Predictability of North Atlantic Ocean Temperatures in observations and CMIP5 models

Understanding the extent to which Atlantic sea surfaces temperatures (SST) are predictable is important due to the strong climate impacts of Atlantic SST on Atlantic hurricanes and temperature and precipitation over adjacent landmasses. However, models differ substantially on the degree of predictability of Atlantic SST and upper ocean heat content (UOHC). In this work, predictability timescales for SST and UOHC are estimated purely from ocean observations, and these observed predictability timescales are compared to those from control runs of CMIP5 models. Observations indicate that predictability timescales for SST and UOHC are longest in the subpolar gyre, with timescales of 4—6 years in this region. Predictability timescales for UOHC, and to a lesser extent SST, depend strongly on the wintertime mixed layer depth, indicating the importance of the ocean thermal memory in setting predictability timescales. Furthermore, on the timescales resolved by the observations (about a 70 year record), neither SST nor UOHC exhibits significant oscillatory behavior. Predictability timescales for SST and UOHC in CMIP5 models are also longest in the subpolar gyre, and in this region timescales range from 4—10 years in different models. Predictability timescales in models also depend on the wintertime mixed layer depth, but the relationship is not as strong as in observations. Furthermore, some models exhibit significant oscillatory variations of SST and UOHC. Factors leading to differences between observed and simulated predictability timescales, including length of data records, external forcing, and model-dependent features, will be addressed.