Modeling habitat suitability for saffron (Crocus sativus L.) using AI-based approaches to support sustainable industrial cultivation

Saffron is a highly strategic medicinal plant in Iran that requires precise habitat suitability analysis to aid expansion of its cultivation areas. The objective of this study was to create a habitat suitability map for saffron in the region, identify the factors that have the most significant influence on saffron growth, and evaluate various machine learning models to determine which algorithm offers the best accuracy for the modeling process. For this purpose, 17 layers of important factors that affect saffron growth were collected and integrated into the model. Three machine learning algorithms were utilized for the modeling process: Classification and Regression Trees (CART), Multivariate Adaptive Regression Splines (MARS), and Mixture Discriminant Analysis (MDA), all implemented in R software. The evaluation of variable importance and machine learning model performance was conducted using the PLS and GLM algorithms and Receiver Operating Characteristic (ROC)Area Under the Curve (AUC), respectively. The results revealed that CART, MARS, and MDA achieved high accuracies of 97.6, 95.1, and 92.6 %, respectively. This study identified several critical factors affecting saffron growth, including rainfall, slope, and elevation. Habitat suitability maps identified 44–49 % of the study area as “high” and “very high” suitability classes, concentrated mainly in the southern and eastern regions. In contrast, 22–26 % of the land is classified as “low” suitability, predominantly located in the central part of the study area, while the remaining 25–34 % is categorized as “moderate” suitability. Based on the suitability map, we devised a strategic plan to expand the cultivation area in regions with high suitability and to avoid areas with low suitability. Integrating these maps into regional agricultural planning enables stakeholders to anticipate environmental challenges, minimize financial risks, and enhance the reliability of yield forecasts. This approach also facilitates efficient allocation of resources and minimizes trial and error when extending cultivation to new areas. • The CART, MARS, and MDA machine learning models demonstrated excellent accuracy in predicting saffron habitat suitability. • Rainfall, elevation, slope, and temperature were identified as the most influential factors for saffron growth. • Approximately 44–49 % of the study area was classified as highly or very highly suitable for saffron cultivation which mainly located in the southern regions. • Habitat suitability maps provide a strategic guide for saffron expansion while minimizing trial and error and improve efficiency. • The integration of AI and spatial modeling in this research offers a scalable, data-driven tool to guide sustainable saffron cultivation.