Machine learning is becoming more and more accessible to the scientific community, with high performance computing capabilities, data collection, and increasing availability of free and highly efficient software packages. Part 1 of this talk discusses the great potential as well as some challenges of using machine learning for climate and weather applications. Challenges include the perceived lack of transparency and the potential for incorrect generalization of these methods. We then discuss strategies for overcoming these challenges, including i) leveraging physics in the AI approach and ii) utilizing visualization tools to help understand the reasoning of these algorithms. Part 2 then discusses machine learning projects that are currently ongoing within NOAA ESRL’s Global System Division (GSD). GSD has several active projects applying different methods of ML to satellite data that will be covered briefly in this talk. One project in particular, a Regions of Interest (ROI) project that uses deep learning to detect cyclonic ROI from water vapor satellite data, will be highlighted at the end.
About the Speakers: Imme Ebert-Uphoff received B.S. and M.S. degrees in Mathematics from the Technical University of Karlsruhe (known today as Karlsruhe Institute of Technology or KIT). She received M.S and Ph.D. degrees in Mechanical Engineering from the Johns Hopkins University. She was a faculty member in Mechanical Engineering at Georgia Tech for over 10 years, before joining the Electrical & Computer Engineering department at Colorado State in 2011 as research professor. Her research interests are in applying data science methods to climate applications. She is also very involved in activities to build bridges between the AI community and the earth science community, including serving on the steering committee of the annual Climate Informatics workshop, and of the NSF sponsored research coordination network (RCN) on Intelligent Systems for the Geosciences. Starting July 1, 2019, she is spending 50% of her time with CIRA to support their machine learning activities.
Christina Kumler comes from an applied mathematics, meteorology, and oceanic science background. She completed her B.S. degree at CU Boulder in applied mathematics in 2013 and then completed her M.S. at University of Miami Florida RSMAS in meteorology and physical oceanography in 2015. She is currently a CIRES scientist and specializes in computational aspects of weather modeling. Over the last two years, her time has been dedicated to applying machine learning techniques to big data problems in the field of weather and climate. In her spare time, she races triathlons, hikes, does semi-professional photography, and loves to cook/bake/eat with friends, family, husband, and dog.