Heterogeneity and Controlling Mechanisms of Soil Organic Carbon Stocks Across Erosion Zones in China Under Future Climate Scenarios

ABSTRACT

Soil organic carbon (SOC), as a key indicator of global carbon cycling and climate change response, plays a critical role in regional ecological security and agricultural sustainability. This study employs three machine learning models including the LightGBM, XGBoost, and random forest to simulate the spatial distribution of topsoil SOC stocks across different erosion types in China, and project their evolutionary trends under future climate scenarios. Model validation demonstrated high accuracy and strong consistency, while factor importance analysis revealed that mean annual temperature, mean annual precipitation, and elevation were the primary environmental drivers influencing the spatial variability of SOC stocks. Under both SSP245 and SSP585 scenarios, SOC stocks in China have a declining trend by 2050 and 2090. Areas affected by freeze–thaw erosion were the most sensitive to rising temperatures, whereas water erosion regions—due to their extensive spatial coverage—are projected to become the dominant zones of carbon loss. Spatially, carbon change hotspots were concentrated in the retreat zones of permafrost regions and mid‐ to low‐latitude water erosion areas, while changes in wind erosion zones remain relatively moderate but were significantly influenced by shifts in precipitation patterns. This study highlights the regional heterogeneity in SOC stocks responses to future climate change across China, underscoring the necessity of integrating natural processes and human activities specific to each erosion type when formulating national‐scale soil conservation and carbon management strategies. These findings provide a scientific foundation for promoting sustainable land management and enhancing ecosystem carbon sequestration capacity in China.