中国黑碳和PM2.5时空分布预测的多源数据驱动机器学习模型的研究

中国快速的工业化与高人口密度导致大气颗粒物排放显著增多，城市空气污染状况与健康风险日益严峻，部分原因在于黑碳（Black Carbon, BC）气溶胶粒子所具有的强烈光吸收特性和高吸附能力。黑碳颗粒能够吸附诸如多环芳烃和重金属等多种有害物质，这进一步增强了PM_2.5颗粒的毒性成分。精准评估黑碳和PM_2.5气溶胶空分布特征，对精细化大气污染防治与气溶胶的气候辐射效应评估至关重要。

卫星遥感技术是目前大范围获取大气气溶胶信息的常用手段，能够弥补常规地面观测在空间覆盖上的局限性。然而，由于非线性辐射传输十分复杂、地表参数在卫星遥感成像的高敏感型以及卫星观测约束较少，传统卫星气溶胶产品的精度不高。

本研究构建了多种机器学习模型用于黑碳的时空分布预测（R²=0.77，RMSE=1.37 μg/m³），汇聚了涵盖多部门污染物排放清单及卫星遥感等多元数据源，为模型构建奠定了坚实的数据基础。利用机器学习技术在解决非线性问题方面的特长，对多种机器学习模型的预测性能进行了细致比较。此外，通过深入剖析模型内部变量的敏感性，提升了对大气污染物时空演变规律的认知与预测精度。研究进一步证实了融合污染物排放清单信息的多源数据模型在预测不同类型污染物时空分布时具有广泛适用性和可靠性，并成功地将这一方法应用于对PM_2.5时空浓度的精确预测。

本研究提出了一种“动态窗口”机器学习策略，该策略进一步耦合了排放清单数据，不仅可用于计算PM_2.5当前浓度，还能预测短期内PM_2.5浓度分布。

研究表明，基于机器学习算法（例如随机森林、XGBoost）构建的预测模型，能有效整合多种数据源，可以提升现有黑碳和PM_2.5浓度时空分布数据的分辨率和准确度，为污染分布和变化特征分析、政策制定以及深入研究黑碳对环境和气候影响提供了有价值的参考数据。

[1] YEGANEH B, KHUZESTANI R B, TAHERI A, et al.Temporal trends in the spatial-scale contributions to black carbon in a Middle Eastern megacity[J].The Science of the total environment, 2021, 792:148364.DOI:10.1016/j.scitotenv.2021.148364.
[2] 秦世广,汤洁,温玉璞.黑碳气溶胶及其在气候变化研究中的意义[J].气象, 2001, 27(11):3-7.DOI:10.7519/j.issn.1000-0526.2001.11.001.
[3] 许黎,王亚强,陈振林,等.黑碳气溶胶研究进展Ⅰ:排放,清除和浓度[J].地球科学进展, 2006, 21(4):9.DOI:CNKI:SUN:DXJZ.0.2006-04-003.
[4] BELLOUIN N, Quaas J, Gryspeerdt E, et al.Bounding Global Aerosol Radiative Forcing of Climate Change[J].Reviews of Geophysics, 2020, 58(1).DOI:10.1029/2019RG000660.
[5] 蒋磊,汤莉莉,潘良宝,等.南京冬季重污染过程中黑碳气溶胶的混合态及粒径分布[J].环境科学, 2017.DOI:10.13227/j.hjkx.201605167.
[6] XING Y F, Xu Y H, Shi M H, et al.The impact of PM2.5 on the human respiratory system[J].Journal of Thoracic Disease, 2016, 8(1):E69-E74.DOI:10.3978/j.issn.2072-1439.2016.01.19.
[7] LIN Y, ZOU J, YANG W, et al. A review of recent advances in research on PM2. 5 in China[J]. International journal of environmental research and public health, 2018, 15(3): 438.
[8] YUAN Q, SHEN H, LI T, et al. Deep learning in environmental remote sensing: Achievements and challenges[J]. Remote Sensing of Environment, 2020, 241: 111716.
[9] POPE C A , Bates D V , Raizenne M E .Health effects of particulate air pollution: time for reassessment?[J].National Institute of Environmental Health Science, 1995(5).DOI:10.1289/EHP.95103472.
