Rockfall Forecasting Using Ensemble Deep Learning and Temporal Gradient-Based Explanations

Rockfalls pose a significant risk for infrastructures and human activities. Accurate forecasting is therefore crucial for prevention. While current methods rely on expert hypotheses, AI models, particularly explainable ones, offer a promising way to improve the accuracy of forecasts and provide transparent, trustworthy results. In this paper, an ensemble of multisource neural networks based on InceptionTime modules whose forecasts are explained using a temporal Grad-CAM approach is being developed for the Mont Saint-Eynard cliff, showcasing the potential of AI in this area.