Interactive comment on “ Sediment management modelling in Blue Nile Basin using SWAT model ”

Comment: The paper has limitation in watershed delineation. If the runoff is simulated at El Diem, Sub basin no. 1 should not be included in the simulation as it drains directly to Sudan rather than through the main Blue Nile outlet or El Diem. The effect is clearly shown in the hydrographs as flow is over estimated in most of the cases. The area of the basin is reported to be 184560 sq.km, which is larger than the accepted range of 150,000 – 170,000 sq.km. Detail can be obtained by referring published articles on the same.


Introduction
The Blue Nile River, which originates from the steep mountains of the Ethiopian Plateau, is the major source of sediment loads in the Nile basin.Soil erosion from the upstream of the basin and the subsequent sedimentation in the downstream area is an immense problem threatening the existing and future water resources development in the Nile basin.The benefits gained by the construction of micro-dams in the Upper Nile are threatened by the rapid loss of storage volume due to excessive sedimentation (El-Swaify and Hurni, 1996;Tamene et al., 2006).Moreover, the green water storage of the Ethiopian highlands, where rainfed agriculture prevails is diminished because of top-soil loss and this has caused frequent agricultural drought (Hurni, 1993;El-Swaify and Hurni, 1996).In the downstream part of the basin (e.g., in Sudan and Egypt) excessive sediment load led to massive operation cost of irrigation canals desilting, and sediment dredging in front of hydropower turbines.For example, the Sinnar dam has lost 65% of its original storage after 62 years operation (Shahin, 1993) and the other dams (e.g., Rosieres and Khashm el Girba) lost similar proportions since construction (Ahmed, 2004).Both the Nile Basin Initiative and the Ethiopian government are developing ambitious plans of water resources projects in the Upper Blue Nile basin, locally called the Abbay basin (BCEOM, 1999;World Bank, 2006).Thus, an insight into the soil erosion/sedimentation mechanisms and the mitigation measures plays an indispensable role for the sustainable water resources development in the region.
Literature review shows there are many catchment models that include the soil erosion/sedimentation processes and simulate the effect of mitigation measures (Merritt et al., Published by Copernicus Publications on behalf of the European Geosciences Union.2003; Borah and Bera, 2003).However, there are a few applications of erosion modelling in the Upper Blue Nile basin.These include Zeleke (2000), Haregeweyn and Yohannes (2003), Mohamed et al. (2004), Hengsdijk et al. (2005), Steenhuis et al. (2009), and Setegn et al. (2010).Zeleke (2000) simulated soil loss using the Water Erosion Prediction Project (WEPP) model and the result slightly underestimated the observed soil loss in the Dembecha catchment (27 100 ha).Haregeweyn and Yohannes (2003) applied the Agricultural Non-Point Source (AGNPS) model and well predicted sediment yield in the Augucho catchment (224 ha).The same AGNPS model was used by Mohamed et al. (2004) to simulate sediment yield in the Kori (108 ha) catchment and the result was satisfactory.Hengsdijk et al. (2005) applied the Limburg Soil Erosion Model (LISEM) to simulate the effect of reforestation on soil erosion in the Kushet -Gobo Deguat catchment (369 ha), but the result raised controversy by Nyssen et al. (2005).The SWAT model was applied for simulation of a sediment yield by Setegn et al. (2010) in the Anjeni gauged catchment (110 ha) and the obtained result was quite acceptable.Steenhuis et al. (2009) calibrated and validated a simple soil erosion model in the Abbay (Upper Blue Nile) basin and reasonable agreement was obtained between the model predictions and the 10-day observed sediment concentration at El Diem located at the Ethiopia-Sudan border.
The application of models to simulate effectiveness of soil and water conservation practices in the Ethiopian highlands is less explored compared to the application of models to estimate soil losses.However, there are a few publications on the observed effects of soil and water conservation practices at a plot and at field scales (Herweg and Ludi, 1999;Descheemaeker et al., 2006;Gebremichael et al., 2005).Herweg and Ludi (1999) investigated the performance of selected soil and water measures at a plot scale and the results showed that a significant reduction in soil loss and runoff.Descheemaeker et al. (2006) reported a high capacity of sediment trapping at a field scale due to reforestation measures.Gebremichael et al. (2005) showed that a considerable sediment reductions by stone bunds at a field scale.
Most of the above modelling applications successfully attempted to estimate the sediment yield at a small catchment scale or evaluate their soil erosion model.However, there is no literature that shows models prediction on the effects of mitigation measures at a large scale in the Blue Nile.Therefore, the objective of this study is to model the spatially distributed soil erosion/sedimentation process in the Upper Blue Nile basin at a daily time step and assess the impact of different catchment management interventions on sediment yield.
A brief description of the Upper Blue Nile Basin is given in the next section, followed by a discussion of the methodology used.The third section presents the model results and discussion of different land management scenarios.Finally, the conclusion summarizes the main findings of the investigations.

