Introduction

Two important resources of the environment are water and land, and their quality degradation would influence on the quality of environment. Land degradation is caused by soil erosion, and the quality of surface water can be deteriorated by a poor management of land use (Singh et al. 2011). Significant soil erosion and land degradation have occured in Iran (Ahmadi et al. 2011). Best management practices can be planned by decision makers for the spatial distribution of watersheds via the identification of soil erosion hot spots. However, precise estimations of runoff and sediment yields are highly needed for any conservation strategies (Erfanian et al. 2015a; Saghafian et al. 2015). Predictions of surface runoff and sediment quantities and rates have been enabled through the development of a number of models. The models related to soil erosion include empirical and physically based models.

The empirical models include the Universal Soil Loss Equation (USLE), the Revised USLE (RUSLE), the Theoretical Modified Universal Soil Loss Equation (MUSLT), the Modified Universal Soil Loss Equation Erosion (MUSLE), Erosion Potential Method (EPM), the Pacific Southwest Interagency Committee (PSIAC) model, the Modified PSIAC (MPSIAC) model, and the United States Bureau of Land Management (US-BLM) model. The processes of runoff and erosion are usually described through hydrology/erosion models such as the EPIC: Erosion-Productivity Impact Calculator (Williams et al. 1983), the WEPP: Water Erosion Prediction Project (Ascough II et al. 1997; Baffaut et al. 1997; Flanagan et al. 1995, 2001), the ANSWERS: Areal Non-point Source Watershed Environment Response Simulation (Beasley et al. 1980), and the SWAT: Soil and Water Assessment Tool (Arnold et al. 1998).

The physically based models depend a great deal on the quality and quantity of available data as well as modeling effort. However, to the best of our knowledge, only empirical models have been employed in much of the previous work conducted in Iran since such a database is lacking. The applied models include the RUSLE (Erfanian et al. 2014, 2015b; Vaezi and Sadeghi 2011), the MUSLE (Amore et al. 2004; Khaledi Darvishan et al. 2010; Sadeghi et al. 2013, 2014), the MUSLT (Khaledi Darvishan et al. 2010; Sadeghi et al. 2013), the EPM (Tangestani 2006), the MPSIAC (Daneshvar and Bagherzadeh 2012), and BLM (Sadeghi 2005). Among the empirical models, the efficiency and accuracy of the MUSLE model is highly dependent on the accuracy of input data, scale (spatial and temporal), and user experiences (Sadeghi et al. 2014). Due to their application for a spatial and temporal distribution of runoff production, soil loss, and sediment yield (Sy), the physically based models are generally considered to be superior to the empirical ones (Maalim et al. 2013).

The Water Erosion Prediction Project (WEPP) is a physically based model which has frequently been applied in the world for soil erosion and runoff prediction on channel, hillslope, and watershed scales (Ascough II et al. 1997; Baffaut et al. 1997; Flanagan et al. 1995, 2001, 2007; Laflen et al. 1991; Liu et al. 1997). The WEPP is a complex and continuous simulation model that incorporates the fundamentals of soil hydrologic and erosion science (Flanagan et al. 2013). Thus, to quantify runoff and sediment yield mainly from agricultural and forested areas, as well as rangelands, this model has been employed to the fields and watersheds at different regions of the world (Laflen and Flanagan 2013; Maalim et al. 2013; Meghdadi 2013; Nearing 2004; Saghafian et al. 2015). The physically based WEPP has been successfully utilized in the USA and many other countries for the assessment of hillslopes and watersheds in terms of important natural resources (Flanagan et al. 2007). This model was applied in the Bautzen dam catchment in Germany, with monthly and daily single events for runoff and sediment yield (Al-Mukhtar et al. 2014). The model was calibrated for 2 years (2005–2007) and validated for another 2 years (2008–2009) against monthly measurements. The desirable performance of Water Erosion Prediction Project (WEPP) for runoff and sediment yield was shown by the results of calibration and validation periods (Al-Mukhtar et al. 2014). The models of WEPP and Soil and Water Assessment Tool (SWAT) were comparatively applied to Zhangjiachong Watershed in China (Shen et al. 2009). Due to the use of an empirical Modified Universal Soil Loss Equation (MUSLE) in the SWAT, its simulations were implemented at a lower level than those of the SWAT, which provided a reasonable confidence for soil loss in most cases though the results were found to be acceptable for both models (Shen et al. 2009). Also, SWAT application contributed to an investigation of the impacts of land use and climate changes on runoff and sediment load in the Loess Plateau of China (Zuo et al. 2016). The results were indicative of the greater effects of both changes on the sediment yield reduction compared to runoff decrease.

