Unraveling the Non-Homogeneous Dispersion Processes in Ocean and Coastal Circulations Using a Clustering Approach

Dispersion processes in environmental flows have been traditionally studied under the strong assumption of homogeneous, isotropic and stationary turbulence. To overcome this limitation, we propose a new approach that combines autocorrelation analysis of simulated Lagrangian trajectories together with unsupervised clustering. To test the approach, we consider several dynamic scenarios around a coastal gulf, subject to different forcing, in order to compare our method with other approaches. Lagrangian trajectories forced by the varying coastal circulation exhibited different behaviors, looping and non-looping paths, and produced a variety of Lagrangian autocorrelation functions. Our approach proves to be able to reveal spatio-temporal variations in ocean dispersion processes without any a priori knowledge of the character of the trajectories. Clusters based on the autocorrelation functions are associated to different inhomogeneous dispersion processes. Finally, we propose a new stochastic model capable of predicting the different forms of autocorrelations.