基于降水遥相关型的夏季降水模式预测改进研究

Study on the Improvement of Summer Precipitation Model Prediction Based on Precipitation Teleconnection

我国月—季降水分布在空间上存在类似于大气环流遥相关的空间遥相关型。本文基于中国近57年夏季降水资料研究了我国夏季降水空间遥相关型的主要空间模态特征及其年代际变化,评估并改进了BCC_CSM模式、ECMWF_SYSTEM4模式以及NCEP_CFSV2模式对中国夏季降水的预测能力。研究结果显示,中国夏季降水实况中存在华北—长江下游、华东—中国中北部、华南—长江流域、西南—东北中部等4类显著的空间遥相关型。动力模式可以预测大尺度的降水分布,而对于不同区域之间降水遥相关这种雨带细节特征的预测能力则较为薄弱,存在着较多虚假相关。为改善模式降水预测技巧,以实况中的降水遥相关型作为约束条件构建了修正方案,以此来修正模式中的降水遥相关型分布。结果显示,经过修正能够有效地改善模式对东北中部、长江下游的预测能力,4年的回报检验结果显示,模式预测的平均距平一致率从47%提高为58%;平均均方根误差从153 mm减小为120 mm;平均趋势异常综合检验(PS)评分从64提高为73。

Spatial distributions of the monthly–seasonal precipitation in China are similar to the atmospheric circulation teleconnections.Based on the summer precipitation data of China in recent 57 years,this paper investigates the main spatial mode characteristics and interdecadal changes of the spatial teleconnection pattern of the summer precipitation in China.Further,this study evaluates and improves the prediction ability of BCC_CSM,ECMWF_SYSTEM4,and NCEP_CFSV2 climate models for the summer precipitation in China.Results show that there are four significant spatial teleconnection patterns of summer precipitation in China:(1)North China:lower reaches of the Yangtze River,(2)East China:Central and Northern China,(3)South China:Yangtze River basin,and(4)Southwest China:Central Northeast China.The climate models can only predict the large-scale precipitation distributions;however,the precipitation remote correlation between different regions cannot be well predicted,and there are many false teleconnections.To improve the precipitation prediction ability of the model,a correction scheme is constructed by taking the precipitation teleconnection pattern in the actual situation as the constraint condition to correct the precipitation teleconnection pattern’s distribution in the model.Results show that the correction can effectively improve the model’s ability for predicting the precipitation of Central Northeast China and the lower reaches of the Yangtze River.Four-year return test results show that the average anomaly consistency rate of model prediction increased from 47%to 58%,the root mean square error decreased from 153 to 120 mm,and the average PS(Prediction Score)score increased from 64 to 73.