Soil organic carbon (SOC) is a key component of ecosystem functioning and climate mitigation, yet its spatial variability in karst landscapes remains insufficiently understood. In this study, we analyzed the spatial pattern of SOC in Guizhou Province using multi-source geospatial datasets, machine-learning models, SHAP-based interpretation, and scenario simulations. A total of 3,000 grid-based observations were extracted from public soil and environmental datasets for the 0–30 cm layer. Among the five tested models, Random Forest (RF) achieved the best predictive performance, with an R 2 of 0.92, an MAE of 0.71 g/kg, and an RMSE of 0.91 g/kg. SHAP analysis identified bulk density (BD), total nitrogen (TN), and mean annual precipitation (MAP) as the most relevant predictors of SOC. Pairwise SHAP interaction analysis indicated that the interaction hierarchy was stable across repeated 5-fold resampling, with BD-TN showing the strongest interaction, followed by BD-MAP and TN-MAP. Scenario simulations further showed that reducing BD by 5 % increased mean SOC by 16.57 %, whereas increasing TN by 10 % increased SOC by 4.02 %. The combined improvement scenario (BD − 5 %, TN + 10 %, MAP + 5 %) produced the largest positive response, with SOC increasing by 16.80 %, while the combined degradation scenario (BD + 5 %, TN − 10 %, MAP − 5 %) reduced SOC by 12.77 %. These results suggest that SOC spatial variability in Guizhou’s karst region is associated with the joint influence of soil physical condition, nutrient status, and hydrothermal background, and that integrated improvement of these conditions may generate stronger positive model responses than single-factor changes alone. Overall, this study provides a scale-explicit and interpretable framework for understanding SOC distribution and scenario-based SOC responses in subtropical karst regions.
Xingfu et al. (Wed,) studied this question.