The purpose of this study was to evaluate the competency of logistic regression (LR) and maximum entropy (MaxEnt) models to predict the distribution of Dorema ammoniacum D. Don. in rangeland habitats in the central region of Iran, Yazd province.

The potential distribution map of Dorema ammoniacum D. Don. was prepared. The homogenous habitats were identified, and vegetation sampling was conducted using a systematic random method. The data including: soil (physical and chemical properties), physiographic (slope, aspect and altitude), and vegetation data (presence and absence) were used. Soil sampling was performed at two depths of 0-30, and 30-60 cm. The required maps were prepared using interpolation method. Statistics were taken from 90 plots along 9 transect both in the presence and absence area. Response curve and Jackknife test (for MaxEnt method) were employed to identify the most important environmental predictive factors. The kappa index was used to determine the agreement between the actual and predicted maps.

The accuracy of predicted map was weak in LR Model (AUC= 0.65), but it was considerably high in the MaxEnt model (AUC=0.87). The agreement between the predicted map of MaxEnt model, and ground truths was very good (kappa=0.74), and the agreement between predicted map generated by LR with the ground-truths was medium (kappa=0.5).

This plant has a limited ecological niche; therefore, the MaxEnt model could take precedence over the LR model because the only data it employs is the presence of the species.