Description of study area
The Upper Blue Nile River basin has a total area of 184, 560 km 2 , and is shown in Fig. 1.The Ethiopian Plateau is deeply incised by the Blue Nile River and its tributaries, with a general slope to the north-west.The elevation ranges from 500 m at Sudan border to 4230 m at the top of highlands.The Didessa and Dabus tributaries drain the south-western part of the basin, and contribute about one third of the total flow.The climate in the Blue Nile is governed by the seasonal migration of the Inter Tropical Convergence Zone ITCZ from the south to the north and back.The annual rainfall varies from 900 mm near the Ethiopia/Sudan border to 2200 mm in the Didessa and the Dabus subbasins.Since the rainfall is highly seasonal, the Blue Nile possesses a highly seasonal flood regime with over 80% of annual discharge occurs from July to October, while 4% of the flow occurs between January and April (Sutcliffe and Parks, 1999).In the basin the minimum and the maximum temperatures are 11 • C and 18 • C, respectively.The dominant soil types are Alisols and Leptosols 21%, followed by Nitosoils 16%, Vertisols 15% and Cambisols 9%.

SWAT model description
The Soil and Water Assessment Tool (SWAT) is a physical process based model to simulate continuous-time landscape processes at a catchment scale (Arnold et al., 1998;Neitsch et al., 2005).The catchment is divided into hydrological response units (HRUs) based on soil type, land use and slope classes that allows a high level of spatial detail simulation.The major model components include hydrology, weather, soil erosion, nutrients, soil temperature, crop growth, pesticides agricultural management and stream routing.
The SWAT model uses the Modified Universal Soil Loss Equations (MUSLE) to compute HRU-level soil erosion.It uses runoff energy to detach and transport sediment (Williams and Berndt, 1977).The sediment routing in the channel (Arnold et al., 1995) consists of channel degradation using stream power (Williams, 1980) and deposition in channel using fall velocity.Channel degradation is adjusted using USLE soil erodibility and channel cover factors.

