Advancing statistical models to reveal the effect of dissolved oxygen on the spatial distribution of marine taxa using thresholds and a physiologically based index

The rapid pace of ocean change has prompted a need to forecast likely future species distributions. Species distribution models are often categorized as either correlative (statistical) or mechanistic, and each has limitations both for advancing understanding and for prediction. Here we sought to benefit from mechanistic understanding of how and why low dissolved oxygen affects species' distributions by applying physiologically informed statistical models to the spatial distribution of sablefish Anoplopoma fimbria, a deep‐dwelling commercially important groundfish. We fit spatial models to trawl‐survey data on catch rate, local temperature and dissolved oxygen, and estimated parameters of the metabolic index, which provided a way to express the temperature‐dependence of oxygen tolerance. We fit generalized linear mixed effects models with Gaussian random fields to capture the latent spatially fixed variables, and included both linear and breakpoint functions for pO2 and the metabolic index. The best fitting models all included breakpoint effects of pO2, and the estimated threshold value of 0.05 atm is close to levels in laboratory studies where metabolism begins to decline. Models based on the metabolic index were not as well supported as those that included pO2, likely because of the decrease in temperature and slight increase in pO2 at deep (> 800 m) depths. These findings illustrate that statistical models of species distributions can be improved by incorporating knowledge of how physiological mechanisms operate. Furthermore, they illustrate that even species with high tolerance for low dissolved oxygen may undergo species distribution shifts in the face of growing oxygen depletion in coastal ocean ecosystems.