The North Atlantic model attractor produced from a 50,000-year integration. The model exhibits both chaotic and more stable regions of phase space, giving rise to rich dynamical behaviour with implications for understanding decadal predictability and atmosphere–ocean coupling in the North Atlantic climate system.
Andrew Nicoll’s paper introduces a data-driven approach for learning how the ocean and atmosphere of the North Atlantic climate system interact on decadal timescales. The framework learns nonlinear differential equations directly from observations, producing interpretable low-order dynamical models rather than relying on “black box” machine learning. This allows us to investigate the dynamical relationships and physical mechanisms underlying climate variability. The models capture interactions between Atlantic Multidecadal Variability (AMV), the North Atlantic Oscillation (NAO), and North Atlantic precipitation, reproduce key features of observed decadal variability, and demonstrate out-of-sample predictive skill. Importantly, the models reveal an active role for North Atlantic precipitation in atmosphere–ocean coupling, highlighting how interpretable data-driven approaches can provide new insights into complex climate dynamics.
For more information, check out Andrew’s paper, just published in ESD.