Regional flood events often have more severe impacts than localized events. However, the spatial dimension of floods is often neglected when deriving flood hazard estimates. Reasons for this neglect are our limited knowledge on processes governing spatial flood dependencies and the scarcity of modeling approaches realistically representing these dependencies. The aim of this talk is therefore threefold: (1) to improve our process understanding of spatial flood dependencies, (2) to propose a stochastic simulation approach for spatial extremes, and (3) to derive regional flood hazard estimates for the United States.

To improve process understanding, we investigate how and why spatial flood dependencies vary seasonally and regionally over the United States. We show that spatial flood dependence is highest in spring, especially in mountainous areas, high in winter at the Pacific coast and the Appalachian Mountains, and high in summer in the Rocky Mountains but generally weak in fall. To improve simulation approaches, we introduce the continuous, stochastic simulation approach PRSim.wave, which reproduces the temporal and distributional characteristics of streamflow at individual sites and retains the spatial dependencies between sites even for spatial extremes. To derive regional flood hazard estimates, we use PRSim.wave to generate long and spatially consistent time series of daily discharge for a large set of catchments in the conterminous United States. We use flood events extracted from these series to estimate how probable it is that a certain percentage of stations within a specific river basin is jointly flooded. We show that there are strong regional differences in the likelihood of joint and potentially widespread flooding. We conclude that the consideration of seasonal and regional variations in spatial flood dependencies is essential for the derivation of reliable, regional flood hazard estimates.