Developing a Real-Time Probabilistic Riverine Flood Inundation Map for the Tonawanda Creek in Western New York

Deterministic flood inundation maps have limitations, because deterministic projections do not directly express many of the uncertainties or probabilities intrinsic to flood forecasting. Users of these maps must make decisions as though inundated terrain, as indicated on the map, is a certainty; an assumption that may not verify. A probabilistic inundation map can communicate uncertainties within flood forecasting and may assist decision-makers to quantify the chances of flooding relative to the risks and costs associated with mitigation efforts.

For this study, a real-time probabilistic inundation map was created for a reach of the Tonawanda Creek, in Erie County, NY, using the US Army Corps of Engineers' HEC-RAS river modeling software. Input parameters such as elevation were determined remotely from a digital elevation model using HEC-GeoRAS in ArcMap. Once constructed, the model was statistically verified relative to the observed output of three historical crests, as determined from archival imagery and discharge records, following procedures from previous studies.

To assess flooding in real-time, the discharge maximum of the individual model runs of the Meteorological Model-based Ensemble Forecast System (MMEFS) were used for the initial discharge in the HEC-RAS model. Inundation output from each of the HEC-RAS runs using MMEFS data were then compiled in ArcMap. The probability of flooding was assessed by reclassifying the overlapping geographic extents of each modeled inundation polygon such that if, for example, 3 out of 10 inundation polygons overlap at a specific gridpoint, the probability of flooding at that location is 30%.

While the true accuracy of the HEC-RAS model for the Tonawanda Creek is yet to be quantified, it is likely that errors intrinsic to digital elevation models and other inaccuracies will compound to adversely affect the accuracy of the output. Despite this however, the versatility of this methodology, especially using remotely-sensed input parameters, allows for inundation maps to be created for more locations than with existing expensive map making procedures.