Predicting marine species distributions: Complementarity of food-web and Bayesian hierarchical modelling approaches
The spatial prediction of species distributions from survey data is a significant component of spatial planning and the ecosystem-based management approach to marine resources. Statistical analysis of species occurrences and their relationships with associated environmental factors is used to predict how likely a species is to occur in unsampled locations as well as future conditions. However, it is known that environmental factors alone may not be sufficient to account for species distribution. Other ecological processes including species interactions (such as competition and predation), and the impact of human activities, may affect the spatial arrangement of a species. Novel techniques have been developed to take a more holistic approach to estimating species distributions, such as Bayesian Hierarchical Species Distribution model (B-HSD model) and mechanistic food-web models using the new Ecospace Habitat Foraging Capacity model (E-HFC model). Here we used both species distribution and spatial food-web models to predict the distribution of European hake (Merluccius merluccius), anglerfishes (Lophius piscatoriusand L. budegassa) and red mullets (Mullus barbatus and M. surmuletus) in an exploited marine ecosystem of the Northwestern Mediterranean Sea. We explored the complementarity of both approaches, comparing results of food-web models previously informed with species distribution modelling results, aside from their applicability as independent techniques. The study shows that both modelling results are positively and significantly correlated with observational data. Predicted spatial patterns of biomasses show positive and significant correlations between modelling approaches and are more similar when using both methodologies in a complementary way: when using the E-HFC model previously informed with the environmental envelopes obtained from the B-HSD model outputs, or directly using niche calculations from B-HSD models to drive the niche priors of E-HFC. We discuss advantages, limitations and future developments of both modelling techniques.