江苏省雷达降水估测集合分析

使用2019年、2020年5—8月江苏省降水分析场及站点观测资料,生成具有定量降水估测(Quantitative Precipitation Estimation,QPE)不确定性时间和空间结构的集合QPE,并用观测降水对集合QPE进行了确定性和概率性检验。确定性验证说明集合QPE能在总体上减小降水量的绝对误差和均方根误差,但也加重了某些区域的降水低估。集合平均能提高估测降水的准确率并减小空报率,也会使漏报增多,这使小雨的TS评分有所降低,但各量级降水TS评分仍能保持在较高水平。集合QPE对各量级降水都有较优的Brier评分,降水量级越大,估测效果越好。集合的离散度较小,且将集合成员排序后,观测值落在两头的频率更高,也反映了离散度偏小。此外,观测值大于集合成员最大值的频率更高,说明集合QPE倾向于低估降水。随着概率阈值的增大,集合估测降水发生的命中率(POD)和假警报率(POFD)逐渐增大,但POD增大的程度比POFD大得多,使相对作用特征曲线为折线。不同概率阈值下的POD和POFD体现了集合QPE对各量级降水都有较高的估测技巧,其中对小雨和中雨分辨能力最好。集合估测小雨和特大暴雨发生概率小于实际频率的情况较为严重,而估测的中雨和大雨发生概率与实际降水的发生频率非常接近,有很高的可靠性,但总体上集合QPE仍是倾向于低估降水的发生概率。

Ensemble analysis of radar precipitation estimation in Jiangsu Province

In this paper,an ensemble QPE is generated in order to improve the effect of estimation of the Quantitative Precipitation Estimation (QPE) in Jiangsu Meteorological Observatory,by combining the spatial structure of error uncertainty represented by the error covariance matrix and the temporal structure represented by the time lag correlation coefficient.The ensemble is generated using data from the Jiangsu Meteorological Observatory from May to August 2019 and May to August 2020,with its deterministic and probabilistic components being verified using the corresponding observed precipitation.It is found that the number of components has little effect on the ensemble QPE and that one can control component numbers between 16 and 50.The deterministic verification results show that ensemble QPE aggravates the underestimation of precipitation in some areas,but reduces the absolute error and root mean square error of precipitation in general.Ensemble mean can improve the accuracy and reduce the rate of false positives,and it will also increase the number of true positives.The ensemble has a good Brier score for precipitation of various magnitudes,with larger precipitation magnitudes having a better estimation effect.Also,the dispersion of the ensemble is small,and after the components of the set are sorted,the frequency of observations falling at both ends increases,which also reflects the small dispersion.In addition,the observed values are more frequent than the maximum values of ensemble components,which indicates that the ensemble QPE tends to underestimate precipitation.With the increase of the probability threshold,the hit rate (POD) and false alarm rate (POFD) of ensemble precipitation estimation gradually increase,but the degree of POD increase is much greater than that of POFD,resulting in a broken ROC curve.POD and POFD with different probability thresholds show that ensemble QPE has high estimation skills for all levels of precipitation,with light rain and moderate rain having the best resolution.It is more concerning that the occurrence probability of light rain and heavy rain estimated by the ensemble is less than the actual frequency,whereas the occurrence probability of moderate rain and heavy rain estimated by the ensemble is very close to the actual frequency of precipitation,which has high reliability;however,the ensemble QPE still tends to underestimate the occurrence probability of precipitation.