).
setup the modeling system for a specific The altitude difference, which is 2408 m for
watershed, this step was forwarded to the Addis Abeba Observation and 1900 m for
future task, which is to assess the impacts of Debre Zeit, could explain their slight distinc-
land use and water resources management tion in temperature between the correspond-
plans in the study area. ing months of the two stations (on average
Presentation of results could be done in differ- Debre Zeit is warmer by 2.2 °C throughout
ent ways, like reports, graphs or even anima- the year).
tions. A clear presentation of the findings is The precipitation data was analyzed from the
essential for a good understanding of the available eight stations in the watershed. The
problems and for reaching consensus length of the recordings varied between sta-
between different stakeholders as well as to tions, with the maximum being 45 years
help decision makers. The results of this (1961-2006). However, the majority lied
study are compiled in the following chapter. between the years 1982 and 2002 with a 20
Postaudit is another way of validation of the year length of record. The analysis showed
model prediction for a specific watershed. that the pattern and character of rainfall
This step should be performed a number of varied in different parts of the country main-
years after the modeling task, which would ly due to its geographic location and topo-
allow adequate time for significant changes graphy (Degefu, 1987; Bekele, 1997). Out of
to occur. The validation would then be the analyzed eight stations, three of them
based on the comparison between the (Addisalem, Ginchi and Tulubolo) falls in
predicted values and the collected new data the regime characterized by a mono-modal
for the corresponding period. In the case of rainfall pattern from March to September.
unsatisfactory outcomes, this step will help The rest of the stations fall in the regime
to redesign the model through changes in characterized by three distinct seasons; wet,
the conceptual model or in the parameter small rain and dry seasons. The wet season
values. starts in June and runs up to September. The
R ESULTS AND D ISCUSSION small rain season runs from February to May
while the dry season usually occurs from
Summaries of the results and discussion October to January. The standard deviation
from the two approaches, statistical and of the rainfall data for each month was cal-
modeling, are presented in this section. The culated based on all stations and it showed
results are compiled in separate sub-sections. that the variation between the stations was
Results from the statistical analysis high during the wet season (37 mm in
August) and low in dry season (3.8 mm in
The analysis on the hydro-meteorological December). The calculated area weighed
data was based on the available temperature, rainfall for Hombole watershed adapted the
precipitation and streamflow measurements mono-modal pattern after the first three
12
Figure 3. Discharge duration curve showing the percent of time in which the mean
annual discharge (177 mm) was exceeded through the 39 years of record (1968-2006).
stations. Still, the amount of rainfall showed showed that the exceedance probability
a distinct boundary between the two rain from the mean annual value was 43%, from
seasons (wet and small rain) while the sepa- which 23% exceeds the mean by 10%
rator, the relatively short ‗dry‘ period as de- (Figure 3). The shape of the daily duration
scribed in Bekele (1997), in between (in this curve, which was flat towards its tail, sug-
case during May) overturned and resulted in gests for a potential of good storage capacity
a single peak rainfall pattern. in the watershed. The average monthly flow
The flow regimes of the Hombole watershed pursued the rainfall trend which is a
was analyzed using 39 years of observed mono-modal curve with a peak in August
stream flow data (1968 - 2006). The average (Figure 4).
annual depth of runoff in the watershed was Spatial distribution and representativeness of
estimated at 177 mm. The wettest year was rainfall data
1971 with a 313 mm discharge and the One of the major limitations to large area
driest, with 86 mm discharge, was the year hydrologic modeling is the spatial variability
1987. The coefficient of variation on annual associated with precipitation. Even though
basis is about 0.29 and the flow regime is eight rainfall gages were used in this study,
characterized by elongated periods of me- the spatial distribution in relation to the
dium to high flow and shorter intermingled watershed area could be considered to be
periods of low flow. The frequency analysis inadequate. According to Arnold et al.
Figure 4. Graphical presentation of average monthly temperature (°C), rainfall (mm),

and streamflow (mm) trends at Homble watershed.
13
Table 2. Correlation coefficients between the observed rainfall and streamflow (Q)
daily data from 1996 to 2000 (the shaded values represent the combinations selected by
SWAT).
Rainfall stations Holeta Q Berga Q Akaki Q Melka Kuntire Q Hombole Q
Addis Abeba Bole 0.22 0.28 0.30 0.32 0.31

Addis Abeba Obs 0.28 0.29 0.35 0.36 0.36
Addisalem 0.17 0.24 0.18 0.35 0.32
Akaki 0.20 0.27 0.33 0.33 0.30
Debre Zeit 0.22 0.25 0.27 0.33 0.36
Ginchi 0.19 0.24 0.20 0.29 0.27
Tulubolo 0.29 0.29 0.27 0.41 0.40
Zukuala 0.17 0.16 0.21 0.25 0.25
Akaki area weighted 0.28 0.33 0.39 0.40 0.39
Melka Kuntire area 0.32 0.37 0.33 0.52 0.49
weighted
Hombole area weighted 0.36 0.41 0.41 0.56 0.54
(1998), using one gage to represent an entire default setup of the model used; SM for the
sub-watershed or even by means of area retention parameter calculation and the vari-
weighed methods for rainfall representation able storage routing method. The hydro-
can cause considerable error in runoff esti- graph produced by this combination was
mation. However, the correlation coefficient dominated by high runoff events during the
of the precipitation data and observed dry and small rainy season while relatively
stream flow might help in explaining the low flow during the main rain season when
model performance in different sub- compared to the observed hydrograph
watersheds. According to the analysis, the (Figure 5). The comparison on daily basis
correlation between the observed stream indicated that the changing of the retention
flow at the Akaki sub-watershed and the parameter calculation method to PT
corresponding rainfall gage considered by improved the fit of the predicted vs.
SWAT (in this case area weighed) was poor observed daily stream flow significantly for
in relation to both Melka Kuntire and some of the sub-watersheds. The significant
Hombole watersheds (Table 2). Holeta and change was on the falling limb and the base
Berga sub-watersheds, both were having flow recession (Figure 5). The change in the
only a single rainfall station had the lowest channel routing method to Muskingum
correlation coefficients relative to the others. improved the results in all sub-watersheds in
combination with both SM and PT
Results from hydrological modeling methods. This method helped to minimize
The model application starts with the sensi- the overestimation of peak flows in some
tivity analysis, both on model structure and cases. The summary from the model per-
flow parameters. This section presents the formance evaluators (Table 3) show that the
analysis from both papers (Paper I and different measures favored different combi-
Paper II) in separate sub-sections according nations, which could be considered as an
to the different procedures. indication for further analysis prior any con-
Sensitivity analysis on model structure clusion. However, the majority of values still
show that Muskingum is giving a better
In Paper II, the analysis of the model struc- performance for this specific study.
ture was based on the different combina-
tions of methods for estimation of retention One common reaction that was observed in
parameter and channel routing. Initially, the Holeta and Berga sub-watersheds was that
un-calibrated simulation results from the the catchments seemed to respond only to
different combinations were compared sta- the surface runoff that came from the heavy
tistically and visually using hydrographs. The rain during the wet season. The simulation
underestimated the fraction of water contri-
14
Figure 5. Hydrographs that compare observed and simulated stream flows

