This dataset contains observation-based pCO2 data and a derived monthly climatology. The observation-based pCO2 fields were created using a 2-step neural network method extensively described and validated in Landschützer et al. 2013, 2014, 2016. The method first clusters the global ocean into biogeochemical provinces and in a second step reconstructs the non-linear relationship between CO2 driver variables and observations from the v2022 release of the Surface Ocean CO2 Atlas (SOCAT, Bakker et al. 2016). This file contains the resulting monthly pCO2 fields at 1°x1° resolution covering the global ocean for the first time including the Arctic Ocean and few marginal seas (see Landschützer et al 2020). The air-sea CO2 fluxes are computed from the air-sea CO2 partial pressure difference and a bulk gas transfer formulation following Landschützer et al. 2013, 2014, 2016. Furthermore, the monthly climatology is created from the monthly average of the period 1985-present.

This dataset contains a combined globally mapped estimate of the air-sea exchange of carbon dioxide (CO2) based on Surface Ocean CO2 Atlas Database (SOCAT) partial pressure of CO2 (pCO2) and calculated pCO2 from Southern Ocean Carbon and Climate Observations and Modeling (SOCCOM) biogeochemistry floats from 1982 to 2017. The pCO2 fields were created using a 2-step neural network technique. In a first step, the global ocean is divided into 16 biogeochemical provinces using a self-organizing map. In a second step, the non-linear relationship between variables known to drive the surface ocean carbon system and gridded observations from the SOCAT dataset (Bakker et al., 2016) starting in 1982 in various combinations with calculated pCO2 from biogeochemical ARGO floats starting in 2014 from the SOCCOM project (Johnson et al., 2017) is reconstructed using a feed-forward neural network within each province separately. The final product is then produced by projecting these driving variables, i.e., surface temperature, chlorophyll, mixed layer depth, and atmospheric CO2 onto oceanic pCO2 using these non-linear relationships. This results in monthly pCO2 fields at 1°x1° resolution covering the entire globe with the exception of the Arctic Ocean and few marginal seas. The air-sea CO2 flux is then computed using a standard bulk formula.

This dataset contains surface-ocean partial pressure of carbon dioxide (pCO2) that the ensemble mean of six two-step clustering-regression machine learning methods. The ensemble is a combination of two clustering approaches and three regression methods. For the clustering approaches, we use K-means clustering (21 clusters) and open ocean CO2 biomes as defined by Fay and McKinley (2014). Three machine learning regression methods are applied to each of these two clustering methods. These machine learning methods are feed-forward neural-network (FFN), support vector regression (SVR) and gradient boosted machine using decision trees (GBM). The final estimate of surface ocean pCO2 is the average of the six machine learning estimates resulting in a monthly by 1° ⨉ 1° resolution product that extends from the start of 1982 to the end of 2016. Sea-air fluxes (FCO2) calculated from pCO2 are also presented in the data. The discrete boundaries of the clustering approach result in semi-discrete discontinuities in pCO2 and fCO2 estimates. These are smoothed by applying a 3 ⨉ 3 ⨉ 3 convolution (moving average) to the dataset in time, latitude and longitude.

This dataset contains continuous monthly maps of sea surface partial pressure of CO2 (pCO2) in the coastal ocean from 1982 to 2020. This product is an updated version of the coastal product of Laruelle et al. (2017) and has been created using a 2-step Self Organizing Maps (SOM) and Feed Forward Network (FFN) method and uses ~ 18 million direct observations from the latest release of the Surface Ocean CO2 database (SOCATv2022, Bakker et al., 2014, 2022). In a first step, the global coastal ocean is divided into 10 biogeochemical provinces using SOM, which group regions with similar environmental properties. Then, for each province, the FFN algorithm reconstructs nonlinear relationships between a set of environmental variables (e.g., sea surface temperature, salinity...) and the observed pCO2. These relationships are then used to perform the spatiotemporal pCO2 extrapolation in regions and time periods where data are lacking. The output consists of continuous monthly pCO2 maps for the coastal ocean, with a spatial resolution of 0.25°, covering the 1982-2020 period. Additionally, this new coastal pCO2 product is used to generate a new coastal air-sea CO2 exchange (FCO2) product for each grid cell at the monthly time scale from 1982 to 2020 using the following equation: FCO2=k∙K0∙∆pCO2∙(1-ice) where FCO2 represents the coastal air-sea CO2 exchange (in mol C m-2 yr-1). By convention a positive FCO2 value corresponds to a CO2 source for the atmosphere. ∆pCO2 represents the difference between the oceanic pCO2 and the atmospheric pCO2 (in atm). K0 (mol C m-3 atm-1) represents the CO2 solubility in sea water which is a function of SST and SSS following the equation of Weiss et al. (1974). k represents the gas exchange transfer velocity (m yr-1) which is a function of the second moment of the wind speed and is calculated using the equation of Ho et al. (2011) and the Schmidt number based on the equation of Wanninkhof et al. (2014). The sea-ice coverage is represented by the term ice and has no units.

