Abstract
Wild Rosa species are valuable genetic resources for biodiversity conservation, functional foods, ornamental horticulture, and agroforestry.We modelled potential habitat suitability for Rosa spinosissima L., Rosa alberti Regel, and Rosa acicularis Lindl.across Kazakhstan using the Maximum entropy algorithm (MaxEnt) and occurrence records from field observations, the Global Biodiversity Information Facility (GBIF), and iNaturalist.Nine non-collinear climatic, topographic, and soil predictors were retained after correlation screening, and model complexity was tuned across feature-class and regularisation-multiplier combinations.Random cross-validation produced mean test values of the Area under the receiver operating characteristic curve (AUC) of 0.942, 0.966, and 0.957 for R. spinosissima, R. alberti, and R. acicularis, respectively.A supplementary four-fold spatialblock sensitivity analysis yielded lower mean AUC values (0.809-0.900), indicating that random partitioning overestimated performance while retaining moderate-to-good discrimination.Slope, elevation, soil organic carbon, temperature variables, and soil water-holding capacity were the principal predictors.High and very high suitability covered 5.0%, 1.3%, and 3.6% of the modelled national area, respectively, with priority zones concentrated in eastern and southeastern mountain systems and central-eastern Kazakhstan.The maps identify targets for field verification, germplasm collection, conservation, and carefully matched agroforestry trials, but represent potential suitability rather than confirmed occupancy.