Refined Source Apportionment Of PM_2.5 Based On Combined Multiple Models

Atmospheric fine particulate matter(PM_2.5)is emitted from complex sources and has serious adverse effects on human health,ecosystems,and climate.PM_2.5 is a prominent regional issue that is affected by local emissions and regional transport.Therefore,identifying PM_2.5 sources and quantifying the regional transport contribution is essential for the development of joint prevention and control measures.The receptor models and numerical models are widely used in PM_2.5source apportionment,but different source apportionment methods have unique advantages and limitations.With the increasing requirements of precision science pollution control,the current PM_2.5 source apportionment methods are unable to meet the higher requirements of precision and timeliness.This study aims to develop a novel PM_2.5source apportionment method system based on a combination of receptor models,numerical models,and machine learning models.It can realize refined source apportionment of PM_2.5,mainly in the aspects of refined source types,potential source regions,and high spatiotemporal resolution,which will provide scientific basis and technical support for air pollution prevention and refined emergency control.In this study,a refined PM_2.5 source method based on the receptor model with extending organic tracers was established.It can distinguish the precise PM_2.5 sources that are usually difficult to identify by incorporating non-polar and polar organic species.Then,a novel PM_2.5 source apportionment method combined with multiple models,including receptor models,Lagrangian particle model,and chemical transport model,was performed to explore the source-receptor relationship and distinguish the contributions of emission sources and source regions.Machine learning algortithm was developed to optimize the numerical simulation prediction,and the pollution concentation prediction in the Beijing-Tianjin-Hebei（BTH）region was corrected from the site to regional scale.On this basis,high spatiotemporal resolution PM_2.5source apportionment and prediction was developed using the Integrated Mobile Source Indicator（IMSI）method,which can realize the rapid and accurate estimation of daily PM_2.5 emission source impacts.The main findings are as follows:（1）Nine types of emission sources based on PMF-MM model were identified,including secondary source,gasoline exhaust,diesel exhaust,coal combustion,industrial source,biomass burning,dust,cooking,and coking.Gasoline and diesel exhausts were distinguished by adding polycyclic aromatic hydrocarbons（PAHs）,C19-C24 n-alkanes as key organic tracers.In addition,levoglucosan and hexadecanoic acid are important additions for identifying biomass burning and cooking,respectively.Secondary sources contributed the most to PM_2.5,accounting for 39.3%.Gasoline and diesel exhausts contributed 16.1%and 7.1%respectively.The contributions of coal combustion（10.2%）,industrial source（12.2%）,and biomass burning（6.7%）were also relatively high.And the contribution of cooking（3.3%）was not negligible.（2）An improved PM_2.5 source apportionment method combined with multiple models was developed to quantify the contribution of various emission sources in different transport pathways including local,south,east,and north,and distinguish the potential source regions of primary or secondary sources.Secondary sources contributed the most to PM_2.5,which was affected by the regional transport of Baoding,Langfang,and Tangshan.Among major primary sources,vehicle source was concentrated locally.Coal combustion contributed a higher proportion locally（15.8%）and in the south（15.0%）,and Baoding,Langfang,and Xingtai along the Taihang Mountains were predominant potential source regions.Biomass burning contributed a higher proportion in the south（19.0%）and northeast（22.6%）,and adjacent southern Hebei and Northeastern areas were the predominant potential source regions.The contribution of industrial processes was relatively low,and Tangshan,Baoding,Xingtai,and Anyang were the predominant potential source regions.（3）Machine learning algorithms showed excellent and stable performance in terms of error correction of pollutants concentration simulated by numerical models,which can better identify the variability in pollution concentrations and their complex correlation with meteorological features.Compared with the WRF-CAMx model,the prediction performance of random forest and support vector regression was excellent in the site optimization.The R values of PM_2.5 and its components were increased to0.71–0.82,and the RMSE and MAE values were decreased by 21%–84%and 4%–82%,respectively.In the regional optimization,the R values of NO₂,CO,and EC predicted by XGBoost model increased to 0.87–0.92 on the ten-fold cross-validation set,and the error indexes decreased by 50%–67%.The predicted spatial distribution of NO₂,CO,and EC concentration fields after XGBoost optimization was basically consistent with that of the observation.The spatial correlation of R values was above0.8,and the error indexes were significantly decreased.（4）Tacking traffic source as an example,a high spatiotemporal resolution PM_2.5traffic source impacts using a comprehensive indicator method was established by combing the optimized high-precision pollutant concentration fields and emission weight factors.The results showed that during the heavy pollution process in January2019,the impacts of traffic source covered the entire BTH region,especially the IMSI index of southern cities along the Taihang Mountains was as high as 6.0.The spatial distribution of the higher contribution of traffic source to PM_2.5 was consistent with the dense road network areas.IMSI traffic source impacts showed a high correlation with the traffic source contribution resolved by the receptor model（R=0.53–0.79）.The spatiotemporal distributions of NO₂,CO,and EC pollutant concentration fields in the next 14 days were close to the observation,and the spatiotemporal variation trend of the predicted IMSI index can better reflect the impacts of daily traffic sources on PM_2.5.This method provides a new ideal for estimating the contributions of emission sources with high spatiotemporal resolution.It is also suitable for the prediction evaluation of other emission source impacts with a small amount of calculation and a short time.