SWAT model setup
The SWAT model inputs are Digital Elevation Model (DEM), landuse map, soil map, and weather data, which is shown in Table 1.The ArcGIS interface (Winchell et al., 2007) of the SWAT2005 version was used to discretize a watershed and extract the SWAT model input files.The DEM was used to delineate the catchment and provide topographic parameters such as overland slope and slope length for each subbasin.The catchment area of the Upper Blue Nile was delineated and discretized into 15 subbasins using a 90 m DEM (http: //srtm.csi.cgiar.org).
The landuse map of the Global Land Cover Characterization (GLCC) was used to estimate vegetation and their parameters input to the model.The GLCC is part of the United States Geological Survey (USGS) database, with a spatial resolution of 1 km and 24 classes of landuse representation (http://edcsns17.cr.usgs.gov/glcc/glcc.html).The parameterization of the landuse classes (e.g.leaf area index, maximum stomatal conductance, maximum root depth, optimal and minimum temperature for plant growth) is based on the available SWAT landuse classes.Table 2 shows the land use and land cover types and their area coverage in the Upper Blue Nile.The land cover classes derived are Residential area 0.2%, Dryland Cropland 17%, Cropland 5.8%, Grassland 2.5%, Shrubland 1.1%, Savanna 68.8%, Deciduous Forest 0.02%, Evergreen Forest 1.6%, Mixed Forest 0.7%, Water Body 2.2%, and Barren 0.4%.
The soil types of the study area were extracted from the SOIL-FAO database, Food and Agriculture Organization of the United Nations (FAO, 1995).There are around 23 soil types, with a spatial resolution of 10 km with soil properties for two layers (0-30 cm and 30-100 cm depth).The soil properties (e.g.particle-size distribution, bulk density, organic carbon content, available water capacity, and saturated hydraulic conductivity) were obtained from Batjes (2002).
The USGS landuse, the FAO soil and the slope class maps were overlaid to derive 1747 unique HRUs.Although the SWAT model provides an option to reduce the number of HRUs to decrease the computation time required for the simulation, we considered all of the HRUs to evaluate the watershed management intervention impact.
Daily precipitation and minimum and maximum temperature data at 17 stations interpolated spatially over the basin were used to run the model.Most of the stations were either established recently or had a lot of missing data.Therefore, a weather generator based on monthly statistics was used to fill in the gaps.Solar radiation and wind speed were generated by the weather generator.
Daily river flow and sediment concentration data measured at El Diem gauging station (see Fig. 1) were used for the model calibration and validation.Although we know that calibrating the model at the subbasin outlets would improve the model parameterization, we could not perform it due to lack of data.The flow observations are available throughout the year, while the sediment concentrations are usually monitored during the main rainy season, which is between June and October.The Blue Nile water is relatively sediment free during the remaining months.
The model was run daily for 12 years; the period from 1990 to 1996 was used for the calibration and the period from 1998 to 2003 was used for the validation.The modelling period selection considered data availability and avoided rapid landuse/cover change that was documented as alarming until the late 1980's by Zeleke et al. (2000) and Zeleke and Hurni (2001).Sensitivity analysis was carried out to identify the most sensitive parameters for the model calibration using One-factor-At-a-Time (LH-OAT), which is an automatic sensitivity analysis tool implemented in SWAT (van Griensven et al., 2006).Those sensitive parameters were automatically calibrated using the Sequential Uncertainty Fitting (SUFI-2) algorithm (Abbaspour et al., 2004;Abbaspour et al., 2007).

Model performance evaluation
Model evaluation is an essential measure to verify the robustness of the model.In this study, three model evaluation methods were used following Moriasi et al. (2007) model evaluation guideline.These methods are (i) Nash-Sutcliffe efficiency (NSE), (ii) percent bias (PBIAS), and (iii) ratio of the root mean square error to the standard deviation of measured data (RSR).The Nash-Sutcliffe efficiency (NSE) is computed as the ratio of residual variance to measured data variances (Nash and Sutcliffe, 1970).The Nash-Sutcliffe is calculated using Eq. ( 1): Where: -X obs i = observed variable (flow in m 3 s −1 or sediment concentration in mg l −1 ).
-X sim i = simulated variable (flow in m 3 s −1 or sediment concentration in mg l −1 ).
-X mean = mean of n values.
n = number of observations.The Percent bias (PBIAS) measures the average tendency of the simulated data to be larger or smaller than their observed counterparts (Gupta et al., 1999).The PBIAS is calculated with Eq. ( 2): The ratio of root mean square error to the standard deviation of measured data (RSR) is calculated as the ratio of the Root Mean Square Error (RMSE) and standard deviation of the observed data (Moriasi et al., 2007), as shown in Eq. ( 3): According to Moriasi et al. (2007) model simulation judged as satisfactory if NSE > 0.5, RSR ≤ 0.70 and PBIAS = ±25% for flow and NSE > 0.5, RSR ≤ 0.70 and PBIAS = ± 55% for sediment.

