應用衛星遙測資料於土地利用回歸模型推估細懸浮微粒之時空分布

隨著工業化及都市化快速發展，空氣污染相關議題逐漸受到重視，其中細懸浮微粒(fine particulate matter, PM2.5)已有研究證實危害人體健康與干擾能見度，因而備受關注。為瞭解PM2.5之時空分布，我國環保署及地方環保局已建置多座空氣品質測站，作為污染濃度及暴露評估之依據；然而，受限於地形和建置成本等因素，使得測站分布不均、數量有限，進而造成估算上的誤差。本研究為解決此問題，以佔我國人口約40%的北部空品區為研究區域，PM2.5為研究目標污染物，利用土地利用回歸模型(land use regression, LUR)整合氣象、土地利用型態、交通等訊息，並結合衛星遙測資料，提升資料覆蓋範圍及時空連續性，以推估更準確的PM2.5濃度之時空分布。研究所使用的衛星遙測資料包含氣膠光學深度(aerosol optical depth, AOD)、衛星影像之土地覆蓋分類(land cover classification)及邊界層高度(boundary layer height, BLH)，其中AOD可反映大氣垂直面的消光性，且時空上與PM2.5濃度高度相關，亦可有效提升LUR模型效能。此外，利用Google Earth Engine (GEE)平台所訓練之土地覆蓋分類，能有效替代傳統土地利用資料，不僅能取得更即時的土地利用訊息，亦具有和傳統LUR模型一樣的推估效能。本研究應用衛星遙測資料所建立的LUR模型，年尺度之Adjusted R2可達0.79，季節尺度則以秋季最高可達0.74；留一交叉驗證法(leave-one-out cross validation, LOOCV)模型驗證結果則顯示，年尺度模型R2為0.79，季節尺度則最高可達0.74，證實土地覆蓋分類結果及AOD等衛星資料的加入，可顯著提升模型效能並建立一穩健的濃度推估模型。此外，LUR模型所推估的濃度分布，能呈現空間上細微的差異性，有助於更準確地辨識各季節污染情形以及其污染熱點，作為空污改善政策之依據。

關鍵字

細懸浮微粒；氣膠光學深度；土地利用回歸模型；土地覆蓋分類；空氣污染

並列摘要

With the rapid development of industrialization and urbanization, air pollution issues are paid more attention. Recent studies have indicated that excessive fine particle matter (PM2.5) concentration in the atmosphere is hazardous to human health and even interferes with visibility. Therefore, the issue of PM2.5 has been receiving much attention in visual feeling. In order to understand the spatial and temporal distribution of PM2.5, Taiwan EPA and local environmental protection bureaus have built air quality monitoring stations to serve as a background for levels of pollutant concentration and exposure assessment. However, due to topographical constraints and construction costs, the geographical distribution of PM2.5 monitoring stations is uneven and the number of stations is limited. These dilemmas have caused assessment bias. In order to solve this problem, the northern air quality control area, where Taiwan’s 40% population live in, was used as the study area. By integrating information such as meteorology, land use inventory, traffic data, etc., an integrated and applicable land use regression model (LUR) for PM2.5 was built. It is worth noting that we added satellite remote sensing data, which was considered to improve the coverage of the data with temporal and spatial continuity information, to estimate more accurate spatio-temporal distribution of PM2.5 concentration. The satellite remote sensing data used in the study include aerosol optical depth (AOD), Landsat 8 land cover classification and boundary layer height (BLH); Among them, AOD reflect the extinction of the vertical plane of the atmosphere, which is highly correlated with the spatio-temporal distribution of PM2.5 concentration, and improve the performance of the LUR model effectively. Land cover classification trained by the Google Earth Engine (GEE) platform can effectively replace traditional land use data, not only to obtain more real-time land use information, but also to have the same performance as the traditional LUR model. The results show that the LUR model established by satellite remote sensing data in this study performs well; the Adjusted R2 of the annual model was 0.79, and the seasonal model performed best in autumn, Adjusted R2 was 0.74. In the model validation results, the annual scale model LOOCV (leave-one-out cross validation) R2 was 0.79, and the seasonal scale model LOOCV R2 reached 0.74. This method and development showing the effects of land cover classification and the addition of remote sensing data such as AOD can effectively improve the model performance and build a robust concentration estimation model.

並列關鍵字

Fine particulate matter ； Aerosol optical depth ； Land use regression model ； Land cover classification ； Air pollution

參考文獻

Abdul-Wahab, S. A., Bakheit, C. S., Al-Alawi, S. M. (2005). Principal component and multiple regression analysis in modelling of ground-level ozone and factors affecting its concentrations. Environmental Modelling Software, 20(10), 1263-1271. https://doi.org/10.1016/j.envsoft.2004.09.001

Google Scholar

Adams, M. D., Kanaroglou, P. S. (2016). Mapping real-time air pollution health risk for environmental management: Combining mobile and stationary air pollution monitoring with neural network models. J Environ Manage, 168, 133-141. https://doi.org/10.1016/j.jenvman.2015.12.012

Google Scholar

Beelen, R., Hoek, G., Vienneau, D., Eeftens, M., Dimakopoulou, K., Pedeli, X., Tsai, M.-Y., Künzli, N., Schikowski, T., Marcon, A., Eriksen, K. T., Raaschou-Nielsen, O., Stephanou, E., Patelarou, E., Lanki, T., Yli-Tuomi, T., Declercq, C., Falq, G., Stempfelet, M., . . . de Hoogh, K. (2013). Development of NO2 and NOx land use regression models for estimating air pollution exposure in 36 study areas in Europe – The ESCAPE project. Atmospheric Environment, 72, 10-23. https://doi.org/10.1016/j.atmosenv.2013.02.037

Google Scholar

Chang, T.-Y., Tsai, C.-C., Wu, C.-F., Chang, L.-T., Chuang, K.-J., Chuang, H.-C., Young, L.-H. (2021). Development of land-use regression models to estimate particle mass and number concentrations in Taichung, Taiwan. Atmospheric Environment, 252. https://doi.org/10.1016/j.atmosenv.2021.118303

Google Scholar

Chen, G., Li, S., Knibbs, L. D., Hamm, N. A. S., Cao, W., Li, T., Guo, J., Ren, H., Abramson, M. J., Guo, Y. (2018). A machine learning method to estimate PM2.5 concentrations across China with remote sensing, meteorological and land use information. Sci Total Environ, 636, 52-60. https://doi.org/10.1016/j.scitotenv.2018.04.251

Google Scholar

主題瀏覽