Revealing spatial patterns and depth-dependent controls of soil organic carbon density across China using interpretable machine learning

Abstract

Soil organic carbon density (SOCD) is a key indicator of soil health and a critical component of terrestrial carbon storage, yet its large-scale spatial organization and environmental controls remain insufficiently understood. Here, we assessed national-scale patterns and depth-dependent environmental relationships of SOCD across China from 2000 to 2020. Multi-source soil observations were integrated with climatic, vegetation, topographic, and soil property data to generate annual SOCD estimates for surface (0–20 cm) and profile (0–100 cm) layers using machine-learning models. Predicted SOCD exhibited pronounced spatial heterogeneity, with consistently higher values in the Northeast Plain and the Qinghai–Tibet Plateau and lower values in arid northwestern regions and intensively cultivated areas of the North China Plain. National mean SOCD showed only minor, statistically insignificant changes over the study period, indicating relative temporal stability at the national scale. Model interpretation and spatial heterogeneity analyses indicated that temperature and the Normalized Difference Vegetation Index were the dominant factors associated with surface-layer SOCD, whereas elevation and soil type gained importance with increasing depth. Interactions among climatic, vegetative, and pedogenic variables further structured regional SOCD patterns. These findings indicate that spatial variability in SOCD across China reflects a depth-dependent interplay between vegetation productivity, hydrothermal conditions, and long-term soil formation processes. By providing an interpretable, national-scale assessment of SOCD patterns and their environmental context, this study supports regionally differentiated soil carbon management and monitoring under ongoing climatic change.