An integrated approach based landslide susceptibility mapping: case of Muzaffarabad region, Pakistan

Abstract

Landslides result in the devastation of property and loss of lives. This study assesses landslide susceptibility by employing geographic information systems (GIS) and machine learning techniques, that is, support vector machine (SVM) and artificial neural network (ANN), with the integration of advanced optimization techniques, that is, particle swarm optimization (PSO). The landslide-inducing factors considered in this study include fault density, lithology, road density, slope, elevation, flow direction, aspect, earthquake intensity, curvature, Normalized Difference Water Index (NDWI), waterways density, rainfall, and Normalized Difference Vegetation Index (NDVI). The resulting landslide susceptibility maps (LSMs) showed that the areas falling under the high and very high susceptibility class have higher rainfall levels, weak lithology, high NDWI, and flow direction. The accuracy assessment of the techniques showed that ANN with an Area Under the Curve (AUC) of 0.81 performed better than SVM with an AUC of 0.78 without the optimization. Similarly, the performance of ANN was also better than SVM using PSO. During the integrated modeling, the AUC of PSO-ANN was 0.87, whereas the AUC of PSO–SVM was 0.84. The accuracy assessment of the produced LSMs also showed a similar trend in terms of accuracy percentage as that of the models.