ML to Accelerate Actionable Climate Information: The Importance of Uncertainty Quantification (Invited Presentation)

To be actionable, climate information should be timely; it should be sufficiently trustworthy in its description of the region and processes of interest; and above all it should be truthful in acknowledging uncertainties. In this presentation I will explore the transformative potential of ML in these three areas by describing a series of case studies from the field of climate modelling and climate prediction. Throughout, the emphasis will be on uncertainty quantification: why it is important, how to achieve reliable ML models, and what we can learn from these models about the predictability of the Earth-system.