Catchment management intervention scenarios
Catchment management intervention involves introducing best management practices (BMPs) to reduce soil erosion and sediment transport.The SWAT model was applied to simulate the impact of BMPs on sediment reduction in the U.S. (Vache et al., 2002;Santhi et al., 2005;Bracmort et al., 2006).These BMPs were represented in the SWAT model by modifying SWAT parameters to reflect the effect the practice has on the processes simulated within SWAT (Bracmort et al., 2006).However, selection of BMPs and their parameters values is site specific and should reflect the study area reality.For this study, we selected appropriate BMPs and their parameters values based on documented local research experience in the Ethiopian highlands (Hurni, 1985;Herweg and Ludi, 1999;Gebremichael et al., 2005).
The scenarios simulated and representations of BMPs in the SWAT are depicted in Table 3.In Scenario 0, the basin existing conditions is considered.In Scenario 1, filter strips were placed on all agricultural HRUs that are the combination of dryland cropland, all soil types and slope classes.The effect of the filter strip is to filter the runoff and trap the sediment in a given plot (Bracmort et al., 2006).Appropriate model parameter for representation of the effect of filter strips is width of filter strip (FILTERW).FILTERW value of 1m was assigned to simulate the impact of filter strips on sediment trapping.This value was modified by editing the HRU (.hru) input table.The filter width value was assigned based on local research experience in the Ethiopian highlands (Hurni, 1985;Herweg and Ludi, 1999).
In Scenario 2, stone bunds were placed on agricultural HRUs that are the combination of dryland cropland, all soil types and slope classes.This practice has a function to reduce overland flow, sheet erosion and reduce slope length (Bracmort et al., 2006).Appropriate parameters for representing the effect of stone bunds are the Curve Number (CN2), average slope length (SLSUBBSN) and the USLE support practice factor (USLE P).We modified SLSSUBSN value by editing the HRU (.hru) input table, whereas USLE P and CN2 values were modified by editing Management (.mgt) input table.The SWAT model assigns the SLSUB-BSN parameter value based on the slope classes.In this application, the SWAT assigned values were 61 m, 24 m and 9.1 m for slope classes 0-10%, 10-20% and over 20%, respectively.The modified parameters values were SLSUB-BSN is equal to 10 m for 0-10% and 10-20% slope classes, USLE P is equal to 0.32, and CN2 is equal to 59 as depicted in Table 3.The SLSUBBSN represents the spacing between successive stone bunds at field condition and the modified value was used as reported by Hurni (1985) and Herweg and Ludi (1999).Similarly, USLE P value was obtained from documented field experience by Gebremichael et al. (2005).The CN2 value was obtained from the SWAT user's manual version 2005 for contoured and terraced condition (Neitsch et al., 2005).
In Scenario 3, we simulated the impact of reforestation on sheet erosion.The reforestation has a function to reduce overland flow and rainfall erosivity.The reforestation effect was simulated by introducing land use change, not by parameters changes.It was deemed impractical to change agricultural land into forest completely.Thus we replaced 8% of the area occupied by cropland, shrubland, barren, mixed forest, and deciduous forest into evergreen forest.The evergreen forest was selected since it has a wider coverage than other types of forest in the study area, as shown in Table 2.The associated parameters (e.g., plant, hydrological and erosion) for the new landuse were changed by the SWAT model from the database.

