Detection and Analysis of Mesoscale Eddies Based on Deep Learning

Abstract

AbstractMesoscale eddies are widely existing in the global ocean, and they can retain and transport salt, heat, and nutrients across ocean basins. Mesoscale eddies have signals on sea surface height (SSH) images, sea surface temperature (SST) images. Previous studies developed automatic eddy identification methods based on SSH or SST. However, single remote sensing data cannot adequately characterize mesoscale eddies. To improve the accuracy and efficiency of eddy detection, we need to develop a new eddy identification method based on multi-source remote sensing data. It is difficult for traditional data mining methods to extract mesoscale eddies from the global, long time series, multi-source remote sensing big data rapidly and accurately. Deep learning has advantages in feature learning and sophisticated relationship modeling, bringing new ideas for detecting global mesoscale eddies based on multi-source satellite images. The paper proposes a deep learning-based model for eddy detection from the synergy of satellite-sensed global SSH and SST big data obtained during the 1993–2015 period. Compared with the previous eddy detection methods, the newly proposed method improves the accuracy and efficiency of eddy detection. Besides, spatio-temporal characteristics of mesoscale eddies are studied based on the model.