A high-performance integrated hydrodynamic modelling framework for large-scale multi-process simulation

Coastal cities are prone to the risks from multiple hazards, e.g., compound floods driven simultaneously by interactive fluvial, pluvial and coastal processes. The overland flow and flooding process of a compound event may further erode soil, move and carry along debris of different size, and pick up and transport pollutants to create secondary hazards to exacerbate the flood impact on assets and environment. When assessing and managing multi-hazard risk, it is essential to have a modelling tool that can depict in detail the flooding and associated processes. However, the traditional flood models seldomly consider and simulate the interactive rainfall-runoff-flooding and associated secondary hazard processes.

This work aims to develop and test a high-performance portable modelling framework to simulate the flooding dynamics triggered by multiple drivers, as well as the relevant cascading processes, to support more comprehensive multi-hazard risk assessment and management. To simulate the complex flooding dynamics from multiple sources, the High-Performance Integrated hydrodynamic Modelling System (HiPIMS) developed at Loughborough University is adopted. HiPIMS solves the full 2D shallow water equations (SWEs) using a Godunov-type finite volume method, implemented with novel variable reconstruction and source term discretisation schemes to handle complex domain topography and wetting and drying to achieve stable and accurate prediction. HiPIMS is further implemented on multiple GPUs to achieve high-performance computing to support large-scale high-resolution simulations. In this work, a new version of more compatible and portable HiPIMS is developed by adopting PyTorch (https://pytorch.org) to distribute GPU threads and reconstruct input data and internal variables, making it easier for interfacing with GIS tools and data pre- and post-processing. To ensure the new HiPIMS is extendable to incorporate new modelling components to achieve multi-hazard and multi-process modelling, the main model code is encapsulated to provide interfaces with easy access to hydrodynamic information, i.e., water depths and velocities, for model coupling. Git (a distributed version control system for programmers to collaboratively develop source codes) is further employed to support long-term flexible model development and maintenance.

The capability of the new HiPIMS is demonstrated and confirmed by application to 1) reproduce a surface water flood event driven by fluvial and pluvial processes across the 2500km² Eden catchment in England; 2) initiation and propagating process of floating debris driven by highly transient flood waves; and 3) wash-off, transport and deposition of non-point-source pollutants driven by rainfall induced overland and flood flows over urban surfaces.