考虑台风事件影响的日降水预报检验与订正

Validation and calibration of daily precipitation forecast considering the influence of typhoon events

【目的】台风带来的强降水常给当地带来洪涝等灾害,在台风事件下提供准确可靠的降水预报是科学研究和业务预报的重要工作。【方法】通过台风路径缓冲区筛选台风场次,检验台风事件和非台风事件分布情况;对美国国家环境预报中心提供的GEFSv12逐日原始降水预报检验数据精度情况,评价原始预报技巧;应用伯努利-伽马-高斯模型对GEFSv12降水预报,在我国粤港澳大湾区格点,设计不同订正方案,在不同降水场景下,评估不同情况下模型订正前后预报性能。【结果】影响我国东南沿海台风登陆时间主要集中于7—9月,台风等级集中在14级以前;台风事件下,利用台风事件样本训练模型订正后预报BIAS接近0,排序概率技巧得分CRPSS平均提高了近15%,α-index相对原始预报提升了约0.16。【结论】伯努利-伽马-高斯模型能有效订正原始预报存在的系统偏差,提高预报的准确性;不同降水场景订正效果有所区别,对出现极值降水较少的非台风事件,订正后效果有更好的表现;台风事件下,利用台风样本训练模型订正效果相对较好,提高了原始预报精度技巧、可靠性,减少预报偏差;订正的效果和相关系数大小有关,相关系数表现好的区域有更好的预报精度技巧得分,经订正后有利于后续水文集合预报应用、模型发展和备灾应灾工作。

[Objective]The heavy precipitation caused by typhoons often leads to floods and other disasters in local areas.[Methods]Typhoon events were screened through the typhoon path buffer zone to check the distribution of typhoon and non-typhoon events.The accuracy of raw daily precipitation forecast data from the GEFSv12,provided by the National Environment Forecasting Center,was evaluated.By designing various calibration schemes,the Bernoulli-Gamma-Gaussian model was employed to calibrate the GEFSv12 precipitation forecast in the Guangdong-Hong Kong-Macao Greater Bay Area.This calibration was performed under different scenarios to assess forecast skill improvement before and after calibration under varying precipitation conditions.[Results]The landfall time of typhoons affecting the southeast coast of China is mainly concentrated from July to September,with typhoon intensities predominantly below grade 14.In the case of typhoon events,the forecast BIAS approached 0 after using typhoon events samples to train the model,result ing in an average increase of nearly 15%in the continuous ranked probability skill score and the alpha-index increased of nearly 0.16.[Conclusion]The Bernoulli-Gamma-Gaussian model can effectively calibrate the systematic deviation of the raw forecast and improve predictive performance.Calibration effects vary across different scenarios.For non-typhoon events with less extreme precipitation,the calibration effect is more pronounced.In the case of typhoon events,training the model with typhoon samples consistently outperforms the other two models,enhancing reliability and forecast skill.The calibration′s impact correlates with the value of the correlation coefficient;areas with higher correlation coefficient exhibit superior prediction skill scores.This attribute bodes for subsequent application in hydrological ensemble prediction and post-calibration model enhancements.