See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/269632081

Analyzing runoff processes through conceptual hydrological modeling in the

Upper Blue Nile Basin, Ethiopia

Article  in  Hydrology and Earth System Sciences · December 2014

DOI: 10.5194/hess-18-5149-2014

CITATIONS READS

39 1,095

8 authors, including:

Mekete Dessie Niko Verhoest

Ghent University Ghent University
19 PUBLICATIONS   441 CITATIONS    309 PUBLICATIONS   9,807 CITATIONS   

SEE PROFILE SEE PROFILE

Valentijn Pauwels Teshager Admasu Negatu

Monash University (Australia) 10 PUBLICATIONS   251 CITATIONS   
145 PUBLICATIONS   4,487 CITATIONS   
SEE PROFILE
SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Geotourism View project

Hydro-climatic variability and flood-based farming in northern Ethiopia View project

All content following this page was uploaded by Niko Verhoest on 16 December 2014.

The user has requested enhancement of the downloaded file.

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014
www.hydrol-earth-syst-sci.net/18/5149/2014/
doi:10.5194/hess-18-5149-2014
© Author(s) 2014. CC Attribution 3.0 License.

Analyzing runoff processes through conceptual hydrological

modeling in the Upper Blue Nile Basin, Ethiopia
M. Dessie1,2 , N. E. C. Verhoest2 , V. R. N. Pauwels5 , T. Admasu3,4 , J. Poesen6 , E. Adgo3 , J. Deckers6 , and J. Nyssen4
1 School of Civil & Water Resources Engineering, Bahir Dar University, P.O. Box 430, Ethiopia
2 Laboratory of Hydrology and Water Management, Ghent University, Coupure links 653, 9000 Gent, Belgium
3 College of Agriculture & Environmental Sciences, Bahir Dar University, P.O. Box 79, Ethiopia
4 Department of Geography, Ghent University, Krijgslaan 281 (S8), 9000 Gent, Belgium
5 Department of Civil Engineering, Monash University, Clayton, Victoria, Australia
6 Department of Earth and Environmental Sciences, KU Leuven, Belgium

Correspondence to: M. Dessie (meketedessie.wossenie@ugent.be)

Received: 19 April 2014 – Published in Hydrol. Earth Syst. Sci. Discuss.: 20 May 2014
Revised: 14 October 2014 – Accepted: 11 November 2014 – Published: 12 December 2014

Abstract. Understanding runoff processes in a basin is of runoff as saturated excess flow prevails from the flat slope
paramount importance for the effective planning and man- areas. Overall, the model study suggests that identifying the
agement of water resources, in particular in data-scarce re- catchments into different runoff production areas based on
gions such as the Upper Blue Nile. Hydrological models rep- topography and including the impermeable rocky areas sep-
resenting the underlying hydrological processes can predict arately in the modeling process mimics the rainfall–runoff
river discharges from ungauged catchments and allow for an process in the Upper Blue Nile Basin well and yields a use-
understanding of the rainfall–runoff processes in those catch- ful result for operational management of water resources in
ments. In this paper, such a conceptual process-based hy- this data-scarce region.
drological model is developed and applied to the upper Gu-
mara and Gilgel Abay catchments (both located within the
Upper Blue Nile Basin, the Lake Tana sub-basin) to study
the runoff mechanisms and rainfall–runoff processes in the 1 Introduction
basin. Topography is considered as a proxy for the variabil-
ity of most of the catchment characteristics. We divided the The Upper Blue Nile Basin, the largest tributary of the Nile
catchments into different runoff production areas using to- River, covers a drainage area of 176 000 km2 and contributes
pographic criteria. Impermeable surfaces (rock outcrops and more than 50 % of the long-term river flow of the Main Nile
hard soil pans, common in the Upper Blue Nile Basin) were (Conway, 2000). The basin (Fig. 1a) drains the central and
considered separately in the conceptual model. Based on southwestern highlands of Ethiopia. The Ethiopian govern-
model results, it can be inferred that about 65 % of the runoff ment is pursuing plans and programs to use the water re-
appears in the form of interflow in the Gumara study catch- source potential of the basin for hydropower and irrigation
ment, and baseflow constitutes the larger proportion of runoff in an effort to substantially reduce poverty and increase agri-
(44–48 %) in the Gilgel Abay catchment. Direct runoff rep- cultural production. The Grand Ethiopian Renaissance Dam
resents a smaller fraction of the runoff in both catchments near the Ethiopian–Sudan border is currently under construc-
(18–19 % for the Gumara, and 20 % for the Gilgel Abay) and tion and several other water resource development projects
most of this direct runoff is generated through infiltration ex- are underway in its sub-basins.
cess runoff mechanism from the impermeable rocks or hard Owing to such rapidly developing water resource projects
soil pans. The study reveals that the hillslopes are recharge in the basin, there is an increasing need for the management
areas (sources of interflow and deep percolation) and direct of the available water resources in order to boost agricultural
production and to meet the demand for electrical power. Sus-

Published by Copernicus Publications on behalf of the European Geosciences Union.

5150 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

tainable planning and development of the resources depend els can benefit from them (Beven, 2001). Istanbulluoglu and
largely on the understanding of the interplay between the hy- Bras (2005) considered topography as a template for various
drological processes and the availability of adequate data on landscape processes that include hydrologic, ecologic, and
river discharges in the basin. However, the available hydro- biologic phenomena. This is more appealing to the Ethiopian
logical data are limited (for example, presently about 42 % highlands, in particular to the Upper Blue Nile Basin, as
of the Lake Tana sub-basin, the source of the Blue Nile, farming and farm drainage methodologies, soil and water
is gauged by the Ministry of Water Resources of Ethiopia). conservation works, soil properties, vegetation, drainage pat-
Furthermore, research efforts performed so far in the Upper terns and density, and even rainfall, are much linked to topog-
Blue Nile Basin with respect to the basin characteristics, hy- raphy in the Ethiopian highlands. Therefore, it remains nec-
drology and climatic conditions are scanty and fragmented essary to investigate the hydrological processes in the Blue
(Johnson and Curtis, 1994; Conway, 1997; Mishra and Hata, Nile Basin taking topography as a proxy for the variability
2006; Antar et al., 2006). Hydrological models that allow for of most of the catchment characteristics. The objective of
a description of the hydrology of the region play an impor- this paper is to study runoff mechanisms in the Upper Blue
tant role in predicting river discharges from ungauged catch- Nile Basin using topography as the dominant landscape com-
ments and understanding the rainfall–runoff processes in the ponent and classify a catchment (as steep, medium and flat
catchments in order to enhance hydrological and water re- slope areas) into different runoff production areas. The study
sources analysis. As such, a number of models have been tries to identify the dominant rainfall–runoff mechanism on
developed and applied to study the water balance, soil ero- the hillslopes (steep and medium slop areas) and the valley
sion, climate and environmental changes in the Blue Nile bottoms (flat areas). A considerable portion of the mountain-
Basin (e.g., Johnson and Curtis, 1994; Conway, 1997; Mishra ous areas in the Upper Blue Nile Basin consists of imperme-
and Hata, 2006; Kebede et al., 2006; Kim and Kaluarachchi, able rocks and hard soil pans, leading to a different runoff
2008; Collick et al., 2009; Steenhuis et al., 2009; Tekleab et process. This paper further investigates the contribution of
al., 2011; Tilahun et al., 2013). such landscapes in the rainfall–runoff process by including
The Soil and Water Assessment Tool (SWAT) and the Hy- a class for these impermeable rock and hard soil surfaces in
drologiska Byråns Vattenbalansavdelning Integrated Hydro- the conceptual hydrological model. This approach has not yet
logical Modelling System (HBV-IHMS) models have been been tested in the Upper Blue Nile Basin. However, similar
applied in the basin (Setegn et al., 2008; Wale et al., 2009; methodologies to the conceptual hydrological model devel-
Uhlenbrook et al., 2010). The SWAT model is based on the opment are discussed by Savenije (2010). Furthermore, it is
Soil Conservation Service (SCS) runoff curve number ap- necessary to obtain better quality river discharge data in the
proach, where the parameter values are obtained empirically basin. In this paper, we will face all these challenges. The
from plot data in the United States with a temperate climate. conceptual hydrological model for the rainfall–runoff stud-
Liu et al. (2008) studied the rainfall–runoff relationships for ies of the basin is calibrated using good-quality discharge
the three Soil Conservation Research Project (SCRP) water- data obtained from recently established measurement sta-
sheds (Hurni, 1984) in the Ethiopian highlands and showed tions. These outcomes positively add to the existing knowl-
the limitations of using such models, developed in temper- edge and contribute to the development of water resources
ate climates, in monsoonal Ethiopia. Adjusted runoff curve plans and decision making in the basin.
numbers for steep slopes with natural vegetation in northern
Ethiopia were reported by Descheemaeker et al. (2008).
Using a simple runoff-rainfall relation to estimate inflows 2 Description of study catchments
to the Lake Tana from ungauged catchments, Kebede et
al. (2006) computed the water balance of Lake Tana. How- The study catchments (Fig. 1b) where the model developed
ever, hills and floodplains were not differentiated in their sim- is applied are located in the Lake Tana Basin, the source of
plified runoff-rainfall relations. Mishra et al. (2004) and Con- the Blue Nile River. The Lake Tana Basin, located in the
way (1997) developed grid-based water balance models for northwestern Ethiopian highlands, with a catchment area of
the Blue Nile Basin, using a monthly time step, to study the 15 077 km2 (including the lake area), consists predominantly
spatial variability of flow parameters and the sensitivity of of the Gilgel Abay, Gumara, Rib and Megech rivers. About
runoff to climate changes. In both models, the role of to- 93 % of the annual inflow to Lake Tana is believed to come
pography was not incorporated, and in the model of Con- from these rivers (Kebede et al., 2006), and better under-
way (1997), soil characteristics are assumed spatially invari- standing of the hydrology of these rivers plays a crucial role
ant. Very few of the model studies discussed above classi- in efficient management of the lake and its basin. Two of the
fied the catchments into different hydrological regimes based sub-catchments (Gumara and Gilgel Abay) were selected for
on the relevant landscape characteristics to study the runoff this study in order to represent the hilly and mountainous
mechanisms and the hydrological processes in the basin. lands of the southern and eastern parts of the sub-basin as
Landscape characteristics can lead into conceptual struc- the bulk of it is located here (Fig. 1b), as well as to optimally
tures and relationships or the conceptual hydrological mod- use the available data. For both sub-catchments, large parts

