ACRONYM模式敏感度與不定性分析研究

Abstract

本研究藉由LHS(Latin Hypercubic Sampling)方法產生適當組數之模式輸 入資料，應用現有較常用的分析技巧，評估ACRONYM模式相應輸出值所具 有之敏感度及不定性。在敏感度分析方面，選擇不同時間及位置之模式輸 出與輸入參數進行迴歸分析。根據多變數迴歸所求出之迴歸方程式，參數 及參數群相應輸出值之敏感度可予以量化，其結果將有助於模式參數率定 效率之提高。在不定性分析方面，步驟與敏感度分析時約略相同，不同之 處乃在分析不同參數合成而得的合成資料。取得輸入參數與相應模式輸出 的迴歸式後，依第一及第二類型平方和之理論分別進行模式輸出不定性受 參數群及參數不定性之影響並予量化，且可得知重要之輸入參數，其結果 並可進一步提供模式使用者瞭解模式模擬結果之可靠度所需之資料。 The purpose of this study is to use suitable sets of input parameters generated by Latin Hypercubic Sampling (LHS) method, and to apply the current analytic technique to evaluate sensitivity and uncertainty of model outputs from ACRONYM. In sensitivity analysis, various model outputs at different times and locations are used to develop multiple regression relationships with model inputs. Based on the developed regression equations the sensitivity of input parameters and input groups on model outputs can be quantified. Information from the sensitivity analysis can be utilized to enhance the efficiency for model parameters calibration. In uncertainty analysis, the procedures are almost the same as that in sensitivity analysis, except that the synthesized data for model input parameters are somewhat different. When the regression equations between model input parameters and model outputs are obtained, the theories associated with type I SS and type II SS are adopted to perform the uncertainty analysis of input groups and input parameters, respectively. From the uncertainty analysis, model output uncertainty can be quantified and important model input parameters can be identified. Furthermore, analysis such as this provides important information on the degree of reliability of simulated results from the model which is useful to model users.