Application of the various soft computing techniques for Landslide susceptibility mapping (Case study: A part of Haraz Watershed)

Abstract

Abstract The objective of this research was to determination the effective parameter on landslide occurrence and compare the landslide susceptibility zoning methods including Support Vector Machine (SVM) and Gaussian Process (GP) regression based on two kernels (Pearson VII and radial basis) and Random Forest (RF) in the part of Haraz watershed, Iran. In present research, nine factors like slope, aspect, elevation, geology, land use, distance of fault, distance of road, distance of river and precipitation were used as key parameters for assessment of landslide susceptibility. Three statistical comparison criteria including Nash–Sutcliffe model efficiency (NSE), Coefficient of Correlation (C.C) and Root Mean Square Error (RMSE) were used to determine the best performing model. The obtained results shown that the Rf model (with C.C = 0.9753, RMSE = 0.1434 and NSE = 0.9176) is more accurate to assess the landslide susceptibility as compare to the other models. Sensitivity analysis suggeste that the factor, aspect, plays the most substantial role in the evaluation of landslide susceptibility. Comparison of results displays that there is no important diversity between observed and predicted values of landslide occurrence and landslide non-occurrence using GP_PUK, GP_RBF, SVM_PUK, SVM_RBF and Random Forest approaches.