Machine Learning Approach to Understand the Relationship Between Melting Arctic Sea Ice Extent and North Atlantic Oscillation

In our study, we construct a phase-space using the velocity and acceleration of the Arctic Sea Ice Extent (SIE) as two variables. Our analysis shows that in recent times the melting arctic SIE resulted in increasing phase-space volume, i.e., the phase-line distribution function has not been constant along the trajectories. We aim to investigate the effect of melting Arctic SIE on the climate particularly on the North Atlantic Oscillation (NAO) index, a measure of variability in the atmospheric pressure at sea level. As we observe that the SIE and SST are highly correlated, we also observe the lead-lag effect of SST and NAO. On a daily, monthly, and yearly basis SST and NAO are anti-correlated to each other. It shows melting SIE and increasing SST has a significant effect on the changing weather pattern of the North Atlantic region, especially in Europe and North America. Based on a Granger causal time series model, we find that the SST and its first and second derivatives (momentum and acceleration) do have a significant (at 0.01% level) effect on the NAO index. One of the criticisms against the statistical model in climate literature is that the predictor's relationships with the target variable do not change over time. We addressed the criticism by developing the Granger Causal Dynamic Linear model, which uses the Bayesian Kalman filter to update the relationship between predictors and the dependent variable dynamically over time. Most of the studies in the previous literature were done on a specific area of the Arctic Sea region. However, we analyzed it on a global/large level. Our statistical learning model hints that a frequent colder climate in Eastern USA and Northern Europe, particularly during the winter season.