Verification of precipitation forecasts from National Weather Service high resolution forecast models
An evaluation of the High Resolution Rapid Refresh (HRRR) model and the Weather, Research, and Forecasting (WRF) model was conducted for the NWS in Melbourne, Florida. The models were verified through the radar precipitation product – One Hour Accumulation (OHA), which is an accumulation, in inches, over the past hour, and updated each radar scan.Three international airport locations were studied from 26 May through 8 July 2015 including: Orlando, Melbourne, and Tampa. The model cycles examined are the 0900Z, 1200Z, 1500Z, and 1800Z. Within each run, the three, six, and nine hour forecasts are compared to radar precipitation estimates. Both 5 and 10 nautical-mile range rings are defined for each location. For each forecast time, the following is recorded: whether or not there was an event, event coverage and maximum precipitation within the range rings. A standard two by two contingency matrix is used to classify the precipitation events (Table 3). Model performance was compared using various skill scores including: Probability of Detection, False Alarm Ratio, Critical Success Index, Percent Correct, and Bias. It was found that, overall, the WRF performed best, with the highest probability of detection, lowest false alarm ratio, higher critical success index, higher percent correct, and reduced bias. This preliminary work suggests that, in terms of local precipitation and short term forecasts, the local WRF may have an advantage over the HRRR when forecasting convective rainfall for the Central Florida region.