Abstract

Using machine learning to make skillful predictions of the wintertime North Atlantic Oscillation, James Keeble, NCAS , J. Keeble, Y. Y. S. Yiu, P. J. Nowack, P. T. Griffiths, and J. A. Pyle. The North Atlantic Oscillation (NAO) has a well-documented effect on wintertime climate in both Europe and North America. It has also been suggested, using statistical techniques and climate models, to possess a degree of predictability on seasonal timescales, an important consideration when planning for potential financial, ecological and human health impacts of wintertime weather. Here, we explore the use of Machine Learning (ML) techniques, specifically Ridge regression, to enhance and understand seasonal predictability of the wintertime NAO index. We find that ML achieves similar predictive skill to other methods (r≈0.6). Furthermore, we show that ML successfully identifies those regions which have been shown in other studies to be important for deriving predictability of the NAO index. We conclude that Ridge regression is a promising alternative technique for making seasonal forecasts of the wintertime NAO and for identifying the sources of this predictability.