Improving MUSLE performance for sediment yield prediction at micro-watershed level using seasonal classified data

Sediment yields from watersheds limit the sustainable use of land resources and is considered as one of the most critical environmental hazards. Sediment yield prediction is important as it not only affects river water quality, reservoir capacity, aquatic habitat, and channel morphology, but is also a good indicator of watershed health and the effectiveness of watershed management conditions (Sadeghi et al. 2007; Noor et al. 2013).

In the absence of actual soil erosion and sediment yield data, hydrologists apply modeling techniques to predict sediment yield at the watershed scale. In this context, the Universal Soil Loss Equation (USLE) and its revised versions (RUSLE), as well as its modified version (MUSLE) have often been applied as the most common soil loss and sediment yield estimators because of their dependence on easily available input data. A large number of the existing erosion and sediment yield models are based on the USLE (Das 2000; Noor et al. 2016; Ben Khelifa et al. 2017). The USLE was basically developed for estimation of soil erosion at plot scale and annual time scale in agricultural areas with a gentle slope. Therefore, its application to storm-wise soil loss at the watershed scale may lead to large errors (Kinnell 2005; Sadeghi et al. 2007; Noor et al. 2010). USLE does not directly consider runoff for sediment yield prediction at the watershed scale (ASCE 1970; Williams 1975). Williams (1975) introduced an improved erosivity factor in terms of storm-wise runoff volume and peak discharge. In other words, this MUSLE erosivity factor takes into account the runoff shear stress effect on soil detachment during individual storm events. Williams (1975) showed that storm-wise prediction of sediment yield at the watershed scale could be simplified using the USLE with its rainfall factor (R) replaced by a runoff index. In this way, it is possible to improve the sediment yield, eliminate the need for delivery ratios, and apply the equation to a single storm (Pongsai et al. 2010). Finally, it can be stated that sediment yield estimation in MUSLE is improved because runoff is a function of antecedent moisture conditions and rainfall energy (Williams 1975; Williams & Berndt 1977; Kinnell 2005; Merritt et al. 2003). Presently, the MUSLE model is preferably applied to the prediction of storm-wise sediment yield in developing countries such as Iran. The major reasons for applying this model are a lack of adequate data and the necessity of having access to accurate sediment yield estimates for running watershed management projects (Sadeghi et al. 2007).

Rangelands are the largest ecosystems on Iran. Covering about 52% of the country, most of these rangelands are located in the arid and semi-arid regions. Unfortunately, rangelands have undergone rapid changes due to factors such as overgrazing, which lead to higher soil and water losses. Therefore, semi-arid rangelands in Iran, due to fragile ecosystem conditions and mismanagement, have a high potential for soil erosion and sediment yield. As a result, evaluation of the soil erosion model, especially MUSLE, at the micro-watershed level in this regard is a prerequisite for soil and water conservation projects in critical areas. Therefore, researchers and managers need to evaluate and calibrate the MUSLE at this fine spatial scale.

In addition to this, soil erosion and sediment yield have seasonal variations. Temporal changes or variability in the watershed conditions may also affect the performance of the model, especially for watersheds with a notable seasonal difference in rainfall (semi-arid rangeland). The issue results mainly from the temporal variations in the variables of the model that exists in the watersheds, but are either not considered or not effectively accounted for. In this case, the model may only perform well in the season when sediment yield is at its maximum.

Based on literature, the MUSLE has been applied at different spatial scales i.e. plots and watersheds (Nicks et al. 1994; Banasik & Walling 1996; Erskine et al. 2002; Kandrika & Venkataratnam 2005; Sadeghi et al. 2007; Pandey et al. 2009; Zhang et al. 2009; Noor et al. 2010; Pongsai et al. 2010; Wang et al. 2010; Ben Khelifa et al. 2017). A short review at the literature verifies that the MUSLE model has not been calibrated based on seasonal classified data or compared with the traditional method. Thus, the present study was conducted to: 1) assess the applicability of the MUSLE in Iran's Sanganeh micro-watershed as a representative semi-arid rangeland in the north east of Iran; and 2) perform a comparative evaluation of two calibration approaches, namely ‘seasonal classification of data’ and ‘traditional approach or no classification of data’.

Considering the importance of soil erosion and the study of sediment processes in semi-arid rangeland ecosystems, the Khorasan Razavi Agricultural and Natural Resources Research Center (KANRRC) assessed some micro-watersheds for the collection of storm-wise runoff and associated sediment. The Sanganeh research micro-watershed is located 100 km from Mashhad City (north east Iran) is one of the watersheds selected for this study. The watershed area, the longest waterways and mean slope of the watershed are 1.2 ha, 145.0 m, and 31.2%, respectively. The study watershed consists of a semi-arid rangeland dominated by Bromus tectorum and Artemisia diffusa, with coverage of 50% (Felegari et al. 2014). The soil is Entisol and Aridosol, young, with a maximum depth of 30 cm. The mean electrical soil conductivity (EC), soil organic matter (OM), clay, sand, silt, and surface rock fragments of soils are 1.81, 1.57, 10.6, 54.7, 34.7, and 5%, respectively. The general features of the study area is shown in Figure 1.

A runoff collection system, which consisted of 3 connected tanks (4.5 m³ total volumes) to capture the maximum runoff volume, was installed at the outlet of the micro-watershed (Figure 1).

