臺灣博碩士論文加值系統

集水區內部降雨、入滲等水文現象在空間上具有不均勻之特性，傳統之
降雨-逕流模式常將集水區視為集塊系統，忽略了水文現象在空間的不均
勻特性。為充分模擬集水區內部水文現象之變化，本文利用分布型降雨-
逕流模式，除使用全域最佳化技巧來率定模式參數，並使用三場降雨事件
來進行模式驗證，發現結果尚為合理。由於爾後模式的應用可能超出率定
事件之範圍，為了解模式輸出由於參數之不確定而引起之誤差，本文針對
模式率定參數所導致之不確定性進行探討，分別使用蒙地卡羅模擬法、拉
丁超立方取樣法、羅森布魯斯點估算法與荷爾點估算法來進行分析，以建
立模式計算歷線之95%信賴區間並提供使用者參考，最後並比較四種方法
之分析結果，得知拉丁超立方取樣法所得之結果比較接近蒙地卡羅法，為
一可替代蒙地卡羅模擬法之分析方式。另外，羅森布魯斯點估算法與荷爾
點估算法所得之變異數有偏大之趨勢。在參數敏感度分析方面，分別使用
局部與全域敏感度分析進行模式三個參數Ks(地表逕流蓄水常數)、Kc(渠
道流蓄水常數)與CH(影響入滲與降雨臨前條件有關)之敏感度研究，由分
析結果得知影響入滲之參數CH為對模式輸出較為敏感之參數。於是進一步
結合較為敏感參數CH與集水區物理特性間之關係，以期減少傳統使用平均
參數所導致逕流模擬之誤差。由分析結果發現利用臨前五日平均流量與CH
值具有較佳之關係，是以建立兩者間之迴歸方程式，並以三場驗證事件來
比較，如果使用迴歸方程式來決定CH參數之模擬結果，確實較傳統使用六
場率定事件之平均參數的方式更能精確模擬流量歷線。關鍵詞：分布型降
雨-逕流模式、不確定性、敏感度分析。

It has been known that hydrological processes (e.g.,
precipitation, infiltra-tion,... , etc.) over basin are
heterogeneous. Traditionallumped rainfall-runo-ff models ignore
the spatial heterogeneity ofhydrological processes. To simula-te
hydrological heterogeneity over basin, distributed rainfall-
runoff models w-ere used in this study, in which global
optimization technique was applied for model calibration. The
validation from three storm events concluded that the d-
istributed model has the ability to simulate the
historicalrainfall-runoff rel-ationship. However, the model may
be applied tostorm events outside of the range of conditions for
which the model has been successfully calibrated and verified.
In order to examine the error of model output caused by
parameters uncertain-ty, four methods, including, Monte Carlo
Method (MCM), Latin Hypercube Sampling Technique, Rosenblueth*s
Point Estimation Method and Harr*s Point Estimation Method, were
used in the study and build 95% confidence interval of
estimatedhydrograph. From the comparison of four methods, Latin
Hypercube Sampling Techn-ique has similar analysis results as
Monte Carlo Method has. The variances esti-mated from
Rosenblueth*s Point Estimation Method and Harr*s Point
Estimation M-ethod are larger than that from MCM. Thesensitivity
of three model parameters, overland flow storage parameter (Ks),
channel storage parameter (Kc) and initi-al infiltration rate
correcting parameter (CH), were further examined by local and
global methods. CH was found to be more sensitive than the other
model para-meters. In order toreduce model errors caused by CH
parameter, which is the mo-st sensitive parameter in the model,
building the relationship between CH and physical properties
over basin is studied. The CH parameter was found to have g-ood
relation with 5-day average flow before the event. The model
performance w-as concluded from three storms that using CH
derived by 5-day average flow bef-ore storm to replace average
values of CH parameter from 6 calibration storms c-an improve
the results of hydrograph simulation.Kekeywords : distributed
rainfall-runoff model, uncertainty, sensitivity analysis.