A Multivariate Statistical Approach for Efficient Compound Flood Hazard Delineation

Flooding in coastal regions is driven by a multitude of factors including mean sea-levels, tides, storm surges, upstream rainfall or river flows, and antecedent soil moisture conditions of the catchment. During many storm events, combinations of various flood drivers interact dynamically and modulate total flood levels, in a process known as compound flooding. Accurate coastal flood hazard delineation requires fully accounting for the interactions of each flood driver in hydrodynamic model frameworks and propagating the impact of their respective probability distributions on overall flood probabilities. However, simulating all possible combinations of storm surge, rainfall, sea-level, etc. in high-resolution hydrodynamic models would be infeasible due to the computational expense. Therefore, there is a need to develop strategies for efficiently estimating compound flood probabilities (i.e., compound flood hazard maps) that can feasibly be applied under current and projected climate and landscape conditions. Here, we propose a statistical framework based on the Joint Probability Method with Optimal Sampling to strategically sample the joint distribution of coastal rainfall, storm surge, mean sea-level, and antecedent soil moisture and select a limited number of hazard scenarios that span the full range of possible combinations. Flooding from the selected hazard scenarios is simulated with a high-resolution coupled hydrodynamic model, and the full probability distribution of compound flooding is reconstructed using a Bayesian statistical approach. This statistical sampling approach can reduce the computational expense of compound hazard delineation by more than ~10x, while maintaining high accuracy compared to expensive Monte Carlo approaches (i.e., in which a large set of naively sampled scenarios is used to estimate compound hazard). We apply our statistical framework to estimate compound flood hazard in the Houston-Galveston region and reveal and the relative flood contributions from rainfall, storm surge, soil moisture conditions, and mean sea-levels. We demonstrate the usefulness of the proposed approach in both estimating overall coastal flood hazard and as a tool for understanding the dynamics of compound flooding in low-gradient catchments.