Model calibration and validation
The most sensitive parameters for flow predictions were curve number (CN2), baseflow alpha factor (ALPHA BF), groundwater delay time (GW DELAY), threshold water depth in the shallow aquifer (GWQMN), ground water "revap" co-efficient (GW REVAP), threshold water depth in the shallow aquifer for "revap" (REVAPMN), soil evaporation compensation factor (ESCO), recharge to deep aquifer (RCHRG DP), channel effective hydraulic conductivity (CH K2), available water capacity (SOL AWC), saturated hydraulic conductivity (SOL K), surface runoff lag time (SURLAG), average slope length (SLSUBBSN) and Manning's "n" value for main channel (CH N2).These flow parameters are used to calculate the amount of flow from the www.hydrol-earth-syst-sci.net/15/807/2011/ Hydrol.Earth Syst.Sci., 15, 807-818, 2011  The model slightly over predicted the flow on the rising limb and slightly under predicted the flow on the receding limb in the calibration and the validation periods (Fig. 2).There could be many reasons for the slight over and under prediction of the flow but most likely it is due to Curve Number (CN2) method that is used to predict the surface runoff.The CN2 method assumes a unique relationship between cumulative rainfall and cumulative runoff for the same antecedent moisture condition.In the Ethiopian Highlands, however, Liu et al. (2008) showed that the ratio of discharge to effective precipitation (Q/(P-ET)) is increasing with cumulative precipitation and consequently the watersheds behave differently depending on how much moisture is stored in the watershed, suggesting that saturation excess processes play an important role in watershed response.The simulated peak flow in August was slightly under predicted in four out of seven years during the calibration period.In the validation period, however, the model slightly over and under predicted the peak flow in four out of five years.There could be various reasons for the peak flow mismatch but it is most likely attributed to precipitation data.Steenhuis et al. (2009) also reported that a limited precipitation data is the main constraint for accurate flow modelling in the Blue Nile.However, it is interesting that the model well simulated the main rainy season flow, which is very important for sediment simulation as shown in Fig. 3.The SWAT sediment predictions were calibrated against measured data from 1990 to 1996 and validated from 1998 to 2003 at El Diem gauging station using daily sediment concentrations, as depicted in Fig. 4.However, sediment concentrations data are available only for the rainy season, which occurs from July to October.The fit between the model sediment predictions and the observed concentrations showed good agreement as indicated by acceptable values of the NSE = 0.88, RSR = 0.35 and PBIAS = −0.05% in the calibration period and NES = 0.83, RSR = 0.61 and PBIAS = −11% in the validation period.The aggregated monthly average sediment concentrations from daily values improved the match between predictions and observed sediment concentrations.This is shown by NES = 0.92, RSR = 0.29 and PBIAS = −0.21%for the calibration and NES = 0.88, RSR = 0.34, and PBIAS = −11% for validation periods.This model performance is comparable to the recent results reported by Steenhuis et al. (2009) and Easton et al. (2010).Steenhuis et al. (2009) results showed that NSE = 0.75 for the calibration and NSE = 0.69 for the vali-dation and Easton et al. (2010) reported NSE = 0.74 for the calibration at El Diem gauging station.
The model well simulated the sediment concentrations on the rising and the falling limbs of the sediment hydrograph in the calibration period (Fig. 4).Although the sediment peak was well captured in most of the calibration years, the model slightly under predicted the sediment peaks in 1993 and 1994.In contrast, the model over predicted the peak concentrations in the validation period except in 1998.The model well simulated the rising limb sediment concentrations in the whole validation period.The predicted sediment concentrations on the falling limb in the validation period duplicated the observation except in 2002 and 2003.

