Benchmarking (multi)wavelet-based dynamic and static non-uniform grid solvers for flood inundation modelling

Abstract

This paper explores static non-uniform grid solvers that adapt three raster-based flood models on an optimised non-uniform grid: the second-order discontinuous Galerkin (DG2) model representing the modelled data as piecewise-planar fields, the first-order finite volume (FV1) model using piecewise-constant fields, and the local inertial (ACC) model only evolving piecewise-constant water depth fields. The optimised grid is generated by applying the multiresolution analysis (MRA) of multiwavelets (MWs) to piecewise-planar representation of raster-formatted topography data, for more sensible grid coarsening based on one user-specified parameter. Two adaptive solvers are also explored that apply the MRA of MWs and of Haar wavelets (HWs) to, respectively, scale and adapt the DG2 (MWDG2) and FV1 (HWFV1) modelled data dynamically in time. The performance of the non-uniform grid and adaptive solvers is assessed in terms of flood depth and extent, velocities, and CPU runtimes, with reference to the raster-based DG2 model predictions on their finest resolution grid. The assessments considered three large-scale flooding scenarios, involving rapid and slow-to-gradual flows. MWDG2 is found to be the most favourable choice when modelling rapid flows, where it excels in capturing small velocity variations. For slow-to-gradual flows, the adaptive solvers deliver less accurate outcomes, and their efficiency can be hampered by overhead costs of the dynamic MRA. Instead, non-uniform DG2 is recommended to capture urban flow interactions more accurately. Non-uniform ACC is 5 times faster to run than non-uniform DG2 but delivers close flooding depth and extent predictions, thus is more attractive for fluvial/pluvial flood simulation over large areas.