Species distribution models and machine learning algorithms for medicinal and industrial plants conservation: Bridging habitat suitability and phytochemical quality mapping

This systematic review synthesizes 90 peer-reviewed studies (2015–2025) that used species distribution models (SDMs) and machine learning (ML) for conserving medicinal and industrial plants, guiding their cultivation, and mapping phytochemical quality. We systematically searched Scopus, Web of Science, ScienceDirect, and Google Scholar using targeted terms for SDMs, precision agriculture, phytochemical mapping, and machine learning. A meta-analysis was not performed due to substantial heterogeneity in study designs, algorithms, and evaluation metrics across the included studies. Studies were included if they applied spatially explicit modeling to medicinal and industrial species and reported model evaluation metrics or quality-related outcomes. We summarized methodological trends, predictor selection, validation strategies, and application pathways across two complementary paradigms: occurrence-based distribution modeling and quality-aware (phytochemical-integrated) modeling. MaxEnt remains the dominant choice for occurrence-only data, whereas Random Forest, Support Vector Machines, and ensemble/hybrid frameworks are increasingly used for occurrence–absence datasets and compound prediction. Integrating edaphic and phytochemical predictors enhances operational relevance for cultivation planning, but geo-referenced phytochemical datasets are scarce, geographically biased, and often lack standardized sampling metadata. Validation practices show an encouraging shift toward spatially structured cross-validation and multi-metric reporting, yet independent external validation is rarely performed. Quality-aware studies frequently document spatial decoupling between high suitability and peak compound concentrations, highlighting the need to co-optimize persistence and product quality. We identified three priority actions: (1) expanded and standardized geo-referenced phytochemical sampling, (2) development of transferable and interpretable modeling workflows, and (3) operational decision frameworks that link SDM outputs to management and value-chain actors. A best-practice checklist for predictor selection, sampling design, model evaluation, and uncertainty communication is provided to guide reproducible, stakeholder-oriented applications.