Earth`s climate is chaotic and noisy. Finding usable signals amidst all of the noise can be challenging: be it predicting if it will rain, knowing which direction a hurricane will go, understanding the implications of melting Arctic ice, or detecting the impacts of human-induced climate warming. Here, will demonstrate how explainable artificial intelligence (XAI) techniques can sift through vast amounts of climate data and push the bounds of scientific discovery. Examples include extracting robust indicator patterns of climate change and identifying Earth system states that lead to more predictable behavior weeks to years in advance. But machine learning models are only as capable as the scientists designing them. I will further discuss how climate science requires the crafting of domain-specific XAI methods, both to gauge the trustworthiness of the XAI’s predictions and quantify uncertainty, but also to uncover predictable signals we didn’t know were there. Explainable AI can open doors to scientific understanding — supporting scientists as we ask new questions about the coupled human-Earth climate system.

Speaker: Elizabeth Barnes

Time: 3:00 p.m.

Format: Hybrid

Virtual: Zoom link provided upon registration

In-person: Columbia Innovation Hub

Free!