懸浮微粒 (Particulate Matter, PM) 是空氣中的微小顆粒物，PM2.5則是直徑小於2.5 μm的細懸浮微粒。PM2.5能深入呼吸道，對人體健康造成嚴重危害，例如增加心血管和呼吸系統疾病的風險。因此，準確地預測民眾所在地的當前PM2.5濃度及未來濃度成為重要議題之一。臺灣擁有約78個國家空氣品質監測站，這些站點使用高精度儀器提供高品質監測數據，但由於成本高昂，其密度較低。相比之下，環境部設置了超過10,000個智慧城鄉空品微型感測器 (以下簡稱：微型感測器) ，雖然這些感測器的分布密度高，但未經實驗室級的校正，準確性有限。此外，空氣汙染預測相關研究多針對單一測站的時間序列資料進行預測，或僅用簡單空間內插法產製推估圖，缺乏更精準的全空間預測圖。本研究以臺中市為研究試區，使用中央研究院開發的AS-LUNG感測器資料，進行微型感測器PM2.5準確性的校正。AS-LUNG經過研究級儀器GRIMM比對校正，其監測資料準確度高。本研究使用2021年12月及2022年1月和2月期間的AS-LUNG高精度PM2.5監測資料，結合溫度、濕度、經緯度以及微型感測器與AS-LUNG之間的距離來建立校正模型。該校正模型採用隨機森林 (Random Forest Regression, RFR) 演算法，以提高微型感測器數據的準確性，並使用資料拆分與十折交叉驗證作為驗證方法。建立校正模型後，再使用校正後之微型感測器PM2.5濃度、溫度、濕度、風速與風向資料建立未來數小時空汙預測模型。本研究使用長短期記憶模型 (Long Short-Term Memory, LSTM) 作為預測方法，該模型能有效地捕捉時間序列中的複雜模式和趨勢，以預測未來24小時的PM2.5濃度。研究中使用2022年1月期間之監測資料建立模型，並使用2022年2月1日之微型感測器資料作為未來數小時空汙預測模型輸入資料，以預測2月2日之PM2.5濃度。最後，本研究使用微型感測器PM2.5之24小時預測值，結合PM2.5相關的土地利用重要變數，其變數資料庫包含國土利用調查資料、遙測植生監測資料庫、地標資料庫、路網數值資料庫、地形資料庫與大型排放源資料庫等，透過RFR演算法建立未來數小時空間變異推估模型，以獲得每小時PM2.5濃度時空變異預測圖，並本研究也以未加入模型的20% 測站資料代表無測站地區，以檢驗未來每小時無測站地區推估成果。結果表明，校正模型之判定係數 (Coefficient of determination, R2) 為0.84，平均絕對誤差 (Mean Absolute Error, MAE) 為3.81 μg/m^3；未來數小時空汙預測模型之R2為0.91，MAE為2.41 μg/m^3；透過未來數小時空間變異推估模型本研究繪製出了全臺中市範圍未來24小時PM2.5時空變異圖，且代表無測站地區的20% 測站資料未來24小時預測值之平均R2為0.95，平均MAE則為2.17 μg/m^3。綜上所述，校正模型顯著提高了微型感測器的數據準確性，使用LSTM未來數小時空汙預測模型則能有效預測未來24小時的PM2.5濃度。此外，未來數小時空間變異推估模型能準確估算全臺中市範圍，即包含無測站區域的PM2.5濃度，顯示出本研究方法在空氣品質監測和預測中的潛力和應用價值。

The study aims to predict PM2.5 concentrations in Taichung City, Taiwan, to mitigate health risks associated with fine particulate matter. Using high-precision AS-LUNG sensor data, the research calibrates the less accurate microsensor data and develops comprehensive spatial and temporal prediction models. The methodology involves calibrating microsensor PM2.5 data with Random Forest Regression (RFR) using AS-LUNG data from December 2021 to February 2022. This calibration improves microsensor accuracy. Subsequently, a Long Short-Term Memory (LSTM) model predicts PM2.5 concentrations for the next 24 hours using the calibrated data and meteorological factors. Finally, an RFR-based spatial estimation model, incorporating land-use/land cover, Normalized Difference Vegetation Index (NDVI), road network, terrain, and emission source data, generates hourly PM2.5 variation maps. Results indicate that the calibration model significantly enhances microsensor data accuracy, achieving an coefficient of determination (R2) of 0.84 and a Mean Absolute Error (MAE) of 3.81 μg/m^3. The LSTM prediction model effectively forecasts 24-hour PM2.5 concentrations with an R2 of 0.91 and an MAE of 2.41 μg/m^3. The spatial estimation model accurately predicts PM2.5 variations, with an average R2 of 0.95 and an MAE of 2.17 μg/m^3 for unmonitored areas. In conclusion, the study's approach markedly improves microsensor data reliability and provides robust PM2.5 predictions, demonstrating its potential for enhancing air quality monitoring and public health protection.

