An mesoscale eddy intelligent detection model with physical constraints

Abstract Marine mesoscale eddies are a common marine ocean phenomenon and they can affect the heat, salinity and water currents in the ocean. Identifying mesoscale eddies can play an important role in shipping, military, and ocean resource development. Compared to traditional recognition algorithms, artificial intelligence methods can identify mesoscale eddies more accurately and efficiently, but they lack physical interpretability due to their training process are driven by data. In this paper, an intelligent detection algorithm with physical constraints was used to realize identification of mesoscale eddies. Firstly, the mesoscale eddy dataset was built from the global ocean physics reanalysis data provided by the Copernicus Marine Environment Monitoring Service (CMEMS), and the PET method was applied to generate labels for the dataset which would be used in intelligent models. Then, the classic EddyNet model was realized and an attention mechanism was added to optimize model performance. Finally, a physical constraint was introduced to make the model more consistent with the physical characteristics of mesoscale eddies: the divergence of the vorticity field was added into the total loss function of the optimized model. By introducing 30% physical constraint weights in the model, the accuracy can be improved from 92.68% to 93.57%. The result illustrated that combining data and physical constraints significantly improves the ability of artificial intelligence methods to recognize mesoscale eddies.