LDEO Surface pCO₂ Database: Dr. Taro Takahashi at LDEO in Palisades, New York started synthesizing global surface ocean CO₂ data in 1997, compiling three decades of observations (∼250000 measurements) to create inaugural monthly global surface pCO₂ maps (Takahashi et al., 1997, 2002). The most recent version (V2019) expanded this dataset to approximately 14.2 million surface water pCO₂ measurements spanning 1957–2019. Distinct from the SOCAT database, the LDEO database reports pCO₂ instead of fCO₂, exclusively from equilibrator-CO₂ analyzer systems, with an average estimated uncertainty of ±2.5 µatm. The database is also interpolated onto a global surface ocean 4°×5° grid for a reference year 2000 (Takahashi et al., 2009) and 2010 (Fay et al., 2024a). Access to the LDEO Surface pCO₂ database (Version 2019) is provided by the Ocean Carbon and Acidification Data System (OCADS, Jiang et al., 2023a) at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0160492.html (last access: 7 January 2026) (Takahashi et al., 2017). Additionally, a dedicated webpage for the LDEO Database is available at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/oceans/LDEO_Underway_Database/ (last access: 7 January 2026).

GLODAPv2: the Global Ocean Data Analysis Project Version 2 (GLODAPv2) aggregates biogeochemical data collected from discrete bottle samples, offering extensive global coverage from the surface to depth (Key et al., 2015; Olsen et al., 2016; Lauvset et al., 2024). While GLODAP is primarily a product for basin-scale hydrographic data, it also includes coastal datasets and observations from a few time-series. The GLODAPv2 data product provides rigorously quality-controlled measurements for 14 essential oceanographic variables: temperature, salinity, dissolved oxygen (DO), nitrate, silicate, phosphate, DIC, TA, pH, chlorofluorocarbons (CFC-11, CFC-12, CFC-113), carbon tetrachloride (CCl₄), and sulfur hexafluoride (SF₆). These variables, excluding temperature, undergo both primary and secondary quality-control procedures to detect outliers and adjust for significant measurement biases. GLODAPv2 was first published in 2016 and was updated annually through a living data process in Earth System Science Data from 2019 through “v2023” which was published in 2024. For these updates, new data (including historical data not previously included in the data product) are quality-controlled and adjusted to the 2016 version (Olsen et al., 2019, 2020; Lauvset et al., 2021, 2022, 2024). Since the global repeat hydrography programs operate with decadal repetitions, the aim is to produce a completely new version of GLODAP, where all cruise datasets will be reevaluated, every decade. Release of the GLODAPv3 data product is planned for 2026, and is expected to evolve the secondary data quality-control practices relative to those used in GLODAPv2. For more information on the secondary quality-control process, refer to Tanhua et al. (2010) and Lauvset and Tanhua (2015). GLODAPv2 offers two kinds of products: the compilation of quality-controlled data from discrete bottle samples taken at sampling location (Key et al., 2015; Olsen et al., 2016, 2019, 2020; Lauvset et al., 2021, 2022, 2024), and a gridded product, interpolated to a 1°×1° grid and the 33 standard depth levels of World Ocean Atlas (WOA) (Lauvset et al., 2016). All versions of the GLODAPv2 data product can be accessed at https://glodap.info/ (last access: 7 January 2026.

GLODAPv2 builds upon three foundational data products: the original GLODAP (Sabine et al., 2004), CARbon dioxide IN the Atlantic Ocean (CARINA, Key et al., 2010), and PACIFic ocean Interior Carbon (PACIFICA, Suzuki et al., 2013). These data products remain available at NCEI: https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0001644.html (last access: 7 January 2026) (GLODAP, Sabine et al., 2005), https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0113899.html (last access: 7 January 2026) (CARINA, Tanhua et al., 2013), and https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0110865.html (last access: 7 January 2026) (PACIFICA, Suzuki et al., 2013), respectively.

Quality Edited Hydrographic Data: the Quality Edited Hydrographic Data product offers both a user-friendly application and a library of ocean profile data curated by Jim Swift (Scripps Institution of Oceanography, La Jolla, California, United States). Similar to GLODAPv2, this data product serves as a comprehensive repository of quality-controlled discrete bottle-based measurements (and limited CTD), spanning from the surface to the depths of the global ocean. Unlike GLODAPv2, this data product does not apply offset corrections. It encompasses a range of oceanographic variables including temperature, salinity, DO, DIC, TA, silicate, phosphate, nitrate, nitrite, CFC-11, CFC-12, and SF₆. To access the application and data, visit: https://joa.ucsd.edu/ (last access: 7 January 2026). Currently, there is not a peer-reviewed paper or public-accessible report for this data product. Cite the data product itself as: Swift and Osborne (2025).

