Spatial Prediction of Rainfall Induced Shallow Landslides Using Adaptive-Network-Based Fuzzy Inference System and Particle Swarm Optimization: A Case Study at the Uttarakhand Area, India

Abstract

Landslides generally occur during rainy season in Himalayas. Most of the landslides observed in the Uttarakhand part of Himalaya, India are of shallow nature. In the present study, we proposed a hybrid model Particle Swarm Optimization based Adaptive-Network-Based Fuzzy Inference System (PSOANFIS), which is a hybrid intelligent approach of Adaptive-Network-Based Fuzzy Inference System (ANFIS) and Particle Swarm Optimization (PSO), for spatial prediction of shallow landslides in part of Uttarakhand State. Firstly, a total of 1295 historical landslide events occurred in the area were identified and mapped from satellite images in conjunction with available historical data from reports to construct a landslide inventory map. In addition, 16 affecting factors (slope angle, slope aspect, elevation, curvature, plan curvature, profile curvature, lithology, soil, distance to lineaments, lineament density, land cover, rainfall, road networks, distance to roads, road density, river networks, distance to river, and river density) were taken into account for landslide spatial modeling. Datasets (training and testing) were then generated from the analysis of the collected data using GIS application. Thereafter, landslide model PSOANFIS was constructed using training dataset for spatial prediction of landslides. Performance of the proposed hybrid model has been compared with another benchmark landslide model namely Support Vector Machines (SVM). Lastly, the predictive capability of the hybrid model was validated using Receiver Operating Characteristic (ROC) curve and Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) indexes. The results of the present study show that the PSOANFIS model performed well for spatial prediction of rainfall induced shallow landslides, thus the PSOANFIS method can also be applied for the development of better landslide predictive models in other landslide prone areas of the world.