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5151

of their territory are intensively cultivated. The lower flood- divisions of the catchment as it is quite inconsistent to sepa-
plains in these catchments with their buffering capacity are rate the groundwater system in the catchment. The catchment
not considered by this study, but were discussed by Dessie et buckets (reservoirs) and the conceptual runoff processes are
al. (2014). depicted in Fig. 2b and c.
The Gilgel Abay catchment (Fig. 1b) covers an area of Jothityangkoon et al. (2001) conceptualized the upper soil
1659 km2 at the gauging station near Picolo, with elevations layer (further referred to as the soil reservoir) as a “leaky
ranging between 1800 and 3524 m a.s.l. Soils are character- bucket”. By adding a groundwater reservoir (Krasnostein
ized by clay, clay loam and silt loam textures, each texture and Oldham, 2004), the conceptual model for modeling the
sharing similar proportions of the catchment area (Bitew and runoff at the catchment outlet was developed.
Gebremichael, 2011). The majority of the catchment is a In Fig. 2, Q1 [mm day−1 ] is the sum of direct runoff and
basalt plateau with gentle slopes, while the southern part has interflow in the soil reservoir; Q2 [mm day−1 ] is the baseflow
a rugged topography. from the groundwater reservoir; QSe2 is the direct runoff
The Gumara catchment covers part of the eastern side of from impermeable surface of the catchment; and the sum
the Lake Tana Basin. At its upper and middle portion, it has of Q1 , Q2 and QSe2 forms the total river discharge, Q
mountainous, highly rugged and dissected topography with [mm day−1 ], at the outlet of a catchment.
steep slopes. The lower part is a valley floor with flat to The water storage at any time t within the soil reser-
gentle slopes. Elevation in the catchment varies from 1780 voir, S(t) in mm, is determined by the precipitation (P , in
to 3700 m a.s.l. At the upper gauging station (Fig. 1b), the mm day−1 ), evapotranspiration (Ea , in mm day−1 ), and other
catchment area is 1236 km2 . Two independent studies found catchment-controlled outputs (Fig. 2c (i–iii)). When the stor-
very homogeneous textures of the soils in this catchment. age depth exceeds the field storage capacity (Sf , in mm),
BCEOM (1998) described it as dominantly clay with sandy precipitation is assumed to be partly transformed into sub-
clay soil at some places in the catchment, while soil data col- surface runoff, to represent inter- or subsurface flow (Qss ,
lected by Miserez (2013) show that texture is clay and clay in mm day−1 ), and partly into deep percolation or recharge
loam. In the hilly catchments, clay soils are essentially Niti- (R, in mm day−1 ) to the groundwater (Fig. 2c (ii)). When
sols, which do not present cracking properties as opposed to the soil reservoir fills completely, and the inflows exceed the
lowland Vertisols (Miserez, 2013). outflows, surface runoff (Qse1 , in mm day−1 ) is generated.
Based on rainfall data from the Dangila and Bahir Dar sta- Quantitatively, the depth of water stored in the soil, S(t),
tions, observed in the period 2000 to 2011, mean annual rain- evolves over time using the water balance
fall is ca. 1500 mm, with more than 80 % of the annual rain-
fall concentrated from June to September. Geologically, the S(t) = S(t − 1t) + (P − Ea − Qss − Qse1 − R)1t, (1)
catchments consist of Tertiary and Quaternary igneous rocks,
as well as Quaternary sediments. The rivers in the hilly areas where P is the precipitation [mm day−1 ], Ea is the ac-
are generally bedrock rivers, whereas in the floodplain the tual evapotranspiration [mm day−1 ], S(t − 1t) is the previ-
rivers meander and sometimes braid (Poppe et al., 2013). ous time step storage [mm], Qss is the interflow or subsur-
face runoff [mm day−1 ], Qse1 is the direct or overland flow
from the soil reservoir [mm day−1 ], R is deep percolation or
3 Model development recharge to the substrata and groundwater [mm day−1 ] and
1t is the time step equal to 1 day.
The model developed is based on a simple water balance Different studies show that part of the interflow water from
approach and the studies by Jothityangkoon et al. (2001), the steep hills appears at the hill bottoms during wet periods
Krasnostein and Oldham (2004) and Fenicia et al. (2008). in the form of increased moisture content or overland flow
The setup of this model is shown in Fig. 2. In this model- (Frankenberger et al., 1999; Bayabil et al., 2010; Mehta et
ing approach, the catchment is first split into soil surface and al., 2004; Tilahun et al., 2013). These findings reveal that
impermeable surface (these are areas with little or no soil the hill bottoms receive additional inputs to the soil reser-
cover and bedrock outcropping in the catchment as well as voir from the steep upper parts of the hills besides the rain-
soils with well-developed tillage pans). The runoff from the fall. In this modeling approach, it is assumed that steep hills
presumed impermeable areas is modeled as infiltration ex- first recharge the medium slope sections, and consequently
cess (Hortonian flow) runoff and is represented as QSe2. The the medium slope surfaces recharge the flat regions (valley
other component of the catchment, recognized as the soil sur- bottoms). The magnitude of the recharge (Qr , in mm d−1 ) is
face, is further divided into three using topographic criteria modeled as
(slope), considering topography as a proxy for the variabil-
ity of most of the catchment characteristics. Here, two reser- Qr = αQss , (2)
voirs are introduced (the soil reservoir and the groundwater
reservoir). The slow-reacting reservoir (or the groundwater where α (-) is interflow partitioning parameter and Qss is as
reservoir) is set to be common to all of the three slope-based defined above. Equation (1) is, therefore, modified for the

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5152 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

Figure 1. The Upper Blue Nile Basin and the Lake Tana sub-basin (a) and the study catchments and the gauging stations in the Lake Tana
sub-basin georeferenced on the SRTM DEM (b).

Figure 2. The modeling approach showing (a) divisions of a catchment into different runoff production areas; (b) conceptual model configu-
ration of the soil surface at an outlet of a catchment; and (c) inflows and outflows for the soil reservoir when the soil water storage capacity is
(i) below field storage capacity (ii) greater than field storage capacity and (iii) greater than the maximum soil water storage (after Krasnostein
and Oldham, 2004).

medium slope and flat surfaces as 3.2 Subsurface runoff

S(t) = S(t − 1t) + (P + Qr − Ea − Qss − Qse1 − R)1t. (3) Subsurface runoff, Qss [mm day−1 ], occurs only when the
storage depth exceeds the field storage capacity (Sf , in mm).
3.1 Actual evapotranspiration It is calculated as the difference between the storage and the
field storage capacity, divided by the response time (Tr ) of the
During wet periods, when the depth of available water ex- catchment with respect to subsurface flow (Jothityangkoon et
ceeds the maximum available soil storage capacity (Sb , in al., 2001):
mm), the actual evapotranspiration is equal to the potential
S(t) − Sf
evapotranspiration (Ep , in mm day−1 ). When S(t) is lower Qss = when S(t) > Sf , (6)
than Sb , Ea is assumed to decrease linearly with moisture Tr
content as follows (Steenhuis and van der Molen, 1986): Qss = 0 when S(t) ≤ Sf . (7)

S(t) The field storage capacity of the soil reservoir, Sf [mm], is

Ea = Ep ( ), (4) calculated using
Sb
Sb = Dφ, (5) Sf = Fc D, (8)

where D is the soil depth [mm] and φ is the soil porosity (-). where Fc (-) is the field capacity of the soil (dimensionless).

