Author(s): Vasilis Bellos; Pierfranco Costabile; Carmelina Costanzo; John Kalogiros

Linked Author(s): Vasilis Bellos, Pierfranco Costabile

Keywords: model benchmarking, 2D shallow water equations. case study

Abstract: The increasing availability of high-resolution topographic data, the development of more and more accurate numerical schemes and the advance in high-performance computing (HPC) techniques fostered the use of two-dimensional Shallow Water Equations (2D-SWEs) as an integrated tool for flood modelling at the catchment scale. According to this approach, hydrological and hydrodynamic flood processes are described entirely within the 2D-SWEs, since the rainfall input and the infiltration losses can be considered as source terms in the mass continuity equation. Although the number of studies focused on the application of integrated 2D-SWEs at the catchment scale is growing quickly, there is a need to provide further evidence about the reliability of this approach in experimental and real-world case studies, highlighting the overall performance in terms of accuracy and computational times. This paper is divided in two parts. In the first one, the numerical parallel code developed by the authors is used within the shallow water inter comparison exercise. Specifically, a finite volume method based on the well-known Roe Riemann solver, first-order accurate in time and space, is used for the computation of the numerical fluxes. An upwind approach is used to discretize the bottom variations and a semi-implicit treatment of the friction source term is implemented. Finally, a robust wet-dry procedure is considered. An unstructured grid, based on irregular triangular elements, is used to obtain the computational domain. The code is written in Fortran90 using OPENMP-MPI directives and it is implemented on a cluster of 5 nodes (each processor is Intel Xeon 10 Core E5-2680v2), using a total of 80 threads. In the second part, the latter solver is used in order to reconstruct the catastrophic flash flood event occurred in November 2017 in Greece, in a small town located near Athens, named Mandra, which had 24 fatalities. The challenges of this benchmark case study are: a) the computational domain is quite complex and is composed by two peri-urban catchments about 20 km2 each, and the urban environment of Mandra; b) the rainfall distribution in space is characterized by significant heterogeneity, due to the orographic mechanism of the storm. The high-resolution topographic data consists of a free 5x5 m Digital Terrain Model generated by the National Cadastre and Mapping Agency of Greece and the boundaries of the building blocks of Mandra generated manually based on the satellite images provided by Google Earth. The rainfall maps are recorded by the mobile weather radar of the National Observatory of Athens with space step equal to 200 m and time step equal to 2 min. Finally, the output of the 2D-SWE solver is compared with post-event field data, which consist of the flood maximum depth at 44 points distributed across the Mandra town.

DOI: https://doi.org/10.3850/IAHR-39WC252171192022SS1864

Year: 2022