The ocean plays a critical role in absorbing and sequestering CO2 from the atmosphere yet understanding where and when this occurs is difficult due to sparse observations. Researchers often use gap-filling machine learning methods to fill in observing gaps; however these methods have been criticized for their black box approach— meaning it is unclear what processes are influencing the resulting gap-filled data product.
In a recently submitted study to Journal of Advances in Modeling Earth Systems, made publicly available prior to peer review via a pre-print, researchers at Columbia University (PI: McKinley) funded by the Climate Observations and Monitoring program employ a new machine learning method that is inclusive of physical constraints— mainly temperatures known effect on surface ocean pCO2. Lead author Val Bennington highlights that the new approach overcomes black box critiques, exhibits realistic physical processes, and is in agreement with other data-based approaches as well as the Global Carbon Budget 2021.
For more information, contact Virginia Selz.
What are pre-prints?
“Pre-prints are freely accessible communications (such as a manuscript) that contains research findings not yet published in a peer-reviewed outlet… Once the final paper is published, it can be linked to a preprint. All ESSOAr preprints receive a DOI that can be used in citations and links.”