September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.

5 51

Modeling streamflow and sediment using SWAT in

Ethiopian Highlands

Hailu Kendie Addis1,2*, Stefan Strohmeier3, Feras Ziadat4,

Nigus Demelash Melaku1,2, Andreas Klik1
(1. Institute of Hydraulics and Rural Water Management, University of Natural Resources and Life Sciences, Vienna 1190, Austria;
2. Gondar Agricultural Research Center, Gondar, Ethiopia; 3. International Center for Agricultural Research in the Dry Areas,
Amman 11110-17198 , Jordan; 4. Food and Agricultural Organization of the United Nations, Rome 00100, Italy)

Abstract: The coincidence of intensive rainfall events at the beginning of the rainy season and unprotected soil conditions after
extreme dry spells expose the Ethiopian Highlands to severe soil erosion. Soil and water conservation measures (SWC) have
been applied to counteract land degradation in the endangered areas, but SWC efficiency may vary related to the heterogeneity
of the landscape. The Soil and Water Assessment Tool (SWAT) model was used to model hydrology and sediment dynamics
of a 53.7 km2 watershed, located in the Lake Tana basin, Ethiopia. Spatially distributed stone bund impacts were applied in
the model through modification of the surface runoff ratio and adjustment of a support practice factor simulating the trapped
amounts of water and sediment at the SWC structure and watershed level. The resulting Nash-Sutcliffe efficiency (NSE) for
daily streamflow simulation was 0.56 for the calibration and 0.48 for the validation period, suggesting satisfactory model
performance. In contrast, the daily sediment simulation resulted in unsatisfactory model performance, with the NSE value of
0.07 for the calibration and –1.76 for the validation period and this could be as a result of high intensity and short duration
rainfall events in the watershed. Meanwhile, insufficient sediment yield prediction may result to some extent from daily based
data processing, whereas the driving runoff events and thus sediment loads occur on sub-daily time scales, probably linked with
abrupt gully breaks and development. The calibrated model indicated 21.08 Mg/hm2 average annual sediment yield, which is
far beyond potential soil regeneration rate. Despite the given limits of model calibration, SWAT may support the scaling up
and out of experimentally proven SWC interventions to encourage sustainable agriculture in the Ethiopian Highlands.
Keywords: SWAT, streamflow, sediment dynamics, soil erosion, soil and water conservation, watershed hydrology
DOI: 10.3965/j.ijabe.20160905.2483

Citation: Addis H K, Strohmeier S, Ziadat F, Melaku N D, Klik A. Modeling streamflow and sediment using SWAT in the
Ethiopian Highlands. Int J Agric & Biol Eng, 2016; 9(5): 51－66.

with the cultivation of the land and thus, inevitably, with

1 Introduction
land degradation[1]. Human interventions, such as
The rise of the human civilizations is directly linked deforestation for agricultural food production, the
cultivation of marginal lands, overgrazing and the
Received date: 2016-03-19 Accepted date: 2016-09-04
Biographies: Stefan Strohmeier, PhD, research interests: soil and
exploitation of soil fertility accelerate soil erosion[2] and
water conservation, Email: s.strohmeier@cgiar.org; Feras Ziadat, subsequent soil depletion is accompanied with reduced
PhD, research interests: soil science, Email: feras.ziadat@fao.org; crop productivity[3]. Ongoing land degradation
Nigus Demelash Melaku, PhD candidate, research interests: soil
endangers the agricultural productivity in many areas
fertility, Email: d.nigus19@gmail.com; Andreas Klik, PhD,
Professor, research interests: soil and water conservation, Email:
around the globe[4], and undoubtedly, the Ethiopian
andreas.klik@boku.ac.at. Highlands are among the most affected. Various
*Corresponding author: Hailu Kendie Addis, PhD candidate, impacts and consequences of the severe land degradation
research interests: soil and water conservation. Mailing address:
in the Ethiopian Highlands have been reported by Hurni
Gondar Agricultural Rresearch Center; P.O. Box 1337, Gondar,
Ethiopia; Tel: +251-918711082, Email: hailukendie@gmail.com. et al.[5]. The extensive famines in 1973 and 1984, as an
 52 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

alarming consequence of droughts and low crop concerning the usability of the semi-distributed
productivity, initiated governmental rethinking eco-hydrological model for remote landscapes. In fact,
[6]
concerning rural land management . The Ethiopian large areas of the Ethiopian Highlands are still under
government responded with large scale rehabilitation investigated and therefore proper model input and
measures and the establishment of various soil and water particularly calibration data (such as streamflow and
conservation (SWC) interventions across the country to sediment yield) are scarce, which might impede proper
[6,7]
counteract the ongoing soil depletion . model calibration and validation in many cases. Various
[13,22]
From the beginning of agricultural activities different studies have shown that advanced erosion models
[8]
SWC techniques have been developed mainly to retain suffer from the lack of available input data especially for
soil fertility and thus crop productivity. Various SWC large scale application. Conclusively, there remains
techniques and their variable impacts have been extensive need to evaluate semi-distributed
intensively discussed in the literatures [7] and [9]. In eco-hydrological watershed modeling in the Ethiopian
particular for the Ethiopian Highlands SWC management Highlands.
through stone bunds was found as sound practice for soil The study reported here was performed in the context
[10]
erosion control . Stone bunds are elevated structures of a multidisciplinary international research project that is
[7]
intersecting a hillslope in specific intervals , resulting in being conducted within the Gumara-Maksegnit watershed
decreased surface runoff and sediment yield through which is located in the Lake Tana basin in the Amhara
slope length reduction and the creation of a small region of Ethiopia. Integrated watershed research is
[11]
retention area . However, SWC interventions are being conducted, including several soil, crop, hydrology
often uniformly applied across landscapes but may only and agro-environmental related analyses, to gain a deeper
be reasonable for certain field conditions. In fact, field insight into watershed scale hydrology and land
conditions are often highly variable in the Ethiopian degradation issues, evaluate various soil and water
[12]
Highlands . Therefore, site specific assessment of the conservation interventions and to aim for an improved
most influential watershed processes may be crucial for livelihood of stakeholders living in the watershed. The
the development of efficient conservation measures. spatial assessment of surface runoff and sediment yield
At present, many models with a broad spectrum of within Gumara-Maksegnit study site using SWAT is a
concepts, which were classified as spatially lumped, key component of the overall research project. The
spatially distributed, empirical, regression, model case study was conducted: (1) to assess the
semi-distributed eco-hydrological model and factorial applicability of SWAT for simulating the key watershed
scoring models, are in use for modelling the processes of a remote and mountainous agricultural
rainfall-runoff-soil erosion and sediment transport watershed, and (2) to evaluate the impact of spatially
processes at different scales[13]. The Soil and Water distributed soil and water conservation (SWC) structures
Assessment Tool (SWAT) is a semi-distributed on surface runoff and soil erosion. Eventually, the study
eco-hydrological model. SWAT is one of the most aims for the establishment of a well-calibrated
widely used watershed models, which was developed by semi-distributed eco-hydrological model as a tool for
the United States Department of Agriculture-Agricultural evaluating multiple land management practices suitable
Research Service (USDA-ARS)[14] and can be used to for reduction of sediment transport, which can be scaled
predict agricultural land management impacts on the up to assess proper SWC strategies and to counteract
hydrological regime of a watershed through simulation of ongoing land degradation at a broader scale.
variable soil, land use and management conditions over
long periods[14,15]. In Ethiopia, SWAT has been used in 2 Materials and methods
a number of studies to predict streamflow and sediment 2.1 Description of the study watershed
[16-21]
yield with different outcomes and recommendations The Gumara-Maksegnit watershed, is located in the
 September, 2016 Addis H K, et al. Modeling streamflow and sediment using SWAT in the Ethiopian Highlands Vol. 9 No.5 53

