Evaluation of Different Supervised Learning Smart Methods and Response Surface Method to Optimize Factors Affecting Erosion (Case Study: Emamzadeh Watershed of Baghmalek)

Evaluation of soil erosion control factors is important regarding the application of management practices. In this study, the effects of non-structural management practices including revision of crop cover (RC) and exclosure (EX) were simulated using WEPP model in ​​Emamzadeh Abdullah watershed of Baghmalek, located in the northeast of Khuzestan Province. Optimization of physical and hydraulic parameters affecting erosion including soil texture and components of the Van Genuchten equation was performed using response surface methodology (RSM), random forest (RF), support vector machine (SVM) and artificial neural network (ANN). Also, the soil erosion rate before and after management practices was defined as the first response (R1) and the second response (R2), respectively. Optimization results by Orange software including random forest methods, support vector machine and artificial neural network showed that the random forest method with MSE, RMSE and R2 equal to 0.991, 0.995 and 0.963 respectively, for the first response and equal to 0.095, 0.307 and 0.974 respectively, for the second response is the most proper method. Also, optimization by response surface method is the most appropriate method with MSE, RMSE and R2 equal to 28.7, 5.37 and 0.999 respectively, for the first response and equal to 4.16, 2.03 and 0.998 respectively, for the second response. Generally, using optimization techniques is a convenient method for evaluating management practices and finally selecting the best one for critical areas. Appropriate management practices based on optimal conditions leading to water and soil loss reduction.