Simulation of sediment yield and evaluation of best management practices in Azuari watershed, Upper Blue Nile Basin

Soil erosion is the process of wearing away the topsoil of a field by erosive forces such as water and wind. Soil erosion by water is one of the most important land degradation problems, and a serious environmental threat in the world (Duru et al. 2018; Bhatti et al. 2021). When soil particles are eroded, they are moved and dumped somewhere else as they are no longer in contact with the original soil (Balasubramanian 2017). Soil erosion and sediment deposition processes combine to produce sediment yield. It is defined as the total sediment output from a watershed for a certain period, as measured at a point of reference (Bashewal & Kamal 2021). Common contributing factors for soil erosion and sediment yield include uncontrolled expansion of agricultural lands, cultivation of steep lands, urbanization, and deforestation without proper management (Kidane & Alemu 2015; Burgan 2022). Inappropriate management and soil conservation strategies lead to excessive erosion of topsoil during heavy precipitations which increases runoff and further increases sediment transport and accelerates sedimentation in lakes and storage reservoirs (Ali et al. 2014).

Soil erosion and sedimentation are major issues in Ethiopia due to a lack of land use planning, over-cultivation, and overgrazing (Demelash 2010; Dibaba et al. 2021). Rivers are observed to carry a full load of sediment showing that a considerable amount of soil is washed away from various watersheds of the country, mostly during the rainy seasons. The soil eroded from uplands is deposited in the downstream areas, which often results in the siltation of dams and water reservoirs, pollution of water sources, and the destruction of fertile agricultural land. Moreover, the deposition of sediment in natural stream channels, drainage ditches, and irrigation canals creates a loss of services and increases cleanout costs. Further, the deposition of sediment in the river channels decreases the channel capacity and results in flooding of the surrounding areas due to overflows.

Large areas of farming and deforestation are common phenomena in Ethiopia's Blue Nile River basin (Yaekob et al. 2020). With the constant change in land use and agricultural activities in the area, sediment generation and transport have become very complex (Tramblay et al. 2010). Deforestation exposes soils to increased water erosion, especially on steep terrain (Nedjai et al. 2013; Brown et al. 2014). The reduction of vegetation cover on lower slopes also increases soil erosion. This does degrade the soil fertility over time and reduces the suitability of land for agricultural use. For instance, the soil losses due to runoff within the Blue Nile River basin are recorded to reach up to 400 t/hayr (Yaekob et al. 2020). The physical removal of top soils in the highlands increases sedimentation downstream (Cerdà et al. 2009). The eroded sediment particles by the flowing water are usually settled in reservoirs, river channels, and irrigation canals (Ali et al. 2014).

Azuari watershed is one of the watersheds in the North Gojam sub-basin of the Upper Blue Nile River basin. The watershed drains to the Azuari River, which is the tributary of the Blue Nile River. The area receives heavy rainfall with mean annual values ranging between 908.68 and 1,539.64 mm. As the topography of the area is mountainous, it is highly susceptible to erosion and environmental degradation. The main causes of such environmental problems in the watershed are the expansion of cultivated land and deforestation (Fisseha et al. 2011). There are many studies on erosion and sediment problems in the different sub-basins of the Upper Blue Nile River basin (Ayele et al. 2017; Moges et al. 2018; Leta et al. 2023). However, very few studies are available on sediment transport and its management in the Azuari watershed. Thus, understanding sediment transport mechanisms, its deposition level downstream, and evaluation of mitigation techniques to control it are crucial for long-term water resource development (Betrie et al. 2011).

