The Avalanche Risk Prediction Intelligent System: Susa Valley Alps Case Study

Snow avalanches pose a significant and escalating natural hazard, especially in climate change and global warming times, demanding advanced predictive capabilities to safeguard lives and infrastructure in mountainous regions. Traditional methods, even when augmented by machine learning applications, frequently contend with inherent data complexities, imbalanced event occurrences, and a critical trade-off between minimizing false negative predictions (safety risks) and reducing economically detrimental false positives in a timely manner. This paper introduces the Avalanche Risk Prediction Intelligent System (ARPIS) implementing a novel data-driven approach designed to address these challenges through a comprehensive, machine learning-centric framework. ARPIS proposes an incremental, feedback-oriented methodology that continuously collects and manages data, improves the model, and infers live avalanche risk prediction, eventually triggering alerts to support decision making. The ARPIS suitability and effectiveness are demonstrated through a full-fledged case study on a dedicated dataset from the high-risk mountainous terrain of the Susa Valley in the Italian Alps, which code and dataset are publicly available.