Amhara region in northwest Ethiopia between 37°33′00″-

37°37′00″E and 12°24′00″-12°31′00″N (Figure 1).
The confined watershed area is 53.7 km2 based on an
ArcGIS watershed delineation using a 90 m grid
Digital Elevation Model (DEM) produced by SRTM
(Shuttle Radar Topography Mission)[23]. The
watershed elevation ranges from 1920 m (outlet) to
2850 m above sea level in the north, while the hillslopes
range from nearly flat (<2%) to extremely steep (>70%)
(Figure 2a). The northern part of the watershed,
Denkez Mountain Ridge, borders to Tekezi Basin, while
the Gumara-Maksegnit watershed is part of the Blue
Nile River Basin. The watershed geology is dominated
by a Trap Series of Tertiary volcanic eruptions[24],
which are commonly described by their degree of
oxidation as exemplified by the frequent dominance of
Figure 1 Overview of the project watershed area in the northwest
ferric over ferrous iron and by the abundant water
Amhara region, Ethiopia
content[24]. The main soils are Cambisol and Leptosol
in the upper and central part of the watershed and
Vertisol in the lower part near the outlet. The
Gumara-Maksegnit River is the main stream of the study
watershed, which part of the Lake Tana drainage basin.
Lake Tana is the origin of the Blue Nile River and the
largest lake in Ethiopia. The Gumara-Maksegnit River
discharges continuously throughout the year and is
characterized by several flood events during the rainy
season versus drastically decreased flow during the dry
season. The climate of the Gumara-Maksegnit
watershed is characterized by the ‘Woina Dega’ zone
(cool semi-humid) between 1920 m to 2400 m above sea
level, and the ‘Dega’ zone (cool) above 2400 m. The
majority of the watershed area is located within the cool
Figure 2 Gumara-Maksegnit watershed maps showing (a) slope
semi-humid zone at an elevation of 1920 m to 2400 m
classes, and (b) elevation data and location of weather stations and
above sea level. The climate is dominated by distinct sub-basins included in stone bund experiment assessment discussed
wet and dry periods. The wet season typically occurs in Section 2.3.5
from June to September and the dry season occurs from 2.2 SWAT model
November to April, while May and October are The SWAT model is a semi-distributed eco-hydrological
transition months. The mean annual rainfall in the continuous event watershed-scale model usable to
watershed is 1200 mm of which more than 90% occurs evaluate the impact of different land management
during the rainy season (June to September). The practices on surface and subsurface water movement,
average monthly maximum and minimum temperatures sediment, and agricultural chemical yields in complex
recorded from 1997 to 2013 were 31.8°C for March and watersheds with different soil, land-use and management
10.8°C for January. conditions[25]. ArcSWAT, as an ArcGIS interface[26],
 54 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

uses GIS spatial algorithms to spatially link multiple watershed study.

model input data, such as watershed topography (DEM), 2.3.1 DEM (Digital Elevation Model)
soil, land use, land management and climatic data. For this study, the 90 m grid cell DEM, produced by
During watershed delineation, the entire watershed is SRTM (Shuttle Radar Topography Mission)[23] was used
divided into different sub-basins. Then, each sub-basin to obtain the topographic characteristics of
is discretized into a series of Hydrologic Response Units Gumara-Maksegnit watershed. Then, the watershed had
(HRUs) as the smallest computation unit of a SWAT been divided into three slope steepness classes, namely:
model, which are characterized by homogeneous soil, 0°-11° (18.77 km2), 11°-28° (17.66 km2) and greater than
land use and slope combinations. Daily climate input 28° (17.26 km2) (Figure 2a).
data for defined locations (mostly related to ground 2.3.2 Climate
weather stations) are spatially related to the different Climate input data required by SWAT includes daily
sub-basins of the model using a ‘nearest neighbor’ GIS precipitation, maximum and minimum temperature,
algorithm. Different model outputs, such as surface relative humidity, half hour rainfall, wind speed and solar
runoff, sediment yield, soil moisture, nutrient dynamics, radiation. Required daily precipitation and maximum/
crop growth etc., are simulated for each HRU, aggregated minimum air temperature data was collected at four
and processed to sub-basin level results on a daily time different weather stations located within (three stations)
step resolution. and slightly outside (one station) the watershed (Figure
SWAT provides different runoff routing techniques 2b). Daily solar radiation, relative humidity, and wind
for both surface runoff and streamflow. In this study, speed data were recorded at a different metrological
surface runoff was computed using the USDA (United station slightly outside the study watershed (Figure 2b).
States Department of Agriculture) NRCS (Natural The SWAT weather generator[30] was used for simulating
Resources Conservation Service) approach[27], while missing daily weather data. The daily climatic data
channel routing was processed by Muskingum routing (from January 1, 1997 to December 31, 2013) recorded at
[28]
method . The NRCS method was chosen to enable the weather station, which was located slightly outside
user friendly and comprehensive consideration of soil and the watershed (Figure 2b) was used to create the monthly
water conservation (SWC) impacts. A number of weather statistics using the weather generator.
methods with varying data requirements for 2.3.3 Land use
evapotranspiration (ET) estimation are incorporated in Land cover map for this research was produced on the
[29]
SWAT: for this study, the Hargreaves formula was pixel based supervised classification of 10 m spot satellite
used. In SWAT, up-land soil erosion is computed based image (Figure 3a). The study watershed has three major
on the Modified Universal Soil Loss Equation land-use classes (Figure 3b) and is mainly covered by
[29]
(MUSLE) , which allows the consideration of a support agricultural land (63.5%) followed by mixed forest
practice factor representing supposed SWC effects on (24.3%), and grazing land (12.2%). The agricultural
sediment loss. land was further subdivided into six major agricultural
2.3 Input data crops: tef (Eragrostis Tef) (30.0%), sorghum (13.2%),
SWAT input data in developing countries (such as barley (6.9%), fava bean (5.6%), winter wheat (4.3%) and
Ethiopia) are usually not readily available and are often chickpea (3.5%). Tef is a minor cereal crop on a global
difficult to collect, and data availability is even more scale, but a major food grain and lovegrass (lovegrass is
limited for good quality calibration and validation data. commonly used as livestock fodder) in Ethiopia and
Amongst the acquisition of various remote sensing Eritrea and this annual crop can be grown under a wide
sources for DEM and land use input preparation, range of conditions[31]. Tef and sorghum are the main
comprehensible field sampling and hydrological staple crops, whereas chickpea is grown in the lower
monitoring were a central task of the Gumara-Maksegnit regions and cannot be grown in the higher altitude.
 September, 2016 Addis H K, et al. Modeling streamflow and sediment using SWAT in the Ethiopian Highlands Vol. 9 No.5 55

2.3.4 Soil 100 cm) were taken for physical and chemical analysis.
SWAT requires multiple soil physical and chemical Undisturbed soil core cylinder samples were taken from
attributes for various soil depths such as soil texture, bulk the topsoil layer to determine bulk density following
density, stone content, organic carbon, hydraulic previously developed procedures[34]. Soil texture was
[32]
conductivity, soil erodibility, etc. At least one measured based on an earlier published method[35], and
software package is available which can be used to organic carbon was determined by a wet oxidation
calculate the spatial distribution of various soil properties method[36]. Available water content and hydraulic
for environmental modeling using selected input conductivity for each layer as well as bulk density for the
[33]
parameters . Nevertheless, good quality field second and third layer were assessed using a
[37]
sampling data may be used preferentially. In this study, pedo-transfer function developed by Saxton and Rawls .
an intensive field sampling campaign was carried out to Nevertheless, the most important soil data impacts were
determine various soil properties in a 500 m by 500 m manually determined based on the previously described
grid over the entire watershed. A total of 234 soil intensive field sampling results. The soil map that
samples were collected using a bucket auger. At each describes the distribution of different soil textural classes
location approximately 2 kg bulk soil samples from of the study watershed is presented in Figure 3c.
different soil layers (0-25 cm), (25-60 cm) and (60-

a. Spot satellite image b. Three major land cover categories c. Soil textural class maps

