Comparing Different Modeling Techniques for Predicting Presence-absence of Some Dominant Plant Species in Mountain Rangelands, Mazandaran Province

In applied studies, the investigation of the relationship between a plant species and environmental variables is essential to manage ecological problems and rangeland ecosystems. This research was conducted in summer 2016. The aim of this study was to compare the predictive power of a number of Species Distribution Models (SDMs) and to evaluate the importance of a range of environmental variables as predictors in the context of rangeland vegetation. In this study, Aflah rangelands with 5721 ha were selected. In this research, predictor variables included climatic, topographic and edaphic parameters. The sampling method was equal random-classification for vegetation and soil. Topographic factors including slope, elevation and aspect were determined in Arc GIS software. In each sample unit, 10 plots were established (total 350 plots) and the lists of the species, their number, their presence or absence were recorded. The efficacy of five different modelling techniques to predict the distribution of five dominant rangeland plant species (Agropyron repens, Festuca ovina, Leucopoa sclerophylla, Stachys lavandulifolia and Tragopogon graminifolius) was evaluated. The models were Generalized Linear Model (GLM), Classification and Regression Trees (CART), Boosted Regression Trees (BRT), Generalized Additive Models (GAM), and Random Forest (RF). Data analysis was done using the R software, version 3.1.1. The results showed that GAM model demonstrated most consistently high predictive power over the species in the rangeland context investigated here. GAM had higher Area Under the Curve (AUC). The AUC (0.67, 0.77, 0.69, 0.64 and 0.60 and Kappa values (0.10, 0.10, 0.19, 0.01 and 0.11) were obtained for Agropyron repens, Festuca ovina, Leucopoa sclerophylla, Stachys lavandulifolia and Tragopogon graminifolius, respectively. GAM model exhibited the most predictive power. The importance analysis of the environmental variables showed that N, pH and aspect were the most important variables in the GAM model. Overall, N, P and C/N soil (0.452, 0.437 and 0.389) were the most important environmental variables.