Decoding spatiotemporal dynamic of atmospheric CO2 concentration under different urban development model

Current research lacks fine-scale assessments of urban XCO₂ dynamics and carbon source/sink identification from spatiotemporal perspectives. Thus, a 1 km-resolution XCO₂ dataset (XCO₂_Cat) for mainland China from 2000 to 2020 was reconstructed using CatBoost model and multi-source data, incorporating spatial and temporal factors. It enabled an in-depth analysis of XCO₂ and XCO₂_ano, offering insights into carbon source-sink dynamics. Furthermore, a carbon risk identification matrix based on XCO₂ and XCO₂_ano was proposed to identify urban carbon risks. Results revealed significant spatiotemporal variations in China, with higher XCO₂ in the east and lower in the west, increasing at a rate of 2–3 ppm/year. Heterogeneity in XCO₂ and XCO₂_ano was observed across urban models. Growing cities had the highest XCO₂ levels. Potentially shrinking, continuously shrinking and smart shrinking cities had lower XCO₂, but still exhibited relatively high values, suggesting that urban shrinkage does not necessarily lead to decarbonization. Growing cities showing the highest XCO₂_ano in most years. Continuously shrinking cities initially exhibited higher carbon source intensity. Substantial regional heterogeneities were found in XCO₂ and XCO₂_ano, emphasizing the need for region-specific carbon reduction strategies. Cities were categorized into four types based on carbon risk identification matrix: HL-type (stable carbon source and constrained carbon sink), LH-type (carbon source to carbon sink and vice versa), HH-type (strong disturbance), and LL-type (weaker carbon source). Key regions and urban development models for each type were identified, and targeted carbon management strategies were proposed. These findings offer critical guidance for implementing effective carbon reduction policies at finer scales.