Landslide susceptibility assessment employing machine learning ensemble models: a study in the most severely battered district of the Southern Western Ghats

Abstract

Landslides are one of the natural catastrophes which are frequently reported in the Western Ghats region and result in severe loss. This modelling intends to demarcate susceptible zones in one of the most impacted districts in the southern Western Ghats. Idukki, being the worst affected district, reported more than 2000 landslides in the year 2018. Two machine learning ensemble models and geoinformation techniques have been employed to identify the susceptibility. Twelve landslide conditioning factors have been utilized for this study. The ROC curve-based validation technique ascertained good and fair prediction capability for the created maps, with AUC scores of 0.821 and 0.776 for the MB-REPTree and AB-REPTree models, respectively. From the validation scores, it is found that the MB-REPTree model is more efficient and of good operational use. The study found 7.81% of the district as very highly susceptible and 16.06% as highly susceptible. So, this study suggests that the MB-REPTree model is the best model to demarcate susceptible zones, not just in the Western Ghats but also in other places with similar climatic and terrain conditions.