利用土地利用回归模型与集成机器学习方法评估中国PM2.5污染时空分布
Assessing the Spatiotemporal Distribution of PM2.5 Pollution in China Using Land Use Regression Model and Ensemble Machine Learning Methods
DOI: 10.12677/aep.2026.1611170, PDF,    科研立项经费支持
作者: 马健容, 廖 鸿, 杨昊林, 崔 潆, 马雪怡, 张 平:西安工程大学环境与化学工程学院,陕西 西安
关键词: PM2.5;土地利用回归模型;集成机器学习方法;空间自相关;PM2.5; Land Use Regression Model; Ensemble Machine Learning Methods; Spatial Autocorrelation
摘要: 随着社会经济快速发展与城市化、工业化进程加快推进,PM2.5污染日益受到广泛关注,准确模拟PM2.5污染空间格局,是制定大气污染防控措施和开展健康效应评估的重要基础。本研究基于随机森林(RF)、类别特征提升(CatBoost)和极端梯度提升(XGBoost)三种机器学习方法,构建土地利用回归–时空加权(LUR-SW)机器学习集成模型,融合2018~2020年气溶胶光学厚度(AOD)、ERA5气象数据、土地利用、污染源及植被指数等多源遥感数据,模拟中国PM2.5浓度的时空分布,为健康暴露风险精确评估与城市可持续发展提供科学支撑。结果表明,LUR建模变量中AOD与PM2.5浓度呈正相关,而气温和降水均与PM2.5浓度呈负相关。CatBoost、XGBoost和RF模型的R2分别为0.86、0.89和0.84,表明三种模型均具有良好的预测性能,其中XGBoost模型的预测性能最优。PM2.5浓度呈现明显的季节性变化特征,冬季最高、夏季最低,且在空间上呈强正自相关性,空间分布以高–高聚集以及低–低聚集特征为主。研究结果可为PM2.5污染精细化模拟、预警区识别与暴露风险量化提供可靠依据。
Abstract: With rapid socio-economic development and the accelerating pace of urbanization and industrialization, PM2.5 pollution has increasingly come under the spotlight. Accurately modelling the spatial distribution of PM2.5 pollution is a crucial foundation for formulating air pollution control measures and conducting health impact assessments. This study utilizes three machine learning methods—Random Forest (RF), Category Feature Boosting (CatBoost) and Extreme Gradient Boosting (XGBoost)—to construct a Land Use Regression-Spartoi-Temporal Weighting (LUR-SW) machine learning ensemble model. By integrating multi-source remote sensing data—including aerosol optical depth (AOD) from 2018 to 2020, ERA5 meteorological data, land use, pollution sources and vegetation indices, to simulate the spatiotemporal distribution of PM2.5 concentrations in China, thereby providing scientific support for the precise assessment of health exposure risks and urban sustainable development. The results indicate that, among the LUR modelling variables, AOD is positively correlated with PM2.5 concentrations, whilst both air temperature and precipitation are negatively correlated with PM2.5 concentrations. The R2 values of the CatBoost, XGBoost, and RF models were 0.86, 0.89, and 0.84, respectively, indicating that all three models exhibited good predictive performance, with XGBoost achieving the best performance. PM2.5 concentrations exhibit distinct seasonal variations, peaking in winter and reaching their lowest levels in summer. Spatially, they display strong positive autocorrelation, with spatial distribution characterized primarily by “high-high” and “low-low” clustering. The findings of this study provide a reliable basis for the detailed modelling of PM2.5 pollution, the identification of early-warning zones, and the quantification of exposure risks.