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5153

The catchment response time is the time taken by the ex-

cess water in the soil to be released from the soil and drained
out from the catchment. This response time depends on the
properties of the soil and the topography of the system, and
the subsurface flow velocity (Vb , in mm day−1 ) can be ex-
pressed as
L
Vb = , (9)
Tr
where L is the average slope length of the catchment [mm].
From Darcy’s law in saturated soils, Vb is also given as
Vb = Ks i, (10)
where Ks is the saturated hydraulic conductivity of the soil
[mm day−1 ] and i is the hydraulic gradient, which is approx-
imated by the average slope gradient (G) of the catchment.
Brooks et al. (2004) analyzed the variability of saturated
Figure 3. Typical surfaces with poor infiltration on hillslopes in
hydraulic conductivity with depth and found large Ks values the Gumara catchment: (a) shallow soil overlying bedrock and (b)
near the surface or root zone layer and the transmissivity that plough pan with typical plough marks. The occurrence of high
decreases exponentially with depth. Accordingly, a variation runoff response on these surfaces is evidenced by the presence of
is made between the upper soil layer (which affects inter- rill erosion (photos: Elise Monsieurs).
flow) and deep soil layer (percolation to groundwater) hy-
draulic conductivities. The permeability (K, in mm day−1 )
of the upper soil layer for the interflow under different soil the total water storage in the soil reservoir at each time
water conditions is modeled as step (Jothityangkoon et al., 2001; Krasnostein and Oldham,
−β S(t) 2004).
S
K = Ks,u (1 − e b ), (11)
S(t) − Sb
where β is a dimensionless parameter and Ks,u [mm day−1 ] Qse1 = when S(t) > Sb
(14)
1t
is the saturated hydraulic conductivity of the upper soil layer, Qse1 =0 when S(t) ≤ Sb
both of which are to be calibrated.
The response time (Tr ) in Eq. (6) is hence approximated 3.4 Surface runoff from the impermeable areas
from Eqs. (9), (10) and (11) as
L Field visits on the Upper Blue Nile Basin (including the study
Tr = , (12) catchments) revealed the existence of exposed surfaces that
GK
cause strong runoff response. These are areas with little or
where L and K are as defined in Eqs. (9) and (11) and G is no soil cover and bedrock outcropping in some parts of the
average slope gradient of the catchment. catchment as well as soils with well-developed tillage pans
The deep percolation or recharge to groundwater (R, in (Temesgen et al., 2012a, b) (Fig. 3). Hence, runoff from these
mm day−1 ) under varying soil water content conditions is almost impermeable areas is modeled as infiltration excess
modeled as (Hortonian flow) runoff with a very small amount of reten-
−γ S(t) tion before runoff occurs (Steenhuis et al., 2009). The surface
Sb
R = Ks,e (1 − e ), (13) runoff from these areas (QSe2, in mm day−1 ) is calculated as
where γ a dimensionless parameter, and Ks,e [mm day−1 ] is
the saturated hydraulic conductivity of the deep soil layer, QSe2 = P − Ep when P > Ep
(15)
which is to be estimated from the aquifer properties of the QSe2 = 0 when P ≤ Ep ,
catchments. This equation is identical to Eq. (11); therefore
in both cases it is assumed that conductivities vary exponen- where P and Ep [mm day−1 ] are as defined above. The im-
tially under varying soil water content conditions but with permeable portion of the catchment area (Ar , in km2 ) is mod-
different magnitudes. eled from the total catchment area (At , in km2 ) as

3.3 Saturated excess runoff Ar = λAt , (16)

Saturated excess runoff or surface runoff (Qse1 , in where λ is the fraction of impermeable surface within the
mm day−1 ) is calculated as the depth of water that exceeds catchment.

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5154 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

3.5 Groundwater reservoir and baseflow to recharge the streams from deep percolation of rainfall on
the catchments that produces baseflow of the rivers/streams.
The introduction of a deep groundwater storage (Fig. 2b) The storage effect of the streams when considered on the ba-
helps to improve low-flow predictions. This baseflow reser- sis of average daily flows of the streams is assumed to be
voir is assumed to act as a nonlinear reservoir (Wittenberg, negligible and hence streamflow routing was not considered
1999) and its outflow, Q2 [mm day−1 ], and storage, Sg [mm], for such smaller streams.
are related as
k1
3.6 Total river discharge
Sg(t)
Q2 = , (17)
1t The total river discharge (Qt , in mm day−1 ) at the outlet of
the catchments is given by
where k1 is a dimensionless model parameter. The water bal-
ance of the slow-reacting reservoir (groundwater reservoir) Qt = Qss + Qse1 + QSe2 + Q2 . (19)
is given by

Sg(t) = Sg(t−1t) + (R − Q2 )1t, (18) 4 Data inputs

where Sg(t) [mm] is the groundwater storage at the given time The data needed for the model are classified into three types:
step, Sg(t−1t) [mm] is the previous time step groundwater topographical, soil and hydrological data.
storage and R [mm day−1 ] is the deep percolation, as given
by Eq. (13). 4.1 Topographical data
In total the model has seven parameters:
Steenhuis et al. (2009) found that overland flow in the Blue
i. Parameters for the recharge (α1 and α2 ): in the three Nile Basin is generated from saturated areas in the relatively
slope classifications, α1 is to consider for the recharge flatter areas and from bedrock areas, while in the rest of
from the steep slope into the medium slope surface and the catchment all the rainfall infiltrates and is lost subse-
α2 is for the recharge from the medium slope surface quently as evaporation, interflow or baseflow. Topographical
into the flat surface. There is no parameter for the steep processes have been found to be the dominant factors in af-
slope surface since there is no surface that recharges it. fecting runoff in the Blue Nile Basin (Bayabil et al., 2010).
Therefore, there are two parameters for the three slope We used topography of catchments as the main criterion to
classifications. divide the catchment into different runoff production sur-
ii. Parameter for the impermeable surface of the catchment faces. Based on slope criteria (FAO, 2006), each study catch-
(λ): the catchment is divided into two surfaces (imper- ment was divided into three sub-catchments as steep (slope
meable surface with no or little soil cover and the soil gradient > 30 %), hilly or medium (slope gradient between
surface). The parameter λ is introduced to represent the 8 and 30 %) and flat (slope gradient < 8 %) to consider spa-
fraction of impermeable surface within the total catch- tial variability in catchment properties and runoff generation
ment and this part of the catchment is not classified as mechanisms (Fig. 4).
steep, medium slopes and flat surfaces since the classi- The 30 m×30 m resolution global digital elevation model
fication of this part of the catchment into such classes is (GDEM) was used to define the topography (downloaded
not important. Thus we have one parameter. from the ASTER website, http://earthexplorer.usgs.gov/).
The GDEM (Fig. 1b) was used to delineate and calculate the
iii. The parameters β, γ , k1 and Ks,u : the parameters β and average slope gradient and average slope length of the catch-
γ are introduced to account variability of permeability ments (topography-related inputs to the model).
and deep percolation of soil with soil water storage. k1
relates discharge and storage for the ground water and 4.2 Soil data
Ks,u is the saturated hydraulic conductivity in the upper
soil layer. We assumed that these parameters are less The model requires data on depth, porosity and field capacity
influenced by topography and each model parameter is of the soils. Soil depth and soil types data (Figs. 5 and 6)
assumed to be same for each slope classification of the were obtained from the Abay River basin integrated master
catchment. Moreover, it is quite inconsistent to sepa- plan study BCEOM (1998).
rate the groundwater system in the catchment. There- In this modeling philosophy, the soil depth is meant to rep-
fore, all the three slope-based classified sub-catchments resent the depth of water stored in the topmost layer (root
share the same groundwater reservoir. zone) of the soil (Fig. 2). The porosity and field capacity
of the soils were derived from the soil texture based on the
In this modeling approach, stream–groundwater interactions work of McWorter and Sunada (1997). From this, we deter-
are assumed to be minimal and the groundwater is assumed mined the soil textures of the study catchments (Table 1).

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5155

Figure 4. The three slope categories for the Gilgel Abay and Gumara catchments.

