基于因子分析的广东省短时强降水预报模型及其业务试验

利用每天4次0.125°×0.125°的ECMWF-Interim再分析资料和广东省2009—2018年地面气象站逐时雨量观测的短时强降水数据集，针对广东不同季节、不同地域的短时强降水，以提高命中率同时控制虚警率为目的，提出基于显著性和敏感性评价的物理量优选和因子分析法，用于构建分期、分区的广东短时强降水概率预报模型。以参数显著性和预测敏感性为标准，在49个待选物理量中挑选18个既与多年平均态存在明显差异，又具有较低虚警率的物理量，应用方差最大正交旋转因子分析法将遴选物理量组合成表征大气不同环境条件的6个因子；为使组合因子更具适应性，基于因子偏离度特征对广东前、后汛期不同区域独立建模，构建分期、分区短时强降水逐6 h格点概率预报模型。汛期业务试验表明，模型对短时强降水发生概率预报效果较好。对2019年汛期模型每天两次起报的12 h预报时效内概率产品进行格点检验，以训练期最优TS评分对应的固定概率作为预测概率阈值，广东省大部分区域TS评分超过0.25，最高超过0.42，平均较ECMWF-Fine业务模式在前、后汛期分别提升0.23与0.21，南部沿海TS评分提升幅度最大，并且模型在提升命中率与降低虚警率之间取得较好的平衡。个例分析表明，对于ECMWF模式常漏报的广东暖区短时强降水，概率预报模型具有明显优势，尤其能为天气尺度弱动力强迫的强降水早期预警提供更多有效信息。 

A probabilistic forecast model of short-time heavy rainfall in Guangdong province based on factor analysis and its operational experiments

In order to improve the hit ratio and reduce the false alarm ratio of short-time heavy rainfall (SHR) forecasts in Guangdong province, a method that combines physical parameters selection based on significance and sensitivity evaluation with factor analysis was proposed to construct probabilistic forecast model in different periods and regions. The ECMWF-Interim reanalysis of 4 times daily data with 0.125° spatial resolution and hourly gauge rainfall dataset in Guangdong province for the period from 2009 to 2018 were used. On the basis of significance and prediction sensitivity evaluation, 18 physical parameters that obviously deviate from their multiyear averages and may possibly reduce the false alarm ratio were selected out of 49 parameters. The varimax orthogonal rotation method was employed to regroup the selected parameters into 6 factors. These factors respectively reflect different environmental conditions. In order to optimize the model, the factor analysis was separately applied to different regions and the pre-flood and post-flood seasons according to the spatial and temporal features of factor deviations. Based on the weighted combination of factors, the probabilistic grid forecast model of SHR in different regions and periods was constructed for SHR forecasting at 6 h intervals. The forecast model yields impressive results in operational experiments during the flood season. During the period from April to September 2019, grid verification was carried out on twice daily forecasts of the probabilistic model at 12 h lead time. A determined probabilistic threshold corresponding to the optimal TS in the training period is taken as the forecast probability threshold of SHR in individual regions and periods, and the calculated threat score (TS) in most of Guangdong province is above 0.25 and the highest value is 0.42. Compared to the operational ECMWF-Fine precipitation forecast, the average TS of the probabilistic model forecasts increases by 0.23 in pre-flood season and 0.21 in post-flood season, with the greatest improvement in the southern coastal area. Moreover, the model achieves a good balance between increasing the hit ratio and decreasing the false alarm ratio. Cases analysis shows that the probabilistic forecast model has obvious superiority in SHR forecasting in warm sector, which is often missed in the ECMWF-Fine precipitation forecasts. The probabilistic forecast model can provide more valuable information for early warnings of SHR under synoptic conditions with weak dynamic forcing.