Causality 101 for Geoscientists
March 30, 2018
Imme Ebert-Uphoff Visiting ATS from CSU Department of Electrical and Computer Engineering
Hosted by Libby Barnes
This talk provides a basic introduction to the principles of causal discovery, i.e. the process of detecting potential cause-effect relationships from data. The primary goal is to teach the audience how to think about causality conceptually, and to provide an intuitive understanding of some of the opportunities and limitations of today's causal discovery methods. We introduce basic concepts, such as graphs as universal language for causality, intervention vs. observation analysis, and Granger vs. Pearl causality. We cover one method for identifying potential cause-effect relationships from observed data in detail, namely constraint-based structure learning based on conditional independence tests. Finally, we illustrate the use of this method for several climate applications. Time permitting, we will highlight common steps to successful collaboration between a geoscientist and a causality scientist, and the important roles that both scientists play in such a collaborative effort.