A Light-Weight ANN Model for Landslide Detection: A Case Study of Idukki, India

This article presents a light-weight ANN model for the creation of a landslide susceptibility map (LSM) for the district of Idukki in the South Indian state of Kerala. The landslide conditioning factors (LCF) considered for the creation, training, validation and testing of the LSM are elevation, slope, aspect, curvature, topographic wetness index (TWI), stream power index (SPI), rainfall, topographic ruggedness index (TRI), geology, soil type and land use and land cover. The Frequency Ratio (FR) analysis has been carried out on the LCFs and those having the highest Predictive Rate (PR) have been determined as aspect, slope, rainfall and soil type. Once the LSM is created, it is tested using landslide and non-landslide points using the proposed ANN model which yields an accuracy of 83.5%. Future scope in this work is to improve the accuracy of the model by using metaheuristic algorithms for optimization of weights of the ANN model.