由研究結果顯示,資料長度為29時,加入一筆極端值資料( 0.3, 0.5, 0.9),不論是 100年或200年,以m= 2 ,3可獲致最佳之參數推估結果。惟當資料長度為49時,加入一筆 200年極端值資料之參數推估結果,均較加入 100年之結果為佳。同時,由於資料長度較長,即使已修正權重至m=4仍然無法推估到極端值其真正之參數值,表示資料長度愈長需要更高的修正權重係數(m),才能反應因極端水文事件發生所造成之低流量頻率分析之影響。因此,建議本方法未來可作為低流量頻率分析與水文情勢研判之參考與應用。 In recent decades, the frequency of extreme flood and drought events has increased in many parts of the world, signifying a shift in global climatic pattern. Conventional flood frequency analysis methods are not readily applicable to drought frequency analysis due to large variability in drought duration. The low-end linear moment method (LL-moments), which assign correction factors (m) to extreme value data, had been suggested for the analysis of low flow observations. In this study, the LL-moments is applied to the fitting of empirical time series data from nine stream gages in Taiwan to the general extreme value (GEV) distribution model. Three out of the nine sets of data demonstrated good fit with the general extreme value distribution (GEV) model.
LL-moments method was then reapplied to simulated time series generated from known GEV model to investigate the effect of varying the value of correction factors on the reliability of parameter estimation. It was found that for data length less then 29, applying the correction factor of 2 or 3 yielded good parameter estimations, even when an extreme value data of return period of 100 or 200 years was added to the time series. For data length 49, applying the correction factor of 2, 3 or even 4 did not yield good parameter estimation. Further study about applying correction factor greater than 4 would be recommended when data length is greater than 49.