Predicting distribution of plant species in arid rangelands of central Iran using probabilistic methods

The quantification of complex relationships between environmental variables and plant habitat distribution is difficult and crucial. The present study employed Logistic Regression (LR), Maximum Entropy (MaxEnt) and Artificial Neural Network (ANN) methods to model plant habitat distribution and identifies the most appropriate modeling approach. The study was conducted in Poshtkouh rangelands, Yazd Province, central Iran. Vegetation was sampled using randomize-systematic sampling method. Soil samples were taken from 0–30 and 30–80 cm depths. The highest values of Kappa index (0.57) belonged to the ANN. Average Kappa values for the MaxEnt and LR were 0.56 and 0.48, respectively. The performance of LR model was higher for species with high marginality and low tolerance, e.g. Cornulaca monacantha, and lower for species with low marginality and high tolerance, e.g. Artemisia sieberi. The ANN and MaxEnt provided better models for species with complex distribution patterns such as widespread species. In fact, differences in the optimal ecological range of plant species, could affect the accuracy of predictive distribution models.