Integrated evaluation of habitat suitability and spatial prediction of medicinal quality for Bletilla striata using species distribution models and machine learning approaches

Bletilla striata (Thunb.) Rchb.f. holds significant value in traditional Chinese medicine and is widely applied in pharmaceuticals, healthcare, and cosmetics. Its unique medicinal properties and increasing market demand have led to extensive artificial cultivation; however, this has also resulted in pronounced spatial variations in medicinal quality, coupled with the decline of wild resources. Addressing this challenge necessitates a systematic assessment framework. This study integrates species distribution models (Maxent and biomod2), active constituent quantification, principal component analysis (PCA), partial least squares regression (PLSR), and over ten machine learning algorithms including XGBoost to evaluate B.striata from three dimensions: ecological suitability, phytochemical variability, and environmental driving mechanisms. Results indicate that the biomod2 model outperforms Maxent in predicting the current and future potential habitat suitability, identifying southwestern regions such as Guizhou and Sichuan as core high-suitability zones. Quantitative analysis of cultivated samples from different regions revealed substantial variation in phenolic glycosides and polysaccharide content, with coefficients of variation reaching up to 48.31 %. Pearson correlation and PLSR analyses revealed that soil factors including selenium content, pH, exchangeable sodium, soil bulk density, gravel proportion, and silt content play dominant roles in influencing the accumulation of active compounds. An XGBoost-based quality classification model was constructed and overlaid with suitability maps to delineate five distinct zones: core, enhancement, consolidation, buffering, and general zones. The integrated framework provides scientific guidance for the conservation, high-quality cultivation, and sustainable utilization of B.striata . More broadly, this innovative framework, combining ecological modeling with machine learning for quality assessment, offers a valuable methodological reference for the sustainable development and quality management of other medicinal plants. • A framework integrated SDMs and machine learning to map B.striata quality zones. • Biomod2 outperformed MaxEnt in suitability prediction with AUC of 0.981. • Five compounds showed 20–48 % spatial variation across 33 sample sites. • Soil Se, pH, Na + , and Mn were key drivers of quality heterogeneity. • Overlay zoning defines core and marginal areas for high-quality cultivation.