Traditional weather and climate models are very effective at producing predictions of the ocean and atmosphere, but require vast amounts of compute power and time to produce their forecasts. We look to develop alternative data-driven models, which learn the dynamics of these systems through machine learning techniques in order to provide forecasts in a faster and more computationally efficient manner. This would enable researchers to run more experiments, giving us better appreciation of uncertainty in our predictions and improving our understanding of the earth system and human impacts on it.

In particular we focus on the ocean, firstly developing data driven models of ocean temperature change, and assessing how well these are able to capture the underlying dynamics of the systems. Later work looks at modelling an ocean channel with jets, and capturing the evolving dynamics of this system.