Application of Generalized Additive Model and Classification and Regression Tree to Estimate Potential Habitat Distribution of Range plant species (Case Study: Khazri Rangelands of Beyaz Plain, Southern Khorasan)

Species distribution models (SDMs) are the basis of informed decisions in vegetation management by quantifying the relationship between species distribution and influential environmental variables. The present study aimed to evaluate the GAM and CART models' performance in estimating the potential habitat distribution as well as recognizing the ecological needs of plant species in the Khezri rangelands of Bayaz plain of southern Khorasan. According to the regional condition and field observation, in an area of about 14500 hectares, vegetation sampling was done using the randomized-systematic method. Eighteen environmental variables including land characteristics, Normalized Difference Vegetation Index (NDVI), and salinity index were used as an estimator to generate maps of predictor variables. After modeling the habitat distribution prediction using CART and GAM methods in R 3.5.2 software, the accuracy of the models was assessed using the subsurface area (AUC) statistics. After determining the threshold by the TSS method, the continuous utility map was converted to the presence/absence map and the degree of conformity of the maps with the kappa index was calculated. Based on the results of the used models, the variables of the base level of the channels network, the vertical distance to the channels network, the depth of the valley, the wetness index, and the relative position of the slope are effective in habitat suitability for species establishment. In general, the GAM method has high accuracy in estimating the habitat distribution range of all species studied (Kappa≥ 0.9). According to the maps obtained from the GAM model, the highest and lowest potential habitats belong to S. rigida and T. serotina species. Therefore, the GAM method can be useful in accurately identifying the ecological needs of plant species and therefore their distribution useful at the local scale. As a result, it is suggested that this method be used as part of a management support system in the protection and restoration of vegetation in the rangelands of the Bayaz plain.