中国年最大致洪暴雨落区随机性描述与预测

Stochastic Description and Prediction of Flood-Causing Annual Maximum Storm Landing Area over China

鉴于中国年最大致洪暴雨落区的随机性，利用1900～2010年全国最大致洪暴雨落区分布资料，结合中国暴雨区划图并进行编号，得出每年暴雨落区的编号，从而形成年最大致洪暴雨落区时间序列。通过对该时间序列进行自相关、趋势、周期等一系列分析，发现该序列为独立序列，序列整体呈减小趋势，且通过了Kendall秩次相关检验，序列存在9年左右的第一主周期、5年左右的第二主周期和30年左右的第三主周期，可见该序列具有一定的规律性和可预测性。利用改进的BP网络方法对该时间序列进行模拟与预测，发现模拟和预测精度均较好，表明中国最大致洪暴雨落区具有一定的可模拟性和可预测性。

It is a random process where the flood causing annual maximum storm （FCAMS） landed in China. Using the 111 years data from 1900 to 2010 and combining with Chinese storm division and numbering it, the numbers of FCAMS landing area are obtained. And then the time series of FCAMS is established. By analyzing the self-correlation, trends, and period of the time series, it is found that the time series is an independent series, which appeared downward trend and passed the Kendall rank related inspection. There are some periods in the time series, such as the first primary period about 9 years, the second primary period about 5 years and the third primary period about 30 years. It shows that the time series is regular and predictable. Improved BP neural network method is used to simulate and forecast the time series and it has high precision, the results show that the FCAMS over China is regular and predictable.