Spatial distribution and primary controlling factors of soil organic carbon under agricultural land based on CatBoost model

Soil organic carbon (SOC) is a fundamental component of the soil carbon pool, and its carbon stock is about 2~3 times that of vegetation carbon stock, which plays an important carbon buffering role in the context of global climate change. In agricultural production, SOC is central to soil fertility, directly impacting soil health and food security. The spatial distribution of SOC is influenced by various geographical environmental variables, including climate, topography, soil-forming parent material, and vegetation, exhibiting a covariant relationship with these factors. Understanding the spatial distribution of SOC in agricultural land and identifying the primary controlling factors is essential for maintaining soil fertility and productivity. The nonlinear relationship between SOC and environmental covariates has been widely demonstrated, but the primary controlling effects of environmental covariates on SOC content and their contribution effects are often neglected. In this study, the primary controlling factors of SOC and their effects were explored by a CatBoost model, a decision tree-based gradient boosting algorithm, to reveal the mechanism of spatial differentiation of SOC at the regional scale, utilizing multi-source data such as measured data and remote sensing. The results showed that the CatBoost model outperforms univariate linear regression models across all independent variables, achieving an overall R² of 0.828, indicating that the model could explain the variations of the target variables well. Total nitrogen (TN), available phosphorus (AP), annual lowest temperature (T), cation exchange capacity (CEC), and available potassium (AK) were, in order, the most significant factors affecting the organic carbon content, with TN ranking the highest with an influence weight of 39.10%. In addition, this study found that threshold effects on SOC were observed for the environmental covariates, and all had two thresholds. Furthermore, no two variables were independent and all had interactive negative effects. It can be concluded that the effect of environmental variables on SOC content was a complex interaction rather than a simple superposition. This indirectly proves that over-fertilization will not achieve the effect of increasing soil fertility, but will result in resource wastage and farmland ecological pollution problems. These findings underscore the importance of considering the interaction effects of environmental covariates to understand the potential processes of SOC accumulation, which are vital for sustainable agricultural development.

Keywords: Soil organic carbon (SOC), cultivated land, CatBoost, primary controlling factors, nonlinear relationships