For the prioritization of the sub-watersheds of the forested watershed of Kasilian in northern Iran in terms of runoff and sediment production, the WEPP model was utilized in a recent study (Saghafian et al. 2015). The results demonstrated that the most critical zones prone to erosion were located at the northern parts of the watershed used for dry farming. Additionally, the highest sediment yield of the whole watershed was found to originate from the sub-watersheds near the main outlet. In another study conducted on this small watershed, the results were indicative of an approximate 32–34% decrease in the sediment yield of the whole watershed via the best management strategies like using soybeans, high levels of fertilization, and a No-Tillage (NT) system (Meghdadi 2013). Furthermore, the model was applied to simulate monthly sediment and runoff in the Emameh watershed, in northern Iran (Akbari et al. 2015). Its performance was reasonably good for sediment yield prediction and slightly inferior for runoff but still acceptable.

By connecting WEPP to ArcGIS software, runoff and sediment rates can be estimated via the development of a geo-spatial interface for WEPP called GeoWEPP tool (Flanagan et al. 2013; Maalim et al. 2013; Renschler 2003; Yu et al. 2009; Yüksel et al. 2008; Zhang et al. 2015). Furthermore, surface runoff, soil loss, and sediment can be estimated on the scales of watershed, hillslope, and channel using this tool (Flanagan et al. 1995). It is also valuable for evaluating the effectiveness of management practices, measures, and conservation on runoff production and soil loss (Akbari et al. 2015; Ebrahimpour et al. 2011; Flanagan et al. 2007; Maalim et al. 2013; Meghdadi 2013; Renschler 2003; Yu et al. 2009). Due to numerous potential changes to the input data, developing and/or organizing all the inputs (topography, soil, and land management practices) to the WEPP model would be more and more complex on the scale of watersheds (Renschler 2003).

In this study, the GeoWEPP was employed to estimate sediment and runoff production in the Lighvanchai watershed in northwest Iran by applying land use change scenarios. The watershed was selected considering problems of land degradation, poor land management, steep slopes, soil losses, and surface runoff. The objectives of this study were: (1) to evaluate the effects of land use change on annual average runoff and sedimentation using the GeoWEPP tool for the Lighvanchai watershed and 2) to select suitable management practices for reducing runoff and sediments in critical regions of the watershed in terms of sediment and runoff production.

Materials and methods

Study area

The Lighvanchai watershed is located between the two villages of Lighvan and Sefidekhan. The watershed area is about 71.86 km2, in the northwest region of Iran. It is located at 46°21′E to 46°28′E longitude and 37°43′N to 37°47′N latitude. The study area is near Tabriz city in a small portion of the Lake Urmia basin (Erfanian and Babaei 2016)(Fig. 1). The Lighvanchai is a river that flows over the northern faces of the Sahand Mountain. Receiving an average annual rainfall of 335.24 mm, the watershed shows an elevation range of 2190–3556 m at the outlet and headwaters, respectively; nearly 83% of the total annual precipitation occurs during the period from October to May with a snowy regime in the highlands. The average annual minimum and maximum temperatures are 7.6 and 18.6 °C, respectively. The watershed climate is cold semiarid based on the Köppen Climate Classification System according to meteorological data at the Lighvan meteorological station from 1982 to 2011. According to a field survey, steep slopes, rocky soils, rangelands with grass species occur in the highlands with a density less than 25% (poor density).This factor and poor land management are the dominant factors contributing to soil erosion and surface runoff in the area.

Fig. 1
figure 1

Geographic location of Lighvanchai watershed, northwest Iran, in the Lake Urmia basin

Full size image

WEPP watershed model

The hillslope model of WEPP is extended into WEPP watershed model. Its capability on a watershed size of larger than 100 km2 has been tested by several studies though it shows a limited maximum size for fields (e.g., Amore et al. 2004). The fundamentals of this model are based on soil physics, plant science, stochastic weather generation, hydrology, hydraulics, infiltration theory, and erosion mechanics. One of its greatest advantages is its capability for estimating daily, monthly, or average annual spatial and temporal distributions of soil loss (Flanagan et al. 2001). The components of the mentioned model include weather generation, irrigation, water balance, overland flow hydraulics, infiltration, erosion, deposition, soil disturbance by tillage, consolidation, plant growth, residue decomposition, frozen soils, and snow accumulation and melt (Flanagan et al. 1995, 2001).

