傳統上,非常態水文序列須經過轉換為常 態水文序列,或經由判定水文序列擾動值之分 布,方能以時間序列模式繁衍水文合成資料。 然而由於資料轉換失真或擾動值之分布誤判, 同時,傳統上合成資料之產生以能保存資料之 樣本統計特性為依據,然樣本統計特性與理論 統計特性可能存在差異,故所繁衍之合成資料 將不具資料之理論特性,因此本論文提出一種 方法,利用Bootstrap resampling之抽樣方法,配合ARIMA 模式可直接繁衍未知分布之水文合成資料,而 資料不須轉換為常態。研究結果顯示,本論文 所提方法可保存資料之理論統計特性,而非僅 保存該組資料之樣本統計特性。 One of the conventional approaches to analyze the non- Gaussian time series is to transform it to a Gaussian time series or to detect the distribution of the noise. The synthetic data are then generated based on the time series model. However, the statistical characteristics of time series may be distorted by using the transformation approach and by drawing a wrong inference of the distribution of noise from the data. Meanwhile, the synthetic data were generated to preserve the statistical characteristics of the samples. The statistical characteristics of samples and population are usually different. Therefore, the statistical characteristics of population will not be preserved by the synthetic data. A method for generating the synthetic data by employing Bootstrap resampling technique and ARIMA models was proposed in this study. The results indicate that the statistical characteristics of population instead of samples were preserved by this proposed method.
