CO2 prediction results

Enhanced urban CO2 monitoring and understanding the spatiotemporal patterns and driving factors of urban CO2 concentrations contribute to the effective management of urban CO2 emissions and the development of strategies to mitigate climate change. This study focuses on accurately predicting and mapping CO2 concentrations in Shenzhen’s road network by integrating vehicle-cruising CO2 observations, street view panoramas, and multisource remote sensing data. By utilizing street view panoramas to capture the surrounding street configuration features and multisource remote sensing data to map nearby urban landscape features, we developed the CO2 prediction model with R² of 0.92 and MAE of 3.297 ppm. Furthermore, we identified eight high-CO2 concentration areas and examined impacts of urban function, urban development, traffic condition, environment condition on CO2 concentrations by explainable machine learning techniques. In urban centers, human activities have a substantial impact on CO2 levels, noticeably increasing during peak commuting times. Effective non-motorway planning, convenient public transport, and diverse urban functions can help reduce CO2 concentrations. Areas with high vegetation cover also show high CO2 concentrations, and the impact of greenery on Shenzhen’s CO2 concentrations is positive in November. This paper offers a new perspective on refined CO2 emission observation and the analysis of complex driving factors, providing novel technical approaches that enhance the precision of CO2 concentrations prediction, and offer interpretable methods for urban CO2 monitoring and management.