Figure 3 The Gumara-Maksegnit watershed

2.3.5 Soil and Water Conservation (SWC) interventions runoff and soil erosion at the field level, but based on their
Different SWC practices have been applied in the local or minor areal extent, these structures have limited
Gumara-Maksegnit watershed such as stone bunds, micro effect on watershed level hydrology or sediment dynamics.
water harvesting ponds, trenches and semi-circular stone Thus, stone bunds were the only SWC interventions
bunds (Figure 4). However, linear (slightly graded) stone considered during watershed modeling and approximately
bunds are the predominant practice, which affect large 50% of the study watershed is presently treated with the
agricultural areas in the central and the lower part of the stone bunds. As described by Bosshart[11], SWC impacts
watershed. Locally installed harvesting ponds (four of stone bunds are mainly related to the reduction of
structures applied within the watershed), trenches and surface runoff and sediment yield by intersecting
semi-circular stone bunds may have a positive effect on hillslope lengths in specific intervals and the ponding
 56 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

effects that occur at each structure. In the course of the landuse type interactions, however, a coarse DEM mesh
Gumara-Maksegnit watershed study, different plot level as used as an input for this study was one of the limitation.
[38]
well as sub-basin level experiments were carried out to The study watershed is composed of a rugged topography
investigate the effects of stone bunds on surface runoff and with different management practices; thus, the 234 soil
soil loss, and moreover, to enable the implementation of sampling points are considered totally different and the
SWC impacted in SWAT modeling. SWAT provides study did not set a threshold that eliminates minor soil
[32]
various options to consider SWC impacts including: (1) types. Therefore, every HRU for the study watershed
surface runoff may be modified through the adjustment of corresponds with an average area of 1.9 hm2. Similarly,
the runoff ratio (Curve Number) and/or the consideration Zabaleta et al.[40] used 165 HRUs for a 4.8 km2 watershed
of a micro-pond (pothole) at the related HRU level, which in Spain, which averaged about 2.9 hm2 per HRU.
will also impact soil erosion, and (2) impacts on sediment The impact of stone bund SWC structures was
yield levels via adjustment of the support practice factor simulated through reduction of the Curve Number (CN_2)
(P-factor) and/or the slope length factor (LS) of the for surface runoff ratio modification as well as the
[39]
MUSLE . The ideal factors that describe the effect of adjustment of the support practice factor (P-factor) to
stone bunds are the USLE support practice factor account for the amount of trapped sediments at the stone
(P-factor), the Curve Number and average slope length bunds. The effect of stone bunds on runoff and soil
(SLSUBBSN). In this study, the SLSSUBSN value was erosion was initially assessed during the erosion plot
modified by editing the HRU (.hru) input table, whereas experimental campaigns in 2012 and 2013, based on the
the P-factor and Curve Number values were modified by comparison of treated and untreated sub-basins located in
editing Management (.mgt) input table. the watershed (this activity is still ongoing). Based on
[41]
the plot experiments carried out in 2013 , stone bund
structures were found to reduce surface runoff by
approximately 60% to 80% and sediment yield between
40% to 80%. This is consistent with other plot
experimental findings reported by Adimassu et al.[42],
where stone bunds reduced sediment yield by roughly
50% compared to untreated plots. However, plot
a experiments tend to reflect optimized stone bund
conditions for just a very limited area. In fact, the stone
bund plot experiments carried out in Gumara-Maksegnit
do not account for cumulative hillslope lengths or the
overall length of the stone bund walls and thus how much
total area those affect, which may lead to considerably
lower SWC impacts at a farm or sub-basin level. For
the sub-basin level experiment (Figure 2b), where the
b area of each sub-basin is approximately 30 hm2, the
Figure 4 Stone bund treated fields (a) and the small channel difference of measured surface runoff between treated
above the stone bund (b) and untreated sub-basins was around 30%. However,
2
The 53.7 km Gumara-Maksegnit watershed was the measured sediment yield declined by only
discretized into 15 sub-basins and 2799 HRUs for the approximately 10% during the 2012 rainy season, which
SWAT simulations. The highest numbers of HRUs for is not consistent with the results reported by
the study watershed occurred as a result of the 234 user [43]
Gebremichael et al. These results include a large
defined soil names, the 3 slope classes as well as the 9 range of uncertainty particularly for sediment yield, but
 September, 2016 Addis H K, et al. Modeling streamflow and sediment using SWAT in the Ethiopian Highlands Vol. 9 No.5 57

also due to only a few synchronically recorded rainfall However, automated calibration is more efficient for
events in the treated and untreated sub-basins (Figure 2b). some applications[46], especially for complex hydrologic
Moreover, the comparability of different sub-basins is models. Different datasets may be required to evaluate
limited as a result of the inherent landscape and rainfall model performance for different environmental
[45]
related variability, even though the sub-basins border conditions . However, the number of attributes and
each other and the soil, slope, and land use conditions are the observation period required for proper consideration
generally homogenous. However, the current SWC of the driving watershed processes may vary from site to
impact research is ultimately designed to provide site. Long term and good quality data is especially rare
comprehensive SWC assessment and conclusive for the Ethiopian Highlands. In the present study, the
modeling parameters. Hence, as an early stage entire simulation period is limited to field observation
assessment, the CN_2 was reduced for agricultural HRUs data from 2011 to 2012 (calibration) and 2013
in the treated areas with the target to achieve overall (validation). The calibration/validation model run was
surface runoff reduction of about 30% on treated HRU’s performed with a warm-up period of seven years to
compared to untreated conditions. The P-factor was set minimize the effect of non-equilibrium initial conditions
equal to 0.85, because: (1) the CN_2 reductions already such as soil moisture or residue cover[47]. In this
leads to reduced soil erosion on the treated areas, and (2) research, daily streamflow and sediment yield recorded at
as a compromise between plot and sub-basin level the outlet of the watershed were used for both calibration
sediment yield ratio outcomes. A small range of and validation of the model.
variability was assigned to the defined CN_2 and P-factor Streamflow was obtained by converting
parameter sets during the calibration procedure, which quasi-continuous water level (m) records (using pressure
allowed additional minor adjustments during the transducer) into flow (m/s) based on an experimentally
automated model optimization. These assumptions developed water level and discharge rating curve[48]
result in the stone bunds essentially replicating the effects (Figure 5). The respective rating curve was established
of terraces[16], in terms of how the average slope length based on water level and manual flow velocity
(SLSUBBSN) is modified to represent terrace effects in measurements using a one-dimensional flow velocity
cropped landscapes. device analyzing several runoff events. The outlet of the
2.4 Calibration and validation data watershed was constructed as a fixed cross section, which
Different calibration approaches can be used in was built from stones, concrete and gabions to ensure an
SWAT with respect to frequency and quantity of explicit and constant relationship between water level and
observation data available for model calibration. discharge. Hysteresis effects related to the different
Nevertheless, the most powerful calibration is usually stages of a peak wave (arriving and leaving) were found
achieved through following a specific calibration order as to have negligible impact on the calculation of the daily
suggested by Arnold et al.[44] In particular, streamflow discharge, considering various sources of uncertainty
data at the sub-basin or watershed level are required to (such as measurement errors and gaps). Moreover, a
perform accurate model hydrologic balance and turbidity sensor was installed at the side wall of the fixed
streamflow calibration, followed by calibration of cross section to gain insight into sediment dynamics of
different pollutants such as sediment load, nutrient yields the main stream. The turbidity meter was calibrated in
and other water quality variables. The calibration the laboratory using on-site sediments to assess the
procedure is typically based on initial sensitivity analysis fraction of suspended soil (g/L) in water related to
results (using a set of sensitive parameters) and is indirect light signal measurement. However,
[44,45]
executed either manually or automatically . considerable data uncertainty has to be taken into account
Calibrations can be performed manually, which can be and the derived sediment concentrations may be used to
[44]
important for clearly understanding some processes . describe general sediment dynamics solely. According
 58 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