[10] PAN B. Application of XGBoost algorithm in hourly PM2. 5 concentration prediction[C]//IOP conference series: earth and environmental science. IOP publishing, 2018, 113: 012127.
[11] LI Y, LIU S, BASHIRI KHUZESTANI R, et al. Emission-Based Machine Learning Approach for Large-Scale Estimates of Black Carbon in China[J]. Remote Sensing, 2024, 16(5): 837.
[12] FUNG P L, ZAIDAN M A, TIMONEN H, et al. Evaluation of white-box versus black-box machine learning models in estimating ambient black carbon concentration[J]. Journal of aerosol science, 2021, 152: 105694.
[13] WEI J, LI Z, CRIBB M, et al. Improved 1 km resolution PM2.5 estimates across China using enhanced space–time extremely randomized trees[J/OL]. Atmospheric Chemistry and Physics, 2020: 3273-3289. http://dx.doi.org/10.5194/acp-20-3273-2020. DOI:10.5194/acp-20-3273-2020.
[14] PARK S, LEE J, IM J, et al. Estimation of spatially continuous daytime particulate matter concentrations under all sky conditions through the synergistic use of satellite-based AOD and numerical models[J]. Science of the total environment, 2020, 713: 136516.
[15] SUN Y, ZENG Q, GENG B, et al. Deep learning architecture for estimating hourly ground-level PM2.5 using satellite remote sensing[J]. IEEE Geoscience and Remote Sensing Letters, 2019, 16(9): 1343-1347.
[16] DE HOOGH K, HÉRITIER H, STAFOGGIA M, et al. Modelling daily PM2.5 concentrations at high spatio-temporal resolution across Switzerland[J]. Environmental Pollution, 2018, 233: 1147-1154.
[17] WEI J, HUANG W, LI Z, et al. Estimating 1-km-resolution PM2.5 concentrations across China using the space-time random forest approach[J]. Remote Sensing of Environment, 2019, 231: 111221.
[18] CHEN Z Y, ZHANG T H, ZHANG R, et al. Extreme gradient boosting model to estimate PM2.5 concentrations with missing-filled satellite data in China[J]. Atmospheric Environment, 2019, 202: 180-189.
[19] BAO F, LI Y, CHENG T, et al. Estimating the Columnar Concentrations of Black Carbon Aerosols in China Using MODIS Products[J/OL]. Environmental Science & Technology, 2020: 11025-11036. http://dx.doi.org/10.1021/acs.est.0c00816. DOI:10.1021/acs.est.0c00816.
[20] AWAD Y A, KOUTRAKIS P, COULL B A, et al. A spatio-temporal prediction model based on support vector machine regression: Ambient Black Carbon in three New England States[J]. Environmental research, 2017, 159: 427-434.
[21] ZHAO J, LIU Y, SHAN M, et al. Characteristics, potential regional sources and health risk of black carbon based on ground observation and MERRA-2 reanalysis data in a coastal city, China[J/OL]. Atmospheric Research, 2021: 105563. http://dx.doi.org/10.1016/j.atmosres.2021.105563. DOI:10.1016/j.atmosres.2021.105563.
[22] BREIMAN L. Random forests[J]. Machine learning, 2001, 45: 5-32.
[23] GAO M , Saide P E , Xin J ,et al.Estimates of Health Impacts and Radiative Forcing in Winter Haze in Eastern China through Constraints of Surface PM2.5 Predictions[J].Environmental Science & Technology, 2017, 51(4):2178-2185.DOI:10.1021/acs.est.6b03745.
[24] ZHONG J, ZHANG X, GUI K, et al. Reconstructing 6-hourly PM2.5 datasets from 1960 to 2020 in China[J]. Earth System Science Data Discussions, 2022, 2022: 1-21.
[25] DI Q, KOUTRAKIS P, SCHWARTZ J. A hybrid prediction model for PM2. 5 mass and components using a chemical transport model and land use regression[J]. Atmospheric environment, 2016, 131: 390-399.