using the Muskingum method combined with the SM method (top) and
PT method (bottom) at Akaki sub-watershed for the year 1996.
buted by the base flow during that time. The In Paper II the sensitivity of the different
hydrographs showed poor base flow reces- hydrological components were analyzed
sion time in comparison to the observed based on the weighting coefficient (cncoef)
flow. This could be the combined result of within the parameter‘s possible range of
their commonly being the steepest and values. This single parameter is associated
dominated by relatively shallow soils in with the PT method, which is used to calcu-
comparison to the others. late the retention parameter for daily curve
number calculations. The sensitivity analysis
Sensitivity analysis of parameters
of the cncoef was performed manually and
In both papers the sensitivity analysis indi- showed that the weighting coefficient and
cated that the stream flow prediction was surface runoff were directly proportional
sensitive to variation in surface water, while the relation with subsurface runoff
ground water and soil parameters. Among was inversely proportional. This result is
the twenty seven hydrological parameters, similar to the study discussed by Kannan et
the identified most sensitive ones were the al. (2008). The default value for cncoef was 1
base flow alpha factor (ALPHA_BF), initial with the possible range between 0 and 2.
SCS curve number II (CN2), soil evapora- The increasing of the parameter value also
tion compensation factor (ESCO), ground increased the mean of the total flow with an
water delay (GW_DELAY), deep aquifer average proportion of about 2.5 mm for
percolation fraction (RCHRG_DP), availa- each 0.1 increment. The base flow fraction
ble soil water capacity (SOL_AWC), satu- accounting problem, which was discussed in
rated hydraulic conductivity (SOL_K), and the sensitivity of the model structure, for the
surface runoff lag time (SURLAG). two smaller sub-watersheds was also tested.
The result shows the improvement achieved
15
Table 3. Un-calibrated model performance measures, using the different combinations

of methods for estimation of retention parameter and channel routing, on daily basis
during the calibration period (1996-1998), values in bold are the relative best in each
case.
2
Watershed Method NSE R PBIAS
Holeta
SM - Variable Storage -0.24 0.13 37.6
SM - Muskingum -0.23 0.13 37.6
PT - Variable Storage -2.26 0.06 49.93
PT - Muskingum -2.16 0.06 49.94
Berga
SM - Variable Storage 0.04 0.14 3.91
SM - Muskingum 0.09 0.16 3.91
PT - Muskingum -0.48 0.13 9.46
Akaki
SM - Variable Storage 0.12 0.21 9.05
SM - Muskingum 0.27 0.29 9.07
PT - Variable Storage 0.25 0.35 10.42
PT - Muskingum 0.41 0.44 10.43
Melka Kuntire
SM - Muskingum -0.44 0.53 114.68
PT - Muskingum -2.08 0.58 120.76
Hombole
SM - Muskingum -0.33 0.52 113.28
PT - Muskingum -1.16 0.65 117.3
by changing cncoef to 0.25 for Holeta sub- calibration, guided by the visual comparison.
watershed in comparison to the default value Generally, the prediction of daily peak flows
and observed flow (Figure 6). The calibrated differed between the two watersheds in
optimal values, based on the comparison of comparison with the observed stream flow.
the base flow fraction between the observed For the Akaki sub-watershed, underestima-
and simulated flow, were 1.05 and 0.6 for tion of the daily peak values was observed
Akaki and Hombole, respectively. during the wet season of the entire period
while some overestimations occurred during
Model calibration and validation the small rain season. For the Hombole
The different calibration schemes, from watershed, the simulated flow showed over-
both papers, produced slightly different estimation of some of the daily peak flows in
hydrographs in both Akaki and Hombole both seasons. The hydrograph comparison
watersheds. The simulated flow from indicated that the PT method was better in
Paper I showed a relatively better prediction peak flow prediction while the SM method
of some peak flows (Figure 7). This was due outperformed in simulating the low flows.
to the final manual adjustment of some of The noticed good agreement between the
the parameters, after performing the auto- observed and simulated discharge of the
16
Figure 6. Hydrograph plots comparing the observed with two simulated

stream flows using different weighting coefficient (cncoef) values in Holeta
sub-watershed during the calibration period (1996-1998).
calibrated model was verified by the values all the cases corresponded to the PT simula-
from the performance evaluators (Table 4). tion from Paper II. On a daily basis, the
The comparison with the simulated results NSE for Akaki and Hombole watershed
from the uncalibrated model (default ranged, respectively, between 0.49 to 0.53
parameters) confirmed that the calibration and 0.72 to 0.83. Similarly, the R2 value
was essential for this area. According to the ranged between 0.50 to 0.53 and
evaluators, the relatively best performance in 0.75 to 0.83 while the PBIAS ranged
Figure 7. Daily observed hydrographs at Akaki (top) and Hombole

