GIS-based support vector machine model in shallow landslide hazards prediction: A case study on Ilam dam watershed, Iran

The SVM algorithm is an applied method that has been considered in recent years to study landslides. The main purpose of this study is to evaluate the mapping power of the GIS-based SVM model with kernel functions analysis for spatial prediction of landslides at the Ilam dam watershed.

According to review sources, 14 underlying factors including elevation, slope, aspect, plan curvature, profile curvature, LS factor, TWI, SPI, Lithologic units, land cover, NDVI, road distance, distance to the drainage channel, distance to fault were selected as factors affecting the occurrence of landslides in the study area and the mentioned layers were prepared in the GIS. In the present study, the non-linear two-class SVM method was used, the two-class SVM requires both datasets representing the occurrence of landslides and non-occurrence of landslides. The landslide inventory was randomly divided into a training dataset of 75% for building the models and the remaining 25% for the validation of the models.

Results and conclusion:

The validation results showed that the area of the prediction-rate curve under the curve (AUC) for landslide susceptibility maps produced by the SVM linear function, SVM polynomial function, SVM radial basic function, and SVM sigmoid function are 0.946, 0.931, 0.912, and 0.871 respectively. To assess the influences of factors on the landslide susceptibility map were used the Cohen’s kappa index of the model. The result shows that the most effective factors are the distance to roads, distance to drainages, and plan curvature in this area.