Improving Computational Efficiency of Compound Flooding Simulations: the SFINCS Model with Subgrid Features

Abstract:

Natural hazards can have significant impacts on coastal communities as demonstrated by recent hurricanes Harvey (2017) in Houston, Texas, and Irma (2017) in Jacksonville, Florida. In both events, flooding was not only driven by marine processes (storm-surge and tides), but also to great extent by pluvial and fluvial processes. To correctly simulate flooding in such compound flooding events requires modelling of all the flood drivers (Leijnse et al., in review). Physics-based numerical models can be used for this task, but are often too computationally-demanding to be used to simulate the large number of scenarios needed for probabilistic forecasting and/or climate variability assessments. With this in mind, the reduced-physics model SFINCS (Leijnse et al., in review) was developed to rapidly simulate compound flooding events. Model studies show that SFINCS has similar accuracy to full-physics models, but at substantially reduced computational expense. However, simulating flooding in complex environments (e.g. creeks and tributaries), requires relatively high-resolution computational grids, and is therefore still computationally-expensive.

To improve the computational efficiency of the model, an updated version of SFINCS with subgrid features was developed. The subgrid functionality allows the computation of fluxes on a coarse grid, while keeping track of details of the high-resolution topography in subgrid cells. For the Harvey case study, this meant using a 5 m resolution DEM to compute flow conveyance of local drainage channels and bayous, while calculating fluxes on a coarser 200 m grid with larger timesteps. This reduced the computation time on a standard laptop from 4 hours to only 1 minute. The accuracy was similar to running without subgrid features and computing fluxes on the finer resolution. Similar results were obtained for the Irma case study.

During this presentation, we will introduce the SFINCS model and the theoretical basis for the subgrid features. Moreover, we will demonstrate that for both cases the possibility of subgrid in SFINCS reduces the computational expense substantially, which enables the possibility to simulate large numbers of scenarios in multi-hazard modeling studies at limited computational expense, while maintaining reliable results.