skip to content

Cambridge NERC Doctoral Training Partnerships

Graduate Research Opportunities
 
Map of temperature profile clusters in the Southern Ocean
Brief summary: 
We will use cutting-edge interpretable machine learning classification techniques to identify and track coherent dynamical and biogeochemical regimes in climate model data (e.g. past climates, possible future climates), and we will relate the behaviour of those regimes to specific mechanisms.
Importance of the area of research concerned: 
In order to make more confident projections of Earth’s future climate, we need a robust set of methods for objectively describing the complex physical and biogeochemical structures that make up the climate system. For example, the definition of the ocean pycnocline, which roughly divides the ocean into upper and lower layers with different air-sea exchange properties, is still somewhat ad-hoc, making comparison across different models challenging. In recent years, researchers have identified unsupervised machine learning as an area with considerable promise for addressing such issues [1,2,3]. Classification methods offer new objective approaches for defining natural regimes and tracking them over time (e.g. the geography of the surface carbon budget). The regimes can be related to climate processes by examining their evolution in different emissions scenarios and in tailored numerical experiments. In addition, many machine learning methods scale up well with dataset size, which is an advantage when handling the tens of Petabytes worth of data produced by modern coupled climate models. Advances in this area will contribute to a more robust description of Earth’s climate system.
Project summary : 
The climate system features rich physical and biogeochemical structures that interact across a wide range of spatial and temporal scales via numerous coupled, nonlinear processes. A key goal of modern climate science is to confidently project the manner in which these structures (e.g. the ocean mixed layer) and processes (e.g. air-sea gas exchange) may evolve under different emissions scenarios, but the highly correlated nature of the climate system makes this challenging. In this project, we will deploy interpretable machine learning techniques to identify and track coherent domains in large, complex Earth system datasets, which is emerging as an active research area [1,2,3]. In particular, we will evaluate different classification techniques based on their suitability for defining ocean water mass structures and for finding boundaries between physical regimes (i.e. ocean fronts).
What will the student do?: 
In this project, the student will explore a cutting edge research frontier: the application of interpretable machine learning techniques to climate model data. The student will evaluate different probabilistic classification techniques in terms of their suitability for objectively identifying robust coherent physical and biogeochemical regimes in climate model data. They will write software to use these methods with climate model data, possibly using big data community geoscience platforms such as pangeo.io. A particular focus will be on producing well documented, readable source code and sharing it in an open source manner in order to improve reproducibility and to encourage the wide use of the objective classification methods developed during the PhD. The student will work as part of the BAS Polar Oceans Team and the BAS Artificial Intelligence Lab (https://www.bas.ac.uk/ai/). They will present their work at national and international scientific meetings in both the geoscience and climate informatics communities. The machine learning and data science skills gained during this project will be relevant for careers in both academia and the private sector.
References - references should provide further reading about the project: 
Sonnewald, M., Dutkiewicz, S., Hill, C., & Forget, G. 2020. Elucidating ecological complexity: Unsupervised learning determines global marine eco-provinces. Science Advances, vol. 6, no. 22, DOI:10.1126/sciadv.aay4740
Sonnewald, M., Wunsch, C., & Heimbach, P. 2019. Unsupervised learning reveals geography of global ocean dynamical regions. Earth and Space Science, vol. 6, pp. 784– 794. DOI:10.1029/2018EA000519
Jones, D.C., Holt, H., Meijers, A., & Shuckburgh, E. 2019. Unsupervised clustering of Southern Ocean Argo float temperature profiles. Journal of Geophysical Research: Oceans, vol. 124, pp. 390– 402. DOI:10.1029/2018JC014629
Applying
You can find out about applying for this project on the British Antarctic Survey (BAS) page.
Dr Francesco Muschitiello
Dr Dan Jones
Andrew Meijers
British Antarctic Survey Graduate Administrator
Professor Peter Haynes