This study focuses on the Kunming municipal area, located in the low-latitude plateau of southwestern China, utilizing remote sensing image data from four distinct periods between 2005 and 2020 to analyze changes in its green landscape patterns. Net primary productivity (NPP) was employed as a metric for carbon sequestration analysis to assess variations in NPP within the Kunming municipal area. Based on Pearson correlation analysis and the XGBoost-SHAP model, the correlations, important indicators, and responses of changes in the green space patterns of the Kunming municipal area to changes in carbon sequestration were analyzed and combined with policy and human factors. The findings indicate the following: (1) From 2005 to 2020, the area proportions of various green space types within the Kunming municipal area were ranked as follows: forest land > grassland > cultivated land > water bodies. (2) Between 2005 and 2015, the patch shapes of green spaces became increasingly complex, with heightened fragmentation among patches. After 2015, this complexity was reduced while connectivity continued to decline alongside an increase in the landscape heterogeneity and richness. (3) Over the period from 2005 to 2020, NPP values for cultivated land, forest land, and grassland exhibited a trend of decreasing and then increasing, reaching their lowest point in 2010. High NPP areas were predominantly found in regions characterized by a hilly topography, elevated altitudes, and substantial natural vegetation cover. (4) There was a significant correlation between green space pattern indices and NPP (p < 0.01), with SHDI, CONTAG, and DIVISION identified as three critical indices influencing NPP. The relationship between landscape patterns and carbon sequestration was most pronounced during the period from 2015 to 2020, followed by that from 2005 to 2010.