傳統污染場址監測仰賴人工實地採樣與化學分析，並將檢體送往實驗室進行化學分析。此類傳統方法雖然在局部點位上具備高精確度，但其製程不僅費用高昂、耗費大量人力與時間，且受限於採樣點位在空間上的離散性，難以連續且全面地揭示污染之動態空間分佈，亦缺乏長時序的監測機制以評估污染之擴散邊界與長期復育成效。為克服傳統監測技術之局限，本研究建構了一套結合「多尺度遙測影像」與「機器學習演算法」之新型環境監測架構，利用地表植被之光譜變異作為間接感測土壤污染介質之科學工具。因此本研究旨在建構一套結合多尺度遙測影像與機器學習之環境監測框架。研究利用公開免費之Sentinel-2衛星影像，結合葉面積指數(LAI)分層機制與隨機森林(Random Forest)演算法，並與DJI Mavic 3M多光譜無人機及手持式XRF進行多尺度交互驗證。
研究結果顯示，LAI為提升模型辨識精確度之關鍵因子。在裸露地LAI 0層級，健康對照組之正確辨識率僅為53%，然而隨著植生密度增加至LAI 3層級時，正確率顯著提升至82%。證實以植被壓力為觀測污染之介質的可行性，而在不同時序比對下本研究透過2018年與2024年之對比分析，為模型之穩定提供了直接實證。在三類分類模型中，2018年與2024年之LAI 0訓練結AUC均維持在0.95(95%)，而在LAI 3高蓋度層級，AUC亦分別達到0.86(86%)與0.90(90%)，呈現穩步提升之勢。四類分類模型之驗證亦顯示一致趨勢，兩年度各LAI層級之AUC表現高度重疊，整體變動幅度控制在10%以內。
上述跨時序驗證數據證明，透過監督式分析訓練之植生指標能有效克服不同年份之氣候與環境波動，建立起穩定的污染壓力辨識指紋。這強化了本研究之科學論點以植生指標作為監測具備極高之可靠度，足資證明模型具備長期環境監測之實務應用潛力，更能夠為未來建立標準化之污染場址預警系統奠定了數據基礎。
最後透過多尺度觀測之相互比對，無人機影像交互驗證衛星機器學習之結果，並結合地面真值XRF化學實證，建立完整之驗證體系。綜上所述，本研究提出之流程不僅擴展了污染監測之時空範圍，亦能降低傳統普查資源投入，為污染區監測提供具經濟效益之科學解決方案。

Traditional monitoring of contaminated sites relies heavily on manual in-situ sampling and chemical analysis, which are not only costly and labor-intensive but also lack long-term temporal monitoring mechanisms to assess the extent of pollution and the effectiveness of land restoration. Therefore, this study aims to construct an environmental monitoring framework that integrates multi-scale remote imagery with machine learning. The research utilizes open- access Sentinel-2 satellite imagery combined with a Leaf Area Index (LAI) stratification mechanism and Random Forest (RF) algorithms, while incorporating DJI Mavic 3M multispectral UAVs and handheld XRF for multi-scale cross-validations.
The results indicate that LAI is a critical factor in enhancing model identification accuracy. In the bare soil (LAI 0) stratum, the classification accuracy for the healthy control group was only 53%. however, as vegetation density increased to the LAI 3 stratum, the accuracy significantly rose to 82%, confirming the necessity of utilizing vegetation as a medium for observing environmental stress in the region.
Furthermore, a comparative analysis between 2018 and 2024 provides direct evidence of the model's stability. In the three-class classification model, the Area Under the Curve (AUC) for the LAI 0 training results remained at 0.95 (95%) for both years, while at the high-coverage LAI 3 level, the AUC reached 0.86 (86%) and 0.90 (90%) respectively, showed a steady upward trend. Validation of the four-class model displayed consistent results, with AUC performance across different LAI levels in both years overlapping significantly and overall fluctuations remained within 10%.
