Will Gregory
|
PhD Title
|
Machine learning tools for pattern recognition in polar climate science |
|
Research Theme
|
Earth, Atmosphere and Ocean Processes |
|
Primary Supervisor
|
 |
|
Primary Institution
|
|
|
Secondary Supervisor
|
 |
|
Secondary Institution
|
|
|
Abstract
|
Arctic sea ice is a thin blanket of ice (typically 2 – 3 metres thick) that covers an area roughly the size of Europe. The ice responds rapidly to changes in oceanic and atmospheric forcing, shrinking every summer to less than half its winter maximum extent. Because of these characteristics, sea ice can be seen as a “barometer” for measuring changes that occur in the atmosphere and the ocean, as well as in various other components of the Arctic climate system. Pattern Recognition is a field dedicated to the development of generalised models which are able to identify regularities within a given data set. In this thesis, two novel techniques will be explored for exploiting such regularities within climate time series data; techniques which encompass aspects of both supervised and unsupervised machine learning and have been used extensively in a number of fields outside of climate science. These methods include the Bayesian inference approach of Gaussian Process Regression, and also Complex Networks. Through the application of these techniques, both independently and in conjunction, the aim is to improve our understanding of the polar climate system by exploiting patterns of sea ice variability and evaluating how they manifest within observational and model products, through seasonal sea ice forecasting, merging of satellite altimetry products, and an assessment of sea ice varibility with the next generation of coupled climate models from CMIP6. |
|
Policy Impact
|
|
|
Background Reading
|
|
|
Publications
|
|
