Verification of QPF Forecasts from the HRRR and RUC

National Weather Service forecasters routinely use quantitative precipitation forecast fields in evaluating model output from various forecast models and in producing forecast products.  The High Resolution Rapid Refresh (HRRR) model and the Rapid Update Cycle (RUC) model are two mesoscale models made available to forecasters on the AWIPS-II workstation, with updated model output available each hour. Precipitation forecasts from selected model runs of the HRRR and RUC were collected during June and July 2016 for a domain covering northeast Colorado. The forecasts were compared to verifying data from the Real-Time Mesoscale Analysis (RTMA) database. Using the Graphical Forecast Editor on the AWIPS-II workstation, a domain of 50 edit areas covering the Front Range foothills and northeast Colorado was defined.  Quantitative precipitation forecasts (QPF) from the HRRR and RUC models, along with quantitative precipitation estimates (QPE) from the RTMA were sampled over the domain of edit areas and collected for statistical analysis. The goal was to assess the accuracy and skill levels of these two models in forecasting the afternoon and evening precipitation regimes common during Colorado summers. Preliminary results show that both models have skill in forecasting the development of convective precipitation with lead times beyond 6 hours and out to 12 or 15 hours.  However, the models have less skill in the placement, timing and intensity of convective precipitation. The data collection and verification schemes developed for this study will be used in an ongoing basis for additional evaluation of model performance and the identification of possible biases in various forecast models.

National Weather Service forecasters routinely use quantitative precipitation forecast fields in evaluating model output from various forecast models and in producing forecast products.  The High Resolution Rapid Refresh (HRRR) model and the Rapid Update Cycle (RUC) model are two mesoscale models made available to forecasters on the AWIPS-II workstation, with updated model output available each hour. Precipitation forecasts from selected model runs of the HRRR and RUC were collected during June and July 2016 for a domain covering northeast Colorado. The forecasts were compared to verifying data from the Real-Time Mesoscale Analysis (RTMA) database. Using the Graphical Forecast Editor on the AWIPS-II workstation, a domain of 50 edit areas covering the Front Range foothills and northeast Colorado was defined.  Quantitative precipitation forecasts (QPF) from the HRRR and RUC models, along with quantitative precipitation estimates (QPE) from the RTMA were sampled over the domain of edit areas and collected for statistical analysis. The goal was to assess the accuracy and skill levels of these two models in forecasting the afternoon and evening precipitation regimes common during Colorado summers. Preliminary results show that both models have skill in forecasting the development of convective precipitation with lead times beyond 6 hours and out to 12 or 15 hours.  However, the models have less skill in the placement, timing and intensity of convective precipitation.  The data collection and verification schemes developed for this study will be used in an ongoing basis for additional evaluation of model performance and the identification of possible biases in various forecast models.