AI maps the North Atlantic’s mixed layer every three hours

A new dataset called Morlenn now tracks the thickness of the ocean’s uppermost, wind-stirred layer across the North Atlantic every three hours, from 2011 to 2023. The product, described in a preprint under review for the journal Earth System Science Data, was built by Lucile Gaultier of OceanDataLab and colleagues, including Clément de Boyer Montégut of the Laboratory of Physical and Spatial Oceanography in Plouzané, France.

The mixed layer is the band of water at the ocean surface that waves and winds churn into a relatively uniform mass. Its depth controls how much heat the ocean exchanges with the atmosphere, and it shapes conditions for marine life. Yet direct measurements of its thickness are sparse, especially at subdaily timescales, which is the gap Morlenn is designed to fill.

Technically, the dataset covers the North Atlantic and adjacent equatorial waters at a spatial resolution of 0.25 degrees, delivering two variables: mixed-layer depth and the maximum buoyancy frequency squared, a measure of the stratification just beneath the mixed layer. The product is generated from satellite observations of the ocean surface, atmospheric fluxes and wave forcing history, processed through a neural network trained on individual Argo float observations of mixed-layer depth and stratification.

Two versions of the dataset are provided. One uses surface inputs alone, while the second additionally incorporates neighbouring Argo data. The team evaluated the product against Argo stations over the full 2011 to 2023 period, complemented by independent glider observations, and examined errors across regions and depth ranges, along with spectral and diurnal-cycle analyses of the variability the product represents.

The headline comparison favours the new product. Scored against Argo stations over 2011 to 2023, the version using neighbouring Argo data reaches a root-mean-square error of 38 metres for mixed-layer depth. On the same stations, the existing ARMOR3D product reaches 45 metres and GLORYS reaches 56 metres. The authors caution, however, that validation of the subsurface stratification estimates is less extensive than the validation of mixed-layer depth.

The dataset also carries documented limitations. The authors note constraints for shallow mixed layers, for extreme depths and for regions with very low salinity, so users should treat those regimes with care. The dataset is available under a Creative Commons Attribution 4.0 license at https://doi.org/10.17882/120058.

The short summary accompanying the preprint notes that the maps reveal how storms deepen the layer, information relevant to weather, climate and marine ecosystems. The dataset’s creators suggest it can support investigations of upper-ocean variability.

It is worth underlining the status of this work. The paper is a preprint posted on 28 October 2026 and is currently under review for Earth System Science Data, so the reported accuracy figures have not yet completed peer review. The error statistics are also specific to the comparison against Argo stations over 2011 to 2023 in the North Atlantic and adjacent equatorial waters, not a global assessment.

For researchers, the practical appeal is temporal resolution. Subdaily variations in mixed-layer depth influence air-sea heat fluxes, and a three-hourly, quarter-degree record spanning more than a decade offers a new basis for studying those variations across the North Atlantic.

The Morlenn dataset, with its documented strengths and stated limitations, now awaits the scrutiny of the peer-review process that its authors have initiated.

Subject of Research: A three-hourly machine-learning dataset of North Atlantic mixed-layer depth and subsurface stratification built from Argo, satellite and atmospheric data.

Article Title: Morlenn: a three-hourly North Atlantic dataset of mixed-layer depth and subsurface stratification, merging ARGO, satellite observations, and atmospheric forcing

Article References: Gaultier, L., Tripathi, P., Collard, F., & de Boyer Montégut, C. (2026). Morlenn: a three-hourly North Atlantic dataset of mixed-layer depth and subsurface stratification, merging ARGO, satellite observations, and atmospheric forcing. https://doi.org/10.5194/essd-2026-807

Image Credits: AI Generated

DOI: 10.5194/essd-2026-807

Keywords: mixed-layer depth, North Atlantic, ocean stratification, Argo floats, neural network, satellite observations, air-sea interaction, ESSD preprint, ocean dataset, machine learning, glider validation, storm response