準確推估雨量對水文設計而言是相當重要的課題,但常因資料在時序上的完整性不足或空間中的代表性不足等因素而使推估結果產生誤差。過去雖有許多研究試圖改善此類問題,然而若考量到我國多個機關各擁站網之現況,結合多站網推估雨量資料以提高精準度亦為一可行之研究方向。高屏地區為我國糧產重鎮之一且四季分明、豐枯季降雨之差距大,故本研究結合該區之中央氣象局與水利署的雨量站網資料,應用通用克利金法針對高屏地區之降雨分布進行空間變異推估。 為減少地形影響非定常性雨量資料的空間變異性,先以逐步迴歸法建立雨量分佈趨勢函數,以求得空間變異性符合定常性的雨量殘差值,再應用通用克利金法對雨量殘差值進行空間變異推估,並以交叉檢定法評估模式之誤差及探討其正確性。最後透過T檢定驗證兩站網之特性是否相同,若不同則以通用共克利金法重新推估之。結果顯示未通過T檢定之資料達53%,因此本研究認為在結合不同雨量站網資料進行空間變異推估時,應利用共克利金法將站網特性納入考量以提高估算之精準度。
Precisely estimating precipitation is very important for hydrological planning. However, lack detailed data on spatial or temporal precipitation distributions frequently cause design errors. Although many researches attempted to reduce the errors, a combination of several rainfall networks is a feasible way to increase the accuracy of estimation. This study applied universal cokriging (UCK) combining two precipitation data observed by Central Weather Bureau (CWB) and Water Resources Agency (WRA) to estimate the precipitation distributions in Gau-Pin region. To reduce the non-stationary effect on spatial variance of rainfall data due to terrain, residual values are first determined between observed rainfall data and spatial trends estimated by stepwise regression analysis. Then, UK was used to estimate spatial variability of precipitation. Finally, a cross-validation procedure was adopted to characterize the estimated errors of individual and combining rainfall networks. A T-test was used to quantify the difference of estimated errors between two rainfall data. If the difference is significant, UCK is more suitable than universal kriging (UK). The analyzed result reveals that about 53% of data do not pass the T-test. Thus, applying UCK to estimate spatial precipitation variability can increase the accuracy of estimation for combining several rainfall networks.