It is known that the sediment transport mechanism is directly related to precipitation and runoff in a hydrological basin (Burgan 2022). Different hydrological models are available for the simulation of sediment and evaluation of different watershed management practices from hydrometeorological data. However, the choice of an appropriate model is largely influenced by the function that the model must fulfill (Ayele et al. 2017). Many recent studies in various parts of the world highlighted the use of the Soil and Water Assessment Tool (SWAT) model to assess sediment yield and evaluate different management practices to control sediments in river basins. Azari et al. (2016) evaluated the watershed sediment yield using the SWAT model in the Northern Forests of Iran. Similarly, Sok et al. (2020) reported the use of the SWAT model to assess sediment yield in the Upper Mekong Basin, China. Recently, Nepal & Parajuli (2022) also reported the good performance of the SWAT model to assess the best management practices (BMPs) to control sediment at a watershed in Mississippi. Furthermore, Ricci et al. (2018) assessed the SWAT model's applicability for simulating runoff and sediment loss in the Carapelle Mediterranean watershed. The SWAT model has also been used in the Blue Nile Basin, Ethiopia, to assess factors such as soil productivity loss, water-induced erosion and unsustainable land management practices, and the impact of land use changes and reforestation on sediment yield in the region (Van Griensven et al. 2012; Sultana et al. 2019; Abebe et al. 2022; Leta et al. 2023).

Therefore, the SWAT model was selected for this study to simulate sediment yield and evaluate the BMPs to control soil loss and sediment transport in high-risk areas due to its computational efficiency, data requirements, and application level (Himanshu et al. 2019).

Meteorological data were obtained from the National Meteorological Agency (NMA) of Ethiopia. The SWAT model requires daily values of rainfall, maximum and minimum temperature, solar radiation, relative humidity, and wind speed. The daily values of these data for the periods from 1991 to 2020 were collected from four meteorological stations in the area, namely, Motta, Gundowin, Debrework, and Robgebeya. The selection of meteorological gauging stations was typically chosen based on physical similarities and regionalization approaches to calibrate and validate the SWAT model.

Daily stream flow data was collected from the Ministry of Water and Energy (MoWE) for the periods of 1988–2012. The data were used for the derivation of sediment yield from measured sediment concentration.

Sediment data for the Azuari River was obtained from the MoWE in a concentrated form. The data were collected from the periods from 1988 to 2012. As the sediment data was in a concentrated form, to change it to sediment yield, a sediment rating curve is required. The sediment rating curve is a plot of stream flow discharge and suspended sediment concentration. It is commonly used for the calculation of average sediment discharge from water discharge when sediment samples are not adequate.

The first step in the SWAT model setup is defining watershed boundaries from a DEM. Thus, the delineation of the studied watershed and its subwatersheds were done using DEM data and the normal SWAT watershed delineation process which includes five major steps, DEM setup, stream definition, outlet and inlet definition, watershed outlets selection, and definition and calculation of sub-basin parameters. For the stream definition, the threshold-based stream definition option was used to define the minimum size of the subwatershed to minimize uncertainty associated with model outputs.

SWAT uses the concept of HRUs which refers to the portion of a subwatershed that possesses unique land use and soil attributes. After watershed and subwatersheds delineation, the HRU analysis was done. The HRU analysis requires land use, soil, and slope data. After adjusting the projection of land use and soil, and by preparing the SWAT code related to the SWAT database, the land use and soil maps together with the multiple slopes map were overlaid to create HRU feature classes. To limit the number of HRUs, a threshold of 5% land cover, 10% slope, and 20% soil was used.

After sediment simulation, a sensitivity analysis was done to identify sensitive parameters for model calibration. A total of nine parameters were selected, and the ranks of sensitive parameters were made depending on global sensitivity analyses p-value and t-statistic. Then, the sensitivity analysis was ranked from most sensitive to least sensitive.

After the sensitive parameters identification, calibration of the model was executed to evaluate the performance of the model simulation using the SWAT_CUP tool. Model calibration and validation were carried out by minimizing the difference between the observed and simulated sediment. Due to limited data, sediment calibration was conducted for the years 1988 and 2004 on a monthly basis. But the first 2 years were considered for a model warm-up period. For validation, sediment data from the year 2005 to 2012 were used.

The performance of the model for sediment simulation was measured using the coefficient of determination (R²), Nash–Sutcliffe efficiency (NSE), and root mean square error observations standard deviation ratio (RSR).

