LISFLOOD-FP 8.2: GPU-accelerated multiwavelet discontinuous Galerkin solver with dynamic resolution adaptivity for rapid, multiscale flood simulation

Abstract

The second-order discontinuous Galerkin (DG2) solver of the two-dimensional shallow water equations in the raster-based LISFLOOD-FP 8.0 hydrodynamic modelling framework is mostly suited for predicting small-scale transients that emerge in rapid, multiscale floods caused by impact events like tsunamis. However, this DG2 solver can only be used for simulations on a uniform grid where it may yield inefficient runtimes even when using its graphics processing unit (GPU) parallelised version (GPU-DG2). To boost efficiency, the new LISFLOOD-FP 8.2 version integrates GPU parallelised dynamic (in time) grid resolution adaptivity of multiwavelets (MW) with the DG2 solver (GPUMWDG2). The GPU-MWDG2 solver performs dyadic grid refinement, starting from a single grid cell, with a maximum refinement level, L, based on the resolution of the Digital Elevation Model (DEM). Furthermore, the dynamic GPU-MWDG2 adaptivity is driven by one error threshold, ε, against normalised details of all prognostic variables. Its accuracy and efficiency, as well as the practical validity of recommended ε choices between 10−4 and 10−3, are assessed for four laboratory/field-scale benchmarks of tsunami-induced flooding with different impact event complexities (i.e. single- vs. multi-peaked) and L values. Rigorous accuracy and efficiency metrics consistently show that GPU-MWDG2 simulations with ε = 10−3 preserve the predictions of the GPU-DG2 simulation on the uniform DEM grid, whereas ε = 10−4 may slightly improve velocity-related predictions. Efficiency-wise, GPU-MWDG2 yields considerable speedups from L ≥ 10 – due to its scalability on the GPU with increasing L – which can be around 2.0-to-4.5-fold. Generally, the bigger the L ≥ 10, the lower the event complexity over the simulated duration, and the closer the ε to 10−3, the larger the GPU-MWDG2 speedups over GPUDG2. The LISFLOOD-FP 8.2 code is open source, under the GPL v3.0 licence, as well as the simulated benchmarks’ setup files and datasets, with a video tutorial and further documentation on https://www.seamlesswave.com/Adaptive (last access: 6 July 2025).