5/14/24: Paul Roebber, Some applications of machine learning in the physical sciences

About the presentation: Interest in machine learning and artificial intelligence is widespread and growing, both in society in general and in the physical sciences. As an example, I am presently engaged in four projects with NOAA in these areas: adaptive numerical weather prediction post- processing (evolutionary programming using neural networks as the “agents”); adaptive post-processing of a deep learning “BLEND” model in support of fire weather forecasting; heavy rainfall nowcasting in the southeastern US (CNN and standard neural networks); verification studies of data-driven ensemble weather prediction models (GraphCast). Some other projects include snow-to-liquid water ratios, coupled Great Lakes lake level modeling, convective forecasting in the US and Taiwan, and reverse engineering of initial conditions in support of ensemble model generation. In this talk, in addition to this overview from my work, I will discuss proposed future directions in machine learning in NOAA, based on a recent survey conducted with support by NOAA’s Office of Science and Technology Integration.

About the speaker: Dr. Paul Roebber is a Distinguished Professor of Atmospheric Sciences at the University of Wisconsin at Milwaukee’s (UWM) School of Freshwater Sciences. Dr. Roebber is an Affiliate Faculty with the Northwestern Mutual Data Science Institute in Milwaukee, where he also serves as the Program Director for the Bachelor of Science in Data Analytics and the Master of Science in Data Science. Since 2016, Roebber has been contributing to the National Weather Service’s Meteorological Development Laboratory data science efforts, which seek to improve weather forecast model information across all of North America, and he holds active grants with the National Oceanic and Atmospheric Administration and the National Science Foundation. During a 2021-22 sabbatical leave, working with NOAA’s Office of Science and Technology Integration, Roebber researched and wrote a report summarizing current machine learning efforts within the National Weather Service, with a view towards removing current obstacles to progress in this area.

Dr. Roebber holds advanced degrees in meteorology and physical oceanography from the Massachusetts Institute of Technology and McGill University. He has edited and published extensively in the scientific research literature, with 80 papers and book chapters in print, and is a co-author of a book published by MIT Press on Expert Forecasting (“Minding the Weather”). In support of this work, Roebber has won 43 grants from Federal and State agencies, as well as the private sector and the Government of Canada. He has directed 33 thesis students at the doctoral and masters level at UWM since 1994.

Dr. Roebber has extensive experience in public and private weather forecasting, synoptic and mesoscale modeling, forecast verification and data science. Paul is the recipient of multiple awards including the American Meteorological Society’s Editors Award, the MIT Club of Wisconsin Individual Tech Award, the UWM Research Foundation Senior Faculty Award, and the UWM Distinguished Undergraduate Teaching Award.