The aforementioned cross-temporal validation data proves that vegetation indices trained through supervised analysis can effectively overcome climatic and environmental fluctuations across different years, establishing a stable diagnostic fingerprint for pollution. This reinforces the scientific argument of this study: that vegetation indices possess high reliability for monitoring, demonstrating the practical potential for long-term environmental surveillance and laying a data-driven foundation for future standardized early warning systems for contaminated sites.
The afore mentioned cross-temporal validation data proves that vegetation indices trained through supervised analysis can effectively overcome climatic and environmental fluctuations across different years, establishing a stable diagnostic fingerprint for pollution pressure. This reinforces the scientific argument of this study: that vegetation indices possess high reliability for monitoring, demonstrating the practical potential for long-term environmental surveillance and laying a data-driven foundation for future standardized early warning systems for contaminated sites.

1. Agisoft. (2026). Agisoft Metashape Professional Edition Features. Cited at:2026.03.11.Available at: https://www.agisoft.com/features/professional-edition/
2. Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32.
3. Chávez, R. O., Clevers, J. G. P. W., Decuyper, M., De Bruin, S., & Herold, M. (2016). 50 years of water extraction in the Pampa del Tamarugal basin: Can Prosopis tamarugo trees survive in the hyper-arid Atacama Desert (Northern Chile)?. Journal of Arid Environments, 124, 292-303.
4. Clark, M. L. (2017). Comparison of simulated hyperspectral HyspIRI and multispectral Landsat 8 and Sentinel-2 imagery for multi-seasonal, regional land-cover mapping. Remote Sensing of Environment, 200, 311-325.
5. Cotrozzi, L., Townsend, P. A., Pellegrini, E., Nali, C., & Couture, J. J. (2018). Reflectance spectroscopy: A novel approach to better understand and monitor the impact of air pollution on Mediterranean plants. Environmental Science and Pollution Research, 25(9), 8249–8267.
6. Ding, C., & He, X. (2004). K-means clustering via principal component analysis. Proceedings of the twenty-first international conference on Machine learning, 225-232.
7. DJI. (2026). DJI Mavic 3M Overview.Cited at:2026.03.05.Available at: https://enterprise.dji.com/zh-tw/mavic-3-m
8. Dutta, S., Mitra, M., Agarwal, P., Mahapatra, K., De, S., Sett, U., & Roy, S. (2018). Oxidative and genotoxic damages in plants in response to heavy metal stress and maintenance of genome stability. Plant signaling & behavior, 13(8), e1460048.
9. Earth Engine Data Catalog. (2026). Sentinel-2: MultiSpectral Instrument, Level-2A. Cited at:2026.03.05.Available at: https://developers.google.com/earth-engine/datasets/catalog/sentinel-2?hl=zh-tw
10. Gitelson, A. A., Gritz, Y., & Merzlyak, M. N. (2003). Relationships between leaf chlorophyll content and spectral reflectance and algorithms for non-destructive chlorophyll assessment in higher plant leaves. Journal of Plant Physiology, 160(3), 271–282.
11. Google Earth. (2026). Google Earth Pro Versions. Cited at:2026.03.11.Available at: https://www.google.com/intl/zh-TW/earth/about/versions/
12. Harris, A., Bryant, R. G., & Baird, A. J. (2006). Mapping the effects of water stress on Sphagnum: Preliminary observations using airborne remote sensing. Remote Sensing of Environment, 100(3), 363–378.
13. Jorge, J., Vallbé, M., & Soler, J. A. (2019). Detection of irrigation inhomogeneities in an olive grove using the NDRE vegetation index obtained from UAV images. European Journal of Remote Sensing, 52(1), 169-177.
14. Kior, A., Sukhov, V., & Sukhova, E. (2021). Application of reflectance indices for remote sensing of plants and revealing actions of stressors. Photonics, 8(12), 582.
