Spatially coherent statistical simulation of widespread flooding events under climate change

Abstract

Simulating rare widespread hydrological events can be difficult even with the use of modelled data such as the UKCP18 12 km regional climate projections. To generate larger event sets for application in catastrophe modelling, two statistical approaches are highlighted and applied to widespread GB-generated flooding events using a grid-based hydrological model and UKCP18 regional projections. An Empirical Copula method was applied on a national scale, generating over 600,000 events across two time-slices (1980–2010 and 2050–2080). This was compared to model-generated events and showed good matching across time-slices and ensemble members, although lacked some ability to describe the least-rare events. The Empirical Copula was also compared to an implementation of a conditional exceedance model. This model was much more computationally intensive so was restricted to Northwest England but offered the ability to be tuned more finely through choices of marginal distributions. Analysing over 11,000 events, it also matched well with the Empirical Copula and model-generated events but under-represented the smallest events. Both approaches require a broad dataset to draw from but showed reasonable efficacy. For simple statistics, the Empirical Copula shows the potential to be a powerful tool in exploring spatial structure over large regions or at a fine spatial resolution.