Seasonal variability of the Pearl River Plume front based on deep learning

Abstract

Ocean fronts are characterized by active vertical shear and mixing, which play a role in primary production, fishery resources, and marine ecosystems. While oceanographers have focused on the dynamics and biogeochemical processes of the Pearl River Plume, the seasonal variations of the plume front remain unclear. In this study, we investigate the seasonal and interannual variations in front intensity within and around the Pearl River Estuary by extracting remote sensing turbidity gradient images using the BOA algorithm. We then apply a deep learning technique to these turbidity gradient images to detect river plume fronts. The deep learning network successfully identifies two types of Pearl River Plume fronts: one with clear and distinct features, and the other displaying diffuse and fragmented characteristics. We further map the seasonal dispersal patterns of the Pearl River Plume front probability, revealing westward and eastward distributions of plume fronts in summer, and a more frequent westward distribution in autumn and winter. A front hotspot is observed near the western coast off the Pearl River Estuary, where the strength of front probability is influenced by the magnitude of the northeasterly wind.