Advancing Urban Flood Predictions Using Weather Ensemble, Machine Learning and Uncertainty Quantification

Flooding is the most pervasive and widely disruptive natural hazard facing many urban communities. Intensifying climate change, aging infrastructure and expanding urbanization are increasing the frequency of damaging flood events and making their prediction more challenging across the globe. Flood predictability is often limited by incomplete knowledge of relevant processes, highly uncertain hydrodynamic model parameters, coarse spatial resolutions, and expensive computational cost for large simulations and uncertainty analysis. Current research and operational efforts in hydrological forecasting aim to develop and implement enhanced forecasting systems to improve the quality of flood forecasts. The key objectives of this study is to advance the predictive understanding of urban flooding by integrating emerging innovations in numerical weather prediction, machine learning and uncertainty quantification. We will discuss the results from: (i) characterization of key physical processes that drives urban flooding, (ii) generation and verification of multimodel ensemble flood forecasts, and (ii) uncertainty propagation and quantification along the hydrometeorological forecasting chain. Reliable prediction of urban floods at high spatiotemporal resolution is crucial to mitigating its threats and for more effective early warning services.