(Keynote) Explainable AI for Climate Prediction, Projection and Discovery

Climate prediction is incredibly challenging. To make progress, the field must explore a wide range of tools, including climate models and data-driven algorithms. Here, I demonstrate how explainable machine learning (or artificial intelligence; XAI) techniques can improve not only prediction skill, but also push the bounds of scientific discovery by uncovering predictable signals we did not know were there. I emphasize that these tools are not at odds with physics-based models, but instead, can be used alongside them to leverage the best of both worlds. Finally, I will speak to the “black box” nature of machine learning and demonstrate how combining domain-knowledge with XAI and careful model development can make them more transparent and understandable to the scientists using them.