to this, quasi-continuous turbidity readings were reveal the challenging monitoring conditions that exist at
controlled and adjusted based on manual bottle sampling the site.
throughout the runoff monitoring period. Streamflow
and sediment yield, which were derived through
multiplying sediment concentration with the according
flow volume, were compiled on a daily basis usable for
SWAT calibration. Figure 6 shows the derived
hydrograph for the main outlet during approximately the
four month rainy season in 2013. However, several
unmeasured sediment concentration and streamflow data, Figure 5 Established rating curve at the outlet of
mainly due to sensor failures or power supply errors, Gumara-Maksegnit Watershed[48]

Figure 6 Hydrograph at the main outlet and precipitation data of the four rain gauge stations in Gumara-Maksegnit watershed

2.5 Model efficiency assessment (R2). However, during the SWAT-CUP calibration
Efficiency criteria are defined as a mathematical multiple simulations are executed accounting for the
measure of how well a model simulation matches user-adjusted set of parameters and related parameter
[45]
corresponding observed data . SWAT calibration ranges. This procedure can result in a very large set of
procedures, including the SWAT-CUP calibration tool, simulations, depending on the number of parameters
provide multiple model efficiency criteria to be used as selected for calibration, the user-adjusted range for
an objective function for model calibration and parameter variation and the selected calibration
[49]
validation . The ‘Sequential Uncertainty Fitting 2’ methodology (including the number of iterations,
(SUFI-2) procedure, available within SWAT-CUP parameter range discretization etc.).
software, was used to perform model sensitivity analysis, 2.5.1 Root mean square error (RMSE)
[49]
calibration and validation procedures through iterative The root mean square error (RMSE) has been used as
variation of user defined parameter sets. The SUFI-2 a standard statistical metric to measure model prediction
algorithm accounts for various sources of uncertainty error in meteorology, air quality, and climate research
such as input data uncertainty, conceptual model studies; a smaller RMSE value indicates better model
[50]
uncertainty and parameter uncertainty . In the present performance[51]. Although RMSE is sensitive to outliers
study, the goodness of the model fit related to streamflow as it places a lot of weight on large errors, it has been
and sediment yield was assessed based on root mean developed to confirm the reliability of models[52]. The
squared error (RMSE), Nash–Sutcliffe efficiency (NSE), RMSE does not provide information about the relative
percent bias (PBIAS) and coefficient of determination size of the average difference and the nature of
 September, 2016 Addis H K, et al. Modeling streamflow and sediment using SWAT in the Ethiopian Highlands Vol. 9 No.5 59

differences comprising them[53]. The RMSE is predicted value[55]. A value of 1 means the predicted
calculated with the following equation: value is equal to the observed value, where a value of

⎡1 n ⎤
1/ 2 zero means there is no correlation between the predicted
RMSE = ⎢ ∑ i =1 ( Ei − Oi ) 2 ⎥ (1)
⎣ n ⎦ and observed values.

2.5.2 Nash-Sutcliffe Efficiency (NSE) 3 Results and discussion

The Nash-Sutcliffe efficiency is a normalized statistic
In the Ethiopian Highlands, erratic and intensive
that determines the relative magnitude of the residual
rainfalls during the rainy season generate several peak
variance (“noise”) compared with the measured data
runoff events (Figure 6), exposing steep sloped areas to
variance (“information”)[54]. The Nash-Sutcliffe
potentially severe soil erosion. In SWAT, rainfall
efficiency is calculated as:
erosive impacts are estimated mainly as a function of the
∑
n
( Ei − Oi )2 intensity and duration of rainfall events. The
NSE = 1 − i =1
(2)
∑
n
i =1
(Oi − O ) 2
hydrograph at the outlet of the study watershed is
The range of E lies between −∞ and 1.0 with E=1 dominated by the short period peak flows, occurring
describing a perfect fit. Values between 0-1.0 are several times weekly whereas mean base flow was
generally viewed as acceptable levels of performance, observed between 1-2 m³/s during rainy season of the
whereas values <0.0 indicate that the mean observed calibration periods. Intense rainfall events correspond to
value is a better predictor than the model[55]. peak flows on daily time scale which states that rainwater
2.5.3 Percent bias (PBIAS) is routed through the watershed in sub-daily time
Percent bias (PBIAS) is defined as the average intervals. This refers to the steep sloped and the rugged
tendency of the observed data compared with their mountainous watershed as well as the convective rainfall
simulated counterparts[56]. The negative values of characteristics in the Ethiopian Highlands. At the outlet,
PBIAS indicate model overestimation bias, and positive peak discharges of about 30 m³/s have been observed
values indicate model underestimation bias. The during the 2012 rainy season whereas extreme floods are

optimal value of PBIAS is 0.0, with low-magnitude expected to exceed this amount several times. In

values indicating accurate model simulation[45]. PBIAS contrast, the SWAT model derives maximum mean daily

is calculated with the following equation: discharges of less than 10 m³/s for the whole calibration
period of the 2011 rainy season. This may be due to the
⎡ ∑ n (Oi − Ei ) × 100 ⎤
PBISA = ⎢ i =1 n ⎥ (3) daily based runoff computation which can’t adequately
⎢
⎣ ∑ i =1
(Oi ) ⎥
⎦ account for intense storms of short duration. Rainfall
2
2.5.4 Coefficient of determination (R ) records for the Aba-Kaloye weather station (2011-2013);
The coefficient of determination R2 is defined as the located in the lower central part of the watershed,
[57]
squared value of the coefficient of correlation . It is suggests that more than 50% of the annual maximum
calculated as follows: daily rainfall occurs within 30 min time periods during
2 intense storms (Table 1).
⎡ ⎤
∑ i =1 (Oi − O)( Ei − E )
n

R2 = ⎢ ⎥ (4) Table 1 Annual maximum series rainfall in units of

⎢ 2 ⎥
∑ i =1 (Oi − O)2 ∑ i =1 (Oi − E ) ⎥⎦
n n
⎢⎣ millimeters for Aba-Kaloye weather station
Year 15 m* 30 m 1h 3h 6h 12 h 24 h 48 h 72 h**
where, n is the number of observations or samples; Oi are
2011 20.2 38.6 42.6 47.4 54.6 68.2 74.6 94.4 119.2
observed values; Ei are estimated values; Ō is mean of
2012 16.8 29.6 37.2 40.4 42.8 54.6 58.8 69.6 83.6
observed values; Ē is the mean of estimated values; i is 2013 15.6 27.8 31.4 36.6 40.2 49.6 52.4 64 79.2

counter for individual observed and predicted values. Note: Durations in the table range from 15-min (15 m*) to 72 h (72 h**).

