A dynamics-weighted principal components analysis of dominant atmospheric drivers of ocean variability with an application to the North Atlantic Subpolar Gyre

This paper describes a framework for identifying dominant atmospheric drivers of ocean variability. The method combines statistics of atmosphere-ocean fluxes with physics from an ocean general circulation model to derive atmospheric patterns optimized to excite variability in a specified ocean quantity of interest. We first derive the method as a weighted principal components analysis and illustrate its capabilities in a toy problem. Next, we apply our analysis to the adjoint of the MITgcm and atmosphere-ocean fluxes from the ECCOv4-r4 state estimate. An unweighted principal components analysis reveals that North Atlantic heat and momentum fluxes in ECCOv4-r4 have a range of spatiotemporal patterns. By contrast, dynamics-weighted principal components analysis collapses the space of these patterns onto a small subset-principally associated with the North Atlantic Oscillation-that dominates interannual SPG HC variance. By perturbing the ECCOv4-r4 state estimate, we illustrate the pathways along which variability propagates from the atmosphere to the ocean in a nonlinear ocean model. This technique is applicable across a range of problems across Earth system components, including in the absence of a model adjoint.