CISL Seminar Series Presents: Imme Ebert-Uphoff

02/12/2015 - 10:00am to 11:00am
Mesa Lab Main Seminar Room

ImmeEbert-Uphoff, Colorado State University

The Potential of Causal Discovery Methods in Climate Science



Causal discovery seeks to identify potential cause-effect relationships from observational data.  A rich theoretical framework for causal discovery exists and has been used for decades in the social sciences and economics, and in recent years in bioinformatics.  Most recently this framework has been introduced to the field of climate science, for example to generate new hypotheses about causal relationships between different climate variables.

The goal of this talk is to raise awareness in the climate science community for the potential of causal discovery. We introduce the basic concepts of causal discovery and study one type of algorithm in detail, namely constraint-based structure learning of Bayesian networks.  We then discuss strengths and limitations of this approach and illustrate it using some recent applications in climate science, such as generating new hypotheses between different climate variables and generating "graphs of information flow" around the globe. Graphs of information flow may allow researchers to gain new insights, for example to identify subtle changes of the planet's dynamics in a warming climate or to evaluate climate models.