区域业务模式6h降水预报检验方案比较

Comparison of Two Verification Methods for 6 h Precipitation Forecasts of Regional Models

对我国华东、华南、华北区域气象中心和中国气象局数值预报中心业务运行的区域模式2011年5-9月的6 h降水预报,采用不同检验结果平均方案进行对比检验。对比结果表明:不同的检验结果平均方案基本不影响与中国气象局数值预报中心模式(NMC-GRA)在相同区域关于TS评分比较的相对检验结论,即当两个模式评分差距较大时,评分高的模式在两个方案中是一样的,但评分比较接近时,若有一个模式对该区大尺度降水预报较好时,则可能在新方案中有较高的TS评分,而此模式原方案评分则可能略低于局地小尺度降水预报较好的模式。但对于较少发生的强降水预报的预报偏差的评价有很大不同,当新方案的结果显示多数模式对强降水的预报偏少,原方案则可能显示偏多,说明模式对大尺度的强降水预报较实况偏少,但对小尺度局地降水的预报则可能偏多。

6 h precipitation forecast for different lead times of operational models for North China Regional Cen- ter, South China Regional Center, East China Regional Center and Numerical Weather Prediction Center of CMA （NWPC） are verified and compared from May to September in 2011. The two methods which are used to average verification results for some period are discussed in detail. The verification results show the forecasting performance is different according to the observational time, the forecasting lead time and coverage area of models. Compared with NWPC GRAPES_Meso model in the same area, the performance for most lead times of operational models of three regional centers is bet- ter, particularly for over 24 h lead time forecast. Furthermore, verification score of East China Regional Center model is the best, TS of this model is higher than that of NWPC model except that 0.1 mm or more precipitation of 18 h forecast with 0600 UTC observation and 24 h forecast with 1200 UTC observation, and the bias is close to 1 for most verification grades. Moreover, TS of intense precipitation of three Re- gional Centers models is higher than that of NWPC model most of the time, but TS of North China region- al center model is lower for 6 h and 18 h forecast with 0600 UTC observation and for 6 h forecast with 1800 UTC observation. On the other hand, the verification results derived from the different averaging ap- proaches have few impacts on TS comparison results for regional model and NWPC model, but the TS val- ue by new method is higher than that of the former method, except when the difference of TS between the two models is very small. For the new method, the model performs better for the large-scale rainfall process perhaps getting higher TS than the model that is only good at meso-scale rainfall. But the differ- ence for heavy rain with little probability, especially for precipitation heavier than 13 mm is significant. If there is more false alarms for the forecast of greater grades small meso-scale precipitation, and there is some missing for the large scale rainfall, the differences between the two methods will be larger. Under this condition, the bias of the new method shows that forecast rainfall is less obviously comparing with ob- servation, but that of the former method indicates more. Such situation usually occurs in North China. Be- cause the two averaging approaches show distinct differences, more studies are needed so that users can understand the differences better and make a correct decision on how to use the forecasts of models properly.