WOD: in addition to the GLODAPv2 (No. 3) and Quality Edited Hydrographic Data (No. 4), users can also access historical and recent original biogeochemical data collected from discrete bottle samples in a uniform format and units, along with their originator quality-control (QC) flags, through the World Ocean Database (WOD) (Mishonov et al., 2024). Like the Quality Edited Hydrographic Data, these measured data remain unaltered. The WOD allows users to filter and subset data by specific variables, platforms, institutions, projects, regions, or time periods (Garcia et al., 2024). Users can visualize sampling locations on a “distribution plot” and access a cruise list for all selected data and variables. Users also have the option of exporting data in NetCDF or Comma-Separated Value (CSV) formats. Additionally, all data in the WOD are reproducible and traceable to their original data sources archived at NOAA's National Centers for Environmental Information (NCEI). The WOD is accessible at https://www.ncei.noaa.gov/products/world-ocean-database (last access: 7 January 2026).

SNAPO-CO₂: Metzl et al. (2024) aggregated over 44 400 measurements of DIC and TA from a series of research cruises and ships of opportunity across various oceanic regions from 1993–2022, under several French research programs, to create a product called “Service National d'Analyse des Paramètres Océaniques du CO₂ (SNAPO-CO₂)”. The majority of the samples were analyzed by the Service National d'Analyse des Paramètres Océaniques du CO₂ (SNAPO-CO₂) at the LOCEAN laboratory in Paris, France. Sampling was performed either from CTD-rosette casts (Niskin bottles) or collected from the ship's flow-through system (intake at roughly 5 m depth). DIC and TA determinations were conducted simultaneously through potentiometric titration in a closed-cell setup, calibrated with certified reference material to achieve an accuracy of ±4 µmol kg⁻¹ for both variables, following Edmond (1970). This methodology was also applied for real-time measurements during OISO cruises, with data from the South Indian Ocean for 1998–2018 included in this compilation. The data are split into two sets: one for the global ocean and coastal zones, and another for the Mediterranean Sea, both accessible in the same format at 10.17882/95414 (Metzl et al., 2023). Additionally, this data product is available at OCADS: https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0285681.html (last access: 7 January 2026) (Metzl et al., 2023).

CODAP-NA: Jiang et al. (2021) synthesized two decades of discrete measurements of carbonate system variables, DO, and nutrient data from the North American continental shelves to generate the first version of Coastal Ocean Data Analysis Data Product in North America (CODAP-NA). The 2021 release encompasses 3391 oceanographic profiles from 61 research cruises spanning the North American continental shelves from Alaska to Mexico in the west and from Canada to the Caribbean in the east. It includes 14 key variables, including temperature, salinity, DO, DIC, TA, pH, carbonate ion, fCO₂, silicate, phosphate, nitrate, all of which have undergone rigorous quality-control. Note that certain datasets meeting the GLODAPv2 QC standards are also included in the GLODAPv2 since its 2022 release (No. 3 above). CODAP-NA is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0219960.html (last access: 7 January 2026) (Jiang et al., 2020).

AZMP Carbon: Gibb et al. (2023) compiled ocean carbonate system variables data from the Canadian Atlantic Zone Monitoring Program (AZMP Carbon) since 2014. More than 100 seagoing missions are represented in this dataset. The sampling strategy generally corresponds to full-depth water samples mostly collected along standardized hydrographic sections. The majority of these data were collected as part of the Atlantic Zone Monitoring Program (AZMP) of Fisheries and Oceans Canada (DFO). Implemented in 1998, the AZMP aims to characterize and understand the causes of oceanic variability at the seasonal, interannual and decadal scales in support of, among other things, fisheries management in the Atlantic Zone (including the Gulf of St. Lawrence, the Scotian shelf and the Newfoundland and Labrador shelf). Since 2014, a minimum of two of the three following carbonate system variables, DIC, TA, and pH, are also acquired by the program at standardized hydrographic stations across the zone (sampled up to three times a year). Each measurement is completed with corresponding temperature, salinity and, when available, nutrients and DO concentration data. This dataset also includes samples collected as part of ships of opportunity, fishing and other scientific trips. The entire dataset comprises 19 531 discrete samples [last updated 21 August 2024]. Among this number, 18 085 have at least two of the three carbonate system variables (e.g., TA, DIC and pH), allowing the derivation of other variables such as the saturation state relative to aragonite and calcite (Ωarag and Ωcalc) and pCO₂ (in µatm). The full dataset of measured and derived variables is available from the Federated Research Data Repository: 10.20383/102.0673 (Cyr et al., 2022) and is updated annually.

