10th International Conference on Climate Informatics
Karen McKinnon
Assistant Professor of Statistics and Environment and Sustainability
University of California, Los Angeles
Bio
Karen McKinnon is an Assistant Professor at the University of California, Los Angeles, joint between the Department of Statistics and the Institute of the Environment and Sustainability. Previously, she was an Applied Scientist at Descartes Labs in Santa Fe, New Mexico, and an Advanced Study Program post-doctoral fellow at the National Center for Atmospheric Research working with Clara Deser. She completed her PhD, advised by Peter Huybers, in 2015 in the Earth and Planetary Sciences department at Harvard University. Her research focuses on developing physical and statistical models for climate variability and change, with a particular interest in the weather-climate continuum.
Abstract
Numerical modeling has allowed for major advances in climate science since the 1950s, when the first general circulation model was published, including the ability to make testable predictions such as the rate of global mean temperature increase due to human influence, and the latitudinal structure of warming. Today, however, we are beginning to run up against the limits of numerical modeling as we increasingly want to create actionable knowledge at human scale. In this talk, I will first present work using statistical methods to make predictions at seven-week lead times for high-impact summer heat waves in the Eastern US, and discuss how artificial intelligence methods have the potential to automate these types of predictions. I will then discuss the challenges of predictability and attribution amidst the substantial internal variability in the climate system, which has been incompletely observed and is challenging to simulate with numerical models. I will conclude with some thoughts on promising future directions at the intersection of climate science and machine learning -- and the challenges we must overcome to pursue those directions.