Decoding prediction of PM2.5 against jointly street-tree canopy size and running vehicle density using big data in streetscapes

Abstract

Running vehicles are one of major sources of PM2.5 emission, and street-tree canopy can function as a green veil to retard diffusion. To decode joint contributions to PM2.5 of host city is a practical approach to predict accurate PM2.5 pollution across results given by different models. In this study, a total of 153 Chinese cities were randomly selected from China where ∼0.3 million of streetscapes were extracted from digital maps in 2023 for analyzing green view index (GVI) and running vehicle density (RVD). Roads were categorized into four classes as coefficients in host cities at varied levels of population urbanization. Air PM2.5 concentration occurred at higher levels (> 60 μg m−3) in northwestern cities and low in southwestern ones (∼10 μg m−3). GVI and RVD showed negative relationships with each other in all road classes in most cities except for five medium sized cities (<1000 thousand population). Multivariate regression models indicated that GVI showed a negative contribution to PM2.5 while RVD contributed positively. City-level PM2.5 was modeled against GVI and RVD using multivariate linear regression, which can be optimized using random forest algorithm (R2 = 0.3062 and 0.9231, accuracy = 71.02 % and 88.17 %, MSE = 90.0327 and 20.4885, MAE = 8.0788 and 3.7173, respectively). GVI was weighted with a higher feature importance than RVD for predicting PM2.5. It was predicted that cities in the centre and along the west edge of mainland China were agglomerated as hotspots with high PM2.5 contamination risks.