AOPP Seminar: What Can Machine Learning Teach Us About Cloud Feedbacks and Deadly Heat?

From cloud processes to the impacts of heatwaves on human health, machine learning offers powerful tools for understanding and modelling complex environmental problems. However, these methods are often criticised for their black-box nature. In this seminar, I will discuss two different applications of machine learning that focus on interpretability, drawing on two research projects.
The first focuses on high clouds, a significant source of uncertainty in climate model projections. Using machine learning to perform cloud-controlling factor analysis, we find evidence that state-of-the-art climate models do not capture an observed physical mechanism (stability iris) affecting high clouds. The second application examines the relationship between heat and human mortality. Using explainable machine learning techniques, we investigate the factors shaping heat-related mortality risk and how this risk varies across locations and temperatures.