A global synthetic multi-peril flood event set

Floods are one of the most common, costly, and deadly natural disasters in many regions of the world. Billions of dollars of damages are caused annually, while most studies predict a further worsening of impacts under a warming climate over the next century. To help mitigate the impacts it is important to understand where, when and the likely severity of flooding that might take place. Recently, the emergence of efficient hydraulic modeling frameworks, able to produce flood hazard maps over the entire world, have provided a vital tool that helps to provide this information to end users. However, these maps are typically ‘static’, offering no information about what a real flood event could look like. This is problematic to, for example, emergency planners who may need to know how large the worst case event might be, or those in the insurance sector who may be interested in estimating tail losses across asset portfolios spanning large spatial regions. To meet these requirements, it is important to consider the spatial dependencies in flood events, i.e., given there is flooding in one region, what is the likelihood we see flooding simultaneously in another. 

In this work we attempt to meet this need through the development of a modeling framework that enables the automated creation of thousands of years of synthetic flood footprints, representing pluvial, fluvial and coastal processes, anywhere in the world. We do this by obtaining global, freely available reanalysis products to use as training data to characterize the flood dependence structures within a multivariate extreme value model at selected locations. The dependence structures are then used to derive synthetic events, interpolated to create event surfaces, which are then used to sample from existing global static hazard layers. The output is a dataset containing thousands of years of synthetic multi-peril (pluvial, fluvial, coastal) flood event footprints around the world. This presentation outlines the key input datasets, methodological steps, and validation procedures implemented. We also highlight important limitations and plans for future development.