MOCHA: Kennedy et al. (2023) curated a comprehensive coastal ocean data product called “Multistressor Observations of Coastal Hypoxia and Acidification (MOCHA)”, encompassing temperature, salinity, DO, ocean carbonate system variables (DIC, TA, pH, pCO₂, fCO₂), nutrients, and chlorophyll measurements from the full water column along the US west coast. The synthesis integrates observations from 71 different sources, including high-resolution autonomous sensors, synoptic oceanographic cruises, and shoreline samples. The MOCHA synthesis spans from the shoreline to well beyond the continental shelf and incorporates observations from CODAP-NA (see No. 7 above), California Cooperative Oceanic Fisheries Investigations (CalCOFI), and other large-scale oceanographic cruises to facilitate linking nearshore, high-resolution observations to broader oceanographic conditions. As of 2025, MOCHA includes 15.9 million temperature readings, 5.0 million salinity measurements, 3.9 million DO records, and 2.3 million pH measurements, along with 8368 DIC, 10 144 TA, and 505 000 pCO₂/fCO₂ measurements, with limited additional chlorophyll and nutrient observations. To reduce the computational load from high-resolution sensors, the synthesis is also available as a “daily aggregated” dataset, with all data sources averaged by day, location, and depth. All data in the MOCHA synthesis product has been quality-controlled to a “plausible and reasonable” standard, but researchers requiring high-precision coastal data may need to apply additional QC tests. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0277984.html (last access: 7 January 2026) (Kennedy et al., 2023), while the methods and the product are described in Kennedy et al. (2024).

ARIOS: the Acidification in the Rias and the Iberian Continental Shelf (ARIOS) project involved compiling and analyzing the historical record of ocean carbonate system measurements and associated variables conducted by the Instituto de Investigacións Mariñas (IIM-CSIC) in Vigo, Spain. This dataset comprises 3343 oceanographic stations and 17 653 discrete samples, combining measurements of pH, TA, and other physical (pressure, temperature, and salinity) and biogeochemical variables (DO, nitrate, phosphate, and silicate) off the northwestern Iberian Peninsula from June 1976 to September 2018 (Padin et al., 2020). The oceanography cruises funded by 24 projects were primarily carried out in the Ría de Vigo coastal inlet, but also in an area ranging from the Bay of Biscay to the Portuguese coast. Robust seasonal cycles and long-term trends were calculated along a longitudinal section, gathering data from the coastal and oceanic zones of the Iberian upwelling system. The data product is available at 10.20350/digitalCSIC/12498 (Pérez et al., 2020).

Marine Inorganic Carbonate Chemistry in the Northern Gulf of Alaska: Monacci et al. (2023) compiled a data product of discrete seawater samples collected each May and September over a 10-year period from 2008 to 2017 along the long-term hydrographic line in the Gulf of Alaska (GAK Line). Samples were collected from a sampling rosette on a profiling CTD. Data variables include profiled seawater temperature, salinity, and DO. Discrete sample variables include DO (i.e., Winkler titrations), macronutrients (nitrate, nitrite, phosphate, silicic acid), DIC, and TA. This data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0277034.html (last access: 7 January 2026) (Monacci et al., 2023), and the synthesis paper can be accessed at 10.5194/essd-16-647-2024 (Monacci et al., 2024).

Coral Reef Carbonate Chemistry Off the Florida Keys: Palacio-Castro et al. (2023) compiled discrete seawater samples from 38 permanent stations located along 10 inshore-offshore transects at the Florida Coral Reef. These samples were collected as part of NOAA's National Coral Reef Monitoring Program (NCRMP) and the South Florida Ecosystem Restoration Research (SFER) cruises. Sampling efforts commenced in 2010, with every two months collections initiated in 2015, resulting in a total of 47 sampling cruises and 1538 discrete seawater samples. For all samples, a minimum of two of the carbonate system variables (TA, DIC) were measured, in addition to salinity and temperature. The Ωarag, pCO₂, and pH were derived from the measured variables using the R package seacarb (Gattuso et al., 2021a). The time-series analysis provides insight into the dynamic carbonate conditions spanning the inshore to offshore gradients, encompassing four distinct regions of the Florida Coral Reef: Biscayne Bay, the Upper Keys, Middle Keys, and Lower Keys. Data is available at https://www.ncei.noaa.gov/access/metadata/landing-page/bin/iso?id=gov.noaa.nodc:NCRMP-CO3-Atlantic (last access: 7 January 2026) (Manzello et al., 2018).

Salish Cruise Data Package and Multi-stressor Data Product: Alin et al. (2025a) compiled data from 61 individual cruise data sets that sampled marine waters of the southern Salish Sea and northern Washington coast (United States) from 2008 to 2024. Since 2014, ongoing seasonal sampling has occurred during April, July, and September for Puget Sound cruises and most frequently during May and October for Sound-to-Sea cruises, which sample from Puget Sound through the Strait of Juan de Fuca to the northern Washington coast. The Salish cruise data package contains observations from water column profiles, with CTD sensor measurements of temperature, salinity, and DO; as well as discrete measurements of DO, nutrients (nitrate, phosphate, silicate, ammonium, nitrite), DIC, and TA. A follow-on data product is also available, containing only samples with complete records for temperature, salinity, and DO from sensors, and DO, nutrients, DIC, and TA from discrete measurements, along with the most commonly used calculated carbonate system variables: pH (total scale), fCO₂, pCO₂, Ωarag, and Ωcalc (Alin et al., 2025b). The data package is available at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/oceans/SalishCruise_DataPackage.html (last access: 7 January 2026). The multi-stressor data product is available at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/oceans/SalishCruises_DataProduct.html (last access: 7 January 2026). Two synthesis papers describing the Salish cruises, as well as seasonality and extreme ocean acidification conditions observed during the 2008–2018 part of the time-series, can be found at https://essd.copernicus.org/articles/16/837/2024/ (last access: 7 January 2026) (Alin et al., 2024a) and https://bg.copernicus.org/articles/21/1639/2024/ (last access: 7 January 2026) (Alin et al., 2024b). A preliminary description of the 2019–2024 Salish cruises can be found at https://www.psp.wa.gov/psmarinewatersoverview.php (last access: 7 January 2026) (Alin et al., 2025c).

