A Methodology for Developing Probabilistic River Flood Inundation Maps

To address the uncertainties of river flood forecasting and to contribute a new visualization for flood mapping, this study outlines a methodology for creating dynamic probabilistic flood inundation maps for a floodplain which currently lacks dynamic flood maps of any kind. To create these maps, a hydrologic model is specially constructed using HEC-RAS for a segment of the Tonawanda Creek in western New York. The model is constructed using pre-existing, publically available data, including a LiDAR digital elevation model, real-time discharge observations from a USGS-maintained gauge on the creek, and other characteristics of the creek's bathymetry derived from previous flood insurance studies. The model is then used to create inundation rasters based on selected initial discharges.

To determine probabilities, predicted peak discharges for the Tonawanda Creek are first collected from the Meteorological Model-based Ensemble Forecast System (MMEFS), which generates predictions of various hydrometeorological parameters in real-time, and are used to create a suite of inundation rasters, one for each individual peak discharge prediction. These inundation rasters are then overlaid and the probabilities are calculated by grid cell based on how many inundations overlap. The resulting map depicts the range of flooding extent probabilities based on the real-time forecast. An alternative probabilistic map is also devised to depict the depth of flooding given a flood event's probability. The 10th, 50th, and 90th percentiles are calculated for the forecast ensemble's peak discharge predictions. These percentiles correspond to benchmarks of flooding likelihood, such that the 10th percentile indicates a 90% likely flood event, 50th percentile a 50% event, and 90th percentile a 10% event. Once completed, the maps are qualitatively evaluated for potential applications in predicting flood events and mitigating flood-induced damages.