(bottom) in comparison to the three different simulated flows during the
calibration (1996-1998) and validation (1999-2000) periods.
17
between (-8.9) to 0.3 and 12.0 to (-1.7). The watershed the maximum overestimation was
negative and positive signs of the PBIAS observed in the year 2000 simulation result.
indicate, respectively, the underestimation The mean annual stream flow decreased
and overestimation of the total simulated gradually from 1998 whereas the average
flow in comparison to the observed flow. weighted rainfall for Hombole watershed
Based on the recommended ranges, the increased from 1999 to 2000. This means
performance of the model lies on a satisfac- that the observed total runoff for the year
tory level for Akaki and good level for 2000 was relatively dry when compared to
Hombole watershed. the observed annual rainfall. The overall
On monthly basis the trend was similar to performance, both during the calibration
the daily time step. The NSE was and the validation periods, might indicate
0.80 to 0.84 and 0.94 to 0.97 for Akaki and that the model performed relatively well
Hombole watersheds, respectively. Likewise, during wet years.
the R2 value was 0.83 to 0.85 and The difference between the two watersheds
0.95 to 0.97 while the PBIAS was in model performance could be due to the
(-8.7) to 0.6 and 12.0 to (-1.6). According to forcing input data. The correlation between
the performance ranges, the model showed rainfall and runoff for Akaki watershed was
good performance in predicting the monthly poorer than for Hombole (Table 2). One of
discharges from both watersheds. the main sources of model uncertainties is
During the validation period (1999-2000), acknowledged (see e.g. Refsgaard and Storm,
the analysis of the model performance 1996; Beven, 2001) to be errors in the input
showed that for the Akaki watershed, the variables such as rainfall and temperature.
underestimation of the observed total Hence, the poor model performance in the
stream flow was in a much larger quantity Akaki sub-watershed could be due to poor
than the calibration period. This might be quality of the gauged climate variables as
explained by the contradiction in the trends well as the very coarse spatial distribution of
of the annual mean rainfall and runoff for weather stations in the sub-watershed.
the year 1999. The mean annual runoff Another possible reason is the aggregation
increased while the rainfall decreased in of the discharges from the different
comparison to 1998 and vice versa when upstream sub-watersheds to a larger area, in
compared to 2000. For the Hombole this case Hombole watershed, which might
Table 4. Model performance statistics from the three different simulations for the
prediction of stream flow in the Akaki and Hombole watersheds. (‘Cal’ stands for the
calibration period, 1996-1998 and ‘Val’ stands for the validation period, 1999-2000, while
PT(I) and PT(II) represents the simulation results from Paper I and II respectively
using the PT method and SM represents the results from Paper II).
Akaki Hombole
Observed Cal mean = 13.2 (m3/s) Observed Cal mean = 52.9 (m3/s)
Observed Val mean = 12.9 (m3/s) Observed Val mean = 39.9 (m3/s)
Evaluator
Daily Monthly Daily Monthly
PT (I) PT(II) SM PT (I) PT(II) SM PT (I) PT(II) SM PT (I) PT(II) SM
Mean Cal 12.8 13.2 11.9 12.7 13.1 11.9 54.2 52.1 59.3 53.7 51.6 58.8
(m3/s)
Val 10.1 10.8 10.1 9.9 10.8 9.9 41.8 41.1 46.2 41.4 40.7 45.6
Cal 0.53 0.53 0.49 0.82 0.84 0.80 0.72 0.83 0.80 0.97 0.96 0.94
NSE
Val 0.55 0.44 0.50 0.82 0.75 0.77 0.71 0.79 0.73 0.91 0.93 0.87
Cal 0.53 0.53 0.50 0.83 0.85 0.83 0.75 0.83 0.82 0.97 0.97 0.95
R2
Val 0.59 0.46 0.55 0.86 0.78 0.84 0.74 0.79 0.76 0.92 0.93 0.91
PBIA Cal -2.9 0.3 -8.9 -2.6 0.6 -8.7 2.3 -1.7 12.0 2.2 -1.6 12.0
S Val -23.4 -17.4 -23.1 -23.3 -17.1 -22.9 3.6 1.86 14.3 3.6 1.8 14.1
18
help in balancing (cancel out) the bias showed that the catchments could, to some
occurred at different steps. extent, stabilize the input signals at this time
scale. The stream flow response seemed to
Model validation at interior points incorporate the influence of the catchment
The model performance on interior points, characteristics and damp the fluctuations of
based on a parent watershed calibration the rainfall, which could explain the better
(Paper I), is summarized in Table 5. The performance of the model at a monthly time
sub-watersheds in the table are listed in or- scale. The uncertainties that come from the
der of increasing drainage area. A noticeable spatial distribution of rainfall is also another
trend was an improvement in performance, possible reason as illustrated by the coeffi-
on both the NSE and R2 values, that was cients of the correlation analysis (Table 2).
directly proportional with the size of the The performance of the model at Akaki
sub-watersheds. The performance was poor sub-watershed was analyzed from the two
for the smaller interior points, in particular scenarios (Table 4 and Table 5); the first one
for Holeta and Berga. The possible explana- was the performance based on the calibra-
tion could be the insufficient capacity of tion of the model using the stream flow data
smaller basins to dampen out input signals of Akaki, and the second one was the per-
and consequent input errors as discussed in formance of the model at Akaki as an
Reed et al. (2004) and Shrestha et al. (2005). interior point. The comparison, based on all
The coefficient of variation for the daily the three model evaluators, showed that the
stream flow data showed that the smaller model performed better in the first case
sub-watersheds exhibited more variability (Table 4) for a daily time scale. The monthly
than the larger ones, which could affect the time scale comparison indicated relatively
accuracy of the simulation. According to poor performance in the first scenario
Hirpa et al. (2010), the statistical analysis on during the calibration period while it showed
daily flow data shows that the degree of better performance during the validation
multifractality of river flow decreases with period. The PBIAS indicated a better
increasing watershed area, which could point performance in all years for the first scena-
out the similarity between larger basins in rio. Even though the performance of the
preserving the different intensities of flow model mostly got better by using the stream
fluctuations. The study concludes that the flow data for calibration, this result might
watershed area is an important factor that also indicate the possibility of using the
controls the long memory of river flow model for ungauged interior points having
fluctuations. Hence, the direct transferring similar catchment characteristics with a
of parameter values should be based on gauged parent watershed.
catchment characteristic analysis that con-
sider the effects of the watershed area. On
the other hand, the monthly variation
Table 5. Model performance statistics for prediction of stream flow at four interior
points (Holeta, Berga, Akaki and Melka Kuntire) based on a parent watershed calibra-
tion (Hombole).
Holeta Berga Akaki Melka Kuntire
Evaluator Period
Daily Monthly Daily Monthly Daily Monthly Daily Monthly
Calibration -1.00 -0.20 0.01 0.59 0.46 0.83 0.63 0.79
NSE
Validation -0.38 0.29 -0.13 0.45 0.37 0.63 0.53 0.76
2 Calibration 0.10 0.39 0.21 0.64 0.47 0.87 0.67 0.89