Line P Marine Carbonate Chemistry Compilation: this dataset contains marine carbonate system measurements collected during 55 Line P cruises from 1990 to 2019 in the subarctic Northeast Pacific. The dataset contains discrete profiles of DIC, TA, seawater temperature, salinity, DO and nutrients. From a total of 27 hydrographic time-series stations, only the five major stations where DIC and TA are routinely sampled were included in this compilation. Among them is the outermost station P26, also known as Ocean Station Papa (Freeland, 2007). Cruises were conducted approximately three times per year, typically in February, May/June and August/September. Each vertical profile was individually inspected and contrasted with the whole pool of data (including historical data) relative to salinity, density, and oxygen to detect and flag poor quality data following the World Ocean Circulation Experiment (WOCE) quality-control convention (Jiang et al., 2022a). Additionally, the recommended cruise-specific adjustments from PACIFICA were applied (Suzuki et al., 2013). The Line P marine carbonate chemistry compilation is described and analyzed in (Franco et al., 2021a) and is publicly available as a single synthesis product at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0234342.html (last access: 7 January 2026) (Franco et al., 2021b). The Line P carbonate chemistry timeseries is maintained by Fisheries and Oceans Canada and continues to the present day. Data are available and continuously updated in the Line P repository, which can be publicly accessed after generating an account at https://waterproperties.ca (last access: 7 January 2026).

Anthropogenic Carbon in the Arctic Ocean: this dataset includes anthropogenic carbon estimates in the Arctic Ocean based on measurements of transient tracers, such as CFC-12 and SF₆ (Terhaar et al., 2020; Tanhua et al., 2009). Using the transient time distribution (TTD) method, anthropogenic carbon estimates were estimated at measurement locations across all basins of the Arctic Ocean between 1983 and 2005. In addition to these estimates, adjusted estimates of anthropogenic carbon at these locations are provided to account for differences in the saturation of transient tracers and anthropogenic carbon in Arctic Ocean surface waters that caused anthropogenic carbon estimates to be biased low (Terhaar et al., 2020). It is recommended to use the adjusted estimates. This dataset can be accessed at 10.17882/103920 (Terhaar et al., 2024a).

HOT: the Hawaii Ocean Time-series (HOT) CO₂ measurement program documents more than 35 years of inorganic carbon dynamics in the open waters of the central North Pacific. Since October 1988, full ocean depth profiles of DIC and TA have been analyzed, and direct measurements of pH have been made over most of this longest-running Pacific Ocean time-series study. The program is based on shipboard observations and experiments conducted on ∼10 expeditions per year to Station ALOHA (22.75° N, 158° W). HOT program background information and details of sampling strategy may be found in Karl and Lukas (1996) and Karl et al. (2001). Results from the HOT CO₂ measurement program can be found in Winn et al. (1994, 1998), Dore et al. (2003, 2009, 2014), and Knor et al. (2023, 2025). The HOT project page, metadata, data streams and data identifiers are listed at https://www.bco-dmo.org/project/2101 (last access: 7 January 2026). A MAPCO₂ system on the Woods Hole Oceanographic Institution Hawaii Ocean Time-series Site mooring (WHOTS; https://www.soest.hawaii.edu/whots/, last access: 7 January 2026) has provided a near-continuous record of surface pCO₂ since 2004, and is anchored by the longer high-accuracy HOT ship-based program (see Sutton et al., 2019 and Knor et al., 2023). A surface ocean data product that includes CO2SYS-calculated values of pCO₂, carbonate mineral saturation states and other derived quantities may be found at https://hahana.soest.hawaii.edu/hot/hotco2/hotco2.html (last access: 7 January 2026) and 10.5281/zenodo.15060931 (Dore et al., 2025).