The range of R2 lies between 0 and 1, and describes Considering the relatively small watershed area,
how much of the observed value is explained by the Gumara-Maksegnit flood events are characterized by
 60 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

relatively short time periods (sub-daily) and distinct peak showed that 48% of the precipitation becoming ET[58],
flows. Based on a simulation of the whole period of while, Gebremicael et al.[59] reported that 53% of the
available climate input data (1997-2013), the calibrated precipitation becoming ET.
model estimates 352 mm of average annual surface runoff, 3.1 Model sensitivity analysis
whereas recharge to the deep aquifer is approximately Sensitivity analysis supports the determination of the
19 mm, and entirely, more than 31% (373 mm) of driving watershed processes and thus the identification of
rainwater balance is used for evapotranspiration. This the most sensitive parameters through the assessment of
low amount of ET in the study watershed was found to the rate of change of model outputs with respect to defined
be attributable to land use/land cover change, mainly changes of model inputs[44]. Fourteen hydrological
from expanding agricultural activities, as it was (Table 2) and eight sediment-related (Table 3) parameters
[60]
described by Alemu et al. Generally, from field were selected for the subsequent SWAT calibration on
observation more water is drained out of the watershed the bases of the sensitivity analysis. In this study, the
as a result of the minimum soil conservation coverage, CN_2 and channel cover factor were found to be the most
land use change and the steep slope nature of the study sensitive parameters with respect streamflow and
[58]
watershed. In contrast, similar study by Yesuf et al. sediment yield, respectively.

Table 2 List of model parameters sensitive to streamflow and fitted values in order of ranking
Adjusted or fitted
Parameter name Description Ranking
parameter value

r__CN2.mgt* Curve number -0.13 1

r__RCHRG_DP.gw** Deep aquifer percolation fraction 0.3 2
r__GWQMN.gw Threshold depth of water in the shallow aquifer required for return flow to occur (mm H2O) -0.13 3
v__ALPHA_BF.gw Base flow alpha factor (days) 0.019 4
r__GW_REVAP.gw Groundwater “revap” coefficient 0.4 5
v__GW_DELAY.gw Groundwater delay time (days) 110 6
v__CH_K2.rte Effective hydraulic conductivity in main channel alluvium (mm/hr) 82.49 7
v__CH_N2.rte Manning’s “n” value for the main channel -0.00783 8
v__ESCO.hru Plant uptake compensation factor 0.63 9
r__SOL_K(1).sol Saturated hydraulic conductivity -0.52 10
r__REVAPMN.gw Threshold depth of water in the shallow aquifer percolation to the deep aquifer to occur (mm H2O) -0.2 11
r__SLSUBBSN.hru Average slope length (m) 0.01 12
v__SURLAG.bsn Surface runoff lag coefficient 0.3 13
r__SOL_AWC(1).sol Soil available water storage capacity 0.28 14
Note: * The qualifier (r__) refers to relative change in the parameter where the value from the SWAT database is multiplied by 1 plus the fitted value), while (v__) means
**
the existing parameter value from the SWAT database is to be replaced by the fitted value. The extension (e.g., .gw) refers to the SWAT input file where the
respective parameter is located.

Table 3 Model parameters sensitive to sediment yield and fitted values in order of ranking
Parameter name Description Fitted parameter value Ranking

v__ CH_COV2.rte* Channel cover factor 0.8 1

v__ CH_COV1.rte Channel erodibility factor 0.15 2
Linear parameter for calculating the maximum amount of sediment that can be re-entrained during
v__ SPCON.bsn** 0.009 3
channel sediment routing
v__ PRF.bsn Peak rate adjustment factor for sediment routing in the main channel 1.4 4
v__ HRU_SLP.hru Average slope steepness (m/m) 0.18 5
v__ SPEXP.bsn Exponent parameter for calculating sediment re-entrained in channel sediment routing 1.35 6
r__ USLE_P.mgt USLE equation support practice factor -0.01 7
v__ RSDIN.hru Initial residue cover (kg/ha) 3400 8
Note: * The qualifier (v__) means the existing parameter value from the SWAT database is to be replaced by the fitted value, while (r__) refers to relative change in the
**
parameter where the value from the SWAT database is multiplied by 1 plus the fitted value). The extension (e.g., .bsn) refers to the SWAT file type where the
parameter occurs.
 September, 2016 Addis H K, et al. Modeling streamflow and sediment using SWAT in the Ethiopian Highlands Vol. 9 No.5 61

3.2 Model calibration and validation predicted for the validation period (Figure 7). Some of
The automated calibration (SWAT-CUP) for the previously published SWAT studies for smaller
streamflow (Figure 7, top) leads to adequate daily watersheds in the northeast and northwest of Ethiopia
calibration results, and validation (Figure 7, bottom) tend to show weaker hydrologic results[18,21], which is an
indicates satisfactory model fit according to the indication that it may be difficult to accurately represent
[45,61]
assessment criteria suggested by Moriasi et al. For processes and thus obtain better results for smaller
2
the calibration period NSE=0.56, PBIAS=6%, R =67 and watersheds. Nevertheless, obvious correspondence of
RMSE=0.62, while for the validation period NSE=0.48, the hydrographs of observed and simulated streamflow
2
PBIAS=18%, R =53 and RMSE=3.4. Meanwhile, the (Figure 7) for both, the calibration and validation period,
measured peak flows on the same day often indicates that SWAT is capable to simulate the
over-predicted for the calibration period and under- hydrological regime of Gumara-Maksegnit watershed.

Figure 7 Observed and simulated daily streamflow hydrograph at the outlet of Gumara-Maksegnit watershed, calibration (top) and
validation (bottom)

In contrast, the sediment simulation results were streamflow about 39% of the data and during the
unsatisfactory, especially during the validation period, validation period about 31% of the data were bracketed
which is shown by the low or even negative NSE values by 95PPU, while during daily sediment yield simulation
(i.e. 0.07 for the calibration period and –1.76 for the around 18% of the data were bracketed for the calibration
validation period). The low sediment yield fit is not period and 13% of the data were bracketed for the
surprising, particularly in highly erosive regions, where validation period by 95PPU. The calculated R-factors
abrupt gully development may affect daily loads for the daily streamflow were 0.51 for the calibration
[16]
significantly. However, Betrie et al. reported that the periods and 0.49 for the validation period, whereas the
fit between the model daily sediment predictions and the R-factors for the daily sediment yield were 0.23 for the
observed concentrations showed good agreement as calibration periods and 0.18 for the validation period.
indicated by very good values of the NSE=0.88 for the The daily sediment data show exceptionally large
calibration period and NSE=0.83 for the validation period prediction uncertainties as compared to stream flow
at El Diem gauging station. During the calibration of prediction. This model uncertainties might be as a result
 62 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

of some errors in the data input sources, data preparation more than 65 Mg/hm2.
and parameterization[62]. Moreover, the uncertainties
might also be as a result of human and instrumental errors
during data processing[63]. Even though kinematic wave
runoff routing is used in the model, peaks of erosional
forces of the channel runoff might be underestimated,
especially in gully regions of changing flow directions
because of gully meanders and/or locally changed flow
conditions. Some of the potential reasons for such
unsatisfactory sediment yield simulations could probably
be, the length of overall measured data, which is quite
Figure 8 Scatterplot of discharge and sediment concentration of
short, strong hydrological heterogeneity and poor
the manual bottle sampling at the main outlet, where dashed lines
monitoring data as well as the use of USLE (or similar)
indicate the lower and upper defined limit of the expected relation
equations in areas where rainfall happens under the form between discharge and sediment concentration
of short intense rainfall events. Nevertheless,
calibration (and validation) of sediment yield on a
monthly basis may give much better results, but due to
plenty of gaps within the observed data, monthly
balancing is not possible for this study. The trends as
well as the order of magnitudes of sediment yield seem to
be achieved through modeling, and therefore, the model
may be able to describe long-term soil erosion Figure 9 Comparison of the observed range of daily sediment
characteristics, even if the event based predictions are yield (manual bottle sampling) and the simulated daily sediment
uncertain. In this study, sediment concentration was yield at the main outlet of the watershed

