Nano-chitosan-mediated salinity tolerance in Calendula officinalis: Nonlinear physiological thresholds and antioxidant responses identified via interpretable hybrid modeling

Soil salinity severely constrains growth and phytochemical quality in medicinal plants. Chitosan-based biostimulants offer a promising strategy for stress mitigation, yet comparative assessments of conventional versus nano-chitosan under salinity, particularly when coupled with interpretable modeling, remain scarce. This study investigated the effects of four salinity levels (0, 50, 100, and 150 mM NaCl) and foliar applications of chitosan and nano-chitosan (0, 0.25, and 0.5 g L⁻¹) on morphological, physiological, and biochemical traits of Calendula officinalis under controlled greenhouse conditions. A hybrid interpretable modeling framework (IGANN–GPR) was developed that integrates generalized additive neural networks with Gaussian process regression, enabling both high-accuracy prediction and principled uncertainty quantification across fifteen plant response variables. The model achieved excellent performance (R = 0.84–0.99), with exceptional accuracy for photosynthetic pigments (R ≥ 0.98) and robust estimation of oxidative stress markers and secondary metabolites. Residual analysis confirmed unbiased predictions and stable generalization. Feature importance analysis revealed that nano-chitosan (35.9%) slightly outperformed salinity (34.0%) and conventional chitosan (30.1%) in influencing plant responses. Interpretable dose–response functions identified a distinct salinity threshold (~75 mM NaCl), beyond which growth and chlorophyll content declined sharply, while proline and phenolic compounds accumulated progressively. Nano-chitosan exhibited rapid sigmoidal activation at lower concentrations, suggesting enhanced bioavailability and more effective modulation of antioxidant enzymes and membrane integrity compared with conventional chitosan. These findings demonstrate that nano-chitosan alleviates salinity-induced oxidative damage by strengthening redox homeostasis and osmotic adjustment. By integrating predictive accuracy with mechanistic interpretability and uncertainty quantification, the proposed hybrid modeling framework offers a robust, transparent, and systems-level tool for understanding stress physiology and supporting precision management of medicinal crops under saline conditions.