ESTOC: the European Station for Time-series in the Ocean (ESTOC) began carbon dioxide monitoring in October 1995, providing a 30-year record on DIC, TA, and pH. This dataset represents the longest continuous monthly record of warming, rising carbon dioxide levels, and acidification in the eastern North Atlantic (González-Dávila and Santana-Casiano, 2023a). ESTOC is located 100 km north of the Canary Islands archipelago (https://plocan.eu/en/installations/ocean-observatory, last access: 7 January 2026). The program includes a ship-based observation system, measuring physical, chemical, and biological variables throughout the 3670 m water column. It also features a moored platform for surface meteorological and oceanic observations as well as subsurface measurements, maintained by the Canary Island Oceanic Platform (PLOCAN, https://plocan.eu/en, last access: 7 January 2026) and the University of Las Palmas de Gran Canaria (https://iocag.ulpgc.es/research/research-units/quima, last access: 7 January 2026). Carbonate system measurements include full-depth bottle sampling for photometric pH, TA, and DIC, conducted monthly from 1995 to 2008, every two months until 2018, and semiannually in recent years due to limited ship time, timed to coincide with moored structure maintenance. ESTOC is also visited every two weeks by a volunteer observing ship, ES-SOOP-CanOA (https://meta.icos-cp.eu/resources/stations/OS_687B, last access: 7 January 2026), part of the European Research Infrastructure ICOS (https://www.icos-cp.eu/observations/ocean/stations, last access: 7 January 2026), which provides real-time surface data on carbon dioxide fluxes and OA. The program also includes the CO₂-ESTOC oceanographic buoy (https://meta.icos-cp.eu/labeling/, last access: 7 January 2026). The full dataset with DOIs is accessible on Pangaea (González-Dávila and Santana-Casiano, 2023a).

Point B Time-series: the Point B Time-series documents the carbonate chemistry at a coastal site in the Bay of Villefranche (43.686200°N 7.314800° E) in Villefranche-sur-mer, France, northwestern Mediterranean Sea. Since January 2007, seawater is sampled weekly at 1 and 50 m, and analyzed for DIC and TA (Kapsenberg et al., 2017). Salinity and temperature are extracted from CTD profiles. Variables of the carbonate system such as pH (total scale) are calculated using the R package seacarb (Gattuso et al., 2021a). Data are available at Pangaea: 10.1594/PANGAEA.727120 (Gattuso et al., 2021b).

Ny-Ålesund Time-series: the Ny-Ålesund Time-series documents the carbonate chemistry at a coastal site of Kongsfjorden, Spitsbergen (78.930660° N 11.920030° E) during the period 2015–2021. It is the first high-frequency (1 h), multi-year (6 years) dataset of salinity, temperature, pCO₂, pH, as well as calculated DIC and TA in the High-Arctic Ocean (Gattuso et al., 2023a). Data are available at Pangaea: 10.1594/PANGAEA.957028 (Gattuso et al., 2023b).

SPOTS: the Synthesis Product for Ocean Time-Series (SPOTS) is a ship-based biogeochemical pilot, aiming at regularly providing high quality data from fixed time-series stations with consistent format and semantics (Lange et al., 2024a). The pilot includes data from 12 fixed ship-based time-series programs with a focus on the Global Ocean Observing System's biogeochemical essential ocean variables. These stations represent unique marine environments across a variety of spatiotemporal resolutions and ranges, with data from 1983 to 2021. While implementing the FAIR principles (Wilkinson et al., 2016) and promoting open data, the metadata of the time-series stations were enhanced to interoperate with the IOC-UNESCO Ocean Data and Information System (ODIS). Additionally, an extensive quality assessment resulted in enhanced intra- and inter-station comparability. Data are available at https://www.bco-dmo.org/dataset/896862 (last access: 7 January 2026) (Lange et al., 2024b).

pCO₂ and pH Time-series from 40 Surface Buoys: Sutton et al. (2019) established a living dataset comprising 40 individual autonomous moored surface ocean pCO₂ time-series established between 2004 and 2013, 17 of which also include autonomous pH measurements. These time-series characterize a wide range of surface ocean carbonate system conditions, across a variety of environments, including 17 oceanic and 13 coastal locations, as well as 10 coral reefs. Data are available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0173932.html (last access: 7 January 2026) (Sutton et al., 2018). Additionally, a dedicated webpage for this project is available at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/oceans/Moorings/ndp097.html (last access: 7 January 2026).

MPI-ULB-SOM-FFN: Landschützer et al. (2020a) created a uniform pCO₂ climatology combining open and coastal oceans. It is a monthly gridded global surface ocean pCO₂ data product without adjusting for a specific reference year. Developed on a higher-resolution 0.25°×0.25° global surface-ocean grid, this product is the result of combining two neural network-based pCO₂ products: the open ocean product described below (i.e., Landschützer et al., 2016) and the coastal product created by Laruelle et al. (2017). Consequently, it represents coastal zones better. Data collected between 1998 and 2015 from the SOCAT database (Version 5) were used to create this data product. The merged climatology product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0209633.html (last access: 7 January 2026). Additionally, a dedicated web page for this project is available at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/oceans/MPI-ULB-SOM_FFN_clim.html (last access: 7 January 2026).