R
Validation 0.05 0.37 0.08 0.38 0.43 0.73 0.55 0.82
Calibration 21.36 20.89 -8.87 -9.05 -17.77 -16.21 -33.37 -33.44

PBIAS
Validation -17.46 -17.37 3.21 3.58 -38.17 -36.91 -32.85 -33.01
19
Water balance of the upper Awash River simulation time scale, the short intensive
Basin storms are likely smoothed (averaged) and
The main components of the upper Awash surface runoff was not generated. On the
River basin simulated water budget were other hand, the study by Chekol (2006),
analyzed at the outlet of the watershed using the SM method in the SWAT model,
(Figure 8). The major component, as calcu- indicates that the base flow is the important
lated from a 100% rainfall input, was evapo- component with the amount of 9.8% from a
transpiration with an average 66% agreed in 100% rainfall input while surface runoff is
all the three simulations (two simulations 6.8%. The comparison indicates that the SM
with the PT method from the two different method consistently simulated the base flow
calibration schemes in Paper I and II and as a dominant contributor on annual basis
one simulation with the SM method in (both in the current study and in Chekol
Paper II). The major difference between the (2006)), while the PT method is in accor-
two soil moisture accounting methods (PT dance with the discussed results from
and SM), for this specific watershed is on Moreda (1999).
the accounting of surface runoff and base The analysis on annual basis (Paper I) for
flow on annual basis. The PT method each simulated year showed that the loss of
(Paper II) resulted in a 24 and 2% of surface water through evapotranspiration was do-
runoff and base flow respectively. The SM minant in all years. The amount of evapo-
method (Paper II) resulted in a 3.3 and transpiration lost from the total annual
20.5% of surface runoff and base flow precipitation reached at its maximum
respectively. These results were compared percentage in the driest year (1997). The
with previous works, Moreda (1999) and foremost contributor to the water yield was
Chekol (2006). According to Moreda (1999), the surface runoff with an exception of the
using a 10-day and a monthly lumped con- year 1999 where the groundwater flow was
ceptual rainfall-runoff model, the dominant greater, by about 1.2 mm. The soil water
flow is the inter flow (lateral flow) with 13% storage (including the vadose zone) was cal-
as compared to the base flow that accounts culated as the difference between the input
2%. The study explained that due to the rainfall and the different output components
Figure 8. Water balance components of the Hombole watershed with the correspond-
ing percentage from 100% precipitation input based on the three different simulations
for the calibration period.
20
(Figure 8). The result indicates a gaining realistic estimates of variables (in this case
storage (positive) with a more pronounced like soil moisture), a reduced uncertainty will
gain in the relatively driest years (1997 and be possible.
1999) and an exception lose (negative) in the The main limitation in using a distributed
year 1996. The change in the soil water model is the large number of parameters
storage exhibited a strong correlation with that needed to be optimized in order to
the surface runoff regime. The increase in reach an acceptable prediction of the output.
the amount of surface runoff was followed Even though the parameters are measurable
by a decrease in soil water storage. The trend in the field, it is not practicable to get error-
of this relation was also similar with the free measurements apart from the time and
other components. However, since the cost constraints. The scale of the measured
simulation was based only on the calibration parameters, that usually does not match the
of the streamflow, these results should model element or discretization scale, which
further be validated with soil moisture is much larger (Beven, 2001), also adds to
measurements. the limitations. During the calibration
The sensitivity of the cncoef parameter was process, the values of the parameters for
also checked on the calibrated model based best fit depend on the initial values, in this
on the annual water budget. The result case the SWAT default parameter values
indicated the possibilities of adjusting the designed for catchments in the United
surface runoff and groundwater flow com- States. The optimal set of parameters might
ponents with a single parameter calibration. be different with improved objective func-
Changing the parameter to 0.5 from the tion values with different initial conditions
calibrated 0.6 shifted some of the surface (Jones, 1997). The auto-calibration method
runoff contribution to the base flow while used in this study gave only one optimum
changing it to 0.7 resulted vice versa. Similar set of parameters. The existence of multiple
results have also been reported by Kannan et optimal parameter sets that can give as good
al. (2008). prediction of stream flow is usually possible,
which is explained in the concept of equifi-
Model prediction performance and
nality of model structures and parameters
uncertainties
(Beven, 1993; Beven, 2001; White and
The overall performance of the model Chaubey, 2005). In this study, the Akaki
depends on different factors. The forcing sub-watershed used the potential parameter
input data (like rainfall and temperature) and space (the SWAT default ranges) during the
the observations used to calibrate and vali- auto-calibration for all the free parameters.
date the model outputs (like the stream flow For Hombole two scenarios were tested: in
data) could be mentioned as the major Paper I a range were defined based on the
uncertainty sources. Apart from data quality, manual calibration and in Paper II all the
in a place like the Awash River basin with an parameter space were used. This might ex-
altitude range of 250 to 3600 m a.s.l., the plain the significant difference in the
extrapolation of rainfall from a distant gauge ALPHA_BF, REVAPMN and RCHRG_DP
is an obvious source of bias. The possibili- parameters between the two watersheds
ties of overcoming some of the drawbacks (Paper I) and also the differences in all the
that concerns the quality and spatial distribu- parameters between different simulations
tion of input data by making use of radar (Table 6).
rainfall data and remotely sensed soil mois-
ture data has been documented (e.g. Jacobs Even though the SWAT model is semi-
et al., 2003; Moon et al., 2004; Jayakrishnan et distributed based on HRU units, the para-
al., 2005). Moreover, in this study, observed meterization in this study was done in a
flow data was the only way of constraining more lumped way, mainly on sub-basin
the calibration and prediction process. scales, during the calibration process. The
Franks et al. (1998) suggested that in the case total number of HRUs was 399 and this
of availability of additional measured or large number made it difficult to calibrate
21
Table 6. The final estimate of, selected flow sensitive SWAT model parameters from
