High-performance simulation technologies for water-related natural hazards

Abstract

Water-related natural hazards, such as flash floods, landslides and debris flows, usually happen in chains. In order to better understand the underlying physical processes and more reliably quantify the associated risk, it is essential to develop a physically-based multi-hazard modelling system to simulate these hazards at a catchment scale. An effective multi-hazard modelling system may be developed by solving a set of depth-averaged dynamic equations incorporating adaptive basal resistance terms. High-performance computing achieved through implementation on modern graphic processing units (GPUs) can be used to accelerate the model to support efficient large-scale simulations. This thesis presents the key simulation technologies for developing such a novel high-performance water-related natural hazards modelling system. A new well-balanced smoothed particle hydrodynamic (SPH) model is first presented for solving the shallow water equations (SWEs) in the context of flood inundation modelling. The performance of the SPH model is compared with an alternative flood inundation model based on a finite volume (FV) method in order to select a better numerical method for the current study. The FV model performs favourably for practical applications and therefore is adopted to develop the proposed multi-hazard model. In order to more accurately describe the rainfallrunoff and overland flow process that often initiates a hazard chain, a first-order FV Godunovtype model is developed to solve the SWEs, implemented with novel source term discretisation schemes. The new model overcomes the limitations of the current prevailing numerical schemes such as inaccurate calculations of bed slope or friction source terms and provides much improved numerical accuracy, efficiency and stability for simulating overland flows and surface flooding. To support large-scale simulation of flow-like landslides or debris flows, a new formulation of depth-averaged governing equations is derived on the Cartesian coordinate system. The new governing equations take into account the effects of non-hydrostatic pressure and centrifugal force, which may become significant over terrains with steep and curved topography. These equations are compatible with various basal resistance terms, effectively leading to a unified mathematical framework for describing different type of water-related natural hazards including surface flooding, flow-like landslides and debris flows. The new depthaveraged governing equations are then solved using an FV Godunov-type framework based on the second-order accurate scheme. A flexible and GPU-based software framework is further designed to provide much improved computational efficiency for large-scale simulations and ease the future implementation of new functionalities. This provides an effective codebase for the proposed multi-hazard modelling system and its potential is confirmed by successfully applying to simulate flow-like landslides and dam break floods.