不同初始场条件对新疆区域数值模式预报性能的影响

初始场条件直接影响到区域模式的预报性能。基于GRAPES_GFS和NCEP_GFS两种初始场，详细比较了两者之间的差异，随后分别利用两种初始场驱动新疆区域数值模式（RMAPS-CA V1.0），对2021年4月整月的数值预报结果以及2021年4月21日一次暴雨过程模拟结果进行了MET（Model Evaluation Tools）对比检验。结果表明：（1）两种资料位势高度扰动场、温度扰动场、湿度扰动场存在明显差异，其相关系数分别为0.26~0.60、0.05~0.24和0.01~0.12，导致次天气尺度上存在着较大差异，并由此造成了模拟结果之间的差异，反映了区域模式对初始场和边界条件的敏感性；（2）从高空位势高度、风速、温度的预报结果看，NCEP_GFS初始场在新疆区域模式中高空要素的预报效果均要优于GRAPES_GFS初始场，均方根误差分别降低35.5~37.2%、7.6~12.6%和6.0~17.2%。从地面常规预报量的检验看，GRAPE_GFS初始场对2 m温度和10 m风速的预报效果则要优于NCEP_GFS初始场，均方根误差分别降低14.3%和6.8%；（3）从降水检验评分看，两种初始场的降水预报整体为漏报现象，NCEP_GFS初始场针对各降水阈值及不同时效的预报降水评分要高于GRAPES_GFS，0.1 mm/6h、6.1 mm/6h和12.1 mm/6h的TS评分分别提高22.5%、16.1%和150.8%；（4）从一次暴雨过程预报的检验结果看，GRAPES_GFS对于24小时为小量级降水预报效果优于NCEP_GFS,准确率分别为61.4%和40.0%；而NCEP_GFS对于大量级的降水预报则要优于GRAPES_GFS,准确率分别为66.7%和33.3%。两种初始场对降水个例检验偏差以空报现象为主，NCEP_GFS的TS评分整体高于GRAPES_GFS。

The Effect of Different Initial Fields Conditions on the Prediction Performance of Regional Numerical Model in Xinjiang

The initial field conditions directly affect the forecasting effect of the regional model. Based on the two initial fields of GRAPE_GFS and NCEP_GFS, the differences between them were compared in detail, and then the two initial fields were used to drive the Xinjiang regional numerical model (RMAPS-CA V1.0), the numerical simulation results of the whole month in April 2021 and the simulation results of a rainfall process on April 21, 2021 were compared and tested by MET (model evaluation tools). The results showed that, firstly, there are obvious differences in the height disturbance field, temperature disturbance field, and humidity disturbance field between the two kinds of data, with correlation coefficients of 0.26~0.60, 0.05~0.24, and 0.01~0.12, respectively, leading to great differences one sub-synoptic scales and thus causing the differences between simulation results, reflecting the sensitivity of the regional model to initial fields and boundary conditions. Secondly, the prediction effect of NCEP_GFS initial field is better than that of GRAPES_GFS initial field in Xinjiang regional model, and the RMSE (root mean square error) decreases by 35.5~37.2%, 7.6~12.6% and 6.0~17.2%, respectively. The results show that GRAPE_GFS initial field is better than NCEP_GFS initial field in predicting temperature at 2 m and wind speed at 10 m. The RMSE decreases by 14.3% and 6.8%, respectively. Thirdly, from the precipitation test score, it can be seen that the precipitation forecasts for the two initial fields are generally underreported, and the TS (threat score) of the initial field NCEP_GFS is higher than that of GRAPES_GFS for different precipitation thresholds and different timings, and the TS of 0.1 mm/6h, 6.1 mm/6h and 12.1 mm/6h were increased by 22.5%, 16.1% and 150.8%, respectively. Finally, according to the test results of a rainstorm process forecast, GRAPES_GFS is better than NCEP_GFS for 24 hours of small-scale precipitation, with accuracy rates of 61.4% and 40.0%, respectively. However, NCEP_GFS is better than GRAPES_GFS for large-scale precipitation forecasting, with accuracy rates of 66.7% and 33.3%, respectively. The BIAS of two initial fields on precipitation cases is mainly due to the phenomenon of empty reporting. The TS score of NCEP_GFS is higher than that of GRAPES_GFS on the whole.