Objective Precipitation Type Forecast Based on ECMWF Ensemble Prediction Product

Ensemble prediction system usually improves forecast skill compared to deterministic model under the same model system. The European Centre for Medium-Range Weather Forecasts (ECMWF) ensemble prediction system precipitation type product (PTYPE) is used with the approach of optimal probability threshold (OPT) to produce precipitation type deterministic forecast. The history dataset of winter half year (October to the next March) from 2016-2017 of ECMWF ensemble prediction system and observations of 2515 surface weather stations are used firstly to estimate optimal probability thresholds for different lead times of rain, sleet, snow and freezing rain under criteria of Optimal HSS (OHSS), optimal TS (OTS) and Optimal forecast bias (OB). Then deterministic precipitation type forecast is calculated from the probabilistic forecast of ensemble prediction system, verified by data of 2018 winter half year and compared with the precipitation type forecasts of ECMWF high-resolution deterministic model (HRD) and ensemble prediction system control forecast (CF). It indicates that optimal probability thresholds under three criteria are different. However, optimal thresholds of snow and freezing rain are the largest which are between 80% and 40%, and optimal thresholds of sleet are the smallest which are under 10%. They all decrease with elongating lead times. Optimal thresholds of rain are small which are between 7% and 25%, increasing with elongating lead times. Verification results show that performances of CF are the lowest with the proportion correct between 92% and 91% and HSS between 0.74 and 0.55. The performance of HRD is better than CF with the proportion correct about 93% and HSS between 0.77 and 0.65. OPT based on ensemble prediction system probabilistic forecast improves the forecast skill of precipitation type significantly. The improvement of OHSS is the most significant with the proportion correct about 94% and HSS from 0.81 to 0.68. From the verification of every kind of precipitation types and case analysis, it demonstrates that the performance of HRD and CF for sleet is similar. However, for the other three precipitation types, the performance of HRD is better than that of CF and the performance of OPT is the best. For freezing rain, the forecast bias of CF and HRD is larger than 2 which means too more false alarm. OPT reduces the forecast area and false alarm of freezing rain and improves the performance. HRD and CF forecast less rains and more snows with poor forecast biases. OPT corrects these errors and makes better forecast bias and TS scores for rain and snow. For sleet, the forecast bias of OPT is less than 1 significantly and the TS score is nearly zero which is worse than HRD and CF.