Machine Learning for Automated Avalanche Terrain Exposure Scale Classification – A Comparison of Two Approaches

Machine learning models can effectively classify avalanche terrain exposure with high accuracy, demonstrating robust generalization across mountain ranges. A study comparing two approaches for Pirin National Park, Bulgaria, found both models achieved over 86% pixel-based accuracy and correctly classified 92-94% of validation polygons. While challenging terrain classes remained difficult to predict, the results highlight the potential of automated methods for avalanche risk management in backcountry areas.

Markov, K., Bühler, Y., Huber, A., & Panayotov, M. (2026). MACHINE LEARNING FOR AUTOMATED AVALANCHE TERRAIN EXPOSURE SCALE CLASSIFICATION – A COMPARISON OF TWO APPROACHES. Proceedings of the International Snow Science Workshop 2026, 1039–1048. https://doi.org/10.15788/1790098776