Mapping forest carbon density and net primary productivity using a stacking-SHAP model toward management zoning

Forest carbon density and net primary productivity (NPP) are two complementary indicators of forest carbon-sink capacity, reflecting long-term carbon storage and annual carbon fixation, respectively. Forests with high NPP do not necessarily exhibit high carbon density, so mapping both indicators is important for carbon-oriented forest management. Stand structure is a physically meaningful intermediate layer linking environmental factors to forest carbon storage. A two-stage interpretable framework is proposed: first, sample plot data from the 9th National Continuous Forest Inventory (NCFI) in Guangdong Province, together with 235 candidate predictors from multi-source datasets, were used to develop an interpretable random forest-SHapley Additive exPlanations (RF-SHAP) model to estimate 10 stand structure factors. Second, this study applied an interpretable Stacking ensemble model (k-nearest neighbors (KNN), RF, eXtreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and multi-layer perceptron (MLP) as base learners with XGBoost as meta-learner) to estimate carbon density and NPP using the 10 stand structure factors. The results showed: (1) The nine continuous structure factors achieved moderate cross-validated performance (R2 = 0.46 - 0.55). The categorical forest species (FS) achieved an overall accuracy of 0.89 (Kappa = 0.88). (2) The Stacking model outperformed individual learners, with a 10-fold cross-validated test R2 of 0.85 for carbon density and 0.77 for NPP. (3) SHAP analysis indicates that stand density (SD) and forest canopy density (FD) contribute most to carbon density estimation, while a FS, mean forest height (FH), and FD show the largest contributions to NPP estimation. (4) Spatial mapping further reveals extensive areas with high NPP but low carbon density. Based on carbon density and NPP patterns, the study area was categorized into four management zones. The “High NPP + Low CD (carbon density)” zone accounts for 31.03% and is prioritized for accumulating carbon while maintaining high productivity. This work provides an interpretable, stand structure-mediated framework for jointly mapping forest productivity and carbon storage to support management zoning.