Spatial modeling of Trigonella elliptica potential habitat using environmental variables and machine learning technique in the Rangelands of Yazd province

Identifying plant species distribution and potential habitats which are under degraded in the rangelands ecosystem is an essential challenge in natural resource science. Performance these studies will support rangeland conservation, restoration, and management measures. In this study, the potential habitat of Trigonella elliptica in rangeland of Yazd province was modeled using one of the advanced machine learning models (Random Forest algorithm). We used 11 variables including land use, soil salinity index, rainfall, minimum and maximum temperature, evaporation, elevation, aspect and degree of slope, distance from river, and topographic wetness index, as well as 103 presence points of T. elliptica to improve the model. 70 % of the T. elliptica presence points were randomly selected for model training and 30% of them for model testing. In order to evaluate the model and the importance of environmental variables were used the area under the receiver operator characteristic (ROC) curve and the Jackknife methods respectively. The evaluation results of the model using the ROC curve (AUC> 0.8) showed a very good performance. Error statistics including Accuracy, Precision, Bias, Probability of Detection and False Alarm Ratio showed 0.9, 0.79, 1, 0.93 and 0.04, respectively, which demonstrate the good performance of the model to prediction. In addition, the results of determining the importance of variables showed that the slope degree and following it, elevation and topographic wetness index are more important than other variables in determining the potential habitat of T. elliptica. The map obtained from the prediction of the potential habitat of T. elliptica can be very useful as accurate information in the rangeland management in order to Reclamation the destroyed habitats of this rangeland plant in Yazd province and be highly regarded by rangeland managers.