Figure 6. Soil depth in the Lake Tana Basin and the study catch-
ments (source: BCEOM, 1998).
Figure 5. Major soil types in the Lake Tana Basin and the study
catchments (source: BCEOM, 1998).
within and around the catchments (www.ethiomet.gov.et).
The location map of the rain gauge stations used for this
The saturated hydraulic conductivity for the deep percola-
study is depicted in Fig. 7. The data for most of the stations
tion (Eq. 12) was estimated using ranges of conductivities
are consistent and continuous, particularly for the first-class
given by Domenico and Schwartz (1990) for the saturated
stations like Dangila, Adet and Debretabor. However, we en-
hydraulic conductivities of a deep soil layer (colluvial mantle
countered gaps in some stations like Sekela station for some
on top of the igneous rock). A summary of the topographic,
periods in the year. In such instances, only the rainfall data
soil and saturated hydraulic conductivity data for the study
from the other stations were considered. Most of the rainfall
catchments is provided in Table 1.
stations in Gilgel Abay catchment are installed at the water
divides, and there is no station in the middle of the catchment.
4.3 Weather data
In this regard, the Gumara catchment has a higher density of
Daily precipitation is the key input meteorological data for rainfall stations. The areal rainfall distribution over the catch-
the model. Other meteorological data like minimum and ments was calculated using the Thiessen polygon method,
maximum air temperature, humidity, wind speed and dura- and the potential evapotranspiration was calculated using the
tion of sunshine hours were also used to calculate the poten- FAO Penman–Monteith method (Allen et al., 1998).
tial evapotranspiration, the other input variable to the model.
All weather data were obtained from the Ethiopian Na-
tional Meteorological Agency (NMA) for 13 stations located

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5156 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

Table 1. Input data on topography, soil and saturated hydraulic conductivities for the study catchments as classified into different hydrological
regimes using topography.

Coverage Average Saturated

from the soil Dominant hydraulic
Slope Average total area depth soil Field conductivity
Catchment class slope (%) (%) (m) texture Porosity capacity Ks,e (m s−1 )
level (≤ 8 %) 3.4 54 0.92 clay 0.46 0.36
Gilgel Abay hilly (8 % < slope ≤ 30 %) 15.9 38 1.29 clay to clay loam 0.42 0.32 9.26 × 10−8
steep (> 30 %) 41.4 8 1.49 clay loam to silt loam 0.4 0.26
level (≤ 8 %) 4.0 24 1.5 clay 0.46 0.36
Gumara hilly (8 % < slope ≤ 30 %) 17.2 60 1.24 loam , silty clay 0.42 0.26 1.16 × 10−8
steep (> 30 %) 41.5 16 1.2 sandy loam 0.25 0.1

ing rating curve equation for the Gumara catchment at the

gauging station (Wanzaye station) is

Q = 44.1h1.965 (R 2 = 0.997, n = 12), (21)

and that of the Gilgel Abay catchment at the Picolo station is

Q = 70.39h2.105 (R 2 = 0.985, n = 14) . (22)

Compared to the discharge data that have been gathered in

the past, the discharge data that are acquired for this study
are of superior quality, since a high time resolution during
the measurement has been used. This minimizes the risk of
missed peaks, particularly during the night. Furthermore, fre-
quent supervision was also conducted during the data collec-
tion campaign. Hence, these data were used for the model
Figure 7. Location map of rainfall stations for the study catchments. calibration. Discharge data collected before December 2011
were obtained for nearby stations from the Hydrology De-
partment of the Ministry of Water Resources of Ethiopia,
4.4 River discharge which has a long data record (since 1960) for these sta-
tions. However, the latter measurements were made using
Starting from July 2011, water level was measured at the staff gauge readings twice a day, with many data gaps and
Wanzaye station (11.788073◦ N, 37.678266◦ E) on the Gu- discontinuities, particularly at the end of the observation win-
mara River and from December 2011 at the Picolo station dow. The discharge data from 2000 to 2005 are relatively bet-
(11.367088◦ N, 37.037497◦ E) on the Gilgel Abay River. The ter and are used to validate the model.
water level measurements were made using mini-divers, au- The 2012 discharge data for Dirma catchment (outlet at
tomatic water level recorders (every 10 min), and manual 12.427194◦ N, 37.326209◦ E), collected in the same way as
readings from a staff gauge (three times a day, at 07:00, 13:00 those of Gilgel Abay and Gumara, were used to assess the
and 18:00), following the procedures described by Amanuel transferability of the model parameters.
et al. (2013).
Discharges were computed from the water levels using
rating curves (Eqs. 21 and 22) for each station. The rating 5 Calibration and validation
curves (Fig. 8) were calibrated based on detailed surveys of
the cross sections of the rivers and measurements of flow ve- The model calibration and validation were performed at a
locity at different flow stages, using the following commonly daily time step, and the hydrological data sets of 2012 and
used expression: 2011–2012 were used to calibrate the Gilgel Abay and Gu-
mara catchments, respectively. Discharge data of 2000–2005
Q = ahb , (20) were used for validation. There are seven calibration pa-
rameters in this model (Table 2), and the calibration was
where a and b are fitting parameters and Q [m3 s−1 ] and h performed using the particle swarm optimization (PSO) al-
[m] are discharge and water level, respectively. The result- gorithm. PSO is a population-based stochastic optimization

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5157

Figure 8. Stage–discharge relationship (rating curves) for Gilgel Abay at Picolo and Gumara at Wanzaye stations.

technique inspired by social behavior of bird flocking or fish lation (positive values indicate overestimation, whereas neg-
schooling (Kennedy and Eberhart, 1995). The advantages of ative values indicate model underestimation bias).
PSO are that the algorithm is easy to implement and that it is n
P
less susceptible to getting trapped in local minima (Scheer- (Qsim,i − Qobs,i )
linck et al., 2009). We carried out 50 iterations and 50 repeti- i=1
PBIAS = n ∗100 % (26)
tions, in total 2500 runs for each catchment to search for the P
Qobs,i
optimal value of the model parameters (Table 2) and 30 parti- i=1
cles were used in the PSO. The criterion in the search for the
optimal value was to minimize the root-mean-squared error The impacts of model parameters on the output of the
(RMSE) as the objective function, given by model when their values are different from the calibrated op-
v timal values were evaluated with respect to the RMSE for
u n
uP Gumara catchment. The sensitivity analysis was made by
u (Qobs,i − Qsim,i )2 randomly selecting parameter values in the region of the opti-
t i=1
RMSE = , (23) mal values obtained from PSO and calculating NSE for each
n
selected value. The applicability of the model to other un-
where Qobs is observed discharge [mm day−1 ], Qsim is simu- gauged catchments outside the study catchments in the Lake
lated or modeled discharge [mm day−1 ] and n is the number Tana Basin was also tested using direct parameter transfer-
of data points. The parameter values corresponding to the ability.
minimum “RMSE” were considered as optimum. From the
optimal model parameters, the performance of the model was 6 Soil and Water Assessment Tool (SWAT) and (FlexB )
also evaluated using (i) the Nash–Sutcliffe efficiency (NSE) models as benchmarks for comparison with
according to Nash and Sutcliffe (1970) and (ii) the coefficient Wase–Tana model
of determination (R 2 ):
n The two models are used as benchmark models to assess the
(Qsim,i − Qobs,i )2
P
performance of the model of this paper (hereafter referred to
NSE = 1 −
i=1
, (24) as the Wase–Tana model, in favor of the project name that
n
funded this study), which tries to use all available informa-
(Qobs,i − Qobs )2
P
i=1 tion and considers topography as a good proxy for the vari-
 2 ability of most of the catchment characteristics in the Upper
n

P
(Qsim,i − Qsim )(Qobs,i − Qobs )  Blue Nile Basin.
i=1
R2 = 
 
s
 n s  , (25) 6.1 SWAT model
n 
(Qsim,i − Qsim )2 (Qobs,i − Qobs )2
 P P 
i=1 i=1 SWAT is a basin-scale and continuous-time model used to
simulate the quality and quantity of surface and ground water
where Qobs [mm day−1 ] and Qsim [mm day−1 ] are the mean and predict the environmental impact of land use, land man-
observed and simulated discharges, respectively. agement practices and climate change (Arnold et al., 1998).
Percent bias (PBIAS) is used as an additional model per- The hydrological model is based on the water balance equa-
formance indicator. It measures the average tendency of the tion
simulated data to be larger or smaller than the observa- t
tions (Gupta et al., 1999). The optimal value of PBIAS is
X
SWt = SW0 + (Ri − Qi − ETi − Pi − QRi )1t, (27)
0, with lower absolute values indicating better model simu- i=1

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5158 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

Table 2. Model parameters, their ranges, and calibrated values found in 2500 iterations in the PSO calibration.

