A framework for avalanche risk probability assessment based on snow and terrain using artificial intelligence

Abstract

Avalanches are a recurring natural hazard in the Indian Himalayas, causing significant loss of life and damage to critical infrastructure. The steep terrain, complex snowpack behavior, and dynamic weather conditions make forecasting snow avalanches highly challenging. This research proposes an AI-based avalanche susceptibility assessment framework that integrates static and dynamic parameters from terrain, snow, and meteorological datasets. Spatial interpolation techniques are employed to generate continuous raster layers, while advanced machine learning models Random Forest, xgboost, and Neural Networks are trained to classify susceptibility zones. The study area, Manali in the North-Western Himalayas, serves as a high-risk testbed for model validation. The framework utilizes a 12.5 m resolution Digital Elevation Model (DEM) from ALOS PALSAR to derive 15 terrain layers, augmented with snow and meteorological features, to build a comprehensive training dataset. Hyperparameter tuning and model evaluation highlights Random Forest and neural networks as strong performers, with ROC-AUC values up to 0.88. A significant innovation is the development of a web-based 3D GIS platform for real-time hazard visualization and decision support. The framework is scalable to other mountainous regions and offers critical benefits to the military, tourism, and civil sectors by supporting proactive hazard zoning, emergency preparedness, and infrastructure risk reduction. newline

Description

Keywords

Citation

item.page.endorsement

item.page.review

item.page.supplemented

item.page.referenced