the three different calibration procedures at Hombole station in accordance with their
default value and calibration range. (PT (I) and PT (II) represents the calibration from
Paper I and II respectively while SM from Paper II).
Final Es-
Parameter Default Calibration Calibration Final Estimate
timate
Name Value Range Range
PT (I) PT (II) SM
ALPHA_BF 0.048 0-0.05 0.049 0-1 0.428 0.898
CN2 * - 20-0% - 17.5 ± 25% 3.1 - 25
ESCO 0.95 0.5-0.95 0.65 0-1 0.04 0.01
GW_DELAY 31 0-10 1 0-500 2 2
RCHRG_DP 0.05 0.05-1 0.305 0-1 0.49 0.05
SOL_AWC ** - 20-5% - 14 ± 25% 18.4 24.9
SOL_K ** - 20-5% - 10.3 ± 25% 16.03 - 25
SURLAG 4 0- 4 0.5 1-24 4.6 8.8
each of these separately. Apart from the could be modeled with good performance
spatial variation, the temporal variation is using the different methods.
also important for areas characterized by a Quantifying or predicting the different
distinct seasonal trend. The calibration hydrological components is the initial step
process for the curve number is especially for sustainable water resources planning and
sensitive for seasonal variation, in that the management. Hence, it is important to
performance of the model would be im- understand the water balance and quantify
proved at the expense of the peak flows and the dominant components in a watershed
vice versa. before planning and implementation. In this
The SM method is reported to predict too study, the water balance investigation shows
much runoff in shallow soils and soils with a disagreement between the two different
low storage (e.g. Neitsch et al, 2005, Kannan methods on the estimation of the contribu-
et al. 2008). According to Kannan et al. tion from the base flow. Even though, the
(2008) the calibrated model using the PT groundwater of this perennial river was
method performs better than the SM expected to sustain in dry seasons (for
method in the areas of low storage soils and instance not a single zero flow was recorded
shallow soils. The result of this study is also during the studied period), it is possible that
in accordance, which shows that the PT the base flow could be recharged from other
method improved the prediction of stream sources. Apart from local infiltration, springs
flow in the areas characterized by medium to and wetlands are among the potential
low storage. The significant factor that was sources. Moreover, faults and fractures
observed from the analysis was the capabili- could direct water from the neighboring
ty of the methods in mimicking the seasonal watersheds. Hence, these calls for further
variation of the rainfall and the retention study that accounts the influence of the
capacity of the soils that in turn govern the geological formations and structural beha-
surface runoff estimation. The quality of vior of the watershed that could assist in
climate data and its representativeness to the quantifying the groundwater potential.
catchment is vital since the core of this me-
thod was to depend more on antecedent C ONCLUSIONS
climate conditions rather than on the soil Distributed hydrological characterization of
storage. Further studies on similar water- a catchment is one of the important infor-
sheds would be necessary to validate the mation components to address the different
outcomes of this study and to help catego- questions in achieving a plan for sustainable
rizing the characteristic of a catchment that water management. Adapting existing
hydrological models, under limited data
22
conditions, is a feasible start to improve the indicated that there was a good agreement
reliability of stream flow prediction particu- between the observed and simulated
larly in developing countries, where data is discharge of the calibrated model, which was
often scarce. Even though, the develop- verified by higher values of coefficient of
ments of hydrological models are well ad- determination (R2) and Nash-Sutcliff effi-
vanced, the challenge in making use of them ciency (NSE) and good agreement in the
is the substantial data requirements. Collect- hydrographs. The model evaluation statistics
ing more data could be the ultimate long for stream flows gave acceptable results that
term solution while making better use of ranged from satisfactory to good, verified by
available information. However, a substan- NSE and R2 > 0.36 and PBIAS ± 25%.
tial improvement is not always related with a Some of the relatively poor model perfor-
bulk of data, rather with the relevancy of the mance could be due to poor quality of the
data and systematized collection system. A gauged climate variables as well as the very
complicating factor is though that the coarse spatial distribution of weather
Awash River passes through six regional stations in the sub-watersheds. Even though
states, which could make it difficult to the model performed in a satisfactory level,
implement a basin wise integrated water the performance level should not be genera-
resources management. For instance, the lized equally for all purposes. The daily
land management and development plans results are much more important than the
are convicted at regional scale or administra- monthly if the simulation is for flood
tive boundaries. The hydro-meteorological analyses and flood protection or prevention
data organization also follows this. Especial- plans. By contrast, for hydropower and irri-
ly, the meteorological stations are grouped gation purposes the monthly results could
according to political regions, which make it help for allocating and planning. Since the
difficult for researchers and practitioners to primary application of the SWAT model is
locate the representative stations in accor- for land management and agriculture, the
dance with the corresponding basins. This calibrated model can be used to analyze the
may hinder the usability of the available effects of change in land use and different
data. Monitoring and evaluating of ongoing management scenarios on stream flow
projects would also be possible with a regimes. The evaluation at the four interior
proper water management system. points indicated that care must be taken on
In the current study, an attempt was made to direct transferring of parameter values in
characterize the hydrological processes of association to watershed characteristics,
the Awash River basin, in Ethiopia, using with emphasis on drainage area.
statistical and deterministic approaches as a Analysis was also made on the different
contribution to fulfill the knowledge gap and options for estimating the retention parame-
the scarce information for future manage- ter used in surface runoff generation with
ment plans. The study focuses on the as- the Soil Conservation Service (SCS) curve
sessment of the applicability of the SWAT number method. One depends on the soil
model in the study area for reliable characte- water content (SM) and the other on antece-
rization of the catchment and the modeling dent climate (PT). While the predicted
uncertainties due to input data and model stream flow hydrographs showed an agree-
structure. The study could also be taken as a ment between the two methods, the