Parameter Explanation Units Minimum Maximum Calibrated values Average value of

Gumara Gilgel Abay both catchments
β parameter to account variability of per- – 1 3 2.445 2.314 2.380
meability of soil with soil water storage
k1 relates discharge and storage for the – 0.1 2 0.971 1.012 0.992
ground water
Ks,u Saturated hydraulic conductivity in the m s−1 0.001 0.1 0.016 0.05 0.033
upper soil layer
γ parameter to account variability of – 0.5 2 1.409 0.9 1.155
deep percolation with soil water
storage
λ coefficient that represents part of catch- – 0.05 0.5 0.149 0.173 0.161
ment that is impermeable
α1 interflow partitioning coefficient for – 0.05 0.8 0.653 0.575 0.614
the steep slope surface
α2 interflow portioning coefficient for the – 0.05 0.8 0.065 0.152 0.109
medium slope surface

where SWt is the soil water content at time t [mm]; SW0 is model and the various equations of the model can be found
the initial soil water content [mm]; 1t is the time step (day) in Fenicia et al. (2008).
and Ri , Qi , ETi , Pi and QRi are the daily amounts of pre- Calibration of this model was made using the PSO tech-
cipitation, runoff, evapotranspiration, percolation and return nique, following similar procedures of the Wase–Tana model
flow [mm day−1 ], respectively. calibration algorithm. The same objective function, RMSE,
In SWAT, a watershed is divided into homogenous hy- is also used in the search for the optimal value.
drologic response units (HRUs) based on elevation, soil,
management and land use, whereby a distributed parameter
such as hydraulic conductivity is potentially defined for each 7 Results and discussion
HRU. Hence, an analyst is confronted with the difficult task
of collecting or estimating a large number of input parame- 7.1 The daily hydrograph and model performance
ters, which are usually not available for regions like the Up-
per Blue Nile Basin. Details of the model can be accessed at 7.1.1 Wase–Tana model performance
the SWAT website (http://swatmodel.tamu.edu).
Automatic calibration and validation of the model was Figures 9 and 10 show a comparison of the modeled with the
made using SWAT-CUP. It is an interface that has been devel- observed discharge data for the two study catchments and for
oped for SWAT automatic calibration and model uncertainty both the calibration and validation periods.
analysis (Abbaspour et al., 2007). R 2 and NSE were used Despite the possible spatial variability of some input data
as objective functions during the calibration process of the (average soil and rainfall data are considered) and the sim-
search for the optimal value. plicity of the model, discharge is reasonably well simulated
during both the calibration and validation periods. This can
6.2 FlexB model be seen from the visual inspection of the hydrographs and
from the model performance indicators (Table 3).
This model is a lumped conceptual type and is character- The NSE of the model is high for both catchments. In
ized by three reservoirs as described by Fenicia et al. (2008): the calibration period, NSE equals 0.86 for Gumara catch-
the unsaturated soil reservoir (UR), the fast-reacting reser- ment and 0.84 for Gilgel Abay catchment, while they are
voir (FR) and the slow-reacting reservoir (SR). The model 0.78 and 0.7, respectively, during the validation period. Fig-
has eight parameters: a shape parameter for runoff genera- ures 9 and 10 also show that the model simulates the over-
tion β [-], the maximum UR storage Sfc [mm], the runoff all behavior of the observed streamflow hydrographs well.
partitioning coefficient D [-], the maximum percolation rate However, an overestimation of the large flood peaks for the
Pmax [mm h−1 ], the threshold for potential evaporation Lp Gilgel Abay catchment is found for the validation period. In
[-], the lag times of the transfer functions Nlag [h], and the the calibration period for this catchment, the model errors
timescales of FR and SR: Kf [h] and Ks [h]. Details of the tend to increase during wetting-up periods for almost all the

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5159

Figure 9. Comparison of predicted and observed discharge and precipitation of the Gumara and the Gilgel Abay catchments for the calibration
period.

Figure 10. Predicted and observed discharges and precipitation of the Gumara and the Gilgel Abay catchments for the validation period.

models. Initially, the soils are relatively dry and most of the 0.79 to 0.85 for the Gilgel Abay catchment, for the validation
rainfall during the beginning of the rainy season is not ef- and calibration periods, respectively. Generally, the modeled
fective to produce runoff in the model as the soil reservoir discharges appear to be less variable over time than the ob-
has to be filled first to generate the faster component of the servations, as shown by the standard deviations in Table 3.
runoff. Besides model uncertainties, the rainfall data quality This is likely due to the fact that data used in the model are
can also affect the model performance, mainly in the case of averaged over the year, while observed river discharges are
the Gilgel Abay catchment. The R 2 values for the time series highly seasonal. We used average daily rainfall data, average
of daily streamflow between simulated and observed values soil data (e.g., porosity, field capacity and soil depth), aver-
were from 0.80 to 0.86 for the Gumara catchment, and from age catchment characteristics data (e.g., slope, slope length)

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5160 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

Table 3. Statistical comparison and model performance of the modelled and observed river discharge (Q) for the two catchments.

Model performance indicators

Standard
Mean Q Deviation RMSE1
[mm day−1 ] [mm day−1 ] [mm day−1 ] NSE2∗ R2 PBIAS3
Gumara
calibration (2011–2012) 2.31 3.79 – – – –
Observed
validation (2000–2005) 2.3 3.75 – – – –
data
Gilgel Abay
calibration (2012) 3.89 5.05 – – – –
validation (2000–2005) 2.33 3.4 – – – –
Gumara
calibration (2011–2012) 2.37 3.56 1.34 0.86 0.86 3.30
Wase-Tana
validation (2000–2005) 1.95 3.05 1.37 0.78 0.8 −11.75
model
Gilgel Abay
calibration (2012) 3.85 4.7 1.85 0.84 0.85 −21.61
validation (2000–2005) 3.14 3.71 1.67 0.7 0.8 34.06
Gumara
calibration (2011–2012) 1.91 3.33 1.55 0.77 0.78 −17.50
SWAT
validation (2000–2005) 1.62 3.11 1.63 0.72 0.75 −29.48
model
Gilgel Abay
calibration (2012) 2.02 3.20 1.40 0.60 0.79 −44.01
validation (2000–2005) 2.45 3.86 2.30 0.55 0.63 5.45
Gumara
calibration (2011–2012) 2.43 3.64 1.54 0.82 0.82 5.30
FlexB
validation (2000–2005) 2.01 3.35 1.47 0.80 0.81 −12.67
model
Gilgel Abay
calibration (2012) 3.81 4.03 1.62 0.80 0.84 5.64
validation (2000–2005) 4.13 4.33 2.15 0.50 0.75 77.67
1. RMSE: Root Mean Squared Error as defined in Eq. (23). 2*. NSE: Nash–Sutcliffe Efficiency as defined in Eq. (24). 3. PBIAS: Percentage Bias as
defined in Eq. (26).

to mention some for the model inputs. Hence, this averaged bration and consequently the validation simulations. On the
condition may be one source of error such that the model other hand, the likely reason for decreased performance of
may not exactly mimic extremes like peak discharges. the FlexB model for the Gilgel Abay catchment is the over-
simplification of the catchment heterogeneity, since it is a
7.1.2 Performance in comparison with the lumped one and the impact is greater when the catchment be-
benchmark models comes larger (Gilgel Abay catchment is larger than Gumara
catchment).
For the calibration period, almost all the three models per- A look at the flow duration curves (Figs. 11 and 12) in-
formed quite well (Table 3). However, an appreciable de- dicates the higher uncertainty of the two benchmark models
crease in model performance has been noticed for the vali- (mainly SWAT model) with respect to low-flow predictions.
dation period in Gilgel Abay catchment for the benchmark In relative terms, Wase–Tana model offers more flexibility
models. SWAT is a physically based complex model, requir- in adapting the model to the catchments based on the vali-
ing extensive input data, which is a challenge for data-scarce dation simulation performances. This can be attributed to the
regions like the Upper Blue Nile Basin. The model simu- consideration of topography-driven landscape heterogeneity
lations can only be as accurate as the input data. This sug- analysis and catchment information extraction for the model,
gests that the coarser data input used for the model in the which strengthens the hypothesis that the topography-driven
study catchments might have significantly affected the cali-

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5161

Figure 11. Predicted and observed flow duration curves of the Gumara and the Gilgel Abay catchments for the calibration period.

