Machine Learning and Artificial Intelligence

The increasing availability of high-resolution observational and model data has driven the integration of Machine Learning (ML) and Artificial Intelligence (AI) into physical oceanography. These techniques allow for automated pattern recognition, predictive modeling, and improved data assimilation, helping researchers analyze complex ocean dynamics more efficiently.

ML and AI methods have been applied to tasks such as eddy detection, ocean front identification, anomaly detection, and data gap filling, complementing traditional physics-based approaches. Recent advances have led to two major trends:

• Deep Learning. Some examples are the following (further reading can be found here):
  • EddyNet – A CNN-based tool for automatic eddy detection.
  • DeepOcean: A General Deep Learning Framework for Spatio-Temporal Ocean Sensing Data Prediction
  • SST-GAN – Uses GANs to enhance the resolution of sea surface temperature data.
• Hybrid Models. Hybrid models that integrate ML/AI with traditional physical models are emerging as powerful tools in oceanographic research. These models combine the rigor of physical laws with the predictive power of ML algorithms, improving the accuracy and efficiency of simulations. A notable example is the approach that combines ML with Atmospheric General Circulation Models (AGCMs), allowing for the capture of dynamic processes that traditional models struggle to represent accurately, as presented here. Additionally, hybrid models incorporating physical constraints into neural networks, known as Physics-Informed Neural Networks (PINNs), have been developed to enhance prediction accuracy in fluid dynamics, as compiled here.