Investigation of potential of erosion prone areas with ICONA models, support vector machine, chad and random forest (Case study: Gonabad Basin)

Experimental erosion estimation models have been developed for a specific area and their calibration is necessary for use in conditions other than their location. Examining the accuracy of experimental models for estimating erosion can lead to better estimates of sediment load and thus better design of soil and water conservation operations. Therefore, identification of high risk areas of erosion to control and reduce erosion and sediment productionIt is necessary. The purpose of this study is to investigate the accuracy and capability of ICONA models, support vector machine, field and random forest in estimating erosion. First, digital layers of effective variables in erosion including slope, geological formation, land use, soil, height, slope direction, surface curvature, waterway network density, distance from waterway, fault density, distance from fault and topographic moisture index (Twi) Turned. In this study, in order to compare different models, statistical indices of correlation coefficient (R) and absolute error (MAE) have been used. The results showed that among the mentioned models, the support vector machine model, ICONA and random forest with M7, M9 and M12 patterns had the highest accuracy with correlation coefficient (0.899), (0.845) and (0.921). And has the lowest mean absolute error value (MAE = 0.711), (MAE = 0.721) and (MAE = 0.628). According to the study of effective factors in soil erosion model, it is concluded that the parameters of slope, geological formation, land use, soil, distance from waterway and topographic moisture index (Twi) are more sensitive to erosion and the factors affecting erosion in These areas are more active.