North Atlantic sea surface temperatures (NASST), particularly in the subpolar region, are among the most predictable locations in the world’s oceans. However, the relative importance of atmospheric and oceanic controls on their variability at multidecadal timescales remain uncertain. Neural networks (NNs) are trained to examine the relative importance of oceanic and atmospheric predictors in predicting the NASST state in the Community Earth System Model 1 (CESM1). In the presence of external forcings, oceanic predictors outperform atmospheric predictors, persistence, and random chance baselines out to 25-year leadtimes. Layer-wise relevance propagation is used to unveil the sources of predictability, and reveal that NNs consistently rely upon the Gulf Stream-North Atlantic Current region for accurate predictions. Additionally, CESM1-trained NNs do not need additional transfer learning to successfully predict the phasing of multidecadal variability in an observational dataset, suggesting consistency in physical processes driving NASST variability between CESM1 and observations.

The Northeast U.S. shelf (NES) is an oceanographically dynamic marine ecosystem and supports some of the most valuable demersal fisheries in the world. A reliable prediction of NES environmental variables, particularly ocean bottom temperature, could lead to a significant improvement in demersal fisheries management. However, the current generation of climate model-based seasonal-to-interannual predictions exhibit limited prediction skill in this continental shelf environment. Here we have developed a hierarchy of statistical seasonal predictions for NES bottom temperatures using an eddy-resolving ocean reanalysis dataset. A simple, damped local persistence prediction model produces significant skill for lead times up to ~6 months in the Mid-Atlantic Bight and up to ~11 months in the Gulf of Maine, although the prediction skill varies notably by season. Considering temperature from a nearby or upstream (i.e. more polewawrd) region as an additional predictor generally improves prediction skill, presumably as a result of advective processes. Large-scale atmospheric and oceanic indices, such as Gulf Stream path indices (GSIs) and the North Atlantic Oscillation index, are also tested as predictors for NES bottom temperatures. Only the GSI constructed from temperature observed at 200 m depth significantly improves the prediction skill relative to local persistence. However, the prediction skill from this GSI is not larger than that gained using models incorporating nearby or upstream shelf/slope temperatures. Based on these results, a simplified statistical model has been developed, which can be tailored to fisheries management for the NES.