The NSE value ranges between 0 and 1. It is 1 if the measured value perfectly matches all forecasts. The forecasts are poor if the NSE is negative, indicating that the average output value is a better estimate than the model forecast (Sahu et al. 2020). Moriasi et al. (2007) stated that an NSE value larger than 0.50 is satisfactory for SWAT model simulation.

RSR varies from the optimal value of 0, which indicates zero RMSE or residual variation and therefore perfect model simulation, to a large positive value. The lower the RSR, the lower the RMSE, and the better the model simulation performance. Moriasi et al. (2007) stated that an RSR value larger than 0.60 is satisfactory for SWAT model simulation.

In this study, the identification of hotspot (sediment-prone) areas was done based on the sediment yield of each subwatershed. Based on Hurni (1985), the severity of erosion-prone areas was evaluated based on Table 2.

Implementing appropriate BMPs in severely affected or hot spot areas is critical for reducing sediment movement and soil erosion in watersheds. But the selection of BMPs and the values of their parameters depend on the study area reality (Betrie et al. 2011). From research experience in the Ethiopian highlands (Hurni 1985), the following five scenarios were considered for the selection of BMPs.

This scenario indicates the present condition of the watershed without any consideration of management practices.

Filter strips reduce overland flow velocity which results in the deposition of particles. Filter strips were placed on all agricultural HRUs, all soil types, and slope classes. An appropriate model parameter used for the representation of the effect of filter strips was the width of the filter strip (FILTERW). FILTERW was modified by editing the HRU (.hru) input table of default 0 value by 1 m filter width value.

A terrace is an embankment within a field designed to intercept runoff and prevent erosion. A terrace is constructed across the slope following the general contour lines. Terraces were placed on all soil types, slope classes, and agricultural HRUs. Appropriate model parameters used for the representation of the effect of terraces are average slope length (SLSUBBSN) and USLE support practice factor (P USLE). The P USLE value was modified by editing the HRU (.hru) and management (.mgt) input table values, respectively. The SWAT model assigns the SLSUBBSN parameter value based on the slope classes. In this application, the SWAT assigned values were 61, 24, 18.3, 15.24, and 9.14 m for slope classes 0–10%, 10–15%, 15–20%, 20–25%, and above 25%, respectively. The modified parameter values for SLSUBBSN are equal to 10 m for all 0–25% and not modified for slopes greater than 25% slope classes. Hurni (1985) recommended that the minimum P factor be adjusted to 0.5 for terracing practice throughout Ethiopia. So PUSLE was modified from the calibrated value of 0.59–0.5 by editing the (.mgt) input table.

Strip cropping is simulated in SWAT by changing Manning's n value for overland flow (STRIP-N) to represent increased surface roughness in the direction of runoff. Curve number (STRIP-CN) was adjusted to account for increased infiltration. USLE Cropping factor (STRIP-C) was adjusted to reveal the mean value for multiple crops within the field. The USLE practice factor (STRIP_P) may also be updated to represent strip cropping conditions. STRIP-N, TERR-CN, STRIP-CN, STRIP-C, and STRIP-P were modified by adding and editing the operations (.Ops) input table of their default values of 0.15, 60, 0.4, and 0.7 by 0.15, 59, 0.2, and 0.5, respectively (Table 3).

Contour planting parameters were modified in SWAT by changing the curve number (CONT-CN) for surface storage and infiltration and the USLE practice factor (CONT_P) for erosion. Initial SCS curve number II (CONT_CN) and contouring PUSLE factor (CONT_P) model parameters were used for adjustment of the effect of contouring. Contouring parameters were modified by adding and editing the operations (.Ops) input table of the default values of 60 and 0.6 for CN and P by 59 and 0.5, respectively (Table 3).

The sensitivity parameters for sediment simulation in SWAT were ranked based on p-value and t-statistics. The larger absolute value t-stat and p-value close to zero is the most sensitive parameter. The most sensitive parameters for sediment identified in this study are Manning's n, the crop cover factor, and the soil erodibility factor (Table 4).

