Articles | Volume 16, issue 2
https://doi.org/10.5194/esd-16-475-2025
https://doi.org/10.5194/esd-16-475-2025
Perspective
 | Highlight paper
 | 
28 Mar 2025
Perspective | Highlight paper |  | 28 Mar 2025

Potential for equation discovery with AI in the climate sciences

Chris Huntingford, Andrew J. Nicoll, Cornelia Klein, and Jawairia A. Ahmad

Data sets

MSI Level-2A BOA Reflectance Product, Collection 1 Copernicus Sentinel-2 (processed by ESA) https://doi.org/10.5270/S2_-znk9xsj

Video abstract

Video highlighting paper "Potential for Equation Discovery with AI in the Climate Sciences" Chris Huntingford et al. https://vimeo.com/1049706218/20d0d72b40?share=copy

Download
Chief editor
Artificial Intelligence (AI) is often criticised for being a "black-box" approach that provides no physical insights into the data being analysed. Very recently, a new branch of AI has emerged, called “AI-led equation discovery”. As the name suggests, it aims to reveal process equations underlying the data. This Perspective Article offers a path to align AI methods with climate research, with a focus on the use of AI-led equation discovery in support of Earth System Models.
Short summary
AI is impacting science, providing key data insights, but most algorithms are statistical requiring cautious "out-of-sample" extrapolation. Yet climate research concerns predicting future climatic states. We consider a new method of AI-led equation discovery. Equations offer process interpretation and more robust predictions. We recommend this method for climate analysis, suggesting illustrative application to atmospheric convection, land–atmosphere CO2 flux, and global ocean circulation models.
Share
Altmetrics
Final-revised paper
Preprint