Towards the next generation probabilistic flood forecasting system for the UK

Flooding is one of the most damaging natural hazards in the UK and worldwide. Building resilience to flooding is key to climate adaptation, with increased demand for accurate and timely flood forecasting required across multiple sectors. Flood dynamics contain complexities that standard forecasting methods struggle to capture, driving the development of more advanced predictive methodologies. Probabilistic flood forecasting can enhance decision-making, such as flood early warning, by offering a better understanding of uncertainty and risk. Uncertainty management is vital since they can significantly impact forecast reliability; uncertainty quantification is crucial for improving flood forecasting models, especially in coupled hydrological-hydrodynamic systems, where multiple interacting variables influence outcomes. This project aims to develop a spatially variable probabilistic flood forecasting model at high-resolution. The work will link different types of models (weather forecast, rainfall-runoff, and flood inundation models), and utilise a Bayesian uncertainty quantification framework. State-of-the-art probabilistic national-scale flood forecasting gives outputs at river locations that have observations or as broad-scale maps (1 km gridded). Here we develop models that can generate maps of flood probabilities with significantly higher resolutions (1-10 m), incorporating these quantified uncertainties and verified against past events. Bayesian Monte Carlo methods will be employed to assess and quantify model uncertainties; sensitivity to a variety of inputs will be assessed through parameter calibration and probabilistic inference to improve model robustness and flood forecast reliability. Through better estimation of uncertainty propagation in coupled hydrological models, this project aids the development of more reliable flood forecasting systems, to support governments and businesses across the UK to become more flood resilient.