Spatiotemporal Heterogeneity of Surface Soil Organic Carbon in China: Novel Insights from Interpretable Machine Learning Coupled with Google Earth Engine

Soil organic carbon (SOC) is crucial in climate change mitigation, yet its spatiotemporal patterns and regional heterogeneity in China remain insufficiently resolved. By integrating field observations with multisource satellite data, we mapped surface SOC dynamics at 1 km for 1986–2023 using XGBoost (R2 = 0.71, RMSE = 1.73 kg C m–2, MAE = 1.21 kg C m–2). Model interpretability was advanced with Shapley additive explanation (SHAP) and piecewise structural equation modeling (piecewiseSEM) to identify key drivers and causal pathways. China’s SOC distribution exhibits pronounced heterogeneity and can be partitioned into four regions. Regions I–II together hold 56.90% of national SOC stock while covering 43.24% of the area. Nationally, SOC increased at 0.015 Pg C yr–1 during 1986–2023, but trends diverged regionally: Region I acted as a net carbon source, whereas Region II functioned predominantly as a sink. Soil properties govern the spatial pattern of SOC (strongest direct effects), while climate exerts the greatest overall influence via both direct and indirect pathways. Total nitrogen (TN), temperature (Temp), and soil organic matter (SOM) were the most influential drivers nationally. These findings underscore two policy priorities: (i) protect existing SOC-rich areas to achieve higher mitigation efficiency, and (ii) adopt region-differentiated conservation and management strategies to enhance land resilience and carbon sequestration under China’s dual-carbon goals.