A High-Performance Integrated Hydrodynamic Modelling System for Real-Time Flood Forecasting

Flooding is the most damaging and costly type of hydrometeorological hazards affecting millions of people every year in most, if not all, of the countries and regions across the world. With ongoing climate change and rapid urbanization, flood risk is expected to significantly increase in the 21st century and beyond. Flood forecasting, and the subsequent warnings, with sufficient lead-time is an important means to enable society to reduce risk and mitigate loss. Over the last few decades, major advances have been made in the fields of remote sensing, numerical weather prediction (NWP) and flood inundation modelling. It is possible to combine high-resolution NWP data directly into river flow/inundation forecast systems to obtain an extended lead time. However, most of existing flood forecasting systems are based on coarse-resolution (~km) hydrological or simplified hydraulic models to predict the hydrological and hydraulic processes. They are not able to resolve the transient flood hydraulics induced by intense rainfall over catchments or cities with complex topography and to provide reliable forecasts of flood inundation.

The fully hydrodynamic models based on numerical solution to the shallow water equations (SWEs), especially those developed using a shock-capturing numerical scheme, represent the current state-of-the-art in flood modelling. However, these full hydrodynamic models are computationally demanding, restricting their application to large-scale simulations across an entire city or catchment, and are therefore rarely used for real-time flood forecasting across a large domain. Harnessing the recent GPU parallel computing technologies, a High-Performance Integrated hydrodynamic Modelling System (HiPIMS) has been recently developed to numerically solves the SWEs using a finite volume Godunov-type shock-capturing scheme for flood modelling. HiPIMS achieves multi-GPU and multi-system (machine) parallelisation through the NVIDIA CUDA parallel computing platform and MPI (Message Passing Interface) to substantially improve its computational efficiency. To support real-time flood forecasting, computing environment/interface has also been developed to automatically convert the numerical weather forecast products (generated by the UKV model) from the UK Met Office to drive HiPIMS to simulate the resulting flooding process. This new high-resolution hydrodynamic flood forecasting system was then applied to reproduce the 2015 Storm Desmond floods over the 2500km²UK Eden Catchment at a 5m resolution; the simulation was run 2.5 times faster than real time, effectively demonstrating HiPIMS’s potential for operational real-time flood forecasting.