Simulation of potential endangered species distribution in drylands with small sample size based on semi-supervised models

Abstract Identifying suitable habitats for endangered species is critical in order to promote their recovery. However, conventional distribution models (SDMs) need large amounts of labeled sample data learn the relationship between and environmental conditions, are difficult fully detangle role environment species, which very sparsely distributed have heterogeneity. This study’s first innovation used semi-supervised model accurately simulate with a small size. The performance was compared three SDMs, namely Maxent, generalized linear model, support vector machine. Applying Populus euphratica (P. euphratica) lower Tarim River basin (TRB), Northwest China. results showed that exhibited better than SDMs an accuracy 85% when only using 443 P. samples. All developed smaller sizes exhibit worse prediction habitat suitability areas while still excellent. mainly near river channel TRB, accounting 13.49% study area. has enormous land potential restoration . here can be evaluate size, provide basis conservation species.