Spatial-Temporal Characteristics And Prediction Of Traffic-Related PM And CO₂ At Urban Neighborhood Scale

With the continuous development of cities and people’s reliance on transportation,traffic pollution has become an important source of urban air pollution.A large part of the public’s daily life is exposed to outdoor commuting,and the exposure to traffic pollutants during commuting will have a negative impact on human health.The prediction of the spatial distribution of traffic pollutants at the urban micro-scale can help refine the urban air environment and improve the public’s understanding of the environmental quality conditions in which they live.In this study,data on traffic-related pollutant and CO₂ concentrations,traffic flow,meteorological conditions,and land use types were obtained by mobile monitoring in the study area of about 4 km² in Xi’an,China,to investigate the relationship between each influencing factor and pollutant concentrations.Firstly,the spatial and temporal distribution characteristics of pollutants and CO₂ are analyzed using the measured data to visualize the measured concentration data.Secondly,the correlation between pollutants and influencing factors is analyzed by correlation analysis,covariance analysis and stepwise regression method for land use regression（LUR）modeling.Finally,in order to obtain simulation results with higher accuracy,more reliability and more reasonable interpretation,a random forest land use model（RF-LUR）was established using the influence factors selected by the land use modeling process to evaluate and predict the spatial distribution levels of each pollutant concentration,and the simulation results of the RF-LUR model were compared and analyzed with those of the common LUR model.High reliability data were obtained from the experimental monitoring,and traffic-related pollutants and CO₂ concentrations showed significant spatial and temporal heterogeneity.In the correlation analysis as well as land use regression modeling,it was found that PM₁₀ and PM_2.5concentrations were significantly correlated with background concentration,wind speed,and relative humidity,and BC and CO₂ concentrations were significantly correlated with relative humidity as well as wind speed.The percentage of buildings and roads in the model is positively correlated with pollutants and CO₂ concentration,while the percentage of greenery and open space is negatively correlated with pollutants and CO₂ concentration.The R² for PM₁₀,PM_2.5,BC and CO₂ in the land use regression modeling were 0.609,0.514,0.106 and 0.255,respectively.The spatial simulation results of the RF-LUR model for pollutant and CO₂concentrations were found to be better in the random forest land use modeling,with R² of 0.826,0.864,0.464 and 0.612 for PM₁₀,PM_2.5,BC and CO₂ models,respectively.The RF-LUR model also has significant improvements in other evaluation indicators,with mean absolute errors reduced by 5.966,4.258,0.362,and 10.005,and root mean square errors reduced by 8.689,4.24,0.374,and 7.157,respectively.The visualization results of the spatial distribution of pollutant concentrations achieved by kriging interpolation of the predicted data derived from the RF-LUR model also better fit the actual monitoring situation.The RF-LUR model combined with mobile monitoring is shown to be more effective in predicting traffic particulate matter and CO₂ concentration distributions at the urban neighborhood scale,which can better visualize the spatial distribution of particulate matter and CO₂ and provide a detailed practical basis for policy formulation and regulatory improvement.