[26] BAO F, CHENG T, LI Y, et al. Retrieval of black carbon aerosol surface concentration using satellite remote sensing observations[J].Remote Sensing of Environment, 2019, 226:93-108.DOI:10.1016/j.rse.2019.03.036.
[27] SHEN J , VALAGOLAM D , MCCALLA S .Prophet forecasting model: a machine learning approach to predict the concentration of air pollutants (PM2.5, PM10, O3, NO2, SO2, CO) in Seoul, South Korea[J].PeerJ, 2020, 8(3):e9961.DOI:10.7717/peerj.9961.
[28] SCHUSTER G L, DUBOVIK O, HOLBEN B N, et al. Inferring black carbon content and specific absorption from Aerosol Robotic Network (AERONET) aerosol retrievals[J]. Journal of Geophysical Research: Atmospheres, 2005, 110(D10).
[29] SCHUSTER G L , DUBOVIK O , HOLBEN B N ,et al.Inferring black carbon content and specific absorption from Aerosol Robotic Network (AERONET) aerosol retrievals[M]. 2005.
[30] WANG L , LI Z , TIAN Q ,et al.Estimate of aerosol absorbing components of black carbon, brown carbon, and dust from ground-based remote sensing data of sun-sky radiometers[J].Journal of Geophysical Research Atmospheres, 2013, 118(12):6534-6543.DOI:10.1002/jgrd.50356.
[31] 包方闻.大气气溶胶光学特性及黑碳浓度卫星遥感反演研究[D].中国科学院大学(中国科学院遥感与数字地球研究所),2018.
[32] 冯进. PM2.5监测技术的发展及测量数据准确性的保障[J]. 计量与测试技术,2014,41(2):52-54,57. DOI:10.3969/j.issn.1004-6941.2014.02.027.
[33] MA S, SHAO M, ZHANG Y, ET AL. Evaluating the performance of chemical transport models for PM2. 5 source apportionment: An integrated application of spectral analysis and grey incidence analysis[J]. Science of The Total Environment, 2022, 837: 155781.
[34] CHEN Q X, HUANG C L, YUAN Y ,et al.Assessment of aerosol types on improving the estimation of surface PM2.5 concentrations by using ground-based aerosol optical depth dataset[J].Atmospheric pollution research, 2019(6).DOI:10.1016/j.apr.2019.07.016.
[35] 沈惠中,王戎,陶澍.近五十年全球大气多环芳烃排放清单[C]//第六届全国环境化学大会暨环境科学仪器与分析仪器展览会摘要集.2011.
[36] GUO J, XIA F, ZHANG Y,et al.Impact of diurnal variability and meteorological factors on the PM2.5 - AOD relationship: Implications for PM2.5 remote sensing[J].Environmental Pollution, 2017, 221:94.DOI:10.1016/j.envpol.2016.11.043.
[37] GUI K, CHE H, WANG Y, et al. Satellite-derived PM2. 5 concentration trends over Eastern China from 1998 to 2016: Relationships to emissions and meteorological parameters[J]. Environmental pollution, 2019, 247: 1125-1133.
[38] CAO J J, ZHU C S, CHOW J C,et al.Black carbon relationships with emissions and meteorology in Xi'an, China[J].Atmospheric Research, 2009, 94(2):194-202.DOI:10.1016/j.atmosres.2009.05.009.
[39] CUI F, PEI S, CHEN M,et al.Absorption enhancement of black carbon and the contribution of brown carbon to light absorption in the summer of Nanjing, China[J].Atmospheric Pollution Research, 2020, 12(2).DOI:10.1016/j.apr.2020.12.008.
[40] JIMENEZ J L, JAYNE J T, SHI Q, et al. Ambient aerosol sampling using the aerodyne aerosol mass spectrometer[J]. Journal of Geophysical Research: Atmospheres, 2003, 108(D7).
[41] EBERT M, INERLE-HOF M, WEINBRUCH S.Environmental scanning electron microscopy as a new technique to determine the hygroscopic behaviour of individual aerosol particles[J].Atmospheric Environment, 2002, 36(39/40):5909-5916.DOI:10.1016/S1352-2310(02)00774-4.
[42] DECARLO P F, KIMMEL J R, TRIMBORN A,et al.Field-deployable, high-resolution, time-of-flight aerosol mass spectrometer.[J].Analytical Chemistry, 2006, 78(24):8281-9.DOI:10.1021/ac061249n.