Scenario analysis
The assessment of the spatial variability of soil erosion is useful for catchment management planning.The soil erosion prone areas in the Upper Blue Nile basin are shown in Fig. 5.The SWAT model simulation shows that the soil erosion extent varies from negligible erosion to over 150 t ha −1 .The soil erosion level in the basin classified into low (0-20 t ha −1 yr −1 ), moderate (20-70 t ha −1 yr −1 ), severe (70-150 t ha −1 yr −1 ) and extreme (≥150 t ha −1 yr −1 ) categories.The low class represents the erosion extent less than the soil formation rates, which is 22 t ha −1 yr −1 in the Ethiopian highlands (Hurni, 1983).The moderate class represents erosion level less than the average soil loss from cultivated land, which is 72 t ha −1 yr −1 (Hurni, 1985).The extreme class represents one fold higher than the average soil loss and the severe class represents two folds higher than average soil loss.The extreme erosion was observed in the cultivated land and low erosion was observed in the savannah land.Extreme erosion was dominant in subbasins 2, 3, and 4. Severe erosion was dominant in subbasins 8, 9, 12, 13 and 15.Moderate erosion was dominant in subbasins 1, 5, and 6; and low erosion was dominant in subbasins 7, 10, 11, and 14.These results show that the erosion level variations within a subbasin and the basin that is very helpful to prioritise BMPs implementation area.Moreover, these results showed that the sediment transport to the main river decreases from the north-east to the south-west of the basin.However, emphasis should be given to relative erosion level than the absolute values because the model was not parameterized at the subbasin outlets due to lack of data.
The observed average sediment yield at the outlet of the Upper Blue Nile was 131×10 6 tyr −1 .The SWAT model predicted 117 × 10 6 tyr −1 for existing conditions.This result is quite comparable with 140 × 10 6 tyr −1 estimate by NBCBN (2005) that includes bed load as well.The bed load approximately accounts for 20-25% of the total load.However, running the model with different catchment management scenarios provided very interesting results.The simulation of filter strips scenario reduced the total sediment yield to 66 × 10 6 tyr −1 from current conditions at El Diem, which is equivalent to 44% reduction.The simulation of stone bunds scenario reduced the total sediment yield to 70 × 10 6 tyr −1 from current conditions, which is equivalent to 41% reduction.The simulation of reforestation scenario showed the least reduction of sediment loads (104×10 6 tyr −1 ) from current conditions at El Diem, which is 11% reduction.This less sediment reductions under reforestation scenario could be attributed to smaller implementation area compared to filter strips and stone bunds implementation area as depicted in Fig. 6.This means, the effect of reforestation scenario on sediment reductions is masked by greater sediment yields from the agricultural land.The filter strips scenario showed greater sediment reductions than stone bunds scenario for equal implementation area.
The impact of BMPs at the subbasin level showed a wider spatial variability on sediment reduction from current conditions as is shown in Fig. 7.The sediment reductions ranged from 29% to 68% under filter strips scenario, 9% to 69% under stone bunds scenario and 46% to 77% under reforestation scenario.The least reductions for filter strips scenario (29%) and stone bunds scenario (10% and 9%) were exhibited in subbasins 3 and 8. Conversely, the reforestation scenario reduced sediment yields by 46% in subbasins 3 and 8.It was observed that filter strips and stone bunds effectiveness became greater as the agricultural area decreased and the proportion of the area for slope class ≤20% increased.This is expected because a higher overland flow concentration occurs as the steepness and a field size increased.The reforestation effectiveness became greater as the percent of agricultural area decreased in a subbasin.This is expected because the sediment yield from agricultural area is higher, and subsequently, masks the effectiveness of the reforestation on sediment reduction.It is important to note that the reforestation effect is greater at the subbasin level than at the basin level.This is attributed to the reforestation implementation area at the subbasin level is greater than at the basin level.This result corroborates Santhi et al. (2005) findings that showed reductions in sediment and nutrient up to 99% at farm level and 1-2% at the watershed level.The sediment reduction effectiveness per ha of each BMP is shown in Table 5.It shows that sediment reductions per ha of each BMP is not consistent across the entire basin.These results revealed that the BMPs became less effective as the subbasins located further from the basin outlet.Furthermore, reforestation measure was more effective than filter strips and stone bunds for the subbasins further from the basin outlet.
The BMPs sediment reductions scenarios from the model were compared to the available literature values to verify the obtained results.The filter strips sediment reductions by 44% at El Diem seems over predicted compared to Verstraeten et al. (2006).These researchers reported low (20%) performance of filter strips at field scale due to overland flow convergence and sediment bypasses of filter strips through ditches.This is expected because filter strips become less effective as the scale increase from plot to field due to flow concentration (Dillaha et al., 1989;Verstraeten et al., 2006).The reason for the higher sediment reductions by the filter strips in this study attributed to the filter strip algorithm uses the same filtering efficiency for sediment and all nutrient forms, and it does not consider flow concentrations at the field scale (White and Arnold, 2009).To obtain a better estimate of the effectiveness of the filter strips, the improved vegetative filter strip (VFS) sub-model of SWAT2009 version should be used than the width of filter strip.
Stone bunds sediment yield reductions were quite comparable to results reported in literature (Herweg and Ludi, 1999;Gebremichael et al., 2005).Herweg and Ludi (1999) reported 72%-100% sediment yield reductions by stone bunds at plot scale in the Ethiopian and the Eritrean highlands.Gebremichael et al. (2005) reported 68% reductions of sediment yields by stone bunds at the field scale in the northern part of Ethiopia.It is worth noting that the scaling effect between a plot and a field is minimal even at field observation.The   2006) that reported the complete sediment yield reductions by reforestation of degraded land in northern Ethiopia.The higher sediment yield reductions observed by Descheemaeker et al. (2006) was due to the reforestation area were located down-slope from cultivated land.