VLIZ-SOM-FFN: Landschützer et al. (2016) employed the Self-Organizing-Map Feed-Forward Network (SOM-FFN) neural network method (Landschützer et al., 2013) to map sea surface pCO₂ from SOCAT (see No. 1 above) (Bakker et al., 2014) to generate monthly pCO₂ fields on a 1°×1° global surface ocean grid, covering the period from 1982 to near present. It is based on the gridded pCO₂ measurements from SOCAT and is updated regularly. The creation of the pCO₂ fields involves a two-step neural network approach, which has been extensively detailed and validated in previous works by Landschützer et al. (2013, 2014, 2016). In the initial step, the global ocean is clustered into biogeochemical provinces, and subsequently, the non-linear relationship between CO₂ driver variables and gridded data from SOCAT (Bakker et al., 2016) is reconstructed. Air–sea CO₂ fluxes are also computed based on the air–sea pCO₂ difference, utilizing a bulk gas transfer formulation as described by Landschützer et al. (2013, 2014, 2016). The product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0160558.html (last access: 7 January 2026). Additionally, a dedicated page for this project is available at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/oceans/SPCO2_1982_present_ETH_SOM_FFN.html (last access: 7 January 2026).

JMA-MLR: Iida et al. (2021) developed a monthly data product for inorganic carbonate variables on a 1°×1° global surface ocean grid for the period 1993–2018. Variables include DIC, TA, pCO₂, air–sea CO₂ flux, pH, and Ωarag. They leveraged data products such as SOCAT.v2019 (Bakker et al., 2016) and GLODAPv2.2019 (Olsen et al., 2019), as well as satellite-based variables, including sea-surface dynamic height (SSDH), mixed layer depth (MLD), and chlorophyll a. The product is updated annually using the latest SOCAT and GLODAPv2 data. The data product can be accessed at https://www.data.jma.go.jp/kaiyou/english/co2_flux/co2_flux_data_en.html (last access: 7 January 2026).

OceanSODA-ETHZ: a.

OceanSODA-ETHZv1 is a monthly gridded global surface ocean data product for multiple ocean carbonate system variables, including DIC, TA, pCO₂, pH (total scale), Ωarag, and Ωcalc (Gregor and Gruber, 2020, 2021, 2023; Ma et al., 2023). This dataset is structured on a 1°×1° global surface ocean grid with monthly resolution from 1982–2022, facilitating research on OA over seasonal to decadal scales. The OceanSODA-ETHZ data product was created by extrapolating in time and space the surface ocean observations of fCO₂ from SOCATv2022 (Bakker et al., 2016) and TA from GLODAPv2.2022 using the newly developed Geospatial Random Cluster Ensemble Regression (GRaCER) method (Gregor, 2021). TA and pCO₂ were then used to calculate the remaining variables of the marine carbonate system with the PyCO2SYS software (Humphreys et al., 2022). Phosphate and silicate from WOA 2018 product was used (Boyer et al., 2018; Garcia et al., 2018a). The OceanSODA-ETHZ data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0220059.html (last access: 7 January 2026).

OceanSODA-ETHZv2 is a surface fCO₂ product with a 0.25°×0.25° spatial resolution and an 8 d temporal resolution, providing estimates starting from 1982 (Gregor et al., 2024a, b). The high-resolution outputs are suitable for investigating the shorter- and finer-scale dynamics of surface fCO₂. Despite sharing a name with its predecessor, OceanSODA-ETHZv2 does not provide TA estimates and employs a different methodology, as described in the following steps: (1) the atmospheric trend of CO₂ is removed by subtracting marine boundary layer CO₂ concentrations from SOCAT fCO₂ producing a new target Δ*CO₂ to reduce the biases at the start and end of the time-series. (2) An 8 d seasonal climatology of Δ*CO₂ is estimated using Gradient Boosted Decision Trees (GBDT), which is later used as a predictor. (3) The non-seasonal thermal component is removed from Δ*CO₂, resulting in a new target, Δ*CO2nonT. (4) The new target is estimated using a feed-forward neural network, with the GBDT as one of the forcing variables. (5) Steps 4 through to 1 are inverted to arrive at fCO₂. (6) Air–sea CO₂ fluxes are computed using ERA5 winds. Data are available at 10.5281/zenodo.11206366 (Gregor et al., 2024b) and are updated annually.

LDEO-HPD fCO₂: the LDEO Hybrid Physics Data (LDEO-HPD) estimates the temporal evolution of surface ocean fCO₂ and air–sea CO₂ exchange, utilizing the strengths of observations and global ocean biogeochemical models (GOBMs) (Gloege et al., 2022). GOBMs are internally consistent, mechanistic representations of the ocean circulation and carbon cycle, and have long been the standard for making spatiotemporally resolved estimates of air–sea CO₂ fluxes. However, there is often a bias between the modelled fCO₂ and available surface ocean measurements (Fay and McKinley, 2021). The LDEO-HPD approach trains an eXtreme Gradient Boosting (XGB) algorithm to learn a non-linear relationship between model-data fCO₂ mismatch and observed predictor variables: sea surface temperature (SST), sea surface salinity (SSS), chlorophyll concentration, mixed layer depth). The GOBM fCO₂ is then corrected with the predicted model-data misfit to estimate real-world fCO₂ for the observation period (Gloege et al., 2022). This results in reconstructed monthly surface ocean fCO₂ and air–sea CO₂ fluxes on a 1°×1° grid in the open ocean beginning in 1982. Additional information can be found at oceancarbon.ldeo.columbia.edu. The data product is available at 10.5281/zenodo.4760205 (Gloege et al., 2021).

