Skillful Decadal Flood Prediction
Simon Moulds, Louise Slater, Nick Dunstone, Doug Smith
Abstract
Abstract Accurate long‐term flood predictions are increasingly needed for flood risk management in a changing climate, but are hindered by the underestimation of climate variability by climate models. Here, we drive a statistical flood model with a large ensemble of dynamical CMIP5‐6 predictions of precipitation and temperature. Predictions of UK winter flooding (95th streamflow percentile) have low skill when using the raw 676‐member ensemble averaged over lead times of 2–5 years from the initialization date. Sub‐selecting 20 ensemble members that adequately represent the multiyear temporal variability in the North Atlantic Oscillation (NAO) significantly improves the flood predictions. Applying this method we show positive skill in 46% of stations compared to 26% using the raw ensemble, primarily in regions most strongly influenced by the NAO. Our findings reveal the potential of decadal predictions to inform flood risk management at long lead times.