Comparison of species distribution models in determining the habitat landscape of Pistacia vera L. specie in Razavi Khorasan province

Introduction
Global climate change has led to change in the ecological amplitude of plant growth, expand plant adaptation to hot climates, and decrease plant adaptation to cold climates. Climate change resulting from human activities occurs at such a speed that many species will not be able to adapt to it. These changes have led to a change in the range of plants growth. Such high-speed changes have caused subsequent changes in the structure and entire ecosystems of the earth, therefore predicting the effect of climate change on the distribution of plant species has become a major field of research for its conservation measures and programs. Changes in the range of distribution of plants are mostly predicted by species distribution models. In this sense, every environmental factor affecting the distribution of plant species has a minimum, maximum and optimal value, which, in combination with other factors, separates the territory of the species and forms an ecological niche. These models are used to investigate species distribution and are based on ecological niche theory. This research was conducted with the aim of determining the potential habitats of Pistacia vera L. species and the factors affecting it in the present and future in Razavi Khorasan province.
 
Materials and Methods
For this purpose, 28 bioclimatic variables including topographic (4 cases), climatic (19 cases), soil (4 cases), and geological (1 case) factors as prediction variables have been analyzed for the correlation coefficient. The variables with high correlation (more than 80 %) have been removed. Environmental variables in ASCII format along with presence points were added for modeling in R software of the desired species. According to the size of the study area, sampling of data points was done based on the field visit during the period 2021-2022 from the introduced areas. through using the Global Positioning System (GPS) of 129 points from 8 regions (as points of presence) were recorded. Then, in order to prevent spatial autocorrelation and reduce the sampling error, the useful areas were converted into 1000×1000 meters grids in ArcGIS 10.5 software, and one presence point was obtained from each cell. In the modeling process, 70 % of the presence points (Pistacia vera L.) were used to generate models and 30 % of the presence points were used to evaluate the performance of the models. Also, to increase the modeling accuracy, the number of repetitions was considered 10. Then all data and points through R software and using Biomed 2 package models including GLM, GBM, CTA, ANN, SRE, FDA, MARS, RF, and MaxEnt Phillips models, in determining the relationship between vegetation and environmental factors in rangelands of Khorasan Razavi province at current and future distribution of this species in 2080-2100 were predicted under climate scenarios ssp1-2.6 and ssp5-8.5 model. The accuracy of the models was evaluated using the values of KAPPA, TSS and ROC indices, which are prominent and widely used indices for determining and identifying the areas of equal potential.
 
Results and Discussion
The variables of climatic factors were removed from the modeling due to the high correlation of 80 %, and the analysis was done using four topographic factors, eight climatic factors, four soil factors and one geological factor. The results of this research showed that according to the accuracy evaluation index, the best modeling for the present time is done by the random forest (RF) model with the ROC, KAPPA, and TSS equal to 100. In the future, the 2.6 and 8.5 scenarios of the random forest model for the ROC, KAPPA, and TSS indicators, with the accuracy of 0.999, 0..982, and 0.989 respectively, have the highest level of accuracy; Also, in the random forest model, the factors that had the greatest impact included: Bio12 (annual precipitation) and Bio15 (seasonal precipitation changes) and land unit at the present time, in the future time under the scenario 2.6 Bio12 (annual precipitation) and Bio15 (seasonal precipitation changes) and DEM and in the scenario 8.5 Bio15 (seasonal precipitation changes) and Bio12 (annual precipitation) and aspect. The results of the relative importance show the great influence of climatic factors on the distribution of this species. It is most present in the habitat with an annual rainfall of 200-285 mm, and more than this amount of rainfall was associated with a decrease in suitability for the establishment of the species. Besides, the height of 800-1300 meters above sea level and rainfall changes up to 7.8 mm in seasonal rainfall also had a positive effect on the suitability of the habitat for the presence of wild pistachio. Also, the most desirable habitat is in low to relatively high hills with a rounded and sometimes flat top consisting of limestone, metamorphic, conglomerate, and shale sandstones and a slope of 40 to 50 % and with shallow to relatively deep gravelly soils. The highest distribution of Pistacia vera L. species is in the northeastern region to the east of Khorasan province. In general, by examining the outputs of the random forest model and comparing the areas prone to the growth of Pistacia vera L. species in the present and future climate scenarios, it can be stated that the trend of stable habitat in the province can be expected.
 
Conclusion
The results of this research can be used to identify areas prone to growth, improvement, development, protection, economic exploitation, and expansion of the habitat of Pistacia vera L. species. From the ecological point of view, the wild pistachio species is considered as one of the most important factors preventing and destroying land in the high mountains of arid and semi-arid regions in many geographical and ecological regions. On the other hand, the economic importance and the income-generating aspect of wild pistachios are also important for local operators. In general, it can be stated that vector machine models provide very good performance for identifying such prone areas. In this research, an attempt was made to evaluate different species distribution vector machine models, and then the most suitable model, which was random forest, was selected.