春季中国日最高气温延伸期预报误差分析及订正

ANALYZING AND CALIBRATING EXTENDED-RANGE FORECAST OF CHINA’S DAILY MAXIMUM TEMPERATURE IN SPRING

数值模式直接输出和经模式后处理得到的预报误差比较,是延伸期逐日要素预报应用基础。针对中国2 583个站点在2020年春季11~30天的日最高温度预报,根据欧洲数值中心的集合预报输出,首先,使用BP-SM(Back-Propagation-Self memory)法和回归法,进行确定性预报订正效果比较;结果表明BP-SM法和回归法都明显降低了预报绝对误差;在11~14天预报中,BP-SM法得到的平均绝对误差为3.3~3.6oC,预报准确率超过35%,订正效果更优。其次,基于模式直接输出和BP-SM法获得的概率预报,使用CRPSS (continuous ranked probability skill score)进行了可预报性分析。结果表明,在地形复杂地区,经过订正,预报准确率明显改善。对于延伸期逐日要素预报,合理的模式后处理方法是降低预报误差和提高预报能力的重要环节。

The comparison between direct model output(DMO)and post-processing product is the key to improving extended-range forecast.Based on the ECMWF ensemble forecasts and the daily maximum temperature forecasts by 2583 stations in China for 11~30 days in spring 2020,the present study uses back-propagation-self memory(BP-SM)method and regression method to compare the effect of deterministic forecast correction.The result shows that the absolute forecast error through the postprocessing method is less than that of the DMO.In the 11~14 d forecast,BP-SM method(absolute error:3.3~3.6℃;accuracy:35%)performs better than linear regression.Then,using the continuous ranked probability skill score,predictability is discussed.The result shows that in areas with complex terrain,the predictability is improved significantly through the BP-SM method.For the 11~30 d daily maximum temperature forecast,the post-processing method plays an important role in reducing forecast errors and improving forecasting capabilities.