Hydrological simulation and prediction of soil erosion using the SWAT model in a mountainous watershed: a case study of Murat River Basin, Turkey

Sedimentation is the main problem for dams and hydraulic structures which results in the reduction of the dam's usable storage volume in the determination of capacity and operation (Morris & Fan 1998). The sedimentation occurring in the reservoir affects the water quality both chemically and physically (Alemayehu et al. 2014). Chemically, sediments act as a distribution mechanism for particular pollutants degrading the water quality within a reservoir. Physically, sedimentation reduces the useful life of the reservoir (Juracek & Stiles 2010). According to Mahmood (1987), each year reservoirs around the world lose their storage volume by 0.5–1%. In Turkey, a lack of measurements for erosion control and land use results in the accumulation of sediments and reduces the capacity of the dam's economic life such as in Medik Dam, Akkaya Dam, and Cubuk-1 Dam (Koycegiz et al. 2021). Each year 154 million tons of sediment are transported by the rivers of Turkey and 84 million tons of this is carried only by the rivers in the Euphrates-Tigris basin (CEM 2018).

In most instances, the cause of sedimentation in reservoirs is soil erosion and occurs mainly on erodible soils, high terrain slopes, and due to intensive rainfall (Beskow et al. 2009). Soil erosion caused by water removes fertile soil layer from the surface, causing accumulation in rivers, lakes, and reservoirs, and can be caused to agricultural, environmental, and economic problems (Wu & Chen 2012; Chandra & Rabin 2019). Moreover, because of climate change, land use change, and human interference, soil erosion has become challenging issue (Zare et al. 2017; Chakrabortty & Chandra 2020; Pal et al. 2021). According to Panagos et al. (2021), mean soil loss rates caused by water erosion would rise by 13–22.5% in agricultural areas of European Union countries compared with 2016 baseline by 2050. However, given a specific location in the watershed, erosion depends on many factors such as regional precipitation patterns, soil texture, land use and management, and topography (Panagos et al. 2015; Roy et al. 2020). Consequently, case studies are encouraged to develop watershed management and soil and conservation studies (Sohoulande Djebou 2018), so efforts can be focused to preserve soil and water resources at the sub-basin scale to minimize the negative impact of soil erosion on the environment.

In order to estimate accurate and reliable streamflow, sediment carried from the land surface into the rivers, and identify the erosion-prone areas to suggest the best management practices in the watershed, hydrological models are used (Briak et al. 2016; Chakrabortty et al. 2022). These models are used effectively to estimate water quality, sediment yield, streamflow, the effects of climate, and land-use changes (Abouabdillah et al. 2014; Abbaspour et al. 2015; Ricci et al. 2018; Van Liew & Mittelstet 2018). The Soil and Water Assessment Tool (SWAT) developed by Arnold et al. (1998) is a physically based hydrological model that is used widely in literature for different purposes. Several researchers have been studied with SWAT for simulating runoff (Jakada & Chen 2020), sediment yield (Rodríguez-Blanco et al. 2016), water quality (Abbaspour et al. 2007; Qi et al. 2020), determine soil erosion areas (Vigiak et al. 2015; Dutta & Sen 2018; Abdelwahab et al. 2018; Daramola et al. 2019), and in different watersheds across the globe. Additionally, the model has been applied to assess the impact of climate change (Zhang et al. 2014; Tzoraki et al. 2015; Pal & Chakrabortty 2019; De Girolamo et al. 2022); land use, and land cover change effect assessment (Jodar-Abellan et al. 2019).