Figure 12. Predicted and observed flow duration curves of the Gumara and the Gilgel Abay catchments for the validation period.

model structure and use of all available information on hy- 7.2 The hydrograph components and hydrological
drology based on topography is a good choice for the Upper response of the catchments
Blue Nile Basin. From a comparison of four model structures
on the Upper Heihe in China, Gao et al. (2014) also con- This hydrological model (the Wase–Tana model) is based on
firmed that topography-driven model reflects the catchment the generation of direct runoff from saturated and imperme-
heterogeneity in a more realistic way. able (degraded surfaces and rock outcrops with little or no
soil cover) areas, interflow from the soil storage in the root
zone layer and baseflow from the deeper layer as groundwa-
ter storage. The understanding of the relative importance of
these processes on the hydrological response of each catch-
ment is still unknown. The mean annual surface runoff (Qse ,

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5162 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

sum of Qse1 and QSe2), interflow or subsurface flow (Qss )

and baseflow (Q2 ) components of the total daily hydrograph
computed by the model for the calibration and validation pe-
riods are given in Table 4.
The total mean annual runoff generated by the model is
in line with the observations for both catchments in the cal-
ibration period (Table 4), while an appreciable difference is
noticed in the values for the Gilgel Abay catchment in the
validation period. One of the problems in accurate model-
ing of the discharge is that precipitation measurements do
not cover well the catchments. This is particularly the case
for the Gilgel Abay catchment, where the rainfall stations
are poorly distributed as most of the meteorological stations
lie near the water divides. The calibration results are better,
since the data from the recently established precipitation sta- Figure 13. One of the springs in Gilgel Abay catchment used as a
tions (e.g., Durbetie) could be used. There are also doubts water supply source for Bahir Dar.
about the representativeness of the discharge data used for
the validation of the model, because the water level measure-
ments were made manually and twice daily (in the morning properties, which clarifies the large groundwater potential. In
and late afternoon), leading to the possibility of missing flash line with this, several large springs exist in the catchment, in-
floods at other moments of the day as the stream discharge is cluding one that is used as a source of water supply for the
very variable. This can be clearly seen from the mean annual city of Bahir Dar (Fig. 13).
observed flows during the calibration and validation periods The other interesting result is that direct runoff is the
for Gilgel Abay. The mean annual observed flow in the val- smallest fraction of the total runoff for both catchments (18–
idation period was found to be much smaller than the cor- 19 % for Gumara and 20 % for Gilgel Abay) and almost all
responding flow during the calibration period (Table 4). The peak flow incidences are associated with direct runoff. More
closer total mean annual runoff values and the better model than 90 % of this direct runoff is found to be from the rel-
performance indicators for the Gumara catchment during the atively impermeable (degraded areas, plough pans or rock
calibration period suggest that the model can perform satis- outcrops with little or no soil cover) surfaces. The calibrated
factorily with better input discharge and precipitation data. result shows that this type of runoff production area covers
From PBIAS results (Table 3), the FlexB model showed 15 % of the Gumara and 17 % of the Gilgel Abay catchments,
overestimated bias and the SWAT model behaved the oppo- respectively. In a similar study, Steenhuis et al. (2009) men-
site for both catchments during the calibration period. tion that the rock outcrops occupy 20 % of the total catch-
Despite the variations in mean annual runoff generated by ment area in the Abay (Blue Nile) catchment at the Ethiopia–
the Wase–Tana model, the partitioning of the total runoff into Sudan border upstream of the Rosaries Dam, which is very
the different components (Table 4) in each period is almost similar to the result of Gilgel Abay catchment in this study.
identical for each catchment, as expected. About 65 % of the The remaining direct runoff is generated from the flat
runoff appears in the form of interflow for the Gumara catch- slopes of the catchments as saturated excess runoff, prob-
ment, and baseflow takes the larger proportion for Gilgel ably near the valley bottoms. The hillslopes (medium and
Abay catchment (44–48 %). Uhlenbrook et al. (2010) found steep slope source areas in this paper) generated almost no
the baseflow to be about 32 % from similar model study re- direct runoff as saturated excess flow. Similar results were
sults for Gilgel Abay catchment. Vogel and Kroll (1992) have obtained by different researchers in the Blue Nile Basin, who
shown that baseflow is a function of catchment area, and ge- identified hillslopes as main recharge areas (Steenhuis et al.,
omorphological, geological and hydrogeological parameters 2009; Collick et al., 2009; Tilahun et al., 2013). Our results
of the catchment have a linear incidence on the discharges. contribute to the debate on the relative importance of sat-
The difference between the baseflow of the two catchments urated excess runoff versus infiltration excess runoff (Hor-
is high, despite their comparable catchment sizes, suggesting tonian overland flow) mechanisms in the Upper Blue Nile
rather the different structure, functioning and hydrodynamic Basin, showing that the rainfall–runoff processes are better
properties of the two catchments. Hence, the model results represented by the soil reservoir methodology. However, fur-
reveal that the groundwater in the Gilgel Abay catchment ther research is necessary that involves rainfall intensity and
receives more recharge and makes a greater contribution to event-based analysis of hydrographs.
the river flow. This is in line with Kebede (2013) and Poppe
et al. (2013), who showed that the largest part of the Gilgel
Abay catchment consists of pumice stones and fractured qua-
ternary basalts with a high infiltration capacity and hydraulic

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5163

Table 4. Model results on the hydrograph components of the catchments.

Runoff components Unit For the calibration period For the validation period
Gumara Gilgel Abay Gumara Gilgel Abay
Total mean annual runoff predicted (Qpr ) mm year−1 864 1405 713 1146
Total mean annual runoff observed (Qob ) mm year−1 843 1420 841 938
Mean annual surface runoff (Qse ) mm year−1 161 280 129 234
% from the total Qpr 19 20 18 20
Mean annual interflow (Qss ) mm year−1 574 508 458 369
% from the total Qpr 66 36 64 32
Mean annual baseflow (Q2 ) mm year−1 128 617 126 548
% from the total Qpr 15 44 18 48

reduced without appreciable impact on the model (Fenicia

et al., 2008). k1, Ks,u and λ are very sensitive parameters in
this model and the model performance drops abruptly if the
parameters exceed a particular threshold value (Fig. 14).
The global sensitivity analysis (Fig. 15), however, shows
interactions among all the input parameters of the model. Al-
though global sensitivity analysis reveals details of the model
behavior in a more general sense through random parame-
ter sampling and that the parameters are all sensitive, the lo-
cal sensitivity analysis indicates that moderate variations in
the parameter values for some parameters can still drastically
change the model performance.
The model parameter transferability to other ungauged
catchments in the basin has been tested by analyzing the
variability among the calibrated parameters of the two catch-
ments. Table 2 shows that the calibrated parameters are
nearly identical for both catchments, except for γ and λ,
which are related to deep percolation and impermeable frac-
tion of the catchment, respectively. As described above, they
Figure 14. Local model parameter sensitivity analysis for Gumara affect the baseflow and direct runoff contributions to the to-
catchment. Parameters are explained in Table 2. tal river flow. However, we showed that the contributions of
these components to the total runoff are relatively small and
γ is poorly sensitive to a wide range of values. Thus the in-
7.3 Transferability of model parameters to other fluence of these parameters is expected to be minimal. This
ungauged catchments and sensitivity is verified by generating flows using the average of the cal-
ibrated parameters of the two catchments and analyzing the
The sensitivity analysis was performed on model parameters effect on the model performance indicators (Table 5). The
for Gumara catchment with respect to the RMSE. model performance obtained using the average model param-
The parameters β, α1 and γ show poor sensitivity for a eter values is similar to the results found using the optimal
wide range of values with respect to the local sensitivity anal- model parameters (Table 3). To further verify the adaptabil-
ysis. The local sensitivity analysis shows the sensitivity of a ity of the average calibrated model parameter values outside
variable to the changes in a parameter if all other parameters the study catchments and see the impacts of scale, we applied
are kept constant at some value (optimal value in this case). the average parameter values to another catchment (Dirma
An increase in the value of β beyond 1.4 showed almost no catchment in the northern part of the Lake Tana sub-basin,
sensitivity, while the model efficiency decreased slightly af- Fig. 1) with an area of 162.6 km2 . Encouraging model effi-
ter an increase in the value of γ from the optimum. This ciency could be obtained, with NSE and R 2 values of 0.58
means that there is little confidence in the model’s correspon- and 0.6, respectively (Table 5). This is to be elaborated fur-
dence with these parameters and that the parameters can be

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5164 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

Figure 15. Global model parameter sensitivity analysis results for Gumara catchment. Parameters are explained in Table 2.

Table 5. Comparison of model performance between the optimal and average model parameters of the three catchments.

