Nonlinear Responses and Decoupling Between Soil Organic Carbon Fractions and Extracellular Enzyme Activity Across Salt-Affected Soils of the Qiangtang Plateau

Cold alpine salt-affected soils may retain carbon through mineral protection or suppressed microbial decomposition, but these processes remain difficult to distinguish. We measured soil physicochemical properties, four extracellular enzyme activities, soil organic carbon (SOC), and its dissolved (DOC), particulate (POC), and mineral-associated (MAOC) fractions across five soil types on the Qiangtang Plateau. Multivariate analyses and five XGBoost models interpreted using Shapley additive explanations (SHAP) characterized carbon-enzyme associations and identified leading predictors and nonlinear response transition points. Model performance varied among response variables (CV-R2 = 0.4094–0.8815). Soda (SD) soil had a distinct carbon-pool composition, and no significant carbon-enzyme correlations remained after Benjamini–Hochberg correction. pH was the leading predictor of SOC and its fractions, whereas total nitrogen and total phosphorus ranked highest for overall enzyme activity. The fitted SHAP contributions changed from negative to positive between pH 8.50 and 9.00 for the carbon variables and at a total nitrogen concentration of 0.71 g·kg−1 for enzyme activity. These model-dependent transition points indicate that carbon-pool restructuring and nutrient-related changes in enzyme activity occur over different environmental ranges. SD soil contained the highest carbon concentrations but the smallest MAOC proportion and the largest POC and DOC proportions, indicating relative enrichment of labile carbon fractions. By jointly analyzing carbon fractions and enzyme activity within an interpretable XGBoost-SHAP framework, this study helps distinguish high carbon stocks from stable sequestration and shows why total SOC alone may overestimate long-term carbon stability in alpine salt-affected soils.