15. Legendre, P. (1993). Spatial autocorrelation: trouble or new paradigm?. Ecology, 74(6), 1659-1673.
16. Loggenberg, K., Strever, A., Greyling, B., & Poona, N. (2018). Modelling water stress in a Shiraz vineyard using hyperspectral imaging and machine learning. Remote Sensing, 10(2), 202.
17. Liu, Y., & Zhu, X. (2024). Tracking mangrove light use efficiency using normalized difference red edge index. Ecological Indicators, 168, 112774.
18. Lee, W. S. (2015). Plant health detection and monitoring. In Hyperspectral Imaging Technology in Food and Agriculture (pp. 275–288). New York, NY: Springer New York.
19. Magaña-Maldonado, R., Chávez-Cortez, E. G., Olascoaga-Arellano, N. K., López-Mejía, M., Maldonado-Leal, F. M., Sotelo, J., & Pineda, B. (2016). Immunological evasion in glioblastoma. BioMed research international, 2016.
20. Mazur, R., Sadowska, M., Kowalewska, Ł., Abratowska, A., Kalaji, H. M., Mostowska, A., ... & Krasnodębska-Ostręga, B. (2016). Overlapping toxic effect of long term thallium exposure on white mustard (Sinapis alba L.) photosynthetic activity. BMC Plant Biology, 16(1), 191.
21. Merzlyak, M. N., Gitelson, A. A., Chivkunova, O. B., & Rakitin, V. Y. (1999). Non‐destructive optical detection of pigment changes during leaf senescence and fruit ripening. Physiologia plantarum, 106(1), 135-141.
22. Muschelli, III, J. (2020). ROC and AUC with a binary predictor: a potentially misleading metric. Journal of classification, 37(3), 696-708.
23. Norman, J. M., & Campbell, G. S. (1989). Canopy structure. In Plant physiological ecology: field methods and instrumentation (pp. 301-325). Springer Netherlands.
24. Page, V., & Feller, U. (2015). Heavy metals in crop plants: transport and redistribution processes on the whole plant level. Agronomy, 5(3), 447–463.
25. Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... & Duchesnay, É. (2011). Scikit-learn: Machine learning in Python. The Journal of machine Learning research, 12, 2825-2830.
26. Pettorelli, N., Vik, J. O., Mysterud, A., Gaillard, J. M., Tucker, C. J., & Stenseth, N. C. (2005). Using the satellite-derived NDVI to assess ecological responses to environmental change. Trends in ecology & evolution, 20(9), 503-510.
27. Rock, B. N., Hoshizaki, T., & Miller, J. R. (1988). Comparison of in situ and airborne spectral measurements of the blue shift associated with forest decline. Remote Sensing of Environment, 24(1), 109-127.
28. Roy, D. P., Boschetti, L., & Trigg, S. N. (2006). Remote sensing of fire severity: assessing the performance of the normalized burn ratio. IEEE Geoscience and remote sensing letters, 3(1), 112-116.
29. Segarra, J., Buchaillot, M. L., Araus, J. L., & Kefauver, S. C. (2020). Remote sensing for precision agriculture: Sentinel-2 improved features and applications. Agronomy, 10(5), 641.
30. Sentinel Online. (2026). Sentinel-2 Missions. Cited at:2026.03.05.Available at: https://sentinels.copernicus.eu/web/sentinel/missions/sentinel-2
31. Sentinelhub. (2026). Sentinel-2 LAI Custom Scripts. Cited at:2026.03.10.Available at: https://custom-scripts.sentinel-hub.com/custom-scripts/sentinel-2/lai/
32. SentiWiki. (2026). Sentinel-2 Mission: Satellite Description. Cited at:026.03.05.Available at: https://sentiwiki.copernicus.eu/web/s2-mission#S2Mission-SatelliteDescription
33. Susmaga, R. (2004). Confusion matrix visualization. Intelligent Information Processing and Web Mining, 107-116.
34. Tan, Y., Sun, J. Y., Zhang, B., Chen, M., Liu, Y., & Liu, X. D. (2019). Sensitivity of a ratio vegetation index derived from hyperspectral remote sensing to the brown planthopper stress on rice plants. Sensors, 19(2), 375.