also manually sampled at three stages of various flood Although stone bunds reduce the slope length, and
events. Although selectively sampled sediment data decrease overland flow and sheet erosion, the calibrated
may not be suitable for daily based model calibration, model still predicted average annual sediment yields
sediment data was used to establish a relation between which were higher than the potential soil regeneration
runoff and sediment concentration (Figure 8). Based on rate. This indicates a need for expanding SWC practices
the manual bottle sampling upper and lower boundaries in the Gumara-Maksegnit watershed to further mitigate
of the expected sediment yield for certain discharge was soil erosion problems.
defined. Though it is commonly accepted that observed Compared to other studies from the literature, Gumara-
[45]
data are inherently uncertain , simulated sediment yield Maksegnit watershed study may provide conclusive
was compared to the expected sediment yield (Figure 9), results, for example, SWAT was applied for streamflow
and the observed sediment yield ranged from 2.9 Mg/hm2 simulation of Gedeb catchment, located at the upper Blue
2
to 27.6 Mg/hm , whereas the calibrated model predicted Nile River basin[12], which resulted in unsatisfactory
10.0 Mg/hm2 sediment yield for the observed period and model performance for both calibration and validation
21.08 Mg/hm annually.2
Similarly, Setegn et al. [19]
used period. However, Koch[12] pointed out various reasons
SWAT to simulate the sediment yield simulations for the for unsatisfactory model results, which seem also valid
Anjeni, a small watershed in the northern highlands of for the Gumara-Maksegnit case study; i.e., poor
Ethiopia, using different slope classifications and the monitoring data, strong hydrological heterogeneity and a
results showed a very high spatial variability for the difficult and remote terrain. In contrast, Setegn et al.[19]
obtained annual sediment yields, which ranged from 0 to reported very good SWAT model performance (NSE
 September, 2016 Addis H K, et al. Modeling streamflow and sediment using SWAT in the Ethiopian Highlands Vol. 9 No.5 63

equal to 0.81 during calibration) for monthly based runoff ratio (Curve Number) and support practice factor
sediment yield of Anjeni-gauged watershed. This may (P-factor) modification. Model calibration executed
indicate a well performing model on one hand, but on the through the SWAT-CUP software resulted in satisfactory
other hand the reasonable calibration result also model performance regarding streamflow. However,
demonstrates typical increasing accuracy of sediment poor agreement between daily observed and simulated
yield prediction for monthly based assessment. sediment yield resulted as indicated by the NSE=0.07 for
Typically, model simulations show a much better fit as the calibration period and –1.76 for the validation period.
[14,64,65]
the comparison time scale increases . There are Nevertheless, overall sediment dynamics and the order of
also a number of previous SWAT studies in Ethiopia, magnitude of various erosion events may be achieved
which documented satisfactory streamflow results through SWAT simulation. Because of acceptable
including studies that report daily comparisons within the streamflow simulation (NSE=0.56 for the calibration
[17,20]
Lake Tana drainage area . However, these are for period and 0.48 for the validation period), but
larger systems with longer overall observed data versus considerable imprecise daily sediment yield prediction at
the smaller Gumara-Maksegnit watershed analyzed in this the same time, it is possible that fluctuating sediment
study with quite short measured data. processes are influenced by abrupt gully bank breaks and
Generally, this study documented insufficiencies for gully network development. Highly variable sediment
matching daily based sediment yield simulation with transport in the main stream may be also a result of
observed data; this might be a result of poor monitoring distinct sub-daily runoff characteristics of the
data (e.g. short observation period, uncertain data Gumara-Maksegnit River, and therefore, daily based
inherent of the measurement technique, occasional data rainfall and streamflow processing may be limited to
gaps, etc.). Moreover, missing records inhibit the model describe variable sub-daily peak wave characteristics,
assessment on a larger time scale (such as monthly or inherently linked with variable sediment yield
yearly), which typically increases the goodness of the characteristics.
model fit. Hence, especially remote watershed Based on the calibrated SWAT model, the long-term
modeling suffers from lack of continuous and good average annual runoff at the main outlet was predicted to
quality data, which has to be considered for be 352 mm, while approximately one third of annual
semi-distributed eco-hydrological based modeling rainfall amount (373 mm) becomes evapotranspiration.
approaches for such areas. The model predicts 21.08 Mg/hm2 as an average annual
sediment yield, which is still alarming and far beyond the
4 Conclusions
potential soil regeneration rate, especially for the situation
In this research, SWAT watershed modeling was of largely applied SWC structures (mainly stone bunds)
performed to describe the driving hydrological and within the watershed. Thus, rethinking of performed
2
sediment transport related processes of a 53.7 km land management strategies and intensification of SWC
watershed in the Ethiopian Highlands. The collected interventions may be needed to achieve sustainable
model input data, either from remote earth observation or agriculture. The Ethiopian Highlands are a fragile
direct field sampling, are supposed to match SWAT ecoregion worthy of protection and physically-based
requirements, but limited monitoring data, strong modeling may be one method to guide scaling up of
hydrological heterogeneity and poor monitoring data as efficient measures to counteract ongoing land degradation.
well as the use of USLE (or similar) equations in areas Eventually, advanced SWC impact assessment may be
where rainfall happens under the form of short intense needed to satisfyingly consider the interaction between
rainfall events are inevitably connected with a large various SWC structures and heterogenic landscape
model uncertainty. Another source of uncertainty is the conditions to support proper decision making in the
simulated stone bund impacts applied through the surface future.
 64 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

Acknowledgements Ethiopian highlands. Applied Geography, 2013; 37(1):

139–149. doi: 10.1016/j.apgeog.2012.11.007
The authors thank the Austrian Development Agency
[10] Nyssen J, Poesen J, Desta G, Vancampenhout K, Daes M,
(ADA) as well as CGIAR Water Land and Ecosystems Yihdego G, et al. Interdisciplinary on-site evaluation of
(WLE) for providing the financial support. We also thank stone bunds to control soil erosion on cropland in northern
the Amhara Regional Agricultural Research Institute Ethiopia. Soil & Tillage Research, 2007; 94(1): 151–163.
doi: 10.1016/j.still.2006.07.011
(ARARI) for constant support and guidance. We thank
[11] Bosshart U. Catchment Discharge and Suspended Sediment
the students from BOKU University Vienna for their kind Transport as Indicators of Physical Soil and Water
assistance and their patient implementation of various Conservation in the Minchet Catchment, Anjeni Research
field experimental trails. Last but not least, we especially Unit, Soil Conservation Research Report 40, University of
Berne, Berne, Switzerland, 1997.
thank the staff from Gondar Agricultural Research Center
[12] Koch F J, Van Griensven A, Uhlenbrook S, Tekleab S, Teferi
for their frequent technical and logistical support and also
E. The Effects of land use change on hydrological
for creating sincere working environment. responses in the Choke Mountain Range (Ethiopia)—A
new approach addressing land use dynamics in the model
SWAT, International Environmental Modeling and
[References]
Software Society (iEMSs). In Proceedings of 2012
[1] Cerdà A, Flanagan D C, Bissonnais Y L, Boardman J. Soil
International Congress on Environmental Modeling and
erosion and agriculture. Soil and Tillage Research, 2009;
Software Managing Resources of a Limited Planet, Sixth
106(1): 107–108. doi: 10.1016/j.still.2009.1
Biennial Meeting, Leipzig, Germany, 1–5 July 2012.
[2] Wolka K, Moges A, Yimer F. Effects of level soil bunds
[13] Vente D J, Poesen J, Verstraeten G, Govers G, Vanmaercke
and stone bunds on soil properties and its implications for
M, Rompaey A V, et al. Predicting soil erosion and
crop production: the case of Bokole watershed, Dawuro zone,
sediment yield at regional scales: Where do we stand?
Southern Ethiopia. Agricultural Science, 2011; 2(3):
Earth-Science Reviews, 2013; 127(2): 16–29. doi:
357–363. doi: 10.4236/as.2011.23047
10.1016/j.earscirev.2013.08.014
[3] Nyssen J, Poesen J, Mitiku H, Moeyersons J, Deckers J.
[14] Gassman P W, Reyes M R, Green C H, Arnold J G. The
Tillage erosion on slopes with soil conservation structures in
soil and water assessment tool: historical development,
the Ethiopian highlands. Soil and Tillage Research, 2000;
applications, and future research directions. Transactions of
57(3): 115–127. doi: 10.1016/S0167-1987(00)00138-0
the ASABE, 2007; 50(4): 1211–1250. doi: 10.13031/
[4] Boardman J. Soil erosion science: Reflections on the
2013.23637
limitations of current approaches. Catena, 2006; 68(2):
[15] Gassman P W, Wang Y K. IJABE SWAT Special Issue:
73–86. doi: 10.1016/j.catena.2006.03.007
[5] Hurni H, Pimentel D. Land degradation, famine, and land Innovative modeling solutions for water resource problems.