LDEO-HPD with Extended Temporal Coverage: building on the work of Gloege et al. (2022), the LDEO-HPD product as mentioned above (No. 28) can be extended back in time to predict fCO₂ for all available model years. Bennington et al. (2022a) find that the largest component of the GOBM corrections is climatological. The smaller corrections at other timescales suggest either that these are well captured by the GOBMs or the data are insufficient. The dominance of climatological corrections supports the extension of the LDEO-HPD fCO₂ product backwards in time. A climatology of model-observation misfits for the best-observed period (2000–present) is applied to the GOBMs for 1959–1981, while an interannually varying correction is used for 1982 onward. (Bennington et al., 2022a). This results in reconstructed monthly surface ocean fCO₂ and air–sea CO₂ fluxes on a 1°×1° grid covering the open ocean, beginning in 1959. Since 2022, the LDEO-HPD Back in Time product has been included in the annual release of the Global Carbon Budget (GCB). Additional information can be found at https://oceancarbon.ldeo.columbia.edu/ (last access: 7 January 2026). The data product can be accessed via Zenodo at 10.5281/zenodo.13891722 (Fay et al., 2024b).

LDEO fCO₂ – Residual Method: a frequently used approach for estimating full-coverage fCO₂ is to train a machine learning algorithm on sparse in situ fCO₂ data and associated physical and biogeochemical observations. While these associated variables have well-known relationships to fCO₂, it is often unclear how they mechanistically drive fCO₂ around the world. The LDEO fCO₂-Residual method takes the basic approach and enhances connections between physical understanding and reconstructed fCO₂. The novel approach used here includes applying pre-processing to the fCO₂ data to remove the direct effect of temperature – a relationship well-documented in literature and lab experiments (Takahashi et al., 2002). This enhances the biogeochemical/physical component of fCO₂ in the target variable (now fCO₂-Residual) and reduces the complexity that the machine learning must disentangle. The resulting algorithm has physically understandable connections between input data and the output biogeochemical/physical component of fCO₂ (Bennington et al., 2022b). This results in reconstructed monthly surface ocean fCO₂ and air–sea CO₂ fluxes on a 1°×1° grid covering the open ocean, beginning in 1982 and extended to the most recent year of available data. Additional information can be found at oceancarbon.ldeo.columbia.edu. The data product can be accessed via Zenodo at 10.5281/zenodo.13941548 (Bennington et al., 2024).

CMEMS-LSCE Surface Ocean Carbonate Data Products: a.

CMEMS-LSCEv1: monthly surface ocean pCO₂ and air–sea CO₂ fluxes on a 1°×1° grid in both the open ocean and coastal seas from 1985–2019 were reconstructed by Chau et al. (2022a). CMEMS-LSCE is short for Copernicus Marine Environment Monitoring Service – Laboratoire des Sciences du Climat et de l'Environnement. This product is generated from an ensemble-based reconstruction of pCO₂ maps trained with gridded data from SOCATv2020 (Bakker et al., 2016). Sea-surface pCO₂ values (converted from the original fCO₂ values in SOCATv2020) were regressed against a set of predictors with non-linear functions, i.e., feed-forward neural network (FFNN) models. The predictors include: sea-surface height (SSH), SST, SSS, MLD, chlorophyll a (Chl a), atmospheric CO₂ mole fraction (xCO₂), and geographical coordinates (longitudes and latitudes). This data product is accessible at https://data.ipsl.fr/catalog/srv/eng/catalog.search#/metadata/a2f0891b-763a-49e9-af1b-78ed78b16982 (last access: 7 January 2026).

CMEMS-LSCEv2: CMEMS-LSCEv2 corresponds to the latest version of the CMEMS-LSCE FFNN. It uses the same ensemble-based reconstruction method for pCO₂ maps as CMEMS-LSCEv1. Improvements include downscaling the spatial resolution to 0.25°×0.25° and reproducing additional surface ocean carbonate system variables on a global grid from 1985 onwards (Chau et al., 2024a). The additional surface ocean carbonate system variables are: pCO₂, DIC, TA, pH, Ωarag, and Ωcalc. Surface ocean pCO₂ is reconstructed based on an ensemble of neural network models mapping gridded observation-based data provided by SOCATv2022 (Bakker et al., 2016). Surface ocean TA is estimated with a multiple linear regression approach (Carter et al., 2016, 2017). The remaining carbonate variables are calculated from pCO₂ and TA using a MATLAB version of CO2SYS (Lewis and Wallace, 1998; Sharp et al., 2023). The CMEMS-LSCE product is updated yearly for surface ocean pCO₂, air–sea fluxes, and the carbonate system variables. Updates are phased with release of the SOCAT database. For surface ocean pCO₂ and air–sea fluxes the temporal coverage is extended to the present date with a latency of 1 month (Chau et al., 2024b). Both the multi-year reconstruction and the near-real time prediction can be accessed through the CMEMS portal: 10.48670/moi-00047 (Chau et al., 2024c).