direction on where to focus from data predicted annual water balance indicated a
collection and quality control aspects. disagreement in quantifying the different
The model was successfully calibrated and hydrological components. The SM method
validated at the outlet and at one interior estimated higher surface runoff contribution
point. The performance of the model was than the groundwater while the PT method
also evaluated at four interior river gauge estimated the other way around. Hence,
stations. The comparison between the further investigation is recommended that
observed and simulated stream flow accounts the geological characteristics and
23
the sources of the base flow to make sure model results is unattainable. Hence, a con-
the occurrence of the groundwater in a suf- tinuous research on quantifying and allocat-
ficient amount for any future development. ing water resources is indispensable for
Overall, the present study indicates that building information databases for decision
proper care must be taken in selecting an makers. Thus, the management plan should
appropriate tool for quantifying the different be based on the availability of the resource,
water balance components, which would be both temporally and spatially, in accordance
used for decision making, especially for with the basin characteristics.
un-gauged catchments where validation of
24
R EFERENCES
Abbott MB, Bathurst JC, Cunge JA, O‘Connel PE, Rasmussen J. 1986. An introduction to the
European hydrological system-Systeme Hydrologique Europeen, SHE. 2: Structure of a physi-
cally based, distributed modelling system. Journal of Hydrology. 87:61-77.
Achamyeleh K. 2003. Ethiopia: Integrated flood management. In: Integrated Flood Management,
Technical support unit (eds.) WMO/GWP Associated Programme on Flood Management.
15p.
Alemayehu T. 2001. The impact of uncontrolled waste disposal on surface water quality in Addis
Abeba. Ethiopian Journal of Science. 24(1):93-104.
Anderson MP, Woessner WW. 1992. Applied Groundwater Modeling: Simulation of Flow and Advective
Transport. Academic Press: New York. 381p.
Ahmad S, Khan IH, Parida BP. 2001. Performance of stochastic approaches for forecasting river
water quality. Water Research. 35(18):4261-4266.
Antar MA, Elassiouti I, Allam MN. 2006. Rainfall-runoff modelling using artificial neural net-
works technique: a Blue Nile catchment case study. Hydrological Processes. 20:1201-1216.
Arnold JG, Sirinivasan R, Muttiah RS, William JR. 1998. Large area hydrologic modeling and
assessment. Part 1, model development. Journal of American Water Resources Association. 34:73-89.
Becker A, Serban P. 1990. Hydrological models for water-resources system design and operation.
WMO Operational hydrology report, Geneva. 34:34-80.
Bekele F. 1997. Ethiopian use of ENSO information in its seasonal forecasts. Internet Journal of
African Studies. 2.
Bergström S, Forsman A. 1973. Development of a conceptual deterministic rainfall-runoff mod-
el. Nordic Hydrology. 4:147-170.
Beven KJ. 2001. Rainfall-Runoff Modeling, The Primer. John Wiley and Sons Ltd.: England. 360 p.
Beven KJ. 1993. Prophecy, reality and uncertainty in distributed hydrological modeling. Advances
in Water Resources. 16:41-51.
Beven KJ, Kirkby MJ. 1979. A physically based variable contributing area model of basin hydrol-
ogy. Hydrological Sciences Bulletin. 24 (1):43–69.
Calver A, Wood WL. 1995. The Institute of Hydrology Distributed Model. In: Computer Models of
Watershed Hydrology, Singh VP (ed.) Water Resources Publications: Colorado; 595–626.
Chekol DA. 2006. Modeling of Hydrology and Soil Erosion of Upper Awash River Basin. PhD Thesis,
University of Bonn. 235 p.
Chow VT, Maidment DR, Mays LW. 1988. Applied Hydrology. McGraw-Hill: New York. 572 p.
Chow VT. 1964. Hydrology and its development. In: Handbook of Applied Hydrology, Chow VT
(ed.) McGraw-Hill: USA; 1-1 - 1-22.
Crawford NH, Linsely RK. 1966. Digital simulation in hydrology: The Stanford Watershed Mod-
el IV. Technical Report No. 39. Stanford University: Palo Alto, CA.
Croton JT, Barry DA. 2001. WEC-C: a distributed, deterministic catchment model—theory, for-
mulation and testing. Environmental Modelling and Software. 16:583–599.
Degefu W. 1987. Some aspects of meteorological drought in Ethiopia. In: Drought and Hunger in
Africa: Denying Famine a Future, Glantz M. (ed.) Cambridge University Press: Cambridge; 23-26.
Desalegn CE, Babel MS, Gupta AD, Seleshi BA, Merrey D. 2006. Farmers' perception of water
management under drought conditions in the Upper Awash Basin, Ethiopia. International Jour-
nal of Water Resources Development. 22(4):589-602.
Eckhardt K, Arnold JG. 2001. Automatic calibration of a distributed catchment model. Journal of
Hydrology. 251:103-109.
25
Emergency Appeal. 2006. Ethiopia: Floods. Preliminary appeal no. MDRET003, 18 August 2006.
International Federation of Red Cross and Red Crescent Societies.
EU. 2000. Directive 2000/60/EC of the European Parliament and of the Council of 23 October
2000, establishing a framework for Community action in the field of water policy. Official Jour-
nal of the European Communities. (L 327):1-72.
FAO. 2005. Irrigation in Africa in figures – AQUASTAT Survey 2005. FAO Water Reports 29,
Rome, Italy.
Flintan F, Tamrat I. 2002. Spilling Blood over Water? The Case of Ethiopia. In: Scarcity and Surfeit,
The Ecology of Africa’s Conflicts, Lind J, Sturman K (eds.) Institute for Security Studies: South
Africa; 242-319.
Franks SW, Gineste P, Beven KJ, Merot P. 1998. On constraining the predictions of a distributed
model: The incorporation of fuzzy estimates of saturated areas into the calibration process.
Water Resources Research. 34(4):787-797.
Govindaraju RS, Rao AR. 2000. Artificial Neural Networks in Hydrology. Kluwer Academic Pub.:
Netherlands. 348 p.
GWP. 2000. Integrated Water Resources Management. TAC Background Papers No. 4, Global
Water Partnership: Stockholm. 67 p.
Hargreaves GL, Hargreaves GH, Riley JP. 1985. Agricultural benefits for Senegal River Basin.
Journal of irrigation and Drainage Engineering. 11(2):113-124.
Hirpa FA, Gebremichael M, Over TM. 2010. River flow fluctuation analysis: Effect of watershed
area. Water Resources Research. 46:W12529.
ICWE. 1992. The Dublin statement and report of the conference. International Conference on
Water and the Environment: Development issues for the 21st century. 26-31 January 1992,
Dublin, Ireland.
IPCC. 2000. Land Use, Land-Use Change, and Forestry. Watson RT, Noble IR, Bolin B, Ravind-
ranath NH, Verardo DJ, Dokken DJ. (eds.) Intergovernmental Panel on Climate Change
(IPCC). Cambridge University Press: Cambridge, UK. 21 p.
Irvine K, Mills P, Bruen M, Walley W, Hartnett M, Black A, Tynan S, Duck R, Bragg O, Rowen
J, Wilson J, Johnston P, O‘Toole C. 2005. Water Framework Directive – An assessment of
mathematical modelling in its implementation in Ireland (2002-W-DS-11). Environmental
Protection Agency, Ireland. 508 p.
Jacobs JM, Myers DA, Whitfield BM. 2003. Improved rainfall/runoff estimates using remotely