To simulate runoff and sediment rates, the user can implement the WEPP model by using both watershed and channel methods. In the first method, the simulation process is carried out for all flow paths (channels) across sub-watersheds. In the second method, the simulation is carried out on a watershed scale. The equation of Green–Ampt–Mein–Larson (GAML) infiltration is used to calculate surface runoff in the hillslope version of WEPP (Ascough II et al. 1997). The effective hydraulic conductivity: Ke (mm h−1) is an important factor for predicting infiltration rate (mm h−1). Rill flow and inter-rill mechanisms are usually derived from hillslope erosion in the soil erosion models. The WEPP model is similar to soil erosion models, which include inter-rill erodibility: Ki (kg m−2 s−1) and rill erodibility: Kr (s m−1). The rill erodibility and critical shear stress (τc) are essential factors for predicting flow detachment capacity. The WEPP model uses specific equation to calculate each factor by relating its quantity to soil properties (sand, clay, organic matter, etc.). The detailed model formulation, basic equations, and theoretical background of WEPP can be found in the literature (e.g., Flanagan et al. 1995; Maalim et al. 2013). Because of spatial variability of these factors at watershed scale, their optimal values are critical for proper estimation of runoff and soil erosion. Moreover, before calibrating and validating WEPP model, the most sensitive factors (Ki, Kr, Ke, and τc) must be identified through a sensitivity analysis (Al-Mukhtar et al. 2014; Meghdadi 2013). In the following sections, the sensitivity analysis of the WEPP model will be explained in more details.

To derive slope inputs for each channel and sub-watershed hillslope profiles in this study, we employed the Topographic Parameterization (TOPAZ) tool incorporated into the GeoWEPP package for ArcGIS 9.3 (Flanagan et al. 2013; Renschler 2003), which can be downloaded for free at: http://www.geog.buffalo.edu. To create a channel network, outline a watershed and its sub-watersheds, and generate the flow paths as the inputs of individual slope profiles, TOPAZ makes use of the digital elevation model (DEM) for a desired region. Moreover, to control the area size of a sub-watershed, it applies the critical source area (CSA) and minimum source channel length (MSCL). However, DEM resolution acts as a determining factor to estimate the CSA and MSCL values (Zhang et al. 2006). After importing the DEM to the TOPAZ, channels, and sub-watershed delineation, the user needs to provide other base maps in txt format (e.g., soil, land use), and to set up soil and land management files for initial model runs. A total of 11,707 flow paths, 64 channels, and 156 sub-catchments were extracted for the Lighvanchai watershed. The output text files generated by the simulation program consist of detailed information on 156 representative hillslopes (out of 662) and 64 channels (out of 664), respectively.

WEPP inputs

Climate data

In this study, daily records of precipitation, maximum, and minimum temperatures at the Lighvanchai weather station were obtained from the East Azarbaijan Regional Water Authority (EA-RWA) over a 30-year period (1982-2011). During this period, the highest and lowest observed air temperatures were +46.1 and −21.6 °C, respectively. Using the monthly mean, standard deviation, skewness, etc., derived from earlier measurements of climatic variables, the WEPP model can provide daily estimates for solar radiation, temperature, wind speed, precipitation, and dew point at a single geographic point via a stochastic Climate Generator (CLIGEN) (Meyer 2011). The statistical weather data provided monthly as an input in a special format are required for CLIGEN. To generate a climate file in WEPP, the monthly climate averages were transformed into a PAR format. Table 1 shows the monthly average for some climate variables for the period 1982–2011. It includes minimum and maximum air temperature (°C), mean air temperature (°C), maximum rainfall over a 30-min period (mm), mean precipitation (mm), number of wet days, the probability of a wet day after a wet day (Pw/w), the probability of a wet day after a dry day (Pd/w), and solar radiation (Langley/day). The monthly average values for solar radiation were calculated in MS Excel using the Hargreaves–Samani method proposed (Erfanian and Babaei 2013; Tabari et al. 2016) due to the lack of measured radiation data.

Table 1 Climatological data from the Lighvan weather station over 30-year period (1982–2011)
Full size table

