Effects of Urban Blue–Green Landscape Patterns on Carbon Sequestration in Chongqing’s Central Urban Area: An XGBoost–SHAP Analysis

Abstract

Against the backdrop of the "Dual Carbon" goals, this study focused on the main urban area in Chongqing. With remote sensing, land use, meteorological data from 2000, 2010, and 2020 -- we used the CASA model to estimate vegetation net primary productivity, or NPP, which we compared to MODIS NPP products. By combining optimal scale analysis, landscape pattern indices, the XGBoost model and SHAP (the Shapley Additive exPlanations) interpretation methods, our study showed that the space relationship between the blue and green landscapes and how they help to reduce the amount of vegetation carbon sinks are associated with one another, not just by the space that the areas are in, but by these other ways. The results indicate that from 2000 to 2020, NPP in the study area showed an overall upward trend. High-value areas were primarily distributed in the peripheral mountains and continuous ecological spaces, while blue-green spaces in the central urban area and densely built river valleys exhibited greater fragmentation. Compared to the baseline model (M0) containing only elevation and slope, the extended model (M1) incorporating landscape pattern indices significantly improved cross-validation R² values: from 0.5954, 0.6248, and 0.5735 to 0.7996, 0.8031, and 0.7311 for the years 2000, 2010, and 2020, respectively.SHAP analysis revealed that among the selected predictors, elevation contributed the most. PLAND (Percentage of Landscape) demonstrated a significant predictive contribution across all three years, whereas the roles of ED (Edge Density), SPLIT (Splitting Index), and AI (Aggregation Index) varied interannually. Nonlinear responses and reference thresholds were identified for elevation, slope, PLAND, LPI (Largest Patch Index), and their contributions to NPP prediction; however, SHAP contributions do not imply direct causal effects. The study concludes that topographic backgrounds, combined with the area, fragmentation, and connectivity of blue-green spaces, collectively correlate with spatial variations in NPP within mountainous cities. These findings provide a reference for optimizing ecological spaces and enhancing vegetation carbon sink potential in the main urban area of Chongqing.