Verification of Precipitation Forecasts from Two Numerical Weather Prediction Models for the Middle- and North-Eastern Region of the USA

Accurate precipitation and temperature forecasts are the pre-requisites to produce skillful flood forecasts. The structures of different precipitation forecasting systems are constantly evolving, for example, with improvements in the forecasting techniques, increases in spatial and temporal resolution, improvements in model physics and numerical techniques, and better understanding of uncertainty. Hence, routine verification is necessary to understand the quality of forecasts at particular times and locations, and as inputs to hydrologic modeling. Hydrologic forecasters in the National Weather Service are evaluating precipitation and temperature forecasts from a wide range of numerical prediction models to improve the streamflow forecasts. To assist in this effort, our goal here is to verify the operational precipitation forecasts from the National Centers for Environmental Prediction 21-member Short Range Ensemble Forecast (SREF) system, together with precipitation reforecasts from the 11-member Global Ensemble Forecast System (GEFS). The verification is done for the middle- and north-eastern region of the United States and for mean areal precipitation forecasts conditioned on precipitation amounts, lead times, seasonality, and accumulation periods. Multi-sensor precipitation estimates are used as observed data. The effect of different basin sizes on forecast quality is also studied by simply choosing areal extents of varying sizes. Although flood forecasting is the main context of this study, separate analyses are presented for moderate and large precipitation events with a view towards providing additional information to forecasters. The summary of verification statistics indicates similar forecasting performance for both SREF and GEFS reforecasts even though GEFS reforecasts are valid for much longer lead times and coarser grid resolution. Precipitation forecasts from both of these models show greater skill in large basins than in relatively smaller ones. Lastly, we identify and highlight the potential of the verified precipitation forecasts as forcing inputs to hydrologic modeling.