Mapping the multi-hazard early warning gap: AI-based susceptibility analysis reveals hotspots where Europe needs integrated warning systems

Across Europe, multiple natural hazards increasingly converge as climate change intensifies the frequency and severity natural hazards, yet early warning systems (EWS) remain organised around single hazards. This institutional and technical fragmentation leaves exposed populations without integrated warnings for compound threats that define contemporary disaster risk. Identifying where multiple hazards converge with high societal exposure is essential for prioritising investments in integrated multi-hazard early warning systems (MHEWS). Here, we present an AI-driven approach combining deep learning-based susceptibility mapping with comprehensive exposure analysis to reveal priority regions where Europe's early warning infrastructure falls short. We address this research gap by adapting a convolutional neural network architecture to European susceptibility mapping of a range of hazards including flood, wildfire, landslide, tsunami, drought, heatwave, extreme wind, volcanic eruption and earthquake. We introduce spatial partitioning to prevent data leakage, generate probabilistic susceptibility outputs, and employ Shapley additive explanations values to interpret model drivers. Our analysis reveals that a quarter of Europeans live in regions susceptible to three or more hazards, with critical exposure hotspots concentrated in coastal and Southern Europe and major river basins where population density, economic assets, and agricultural infrastructure converge with high multi-hazard susceptibility. These findings provide an evidence base for strategic allocation of resources toward integrated EWS in regions where single-hazard approaches are demonstrably insufficient.