Evaluation of MaxEnt method for habitat distribution modeling of three plant species in Garizat rangelands of Yazd province, Iran

Abstract

This study aimed to predict geographical distribution of Tamarix ramosissima, Seidlitzia rosmarinus and Cornulaca monocantha in Poshtkouh rangelands and to find the influential variables in the distribution of these species in desert rangelands of central Iran. Eleven environmental factors used to explore the effective environmental variables on given species distribution. Maps of the environmental variables were generated using GIS and Geostatistics facilities. Predictive maps of distribution were produced with maximum entropy method (MaxEnt). Accuracy of model output was assessed by using area under the curve (AUC) and withholding 25 per cent of the data. The agreement of predictive map with actual map was checked by calculating Kappa coefficient. The results indicated that vegetation distribution pattern was mainly related to soil characteristics such as EC, available moisture (AW), lime, organic matter (OM) and elevation. AUC values indicated the high power of MaxEnt to create habitat distribution maps of plant species except C. monocantha (S. Rosmarinus = 0.98, T. ramosissima = 0.99, and C. monocantha = 0.78). Correspondence of actual map with predictive map for S. rosmarinus, C. monocantha and T. ramosissima was assessed at very satisfactory (Kappa=0.76), good (Kappa= 0.61) and poor (Kappa= 0.31) level, respectively.