Assessing the Climate Sensitivity of Soil Organic Carbon in China Based on Machine Learning and a Bottom-Up Framework

Soil organic carbon (SOC) plays a crucial role in the terrestrial carbon cycle and climate regulation. Quantifying the sensitivity of SOC to climate change is essential for developing effective strategies to address climate change and optimizing agricultural production. This study compares the performance of four machine learning models in assessing SOC, ultimately selecting the optimal Extreme Gradient Boosting model for spatial predictions of surface SOC (0–30 cm) across the country. The results indicate that areas with higher organic carbon density are primarily concentrated in the Tibetan Plateau and northeastern regions. Notably, regions with high uncertainty in predictions correspond to areas of elevated organic carbon density. Average temperature, average precipitation, and the Normalized Difference Vegetation Index were identified as the most influential factors across all models. Based on the predictions from the optimal model and a bottom-up framework, various potential climate change scenarios were considered, allowing for the quantification of SOC sensitivity to climate change. Under scenarios of increased temperatures and decreased precipitation, SOC loss intensified, hindering SOC accumulation. When the average temperature rose by 1.45 °C and precipitation decreased by 14.67%, a loss of 10% in SOC was projected for most regions of China. These findings provide critical insights for the proactive formulation of climate adaptation strategies, soil health preservation, and the maintenance of ecosystem stability.