What can machine learning tell us about Southern Ocean heat content?

Institution: British Antarctic Survey

Lead Supervisor: Dr Dan Jones

Project Description

Since the 1970s, the ocean has absorbed more than 90% of the total extra thermal energy added to the climate system via anthropogenic greenhouse gas emissions. The Southern Ocean has been an especially important region of heat uptake and storage, having absorbed more than 75% of the excess heat. In this project, we will apply a well-tested machine learning method (i.e. unsupervised classification using a Gaussian Mixture Model, or GMM) to autonomous float data to better understand the three-dimensional structure of Southern Ocean heat content.

Links to activity:
The student will work as part of the BAS Polar Oceans team. They will be encouraged to attend science seminars and group/project meetings. The student will have opportunities to interact with scientists involved in oceanographic field work, high-level numerical analysis, and ocean/climate modelling. They will work as part of the ORCHESTRA project, a 5-year, cross-centre NERC project focused on Southern Ocean heat storage and transports (www.bas.ac.uk/project/orchestra/). The student will be supervised by Dan Jones (lead), Andrew Meijers, and Emily Shuckburgh. BAS will provide a computer, desk space, and a software license if needed. The student will also have access to the BAS high performance computing platform (scihub).

Other elements:
The student will perform a literature review to put the project into a broader scientific context. They will also meet with other scientists, including some heavily involved with fieldwork. The student will be encouraged to interact with the rapidly expanding community of machine learning experts at both BAS and the University of Cambridge.

Enhanced skills:
The student will gain competence with an important machine learning technique, as well as familiarity with handling, visualising, and analysing large oceanographic datasets. They will also learn more about oceanography as an active area of scientific research.

Indicative timescale for project (subject to change)

• Week 1: Introduction/welcome to BAS, start literature review (supervisors will provide a list
of papers to start with)
• Week 2: Finish literature review. Learn how to read/manipulate/visualise Argo float data
• Week 3: Learn about unsupervised classification methods, try simple examples
• Week 4: Learn about Gaussian Mixture Models (GMM), try simple examples
• Week 5-8: Write and implement code to apply unsupervised classification to Argo data
• Week 9: Work on plotting, analysing, and summarising the results
• Week 10: Write up summary, give short presentation to small group at BAS

The student will be encouraged to meet with other scientists and machine learning experts at BAS and the University of Cambridge throughout the 10-week project.

Powered by WordPress. Designed by WooThemes