Evaluating Ecological Niche Models: A Comparison Between Maxent and GARP for Predicting Distribution of Hevea brasiliensis in India

Abstract

Selection of appropriate ecological niche model for predicting species niche distribution has been a challenge considering the type of species and input variables. Therefore, two ecological niche modelling approaches (Maxent and GARP) are employed to predict the present distribution of a planted species (Hevea brasiliensis Muell. Arg.) in two bio-geographical regions of India: Western Ghats (WG) and North east (NE) regions. The difference between two approaches is in the algorithm and kinds of species data (presence-only or presence and absence) used for model training. GARP over-estimates were observed more in NE as compared to that of WG. Maxent predicts Hevea distribution more accurately in both regions as it considers presence-only data, which appears to be more accurate for this species. The over-prediction of Hevea niche distribution by GARP especially in NE may be attributed to inaccurate and insufficient ‘absence data’ as compared to ‘presence data’. The model accuracy estimator, AUC failed to attribute the difference in model predictability between Maxent and GARP, whereas partial-AUC is found to be better estimator of model spatial accuracy. Therefore, Maxent is found to be more appropriate model for predicting the niche distribution of a plantation species like Hevea.