Hydrology and sediment data

Sediment loads were measured by standard sediment samples that were collected from the watershed outlet at Lighvan station (water and sediment gauge station) by the EA-RWA at irregular intervals. The dry weights of sediment samples were measured. The river flow measurements at this station are based on water level readings approach. All recorded daily data of flow discharges and suspended sediment concentrations (SSCs) were collected from the EA-RWA for the time period of 1982–2011, and data quality control was carried out. Using a simple program (Arnold and Allen 1999) available at http://swat.tamu.edu/software, the base flow discharge was generated from the total stream flow. Using the watershed drainage area, the suspended load of the sediment (mg/L) was converted into the suspended sediment (ton/ha) as well. Due to the lack of an automatic recording rain gauge, we used daily meteorological data measured at Lighvan weather station near the watershed outlet. The volumes of the direct runoff discharge were divided by watershed area to calculate the surface runoff depth. To estimate the suspended sediment concentration (SSC) for subsequent flow calculations, the rating curves of sediments are usually applied in the absence of an actual SSC measurement (Horowitz 2003; Isik 2013). Therefore, we developed the best sediment rating curve (not presented in this paper) to estimate annual average of suspended sediment load (6512.9 ton/yr) for the time period of 1982–2011. In addition, the average suspended sediment load of 4996 ton/yr at Lighvan station was reported in a previous study for the period 1961–1994 (Arabkhedri et al. 2004). Figure 2 shows monthly precipitation values (mm) which mostly takes place from March to May, whereas the maximum amounts of stream flow (m3/s) are observed during April–August (Fig. 3). The annual average of daily flow discharge and annual average of stream flow volume were about 0.8 m3/s and 25.42 Mm3, respectively.

Fig. 2
figure 2

Box-plot of monthly mean precipitation at Lighvan weather station (1988–2011)

Full size image
Fig. 3
figure 3

Box-plot of monthly mean stream flow at Lighvan hydrometric station (1988–2011)

Full size image

Topographic and soil map

1:25,000 scale of a digital topographic map (Fig. 4a) provided by the National Cartographic Center (http://www.ncc.org.ir) was used to build a digital elevation model (DEM) with a spatial resolution of 30 m (Fig. 4b). The raster map was converted into ASCII format which is utilized by the TOPAZ to delineate the watershed and sub-watersheds, and to generate channels and flow paths for individual profile slope inputs. To model soil erosion, a soil map provides a major input. Using the randomly collected samples of 77 points across the sub-watersheds, some of soil properties were measured in this study. The soil input file includes soil texture and depth, sand, silt, clay, organic matter ratio, cation exchange capacity saturation level, and albedo (Akbari et al. 2015). The WEPP model enables us to calculate the values of rill and inter-rill erodibilities, critical shear, and effective hydraulic conductivity. We used a soil type map in a scale of 1:250,000 provided by the National Soil and Water Research Institute (http://www.swri.ir), in which the soil depth varies from shallow to deep with a mostly coarse texture. As shown in Table 2, soil texture type is sandy loam for soil types of C-5, C-31, E-42, and E-45 (more than 90% of area). The map shown in Fig. 4c shows five soil types with different slope classes: C-5: calcaric regosols and lithic leptosols with a sandy loam texture (slope 40–100%); C-31: calcaric regosols (slope 15–40%); E-42: calcaric regosols with variable amounts of clay (slope 15–40%); E-45: calcaric regosols (slope 40–100%), and C-58 is the calcaric soil with a loamy sand texture (slope 15–40%). First, by referring to a soil raster map in ASCII format, the map layer of the soil was generated to run GeoWEPP. Then, based on the relevant land cover types, a text file of the soil was developed. To determine the description corresponding with the soil map layer in GeoWEPP, this file is utilized by GeoWEPP and WEPP/TOPAZ translator. According to WEPP user’s guide, special USGS codes were assigned in the text file to represent the values of the raster cell in the map layer of the soil (Zuo et al. 2016).

Fig. 4
figure 4

Lighvanchai watershed: topography (a), digital elevation model (b), soil type map (c), land use (d)

Full size image
Table 2 Soil characteristics based on sampling across the study area required for initial model run
Full size table

Land use map

To prepare a land use/land cover (LULC) map, we used a satellite image from Landsat 8 (OLI bands), available for free download at: https://earthdata.nasa.gov under a clear sky condition during the plant growing season in the study area, acquired on July 10, 2013 (Pass No. 168 and Row No. 34). The authors performed the common standard geo-referencing; geometric and radiometric corrections; a supervised image classification using the maximum likelihood method as well as post-processing techniques in the ENVI 5.1 software (Congalton 1991). The accuracy assessment was carried out using 198 ground control points (GCPs) as reference data acquired by a GPS device during field survey. The user accuracy (%), producer accuracy (%), overall accuracy (%), and kappa statistics (Ka) for the classified image were calculated from a confusion matrix obtained (Congalton 1991). Finally, the LULC map (Fig. 4d) was created with an overall accuracy of 91.41% and kappa statistics (Ka) of 0.88. The description of land use types in the Lighvanchai watershed is presented in Table 3. The land use map was classified according to the USGS classification and resulted in a residential area (0.6%), agricultural lands (1.33%), fallow lands (4.03%), rangelands (37.51%), bare soil (51.12%), and mixed forests and trees (5.41%). The raster maps of the Lighvanchai watershed shown in Fig. 4 have a 30 m spatial resolution based on the UTM projection coordinate system (Zone 38 N).

