Stable machine-learning parameterization of subgrid processes for climate modeling at a range of resolutions

OneNOAA Science Seminar

Note: This seminar will be presented online only.

Presenter:

Janni Yuval, MIT

Co-author: Paul O'Gorman, MIT
Sponsor: STAR Science Seminar Series

Remote Access:
https://noaa-nesdis-star.webex.com/noaa-nesdis-star/j.php?MTID=m8ac1fa6af0ac0738324b961f166a93f6
Meeting number: 199 929 9230
Password: STARSeminar

Join by phone
+1-415-527-5035 US Toll

Access code: 199 929 9230

Abstract: Global climate models represent small-scale processes such as convection using subgrid models known as parameterizations, and these parameterizations contribute substantially to uncertainty in climate projections. Machine learning of new parameterizations from high-resolution model output is a promising approach, but such parameterizations have been prone to issues of instability and climate drift, and their performance for different grid spacings has not yet been investigated. Here we use a random forest to learn a parameterization from output of a three-dimensional high-resolution idealized atmospheric model. The parameterization leads to stable simulations at coarse resolution that replicate the climate of the high-resolution simulation. Retraining for different coarse-graining factors shows the parameterization performs best at smaller horizontal grid spacings. Our results yield insights into parameterization performance across length scales, and they also demonstrate the potential for learning parameterizations from global high-resolution simulations that are now emerging.

Bio:
Janni Yuval is a post-doctoral fellow at MIT at the department of Earth, Atmospheric and Planetary Sciences. At MIT he works with Paul O'Gorman on machine learning parameterization. He is a diverse person with a wide spectrum of interests and skills. He has a BSc. in theoretical physics, an MSc in theoretical soft matter physics, and a PhD in atmospheric dynamics. Furthermore, after finishing his PhD he worked as an algorithm developer at Mobileye. Later, he worked as a data scientist at Clalit Research Institute, where he used machine learning, and causal inference methods to develop personalized medicine. Nowadays, he is excited about the possibility to use machine learning for reducing the uncertainty in climate projections. The work he will present is accepted to Nature Communications (in press).

Seminar Contact:
Stacy Bunin, stacy.bunin@noaa.gov

Subscribe to the OneNOAA Science Seminar weekly email:

Send an email to OneNOAAscienceseminars-request@list.woc.noaa.gov with the word 'subscribe' in the subject or body. Visit the OneNOAA Science Seminar Series website for more information. https://www.nodc.noaa.gov/seminars/
{Janni Yuval, MIT}