Ayturan, Y. A., Ayturan, Z. C., ALTUN, H. O., Kongoli, C., Tuncez, F. D., Dursun, Ş., & ÖZTÜRK, A. (2020). Short-term prediction of PM2. 5 pollution with deep learning methods. https://doi.org/10.30955/gnj.003208
Brook, R. D., Rajagopalan, S., Pope III, C. A., Brook, J. R., Bhatnagar, A., Diez-Roux, A. V., ... & Kaufman, J. D. (2010). Particulate matter air pollution and cardiovascular disease: an update to the scientific statement from the American Heart Association. Circulation, 121(21), 2331-2378. https://doi.org/10.1161/CIR.0b013e3181dbece1
Chae, S., Shin, J., Kwon, S., Lee, S., Kang, S., & Lee, D. (2021). PM10 and PM2. 5 real-time prediction models using an interpolated convolutional neural network. Scientific Reports, 11(1), 11952. https://doi.org/10.1038/s41598-021-91253-9
Dai, X., Liu, J., & Li, Y. (2021). A recurrent neural network using historical data to predict time series indoor PM2. 5 concentrations for residential buildings. Indoor air, 31(4), 1228-1237. https://doi.org/10.1111/ina.12794
Ding, W., & Sun, H. (2023). Prediction of PM2. 5 concentration based on the weighted RF-LSTM model. Earth Science Informatics, 16(4), 3023-3037. https://doi.org/10.1007/s12145-023-01111-7
Dominici, F., Peng, R. D., Bell, M. L., Pham, L., McDermott, A., Zeger, S. L., & Samet, J. M. (2006). Fine particulate air pollution and hospital admission for cardiovascular and respiratory diseases. Jama, 295(10), 1127-1134. https://doi.org/10.1001/jama.295.10.1127
Fleischer, N. L., Merialdi, M., van Donkelaar, A., Vadillo-Ortega, F., Martin, R. V., Betran, A. P., ... & Marie S. O´ Neill. (2014). Outdoor air pollution, preterm birth, and low birth weight: analysis of the world health organization global survey on maternal and perinatal health. Environmental health perspectives, 122(4), 425-430. https://doi.org/10.1289/ehp.1306837
Gul, S., Khan, G. M., & Yousaf, S. (2022). Multi-step short-term PM 2.5 forecasting for enactment of proactive environmental regulation strategies. Environmental Monitoring and Assessment, 194(5), 386. https://doi.org/10.1007/s10661-022-10029-4
Karimian, H., Li, Q., Wu, C., Qi, Y., Mo, Y., Chen, G., ... & Sachdeva, S. (2019). Evaluation of different machine learning approaches to forecasting PM2. 5 mass concentrations. Aerosol and Air Quality Research, 19(6), 1400-1410. https://doi.org/10.4209/aaqr.2018.12.0450
Kristiani, E., Lin, H., Lin, J. R., Chuang, Y. H., Huang, C. Y., & Yang, C. T. (2022). Short-term prediction of PM2. 5 using LSTM deep learning methods. Sustainability, 14(4), 2068. https://doi.org/10.3390/su14042068
Li, X., Peng, L., Yao, X., Cui, S., Hu, Y., You, C., & Chi, T. (2017). Long short-term memory neural network for air pollutant concentration predictions: Method development and evaluation. Environmental pollution, 231, 997-1004. https://doi.org/10.1016/j.envpol.2017.08.114
Li, T., Zhang, Y., Wang, J., Xu, D., Yin, Z., Chen, H., ... & Shi, X. (2018). All-cause mortality risk associated with long-term exposure to ambient PM2· 5 in China: a cohort study. The Lancet Public Health, 3(10), e470-e477. https://doi.org/10.1016/S2468-2667(18)30144-0
Li, L., Girguis, M., Lurmann, F., Pavlovic, N., McClure, C., Franklin, M., ... & Habre, R. (2020). Ensemble-based deep learning for estimating PM2. 5 over California with multisource big data including wildfire smoke. Environment international, 145, 106143. https://doi.org/10.1016/j.envint.2020.106143
Li, C., van Donkelaar, A., Hammer, M. S., McDuffie, E. E., Burnett, R. T., Spadaro, J. V., ... & Martin, R. V. (2023). Reversal of trends in global fine particulate matter air pollution. Nature Communications, 14(1), 5349. https://doi.org/10.1038/s41467-023-41086-z
Lin, L., Chen, C. Y., Yang, H. Y., Xu, Z., & Fang, S. H. (2020). Dynamic system approach for improved PM 2.5 prediction in Taiwan. IEEE Access, 8, 210910-210921. https://doi.org/10.1109/ACCESS.2020.3038853
Lo, C. Y., Huang, W. H., Ho, M. F., Sun, M. T., Chen, L. J., Sakai, K., & Ku, W. S. (2020). Recurrent learning on PM 2.5 prediction based on clustered airbox dataset. IEEE Transactions on Knowledge and Data Engineering, 34(10), 4994-5008. https://doi.org/10.1109/TKDE.2020.3047634
