Robust Ocean Subgrid-Scale Parameterizations Using Fourier Neural Operators

Saved in:
Bibliographic Details
Title: Robust Ocean Subgrid-Scale Parameterizations Using Fourier Neural Operators
Authors: Mangeleer, Victor, Louppe, Gilles
Contributors: Montefiore Institute - Montefiore Institute of Electrical Engineering and Computer Science - ULiège
Source: Machine Learning and the Physical Sciences, NeurIPS 2023, New Orleans, United States [US], December 15, 2023
Publication Year: 2023
Subject Terms: Computer Science - Learning, Physics - Atmospheric and Oceanic Physics, Deep Learning, Fourier Neural Operators, Parameterizations, PyQG, Engineering, computing & technology, Electrical & electronics engineering, Ingénierie, informatique & technologie, Ingénierie électrique & électronique
Description: In climate simulations, small-scale processes shape ocean dynamics but remaincomputationally expensive to resolve directly. For this reason, theircontributions are commonly approximated using empirical parameterizations,which lead to significant errors in long-term projections. In this work, wedevelop parameterizations based on Fourier Neural Operators, showcasing theiraccuracy and generalizability in comparison to other approaches. Finally, wediscuss the potential and limitations of neural networks operating in thefrequency domain, paving the way for future investigation.
Document Type: conference poster not in proceedings
http://purl.org/coar/resource_type/c_18co
conferencePoster
peer reviewed
Language: English
Relation: https://arxiv.org/abs/2310.02691; https://arxiv.org/abs/2310.02691
Access URL: https://orbi.uliege.be/handle/2268/309230
Rights: open access
http://purl.org/coar/access_right/c_abf2
info:eu-repo/semantics/openAccess
Accession Number: edsorb.309230
Database: ORBi
Description
Abstract:In climate simulations, small-scale processes shape ocean dynamics but remaincomputationally expensive to resolve directly. For this reason, theircontributions are commonly approximated using empirical parameterizations,which lead to significant errors in long-term projections. In this work, wedevelop parameterizations based on Fourier Neural Operators, showcasing theiraccuracy and generalizability in comparison to other approaches. Finally, wediscuss the potential and limitations of neural networks operating in thefrequency domain, paving the way for future investigation.