Data-driven avalanche forecasting using weather and satellite data

Abstract

Avalanche forecasting is essential for safety in mountainous regions where avalanches threaten human life and infrastructure. Traditional methods for assessing avalanche risk, such as snow pit analysis, are challenging to apply over extensive areas due to their intensive labor and resource requirements. In this study, we explore the potential of satellite-based data for avalanche forecasting by leveraging a dataset of nearly half a million avalanche detections in Norway from 2016 to 2020. This dataset enables a data-driven approach to identifying meteorological precursors to avalanches. We present the methodology for integrating time series of avalanche activity with numerical weather prediction (NWP) data using spatio-temporal deep learning models. We introduce a prototype model and discuss the primary challenges in training this architecture. This framework lays the foundation for improved avalanche forecasting over large regions where in-situ measurements are sparse or unavailable.