Exploring the main driving factors of gross primary production in different climate zones of China using the XGBoost–SHAP model 后印本

摘要: Understanding the drivers of gross primary production (GPP) is essential for assessing vegetation productivity dynamics under climate change, particularly across regions with strong climatic heterogeneity. China spans diverse climate zones and ecosystems, yet the relative importance of climatic, environmental, and anthropogenic factors regulating GPP has remained poorly resolved. In this study, we investigated the spatiotemporal patterns of GPP across China from 2001 to 2020 and quantified the contributions of multiple driving factors across different climate zones. We combined ridge regression with an interpretable machine learning framework based on Extreme Gradient Boosting (XGBoost) and SHapley Additive exPlanations (SHAP) to disentangle the long-term linear controls on and short-term nonlinear responses driving GPP. Ridge regression was employed to address multicollinearity among predictors and to quantify their interannual contributions, while SHAP analysis was used to quantify feature contributions in nonlinear model predictions. Our results indicated that leaf area index (LAI) and human footprint dominated the long-term variability of GPP in most climate zones, whereas temperature and solar radiation exerted stronger influences on instantaneous GPP responses. The relative importance of drivers varied markedly among climate zones, reflecting region-specific climatic constraints and vegetation physiological characteristics. In addition, the contribution of atmospheric CO2 to GPP variability was notably limited in the alpine climate zone and showed a declining fertilization effect nationally, suggesting increasing constraints imposed by water availability and nutrient limitations. By integrating linear attribution and nonlinear interpretability, this study provides a comprehensive assessment of the controls on GPP dynamics across China and highlights the importance of accounting for climatic heterogeneity and temporal scales when evaluating vegetation productivity responses to environmental change.