Clearing the noise to predict the rhythm of the North Atlantic climate

Decadal prediction of the North Atlantic Oscillation (NAO) is essential for developing decadal climate services and improving climate forecasts across the North Atlantic region. However, extracting robust predictive skill at these timescales remains challenging due to internal variability and ensemble limitations. Here, we introduce an innovative postprocessing framework called temporally aligned averaging (TAA) that substantially enhances the prediction skill of the winter NAO, increasing the explained variance from 34 to 69% over 1961-2023, and providing reliable forecasts even for short-term (4 to 5 years) averages. Combined with a subsampling strategy that fully exploits large ensembles from a single model to circumvent inconsistencies and biases arising from multimodel mixtures, the TAA method notably strengthens decadal precipitation forecasts, enhancing its explained variance from 16 to 54% over Scandinavia and from 1 to 58% over the Mediterranean basin. This substantial advancement in decadal forecasting could have broad implications for various sectors by improving climate risk assessments and adaptation strategies.