sensed soil moisture. Journal of the American Water Resources Association (JAWRA). 39(2):313-324.
Jakeman AJ, Hornberger GM. 1993. How much complexity is warranted in a rainfall-runoff
model? Water Resources Research. 29(8):2637-2649.
Jayakrishnan R, Srinivasan R, Santhi C, Arnold JG. 2005. Advances in the application of the
SWAT model for water resources management. Hydrological Processes. 19:749-762
Jones JAA. 1997. Global Hydrology: processes, resources and environmental management. Addison Wesley
Longman Limited: England. 399 p.
Kannan N, Santhi C, Williams JR, Arnold JG. 2008. Development of a continuous soil moisture
accounting procedure for curve number methodology and its behaviour with different evapo-
transpiration methods. Hydrological Processes. 22:2114-2121.
Kebbede G. 2004. Living With Urban Environmental Health Risks: The Case of Ethiopia. Ashgate Pub-
lishing: England. 251p.
Kusre BC, Baruah DC, Bordoloi PK, Patra SC. 2010. Assessment of hydropower potential using
GIS and hydrological modeling technique in Kopili River basin in Assam (India). Applied Ener-
gy. 87:298-309.
26
Merritt WS, Letcher A, Jakeman AJ. 2003. A review of erosion and sediment transport models.
Environmental Modelling and Software. 18:761-799.
Minns AW, Babovic V. 1996. Hydrological modelling in a hydroinformatics context. In: Distri-
buted Hydrological Modelling, Abbott MB, Refsgaard JC (eds.) Kluwer Academic Pub.: Nether-
lands, 297-312.
Molle F, Jayakody P, Ariyaratne R, Somatilake H. 2005. Balancing irrigation and hydropower: A
case study from southern Sri Lanka. Research Report 94. International Water Management In-
stitute (IWMI): Colombo, Sri Lanka.
Moon J, Srinivasan R, Jacobs JH. 2004. Stream flow estimation using spatially distributed rainfall
in the Trinity River Basin, Texas. Transactions of the ASABE. 47(5):1445-1451.
Moreda F. (1999). Conceptual rainfall-runoff models for different time steps with special consideration for semi-
arid and arid catchments. PhD Thesis, Vrije Universiteit Brussel. 226 p.
Moriasi DN, Arnold JG, Van Liew MW, Bingner RL, Harmel RD, Veith TL. 2007. Model evalua-
tion guidelines for systematic quantification of accuracy in watershed simulations. Transactions
of the ASABE. 50(3):885-900.
Motovilov YG, Gottschalk L, England K, Rodhe A. 1999. Validation of distributed hydrological
model against spatial observations. Agricultural and Forest Meteorology. 98:257-277.
Mulligan M, Wainwright J. 2004. Modelling and model building. In: Environmental modeling: Finding
Simplicity in Complexity, Wainwright J, Mulligan M (eds.) John Wiley and Sons: England; 7-73.
Neitsch SL, Arnold JG, Kiniry JR, Williams JR. 2005. Soil and Water Assessment Tool Theoretical Do-
cumentation: version 2005. USDA Agricultural Research Service and Texas A&M Blackland Re-
search Center: Temple, TX. 476 p.
Piguet F. 2002. Afar region: A deeper crisis looms. United Nations, Emergencies Unit for Ethi-
opia. 13 p.
Piguet F. 2001. Even after good rains, Afar Pastoralists remain vulnerable. United Nations De-
velopment Programme, Emergencies Unit for Ethiopia. 42 p.
Prabu PC. 2009. Impact of heavy metal contamination of Akaki River of Ethiopia on soil and
metal toxicity on cultivated vegetable crops. Electronic Journal of Environmental, Agricultural and
Food Chemistry. 8(9):818-827.
Reed S, Koren V, Smith M, Zhang Z, Moreda F. 2004. Overall distributed model intercompari-
son project results. Journal of Hydrology. 298:27-60.
Refsgaard JC. 1996. Terminology, modeling protocol and classification of hydrological model
codes. In: Distributed Hydrological Modelling, Abbott MB, Refsgaard JC (eds.) Kluwer Academic:
Netherlands; 17-39.
Refsgaard JC, Storm B. 1996. Construction, calibration and validation of hydrological models. In:
Distributed Hydrological Modelling, Abbott MB, Refsgaard JC (eds.) Kluwer Academic: Nether-
lands; 41-54.
Salas JD. 1993. Analysis and modeling of hydrologic time series. In: Handbook of Hydrology, Maid-
ment DR (ed.) McGraw-Hill: USA; 19.1-19.72.
Shrestha RK, Sayama T, Tachikawa Y, Takara K. 2005. Use of disaggregated rainfall data for dis-
tributed hydrological modeling in Yodo River basin. Annuals of the Disaster Prevention Research In-
stitute, Kyoto University. 48B:43-51.
Sorooshian S. 1991. Parameter estimation, model identification and model validation: Conceptual
type models. In: Recent advances in the modeling of Hydrological Systems, Bowels DS, O‘Connell PE
(eds.) Kluwer Academic: Netherlands; 443-467.
Taddese G, Sonder K, Peden D. 2003. The water of the Awash River Basin a future challenge to
Ethiopia. International Livestock Research Institute (ILRI). Addis Abeba, Ethiopia. 13 p.
27
Tesfaye YG, Meerschaert MM, Anderson PL. 2006. Identification of periodic autoregressive
moving average models and their application to the modeling of river flows. Water Resource Re-
search. 42:W01419.
USDA Soil Conservation Service (SCS). 1972. National Engineering Handbook Section 4 hydrology.
United States Department of Agriculture (USDA): Washington, DC; 10.1-10.24.
UN 2000. United Nations Millennium Declaration. General Assembly, fifty fifth session, Agenda
item 60(b), A/RES/55/2.
UNEP. 2001. The Mesopotamian Marshlands: Demise of an Ecosystem. Partow H (ed.) Early
Warning and Assessment Technical Report, UNEP/DEWA/TR.01-3 Rev. 1. Division of Ear-
ly Warning and Assessment. United Nations Environment Programme. Nairobi, Kenya.
van Griensven A, Meixner T. 2007. A global and efficient multi-objective auto-calibration and
uncertainty estimation method for water quality catchment models. Journal of Hydroinformatics.
9(4):277-291.
van Griensven A, Meixner T, Grunwald S, Bishop T, Diluzio M, Srinivasan R. 2006. A global
sensitivity analysis method for the parameters of multi-variable catchment models. Journal of
Hydrology. 324:10-23.
Van Liew MW, Veith TL, Bosch DD, Arnold JG. 2007. Suitability of SWAT for the conservation
effects assessment project: Comparison on USDA agricultural research service watersheds.
Journal of Hydrologic Engineering. 12(2):173-189.
White KL, Chaubey I. 2005. Sensitivity analysis, calibration, and validation for a multisite and
multivariable SWAT model. Journal of the American Water Resources Association (JAWRA).
41(5):1077-1089.
Zinabu GM, Pearce NJG. 2003. Concentrations of heavy metals and related trace elements in
some Ethiopian rift-valley lakes and their in-flows. Hydrobiologia. 429:171-178.
Other references
ERSDAC 2009. ASTER Global Digital Elevation Model. Earth Remote Sensing Data Analysis
Center (ERSDAC). http://www.gdem.aster.ersdac.or.jp/index.jsp. Accessed in March 2010.
MDG reports 2010. Assessing Progress in Africa toward the Millennium Development Goals.
United Nations Development Programme (UNDP).
http://www.undp.org/mdg/reports.shtml Accessed in January 2011.
Ministry of Water and Energy, 2010. Projects and Programs: Dams and Hydropower. Ministry of
Water and Energy. Addis Abeba, Ethiopia.
http://www.mowr.gov.et/index.php?pagenum=4.3&pagehgt=1000px.
Accessed in February 2011.
28

