Vertical structure of mesoscale eddies is key to the eddy‐induced heat/material transport that further affects the climate and marine ecosystem. This study explores the vertical structure of mesoscale eddies in the Kuroshio‐Oyashio Extension region (KOE) and its underlying dynamics. By applying the hierarchical ascending classification to the observational and reanalysis data sets, we classify mesoscale eddies with three distinct kinds of vertical structures. Each kind of eddies exhibits clear spatial aggregation along a distinct zonal band. Eddies have core depths of 100–300 m in the northern part of the KOE and core depths of 300–500 m and 0–100 m in the southern. The eddy splitting or merging does not introduce new kind of eddy vertical structure but causes large intra‐kind variability. The different kinds of eddy vertical structures can be partially accounted for by the baroclinic instabilities at the eddy generation sites and local adjustment process.

Graph Convolutional Neural Network (GCN), a widely adopted method for analyzing relational data, enhances node discriminability through the aggregation of neighboring information. Usually, stacking multiple layers can improve the performance of GCN by leveraging information from high-order neighbors. However, the increase of the network depth will induce the over-smoothing problem, which can be attributed to the quality and quantity of neighbors changing: (a) neighbor quality, node's neighbors become overlapping in high order, leading to aggregated information becoming indistinguishable, (b) neighbor quantity, the exponentially growing aggregated neighbors submerges the node's initial feature by recursively aggregating operations. Current solutions mainly focus on one of the above causes and seldom consider both at once. Aiming at tackling both causes of over-smoothing in one shot, we introduce a simple Two-Sided Constraint (TSC) for GCNs, comprising two straightforward yet potent techniques: random masking and contrastive constraint. The random masking acts on the representation matrix's columns to regulate the degree of information aggregation from neighbors, thus preventing the convergence of node representations. Meanwhile, the contrastive constraint, applied to the representation matrix's rows, enhances the discriminability of the nodes. Designed as a plug-in module, TSC can be easily coupled with GCN or SGC architectures. Experimental analyses on diverse real-world graph datasets verify that our approach markedly reduces the convergence of node's representation and the performance degradation in deeper GCN.

Mesoscale eddies are known to enhance diapycnal mixing in the ocean, yet direct observation of this effect remains a significant challenge, especially in the robust Antarctic Circumpolar Current (ACC). To quantify the diapycnal mixing induced by mesoscale eddies in the standing meander of the ACC, satellite altimeter and Argo profile data were combined to composite eddies, where the 1.6 m dynamic height contour was used for the first time instead of the climatological Northern Sub-Antarctic Front (SAFN) to define the northern boundary of the ACC to eliminate the influence of frontal shift. The 3D structures of the composite anticyclonic/cyclonic eddy (CAE/CCE) were obtained. Both the CAE and CCE were similar in shape to Taylor columns, from sea surface to the neutral surface of 28.085 ${\mathrm{kgm}}^{-3}$ (1689 ± 66 dbar) for the CAE, and from sea surface to 28.01 ${\mathrm{kgm}}^{-3}$ (1491 ± 202 dbar) for the CCE. On the same neutral surface, the diffusivity ($\kappa$) inside the CCE was one to two orders of magnitude higher than that inside the CAE. Vertically, the maximum influence depth of the CCE on $\kappa$ reached 1200 dbar, while for the CAE, it reached 800 dbar, where $\kappa$ exceeded $O\left({10}^{-4}\right) {\mathrm{m}}^2{\mathrm{s}}^{-1}$, and $\kappa$ gradually decreased from these depths downwards.