A comprehensive event-based quantification of global flood risk

Global flood losses have exceeded one trillion dollars this century, driven by the twin forces of climate change and floodplain urbanisation. However, our collective understanding of these risks remains fundamentally constrained by the observational record, which is too short to adequately sample extremes, and by existing global models that fail to account for the spatial dependence of hazard events. In this study, we present the first high-resolution global catastrophe model for fluvial, pluvial, and coastal flooding, offering a comprehensive quantification of property flood risk that alleviates the limitations of historical data and existing models.

Our framework couples existing 30 m global flood hazard maps – simulated using a 1D/2D inertial formulation of the shallow water equations – with a stochastic event set representing 10,000 years of plausible climate conditions. By using a conditional extremes statistical model trained on ERA5 reanalysis data, we generated 9.7 million synthetic global flood events that capture the complex spatio-temporal and multi-peril dependencies often ignored in traditional assessments. This hazard event set is intersected with a spatially granular global economic exposure database, where regional capital stock models for residential, commercial, and industrial assets are downscaled onto the Global Human Settlement Layer and assessed using engineering-based vulnerability functions.

We estimate that the global Annual Average Loss (AAL) from flooding is $144 billion under current climate conditions. Crucially, the model reveals that in a 1-in-250-year extreme year (0.4% annual exceedance probability), global direct losses reach $531 billion: equivalent to 0.5% of global GDP and approximately 2.6 times the economic losses of Hurricane Katrina. The analysis highlights large risk inequities: while North America and East Asia face the highest absolute losses (~$350 billion in extreme years), the mega-deltas of Southeast Asia emerge as critical hotspots where losses are highest relative to capital stock.

Furthermore, we demonstrate the high volatility of flood losses, showing that even (a purely theoretical) 50 years of stationary observations cannot constrain national-scale AAL estimates to within a factor of three. These results underscore the necessity of synthetic event-based modelling for robust resilience planning, justifying large-scale mitigation investments, and helping to bridge the disaster insurance protection gap.