[43] KHLYSTOV A, WYERS G P, SLANINA J. The steam-jet aerosol collector[J]. Atmospheric Environment, 1995, 29(17): 2229-2234.
[44] NG N L, HERNDON S C, TRIMBORN A, et al. An Aerosol Chemical Speciation Monitor (ACSM) for routine monitoring of the composition and mass concentrations of ambient aerosol[J]. Aerosol Science and Technology, 2011, 45(7): 780-794.
[45] SUN Y, WANG Z, DONG H,et al.Characterization of summer organic and inorganic aerosols in Beijing, China with an Aerosol Chemical Speciation Monitor[J].Atmospheric Environment, 2012, 51(none):250-259.DOI:10.1016/j.atmosenv.2012.01.013.
[46] KOCH D, SCHULZ M, KINNE S, et al. Evaluation of black carbon estimations in global aerosol models[J]. Atmospheric Chemistry and Physics, 2009, 9(22): 9001-9026.
[47] ZHANG Y , FU R , YU H ,et al.A regional climate model study of how biomass burning aerosol impacts land‐atmosphere interactions over the Amazon[J].Journal of Geophysical Research Atmospheres, 2008, 113(D14).DOI:10.1029/2007JD009449.
[48] REDDINGTON C L, MORGAN W T, DARBYSHIRE E,et al.Biomass burning aerosol over the Amazon: analysis of aircraft, surface and satellite observations using a global aerosol model[J].Atmospheric Chemistry and Physics, 2018:1-32.DOI:10.5194/acp-2018-849.
[49] GOGOI M M, BABU S S, IMASU R, et al. Satellite (GOSAT-2 CAI-2) retrieval and surface (ARFINET) observations of aerosol black carbon over India[J]. Atmospheric Chemistry and Physics, 2023, 23(14): 8059-8079.
[50] BI J, KNOWLAND K E, KELLER C A,et al.Combining Machine Learning and Numerical Simulation for High-Resolution PM[J].Environmental science & technology, 2022, 56(3):1544-1556.DOI:10.1021/acs.est.1c05578.
[51] STEKHOVEN D J. missForest: Nonparametric missing value imputation using random forest[J]. Astrophysics Source Code Library, 2015: ascl: 1505.011.
[52] CHEN T, GUESTRIN C. Xgboost: A scalable tree boosting system[C]//Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining. 2016: 785-794.
[53] FRIEDMAN J H.Stochastic gradient boosting[J].Computational Statistics & Data Analysis, 2002.DOI:10.1016/S0167-9473(01)00065-2.
[54] SCHAPIRE R E.The strength of weak learnability[J].Proceedings of the Second Annual Workshop on Computational Learning Theory, 1989, 5(2):197-227.DOI:10.1007/BF00116037.
[55] NATEKIN A, KNOLL A. Gradient boosting machines, a tutorial[J]. Frontiers in neurorobotics, 2013, 7: 21.
[56] FISHER A, RUDIN C, DOMINICI F.All Models are Wrong, but Many are Useful: Learning a Variable's Importance by Studying an Entire Class of Prediction Models Simultaneously[J]. 2018.DOI:10.48550/arXiv.1801.01489.
[57] 黄锴. 基于排放源的中国黑碳气溶胶浓度机器学习空间预测方法[D]. 深圳. 南方科技大学,2022.
[58] WANG R, TAO S, CIAIS P, et al. High-resolution mapping of combustion processes and implications for CO 2 emissions[J]. Atmospheric Chemistry and Physics, 2013, 13(10): 5189-5203.
[59] MENG W, ZHONG Q, YUN X, et al. Improvement of a global high-resolution ammonia emission inventory for combustion and industrial sources with new data from the residential and transportation sectors[J]. Environmental science & technology, 2017, 51(5): 2821-2829.
[60] 贾小龙, 陈丽娟, 高辉, 等. 我国短期气候预测技术进展[J]. 应用气象学报, 2013, 24(6): 641-655.
[61] 孙淑清,高守亭.现代天气学概论[M].气象出版社,2005.