Compounding factors for extreme flooding around Galveston Bay during Hurricane Harvey

Coastal hazard is rarely driven by only one source, as exemplified by the compound flooding from Hurricane Harvey in Galveston Bay in 2017. A 3D creek-to-ocean model is developed to explicitly resolve, without grid nesting, the marine (combination of atmospheric forcing and tides), fluvial and pluvial extremes for this extreme event. We first thoroughly assess the model skills using all available observations in the Galveston Bay region, including High Water Marks (HWMs) and field estimates of maximum inundation extent in the watershed. Subject to uncertainties in the river flows, atmospheric forcing, initial condition of salinity and temperature, and digital elevation model of bathymetry-topography, the model is shown to generally exhibit good skills for predicting inundation and compound surges, with a hit rate for inundation extent of 0.92, average mean-absolute-errors of 0.65 m for HWMs, 1.7 psu for salinity, and 1.4 °C for temperature. We then apply the model to quantify the individual contributions from the three major forcings (ocean, river and precipitation). Comparison of results (in the form of ‘compound ratio’) from the simulations with three factors being applied individually with those from the baseline simulation with all factors included in a single model reveals the nonlinear compounding effects in most of the areas in Galveston Bay, and indicates that the compound flooding problems are best simulated using a single model that integrates across all factors because the interactions among processes are very complex and highly nonlinear; in other words, summing up the results from individual forcings would lead to large errors. For example, the hydrodynamic model results forced by river inflows at boundary and oceanic and atmospheric forcings, without explicitly accounting for the direct precipitation in the coastal watersheds, would severely underestimate the resultant flow and surge by up to 90%. ‘Regions of dominance’ are also identified for each forcing factor from the sensitivity results. These concepts are applicable to other compound flooding studies as well.