Climate Covariate Choice and Uncertainty in Projecting Species Range Shifts: A Case Study in the Eastern Bering Sea

Species distribution models (SDMs) are critical to the adaptive management of fisheries under climate change. While many approaches projecting marine species range shifts have incorporated the effects of temperature on movement, there is a need to incorporate a wider suite of ecologically relevant predictors as temperature-based SDMs can considerably under- or over-estimate the rate of species responses to climate shocks. As a subarctic ecosystem at the sea ice margin, the Eastern Bering Sea (EBS) is warming faster than much of the global ocean, resulting in the rapid redistribution of key fishery and subsistence resources. To support long-term planning and adaptation, we combine 40 years of scientific surveys with a high-resolution oceanographic model to examine the effects of bottom temperature, oxygen, pH and a regional climate index (the extent of the EBS 'cold pool') on range projections through the end of the century. We use multimodel inference to partition uncertainty among earth systems models, climate scenarios and distribution model parameterizations for several ecologically and economically important EBS groundfish and crabs. Covariate choice is the primary source of uncertainty for most species, with models that account for spatial responses to the cold pool performing better and suggesting more extensive northward movements than alternative models. Models suggest declines in the probability of occurrence at low pH and oxygen concentrations for most species. We project shifts that are directionally consistent with, yet larger than those previously estimated for most species, suggesting that accounting for large-scale climate variability in species distribution models may substantially alter range projections.