跳到主要內容 :::

年度	112	專案性質	實驗性質
專案類別	模場試驗	研究主題	整治
申請機構	國立中興大學	申請系所	環境工程學系
專案主持人	張書奇	職等/職稱	副教授
專案中文名稱	南崁溪污染底泥整治模場試驗計畫 （結合人工智慧之多溴二苯醚整治技術研發）
中文關鍵字	底泥整治、南崁溪、模場試驗、相反轉法、多溴二苯醚、人工智慧
專案英文名稱	A Pilot study of the remediation of contaminated sediments – The technology development of integrated artificial-intelligence and the remediation of sediments contaminated by polybrominated diphenyl ether
英文關鍵字	Remediation, Artificial intelligence, polybrominated diphenyl ether, Mercury
執行金額	3,500,000元
執行期間	2023/9/1 至 2024/8/31
計畫中文摘要	工業發達國家之底泥污染情況相當普遍，河川下游之底泥與河口附近海水底泥 均已遭受嚴重污染，其中尤以疏水性有機物及重金屬污染最為嚴重，而台灣之底泥 污染物檢測濃度在全世界相關文獻報導中屢屢名列前茅。如台南安順場址海水池底 泥中戴奧辛之污染、南崁溪底泥中之多溴二苯醚污染。這些污染也可藉由食物鏈破 壞生態系及影響人體健康，急需發展經濟有效之整治技術。本研究團隊於 107-108 年度計畫以相反轉技術結合微生物篩選及降解技術(ISPIE/BiRD)進行底泥整治之 現地模場試驗及土壤污染整治之砂箱試驗，不僅對於已經風化之多氯聯苯污染物 Aroclor 1254 與六氯苯可達 98.0%以上去除率，且對土壤中 DDT 可達到 99.9%以上 之去除率，Lindane 可達 92.2%以上之去除率。為進一步優化此技術並結合人工智 慧之深度機器學習技術，針對底泥中多溴二苯醚污染整治技術進行 2 年期之技術研 發計畫，本年度為第 2 年計畫。第 1 年計畫之預先採樣已確認南崁溪底泥中有高濃 度之 BDE-209 之污染，且批次實驗與管柱實驗均確認 ISPIE/BiRD 之技術可行性且 確認 pH 值與有機質為主要控制因子。本計畫為第 2 年計畫，是以現地模場試驗驗 證第 1 年計畫成果並且針對第三代基因定序分析結果進行機器學習以建立完善之 ISPIE/BiRD 底泥整治之預估模式與關鍵微生物指標。本計畫至目前已經完成底泥 採樣與微生物馴養、ISPIE 操作、BiRD 連續監測、台越線上研討會、機器學習等 工作項目。最終結果顯示:ISPIE 操作部分風化組中最佳為 FL(添加組下層)，次 佳為 FU(添加組上層)，其次為 WL(風化組下層)，最差者為 WU(風化組下層)。 將 ISPIE 與 BiRD 兩階段脂去除率合併計算，最佳之總移除率可達 84%。風化組中 以 WNR-L 最佳(k = 0.0103 d-1, t1/2 = 67 d)，WBS-L 次佳(k = 0.0032 d-1, t1/2 = 218 d); 添加組也以 FNR-L 最佳(k = 0.0162 d-1, t1/2 = 43 d)，FBS-L 次佳(k = 0.0157 d-1, t1/2 = 44 d)，目標污染物之移除率已經優於第 1 年計畫之批次實驗與管柱實驗成果。原 本之適合風化污染物生物還原脫溴之菌群可能也有助於汞之去甲基化。TGS 資料分 析顯示有進行 ISPIE 技術之組別可觀察到 Clostridium 屬豐度提升，而 Clostridium 豐度之提升對於厭氧還原脫溴效果有正相關，Alpha 多樣性分析顯示降解能力佳之 組別似乎有 Shannon index 與 observed feature(ASVs 數)均偏低之情況。TGS 資 料機器學習研究結果顯示 LASSO 最為適合本計畫資料之機器學習模式，其交叉驗 證階段之 MSE、RMSE 及 R2 分別可達 0.0019、0.0435 與 1.0000。本計畫施行符合 預定進度，各項查核點已如期完成。
計畫英文摘要	Sediment pollution in industrially developed countries is quite common. Hydrophobic organic compounds have severely contaminated the sediment in the lower reaches of rivers and the estuarine. In Taiwan, the detection concentration of sediment pollutants has been repeatedly ranked among the top in relevant literature reports worldwide. For example, there is dioxin pollution in the bottom mud of the seawater pool at the Anshun site in Tainan and polybrominated diphenyl ether pollution in the sediment of the Nankan River. These pollutions can also damage the ecosystem and affect human health through the food chain, and there is an urgent need to develop cost-effective remediation technologies. From 2018 to 2019, our research team developed in situ phase inversion emulsification and biological reductive dehalogenation (ISPIE/BiRD) to remediate contaminated sediment and soil. We achieved a removal rate of more than 98.0% on Aroclor 1254 and hexachlorobenzene in sediment, more than 99.9% on DDT, and 92.2% on Lindane in soil. To optimize this technology further and combine it with deep machine learning technology, a two-year technology research and development plan is being carried out to treat polybrominated diphenyl ether pollution in sediments. This year is the second year of the plan. The pre-sampling planned in the first year has confirmed that there is a high concentration of BDE-209 contamination in the sediment of Nan-Kan River, and both batch experiments and column experiments have confirmed the technical feasibility of ISPIE/BiRD and confirmed that the pH value and organic matter are main control factors. This project is the second year of the two-year program. We conducted a field pilot study to verify the project's first-year results. We applied machine learning to the third-generation sequencing (TGS) results to establish complete ISPIE/BiRD sediment remediation patterns and key microbial indicators. This project has so far completed sediment sampling, microbial domestication, ISPIE operation, BiRD monitoring, Taiwan-Vietnam webinar, and machine learning. The final results show that in the ISPIE operation, in the weathered group, the best is WNR-L, and the second best is WNR-U; in the fresh-added group, the best is FBS-U, and the second best is FBS-L. Combining the two-stage fat removal rates of ISPIE and BiRD, the best total removal rate can reach 84%. In the weathering group, WNR-L was the best (k = 0.0103 d-1, t1/2 = 67 d), and WBS-L was the second best (k = 0.0032 d-1, t1/2= 218 d). For the fresh-added group, FNR-L is the best (k = 0.0162 d-1, t1/2= 43 d), and FBS-L is the second best (k= 0.0157 d-1, t1/2 = 44 d). The target contaminant removal rate is better than the results of batch experiments and column experiments conducted in the first year. The bacterial flora originally suitable for the bioreductive debromination of weathered pollutants may also contribute to the demethylation of mercury. TGS data analysis shows that an increase in the abundance of Clostridium can be observed in the group with ISPIE technology, and the rise in Clostridium abundance is positively correlated with the effect of anaerobic reductive debromination. Alpha diversity analysis shows that the group with good degradation ability has a lower Shannon index, and the observed feature (number of ASVs) is lower, too. The machine learning results on TGS data showed that LASSO is the most suitable model for this project's data. In the cross-validation stage, its MSE, RMSE, and R2 can reach 0.0019, 0.0435, and 1.0000, respectively. The implementation of this project is in line with the scheduled progress, and all checkpoints have been completed as planned.