使用開徑式傅立葉轉換紅外線光譜儀及逆演算空氣擴散模式定位逸散源之方法驗證

Abstract

過去許多研究提出逆演算空氣擴散模式技術，以定位不明污染逸散源位置之方法。本研究使用開徑式傅立葉轉換紅外線光譜儀(Open-Path Fourier Transform Infrared)收集下風處濃度資料，在大範圍之空間區域中，相對於使用多個定點採樣器，可更快地獲得具有代表性之資料。本研究在工業區進行三場次追蹤氣體實地實驗，於下風處架設三條OP-FTIR監測線(長度分別為123、127、127公尺)，追蹤氣體釋放源有三個點，分別為逸散源A(Source A)與三道監測線之距離大約為105、315、與530公尺(期間: 295分鐘)，逸散源B(Source B)與三道監測線之距離大約為355、565、與780公尺(期間:180分鐘)，逸散源C(Source C)與三道監測線之距離大約為400、615、與825公尺(期間: 483分鐘)。為從理論上驗證此技術之可行性，也於電腦程式中進行模擬進行前向(forward)方法預測下風處濃度，並使用最佳化演算法逆演算美國環保署ISCST3(Industrial Source Complex – Short Term 3)模式與AERMOD(American Meteorological Society Environmental Protection Agency Regulatory Model)模式去回推逸散源位置，另嘗試以風向直接進行向量(vector)法估計出可能之逸散源位置。研究結果發現逆演算法雖然無法精準回推釋放點，但藉由適當之篩選條件可得到一可能釋放區域(uncertainty area)，ISCST3模式於僅用1個風向下Source B真實點與預測點之平均距離為60.40公尺(CCF>0.5)，而AERMOD模式於Source B真實點與預測點平均距離為49.44公尺(CCF>0.75) 皆為經過篩選後之最佳結果，而其流量之預測結果分別為低估22.6%與51.0%。另外兩個釋放源由於實驗場地與資料限制，其結果不盡理想，1個風向之回推結果平均距離於Source A為128.22-340.42公尺，而Source C為156.93-295.38公尺。研究結果亦顯示氣象資料及模式參數值會影響回推值，因此未來研究需計算與取得適和台灣環境之參數值，以使模式預測或回推更為準確。

In many previous studies, the technique of inversing air dispersion model technology was presented to locate unknown emission sources locations. In this study, wWe collected the downwind concentration data using optical remote sensing (ORS)open-path fFourier transform infrared spectroscopy (OP-FTIR) instrument thatwhich can obtainprovide representative data faster than using many point samplers among ain large spatial areas. In our field experiments of releasing tracer gases, we set up three discrete OP-FTIR monitoring lines (lengths of lines were 123, 127, and 127m) at the downwind sites of the survey area near an industrial complex to locate two three artificially released emission sources. ForTo verifying the inversion algorithmtheory, we also conducted computer simulation studies. combine path integrated concentration data and meteological data as input data, and the uncertainty areas of unknown emission source are estimated. For Source A, the distance between the source and the monitoring lines was 355, 565, and 780m, respectively. On the other hand, that distance for Source B was 105, 315, and 530m. The experiment study was conducted with three discrete monitoring lines as the field experimental setup in the reconstruction process. The distance between the Ssource A (duration: 295minutes) and the monitoring lines was 105, 315, and 530m, respectively. The distance between the Ssource B (duration: 180minutes) and the monitoring lines was 355, 565, and 780m, respectively. The distance between the Ssource C (duration: 438minutes) and the monitoring lines was 400, 615, and 825m, respectively. An oOptimization algorithm was used to inverse the U.S. EPA ISCST3 and AERMOD models for to traceing back the source locations considering different scenarios including different wind directions, emission rates and source locations. Previous studies showed that the screening criteria with efficient downwind PICdata and wind direction could improve the reconstruction result. For verifying the theory, we demonstrated the results of uncertainy area with different screening criteria, but we found that screening criteria of CCF and wind direction were not very robustious . The results showed that the true source locations could not be identified exactly but they could be covered by the uncertainty areas.We could estimated the uncertainy area of possible source location with reconstruction procedure of ISCST3 and AERMOD model. The average distance between Source B and the predicted source location was 49.44m (AERMOD model) and 60.40m (ISCST3 model), and Source B was provided the best result after screening for CCF larger smaller than 0.75. The estimated emission rates were underestimated from real emission rate forby 22.6% (ISCST3 model) and 51.9% (AERMOD model). The other emission sources were gave obtained poorworse results because of limitations of the experimental setup. The average distance of errors for Source A ranged from 128.22-340.42m, and Source C was 156.93-295.38m. The poor results were because we could notdue to not offer thehaving suitable proper model parameters and meteorological data for model process. Future studies should obtain local data to improve the performance of the modeling and inversion techniques.