December 4, 2025Open Access

Intra-Urban CO2 Spatiotemporal Patterns and Driving Factors Using Multi-Source Data and AI Methods: A Case Study of Shanghai, China

LPLeyi PanSoutheast University QFQingyan Fu FYFan Yang

Key Points

CO2 concentrations show significant spatiotemporal patterns, being higher in central urban districts than suburbs.
Data from OCO-2 satellites and ground measurements led to results identifying airports and industrial emissions as major contributors to CO2 levels.
Machine learning and OLS regression methods demonstrated close ties between spatial CO2 variability and environmental factors.
Findings highlight the need for effective environmental planning to address urban CO2 emission challenges.

Abstract

Cities are major sources of anthropogenic carbon dioxide (CO2) emissions, making the study of intra-urban CO2 concentration patterns an emerging research priority. However, limited data availability and the complexity of urban environments have impeded detailed spatiotemporal analyses at the city scale. To address these challenges, an analysis supported by multi-source data and GeoAI methods is carried out to examine the spatial distribution, vertical variation, temporal dynamics, and driving factors of CO2 concentrations in urban areas. We combined OCO-2 satellite-derived XCO2 data (2014–2024) with ground-based measurements from the Shanghai Tower (August 2024 to March 2025), alongside meteorological and socioeconomic variables. The analysis employed spatial interpolation (inverse distance weighting), nonparametric testing (Mann–Whitney U test), time series decomposition, ordinary least squares (OLS) regression, and machine learning techniques including random forest and SHAP (SHapley Additive exPlanations) analysis. Results reveal that CO2 concentrations are significantly higher in central urban districts compared to suburban areas, with notable spatial heterogeneity. Elevated levels were detected near ports and ferry routes, with airports and industrial emissions identified as principal contributors. Vertically, CO2 concentrations decline with increasing altitude but exhibit a peak at mid-level heights. Temporally, a pronounced seasonal pattern was observed, characterized by higher concentrations in winter and lower levels in summer. Both OLS regression and machine learning models highlight proximity to emission sources, wind speed, and temperature as key determinants of spatial CO2 variability, with these factors collectively explaining 67% of the variance in OLS models. This study demonstrates how multi-source data and advanced methods can capture the spatial, vertical, and seasonal dynamics and driving factors of urban CO2 concentrations, offering insights for policy, planning, and mitigation.

KI fragen

Bookmark

View Full Paper