Soil is among the most important natural resources (Chakrabortty et al. 2020). Besides, land degradation due to soil loss has a negative impact on crop production and agriculture in many countries around the world (Pal & Shit 2017). According to a global study (Wuepper et al. 2020), Turkey is among the countries that are more severely affected by soil erosion. Turkey is a mountainous country with the mean elevation is nearly 1,250 m and 62.5% of total surface has more than 15% slope (Imamoglu & Dengiz 2017). Hilly topography and vegetation are among the two most important factors of soil erosion causing land degradation and nearly 12.7% of the total area of Turkey is exposed to severe and very severe soil erosion (CEM 2018). In the Euphrates-Tigris basin, which has Turkey's largest and highest water potential basin, occurs nearly a quarter of the total erosion of Turkey and half of the total sediment yield of its (CEM 2018). In this case, sustainable management and a convenient eco-friendly solution should take account for the Euphrates-Tigris basin for reducing the sedimentation in the reservoir and identifying the erosion-prone areas.

Several studies have been conducted by using the SWAT model around the world, but these studies are limited in Turkey (Guzel 2010; Akiner & Akkoyunlu 2012; Cuceloglu et al. 2017; Yıldırımer 2018; Peker & Sorman 2021) especially in the Euphrates-Tigris basin. The Murat River Basin is one of the headwater of Euphrates River and it is the source of several large dams in downstream. Agriculture and livestock are the main livelihoods of the region. Hence, the streamflow, sediment yield, and soil erosion model analysis can be a suitable tool in this basin for watershed management planners. To the best of our knowledge, there is no study in the Murat River Basin with spatial and temporal analysis of sediment yield and soil erosion performed. Hence, the novel contribution of this study will guide the hydrologist, water resource planners, and watershed management applications in the Murat River Basin. For these reasons, the SWAT model has been selected as a tool to simulate hydrological processes and sediment yield. Besides, the detailed objectives of this study are given as follows: (1) assessing the applicability of the SWAT model for the hydrological analysis in the Murat River Basin, (2) compare the model prediction of streamflow based on daily time interval, and (3) determine soil erosion-prone areas within the studied watershed where measures to reduce soil erosion are needed.

The spring months, March to May, is the rainy season of the watershed and occurred the peak flows in this period. The winter is generally snowy where snowmelt is the main source feeding the river and the maximum snowmelt also occurred in the spring months. The yearly average precipitation of the studied watershed is 774.5 mm. The maximum temperature was recorded in the studied period (2005–2013), in August 2006 as 38.1 °C and the minimum temperature in January 2012 as −33.7 °C.

ArcSWAT is a user-friendly software program which is an extension of ArcGIS-ArcView with a graphical user input interface for SWAT.

The land use/cover map was obtained from the Coordination of Information on the Environment Land Cover datasets (CORINE 2012) with a 100 m resolution. Since the study period was between 2005 and 2013, the year of land cover data was chosen as 2012 to better represent the calibration/validation period. The dominant land cover was observed as a pasture with a portion of 47.43% and agricultural land (26.48%), barren (22.91%), water (1.41%), forest (0.33%), residential (1.02%), and wetland (0.42%) as seen in Figure 2(b). The soil map was obtained from the Harmonized World Soil Database v1.2 (HWSD) developed by the Food and Agriculture Organization (FAO/UNESCO). The grid-based soil data obtained from the Digital Soil Map of World (DSMW) database was added into the model in the same projection as the DEM and land use/land cover data after adding the properties of the DSMW soils to the model database file. According to the FAO's soil classification system, Eutric Cambisols (Be115-2-3), Lithosols-Eutric Cambisols (I-Be-c), Lithosols-Calcaric Regosols-Calcic Xerosols (I-Rc-Xk-c), Haplic Kastanozems (Kh1-2ab), Calcic Kastanozems (Kk16-2b), and Haplic Xerosols (Xh31-3a) were defined in the study area (Figure 2(c)). The whole watershed was categorized into five slope classes using the ArcSWAT interface (Figure 2(d)). Then, all the maps, land use/cover, soil, and slope were overlaid to create the HRU's. Since the studied watershed is snow-dominated mountainous, ten elevation bands were added to increase the model accuracy. The purpose of this process is to enable the model to simulate snowfall and snowmelt separately for each elevation band by taking into account the changes in precipitation and temperature depending on the altitude in each sub-basin (Abbaspour et al. 2007). Previous studies also refer that adding elevation bands increases the model's accuracy for mountainous watersheds and there is no certain number of elevation bands recommended (Peker & Sorman 2021).