文章引用:马健容, 廖鸿, 杨昊林, 崔潆, 马雪怡, 张平. 利用土地利用回归模型与集成机器学习方法评估中国PM2.5污染时空分布[J]. 环境保护前沿, 2026, 16(11): 1692-1704. https://doi.org/10.12677/aep.2026.1611170

参考文献

[1] Li, R., Ma, T., Xu, Q. and Song, X. (2018) Using MAIAC AOD to Verify the PM2.5 Spatial Patterns of a Land Use Regression Model. Environmental Pollution, 243, 501-509.
https://doi.org/10.1016/j.envpol.2018.09.026
[2] Wei, J., Huang, W., Li, Z., Xue, W., Peng, Y., Sun, L., et al. (2019) Estimating 1-Km-Resolution PM2.5 Concentrations across China Using the Space-Time Random Forest Approach. Remote Sensing of Environment, 231, Article 111221.
https://doi.org/10.1016/j.rse.2019.111221
[3] Wu, C., Zeng, Y. and Lung, S.C. (2018) A Hybrid Kriging/Land-Use Regression Model to Assess PM2.5 Spatial-Temporal Variability. Science of The Total Environment, 645, 1456-1464.
https://doi.org/10.1016/j.scitotenv.2018.07.073
[4] Huang, L., Zhang, C. and Bi, J. (2017) Development of Land Use Regression Models for PM2.5, SO2, NO2 and O3 in Nanjing, China. Environmental Research, 158, 542-552.
https://doi.org/10.1016/j.envres.2017.07.010
[5] Han, L., Zhao, J., Gao, Y., Gu, Z., Xin, K. and Zhang, J. (2020) Spatial Distribution Characteristics of PM2.5 and PM10 in Xi’an City Predicted by Land Use Regression Models. Sustainable Cities and Society, 61, Article 102329.
https://doi.org/10.1016/j.scs.2020.102329
[6] Briggs, D.J., Collins, S., Elliott, P., Fischer, P., Kingham, S., Lebret, E., et al. (1997) Mapping Urban Air Pollution Using GIS: A Regression-Based Approach. International Journal of Geographical Information Science, 11, 699-718.
https://doi.org/10.1080/136588197242158
[7] 刘炳杰, 彭晓敏, 李继红. 基于LUR模型的中国PM2.5时空变化分析[J]. 环境科学, 2018, 39(12): 5296-5307.
[8] 杨明亮, 朱宗玖. 基于LUR模型下PM2.5浓度的空间分布模拟分析[J]. 大气与环境光学学报, 2022, 17(3): 347-359.
[9] Wang, Z., Li, R., Chen, Z., Yao, Q., Gao, B., Xu, M., et al. (2022) The Estimation of Hourly PM2.5 Concentrations across China Based on a Spatial and Temporal Weighted Continuous Deep Neural Network (STWC-DNN). ISPRS Journal of Photogrammetry and Remote Sensing, 190, 38-55.
https://doi.org/10.1016/j.isprsjprs.2022.05.011
[10] Wei, J., Liu, S., Li, Z., Liu, C., Qin, K., Liu, X., et al. (2022) Ground-Level NO2 Surveillance from Space across China for High Resolution Using Interpretable Spatiotemporally Weighted Artificial Intelligence. Environmental Science & Technology, 56, 9988-9998.
https://doi.org/10.1021/acs.est.2c03834
[11] Li, T., Wang, Y. and Wu, J. (2024) Deriving PM2.5 from Satellite Observations with Spatiotemporally Weighted Tree-Based Algorithms: Enhancing Modeling Accuracy and Interpretability. npj Climate and Atmospheric Science, 7, Article No. 138.
https://doi.org/10.1038/s41612-024-00692-4
[12] Ho, C., Chan, C., Cho, C., Lin, H., Lee, J. and Wu, C. (2015) Land Use Regression Modeling with Vertical Distribution Measurements for Fine Particulate Matter and Elements in an Urban Area. Atmospheric Environment, 104, 256-263.
https://doi.org/10.1016/j.atmosenv.2015.01.024
[13] 吴健生, 谢舞丹, 李嘉诚. 土地利用回归模型在大气污染时空分异研究中的应用[J]. 环境科学, 2016, 37(2): 413-419.
[14] 张平, 张凤倩, 朱慧敏, 等. 基于FA-SVM优化LUR模型的汾渭平原PM2.5时空格局模拟[J]. 西安工程大学学报, 2025, 39(3): 89-101.
[15] Yu, W., Li, S., Ye, T., Xu, R., Song, J. and Guo, Y. (2022) Deep Ensemble Machine Learning Framework for the Estimation of PM2.5 Concentrations. Environmental Health Perspectives, 130, Article 37004.
https://doi.org/10.1289/ehp9752