Redefining landslide susceptibility under extreme rainfall events using deep learning

Abstract

The apparent changes in the Indian summer monsoon rainfall pattern and the nature of extreme rainfall events (EREs) in the southern Western Ghats (WG) caused widespread landslides across the region. Landslide susceptibility maps generated using the past landslide inventory are one of the efficient tools that can help mitigate the deleterious effects of landslides. Landslide susceptibility maps produced using the landslide inventory of normal and extreme rainfall years by deep neural network (DNN) modelling vary in terms of their degree of susceptibility, and the differences are prominent in the moderate and high susceptibility zones. The DNN model trained by landslide inventory of normal rainfall years is capable of predicting landslides during EREs (with Area Under the Curve (AUC) value >0.90). The inclusion of the landslides that occurred during recent EREs (since 2018) into the existing landslide inventory provides a more accurate and refined prediction of landslide susceptibility, which facilitates risk-informed landscape planning and development of the region. In addition, the result reveals that 13 % of the Kerala state is extremely susceptible for landslide occurrence, among this Idukki, Palakkad, Malappuram, Pathanamthitta, and Wayanad districts are highly vulnerable to the occurrence of landslides. Besides, the study also shows an increase of 3.46 % area in extreme susceptibility zone after the 2018 ERE. The updated landslide susceptibility map of the region may be used as a vital tool for planning landslide mitigation activities in the wake of recurrent EREs and associated landslide occurrences.