Diagnosing coastal processes using machine learning and ocean buoyancy gliders

Abstract

Ocean buoyancy gliders provide a comprehensive view of the water column, offering more than simply a snapshot of a single moment in time or space. In this study, we applied the established machine learning method, k‐means clustering, to a glider dataset collected in the summer of 2015 in the northern Gulf of Mexico. Clustering analysis of chromophoric dissolved organic matter and salinity revealed the physical structure of water masses, both vertically within the water column and horizontally along the shelf. Supplementary statistical analyses, including principal component analysis and ANOVA, of individual clusters confirmed the clusters were statistically distinct from one another and provided insights into the factors contributing to their differentiation. The clusters identified in the glider dataset represent water masses variously distinguished by river plumes, wind‐induced upwelling effects, shifts in currents, density‐induced stratification, and biological processes. This study demonstrates that applying machine learning clustering methods to subsurface glider data is a novel technique that enhances the analytical capabilities of both glider and other oceanographic datasets.