The resource considerations for the implementation of watershed management programmes may limit the implementation to a few watersheds. Thus, it is always better to start management measures from the highest priority subwatershed. There are different sediment mitigation measures in the SWAT model. In this study, four BMPs (scenarios) were evaluated on the six critical subwatersheds, namely SW-1, SW-2, SW-4, SW-5, SW-6, and SW-7. The four BMPs applied on these subwatersheds are filter strip, terracing, strip cropping, and contouring (Table 7). The effects of BMPs on sediment reduction were compared with a baseline scenario (scenario 0) (no management practices).

Without receiving any management practices, the average simulated sediment yield of the watershed was 10.81 t/ha/yr. However, with the application of the filter strips with a 1 m width of strips (scenario 1), the mean sediment yield was found to be 6.96 t/ha/yr, which exhibited a reduction of the sediment by 35.61% from the baseline. By applying terracing (scenario 2), the average sediment yield was found to be 8.6 t/ha/yr, which is a reduction of 20.44%. Similarly, with the applications of terracing (scenario 2), strip cropping (scenario 3), and contouring (scenario 4), the average sediment yields were found to be 8.6, 6.04, and 6.1 t/ha with corresponding reductions of the sediment from the baseline scenario by 20.44, 44.12, and 43.6%, respectively. Thus, the critical subwatersheds are suggested to be better managed using strip cropping with a strip width of 1 m on agricultural lands for the control of sediments. Strip cropping is one of the biological SWC measures. As the biological measures are generally more economical than the structural measures, the use of strip cropping is also advantageous from the viewpoint of cost.

The SWAT model has gained widespread acceptance in recent years as a tool for analyzing sediment production in river basins and watersheds. The model has been applied in numerous studies to evaluate sediment yield in various contexts. Modeling using SWAT is significant because it offers an in-depth understanding of the effects of water erosion on agricultural land and water resources. This study was conducted to simulate the sediment yield and develop BMPs in the Azuari watershed, Upper Blue Nile basin, Ethiopia using the SWAT model. The necessary hydrological and metrological data for model inputs were collected from MoWE. As the sediment data was in concentrated form, the sediment yield was derived from measured stream flow using a rating curve. The performance of the SWAT model for sediment simulation was checked by the R², NSE and RSR. The study area was delineated into 19 subwatersheds and the sediment yield was estimated in each subwatershed using the MUSLE in ArcSWAT. The model parameters for sediment were calibrated using sediment load data from 1988 to 2005 and were validated using data from 2006 to 2012. The model performed well for monthly sediment yield with R², NSE, and RSR values of 0.65, 0.64, and 0.6 during calibration and 0.58, 0.58, and 0.65 during validation, respectively. The sediment yield in the subwatersheds was found to vary from 0.72 to 36.56 t/ha/yr with an average sediment yield of the watershed of 10.81 t/ha/yr. The sediment yields in the 13 subwatersheds were found lower than the acceptable soil loss rate, whereas six subwatersheds which are coded as SW-1, SW-2, SW-4, SW-5, SW-6, and SW-7 were identified as critical or hotspot areas as they exhibited high to severe soil erosion. Four soil and conservation measures were evaluated in SWAT as BMPs to control sediment in the identified hotspot areas, namely filter strip, terracing, strip cropping, and contouring. The effect of applying these BMPs was checked against the baseline scenario. It was found that filter strip reduces sediment up to 35.61%, terracing up to 20.44%, strip cropping up to 44.12%, and contouring up to 43.6% from the baseline. Thus, the critical subwatersheds are advised to be managed using strip cropping for effective control of sediments in the area. Overall, using the SWAT model to examine sediment yield offers useful insights into how different land use and soil management methods affect the health of ecosystems and water quality. The findings of this study will also help to make well-informed decisions for developing the best watershed management strategies.

The authors would like to thank the Ministry of Water and Energy (MoWE) of Ethiopia for providing the necessary data for the research work and the Ministry of Education for the financial support.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.