35. The XRF Company. (2026). Olympus DELTA DS-4000 Handheld XRF Analyzer Specifications. Cited at:026.04.14. Available at: https://shop.xrfcompany.com/products/xrf-company-certified-preowned-used-handheld-xrf-analyzer-olympus-delta-ds-4000-cash-for-gold-soil-sku-695
36. Vera-Esmeraldas, A., Pizarro-Oteíza, S., Labbé, M., Rojo, F., & Salazar, F. (2025). UAV-Based Spectral and Thermal Indices in Precision Viticulture: A Review of NDVI, NDRE, SAVI, GNDVI, and CWSI. Agronomy, 15(11), 2569.
37. Yadav, S. K. (2010). Heavy metals toxicity in plants: an overview on the role of glutathione and phytochelatins in heavy metal stress tolerance of plants. South African Journal of Botany, 76(2), 167–179
38. Zha, H., He, X., Ding, C., Gu, M., & Simon, H. (2001). Spectral relaxation for k-means clustering. Advances in neural information processing systems, 14.
39. Zhai, Y., Zhou, L., Qi, H., Gao, P., & Zhang, C. (2023). Application of visible/near-infrared spectroscopy and hyperspectral imaging with machine learning for high-throughput plant heavy metal stress phenotyping: A review. Plant Phenomics, 5, 0124.
40. Zhang, Z., Liu, M., Liu, X., & Zhou, G. (2018). A new vegetation index based on multitemporal Sentinel-2 images for discriminating heavy metal stress levels in rice. Sensors, 18(7), 2172.
41. 沈再木。(2006)。土壤污染與農產品安全。農業部農業知識入口網。引用時間2026.03.02.。Available at: https://kmweb.moa.gov.tw/knowledgebase.php?id=13300
42. 陳吉村。(1993)。重金屬對農業的影響。花蓮區農業專訊，6:8-10。
43. 監察院。(2010)。煉鋼爐碴、集塵灰非法棄置，監察院糾正環保署、工業局及高雄縣、市政府。引用時間 2026.03.10。Available at:https://www.cy.gov.tw/News_Content.aspx?n=124&sms=8912&s=5208
44. 監察院。(2019)。廢棄物非法棄置監控不力糾正案。引用時間 2026.03.02。Available at: https://www.cy.gov.tw/News_Content.aspx?n=124&sms=8912&s=5208
45. 環境部。(2009)。環保署說明高雄大坪頂地區7處非法棄置場址調查結果。引用時間 2026.03.10。Available at: https://enews.moenv.gov.tw/Page/3B3C62C78849F32F/ad51b016-aa22-4666-b94c-dad174f86960
46. 環境部。(2020)。世紀之毒：安順廠污染案與鹿耳門悲歌。引用時間 2026.03.10。Available at: https://topic.moenv.gov.tw/edcs/cp-362-8767-1ae4d-6.html
47. 環境部。(2023)。土壤污染評估調查及檢測作業管理辦法。引用時間 2026.03.02。Available at: https://law.moj.gov.tw/LawClass/LawAll.aspx?pcode=O0020042
48. 環境部。(2017)。化學工業的歷史共業：戴奧辛。引用時間 2026.03.02。Available at: https://www.cha.gov.tw/cp-263-2730-6fd8b-1.html
49. 環境部環境檢驗所。(2009)。土壤檢測方法總則 (NIEA S103.61C)。引用時間 2026.03.02。Available at: https://www.nera.gov.tw/upload/old/fe7b4ed0-fbab-499f-8cdf-5de95542e238/S10361C.pdf
50. 蔡佳均。(2022)。機載高光譜影像分析植物光譜特徵與戴奧辛濃度關聯性-以中石化安順廠為例，國立成功大學資源工程系碩士論文。