Untangling the web: Using machine learning to understand climate critical dynamics in the Southern Ocean

The Southern Ocean surrounding Antarctica is the principle region where the deep ocean, cryosphere and atmosphere may freely exchange properties with one another. This is the major pathway for heat, carbon and nutrients into the ocean interior and has a disproportionately large impact on global climate. However, such exchanges of active tracers such as heat, freshwater and CO2 within a complex dynamical system presents considerable potential for difficult-to-predict coupled feedbacks that may profoundly influence both regional and global climates. For example the vertical upwelling of warm water from the deep ocean may be increased due to anthropogenically driven wind changes. This brings more warm water to the surface, melting sea ice and adding freshwater to the system. In turn the freshwater drives stronger stratification of surface waters and may reduce subsequent downwelling of heat and carbon enriched waters, reducing the ocean carbon/heat sink and feeding back atmospheric CO2, warming and wind increases. Presently IPCC (Intergovernmental Panel on Climate Change) class climate models do not produce coherent future projections for the Southern Ocean, largely due to differences in how such dynamical systems are modelled. This represents a major source of uncertainty for global predictions of surface warming and sea level rise and needs to be addressed to improve regional and global climate forecasting.

This project will utilise emerging data analysis techniques and algorithms to examine the IPCC suite of state-of-the-art climate models and identify and characterise the key dynamical relationships between southern ocean-atmosphere-ice variables that set the wide range of future polar climate projections currently introducing uncertainty into future climate projections. The field of 'climate informatics' is an exciting and newly emerging one (Monteleoni et al. 2012), with great untapped potential for applying recent advances in machine learning and data science to the large and multi-dimensional datasets that climate data and models represent. This work will apply such techniques to the Southern Ocean climate system in coupled climate models to discover the key parameters governing the response of the region to both natural and anthropogenic forcing, and in doing so work to understand the dynamics of the real system and reduce the present uncertainty in climate projections.