resource scenarios in Ethiopia. In: Pimentel D. (Ed.) World Int J Agric & Biol Eng, 2015; 8(3): 1–8. doi: 10.3965/

soil erosion and conservation. Cambridge University Press, j.ijabe.20150803.1763

1993; pp. 27–61. [16] Betrie G D, Mohamed Y A, van Griensven A, Srinivasan R.

[6] Hurni H. Degradation and conservation of the resources in Sediment management modelling in the Blue Nile Basin
the Ethiopian highlands. Mountain Research and using SWAT model. Hydrology and Earth System Sciences,
Development, 1988; 8(2/3): 123–130. doi: 2011; 15(3): 807–818. doi: 10.5194/hess-15-807-2011
10.2307/3673438 [17] Mengistu D T, Sorteberg A. Sensitivity of SWAT
[7] Herweg K, Ludi E. The performance of selected soil and simulated streamflow to climatic changes within the Eastern
water conservation measures—case studies from Ethiopia Nile River basin. Hydrology and Earth System Sciences,
and Eritrea. Catena, 1999; 36(1): 99–114. doi: 10.1016/ 2012; 16(2): 391–407. doi: 10.5194/hess-16-391-2012
S0341-8162(99)00004-1 [18] Schmidt E, Zemadim B. Expanding sustainable land
[8] Morgan R P C. Soil erosion and conservation. Longman management in Ethiopia: Scenarios for improved agricultural
Scientific & Technical, J. Wiley, 1995. ISBN: water management in the Blue Nile. Agricultural Water
978-1-4051-4467-4 Management, 2015; 158: 166–178. doi: 10.1016/j.agwat.
[9] Teshome A, Rolker D, Graaff J D. Financial viability of 2015.05.001
soil and water conservation technologies in northwestern [19] Setegn S G, Dargahi B, Srinivasan R, Melesse A M.
 September, 2016 Addis H K, et al. Modeling streamflow and sediment using SWAT in the Ethiopian Highlands Vol. 9 No.5 65

Modeling of sediment yield from Anjeni-Gauged watershed, [31] Ayele M, Doleže J, Van Duren M, Brunner H, Zapata-Arias
Ethiopia using SWAT model. Journal of the American F J. Flow cytometric analysis of nuclear genome of the
Water Resource Association, 2010; 46(3): 514–526. Ethiopian cereal tef [Eragrostis tef (Zucc.) Trotter].
doi: 10.1111/j.1752-1688.2010.00431.x Genetica, 1996; 98(2): 211–215. doi: 10.1007/BF00121369
[20] Wosenie M D, Verhoest N, Pauwels V, Negatu T A, Poesen J, [32] Neitsch S L, Arnold J G, Kiniry J R, Williams J R. Soil
Adgo E, et al. Analyzing runoff processes through and Water Assessment Tool–Theoretical Documentation,
conceptual hydrological modeling in the Upper Blue Nile Version 2005. Grassland, Soil and Water Research
Basin, Ethiopia. Hydrology and Earth System Sciences, Laboratory, Agricultural Research Service, Blackland
2014; 18(12): 5149–5167. doi: 10.5194/hessd-11-5287- Research Center, Texas Agricultural Experiment Station,
2014 Texas. 2005.
[21] Yesuf H M, Assen M, Alamirew T, Melesse A M. [33] Ziadat F M, Dhanesh Y, Shoemate D, Srinivasan R,
Modeling of sediment yield in Maybar gauged watershed Narasimhan B, Tech J. Soil-Landscape Estimation and
using SWAT, northeast Ethiopia. Catena, 2015; 127: Evaluation Program (SLEEP) to predict spatial distribution of
191–205. doi: 10.1016/j.catena.2014.12.032 soil attributes for environmental modeling. International Int
[22] Rompaey A J V, Govers G. Data quality and model J Agric & Biol Eng, 2015; 8(3): 158–172. doi: 10.3965/
complexity for regional scale soil erosion prediction. j.ijabe.20150803.1270
International Journal of Geographical Information Science, [34] Smith K A, Mullins C E. Soil analysis: physical methods.

2002; 16(7): 663–680. doi: 10.1080/13658810210148561 New York: Marcel Dekker, Inc.. 1991.

[23] Jarvis A, Reuter H I, Nelson A, Guevara E. Hole-filled [35] Gee G W, Or D. 2.4 Particle-size analysis. In Dane J H,

