Multi-model integration for karst carbon storage dynamics in Guizhou, China: A machine learning-driven approach with policy implications under carbon neutrality goals

Karst regions are among the most complex and fragile ecosystems globally, playing a critical role in the carbon cycle. However, systematic assessments of carbon storage (CS) dynamics and their driving mechanisms remain limited. This study focuses on Guizhou Province, a global karst hotspot, using it as a case study. By integrating multi-source remote sensing data with the PLUS and InVEST models, and incorporating the ecological protection red line into future predictions, we systematically analyze the spatiotemporal dynamics of CS from 1990 to 2060. A CatBoost-SHAP-PDP machine learning framework is applied to effectively reveal the nonlinear relationships and interactive driving mechanisms influencing CS changes in these ecosystems. The results show that between 1990 and 2020, the overall CS in Guizhou exhibited a fluctuating downward trend, decreasing by 1.4 %, with the Karst region showing a more significant decline (2.08 %), far exceeding that of non-Karst regions (0.35 %). Future scenario predictions suggest that under natural development and cropland protection scenarios, CS would further decrease by 8.25 % and 6.85 % respectively by 2060, while ecological protection scenarios could restore CS to near-1990 levels. The driving mechanism analysis identified NDVI, evaporation, and SPEI as the primary driving factors, explaining 68.2 % of the variation in CS. These factors demonstrate complex nonlinear relationships and interactive effects. CS increases under favorable ecological conditions, including healthy vegetation cover (NDVI > 0.5), moderate evaporation, adequate moisture, and warm temperatures. Peak CS occurs when dense vegetation is coupled with optimal climate and water availability. This research provides a methodological framework for long-term CS assessment and mechanistic analysis in karst regions, while offering scientific support for land-use optimization and ecological protection policy formulation under regional “dual carbon” goals.