A globally-applicable framework for compound flood hazard modeling

Abstract. Coastal river deltas are susceptible to flooding from pluvial, fluvial, and coastal flood drivers. Compound floods, which result from the co-occurrence of two or more of these drivers, typically exacerbate impacts compared to floods from a single driver. While several global flood models have been developed, these do not account for compound flooding. Local scale compound flood models provide state-of-the-art analyses but are hard to scale up as these typically are based on local datasets. Hence, there is a need for globally-applicable compound flood hazard modeling. We develop, validate and apply a framework for compound flood hazard modeling, which consists of the local high-resolution 2D hydrodynamic flood model SFINCS, which is automatically set up from global datasets and loosely coupled with a global hydrodynamic river routing and flood model, as well as a global surge and tide model to account for interactions between all drivers. To test the framework, we simulate two historical compound flood events, cyclones Idai and Eloise, in the Sofala province of Mozambique, and compare the flood extent to observations from remote sensing and to the global quasi 2D CaMa-Flood model. The results show that while the global and local model have similar skill in terms of the critical success index, they result in rather different flood maps. On the one hand, the local model has a higher hit ratio due to the representation of direct coastal and pluvial flooding (rain on grid) and a higher floodplain connectivity. It also shows a faster response to coastal drivers within the estuaries and more realistic flood depth maps. On the other hand, the local model has a higher false alarm ratio, which is partly explained by the inclusion of direct pluvial flooding without sufficient representation of small scale (subgrid) drainage capacity. To showcase a possible application of the framework, we also determine the dominant flood drivers and transition zones between flood drivers for both events. These vary significantly between both events because of differences in the magnitude of and time lag between the flood drivers. We argue that a wide range of plausible events should be investigated to get a robust understanding of compound flood interactions, which is important to understand for flood adaptation, preparedness, and response. As the model setup and coupling is automated, reproducible, and globally applicable, the presented framework is a promising step forward towards large scale compound flood hazard modeling.