Model performance for the

Model performance for the average of the parameters of
Catchment optimal model parameters optimal model the two catchments
RMSE RMSE
[mm day−1 ] NSE R2 [mm day−1 ] NSE R2
Gumara calibration period 1.34 0.86 0.86 1.48 0.84 0.86
validation period 1.37 0.78 0.80 1.82 0.76 0.77
Gilgel Abay calibration period 1.85 0.84 0.85 1.98 0.83 0.84
validation period 1.67 0.70 0.80 1.93 0.68 0.78
Dirma for the 2012 discharge – – – 1.79 0.58 0.60

ther in the future, involving more catchments and more years ies (Sivapalan, 2009; Savenije, 2010; Gao et al., 2014). The
of data. results suggest the possibility of directly using the average
In general, transferability results showed good perfor- model parameter values for other ungauged catchments in
mance of the daily runoff model in the two study catchments the basin, even though further tests on such catchments are
and an average performance in the test catchment (Dirma still recommended. However, we believe that this is a useful
catchment). This can be explained by the fact that effort was result for operational management of water resources in this
made to incorporate more knowledge in the model structure data-scarce region.
to increase model realism. We based our model strongly on
the soil storage characterization of the soil reservoir in the
rainfall–runoff process and representation of the maximum 8 Conclusions
storage of the unsaturated reservoir at the catchment scale,
which is closely linked to rooting depth and soil structure In this paper, a simple conceptual semi-distributed hydrolog-
and strongly depends on the ecosystem. Transferability of ical model was developed and applied to the Gumara and
the model has benefited from this in that we were able to Gilgel Abay catchments in the Upper Blue Nile Basin, Lake
derive most of the input data from the test catchments. The Tana sub-basin, to study the runoff processes in the basin.
consideration of topography-driven landscape heterogeneity Good-quality discharge data were collected through a field
analysis and catchment information extraction based on to- campaign using automatic water level recorders with high
pography (slope) for the model is another reason for the bet- time resolution. We used the topography and soil texture data
ter performance of the model transferability. The role of to- of the catchments as the dominant catchment characteris-
pography in controlling hydrological processes and its link- tics in the rainfall–runoff process. In the model, a distinc-
age to geology, soil characteristics, land cover and climate tion is made between impermeable surfaces (degraded sur-
through coevolution have been indicated in different stud- face or exposed rock with little or no soil cover) and per-
meable (soil) surfaces as different types of source areas for

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5165

runoff production. The permeable surfaces were further di- Acknowledgements. We wish to acknowledge the efforts of the
vided into three subgroups using topographic criteria such as field staff and data collectors for their help during the installation of
flat, medium, and steep slope areas. The rainfall–runoff pro- monitoring stations and data recording. We thank the project staff
cesses were represented by two reservoirs (soil and ground- and MSc students for the logistic help and valuable field inputs.
water reservoirs) and the water balance approach was used to We are grateful to the Ministry of Water and Energy and National
Meteorological Agency of Ethiopia for making data available. This
conceptualize the different hydrological processes in each of
research was supported by the Belgian Development Cooperation
the two reservoirs. Such a detailed form of modeling, using (VLIR-UOS, WASETANA project).
topography as a dominant landscape characteristic to classify
a catchment into different hydrological regimes, has not been Edited by: R. Merz
applied yet in the Upper Blue Nile, Lake Tana sub-basin.
We demonstrated that the model performs well in simulat-
ing river discharges, irrespective of the many uncertainties.
Model validation indicated that the Nash–Sutcliffe values for References
daily discharge were 0.78 and 0.7 for the Gumara and Gilgel
Abbaspour, K. C., Yang, J., Maximov, I., Siber, R., Bogner, K.,
Abay catchments, respectively.
Mieleitner, J., Zobrist, J., and Srinivasan, R.: Modelling hydrol-
We were able to partition the total runoff into a fast com- ogy and water quality in the pre-alpine/alpine Thur watershed
ponent (direct runoff and interflow) and a slow component using SWAT, J. Hydrol., 333, 413–430, 2007.
(baseflow) and estimated the contributions of each compo- Allen, R. G., Pereira, L. S., Raes D., and Smith, M.: Crop evapotran-
nent for the catchments. About 65 % of the runoff appears in spiration. Guide- lines for computing crop water requirements,
the form of interflow for the Gumara catchment, and base- FAO Irrigation and Drainage Paper 56, FAO, Rome, 1998.
flow is responsible for the larger proportion of the discharge Antar, M. A., Elassiouti, I., and Allam, M. N.: Rainfall-runoff mod-
for the Gilgel Abay catchment (44–48 %). Direct runoff gen- eling using artificial neural networks technique: a Blue Nile
erates the lower fraction of runoff components in both catch- catchment case study, Hydrol. Process., 20, 1201–1216, 2006.
ments (18–19 % for the Gumara and 20 % for the Gilgel Arnold, J. G., Srinivasin, R., Muttiah, R. S., and Williams, J. R.:
Large Area Hydrologic Modeling and Assessment: Part I. Model
Abay) and almost all peak flow incidences are associated
Development, JAWRA J. Am. Water Resour. Assoc., 34, 73–89,
with direct runoff. More than 90 % of this direct runoff is
1998.
found to be from the relatively impermeable (plough pan Bayabil, H. K., Tilahun, S. A., Collick, A. S., and Steenhuis, T. S.:
or rock outcrops with little or no soil cover) source areas. Are runoff processes ecologically or topographically driven in
The hillslopes (medium and steep slope source areas) are the Ethiopian Highlands? The case of the Maybar, Ecohydrology,
recharge areas (sources of interflow and deep percolation) 3, 457–466, doi:10.1002/eco.170, 2010.
and generated almost no direct runoff as saturated excess BCEOM: Abay River Basin Integrated Development Master Plan
flow. – Phase 2 – Land Resources Development – Reconnaissance
The results of this study, with comparisons to two bench- Soils Survey, Ministry of Water Resources: Addis Ababa, 208
mark models, clearly demonstrate that topography is a key pp., 1998.
landscape component to consider when analyzing runoff pro- Beven, K.: On landscape space to model space mapping, HP today,
Hydrol. Process., 15, 323–324, 2001.
cesses in the Upper Blue Nile Basin. Generally, runoff in the
Bitew, M. M. and Gebremichael, M.: Assessment of satellite rainfall
basin is generated both as infiltration and saturation excess products for streamflow simulation in medium watersheds of the
runoff mechanisms. A considerable portion of the landscape Ethiopian highlands, Hydrol. Earth Syst. Sci., 15, 1147–1155,
in the Upper Blue Nile Basin consists of impermeable rock doi:10.5194/hess-15-1147-2011, 2011.
outcrops and hard soil surfaces (15–17 % of the total catch- Brooks, E. S., Boll, J., and McDaniel, P. A.: A hillslope-scale exper-
ment area as per the results of this study) and they are the iment to measure lateral saturated hydraulic conductivity, Water
sources of most of the direct runoff. This conceptual model, Resour. Res., 40, W04208, doi:10.1029/2003WR002858, 2004.
developed to study the runoff processes in the Upper Blue Collick, A. S., Easton, Z. M., Ashagrie, T., Biruk, B., Tilahun, S.,
Nile Basin, may help to predict river discharge for ungauged Adgo, E., Awulachew, S. B., Zeleke, G., and Steenhuis, T. S.:
catchments for a better operation and management of wa- A simple semidistributed water balance model for the Ethiopian
ter resources in the basin, owing to its simplicity and parsi- highlands, Hydrol. Process., 23, 3718–3727, 2009.
Conway, D.: A water balance model of the upper Blue Nile in
monious nature with respect to parameterization. The runoff
Ethiopia, Hydrol. Sci., 42, 265–282, 1997.
processes in the basin are also found to be affected much by Conway, D.: The Climate and hydrology of the Upper Blue Nile
the rainfall, as the performance of the model was better for River, The Geographical J., 166, 49–62, 2000.
those study catchments where coverage of rainfall stations Descheemaeker, K., Poesen, J., Borselli, L., Nyssen, J., Raes, D.,
was good. Hence a better spatial and temporal resolution of Mitiku Haile, Muys, B., and Deckers, J.: Runoff curve numbers
rainfall data is required to further improve the model perfor- for steep hillslopes with natural vegetation in semi-arid tropical
mance and to further enhance the understanding of the runoff highlands, northern Ethiopia, Hydrol. Process., 22, 4097–4105,
processes in the basin. doi:10.1002/hyp.7011, 2008.

