A systematic review of spatial habitat associations and modeling of marine fish distribution: A guide to predictors,...

As species distribution models, and similar techniques, have emerged in marine ecology, a vast array of predictor variables have been created and diverse methodologies have been applied. Marine fish are vital food resources worldwide, yet identifying the most suitable methodology and predictors to characterize spatial habitat associations, and the subsequent distributions, often remains ambiguous. Our objectives were to identify knowledge gaps in fish guilds, identify research themes, and to determine how data sources, statistics, and predictor variables differ among fish guilds. Data were obtained from an international literature search of peer-reviewed articles (2007-2018; n = 225) and research themes were determined based on abstracts. We tested for differences in data sources and modeling techniques using multinomial regressions and used a linear discriminant analysis to distinguish differences in predictors among fish guilds. Our results show predictive studies increased over time, but studies of forage fish, sharks, coral reef fish, and other fish guilds remain sparse. Research themes emphasized habitat suitability and distribution shifts, but also addressed abundance, occurrence, stock assessment, and biomass. Methodologies differed by fish guilds based on data limitations and research theme. The most frequent predictors overall were depth and temperature, but most fish guilds were distinguished by their own set of predictors that focused on their specific life history and ecology. A one-size-fits-all approach is not suitable for predicting marine fish distributions. However, given the paucity of studies for some fish guilds, researchers would benefit from utilizing predictors and methods derived from more commonly studied fish when similar habitat requirements are expected. Overall, the findings provide a guide for determining predictor variables to test and identifies novel opportunities to apply non-spatial knowledge and mechanisms to models.