Due to the lack of available datasets (such as relative humidity, wind speed, and solar radiation) for the study area, globally available weather data were used for SWAT modeling. Climate Forecast System Reanalysis (CFSR) by the National Centers for Environmental Prediction was used for climate data. Twelve points were detected for obtaining climate data from CFSR in the studied area (Figure 2(a)). The observation data, suspended sediment, and discharge were obtained from Turkish State Hydraulics Works (DSI). But, the agency collects the suspended sediment sampling once or twice a month from the gages. Therefore, these data may not be sufficient for annual sediment estimates. The LOADEST estimator, version 2012 (Runkel et al. 2004) was used to convert suspended sediment data to monthly sediment load. LOADEST can be used to generate a regression model for predicting constituent loads in streams and rivers. The model estimates constituent loads using three statistical methods. In this study, the Adjusted Maximum Likelihood Estimation (AMLE) method was used. The sediment load outputs obtained from the LOADEST estimator used as an input for the SWAT model. After entering all input parameters into the interference, the SWAT model was set and run on a daily and monthly time period for the years between 2005 and 2013 which is one year selected as a warm-up period.

Several parameters can be used for calibration and validation procedures. However, some parameters are affecting the targeted variable more than the other parameters, defined as sensitive parameters (Panda et al. 2021). SWAT Calibration Uncertainty Procedure (SWAT-CUP) program was used to compare flow and sediment prediction in this study. SWAT-CUP program is capable of performing sensitivity analysis, calibration, validation, and uncertainty analysis of the SWAT model (Abbaspour 2015). Besides, to predict the parameters about streamflow and sediment yield, the Sequential Uncertainty Fitting Version 2 (SUFI-2) algorithm was selected (Abbaspour et al. 2004). Two statistical parameters as p-factor and r-factor were used in the SUFI-2 algorithm. p-factor (range 0–1) is the percentage of the actual data covered 95 PPU (95% prediction uncertainty). In other words, (1 − p-factor) can be defined as a model error (Abbaspour et al. 2017). The r-factor (range 0–∞) is the ratio of the average thickness of 95 PPU to the standard deviation of observed data. The results having higher p-factor and lower r-factor represent better model performance (Abbaspour et al. 2007). The reference p-factor and r-factor ranges to be used for sediment and streamflow according to the study of Abbaspour (2022) are given in Table 2.

In sensitivity analysis, p-value and t-stat statistical parameters are used to assess sensitive parameters in the SUFI-2 algorithm. While the p-value approaching zero and larger absolute t-stat values are considered more sensitive the parameters that entered to SWAT-CUP. In this study, the Latin Hypercube One Factor-At-a Time (LH-OAT) (van Griensven & Meixner 2006) sampling technique was used to perform global sensitivity analysis. The SWAT-CUP model was performed with 500 simulations with 25 parameters. Fifteen most sensitive parameters were selected for streamflow calibration and validation process and three hydrological parameters selected for after streamflow calibration for sediment calibration and validation (Table 3).

Calibration can be defined as adjusting model parameters to achieve the best simulation match with observation data. In this study, the parameters which are precipitation lapse rate (PLAPS) and temperature lapse rate (TLAPS) were fixed and removed from the calibration procedure before starting the calibration procedure with the sensitive parameters. Then, the same process was applied for the following snow parameters: snowfall temperature (SFTMP), snowmelt base temperature (SMTMP), maximum and minimum melt rate for snow during the year (SMFMX, SMFMN), and snowpack temperature lag factor (TIMP). Finally, the calibration and validation procedure was performed with 500 simulations with selected sensitive parameters.

All kinds of soil protection methods, such as terracing, grazing, and transverse structure, are considered as inputs to support practice factor (USLE_P). USLE_P value was assigned to be 1, since there was no protection method in the studied area. The value of soil erodibility factor, USLE_K, was set the value of 0.347 after the calibration. Channel cover factor (CH_COV) has an impact on the soil loss from channels (Dutta & Sen 2018) and it was set to 0.005.

