UExP-FNN-U full surface ocean carbonate system

Monthly global surface-ocean carbonate fields and uncertainties

Source
Zenodo
Published
2026
Region
Global
Format
multiple formats
Access
Open access
License
cc-by-4.0
Themes
Oceans & Sea Level, Climate & Temperature, Greenhouse Gases, Geospatial & GIS Data

Get the data from Zenodo

About this dataset

Product Information

Product name

UExP-FNN-U

SOCOM-style name

UExP-FNN-U

Product version

v2026-1

Changelog at end of repository

Coverage

January 1980 – December 2025

Global ocean (including under ice regions) at ~0.2 m depth

Resolution

Monthly 1° x 1°

Contact

Daniel J. Ford

Jamie D. Shutler

Traceable code and inputs

Information on the input datasets and code can be found after the changelog

How to reference these data?

If these data are used please reference Ford et al. (2024a) and this repository with the version (the version is key to tracability) .

Note: If you use these data for a new analysis or publication, please contact Daniel J. Ford ( ). This will allow the opportunity to contribute with detailed information on the dataset and depending on the level of involvement be considered for co-authorship.

Product Description

The UExP-FNN-U approach is described in detail within Ford et al. (2024a) and therefore we provide a summary of the algorithm for interpolating the fugacity of CO 2 in seawater (fCO 2 (sw) ). The UExP-FNN-U is a two step neural network interpolation technique, the self-organising map feed forward neural network (SOM-FNN) (Landschützer et al., 2014, 2016). The first step is a self-organising map (SOM) which was used to divide the global oceans into regions, or provinces, of similar oceanic conditions. The inputs to this step were monthly climatology of sea surface temperature (SST) from the European Space Agency Climate Change Initiative (ESA-CCI) SST, merged sea surface salinity (SSS) from the CCI and the CMEMS reanalysis (GLORYS12V1; merged using a hierarchy approach described in Gregor et al., 2024), CMEMS GLORYS12V1 mixed layer dapth (MLD), and spatailly and temporally complete OC-CCI chlorophyll-a (chl-a) (Ford et al. 2026a; Sathyendranath et al. 2019) and the Takahashi et al. (2009) fCO 2 (sw) climatology updated in Fay et al. (2024). The SOM produces 16 provinces, and two manual provinces are implemented to cover the Arctic Ocean and Mediterranean + Red Sea using Longhurst biogeochemical provinces (Longhurst, 1998). The second step uses a feed forward neural network (FNN) ensemble (10 members) for each province to estimate the relationships between the target variable (i.e in situ fCO 2 (sw) from the recalculated SOCAT dataset; Bakker et al., 2016; Ford et al., 2025; Ford et al. 2026c) and oceanic properties that likely control their variability. For the UExP-FNN-U v2026-pre1 these were SST, SSS, MLD, chl-a and xCO 2 (atm) and anomalies of each.

In v2026-pre1, we implemented a probability based SOM approach, which provides a probability that each monthly 1 degree pixel is assigned to each of the 18 provinces. To achieve the probabilities, the SOM training occurs as in Ford et al. (2024a) and then the inputs are perturbed within the limits of their uncertainties and the resulting SOM provinces retrieved for 400 ensembles.

How the data was collected

Product Information

Product name

UExP-FNN-U

SOCOM-style name

UExP-FNN-U

Product version

v2026-1

Changelog at end of repository

Coverage

January 1980 – December 2025

Global ocean (including under ice regions) at ~0.2 m depth

Resolution

Monthly 1° x 1°

Contact

Daniel J. Ford

Jamie D. Shutler

Traceable code and inputs

Information on the input datasets and code can be found after the changelog

How to reference these data?

If these data are used please reference Ford et al. (2024a) and this repository with the version (the version is key to tracability) .

Note: If you use these data for a new analysis or publication, please contact Daniel J. Ford ( ). This will allow the opportunity to contribute with detailed information on the dataset and depending on the level of involvement be considered for co-authorship.

Product Description

The UExP-FNN-U approach is described in detail within Ford et al. (2024a) and therefore we provide a summary of the algorithm for interpolating the fugacity of CO 2 in seawater (fCO 2 (sw) ). The UExP-FNN-U is a two step neural network interpolation technique, the self-organising map feed forward neural network (SOM-FNN) (Landschützer et al., 2014, 2016). The first step is a self-organising map (SOM) which was used to divide the global oceans into regions, or provinces, of similar oceanic conditions. The inputs to this step were monthly climatology of sea surface temperature (SST) from the European Space Agency Climate Change Initiative (ESA-CCI) SST, merged sea surface…

In v20…

Browse sustainability datasets on SusanHub