极端降水事件概率分布识别方法对比研究

Comparisons on probability distributions of extreme precipitation events identified by different methods

识别极端降水的概率分布对于区域水旱灾害的预测防治具有重要意义。研究针对广泛采用的L-矩分析、KS检验和卡方检验3种方法,基于四川盆地24个气象站1951—2011年的日数据,对最大日降水量、连续最长无雨天数的概率分布分别进行了识别和误差分析。结果表明:KS检验结果的相对误差最小且大都小于0.05,并且与基于卡方检验的结果接近;基于L-矩分析识别结果的误差较大,尤其是连续最长无雨天数的平均误差在多个回归水平下超过10%;此外,具有相同概率分布的站点,基于KS和卡方检验结果的空间连续性更好。上述结果和一些研究优先推荐L-矩分析识别水文序列概率分布的结论不一致,原因是这些研究主要针对径流而非最极端的降水情况,且降水序列的空间异质性大。

Identifying the probability distribution of extreme precipitation events is crucial to regional floods and droughts prediction and prevention. In this study, three widely used methods （L-momentum analysis, KS test and Chi-Square test） are compared by test the probability distribution of maximum rainfall in a day, Maxl d, and maximum consecutive drought days, Max CDD, on the basis of daily data from 1951-2011. The results reveal that the KS test exhibits the minimum relative error of less than 0.05 in most situations, Chi-sqnare test shows close level. L-momentum analysis displays the largest error, especially in Max CDD which is more than 0.1 in several return periods. On the other hand, those stations showing the same probability distribution exhibit a better spatial continu- ity when the results are based on the KS test and Chi-square test. These results also reflect that KS test and Chisquare test are more appropriate than L-momentum analysis when identify the probability distribution of extreme precipitation events. However, this is different from the conclusions of some literatures, which recommend L-momentum analysis on testing the probability of hydrological time series. The main reason is that those studies are mostly based on runoff instead of extreme precipitation, and the precipitation time series also exhibits substantial spatial heterogeneity.