Iranian Journal of Applied Ecology

Status determination of wildlife habitats is very important in conservation programs and management of wildlife. So, in this study Ursus arctos habitat suitability was modeled using maximum entropy algorithm (MaxEnt) in Shimbar protected area. In order to model the habitat suitability, after investigating and resolving the spatial autocorrelation of occurrence records, spatially independent localities were divided into the calibration and evaluation sets and then were combined with 10 environmental variables (VIF<10) selected by MMS software. The performance of predictive models was tested using AUC and jackknife validation test. So, we applied two different thresholds, the LPT threshold and 10% threshold to generate presence/absence map. Also other Jackknife tests applied to measure variables importance. The results showed that predictive model was more efficient than random model (AUC=0.980). In addition, the potential suitable areas cover 20.75% of study area. The MaxEnt model had 88.46% success rate and was statistically significant (P = 0.000). Results of Jackknife showed that ‘plant type’ variable alone contains valuable information for modelling. Our study demonstrated that habitat suitability was successfully predicted by MaxEnt modelling, so this methodology might provide a powerful tool for improving the wildlife habitats information.

Several modelling techniques have been developed for habitat suitability modelling. In the meantime, the Fuzzy Inference System (FIS) with ability to model uncertainty of input variables is an effective method to model wildlife species habitat suitability. So, Persian Leopard habitat suitability was predicted in Shimbar Protected Area using FIS. Therefore, the effective environmental variables were determined. We also defined and determined the linguistic variables, linguistic values, and range of them. Then, we designed the membership functions of the fuzzy sets of the input and output variables. Also, the definition of the fuzzy rules in the system was performed. Finally, the defuzzification of output was carried out. The accuracy of the predictive model was tested using AUC. Also, 11 FISs were developed to determine sensitivity of the models and important variables in modelling. The results showed that the predictive model was more efficient than the random model (AUC=0.960). In addition, the ‘distance to capra’ was the most important predictor. According to the success of FIS in Persian Leopard habitat suitability modelling, we suggest this method to improve and complete the existing spatial information of wildlife habitats in Iran, especially about regions and species that have been less studied.