Coastal Urban Flood Driver Decomposition via a Novel Ensemble Modeling Framework

Abstract Submission: Flood-predictive storm event characteristics are uncertain and spatially varied, especially in complex watersheds with multiple possible hydrologic drivers. Traditional flood mitigation designs rely on a limited set of synthesized storm events, often overlooking uncertainties in flood-inducing conditions. To account for this uncertainty and improve flood mitigation design, we developed and tested an ensemble-based framework for flood probability mapping that also enables a deeper understanding of how rainfall and storm surge contribute individually and jointly to flooding in a 1.9 km2 coastal urban watershed in Norfolk, Virginia.

First, we generated 1,000-years of potentially flood-inducing compound storm events accounting for relationships between rainfall and storm surge event characteristics. Each event was simulated using a novel 1D/2D hydrodynamic model, TRITON-SWMM, under three scenarios: 1) rainfall-only with median boundary condition water levels; 2) variable water levels (tidal patterns plus storm surge) with no rainfall; and 3) combined rainfall and variable water levels. We then decomposed the contributions of rainfall, variable water level boundary conditions, and their interaction to peak flood depths in each grid cell and event by comparing results from scenarios one and two with scenario three. We also derived empirical depth-exceedance probability curves for each grid cell using scenario three results. Combining these two analyses allowed us to quantify key flood-inducing event characteristics and flood attribution by watershed location and probability.

By revealing the hydrologic sources and event characteristics causing flooding by location and return period, our framework offers actionable insights for enhancing flood mitigation infrastructure.

Learning Objectives/Expected Outcome (Optional) :