Modeling species habitats of Juniperus excelsa M.Bieb. using remote sensing and geographic information system in Ardabil and Zanjan province

Remote sensing data provide a high ability to represent habitat characteristics and use in species distribution models. The purpose of this study is to determine the most important of remote sensing predictors, including climatic indices (precipitation and temperature), primary and secondary topographic indices (elevation, slope, direction, Topographic Position Index (TPI) and Terrain Roughness Index (TRI) and remote sensing indices (Global Environmental Monitoring Index (GEMI), Leaf Area Index (LAI), Modified Normalized Difference Water Index (MNDWI), Modified Simple Ratio Index (MSR), Normalized Burn Ratio Index (NBR) and Visible Atmospherically Resistant Index (VARI)) using two species distribution models (Boosted Regression Tree and Random Forest) to predict the presence of J. excelsa in Khalkhal County of Ardabil province and northern part of Zanjan province using SAHM software. The evaluation of prediction models using AUC chart (Area under curve) showed that it is at an excellent level for both the BRT model (0.991) and the RF model (0.974). The most important affecting habitat desirability based on the BRT method include annual precipitation, slope, digital elevation model, temperature, GEMI index and TRI index variables respectively. The most important variables affecting habitat desirability based on the RF method, respectively, include annual precipitation, digital elevation model, GEMI index, slope, VARI index, temperature, MSR index, TRI index, TPI index, NBR index, MNDWI index, LAI index and aspect. The region mapped in the study as suitable habitats’ for the species could be used in the planning strategies with the aim of evaluating the susceptible habitats, the possibility of conservation, reproduction and breeding. Considering that the modeling method choice is the main source of variability in predictions and choosing the best prediction model is not simple, therefore, it is suggested to use a combination of these models instead of relying on the outputs of a single model.