Comparison of GARP and Maxent in modelling the geographic distribution of Bacillus anthracis in Zimbabwe

Abstract

A number of presence-only models can be used in the prediction of the geographic distribution of diseases and/or their vectors. The predictive performance of these models differs depending on a number of factors but primarily the modeled species’ ecological traits. In this study, the performance of GARP and Maxent, two of the most commonly used modelling methods were compared in predicting presence and absence of anthrax in Zimbabwe using accuracy, sensitivity, specificity, Kappa statistic and the Jaccard coefficient as measures of model performance. The results showed that GARP had higher accuracy than Maxent (GARP = 0.70, Maxent = 0.67). Both methods had equal sensitivity (sensitivity = 0.71), but GARP had higher specificity (GARP=0.70, Maxent=0.67). Both Kappa and the Jaccard coefficient were also higher for GARP (0.335; 0.36) than for Maxent (0.295; 0.34). The results imply that GARP has superior performance over Maxent and is recommended for modelling species habitat suitability.

Keywords: ENMs, GARP, Maxent, Anthrax