Surrogate flood models for compound flood risk assessments and early warning

Probabilistic flood risk assessments (PFRA) and flood early warning systems (FEWS) are essential tools for managing and responding to disastrous flood events, particularly in coastal deltas where flooding is often compound, resulting from the interplay of coastal and riverine water levels and local rainfall. PFRA and FEWS require assessing the compound flood hazard under a broad range of plausible or forecasted hydro-meteorological conditions. While efficient hydrodynamic models for compound flooding have been developed, such as SFINCS (Leijnse et al. 2021; van Ordmondt et al. 2024), trade-offs in the model resolution or number of simulations or stochastic variables are often required for PFRA and FEWS, at the cost of model accuracy.

The physics-guided  hybrid  LSG model (Fraehr et al. 2022, 2023) uses a Sparse Gaussian Process model trained on Empirical Orthogonal Functions (EOF) derived from simulations with a high- and low- resolution hydrodynamic model. For new events to simulate, the approach combines a simulation of the low-resolution hydrodynamic model with the trained surrogate model, to predict high-resolution water depths at low computational costs. While this model has successfully been applied for riverine flooding, it has not yet been used to predict compound flooding from multiple drivers.

This study tests the surrogate SFINCS-LSG model for compound PFRA and FEWS. We investigate the optimal choice of events to train the model and test the model for case studies in Brisbane, Australia and Charleston (NC), USA. We validate the surrogate model against the high-resolution SFINCS model for different historical compound events, hypothetical compound flood scenarios, and compound PFRA.

Based on preliminary results, we find that compared to the course-resolution SFINCS model, the surrogate model provides a significant improvement at very low computation costs, while compared to the high-resolution SFINCS model it achieves a large speedup with only a small drop in accuracy. While the results are promising for individual simulations, the surrogate model struggles to capture the transition zone based on the difference between model simulations. Nonetheless, the surrogate SFINCS-LSG models seem a promising approach to improve compound PFRA and FEWS.

References

Fraehr et al. (2022). Upskilling low-fidelity hydrodynamic models of flood inundation through Spatial analysis and Gaussian Process learning. WRR, https://doi.org/10.1029/2022WR032248

Fraehr et al (2023). Development of a fast and accurate hybrid model for floodplain inundation simulations. WRR, https://doi.org/10.1029/2022WR033836

Leijnse et al. (2021). Modeling compound flooding in coastal systems using a computationally efficient reduced-physics solver: Including fluvial, pluvial, tidal, wind- and wave-driven processes. Coastal Engineering, https://doi.org/10.1016/j.coastaleng.2020.103796

van Ormondt et al. (2024). A subgrid method for the linear inertial equations of a compound flood model, EGUsphere [preprint], https://doi.org/10.5194/egusphere-2024-1839