Predicting ecological regulators of thymol and carvacrol biosynthesis in Oliveria decumbens Vent. using a hybrid ensemble machine learning model (RF+SVR-RBF) in arid regions of Iran

Dryland ecosystems, encompassing arid to semi-arid regions, impose strong climatic and edaphic constraints that profoundly shape plant functional traits and secondary metabolism. Understanding how environmental factors regulate phytochemical biosynthesis is essential for biodiversity conservation and sustainable resource management under increasing aridity. Oliveria decumbens Vent., an endemic medicinal species of the drylands of Fars Province, Iran, provides an excellent model for exploring the ecological determinants of metabolite variability in water-limited habitats. We integrated ecological predictors with machine learning to model the spatial variation of thymol and carvacrol concentrations across 59 georeferenced populations of O. decumbens . Three predictive models—Random Forest (RF), Support Vector Regression (SVR) with a Radial Basis Function (RBF) kernel (SVR-RBF), and a hybrid ensemble (RF+SVR-RBF)—were developed and evaluated. Model performance was quantified using root mean squared error (RMSE), mean absolute error (MAE), coefficient of determination ( R 2 ), and the concordance correlation coefficient (CCC). Generalized Linear Model (GLM) was applied to identify key environmental variables regulating metabolite biosynthesis. The hybrid ensemble consistently outperformed individual model, achieving the highest predictive accuracy ( R 2 =0.82 for thymol and R 2 =0.80 for carvacrol). Spatial mapping revealed pronounced heterogeneity in metabolite distribution, with distinct functional hotspots in the northern and western semi-arid regions of Fars Province. GLM analysis indicated that mean annual temperature, slope aspect, slope degree, and sand content were strong positive predictors of thymol and carvacrol accumulation, whereas high soil potassium, clay percentage, and alkaline pH constrained metabolite production. This study shows that hybrid ensemble modeling effectively captures how environmental gradients regulate secondary metabolism in dryland plants. The proposed framework, combining RF with SVR-RBF, is transferable across arid environments. These findings support trait-based ecological predictions and offer practical insights for conservation and sustainable cultivation of high-value medicinal plants in water-limited ecosystems.