Conclusions
The SWAT model was applied to model spatially distributed soil erosion/sedimentation processes at daily time step and to assess the impact of three Best Management Practices (BMPs) scenarios on sediment reductions in the Upper Blue Nile River basin.The model showed that the erosion prone area at Hydrological Response Units (HRUs) level, which is already useful information for catchment management planning.For existing conditions scenario, a reasonable agreement was obtained between the model sediment yields predictions and measured sediment yields at the basin outlet.
The simulation results showed that applying filter strips, stone bunds and reforestation scenarios reduced the current sediment yields both at the subbasins and the basin outlets.
The effectiveness of each BMP, however, depends upon the percentage of land available, and local topographical conditions in the basin.The potential effect of the BMPs could be obtained by implementing reforestation in steep areas, and filter strips and stone bunds in low slope areas of the catchment.These results indicate that applying BMPs could be effective in reducing sediment transport for sustainable water resources management in the basin.However, any implementation of catchment management measures to reduce sediment yields involves the use of resources and willingness of decision makers.This study shows that modelling approach could be helpful for decision makers to evaluate the cost and benefits of particular BMP measures.A definitive interpretation of the quantitative results may not be appropriate because some processes are not well represented in the SWAT model (e.g., deterioration of the BMPs, flow concentrations in filter strip and gully erosion) and lack of model parameterization at local scale (e.g., subbasin or HRU level).While this study was able to give relative estimates of the erosion measures, further model parameterization at a local scale should be done as more data and information become available.

Fig. 1 .
Fig. 1.Location map of the Upper Blue Nile.

Fig. 3 .
Fig. 3. Observed and simulated daily wet season (from June to October) flow hydrograph at El Diem station, calibration (top) and validation (bottom).

Fig. 5 .
Fig. 5. Relative erosion prone areas (predicted sediment yield at each HRU by the SWAT model) for existing conditions in the Upper Blue Nile.

Fig. 7 .
Fig. 7. Percent reductions in sediment yield due to BMPs at subbasins level of the Upper Blue Nile basin (the basin outlet is located in the subbasin 4).

Table 1 .
Spatial model input data for the Upper Blue Nile.

Table 2 .
Land use/Land cover types and area coverage in the Upper Blue Nile.

Table 3 .
Scenarios description and SWAT parameters used to represent BMPs.The extensions, .hruand .mgtare input files, where parameter value was edited.

Table 4 .
SWAT sensitive parameters and fitted values.

Table 5 .
Sediment reductions effectiveness per ha of BMPs.