SRTM for the globe Version 4. Available CGIAR-CSI Topp G C (eds). Methods of soil analysis. Part 4. Physical
methods. Madison, WI: Soil Sci. Soc. Am. Book Ser. 5.
SRTM 90m Database Httpsrtm Csi Cgiar Org, 2008.
2002; pp. 255-293.
[24] Mohr P A. The geology of Ethiopia. University
[36] De Vos B, Lettens S, Muys B, Deckers J A. Walkley–Black
College of Addis Ababa Press, Ethiopia. 1963.
analysis of forest soil organic carbon: recovery, limitations
[25] Arnold J G, Srinivasan R, Muttiah R S, Williams J R. Large
and uncertainty. Soil Use and Management, 2007; 23(3):
area hydrologic modeling and assessment part I: Model
221–229. doi: 10.1111/j.1475-2743.2007.00084.x
development. Journal of the American Water Resources
[37] Saxton K E, Rawls W J. Soil water characteristic estimates
Association, 1998; 34(1): 73–89. doi: 10.1111/j.1752-1688.
by texture and organic matter for hydrologic solutions. Soil
1998.tb05961.x
Science Society of America Journal, 2006; 70(5): 1569–1578.
[26] Olivera F, Valenzuela M, Srinivasan R, Choi J, Cho H, Koka
doi: 10.2136/sssaj2005.0117
S, et al. ARCGIS-SWAT: A geodata model and GIS
[38] Brenner C, Strohmeier S, Ziadat F, Klik A. Soil
interface for SWAT. Journal of the American Water
conservation measures in the Ethiopian Highlands: The
Resources Association, 2006; 42(2): 295–309. doi:
effectiveness of stone bunds on soil erosion processes. In
10.1111/j.1752-1688.2006.tb03839.x
EGU General Assembly Conference Abstracts, 2013; vol. 15,
[27] Baker T J, Miller S N. Using the Soil and Water
pp. 5392. Vienna.
Assessment Tool (SWAT) to assess land use impact on water
[39] Williams J R. Sediment routing for agricultural watersheds.
resources in an East African watershed. Journal of
Journal of the American Water Resources Association
Hydrology, 2013; 486(8): 100–111. doi: 10.1016/j.jhydrol.
(JAWRA), 1975; 11(5): 965–974. doi: 10.1111/j.1752-
2013.01.041
1688.1975.tb01817.x
[28] Neitsch S L, Arnold J G, Kiniry J R, Williams J R. Soil and [40] Zabaleta A, Meaurio M, Ruiz E, Antigüedad I. Simulation
Water Assessment Tool Theoretical Documentation Version climate change impact on runoff and sediment yield in a
2009: TR-406. Texas Water Resources Institute, College small watershed in the Basque Country, northern Spain.
Station, TX. 2011. Journal of Environmental Quality, 2014; 43: 235–245.
[29] Williams J R, Berndt H D. Sediment yield prediction based doi:10.2134/jeq2012.0209
on watershed hydrology. Transactions of the ASABE, 1977; [41] Rieder J, Strohmeier S, Demelash N, Ziadat F, Klik A.
20(6): 1100–1104. doi: 10.13031/2013.35710 Investigation of the impact of stone bunds on water erosion in
[30] Schuol J, Abbaspour K C. Using monthly weather statistics northern Ethiopia. In EGU General Assembly Conference
to generate daily data in a SWAT model application to West Abstracts, 2014; vol. 16, p. 3885. Vienna.
Africa. Ecological Modelling, 2007; 201(3): 301–311. doi: [42] Adimassu Z, Mekonnen K, Yirga C, Kessler A. Effect of
10.1016/j.ecolmodel.2006.09.028 soil bunds on runoff, soil and nutrient losses, and crop yield
 66 September, 2016 Int J Agric & Biol Eng Open Access at http://www.ijabe.org Vol. 9 No.5

in the central highlands of Ethiopia. Land Degradation & performance. Bulletin of the American Meteorological
Development, 2012; 25(6): 554–564. doi: 10.1002/ldr.2182 Society, 1982; 63(11): 1309–1313.
[43] Gebremichael D, Nyssen J, Poesen J, Deckers J, Haile M, [54] Nash J E, Sutcliffe J V. River flow forecasting through
Govers G, et al. Effectiveness of stone bunds in controlling conceptual models: Part I. A discussion of principles.
soil erosion on cropland in the Tigray Highlands, northern Journal of Hydrology, 1970; 10(3): 282–290. doi:
Ethiopia. Soil Use and Management, 2005; 21(3): 287–297. 10.1016/0022- 1694(70)90255-6.
doi: 10.1111/j.1475-2743.2005.tb00401.x [55] Krause P, Boyle D P, Bäse F. Comparison of different
[44] Arnold J G, Moriasi D N, Gassman P W, Abbaspour K C, efficiency criteria for hydrological model assessment.
White M J, Srinivasan R, et al. SWAT: Model use, Advances in Geosciences, 2005; 5(5): 89–97.
calibration, and validation. Transactions of the ASABE, [56] Gupta H V, Sorooshian S, Yapo P O. Status of automatic
2012; 55(4): 1491–1508. doi: 10.13031/2013.42256 calibration for hydrologic models: Comparison with
[45] Moriasi D N, Arnold J G, Van Liew M W, Bingner R L, multilevel expert calibration. Journal of Hydrologic
Harmel R D, Veith T L. Model evaluation guidelines for Engineering, 1999; 4(2): 135–143.
systematic quantification of accuracy in watershed [57] Stigler S M. Francis Galton’s account of the invention of
simulations. Transactions of the ASABE, 2007; 50(3): correlation. Statistical Science, 1989; 4(2): 73–79.

885–900. doi: 10.13031/2013.23153 [58] Yesuf H M, Melesse A M, Zeleke G, Alamirew T.

[46] Balascio C C, Palmeri D J, Gao H. Use of a genetic Streamflow prediction uncertainty analysis and verification

algorithm and multi-objective programming for calibration of of SWAT model in a tropical watershed. Environmental
Earth Sciences, 2016; 75(9): 1–6.
a hydrologic model. Transactions of the ASABE, 1998;
[59] Gebremicael T G, Mohamed Y A, Betrie G D, Zaag P V D,
41(3): 615–619. doi: 10.13031/2013.17229
Teferi E. Trend analysis of runoff and sediment fluxes in
[47] Gitau M W. A quantitative assessment of BMP
the Upper Blue Nile basin: A combined analysis of statistical
effectiveness for phosphorus pollution control: The Town
tests, physically-based models and landuse maps. Journal of
Brook Watershed. NY. Pennsylvania State University.
Hydrology, 2013; 482(2): 57–68. doi: 10.1016/j.jhydrol.
2003.
2012.12.023
[48] Zehetbauer I, Strohmeier S, Ziadat F, Klik A. Runoff and
[60] Alemu H, Kaptué A T, Senay G B, Wimberly M C, Henebry
sediment monitoring in an agricultural watershed in the
G M. Evapotranspiration in the Nile Basin: Identifying
Ethiopian Highlands. In EGU General Assembly
Dynamics and Drivers, 2002–2011. Water, 2015; 7(9):
Conference Abstracts, 2013; vol. 15, p. 5640. Vienna.
4914–4931. doi: 10.3390/w7094914
[49] Abbaspour K C, Vejdani M, Haghighat S. SWATCUP
[61] Moriasi D N, Gitau M W, Pai N, Daggupati P. Hydrologic
calibration and uncertainty programs for SWAT. In Proc.
and water quality models: Performance measures and
Intl. Congress on Modelling and Simulation (MODSIM’07),
evaluation criteria. Transactions of the ASABE, 2015;
2007; 1603-1609. L. Oxley and D. Kulasiri, eds. Melbourne,
58(6): 1763–1785. doi: 10.13031/trans.58.10715
Australia: Modelling and Simulation Society of Australia and
[62] Freer J, Beven K, Ambroise B. Bayesian estimation of
New Zealand.
uncertainty in runoff prediction and the value of data: An
[50] Gupta H V, Beven K J, Wagener T. Model calibration and application of the GLUE approach. Water Resource
uncertainty estimation. In Encyclopedia of Hydrological Research, 1996; 32(7): 2161–2173.
Sciences, 2006; 11–131. [63] Singh V, Bankar N, Salunkhe S S, Bera A K, Sharma J R.
[51] Chai T, Draxler R R. Root mean square error (RMSE) or Hydrological stream flow modelling on Tungabhadra
mean absolute error (MAE)?–Arguments against avoiding catchment: parameterization and uncertainty analysis using
RMSE in the literature. Geoscientific Model Development, SWAT CUP. Current Science, 2013; 104(9): 1187–1199.
2014; 7(3): 1247–1250. doi: 10.5194/gmd-7-1247-2014 [64] Engel B, Storm D, White M, Arnold J, Arabi M. A
[52] Hernandez-Stefanoni J L, Ponce-Hernandez R. Mapping hydrologic/ water quality model application protocol.
the spatial variability of plant diversity in a tropical forest: Journal of the American Water Resources Association.
comparison of spatial interpolation methods. 2007; 43(5): 1223–1236.
Environmental Monitoring and Assessment, 2006; 117: [65] Gassman P W, Sadeghi A M, Srinivasan R. Applications of
307–334. doi: 10.1007/s10661-006-0885-z the SWAT model special section: overview and insights.
[53] Willmott C J. Some comments on the evaluation of model Journal of Environmental Quality, 2014; 43(1): 1–8.