Terrain-based avalanche susceptibility mapping in a Manali region of Himachal Pradesh, India: machine learning approaches

Abstract

Avalanches are natural events that can lead to significant risks to both human life and property. The creation of an avalanche susceptibility map is a valuable tool for effectively managing the avalanche prone areas. The primary objective of this paper is to utilize and analyse machine learning models for susceptibility mapping, with the goal of classifying avalanche-prone regions based on terrain parameters extracted from a digital elevation model. In other word, to explore the capability of Tree-based machine learning methods to handle the GIS dataset. Fifteen data layers have been extracted, stacked, and processed to create training and testing data using the avalanche inventory. Three tree-based machine learning models has been trained and tuned using grid search on dataset that has been split into 80:20 for model calibration and validation. Results indicated that both advanced models had an excellent performance in terrain-based avalanche modelling (ROC-AUC > 85%), although true positive and true negative analysis demonstrated the superior performance of Random Forest. Feature importance analysis indicated that elevation and aspect are the top effective and most common feature among all the variables and models, respectively. Building a high-quality and informative database is a crucial part, and avalanches inventory classification before susceptibility assessment is a key step in enhancing the accuracy of the model. The study’s findings can offer valuable insights for land use planning, enabling the control of avalanche paths and mitigating potential hazards. Additionally, these results can serve as a valuable reference for future studies focused on snow avalanche hazards modelling.