Predicting Flood Severity for New England Basins with a Semi-Distributed Hydrologic Model

In an effort to modernize the state of practice of flash flood forecasting, recent research has shown promise in utilizing regionalized, continuous, distributed hydrologic models. Additional avenues of refining the forecasting methods have included attempting to forecast event frequencies in lieu of relying on flood magnitudes generated by the models. It is anticipated that this additional post processing of the distributed modeling results can alleviate some modeling errors inherent with trying to represent any natural process. This study examines the application of a regional distributed hydrologic model of lower New England, specifically the results calculated at internal sub-basins by comparing those results to historical gauged data. Then, through frequency transform methods, the forecasting potential is reassessed to determine if frequency predictions can increase confidence in predicting a flash flood. This study further addresses the sensitivity of the spatial scale of the subject catchments as well as the temporal scale in determining the effectiveness of both the distributed model and the frequency prediction. It was found that by post processing the predicted data, the bias in the forecasted events was greatly reduced as compared to the raw output from the modeling. This bias was also sensitive to the resolution of the time step, with the error directly related to that resolution. On the spatial scale, it was shown that the variation in the catchment size did not have a significant impact on the results. Overall, it is shown that there is value to post processing hydrologic modeling results from a continuous, distributed model in order to predict a probability of exceedance as opposed to basing flood warnings on raw flow magnitude calculations