A Deep-learning-based Frontal Detection in the China Sea

The oceanic front is a relatively stable structure formed by the rapid changes of marine dynamic environment elements in the water body, which affects the material exchange process on both sides of the gradient. Accurate detection of the position and intensity of the front is of great significance for climate change and fishery production. This study proposes a deep-learning semantic segmentation model for frontal detection, especially since the multi-task module enables the model to detect the position and intensity of the front simultaneously. Besides, some improvement was considered in the deep-learning model, i.e., the convolutional block attention mechanism was integrated into the encoder and decoder of the model to capture the frontal feature at the channel and spatial level to achieve high-precision detection. We collect daily grid data of sea surface temperature (SST) and chlorophyll concentration (CHL) from satellite observations and build a frontal dataset to train and test the model. Compared with other classic semantic segmentation methods, the model can extract the position more precisely. Finally, we use the model to detect the frontal intensity and occurrence frequency in the China Sea (Yellow Sea and East China Sea). The results show certain differences in two different data sources between SST and CHL, and some possible related reasons are discussed further.