Capturing Mesoscale and Submesoscale Ocean Fronts in the Gulf of Mexico With Gaussian Mixture Modeling and Satellite Observations

Abstract This study advances the detection of ocean fronts in the Gulf of Mexico (GoM) by comparing traditional front detection algorithms with a machine learning model. Specifically, we evaluate the performance of three widely used methods, Canny Edge Detection (referred to as Canny), the Cayula‐Cornillon Algorithm (CCA) (referred to as CCA), and the Belkin‐O’Reilly Algorithm (referred to as BOA), against a machine learning–based Gaussian Mixture Model (referred to as GMM). These methods are applied to a suite of satellite‐derived oceanographic variables, including Sea Surface Temperature (SST), Sea Surface Salinity (SSS), Chlorophyll‐a (Chl‐a), and Absolute Dynamic Topography (ADT). To distinguish mesoscale from submesoscale features, we incorporate high‐resolution data sets from the Surface Water and Ocean Topography (SWOT) mission and the NASA's Plankton, Aerosol, Cloud, ocean Ecosystem (PACE) mission. The algorithms are tested through case studies capturing a range of oceanographic conditions, including seasonal variations, flood and drought events, and different phases of the Loop Current (LC). The GMM approach offers a statistical, machine‐learning based alternative to traditional gradient‐ and histogram‐based methods. By benchmarking these techniques, this study provides a comprehensive assessment of their strengths and limitations in identifying dynamic frontal structures. These features are key to understanding vertical and lateral exchanges of heat, nutrients, and carbon, and play a critical role in regional marine ecosystems and climate processes.