Quantitative mechanisms and spatial configuration modes of health-promoting subtropical forests based on BVOC emissions and machine learning.

Forest therapy has gained widespread global recognition as a nature-based health intervention, with biogenic volatile organic compounds (BVOCs) proposed as one of the key components linking forest exposure to health outcomes. However, the quantitative mechanisms underlying how forest structure and environmental factors influence these effects remain poorly understood. Existing research lacks quantitative indicators and nonlinear threshold analyses, resulting in an absence of guidelines for health-promoting forest design. To address these gaps, this study developed an analytical framework integrating machine learning and SHAP (SHapley Additive exPlanations) interpretability analysis to identify and quantify the key drivers of the BVOC-related health-benefit potential in subtropical forests. Data on BVOC emissions, stand structural indices, and environmental variables were collected across five forest types. By calculating the concentrations of major BVOC components (terpenes and aromatic compounds), we developed a Health Benefit (HB) index as a proxy indicator of BVOC-related health benefit potential, which was subsequently utilized as the response variable for modeling. The dominant driving factors were identified through this framework, and a conditional association analysis based on causal forests was then used to test whether the structural drivers retained non-redundant signal after conditioning on the other predictors. Generalized additive models (GAMs) fitted to the SHAP response surfaces yielded model-derived thresholds, that is, the values at which a driver's fitted contribution changes sign, for humidity (68.25%, bootstrap 95% CI 66.77-68.71%), CO2 (396.22 ppm, CI 392.35-400.30 ppm; 427.61 ppm, CI 423.03-431.93 ppm), competition index (6.84, CI 6.52-7.20), mingling index (0.49, CI 0.47-0.51), and aggregation index (1.04, CI 1.03-1.05). Furthermore, four forest configuration modes (evergreen broad-leaved mixed, deciduous broad-leaved mixed, coniferous mixed, and coniferous-broad-leaved mixed) were constructed, providing threshold-informed conceptual templates for health-promoting forests planning. This framework provides an analytical pathway from identifying driving factors to developing structural configuration schemes, advancing the scientific foundation for subtropical forest wellness planning.