ARGO Float Data Analysis Using a Generalized Spectral Clustering Method

Spectral clustering method is a powerful tool for identifying Lagrangian coherent clusters from lagrangian trajectories. These coherent clusters group trajectories that are most similar to each other within the same cluster while being most dissimilar from trajectories in other clusters. Traditional spectral clustering defines similarity based on the physical distance between particles. Here, we generalize the spectral clustering technique to incorporate other physically significant properties, such as water density, temperature, or salinity  into the similarity definition between trajectories.

We apply the generalized spectral clustering method to the global ARGO float dataset and compare the resulting coherent clusters to those identified using other dynamical systems techniques for Lagrangian coherent structures identification, including FTLEs, LAVDs, and encounter volume. Different methods may identify different clusters because they use different definitions of coherence, making them most effective when used together. Also, we investigate the temporal evolution of these clusters and assess their consistency over time to understand changes or stability in water masses and ocean properties over the past decade.