Spatial Analysis and Prediction of Soil Erosion in a Complex Watershed of Cameron Highlands, Malaysia

Abstract

Land degradation in the form of erosion is a serious geo-hazard threatening land and water resources sustainability. Its socioeconomic and ecological impacts necessitated its geospatial prediction via susceptibility analysis. Accuracy of susceptibility mapping depends largely on modeling techniques and causative factors (CFs) considered. The study implements scarcely used CFs and support vector machine (SVM) technique for geospatial prediction of soil erosion for low and peak rainfall cycles in a complex watershed of Cameron Highlands Malaysia. The CFs considered are non-redundant static (drainage density, length slope, line ament density, and erodibility) along with some dynamic (surface temperature, soil moisture index, rainfall erosiv ity, and vegetation index) CFs. Four kernel functions of SVM with optimized kernel parameters were used for spatial prediction of erosion for both rainfall cycles. The model performances were validated using the commonest evaluation criteria. The results indicated that polynomial SVM outperformed other models and was used in devel oping soil erosion susceptibility maps. The distribution pattern of erosion showed that majority of the area are “low and moderately” susceptible. The analysis shows that most of the highly susceptible locations are within the urban and agriculture land-use. The study provides information on erosional land forms that could be advanced to gullies. This medium-scale susceptibility map produced could be useful for sustainable watershed management for mitigating erosion.