汛期分期变点分析方法的原理及验证

Change Point Analysis Methods for Flood Seasonality Determination

目前常用的汛期分期方法存在一定的主观性,划分汛期也不够精确,为此提出了用于汛期分期计算的3种变点分析模型:基于日最大值取样的均值变点分析模型、基于超定量取样的概率变点分析模型以及基于年最大值取样的概率变点分析模型。应用统计试验方法对这3种汛期分期模型进行效率评价与比较,结果表明:3种方法都可以客观、精确地进行汛期分期计算,在一定程度上避免了分期中的主观随意性;其中以基于超定量取样的概率变点分析模型具有较好的无偏性和有效性,最适宜在汛期分期计算中应用。

Since at present the methods of flood season-ality determination are subjective and inaccurate, this paper focuses on comparing and evaluating three models for determination of flood seasonality. The change point analysis theory is used to divide the flood period into several segmentations through three different sampling methods, which include daily maximal （DM）, annual maximal （AM） and the peaks-over-threshold （POT） models. Monte Carlo experiments show that the change point analysis is an objective method and can obtain rational and accurate flood seasonality. It＇s also shown that change point analysis with the POT sampling method can obtain best effect with the least bias and highest efficiency among these models.