應用概似不確定性估計於溼地參數之不確定性分析

Abstract

溼地為一種尊重自然、遵循生態平衡的自然淨化單元。早期溼地研究多半關注在水質淨化成效上，近年來，針對溼地內部機制之模式研究不斷增加，但要能完整呈現溼地複雜機制卻相當困難。模式中參數的維度因模式結構複雜度而異，並且因季節、溫度與空間的變化，對模式預測結果之影響甚巨，為模式不確定性的主要來源之一。然而在眾多溼地模式的應用中，以系統動力模式可清楚描述溼地系統中物質傳輸機制，讓使用者清楚了解溼地系統結構和影響去除效能的參數，因此本研究運用概似不確定性估計（Generalized Likelihood Uncertainty Estimation,GLUE）進行系統動力溼地模式之參數不確定性分析，提供未來溼地研究與應用上的參考。 本研究選定成功大學人工溼地場址收集之水質數據，來探究溼地機制並預測溼地中溶氧（DO）、五日生化需氧量(BOD5)、總氮(TN)、總懸浮固體(TSS)、總磷(TP)之水質變化情形。經過模式的率定及驗證後，探討模擬結果的表現及各營養鹽預測時之敏感性參數，並透過對合宜模擬參數解集合(behavioral simulations)分析參數不確定性對模式預測之影響程度。研究結果顯示，系統動力溼地模式已能初步掌握溼地模場水質變化之趨勢。在同時擾動多個參數下，對溼地各營養鹽預測結果有明顯影響的參數為BOD中含碳比例(BODC)、Arrhenius常數(θ1)、顆粒粒徑大小(dss)、懸浮顆粒中含磷比例(FacP/SS )四種。而當參數解集合中BODC範圍在0.6至0.72、θ1在0.85至1.10、dss大於4.95E-6與dss及FacP/SS參數值分別小於3.5E-6及0.1313時，模擬結果較佳，並進一步透過多目標分析探討較佳之參數解集合。綜觀以上結果，研究分析各別營養鹽之參數敏感度及水質變化的不確定性邊界，提供後續溼地工程施作及操作上，更多參考依據，期望能對人工溼地的應用發展有幫助。

Wetland is a kind of natural treatment that always follow the balance of ecosystem and respect the nature. Wetland researches concerned about the cleaning ability of westewater in the beginning, but the studies of the interaction inside the wetland system are increace a lot in these few years. Unfortunately, it is hardly to spread out the whole system of wetland in models. The complexity of model structure can effect the dimension of model parameters. With the variation of seasons, temperature, and spaces. The parameters will have an extremely influence of the model prediction. It is one of the factor of uncertainty sources. In many kinds of wetland model applications, system dynamic(SD) model could describe the transformation of wetland structure much more clearly, let the user could easily understand the wetland system and the parameters in SD model. This research applications the Generalized Likelihood Uncertainty Estimation(GLUE) to uncertainty analysis of system dynamic wetland model‘s parameter. Hope to provide some informations for wetland researchs in the future. The water quality data use in theis research are collected from the construct wetland in National Cheng Kung university, to analysis the wetland interactions and predicts the water quality changing about dissolve oxygen(DO), biochemical oxygen demand(BOD5), total nitrogen(TN), total suspended solid(TSS), and total phosphorous(TP). Through the model calibration and validation steps, to determine the results performance and the sensitivity parameters for the prediction of every nutrients. Then use the behavioral simulations of the divided groups to analysis the influence of uncertainty for the model predictions. According to the results, the system dynamic wetland model can already fit the water quality changing of construct wetland. If disturb the parameters at the same time, there were four parameters are more sensitive such as the carbon in the BOD(BODc), Arrhenius constant(θ1), the diameter of suspended solid(dss), and the phosphorous in the suspended solid(FacP/SS).when the value of BODC between 0.6 to 0.72、θ1 between 0.85 to 1.10、dss value is bigger than 4.95E-6 and the parameter value of dss and FacP/SS smaller than 3.5E-6 and 0.1313 that the performance are comparatively good. Finally, use the method of multi-objective analysis to find the better results of parameters. This research analysis the sensitivity of parameters for each nutrients and find the better parameter results for predict the water quality performance. Give more important informations to wetland construction and operation. Hope can help and expanding the applications for construct wetland.