An auto-tracking algorithm for mesoscale eddies using global nearest neighbor filter

Abstract

Many tracking algorithms have been developed to automatically track mesoscale ocean eddies. They are successful in most situations except when there is more than one successor candidate. This study presents a tracking approach using the global nearest neighbor filter (GNNF) to tackle this problem. The GNNF method implements the Kalman filter to model and track the process of ocean eddies, and then employs an optimization method to identify the most possible successor from the multiple candidates. The method was evaluated using an eddy dataset from the South China Sea (SCS) and its performance was compared against the distance-based search (DBS) and the overlap-based search (OBS) methods. Results show that GNNF is the most successful method to correctly identify a successor for a specific eddy with multiple potential candidates (accounts for nearly 2% of the total eddies in our dataset). We also evaluated the methods using synthetic eddy tracks and results show that the performance of all three methods is strongly affected by the number of tracks and the variations of eddy propagation velocity. The average pairing error of GNNF, DBS, and OBS are about 0.2%, 0.4%, and 0.5%, respectively, when the synthetic tracks were generated with experiment parameters best fit the survey results of ocean eddies in the SCS. The GNNF method is still the most successful algorithm in identifying the correct successor regardless of the complexity of synthetic tracks.