This dataset contains the partial pressure of carbon dioxide (pCO2) climatology that was created by merging 2 published and publicly available pCO2 datasets covering the open ocean (Landschützer et. al 2016) and the coastal ocean (Laruelle et. al 2017). Both fields were initially created using a 2-step neural network technique. In a first step, the global ocean is divided into 16 biogeochemical provinces using a self-organizing map. In a second step, the non-linear relationship between variables known to drive the surface ocean carbon system and gridded observations from the SOCAT open and coastal ocean datasets (Bakker et. al 2016) is reconstructed using a feed-forward neural network within each province separately. The final product is then produced by projecting driving variables, e.g., surface temperature, chlorophyll, mixed layer depth, and atmospheric CO2 onto oceanic pCO2 using these non-linear relationships (see Landschützer et. al 2016 and Laruelle et. al 2017 for more detail). This results in monthly open ocean pCO2 fields at 1°x1° resolution and coastal ocean pCO2 fields at 0.25°x0.25° resolution. To merge the products, we divided each 1°x1° open ocean bin into 16 equal 0.25°x0.25° bins without any interpolation. The common overlap area of the products has been merged by scaling the respective products by their mismatch compared to observations from the SOCAT datasets (see Landschützer et. al 2020).

This dataset contains global surface water partial pressure of CO2 data, pCO2w data, and sea-air CO2 fluxes based on the extremely randomized trees or extra trees, ET machine learning approach. The products are labeled as AOML_ET. The outputs are monthly pCO2 and sea-air CO2 flux fields at 1°x1° resolution covering the global ocean from 1998 to 2023. The results of several different permutations of AOML-ET are provided. They all use the following predictor variables at monthly and at 1°x1° resolution. The variables are time, location, sea surface temperature, SST, sea surface salinity, SSS, mixed layer depth, MLD, and chlorophyll-a, chl-a. The training is performed on the v2020 and v2023 releases of the Surface Ocean CO2 Atlas (www.SOCATinfo). The sea-air CO2 fluxes are computed from the air-sea CO2 partial pressure difference, ∆pCO2, and a bulk gas transfer formulation with windspeed. This data holding includes files of the monthly 1°x1° predictor variables. Several AOML_ET products are provided in this accession. SOCATv2020 and SOCATv2023 are used for training; pCO2w and ∆pCO2 are target variables; two different datasets for air CO2 are used for the calculation of sea-air CO2 fluxes, and three different gas exchange parameterizations are applied to determine the sea-air CO2 fluxes. Details are found in the readme file.

The ocean is a sink for ~25% of the atmospheric CO2 emitted by human activities, an amount in excess of 2 petagrams of carbon per year (PgC yr−1). Time-resolved estimates of global ocean-atmosphere CO2 flux provide an important constraint on the global carbon budget. However, previous estimates of this flux, derived from surface ocean CO2 concentrations, have not corrected the data for temperature gradients between the surface and sampling at a few meters depth, or for the effect of the cool ocean surface skin. Here we calculate a time history of ocean-atmosphere CO2 fluxes from 1992 to 2018, corrected for these effects. These increase the calculated net flux into the oceans significantly.

This dataset provides 1 km gridded monthly estimates of surface ocean partial pressure of CO2 (pCO2) and air-sea flux of CO2 (CO2 flux) for the northern Gulf of America for the period 2006 through 2010. Estimates of pCO2 were derived from MODIS/Aqua satellite imagery in combination with ship-based observations. Estimates of CO2 flux were derived from estimates of seawater pCO2, wind fields, and atmospheric pCO2.