Comparative Analysis on Precipitation Forecasting Capabilities of Two Ensemble Prediction Systems Around Qinling Area

Using precipitation forecast data of ECMWF and NCEP ensemble prediction systems, hourly rainfall data fusion by CMORPH (NOAA Climate Prediction Center Morphing Method) satellites and 30000 automatic weather stations, the precipitation and precipitation probability forecasting capability of ECMWF and NCEP ensemble prediction systems around Qinling area are comparatively analyzed from June to October in 2013 and 2014, mainly based on classic skill score and ROC (relative operating characteristic) statistical method. Results show that the precipitation spatial distribution pattern can be better described by both ECMWF and NCEP ensemble prediction systems with the disadvantages that forecasted high value center is larger and the precipitation amplitude is small; the correlation coefficient of ECMWF control forecast and perturb forecast with observations is higher, the standard deviation ratio is close to 1.0 in previous 10 days, which is better than NCEP, but NCEP forecast skill score has better performance than ECMWF in 264-360 hours.The ensemble mean skill score for heavy rain of ECMWF ensemble prediction system is better than NCEP in 0-120 hours. The forecast skill score of ensemble mean is lower than control forecast and perturb forecast for both two systems. Ensemble mean significantly reduces the standard deviation of precipitation amplitude and this is not conducive to the accuracy of synoptic scale precipitation prediction. Ensemble mean significantly increases forecast bias of light rain and increases the false rate, while the forecast bias of heavy rain and the fail rate decrease. This phenomenon is more remarkable when ensemble mean contains more perturb members and forecast skill is roughly the same between different members, and this makes ECMWF ensemble mean skill scores for light rain lower than NCEP.Overall, ECMWF probability forecast effect is better than NCEP. When precipitation threshold increases, BS score of both two models increases sharply while the forecast capacity significantly reduces, for storms, the ROC area is smaller than climate probability sometimes. The ROC analysis show that as the forecast probability improves, ECMWF ensemble prediction system slightly decreases the hit rate and significantly reduces the false rate, however, NCEP have a high false rate and higher hit rate. So depending on user's requirements, different model can be chosen as reference.