本文基于1961-2018年中国2400余个国家气象站点的观测数据插值得到的逐日降水数据，利用回归分析、功率谱分析等统计方法，总结归纳了华南地区暴雨与区域性暴雨的变化趋势和周期特征，并选取近年四个典型的区域性暴雨过程，分析了业务气候模式对此类强降水过程的预报能力，探究了模式集合对区域性暴雨过程的预报水平。主要结论包括：

1961-2018年，广西北部地区和华南沿海地区年平均的暴雨日数与暴雨雨量最多，夏季华南地区总的暴雨日数与暴雨雨量最多，其次是春季。全年的暴雨雨量、暴雨日数和暴雨强度在广西北部地区、广东南部地区、福建省和海南省明显增加和增强，四季中，夏季华南地区平均值的增长速率最大，之后是秋季。1月-12月，58年间平均的区域性暴雨与大暴雨日数、过程数呈单峰型分布，其中区域性暴雨在5月-7月频发，而区域性大暴雨主要出现在6月-8月。全年华南地区平均的区域性暴雨日数与过程数分别为28 days和17 times，还分别以0.15 days/a和0.097 times/a的速率增加，四季均呈现出增加的趋势，其中夏季增加速率最快，秋季最慢。区域性大暴雨日数与过程数除了在春季减少外，全年和其他季节均显著增加。单次过程最大和平均持续日数在春季减少，而冬季以0.015 days/a的速率增加。全年、冬季和夏季的最大单次过程综合强度指数增强，特别是冬季。周期性结果显示，全年和四季中华南地区的暴雨与区域性暴雨表现出不同程度的准3 a、准8 a、准14 a和准18 a的周期特征，全年的暴雨与区域性暴雨在21世纪基本呈现出显著的准3 a和准18 a的振荡周期。

与观测值相比，CMA、KMA、NCEP和UKMO模式均能够较好地预报出华南区域性暴雨过程大致的降水区域，但是预报的累积降水量均低于观测值。区域性暴雨过程的局地性越强、降水量越大，模式预报结果与观测值的差异越大，难以准确预报降水中心的累积雨量。随着预报时效的增长，模式的预报水平降低，当预报时效超过15天时，模式基本不能预报出华南区域性暴雨过程发生的情况。四个模式中，NCEP模式对华南区域性暴雨过程的预报能力最优。将四个模式的预报结果进行集合，通过等权重平均集合方法得到的预报结果削弱了部分模式的优势，而CMA、KMA、NCEP和UKMO模式以权重系数0.346、 0.415、 0.423 和 0.401进行不等权重集合后预报的降雨落区更接近于观测值，同时还缩小了模式预报的降雨量与观测值间的差异。NCEP/CFSv2模式每日可进行四次预报，单个时次的预报存在不稳定性，通过等权重平均和不等权重平均集合后的预报结果相近，二者的预报水平优于单个时次的预报。

Based on the daily precipitation data interpolated from more than 2400 national meteorological stations in China during 1961-2018, the spatial and temporal variation characteristics in rainstorms and regional rainstorms in South China are studied. By using statistical methods such as linear regression and power spectrum analysis, the interannual variation regularity and periodic characteristics of rainstorms in South China are explored. In addition, the prediction abilities of the operational climate models for regional rainstorm processes in South China are analyzed, and the prediction abilities of ensemble models are explored. The main conclusions are as follows:

From 1961 to 2018, the numbers of rainstorm days and the amounts of rainstorm rainfall in coastal areas of South China, as well as northern Guangxi Province are significantly higher than those in other areas in South China. The numbers of rainstorm days and the amounts of rainfall in the whole South China reach the maximum in summer, and followed by spring. The numbers of annual rainstorm days, rainfall and intensity increased most significantly in northern Guangxi, southern Guangdong, Fujian and Hainan Provinces. The average increase rates in South China reach the maximum in summer, and followed by autumn. From January to December, the numbers of regional rainstorm days and processes, as well as regional heavy rainstorm, show single-peak patterns. The regional rainstorms occur frequently from May to July, while the regional heavy rainstorms mainly occur from June to August. During 1961-2018, the average numbers of regional rainstorm days and regional rainstorm processes in South China are 28 days and 17 times, respectively. The annual increase rates of regional rainstorm days and processes are 0.15 days/a and 0.097 times/a, respectively. And both of them show upward trends in all seasons, which with the largest trend in summer and the smallest trend in autumn. The numbers of regional heavy rainstorm days and processes increase significantly in annual and other seasons except in spring. The maximum and average single process duration decrease in spring, but increase significantly in winter at the rate of 0.015 days/a. The maximums of comprehensive intensity index of regional rainstorm processes show an upward trend in annual, winter and summer, and the increasing rate is the largest in winter. The results of periodic analysis show that the annual and seasonal heavy rainstorms and regional heavy rainstorms in South China show quasi-3a, quasi-8a, quasi-14a and quasi-18a cycles to varying degrees. Annual rainstorms and regional rainstorms show significant quasi-18a and quasi-3a oscillation cycles in the 21st century.

Compared with the observed values, CMA, KMA, NCEP and UKMO models can better forecast the rainfall areas, but the predicted cumulative precipitation amounts are lower than the observed values. If the rainfall area is more concentrated and the precipitation amount is greater, it is difficult to accurately forecast cumulative precipitation in rainfall center. The prediction abilities decrease as the lead times increase, and when the lead times are greater than 15 days, the models cannot predict the occurrences of regional rainstorms. Among the four models, NCEP/CFSv2 has the best prediction ability for rainstorm process in South China. The prediction results that are obtained by the four models by using the equal-weight mean integration method are not good, because the advantages of some models are weakened. However, the prediction results that are obtained by the CMA, KMA, NCEP and UKMO models, which have different weight coefficients of 0.346, 0.415, 0.423 and 0.401, respectively, are not only closer to the observed values in rainfall areas, but also reduce the biases between the predicted precipitation amounts and the observations. The NCEP/CFSv2 model can make real-time forecast four times a day, and the single time forecast at 12, 18, 00 and 06 UTC is unstable. The results of the integration of equal-weight average and unequal-weight average are similar, and the prediction results of the two methods are better than the prediction at a single time.