Lung, S. C. C., Wang, W. C. V., Wen, T. Y. J., Liu, C. H., & Hu, S. C. (2020). A versatile low-cost sensing device for assessing PM2. 5 spatiotemporal variation and quantifying source contribution. Science of The Total Environment, 716, 137145. https://doi.org/10.1016/j.scitotenv.2020.137145
Luo, Z., Huang, F., & Liu, H. (2020). PM2. 5 concentration estimation using convolutional neural network and gradient boosting machine. Journal of Environmental Sciences, 98, 85-93. https://doi.org/10.1016/j.jes.2020.04.042
Murray, C. J., Aravkin, A. Y., Zheng, P., Abbafati, C., Abbas, K. M., Abbasi-Kangevari, M., ... & Borzouei, S. (2020). Global burden of 87 risk factors in 204 countries and territories, 1990–2019: a systematic analysis for the Global Burden of Disease Study 2019. The lancet, 396(10258), 1223-1249. https://doi.org/10.1016/S0140-6736(20)30752-2
Ni, J., Chen, Y., Gu, Y., Fang, X., & Shi, P. (2022). An improved hybrid transfer learning-based deep learning model for PM2. 5 concentration prediction. Applied Sciences, 12(7), 3597. https://doi.org/10.3390/app12073597
Peralta, B., Sepúlveda, T., Nicolis, O., & Caro, L. (2022). Space-time prediction of PM2. 5 concentrations in Santiago de Chile using LSTM Networks. Applied Sciences, 12(22), 11317. https://doi.org/10.3390/app122211317
Pope Iii, C. A., Burnett, R. T., Thun, M. J., Calle, E. E., Krewski, D., Ito, K., & Thurston, G. D. (2002). Lung cancer, cardiopulmonary mortality, and long-term exposure to fine particulate air pollution. Jama, 287(9), 1132-1141. https://doi.org/10.1001/jama.287.9.1132
Rentschler, J., & Leonova, N. (2023). Global air pollution exposure and poverty. Nature communications, 14(1), 4432. https://doi.org/10.1038/s41467-023-39797-4
Shapley, L. S. (1953). A value for n-person games. https://doi.org/10.1515/9781400829156-012
Wang, W. C. V., Lung, S. C. C., & Liu, C. H. (2020). Application of machine learning for the in-field correction of a PM2. 5 low-cost sensor network. Sensors, 20(17), 5002. https://doi.org/10.3390/s20175002
Wong, P. Y., Lee, H. Y., Chen, Y. C., Zeng, Y. T., Chern, Y. R., Chen, N. T., ... & Wu, C. D. (2021). Using a land use regression model with machine learning to estimate ground level PM2. 5. Environmental Pollution, 277, 116846. https://doi.org/10.1016/j.envpol.2021.116846
Wu, C. D., Zeng, Y. T., & Lung, S. C. (2018). A hybrid kriging/land-use regression model to assess PM2. 5 spatial-temporal variability. Sci Total Environ, 645, 1456-1464. https://doi.org/10.1016/j.scitotenv.2018.07.073
Xayasouk, T., Lee, H., & Lee, G. (2020). Air pollution prediction using long short-term memory (LSTM) and deep autoencoder (DAE) models. Sustainability, 12(6), 2570. https://doi.org/10.3390/su12062570
Xu, X., Shi, K., Huang, Z., & Shen, J. (2023). What factors dominate the change of PM2. 5 in the world from 2000 to 2019? A study from multi-source data. International Journal of Environmental Research and Public Health, 20(3), 2282. https://doi.org/10.3390/ijerph20032282
Zhou, S., Wang, W., Zhu, L., Qiao, Q., & Kang, Y. (2024). Deep-learning architecture for PM2. 5 concentration prediction: A review. Environmental Science and Ecotechnology, 100400. https://doi.org/10.1016/j.ese.2024.100400
中華民國內政部戶政司，(2024)。戶籍人口統計速報。Retrieved 7/4 from https://www.ris.gov.tw/app/portal/346
中華民國交通部公路局，(2024)。機動車輛登記數。Retrieved 7/4 from https://stat.motc.gov.tw/mocdb/stmain.jsp?sys=100&funid=a3301
環境部，(2024)。細懸浮微粒一日變化特徵。Retrieved 7/4 from https://stat.thb.gov.tw/hb01/webMain.aspx?sys=100&funid=11100
環境部，(2024)。臺灣空汙哪裡來。Retrieved 7/4 from https://air.moenv.gov.tw/Encyclopedia/Classification_1.aspx
環境部，(2024)。中央監測站背景介紹。Retrieved 7/6 from。https://airtw.moenv.gov.tw/CHT/EnvMonitoring/Central/Background_Intro.aspx
環境部，(2024)。空氣品質國家監測站、智慧城鄉微型感測器與校園民間微型感測器的差異。Retrieved 7/6 from https://airtw.moenv.gov.tw/cht/Encyclopedia/AirSensor/AirSensor_2.aspx
交通部中央氣象署，(2021)。氣候月平均。Retrieved 7/12 from https://www.cwa.gov.tw/V8/C/C/Statistics/monthlymean.html