Revealing nonlinear and interaction mechanisms of resource and environmental carrying capacity through explainable machine learning

Existing studies on resource and environmental carrying capacity (RECC) have primarily focused on comprehensive evaluation and linear identification. But the quantification of driving effect on RECC have not been fully explored, especially the interaction and nonlinear effects. Therefore, this paper constructs an explainable machine learning framework, to comprehensively assess the RECC index of 31 provincial regions in China and to reveal nonlinear and interaction mechanisms of influence factors. Firstly, the XGBoost model is utilized to surrogate the evaluation results, which demonstrates outstanding predictive performance (R2 = 0.95, RMSE = 0.08) and stability. The results shows that the average RECC index in China dropping from 1.18 to 0.39, marking a transition from a high-risk state to a safe and sustainable state. Then, the SHAP model quantifies the nonlinear and threshold effects of key influence factors. The results indicate that the total emissions of pollutants in wastewater (X2) and industrial sulfur dioxide emissions (X3) are the two most critical factors, with a cumulative contribution rate of 49%. 50000 tons is a threshold for X2, beyond which the RECC index significantly increases with the increase of X2. 10 billion cubic meters is the marginal contribution of X3, with the greatest impact on RECC index. The interaction effect indicates a significant positive synergistic effect between X2 and X3, with the maximum increase in SHAP value reaching 0.1. There is a significant antagonistic effect between resource availability and environmental pressure. This study offering new directions for understanding the underlying mechanisms of complex environmental systems.