Performance evaluation of machine learning and statistical techniques for modelling landslide susceptibility with limited field data

Due to the unique blend of physiography, climate, and socio-economic factors, the windward slopes of the southern Western Ghats (WG) are highly susceptible to the occurrence of landslides. The present study compares the efficiency of advanced deep learning and machine learning techniques, such as deep neural network (DNN), and random forest (RF) with conventional methods, viz., frequency ratio (FR) and analytical hierarchy process (AHP) method for mapping the landslide susceptibility in a highland region of the western slopes of the southern WG, where the field observations are limited. Various factors controlling the occurrence of landslides, such as lithology, geomorphology, land use/ land cover, soil texture, proximity to lineaments, roads, and streams, slope angle, rainfall, topographic wetness index (TWI) and curvature are used for the landslide susceptibility mapping. The results of the present study indicate comparable performances between machine learning models and conventional approaches in field data-scarce regions. Moreover, this study has extended implications in landslide risk management of the southern WG in the context of repeated extreme rainfall events.

Automated summary

This study compares the performance of machine learning models and conventional methods in mapping landslide susceptibility in data-scarce regions of the southern Western Ghats, and finds similar results between the two approaches.