Table 3 Land use types in the Lighvanchai watershed based on Landsat 8 image classification
Full size table

In the present study, the required data of management file were entered into WEPP based on field measurements and a study of the previous reports. Using a management file, the layer of land cover was first generated in ASCII format to run GeoWEPP. Then, referring to the related types of land use, the text file of land cover was defined. For determining the description corresponding to the layer of land cover in GeoWEPP, GeoWEPP, and WEPP/TOPAZ translator apply this text file. According to WEPP user’s guide for the text file, special USGS codes were assigned to represent the values of the raster cell in the layer of land cover (Zuo et al. 2016).

Model evaluation

Calibration, sensitivity analysis, and validation are the procedures used for the model evaluation. To reduce uncertainty when simulating a model, WEPP calibration should be performed against the measured data in the same way as any other hydrologic models (Saghafian et al. 2015). Sensitivity analysis can be associated with the phases of formulation, calibration, and validation of a model. In this study, the model sensitivity analysis was conducted based on the quantity of four factors consist of the rill erodibility (Kr), the inter-rill erodibility (Ki), the effective hydraulic conductivity (Ke), and the critical shear stress (τc). The initial range of each factor was selected by referencing to its pre-defined values in the literature and using a priori information obtained from the soil measurements in the watershed (Akbari et al. 2015; Al-Mukhtar et al. 2014; Meghdadi 2013; Nearing 2004; Saghafian et al. 2015; Zhang et al. 2006; Zuo et al. 2016) and were adjusted by several simulations until a minimum value for the objective function (RMSE) was obtained.

To identify the most sensitive factors (Kr, Ki, Ke, and τc), they were separately analyzed in terms of sensitivity. WEPP model’s sensitivity to the quantity of each factor was specified based on runoff and sediment productions. In this study, the value of each factor was changed within a range of ±50% of the pre-defined values in the literature and model documentation, while keeping others constant. Subsequently, the model sensitivity was classified as being non-sensitive, or having a low sensitivity or a high sensitivity for surface runoff and/or sediment yield prediction with respect to the degree of its sensitivity to reasonable changes in each factor value, defined by the user in a trial and error approach. A relative sensitivity index can be used to define the rate of a model’s sensitivity through the quantification of the output variation impact over different values of a model parameter (Baffaut et al. 1997).

The available monthly values of total sediment loads and surface runoff depths measured for model calibration (2005–2007) and validation (2008–2010) in recent years were exclusively utilized due to the lack of hourly records of rainfall and regular daily loads of sediment at the watershed outlet. We attempted to evaluate the model during three representative wet, dry, and average precipitation years for both calibration and validation steps. The selection of the three representative years is useful for better calibration which would result in erosion estimates that imitate long-term average trends (Nearing 2004). The model calibration was carried out based on factors which were sensitive for proper estimation of runoff and sediment. If the model was not sensitive to a specific factor, then a pre-defined value for that factor is used. The calibration was manually performed based on a trial and error method. As previously mentioned, after adjusting each factor, the model was run frequently to simulate monthly surface runoff and sediment loads until a minimum discrepancy between the observed and simulated values was obtained.

The statistical criteria of the coefficient of determination (R 2), Nash–Sutcliffe efficiency (NSE), mean bias error (MBE), and root-mean-square error (RMSE) were applied to quantitatively evaluate the WEPP model. The formulation and description of the criteria for watershed model evaluation is presented in Table 4. The NSE value can range between −∞ and +1, while +1 indicates a perfect accuracy for a model simulation. A value >0.60 and >0.50 for R 2 and NSE, respectively, was considered acceptable (Moriasi et al. 2007). The average deviation of the predicted values from their corresponding measured data is represented by MBE measure, which provides some information on a model’s long-term performance in a way that a lower MBE demonstrates a better long-term prediction. A model’s underestimation or overestimation degree is represented by a positive or negative MBE value. The closer to zero RMSE is, the more accurate a model prediction will be (Moriasi et al. 2007).