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

5166 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling

Dessie, M., Verhoest, N. E. C., Admasu, T., Pauwels, V. R. N., Poe- Mehta, V. K., Walter, M. T., Brooks, E. S., Steenhuis, T. S., Wal-
sen, J., Adgo, E., Deckers, J., and Nyssen, J.: Effects of the flood- ter, M. F., Johnson, M., Boll, J., and Thongs, D.: Application of
plain on river discharge into Lake Tana (Ethiopia), J. Hydrol., SMR to modeling watersheds in the Catskill Mountains, Environ.
519, 699–710, doi:10.1016/j.jhydrol.2014.08.007, 2014. Model. Assess., 9, 77–89, 2004.
Domenico, P. A. and Schwartz, F. W.: Physical and Chemical Hy- Miserez, A.: Soil erodibility and mapping in different hydrologi-
drogeology, John Wiley & Sons, New York, p. 824, 1990. cal land systems of Lake Tana basin, Ethiopia, MSc thesis study,
Fenicia, F., Savenije, H. H. G., Matgen, P., and Pfister, L.: Department of Soil and Water Management, KULeuven, Leuven,
Understanding catchment behavior through stepwise model Belgium, 2013.
concept improvement, Water Resour. Res., 44, W01402, Mishra, A., Hata, T., and Abdelhadi, A.W.: Models for recession
doi:10.1029/2006WR005563, 2008. flows in the upper Blue Nile River, Hydrol. Process., 18, 2773–
Food and Agriculture Organization of the United Nations, Rome 2786, 2004.
(FAO): Guidelines for soil description, Rome, Italy, 2006. Mishra, A. and Hata, T.: A grid-based runoff generation and flow
Frankenberger, J. R., Brooks, E. S., Walter, M. T., Walter, M. F., and routing model for the upper Blue Nile Basin, Hydrol. Sci. J., 51,
Steenhuis, T. S.: A GIS-Based Variable Source Area Hydrology 191–206, 2006.
Model, Hydrol. Process., 13, 804–822, 1999. Nash, J. E. and Sutcliffe, J. V.: River flow forecasting through con-
Gao, H., Hrachowitz, M., Fenicia, F., Gharari, S., and Savenije, H. ceptual models part I: a discussion of principles, J. Hydrol., 10,
H. G.: Testing the realism of a topography-driven model (FLEX- 282–290, 1970.
Topo) in the nested catchments of the Upper Heihe, China, Hy- Poppe, L., Frankl, A., Poesen, J., Teshager Admasu, Mekete
drol. Earth Syst. Sci., 18, 1895–1915, doi:10.5194/hess-18-1895- Dessie, Enyew Adgo, Deckers, J., and Nyssen, J.: Geomorphol-
2014, 2014. ogy of the Lake Tana basin, Ethiopia, J. Maps, 9, 431–437,
Gupta, H. V., Sorooshian, S., and Yapo, P.O.: Status of automatic doi:10.1080/17445647.2013.801000, 2013.
calibration for hydrologic models: Comparison with multilevel Savenije, H. H. G.: HESS Opinions “Topography driven conceptual
expert calibration, J. Hydrologic Eng. 4, 135–143, 1999. modelling (FLEX-Topo)”, Hydrol. Earth Syst. Sci., 14, 2681–
Hurni, H.: Third Progress Report, Soil Conservation Research 2692, doi:10.5194/hess-14-2681-2010, 2010.
Project, Vol. 4. University of Berne and the United Nations Uni- Scheerlinck, K., Pauwels, V. R. N., Vernieuwe, H., and Baets, B.
versity, Ministry of Agriculture, Addis Ababa, 1984. De.: Calibration of a water and energy balance model: Recursive
Istanbulluoglu, E. and Bras R. L.: Vegetation-modulated land- parameter estimation versus particle swarm optimization, Water
scape evolution: Effects of vegetation on landscape processes, Resour. Res., 45, W10422, doi:10.1029/2009WR008051, 2009.
drainage density, and topography, J. Geophys. Res., 110, F02012, Setegn, S. G., Ragahavan, S., and Bijan, D.: Hydrological modelling
doi:10.1029/2004JF000249, 2005. in the Lake Tana Basin, Ethiopia using SWAT model, The Open
Johnson, P. A. and Curtis, P. D.: Water balance of Blue Nile River Hydrol. J., 2, 49–62, 2008.
Basin in Ethiopia, J. Irrig. Drain. Eng. ASCE, 120, 573–590, Sivapalan, M.: The secret to “doing better hydrological sci-
1994. ence”: Change the question!, Hydrol. Process., 23, 1391–1396,
Jothityangkoon, C., Sivapalan, M., and Farmer, D. L.: Process con- doi:10.1002/hyp.7242, 2009.
trols of water balance variability in a large semi-arid catchment: Steenhuis, T. S. and van der Molen, W. H.: The Thornthwaite-
downward approach to hydrological model development, J. Hy- Mather procedure as a simple engineering method to predict
drol., 254, 174–198, 2001. recharge, J. Hydrol., 84, 221–229, 1986.
Kebede, S., Travi, Y., Alemayehu, T., and Marc, V.: Water balance Steenhuis, T. S., Collick, A. S., Easton, Z. M., Leggesse, E. S.,
of Lake Tana and its sensitivity to fluctuations in rainfall, Blue Bayabil, H. K., White, E .D., Awulachew, S. B., Adgo, E., and
Nile basin, Ethiopia, J. Hydrol., 316,233–247, 2006. Ahmed, A. A.: Predicting discharge and erosion for the Abay
Kebede, S.: Groundwater in Ethiopia: Features, Numbers (Blue Nile) with a simple model, Hydrol. Process. 23, 3728–
and Opportunities, ISBN 3642303919, Springer, p. 297, 3737, 2009.
doi:10.1007/978-3-642-30391-3, 2013. Tekleab, S., Uhlenbrook, S., Mohamed, Y., Savenije, H. H. G.,
Kennedy, J. and Eberhart, R.: Particle Swarm Optimization, Pro- Temesgen, M., and Wenninger, J.: Water balance modeling of
ceedings of IEEE International Conference on Neural Networks, Upper Blue Nile catchments using a top-down approach, Hy-
IV., 1942–1948, 1995. drol. Earth Syst. Sci., 15, 2179–2193, doi:10.5194/hess-15-2179-
Kim, U. and Kaluarachchi, J. J.: Application of parameter estima- 2011, 2011.
tion and regionalization methodologies to ungauged basins of the Temesgen, M., Savenije, H. H. G, Rockström, J., and Hoogmoed,
Upper Blue Nile River Basin, Ethiopia, J. Hydrol., 362, 39–52, W. B.: Assessment of strip tillage systems for maize production
2008. in semi-arid Ethiopia: Effects on grain yield, water balance and
Krasnostein, A. L. and Oldham, C. E.: Predicting wet- water productivity, Phys. Chem. Earth, Parts A/B/C, 47–48, 156–
land water storage, Water Resour. Res., 40, W10203, 165, 2012a.
doi:10.1029/2003WR002899, 2004. Temesgen, M., Uhlenbrook, S., Simane, B., van der Zaag, P., Mo-
Liu, B. M., Collick, A. S., Zeleke, G., Adgo, E., Easton, Z. M., hamed, Y., Wenninger, J., and Savenije, H. H. G.: Impacts of
and Steenhuis, T. S.: Rainfall-discharge relationships for a mon- conservation tillage on the hydrological and agronomic perfor-
soonal climate in the Ethiopian highlands, Hydrol. Process., 22, mance of Fanya juus in the upper Blue Nile (Abbay) river basin,
1059–1067, 2008. Hydrol. Earth Syst. Sci., 16, 4725–4735, doi:10.5194/hess-16-
McWorter, D. B. and Sunada, D. K.: Groundwater Hydrology and 4725-2012, 2012b.
hydraulics, Water Resources Publications, Ft. Collins. Co., 1997.

Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014 www.hydrol-earth-syst-sci.net/18/5149/2014/

 M. Dessie et al.: Analyzing runoff processes through conceptual hydrological modeling 5167

Tilahun, S. A., Guzman, C. D., Zegeye, A. D., Engda, T. A., Wale, A., Rientjes, T. H. M., Gieske, A. S. M., and Getachew, H. A.:
Collick, A. S., Rimmer, A., and Steenhuis, T. S.: An effi- Ungauged catchment contributions to Lake Tana’s water balance,
cient semi-distributed hillslope erosion model for the subhumid Hydrol. Process., 23, 3682–3693, 2009.
Ethiopian Highlands, Hydrol. Earth Syst. Sci., 17, 1051–1063, Wittenberg, H.: Base flow recession and recharge as nonlinear stor-
doi:10.5194/hess-17-1051-2013, 2013. age processes, Hydrol. Process. 13, 715–726, 1999.
Uhlenbrook, S., Mohamed, Y., and Gragne, A. S.: Analyzing catch- Zenebe, A., Vanmaercke, M., Poesen, J., Verstraeten, G., Ni-
ment behavior through catchment modeling in the Gilgel Abay, gussie Haregeweyn, Mitiku Haile, Kassa Amare, Deckers, J., and
Upper Blue Nile River Basin, Ethiopia, Hydrol. Earth Syst. Sci., Nyssen, J.: Spatial and temporal variability of river flows in the
14, 2153–2165, doi:10.5194/hess-14-2153-2010, 2010. degraded semi-arid tropical mountains of northern Ethiopia, Z.
Vogel, R. M. and Kroll, C. N.: Regional geohydrologic-geomorphic Geomorphol., 57, 143–169, 2013.
relationships for the estimation of low-flow statistics, Water Re-
sour. Res., 28, 2451–2458, 1992.

www.hydrol-earth-syst-sci.net/18/5149/2014/ Hydrol. Earth Syst. Sci., 18, 5149–5167, 2014

View publication stats