Local forest structure modifies realized climate vulnerability under macroclimatic forcing: Attribution from a national forest inventory

Forests are a major component of Nature-based Climate Solutions, but their mitigation value depends on how climate vulnerability is realized beyond macroclimatic exposure. Existing large-scale assessments remain largely exposure-based or correlative, making it difficult to distinguish direct climatic forcing from effects associated with local ecological conditions. Using two decades of China’s National Forest Inventory (NFI) comprising nearly one million tree records across 33 dominant tree species (groups), we developed an attribution framework integrating Structural Causal Models (SCM) and Double Machine Learning (DML) to estimate the conditional effect of individual climate drivers on annual DBH growth across observed spatial climatic gradients, after adjustment for measured local ecological factors. Identification of these conditional effects rests on the assumption of no residual unmeasured confounding after adjustment for measured local factors. The framework characterizes realized climate vulnerability through individual-tree diameter growth responses and quantifies individual-level heterogeneity in climate sensitivity. Species-level average responses masked substantial individual-level divergence: trees exposed to similar macroclimatic conditions often showed contrasting climate–growth responses depending on local topography and stand structure. Local ecological conditions modified climate–growth responses in non-linear and context-dependent ways, with topographic factors more influential under thermal stress and stand structural attributes more influential under hydrological stress. Density-associated empirical ranges were further identified within which growth-based climate vulnerability was relatively lower under observed stand structural conditions. These ranges are descriptive features of the fitted relationships within the observed climatic and structural envelope, and the nonparametric estimator does not support projection beyond this envelope. This study advances forest vulnerability assessment from exposure-based prediction toward conditional attribution of heterogeneous climate–growth responses, providing empirical evidence relevant to context-dependent forest adaptation under global environmental change.