AI Model Improves Sea Surface Temperature Downscaling Accuracy

A new deep learning framework, EddyFlow, has been introduced to enhance the downscaling of sea surface temperature (SST) data. Traditional deep learning models often smooth out important mesoscale features, which are crucial for understanding regional ocean dynamics, even while achieving numerical accuracy. EddyFlow addresses this by focusing on physics-informed representation learning, ensuring that the model not only predicts accurately but also maintains the spectral fidelity of these smaller-scale ocean structures. The framework was trained using data from the Gulf of St. Lawrence and then tested on unseen domains like the Bay of Fundy and the Gulf of Mexico in zero-shot and few-shot scenarios. EddyFlow significantly reduced prediction errors and achieved high skill relative to persistence models, demonstrating its ability to generalize across different geographical regions while preserving essential oceanographic details.