Spatiotemporal variations of carbon dioxide (CO2) at Urban neighborhood scale: Characterization of distribution patterns and contributions of emission sources

•Spatiotemporal distributions were analyzed for urban CO2 concentration via field study.•CO2 in urban neighborhoods exhibited notable daily and seasonal variations.•The CO2 showed a gradual decrease in roadside, residential, and green space areas.•Random Forest and eXtreme Gradient Boosting were used for source contribution analysis.•Traffic source had the largest contribution to the CO2 followed by green space. The gradual increase in atmospheric carbon dioxide (CO2) concentrations has attracted worldwide attention for its strong relationships with global climate change. Considerable efforts are being undertaken to characterize spatiotemporal variations of CO2 at a city, regional and national level, aiming at providing pipelines for carbon emission reduction. However, there is scarce knowledge of how CO2 at the urban neighborhood scale is produced and distributed in the context of time and space, which is useful to accurately target source contributions from the ground up and reduce carbon emissions at a fine-grained scale. In this study, mobile measurements of CO2 concentrations were made in a 2 km × 2 km urban area covering different land use types to separately characterize the spatiotemporal distribution patterns of CO2 in roadside, residential and green space areas. The results show that CO2 concentrations in the late afternoon (Local time, UTC+8, LT 17–18) were higher than those at noon (LT 11–12), and that CO2 concentrations in winter were higher than those in summer. The roadside areas exhibited the highest CO2 concentration level of 452.66 ± 20.59 ppm, followed by residential areas (436.34 ± 27.02 ppm) and green space areas (428.98 ± 20.49 ppm). The result indicates that traffic sources brought more carbon emissions and contributed to a significant increase in CO2 concentrations, while urban greenery caused more carbon absorptions and reduced CO2 concentrations. This can be further confirmed by the observations that CO2 concentrations in the roadside neighborhood showed a strong positive correlation (R2 = 0.86) with ambient traffic flow. Then two machine learning models, i.e., Random Forest and eXtreme Gradient Boost, were developed to quantify the individual contribution from different carbon emission sources to the CO2 distributions, including traffic flow, greening rate, and domestic energy consumption. The results show that traffic-related carbon emissions were the most important influencing factor and accounted for approximately 60% of ambie