Table 4 Evaluation criteria for accuracy quantification of the GeoWEPP for simulating surface runoff and sediment yield
Full size table

Long-term continuous simulation

To this point, we presented the model description and input data for model application. The WEPP model similar to other process-based erosion models can act as a very valuable tool for a variety of purposes. For instance, it can be used for choosing proper conservation practices, tracking erosion changes over time, regulating land activities aimed at conservation compliance, designing storm responses, and estimating long-term sediment loadings to streams with fair and reasonably accurate predictions (Nearing 2004). Therefore, despite the problems and the limitations of natural variability in erosion and long simulation in order to obtain stable erosion values, we employed the calibrated WEPP model in the Lighvanchai watershed using the current land use and climate data for a 30-year period from 1982 to 2011.

The reports by the GeoWEPP include different types of information, records, and the tables which report total sediment yield, sediment delivery ratio (SDR) and runoff volumes for the hillslopes, sub-watersheds, and channels delineated. The txt reports and output maps of sediment and runoff production were obtained from the current land use map (scenario 1 with assumption of no-change in past and future). To reduce the annual average surface runoff and sediment yield at the watershed scale, two management scenarios of land use changes were defined by identifying the critical hillslopes. To assess the impacts of land use changes based on the reductions of sediment and runoff productions, the process of simulation was repeated for each management scenario. The best management practices in the Lighvanchai watershed were determined through the analysis of model outputs and a field survey.

Results and discussion

Sensitivity analysis

Table 5 shows the results of the sensitivity analysis for 2005–2007 as the 3 representative years of precipitation. As can be seen, both surface runoff and sediment yield are highly sensitive to the variations of the effective hydraulic conductivity (Ke). Runoff was found to have a low sensitivity to the rill erodibility (Kr), the critical shear stress (τc), and the inter-rill erodibility (Ki) values, whereas the sediment load predicted by the GeoWEPP is highly sensitive to these factors. Since the dominant soil texture in the study area is a sandy loam, the calibrated values of the Ke, Ki, Kr, and τc were obtained as 0.18 (mm h−1), 5.1E + 06 (kg.s.m−4), 0.06 (s m−1), and 3.1 (Pa), respectively. The relative sensitivity ratio was zero for the Ki, Kr, and τc to predict surface runoff (mm), but the values for Ke were −0.135 and −0.32 for the runoff and sediment yields, respectively. The sensitivity ratios for Ki, Kr, and τc to simulate sediment yield were 0.004, 0.061, and −0.12, respectively. An increase in the value of Kr was found to enhance the predicted sediment yield and vice versa. Also, the critical shear stress had a significant impact on the predicted sediment load. The results of sensitivity analysis were in an agreement with previous studies (Al-Mukhtar et al. 2014; Baffaut et al. 1997; Meghdadi 2013; Saghafian et al. 2015). The calibrated Ki, Kr, and τc values were consistent with the range of values reported in some previous studies (Akbari et al. 2015; Al-Mukhtar et al. 2014; Saghafian et al. 2015).

Table 5 Sensitivity classifications of soil related factors for different model runs during 2005–2007
Full size table

Performance of the GeoWEPP model

The simulated monthly sediment yield and runoff during the calibration period (2005–2007) was compared to the observed values for a few months (due to a data availability problem) in the watershed outlet as shown in Table 6. The model simulation of monthly surface runoff (mm) was satisfactory with a determination coefficient of 0.86, Nash–Sutcliffe efficiency coefficient of 0.63, RMSE of 19.16 mm, and MBE of −1.65 mm. Moreover, the model simulated well the monthly sediment yield with a determination coefficient of 0.93, Nash–Sutcliffe efficiency coefficient of 0.59, RMSE of 0.06 (ton/ha), and MBE of −0.01 (ton/ha). Furthermore, the monthly sediment yield and runoff were successfully predicted for the validation period (2008–2010). Monthly runoff simulation was performed with a determination coefficient of 0.89, Nash–Sutcliffe efficiency coefficient of 0.61, RMSE of 1.96 (mm), and MBE of −0.72 (mm). The criteria for the monthly sediment yield were 0.91, 0.68, 2.73, and −0.02, respectively. Both monthly surface runoff and sediment yield were found to be under-predicted by the GeoWEPP model as shown through the MBE results. However, the NSE > 0.60 and R 2 > 0.80 shows a satisfactory agreement between observations and the simulation (Moriasi et al. 2007). Some discrepancies observed for surface runoff are most likely associated with the snowmelt runoff during the spring and early summer and originating from the highland regions in the Lighvanchai watershed. The under- and over-predictions of 13 and 27% of monthly runoff and sediment yield were, respectively, resulted from the GeoWEPP model in a similar study conducted in Orcan Creek Watershed in Kahramanmaras, Turkey. Moreover, the R 2 values of 0.93 and 0.94 were, respectively, revealed by the statistical comparison of the predicted runoff and sediment yield against the observed data. The RMSE of 8.43 and 2.96 were calculated for the runoff and sediment yield prediction, respectively. In addition, more and slightly less accurate sediment yield and runoff predictions were obtained as the monthly predictions in Orcan Creek watershed, respectively (Yüksel et al. 2008). In addition, the WEPP performance in the Emameh watershed in northern Iran was reasonably high for sediment yield prediction and slightly inferior for runoff. However, the results were still acceptable and some discrepancies for surface runoff were assumed to be associated with the numerous upstream of the Emameh watershed that drains into the main river (Akbari et al. 2015). The WEPP model in the Bautzen dam catchment in Germany was calibrated for 2 years (2005–2007) and validated for another 2 years (2008–2009) against monthly measurements. Nevertheless, WEPP showed a satisfactory performance for sediment yield and runoff during the calibration and validation periods (Al-Mukhtar et al. 2014).

