Modeling the Distribution of Juniper Species (Juniperus excelsa M.Bieb.) Using Multivariate Adaptive Regression Splines (MARS) and Generalized Linear Model (GLM) in Southern Ardabil and Northern Zanjan Provinces

Remote sensing data, with its comprehensive and accessible nature, facilitates the understanding of plant-environment interactions. Integrating such information into species distribution models can create global maps of species distributions and aid in their conservation and restoration. This research aims to determine the remote sensing predictors—including climatic, topographic, and remote sensing indices—to predict the distribution of J. excelsa in Khalkhal County, Ardabil Province, and Northern Zanjan Province.

The study area includes Khalkhal County in Southern Ardabil Province and the northern parts of Zanjan Province (Zanjan, Tarem, and the northeastern part of Mahenshan), covering approximately 1,036,742 hectares. Suitable locations were selected for sampling based on a land use map and preliminary field surveys. Coordinates of J. excelsa presence and absence were recorded using GPS, resulting in 1,577 recorded points. Areas with at least one square kilometer of dominance and sampling sites at least one kilometer apart were considered. To reduce sampling bias and autocorrelation, SDM toolbox and multi-distance methods in ArcGIS software were used. Pseudo-absence points were created randomly using the Random Selection tool in ArcGIS, with approximately 1 km intervals, and verified through field visits. Digital elevation model, slope, and aspect maps were prepared using topographic maps at a 1:25000 scale and the Spatial Analyst tool in ArcGIS. Precipitation and temperature layers were prepared using satellite images with high temporal resolution. Remote sensing indices (GEMI, LAI, MNDWI, and VARI) were calculated and rasterized using ENVI software. MARS and GLM models were performed in SAHM software.

The area under the curve (AUC) was 0.967 for GLM and 0.984 for MARS, indicating excellent model performance. Key variables for GLM included VARI, slope, temperature, digital elevation model, LAI, MNDWI, annual precipitation, and GEMI indices. For MARS, important variables were VARI, annual precipitation, digital elevation model, slope, MNDWI, temperature, LAI, GEMI, and aspect. Suitable habitats for J. excelsa were estimated to be 349,569 hectares (33.7% of the area) by GLM and 340,610 hectares (32.8% of the area) by MARS. Habitat elevation ranged from 450 to 2800 meters, with species presence increasing with slope percentage. Species presence was observed within precipitation ranges of 300 to 800 mm and temperatures between 6 to 18°C. GEMI index showed highest presence at 0.18 and lowest at 0.24. LAI index indicated highest presence at 0 and lowest between 0.05 to 0.15. MNDWI index remained constant from -0.5 to -0.1, then decreased. VARI index showed increasing presence from -0.18 and remained constant from -0.18 to 0.3.

Due to the different algorithms used, it is more reliable to employ multiple prediction methods rather than a single one. The models presented are valid within the study area's habitat conditions and should be tested in other similar regions for validation.