In this study, the calibration and validation of the SWAT model were applied with the observed data obtained from two gauging stations (E21A074 and E21A077) on the Murat River basin. The calibration and validation period were determined between 2006–2010 and 2011–2013, respectively, while 2005 was selected as the warm-up period.

The monthly average discharge and sediment data were used based on monthly time interval in the model for selected two gauging stations. Besides, only daily discharge data was used for the daily-based model. 500 simulations were performed with Nash–Sutcliffe Efficiency (NSE) that was selected as an objective function until reach the satisfactory criteria. The model evaluation methods given by Moriasi et al. (2007), who defined standards for model evaluation and presented value ranges and corresponding performance ratings for suggested statistical parameters, are referred to in this study (Moriasi et al. 2007; Table 4). According to their study, the NSE and R² (coefficient of determination) value greater than 0.5 are the acceptable values for streamflow and sediment models.

After reaching the valid statistical criteria of streamflow estimation, sediment yield can be calibrated and validated because the runoff and streamflow is the main driven factor of sediment transportation (Arnold et al. 2012). After having a satisfactory calibration results, the validation processes were performed with the same calibrated parameters to simulate streamflow and sediment yield. The average and standard deviation of the streamflow and sediment yield data used in the calibration and validation period should not be much different from each other (Abbaspour et al. 2017). The statistical correlation obtained between calibration and validation data for two gaging stations that are used in this model given in Table 5.

Results from calibration and validation must meet some criteria to be acceptable (Table 4). Three statistical evaluation criteria including R², NSE (Nash & Sutcliffe 1970), and Percent Bias (PBIAS) (Gupta et al. 1999) were selected to evaluate the performance of the SWAT model based on the accuracy of predicted and observed streamflow and sediment yield.

Besides, for the station E21A074, the NSE was found 0.57 and 0.50, R² was 0.57 and 0.53, and PBIAS was −7.4 and 15.6% (Table 6). According to the evaluation criteria by Moriasi et al. (2007), the performance of the model is in the acceptable range for the studied watershed (Table 4). The same procedures were applied to the daily-based simulation. The model produced less effective performance values on the daily simulation than on the monthly simulation. Similar results were found in the literature as well (Yang et al. 2016; Duru et al. 2018). However, the results are still in the acceptable ranges with NSE: 0.56, 0.55, R²: 0.57, 0.56, and PBIAS: −15, 1.2% in calibration and validation periods, respectively, for the station E21A074 (Table 6). The detailed results of calibrated and validated streamflow for two stations are given in Table 6.

Runoff is one of the main factors that produce sediment yield. The maximum runoff was observed at the sub-watershed 139 with 46.11 mm, and the next sub-watershed 137 with 46.08 mm. The maximum average sediment yield was also obtained at the sub-watershed 76 with 42.5 mm runoff value. It is seen from Figure 8 that generally, the sediment yield value is high in the sub-watersheds where the runoff is high. The correlation coefficient between monthly average surface runoff and sediment yield was high (R² = 0.98) and statistically significant (P < 0.05). Similar results also indicated by Mosbahi & Benabdallah (2013) and Tibebe & Bewket (2011), that estimated sediment yield were generally higher where the runoff generation was the higher in the watershed. Figure 9(a) indicates that 61% of barren land was under very severe (48%) and severe (13%) soil erosion class, whereas 31% of forest area, 27% of pasture area, and 10% of agricultural area are under very severe soil erosion conditions. A total of 73% of agricultural area on the watershed was under the very slight (47%) and slight (26%) soil erosion class. The average soil erosion on the total area of barren land is 13.7 t/ha/y, while 5.8 t/ha/y in the agricultural area. These results show that the lowest soil erosion occurred in the agricultural area while the highest was seen in the barren area. The reason for that the cover factor has an influence on the soil erosion. Hence, in barren areas, there is no vegetation cover meanwhile in agricultural areas the vegetation provides a protection of soil.

