Coastal communities are at risk from long-term sea level rise and decadal variations, particularly in the North Atlantic. This study uses a self-organizing map (SOM) framework and 5000-year sea level anomaly data from climate model simulations to assess North Atlantic sea level variability and predictability. The analysis identifies preferred transitions among patterns of variability, revealing long-term predictability on decadal timescales linked to shifts in Atlantic meridional overturning circulation (AMOC) phases. Combining the SOM with model-analog techniques demonstrates prediction skill comparable to initialized hindcasts for large-scale sea level patterns and low-frequency coastal variations. Additional short-term predictability is found after removing low-frequency signals, stemming from slow gyre circulation adjustments triggered by North Atlantic Oscillation-like variability. The study highlights machine learning's potential for assessing predictability sources and enabling long-term climate prediction.

Qinxue Gu, Liping Zhang, Liwei Jia, Thomas L. Delworth, Xiaosong Yang, Fanrong Zeng, William F. Cooke, Shouwei Li