Modelling potential distribution of the threatened tree species Juniperus oxycedrus: How to evaluate the predictions of different modelling approaches?
Questions: How can predictions of potential species distribution derived from presence-only data and different modelling algorithms be compared and evaluated? Where does suitable habitat for Juniperus oxycedrus exist within the study area and which bioclimatic variables prove to be most important in the prediction of J. oxycedrus potential distribution? Location: Central High Atlas, Morocco. Methods: Ecological niche factor analysis (ENFA), maximum entropy approach (MAXENT) and generalized linear models (GLM) were applied to either presence-only data of J. oxycedrus (ENFA and MAXENT) or presence-absence data (GLM), using bioclimatic variables as predictors. Model accuracy of ENFA, MAXENT and GLM was assessed using their specific evaluation measures and prediction success by means of the minimal predicted area (MPA). Finally, the three maps of potential species distribution were intersected to investigate their consistency in the geographic distribution of predicted suitable habitat. Results: Species distribution models calculated by ENFA, MAXENT and GLM show good model quality according to their evaluation measures. However, calculation of MPA revealed considerable discrepancies in geographic distribution and spatial extent of areas predicted as suitable habitat for J. oxycedrus. Conclusion: Suitable habitat for J. oxycedrus is predicted using all applied modelling approaches in 8\% of the study area (mostly the north-western part). MAXENT gave the best results considering model accuracy and prediction success according to the MPA scores. Predictors representing maximum temperature and potential radiation were selected in all methods to explain species occurrences. This points to a possible shift of the species' potential distribution in cases of a rise in summer temperature due to climate change.