Advancing the Understanding of AMOC Dynamics and Stability through Manifold Learning

The Atlantic Meridional Overturning Circulation (AMOC) is a vital component of Earth’s climate system, significantly influencing global temperature and precipitation patterns. Concerns persist about a potential AMOC weakening or collapse, but models and reanalysis products reveal considerable uncertainty about its behavior. Addressing this complexity requires exploring novel approaches that deepen our understanding of this phenomenon. Advances in dynamical systems theory enable high-dimensional climate systems to be represented on reduced-dimensional manifolds, where the system’s state at any time t can be described by a state vector parameterized by N variables and their spatial variability. This study applies these methods to examine the AMOC across four ocean reanalysis products (GODAS, SODA3.3.2, ORAS5, GLORYS12V1) and two simulations from the Energy Exascale Earth System Model (E3SM) at different resolutions. By employing linear and non-linear dimensionality reduction techniques, this research constructs the AMOC attractor and investigates how system dynamics vary across resolutions. Understanding these dynamics is crucial for refining climate model projections, providing insights into AMOC stability under changing conditions, and highlighting key variables driving variability, thereby supporting more accurate climate assessments.