In this study, water flow and sediment yield were monitored at the main outlet of the micro-watershed from December 2006 to April 2016. The runoff volume was calculated after each storm event through multiplying the depth of collected water, measured using an iron ruler at five points in the tank (corners and central), by the surface area of the collector. The collected runoff and sediment was then mixed thoroughly and one sample was taken to determine sediment concentration and sediment yield.

The erosivity factor was computed for each individual rainfall event as a reduced form of the volume. Peak rates of runoff were also monitored at the mean outlet of the watershed and calculated according to the triangular hydrograph theory (Das 2000).

The MUSLE was then run using the data set collected for 41 individual storm events that occurred during the study period (2006–2016). The storm-wise sediment yield predictions were compared with observed data based on the criteria of the coefficient of determination (R²) and relative estimation error (RE) (Green & Stephenson 1986).

The present study was conducted in the Sanganeh research micro-watershed to assess the applicability of the MUSLE for prediction of storm-wise sediment yield. The model was applied to 41 storm events that occurred from December 2006 to April 2009. Figure 2 shows storm-wise sediment yields in different seasons in the Sanganeh micro-watershed. As depicted in Figure 2, the maximum sediment yield was transported in spring due to high rainfall intensity, sediment availability, and lack of sufficient vegetation growth. Also, in summer, due to the scarcity of rainfall, runoff and soil erosion did not occur.

The average weighted values of 0.3 and 1 were assigned to the watershed factors of soil erodibility (K) and conservation practice (P), respectively. Also, based on season and vegetation growth stage, the crop management factor (C) was assigned to be 0.03–0.2. The results of a comparative evaluation between measured and estimated sediment yield data are shown in Figure 3. Comparison of the estimated and observed values (Figure 3) showed that the MUSLE significantly overestimated the sediment yield in the studied micro-watershed. Considering this overestimation of the model, the findings of this study are in agreement with those of several previous studies. For example, Johnson et al. (1986), Epifanio et al. (1991), Sadeghi et al. (2007), Pongsai et al. (2010), and Noor et al. (2010) all concluded that the MUSLE overestimated storm-wise sediment yield prediction. The results of graphical presentation and statistical analysis prove that the MUSLE does not produce reasonable estimates of sediment yield in the Sanganeh micro-watershed.

The resulting underestimation and overestimation depend on various site-specific conditions such as rainfall characteristics, watershed size, land use, and the reliability of observed sediment data (Kandrika & Venkataratnam 2005; Sadeghi et al. 2007; Pongsai et al. 2010). The overestimation tendency of the predictions is probably caused by the topography factor, especially on mountains with steep slopes and in very small watersheds with conditions fundamentally different from those under which the MUSLE (Williams 1975) was originally developed. Nevertheless, no specific erosion model is currently available that simulates sediment yield accurately without calibration. Figure 4 shows a correlation between observed and predicted sediment yield based on seasonal classified data and total data. Although the MUSLE model did not perform well in the case of sediment yield estimation for the Sanganeh micro-watershed, the high correlation between two observed and estimated values (Figure 4) verifies the potential calibration of the model.

Therefore, regression models were then developed contrasting the estimated and 1) total measured sediment yields (traditional approach) and 2) seasonal classified data. The best-fit models between predicted (MUSLE) and observed sediment values were selected based on the maximum R² and minimum RE criteria summarized in Table 1.

The estimation RE of the traditional method and seasonal classification of data were calculated as 62% and 34%, respectively, suggesting an improved MUSLE performance using the season classification of data. The estimation RE of the seasonal classification of data falls within the acceptable range of soil erosion and sediment yield modeling (Sadeghi et al. 2007; Noor et al. 2010). The results of the comparative evaluation of measured and estimated sediment yield data are presented in Figure 5.

The scatter plot of predicted and observed sediment yield data illustrates a very good agreement between two data sets. Scrutinizing the results in Figure 5 and Table 1 shows that the revised MUSLE performed well in the prediction of storm-wise sediment yield in the Sanganeh micro-watershed in Iran.

Also, the seasonal evaluation of model (Table 1) showed that the estimation REs of 31%, 40%, and 33% in spring, autumn, and winter, respectively, which are within or close to the acceptable range. Also, as can be inferred from Figure 2 and Table 1, the model has a good prediction in the high sediment period (i.e., spring). Similar to the findings of present study, several studies (Porretta-brandyk et al. 2011; Zhang et al., 2015; Vigiak et al. 2017; Gao et al. 2018) using other hydrologic models have shown improved model efficiency by using seasonal separating data.

The satisfactory performance of the revised MUSLE suggested its application for the study area and probably for other areas with similar agroclimatologocal conditions, owing to its simplicity and accessibility of required inputs as emphasized by Hrissanthou (2005). The capability of the revised MUSLE in the above evaluation without direct involvement of rainfall characteristics in estimation of sediment yield agrees with the studies of Williams (1975), Williams & Berndt (1977), Hrissanthou (2005), Mishra et al. (2006) and Sadeghi et al. (2007), who stated sediment yield from upland areas is generally better correlated with observed runoff than with rainfall. However, more elaborate and comprehensive studies are essential to further increase understanding of this issue.

It can be concluded from the results of the present study that the original MUSLE did not perform well in the prediction of sediment yield from the study watershed, while its revised version developed during the study significantly improved the performance of the model. Also, seasonal separating data and then evaluation of MUSLE showed a significant increase in the performance of this model compared with the traditional approach. The results of the present study can also be applied for a better understanding of the hydrological processes on the study areas; i.e., north east Iran. Nevertheless, a longer and widespread record of sediment loading is needed to better define the natural conditions and the response of sediment yield.