Unsupervised Classification Applied to Earth System Data

Overview

Unsupervised Classification attempts to identify groups or structures in large datasets. Over the last several years, many of my collaborators and I have been applying unsupervised classification techniques to analyze complex Earth system datasets. By grouping data into coherent classes, this approach reveals underlying patterns in a variety of geophysical processes, from oceanographic properties to atmospheric phenomena.

The project has been applied to a broad range of Earth system data, including:

• Southern Ocean temperature profiles
• Weddell Gyre thermohaline structures
• Global atmospheric ozone profiles
• Sea level data in the Nordic Seas
• Extratropical cyclones

This work has led to new insights into the structure and dynamics of these complex systems, including the identification of previously unnoticed patterns in the data.

Key Objectives:

• Apply unsupervised classification methods (such as Gaussian mixture models and clustering algorithms) to Earth system data
• Identify new patterns and structures in datasets like ocean temperature, salinity, ozone, and sea level data
• Improve the understanding of oceanic and atmospheric processes by identifying coherent regimes and transitions
• Collaborate with scientists and institutions to expand the application of unsupervised classification in environmental sciences

My Role:

As a key collaborator and one of the pioneers of applying unsupervised classification to Earth system data, I lead the development of machine learning algorithms and their application to large, complex environmental datasets. My role involves designing and implementing the classification frameworks, interpreting the results, and contributing to the broader scientific discussions on the implications of these findings.

Impact:

This work has advanced the application of unsupervised learning to Earth system sciences, helping to uncover new patterns and improving our understanding of natural processes like ocean circulation, atmospheric composition, and extreme weather events. By applying these methods to diverse datasets, we hope to improve predictions of environmental phenomena and contribute to climate change research.

Related Publications:

• Poropat, L., Jones, D., Thomas, S. D. A., and Heuze, C. (2024). “Unsupervised classification of the northwestern European seas based on satellite altimetry data.” Ocean Science, 20, 201-215.
• Jones D.C., Sonnewald M., Zhou S., Hausmann U., Meijers A. J. S., Rosso I., Boehme L., Meredith M. P., and Naveira Garabato A. C. (2023). “Unsupervised classification identifies coherent thermohaline structures in the Weddell Gyre region.” Ocean Science, 19:857-885.
• Fahrin, F., Jones, D. C., Wu, Y., Keeble, J., and Archibald, A. T. (2023). “Technical note: Unsupervised classification of ozone profiles in UKESM1.” Atmospheric Chemistry and Physics, 23, 3609-3627.

Future Directions:

Future work will focus on expanding the application of unsupervised classification to additional datasets, including future climate models, global carbon cycles, and ecosystem dynamics. The goal is to refine the techniques and apply them to more comprehensive Earth system models to improve predictive capabilities and inform climate adaptation strategies.