Abstract: Urban green spaces, a vital component of urban carbon sink system, play a crucial role in achieving the dual carbon goals through the coupling between their spatial patterns and carbon sequestration capacity. Taking central urban area of Chaoyang City as a case study, we investigated the intrinsic relationship between the spatial patterns of green spaces and their carbon sequestration efficacy, aiming to provide reference for urban green space system planning in arid and semi-arid regions. Sentinel-2A remote sensing imagery, NDVI data, and field surveys were used to identify the distributions of green space patches in the central urban area. Net primary production (NPP) of vegetation was estimated using the CASA model, and then converted into carbon stock to evaluate carbon sequestration capacity. Cluster analysis was used to examine the spatial variation in carbon sequestration capacity. Pearson correlation analysis and XGBoost-SHAP method were used to assess the importance of different indicators in terms of carbon sequestration capacity and underlying mechanisms. Finally, the mismatched areas were identified by coupling spatial patterns of carbon sequestration with landscape index, and then were classified into three zones: potential restoration area, transitional improvement area, and core carbon sequestration area. Such efforts would provide spatial decision support for green space system planning. Results showed that: (1) In 2023, the average annual vegetation NPP in Chaoyang City’s central urban area was 319.59 g C·m^-2·a^-1, with a total annual carbon sequestration of 1272.524 t. (2) At the landscape level, landscape fragmentation (DIVISION), maximum patch index (LPI), edge density (ED), and landscape shape index (LSI) significantly influenced carbon sequestration capacity. At the type level, area proportion (PLAND), aggregation index (AI), fragmentation (DIVISION), and cohesion index (COHENSION) were the most critical factors influencing carbon sequestration capacity. (3) Spatial coupling analysis identified nine corresponding spatial categories. Moderately impacted spaces and moderately carbon-sequestering spaces exhibited the highest number of grid cells (31.43%), while highly impacted spaces and low carbon-sequestering spaces had nearly zero grid cells. Based on spatial matching, potential restoration zones accounted for 40.7%, transitional improvement zones for 34.7%, and core carbon-sequestering zones for 25.6% of the total area. Based on these findings, targeted improvement recommendations are proposed for the three green space categories in Chaoyang City’s central urban area from both landscape and type dimensions. These recommendations aim to enhance the carbon sequestration capacity of Chaoyang’s green spaces, strengthen carbon sink functions in urban-rural planning, and advance the progress toward achieving carbon neutrality goals.

Key words: urban green space, carbon sequestration capacity, machine learning, spatial differentiation, Chaoyang City