The results were analyzed at the sub-watershed scale and found that especially the sub-watersheds 65, 66, 74, 75, 76, 137, and 139 produced more values for runoff, sediment yield, and soil erosion. These findings will guide decision-makers on the watershed management by focusing on these areas to take a precaution. With these measures to be taken, the sustainability of the basin will be ensured and less sediment transport will contribute to the economic life of the dams.

Figure 9(b) shows the ratio of soil erosion for the different soil types on the watershed. Loam soil texture covers 84% of the total watershed area and 18% of this area is in the category of very severe soil erosion. The rest of the watershed (16%) is a clay-loam texture with very severe (35%), severe (34%) soil erosion classes. The average sediment yield is higher in the Clay-Loam soil group (10.24 t/ha/y) than the Loam soil group (5.89 t/ha/y) since mostly Clay-Loam soils is located the downstream area where the mostly sediment yield occurred. Since the other factors such as rainfall erositivity, soil cover, land slope, and management practices have a major influence on soil erosion, it can not concluded that soil type is the most important driver in soil erosion. Similar results reported by Ricci et al. (2018). It is expected in Figure 9(c) that a higher value of slope produces more soil loss. 21% of the total area of the basin has a slope of greater than 25%, while 62% of its under the very severe (30%) and severe (32%) soil erosion class. In addition, the average sediment yield on a slope greater than 25% was found as 18.5 t/ha, and on the slope of 15–25% as 9.54 t/ha, and with the lowest value as 0.38 t/ha on the slope of 0–3%. It can be concluded that slopes have an influence on sediment yield similar to results reported the study by Yuan & Forshay (2019). With all these findings from the figures and the results of sediment yields, it can be assumed that soil erosion is quite dependent on land use and soil type.

This study tried to demonstrate the performance of the SWAT model for estimating streamflow, sediment yield, and identification of soil erosion-prone areas in the Murat River basin, which is a part of the Euphrates-Tigris basin, the largest basin in Turkey. First, the sensitivity parameters were selected by using the Latin Hypercube One-Factor-At-A-Time (LH-OAT) algorithm and found the CN2 as the most sensitive parameter for daily and monthly streamflow. Then, the model was calibrated and validated for both streamflow and sediment yield by using the SWAT-CUP program and compared with the data obtained from two gaging stations. Based on the statistical results, NSE, R², and PBIAS, the SWAT model was revealed to be a reliable tool to estimate streamflow and sediment yield in the Murat River Basin.

Terrain slope was found having a great influence on soil erosion. 63% of the area with a slope of greater than 25% are in the very severe erosion class. The hydrological components were analyzed based on the annual average values obtained from the SWAT model. Actual evapotranspiration (ET) generated most water loss from the watershed, accounting for around 64% of total water loss. The spatial distribution of soil erosion at the sub-watershed scale was also analyzed and the generated soil erosion map can clearly illustrate the prioritization of the basin's erosion-prone areas. The barren land is mostly observed in the area with a slope greater than 25% and consequently, sediment yield and soil erosion values are excessive. The average sediment yield was obtained as 13.72 t/ha/y on barren land use. Especially, the sub-watersheds 65, 66, 74, 75, 76, 137, and 139 produce a high amount of surface runoff, and sediment yield, so precautions should be focused on these areas. Indeed, sediment yield mostly observed in the downstream part of the basin. The findings of this study can help watershed managers to plan about where the best management techniques and conservation measures can be applied efficiency. Because of limited spatial studies about soil erosion and sediment yield in the Murat River Basin, this study will guide researchers who will work in this basin and nearby. Besides, the model could be also applied in future research to determine the impact of climate change on water resources, soil erosion, and land use.

This paper has been derived from the doctoral dissertation of Erkan Karakoyun.

Conceptualization, Methodology, Software, Writing first draft: E.K. Visualization, Review and Editing, Supervision: N.K.

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

The authors declare there is no conflict.