Table 6 Model evaluation results for calibration and validation periods based on monthly data
Full size table

The sediment yields simulated and observed for the calibration period about the 1:1 line are illustrated in Fig. 5. Except for the over-prediction (points above the line) or under-prediction (points below the line) for some months, the simulated sediment yields can be seen to be closely distributed about the ideal line. The monthly observed and simulated runoffs were graphically compared as shown in Fig. 6. For higher value for monthly observed runoff, the simulated value was below the 1:1 line, indicating that WEPP under-predicted the surface runoff. Observed and simulated sediment yield and the surface runoff values for the calibration period (2008–2010) plotted along the 1:1 lines (Fig. 7). The simulated sediment yield (Sy) can be seen to be under-predicted. Additionally, the simulated runoff values can be observed to be distributed about the 1:1 line as exhibited by Fig. 8. In general, WEPP performed satisfactory with NSE > 0.60 and R 2 > 0.80 to predict sediment yield and surface runoff, respectively.

Fig. 5
figure 5

Comparison between observed and simulated monthly sediment yields (2005–2007)

Full size image
Fig. 6
figure 6

Comparison between observed and simulated monthly surface runoff (2005–2007)

Full size image
Fig. 7
figure 7

Comparison between observed and simulated monthly sediment yields (2008–2010)

Full size image
Fig. 8
figure 8

Comparison between observed and simulated monthly surface runoff (2008–2010)

Full size image

The differences between validation and calibration results were expected since the precipitation and stream flow conditions were not directly comparable. Despite the unavoidable uncertainties with the WEPP model, the model evaluation results imply that the WEPP model works relatively well and it can be of use in the Lighvanchai watershed for a long-term continuous simulation to identify critical areas with respect to surface runoff and soil erosion production, for best management practices. The following section presents the results for the WEPP model for a long-term simulation (1982–2011) under current land use and management.

Simulated effects of current land uses on soil erosion

The land use, soil properties, and climate data were compiled into relevant input files, and the WEPP model was implemented for the 30-year period (1982–2011). As mentioned earlier, the WEPP model uses two methods for simulation (watershed and channel modes) and provides raster maps and textual reports for the representative hillslopes and channels. The report txt files include runoff volume, soil loss, sediment yield, and average annual delivery quantities from the channel outlet (total hillslope or channel soil loss, and sediment delivery ratio (SDR), and total sediment). Table 7 shows the results from a long-term continuous simulation based on current land use condition (Scenario 1) during the period 1982–2011 in which the annual average of sediment yield at the watershed outlet was predicted to be about 4340 ton/year and the sediment delivery ratio (SDR) to be 0.156. As previously mentioned, the annual average of suspended sediment load predicted by a previously developed rating curve (not presented in this paper) at Lighvan station was estimated to be about 6512 ton/year for the period 1982–2011. In a similar work on developing the rating curve models at this station, the average annual suspended sediment load was estimated as 4996 ton/year for the time period 1961–1994 (Arabkhedri et al. 2004). The average annual runoff volumes passing through the watershed outlet for a number of storm events (simulated by the WEPP from daily rainfall data) with surface runoff and sediment loads was about 93,584 m3/year and the average annual sediment delivery per unit area (specific sediment yield) was 0.6 ton/ha/year.

