Duration: 24 months
In October 2011, the Bernese Alps experienced a devastating flood due to rapid snowmelt exacerbated by heavy rainfall, causing approximately CHF 90 million in damages [1]. To effectively plan for flood hazards, it's crucial to understand the compounding event paths leading to the flood along with their probabilities. Rain-on-snow flooding is just one example of a compound climate-related event affecting communities and infrastructure, underscoring the need for an analysis framework [2]. Advancements in earth observation data provide detailed insights into the combinations of events leading to disasters [3]. However, existing computational methods have limitations. Physical models like SNOWPACK and CRHM are sensitive to calibration parameters, potentially misrepresenting snowpack structure [4,5]. While statistical models offer some improvement [6], integrating their strengths remains challenging. Classical statistical methods, such as copula and extreme value models, excel at extrapolation but lack high-dimensional pattern recognition. In contrast, machine learning excels at pattern recognition in known scenarios but struggles with rare combinations. This research will combine both statistical approaches to create a model that is strong in both areas, underpinning an interactive map that offers insights into compound rain-on-snow flooding across the Alps. New combinations of flood hazards can be explored along with their likelihoods, allowing for scenario exploration. For example, the tool can predict the most likely combination of snowmelt and rainfall for a given flood size, and assess flood risk under changing rainfall and snowpack conditions. This solution will enhance flood prediction, support disaster preparedness, and pave the way for modelling other compound climate events. By leveraging satellite data, our project aligns with ESA’s mission to develop satellite-based technologies and services for climate resilience.
