Comparison of two modelling strategies for 2D large-scale flood simulations

Abstract

In this paper, two emerging strategies for the reduction of the computational time of 2D large-scale flood simulations are compared, with the aim of evaluating their strengths and limitations and of suggesting guidelines for their effective application. The analysis is based on two state-of-the-art raster flood models with different governing equations and parallelization strategies: PARFLOOD, a GPU-accelerated code that solves the fully dynamic shallow water equations, and LISFLOOD-FP, which combines a parallel implementation for CPU with simplified equations (local-inertial approximation). The results of two case studies (a river flood propagation, and a lowland inundation) suggest that, at coarse grid resolutions, the parallelized simplified model LISFLOOD-FP can represent a good alternative to fully dynamic models in terms of accuracy and runtime, while the GPU-parallel code PARFLOOD is more efficient in case of high-resolution simulations with millions of cells, despite the greater complexity of the numerical scheme.