The vertical tilt of mesoscale eddies is a pervasive oceanic phenomenon. Here, we employ a Transformer-based deep learning model, EVTNet, to reconstruct this three-dimensional structure using global satellite and Argo data (2004–2021). EVTNet accurately reconstructs eddy vertical tilt characteristics, demonstrating robust agreement with independent in situ observations. We identify a depth-dependent divergence in tilt direction, which is opposite (~74%) between the upper (10–300 dbar) and deep (300–1000 dbar) oceans. This stratified behavior likely stems from wind forcing in the upper layer versus free baroclinic Rossby wave propagation below. The upper-ocean tilting distance is, on average, 2.5 times that observed in the deep ocean. Temporal evolution of eddy tilting distance follows three distinct patterns (decreasing, increasing then decreasing, and increasing), potentially modulated by eddy-eddy interactions. Furthermore, simulations suggest that eddy divergent tilting may facilitate submesoscale motion development. These findings offer new insights into the complex dynamics of mesoscale eddies.

暂无