四种定量降水预报客观订正方法对比研究

A comparative study of four objective quantitative precipitation forecast calibration methods

中文关键词:  频率匹配法  最优TS评分  最优百分位  概率匹配  定量降水预报  偏差订正

英文关键词:Frequency matching method  Optimal threat score  Optimal percentile  Probability matching  Quantitative precipitation forecast  Bias correction

中文摘要:

      基于2019年全年、不同季节、不同预报时效的欧洲中期天气预报中心模式的定量降水预报，检验评估了频率匹配、最优TS评分、最优百分位、概率匹配4种定量降水预报客观订正法的综合性能。利用理想模型研究了不同雨带位移偏差和干湿偏差情形下频率匹配法与最优TS评分的表现，并通过个例订正展示了4种定量降水预报订正法的基本特征。结果表明：频率匹配与最优TS评分仅能对确定性预报的降水量级进行调整，最优百分位和概率匹配法通过引入集合预报信息可在一定程度上改变预报的降水落区。频率匹配法以频率偏差最优为目标，可以很好地消除模式的干湿偏差，但仅在位移偏差较小且存在较大干湿偏差时提升原始预报的TS评分。最优TS评分法难以改进存在弱湿偏差的中雨预报的TS评分，而最优百分位法利用集合预报信息可以显著提升所有降水等级的TS评分，在较长预报时效下优势尤其明显，但也存在春、夏两季湿偏差较大的问题。概率匹配法由于没有使用历史实况信息，在暴雨订正中干偏差较大。经济价值模型检验评估表明，最优百分位法在暴雨量级的风险决策中具有较高的参考价值。

英文摘要:

      Four objective quantitative precipitation forecast (QPF) calibration methods, including frequency matching method (FMM), optimal threat score (OTS), optimal percentile (OP) and probability matching (PM), are comprehensively verified based on annual and seasonal ECMWF QPFs at different forecast lead times. An ideal model is proposed to study the performance of FMM and OTS under different scenarios of spatial displacement and dry/wet biases. A heavy rain case is used to demonstrate basic characteristics of the four different calibration methods. Results show that FMM and OTS can only adjust the magnitude of deterministic QPF, while OP and PM can change the pattern of QPF to some extent by using ensemble forecast information. Aiming at optimizing the frequency bias, FMM can well eliminate the dry/wet bias of QPFs, but it can only improve the threat score (TS) of original QPFs when the displacement error is small and the dry/wet bias is large. OTS has limited skills in improving the TS of moderate rain with weak wet bias. By contrast, OP can improve the TS of all precipitation thresholds, benefiting from using ensemble forecast information, especially for longer forecast lead times. However, OP shows large wet biases during spring and summer seasons, while PM suffers from large dry biases for torrential rain events due to the lack of historical observation information. The evaluation of economic value model shows that OP has relatively higher reference value for torrential rain in risk decision making.