FULLTEXT01

Uploaded by

Document Informationclick to expand document information

Document Informationclick to expand document information

Copyright:

Available Formats

FULLTEXT01

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

FULLTEXT01

Uploaded by

Copyright:

Available Formats

HYDROLOGICAL MODELING AS A TOOL

FOR SUSTAINABLE WATER RESOURCES

© Selome M. Tessema 2011

Reference to this publication should be written as: Tessema, SM (2011) Hydrological

Key words: Hydrological characteristics; Distributed hydrological modeling;

I NTRODUCTION pollution that further deteriorates the water

HBV (Bergström and Forsman, 1973), Stochastic models attempt to establish a

Figure 1. Location map of the study area.

investments in dams for hydropower pur- basin is estimated to be around 14 million as

Define Purpose Figure 2. The proce-

Sensitivity analysis and

Table 1. Summary of the physical characteristics of the five sub–watersheds and

Figure 4. Graphical presentation of average monthly temperature (°C), rainfall (mm),

Addis Abeba Bole 0.22 0.28 0.30 0.32 0.31

Figure 5. Hydrographs that compare observed and simulated stream flows

Table 3. Un-calibrated model performance measures, using the different combinations

Figure 6. Hydrograph plots comparing the observed with two simulated

Figure 7. Daily observed hydrographs at Akaki (top) and Hombole

PT (I) PT(II) SM PT (I) PT(II) SM PT (I) PT(II) SM PT (I) PT(II) SM

2 Calibration 0.10 0.39 0.21 0.64 0.47 0.87 0.67 0.89

Calibration 21.36 20.89 -8.87 -9.05 -17.77 -16.21 -33.37 -33.44

You might also like