Landslide Susceptibility Assessment Using an Optimized Group Method of Data Handling Model

Abstract

Landslides can cause considerable loss of life and damage to property, and are among the most frequent natural hazards worldwide. One of the most fundamental and simple approaches to reduce damage is to prepare a landslide hazard map. Accurate prediction of areas highly prone to future landslides is important for decision-making. In the present study, for the first time, the group method of data handling (GMDH) was used to generate landslide susceptibility map for a specific region in Uzbekistan. First, 210 landslide locations were identified by field survey and then divided randomly into model training and model validation datasets (70% and 30%, respectively). Data on nine conditioning factors, i.e., altitude, slope, aspect, topographic wetness index (TWI), length of slope (LS), valley depth, distance from roads, distance from rivers, and geology, were collected. Finally, the maps were validated using the testing dataset and receiver operating characteristic (ROC) curve analysis. The findings showed that the “optimized” GMDH model (i.e., using the gray wolf optimizer [GWO]) performed better than the standalone GMDH model, during both the training and testing phase. The accuracy of the GMDH–GWO model in the training and testing phases was 94% and 90%, compared to 85% and 82%, respectively, for the standard GMDH model. According to the GMDH–GWO model, the study area included very low, low, moderate, high, and very high landslide susceptibility areas, with proportions of 14.89%, 10.57%, 15.00%, 35.12%, and 24.43%, respectively.