Inferring distributions of chirodropid box-jellyfishes (Cnidaria: Cubozoa) in geographic and ecological space using ecological niche modeling

Abstract

Geographic distributions of many marine species are poorly documented or understood, which is particularly true for marine invertebrates. Ecological niche modeling (ENM) offers a means to address this issue, but to date most studies using ENM have focused on terrestrial taxa. In general, ENM relates environmental information to species’ occurrence data to estimate the ecological niche of a species, rather than just interpolating a geographic distribution. This process leads to predictions of suitable habitat that generally exceed the range actually inhabited by a single species: such areas of geographic over-prediction (commission) may be inhabited by closely related species, and the model thus offers the inferential power to predict the potential distributions of these species as well. We explored the utility of ENM to investigate potential distributions of chirodropid box-jellyfishes (Cnidaria: Cubozoa), a group of highly toxic invertebrates whose biogeography is poorly understood. We were able to predict reported occurrences of box-jellyfishes throughout the Indo-Pacific from data of closely related species. By doing so, we demonstrate that geographic over-prediction in ENM can be desirable when concerned with predictions beyond current knowledge of species’ distributions. Several methods are used for ENM; here, we compared the 2 most commonly used methods, the Genetic Algorithm for Rule-Set Predictions (GARP) and a maximum entropy approach (Maxent). Our comparison shows that Maxent may be more prone to overfitting, whereas GARP tends to produce broader predictions. Transforming continuous Maxent predictions into binary predictions remedies problems of overfitting, and allows for effective extrapolation into unsampled geographic space.