Future Rainfall and Flood Extremes (FURFLEX)

Extreme rainfall and flooding cause some of the largest impacts on society out of all meteorological events, and these are predicted to be strongly exacerbated by climate change. In the UK, flood defence planning requires understanding the severity of events at and beyond the 100 year return level (i.e. the magnitude that would be expected to be exceeded once per 100 years on average). Existing predictions are inadequate, relying on simulations of future rainfall changes from small samples of coarse-resolution climate model output and simple statistical flood prediction methods. These approaches do not capture important effects such as rainfall becoming more concentrated in shorter bursts in a warmer climate. The flood prediction datasets are opaque and lead to predictions of financial losses due to flooding three times those observed, giving little confidence in their use to quantify extreme thresholds and project climate change impacts.

These problems can be addressed by using physically-based modelling of high-resolution rainfall and flooding, based on fundamental laws. This combines global climate models that simulate large-scale weather states, local-scale weather models to predict detailed precipitation for individual river catchments, hydrological models to predict streamflow in rivers and flood models to determine flood extent and losses. It has not previously been possible to study extreme rainfall and flood events with this approach due to the computational expense of sampling enough of these rare events. Our recent advances have overcome this.

We will produce physically-based simulations that quantify extreme high-resolution rainfall, streamflow and flood risks at ~100-1000 year return levels across the UK for the first time, addressing the key policy-relevant events. We will do this for the present up to 2080 and at policy-relevant global warming levels. We will also show the robustness of projections across different models for the first time. We will do this using the following groundbreaking advances made by the project team:

- local-scale (2.2km) precipitation projections produced with the "UKCP Local" UK climate model that can capture strongly convective rainfall systems, which have a critical role in flood risk. These have recently produced a great advance in the quality of rainfall simulations.

- a very efficient emulator of these high-resolution rainfall simulations based on cutting edge machine learning. This enables large samples of predictions to be produced for studying extremes at low cost, based on existing multi-thousand year, coarse-resolution climate model runs. Unlike previous statistical approaches, the method can produce rainfall predictions with realistic spatial structure, as required for realistic flood modelling.

- national-scale hydrological and flood modelling at ~20m resolution, combined with exposure and vulnerability data, which can translate these rainfall predictions into river flows and flood risk, enabling decision-making at the local scale.

We will also use our rainfall emulator to show the range of plausible changes in extremes across different climate and hydrological models for the first time, which is necessary for anticipating the most severe possible outcomes and mitigating the associated risks.

Once demonstrated, our methods could be applied to a wide range of other phenomena and locations, greatly increasing access to local-scale climate impacts modelling. We will work with our Met Office, Environment Agency and Fathom risk consultancy partners to use our findings to improve flood risk quantification and mitigation for industry and government.