Application of a generalized spectral clustering method for characterizing water masses using Argo floats

We apply  generalized spectral clustering methods to the global Argo dataset and compare the identified clusters with those obtained from established dynamical systems approaches, including finite-time Lyapunov exponents (FTLEs), Lagrangian-averaged vorticity deviation (LAVD), encounter volume, and a newly introduced tool— retention volume.

Spectral clustering provides a powerful framework for identifying Lagrangian coherent clusters from particle trajectories, grouping together trajectories that evolve similarly while remaining distinct from others. Traditionally, spectral clustering relies on physical proximity to define similarity between particles. Here, we extend this approach by incorporating additional oceanographic properties—such as temperature, salinity, density, and spiciness—into the similarity measure. This generalization allows us to detect coherent water masses that are not only spatially coherent but also share key physical characteristics.

Our results highlight the potential of the generalized spectral clustering method, combined with Argo measurements, to provide new insights into ocean transport and water mass transformations.