Seasonal Predictability of Sea Ice and Bottom Temperature Across the Eastern Bering Sea Shelf

Seasonal sea ice plays a key role in shaping the ecosystem dynamics of the eastern Bering Sea shelf. In particular, it leads to the formation of a characteristic pool of cold water that covers the bottom of the shelf from winter through summer; the extent of this cold pool is often used as a management index for distribution, productivity, recruitment, and survival of commercially important fish and shellfish species. Here, we quantify our ability to seasonally forecast interannual variability in Bering Sea bottom temperature and sea ice extent. Retrospective forecast simulations from two global forecast models are downscaled using a regional ocean model; the retrospective forecast simulations include 9‐month to 12‐month forecasts spanning 1982–2010. We find that dynamic forecasting can predict summer bottom temperatures across the eastern Bering Sea shelf with lead times of up to 4 months. The majority of the prediction skill derives from the persistence signal, and a persistence forecast is comparably skillful to the dynamic forecast at these lead times. However, forecast skill of sea ice advance and retreat is low when a forecast model is initialized before or during the ice season (October–February); this limits the ability of either dynamic or persistence models to predict summer bottom temperatures when initialized across the late fall to early spring months. Plain Language Summary The Bering Sea region is home to a large number of important fisheries. Fluctuations in fish distribution, abundance, and productivity can be influenced by environmental factors; in the Bering Sea, the temperature of the deep shelf water can affect where groundfish are located, how much of their preferred prey is available, and how many juvenile fish and shellfish survive each year. In this study, we test whether we can successfully predict whether bottom water will be colder or warmer than average in a given year. We do this by running forecast models for previous years and comparing the model output to observations collected during the same period. We find that we can skillfully predict summer bottom temperature if we start a forecast four or fewer months in advance of the summer period. However, beginning the forecast during the winter (when sea ice is present) or during the prior fall results in low skill. The 4‐month forecasts can be useful when planning for summer surveys and to provide advance notice to the North Pacific Fishery Management Council of any unusual sum