Table 7 Model simulation results for the period 1982–2011 under current land use
Full size table

The effect of land use change scenarios

The critical hillslopes in terms of soil losses and runoff were selected by analyzing the output maps and report files of the WEPP model (Fig. 9). The simulation results are based on the current land cover and management. In the next step, the best management practices as proposed management scenarios (land use changes) were applied to susceptible hot spot zones throughout the watershed only. Given that the dominant crops grown in the Lighvan and Sefidekhan villages are wheat, barley, potatoes, cucumbers, and forage crops (alfalfa and trifolium), considering the dominant soil type (sandy loam and loamy sand), climate conditions, and consultation with local experts and people in the watershed (farmers and ranchers), we proposed two land management scenarios: (a) alfalfa cultivation with suitable soil tillage (scenario 2) and (b) grassland development (with vegetation density more than 50%) for grazing and soil conservation (scenario 3). The corresponding values for each scenario were defined for the susceptible hillslopes in the land management file of the GeoWEPP. For other regions, the model pre-defined values of soil and land use related factors were applied. Accordingly, the simulation was carried out separately for each scenario during the period 1982–2011.

Fig. 9
figure 9

Location of selected critical hillslopes for management scenarios

Full size image

As can be seen in Table 8, the second and third scenarios demonstrate a reduced sediment delivery ratio (SDR) of the current land use from 0.156 to 0.038 and 0.036 for a long-term simulation at the watershed scale, respectively. The rate of annual runoff volume (m3/year) and total sediment (ton/ha) in the second scenario were 42,096 and 429, respectively. For the third scenario, the model outputs from the watershed outlet were predicted to be 30,239 and 226, respectively. Compared to the first scenario, the reduction rates in the annual average sediment load (ton/year) were about 90 and 94%, respectively. Moreover, for the second and third management scenarios, the reduction rates in the annual average runoff production were about 55 and 67%, respectively. Hence, applying the third scenario could significantly reduce the annual values for sediment and runoff production from the Lighvanchai watershed. The spatial distribution of soil loss under the third management scenario was compared to the first scenario (without changing the current use) by the watershed method (Fig. 10) and channel method (Fig. 11). As shown in Fig. 10, the soil losses for the most of sub-watersheds are reduced to below 2 ton/ha/year in the study area in the watershed method run and the remaining critical regions with soil loss above 2 ton/ha/year are mostly related to highlands areas or steep lands with a low vegetation density. However, the results of reducing the rates of annual runoff volume and sediment yield from the watershed outlet support the importance of critical hot spot identification to select the best planning for soil and water conservation in the study area. The decreased values of surface runoff and sediment yield through land management, land use change scenarios, and biological or structural practices like using check dams and soil tillage have been also reported in the previous studies (Akbari et al. 2015; Al-Mukhtar et al. 2014; Maalim et al. 2013; Meghdadi 2013; Saghafian et al. 2015).

Table 8 Comparison of significant simulation results under different land use (management scenarios)
Full size table
Fig. 10
figure 10

Comparison of soil loss (watershed method) for current land use (left) and the third management scenarios (right)

Full size image
Fig. 11
figure 11

Comparison of soil loss (channel method) for current land use (left) and the third management scenarios (right)

Full size image

Conclusions

The results of long term simulations (1982–2011) by the WEPP model for current land use and two land use change scenarios in the Lighvanchai watershed can be summarized as follows:

(1) Runoff production: The annual runoff volume (with sediment yield) from watershed outlet in the first scenario (current land use) was predicted to be 93,584 m3/year. In the third scenario (grassland development), the annual runoff volume is reduced to 30,239 m3/year (55% reduction), whereas in the second scenario (alfalfa with appropriate tillage), it is reduced to 42,096 m3/year (67% reduction), which shows the effects of this plowing and tillage system in reducing the annual runoff volume.

(2) Soil losses and sediment yield: The watershed sediment yield under current land use was predicted to be about 4340.4 ton/yr, while for the second and third scenarios it was simulated as 429 ton/yr (90% reduction) and 226 ton/yr (94% reduction), respectively. The lowest value of the Sy (ton/ha) was simulated for the third scenario (grassland without soil tillage). It is worth noting that under the third scenario, the reduction rate of 94% in Sy confirms the importance of selecting critical representative hillslopes for targeting management actions in a watershed. Therefore, alfalfa cultivation and grassland development could significantly impact on the annual runoff and sediment response. The analysis of maps and reports of the WEPP model for the watershed showed that tree-covered slopes (trees and shrub) have the lowest rate of soil loss, while the highest rate of runoff production occurs in the residential areas and other lands with impervious surfaces particulary especially in areas with steep slopes.

This study emphasizes the applicability of the GeoWEPP tool for evaluating the impact of land use changes on annual surface runoff and sediment loads from an entire watershed. However, it is recommended that the performance of the model is assessed at hillslope and plot scales by the direct measurement of rainfall, runoff, and sediment loads.