本研究之主要目的係使用目前最新的時頻分析方法-希爾伯特-黃轉換(Hilbert-Huang Transform, HHT)做為主要時頻分析工具，並將時頻圖以熵值量化，藉以探討氣候變異是否對濁水溪流域之降雨產生影響。 本研究首先以日雨量資料觀點，直接對氣候變異是否對降雨型態產生影響做探討，此外也加入不同於傳統的邊際頻率、邊際時間與熵值等三種觀點，再進一步搭配有效日降雨強度、降雨分佈情形等水文指標性數值，力求以不同的角度探討降雨變化之存在與否。此外，本研究另設計一檢定準則，以判定在不同觀點下，降雨是否因氣候變異而發生變化。經由此檢定方式亦可降低直接由圖面判定變化，造成結論過於主觀之情況發生。本研究並列出檢定觀測值所對應之P-value，來表示各測站降雨量於不同月份受氣候變異影響之程度。 將時頻圖利用熵值量化後可以發現，雨量分佈較集中，不確定性較小，熵值會因而較低；反之若雨量分布較均勻，不確定性較高，熵值則相對較高。由檢定結果得知，氣候變異使降雨型態有了某些改變，可是這些改變的訊息在雨量觀點中，卻僅能提出二月份有變化之結論，其他月份則無明顯變化發生。 相對地，藉由邊際頻率、邊際時間與熵值等觀點，則可看出其他月份其實是有變化存在的，只是此變化訊息不易由雨量觀點上直接看出，需經由其他觀點才能看出其變化之存在。以上利用不同的觀點搭配本研究所設計之檢定準則來探討降雨變化之存在性，為本研究所提出之主要論點，提供日後水文分析與氣候變異研究之參考與應用。 This study investigated the application of Hilbert-Huang Transform (HHT) in detecting changes in precipitation patterns brought by global climate change. Marginal frequency spectrum analysis, marginal time spectrum analysis, and calculation of time-frequency entropy were applied to precipitation data from Zhuoshuixi watershed, located in Taiwan, to detect and identify possible changes in precipitation patterns. In addition, additional hydrological data such as effective mean rainfall intensity and the distribution of precipitation were also analyzed as other approaches of showing the effects of climate change. As a method of reference for comparison with HHT, the conventional null hypothesis tests were also applied to verify possible changes in the amount of precipitation in different months of the year.
In the null hypothesis tests, significance values (P-value) were determined to signify the degree of climate change impact on the amounts of precipitations in different months of the year. The results of the null hypothesis tests suggested that the changes in precipitation patterns are only statistically significant in the month of February. In comparison, Hilbert-Huang Transform (HHT) plus the inspection of additional hydrological data were able yield more information about changes in precipitation patterns. Time-frequency entropy value was found to be small for concentrated precipitation with less uncertainty, while large for distributed precipitation with higher uncertainty.