CarboScope (Jena-MLS): the Jena Mixed-Layer Scheme (within the CarboScope family of data-based estimates of carbon-cycle variability) is based on observed sea surface pCO₂ from SOCAT (see above No. 1) (Bakker et al., 2014). It provides daily global fields of pCO₂ and air–sea CO₂ fluxes from 1957 to the year before present, on a resolution of 2.5°×2° degrees. In the original method (Rödenbeck et al., 2013), a diagnostic model of the carbon balance in the ocean mixed layer is being fitted to the pCO₂ data, by adjusting the ocean-interior sources and sinks of carbon of the mixed layer. The multi-decadal trend is derived from the data-based Ocean Circulation Inverse Model (OCIM) estimate provided by DeVries (2022b). Since a later extension described in Rödenbeck et al. (2022), the variability in the ocean-interior sources and sinks is first regressed against variability in SST and wind speed. The regression step is followed by a correction step with explicit temporal variability, to also represent data variability not yet represented by the predictors of the regression. The CarboScope product is updated yearly. The results from current and previous releases can be downloaded from https://www.bgc-jena.mpg.de/CarboScope/ (last access: 7 January 2026).

UOEx-Watson: this product is an estimate of the atmosphere-ocean flux of CO₂ that takes into account near-surface temperature deviations (Watson et al., 2020). Most estimates use data on surface ocean pCO₂ without considering corrections due to temperature gradients within the uppermost few millimeters of the sea surface (“Skin temperature effects”) or small effects due to changes in temperature that occur during sampling and measurement, especially when the measurement is from a commercial vessel rather than a research ship. This product takes these effects into account by recalculating pCO₂ from the SOCAT data base (v2019) using co-located satellite observations of skin temperature. The result is a substantial increase in the calculated net global uptake of CO₂. In other respects, the methodology for this data product follows the two-step neural network approach described by Landschützer et al. (2013, 2014). The gridded data set of sea surface fCO₂ adjusted to satellite-derived subskin surface temperature, is available at 10.1594/PANGAEA.905316 (Holding et al., 2019). Ocean-atmosphere fluxes interpolated to monthly and 1°×1° spatial resolution is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0301544.html (last access: 7 January 2026).

NIES-ML3: the ensemble product of three machine learning methods (ML3) from the National Institute for Environmental Studies (NIES), Japan, includes monthly global surface ocean fCO₂ in 1982–2024 on 1°×1° grids. Using a leave-one-year-out (LOYO) validation method and three machine learning models, Zeng et al. (2022) found that the time variant trends of ocean CO₂ could be estimated approximately by a harmonic function fitting of the annual atmospheric CO₂. They removed the estimated trends from the ocean CO₂ and applied the LOYO to the trend-removed data to obtain the trend that could not be approximated by the fitting for trend correction. The trend-removed data by the corrected trends were used to train the models. The gap-filled CO₂ maps were constructed by adding the trends to model predictions. The product is available at NIES: 10.17595/20220311.001 (Zeng, 2022).

CSIR-ML6: provides monthly 1°×1° estimates of surface pCO₂ (Gregor et al., 2019a). The approach uses the conceptual two-step approach of clustering and performing regressions for each cluster as Landschützer et al. (2016). CSIR-ML6 investigates the efficacy of various machine learning (ML) methods in estimating surface pCO₂, namely, feed-forward neural networks (FFNN), extremely randomized trees (ERT), gradient boosting machines (GBM), and support vector regression (SVR). It is found that the ensemble of all but the ERT method resulted in the best estimate, highlighting the fact that various ML methods do not produce the same outcome, particularly when data is sparse. Further, the variance between ensemble members can inform us about regions where uncertainty may be large due to methodological differences. Despite this, all methods achieve roughly the same uncertainty – a barrier, or wall beyond which the community has yet to overcome. The data are available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0206205.html (last access: 7 January 2026) (Gregor et al., 2019b). The product is one of the six ensemble members of the SeaFlux dataset.

Stepwise-FFNN: Zhong et al. (2022) constructed a monthly global 1°×1° surface ocean pCO₂ product from January 1992 to December 2024, by combining the stepwise regression algorithm and a feed-forward neural network (FFNN) to select predictors of pCO₂ based on the mean absolute error in each of the 11 biogeochemical provinces defined by the self-organizing map (SOM) method. The methodology for this data product used regionally optimal predictors to account for differences in pCO₂ drivers, lowering local biases relative to a single global predictor set. The developed data product is available at 10.12157/IOCAS.20250814.001 (Zhong, 2025).

AOML-ET: Wanninkhof et al. (2024, 2025) developed a monthly global ocean data product of seawater pCO₂ and air–sea CO₂ fluxes, referred to as AOML-ET, using an extremely randomized trees (ET) machine learning technique. These maps are created on 1°×1° spatial grids, providing global surface ocean coverages from 1998 to 2023. AOML-ET incorporates several predictor variables, including time, location, SST, SSS, MLD, and chlorophyll a. The model was trained using the v2020 and v2023 releases of the SOCAT data product (No. 1). Air–sea CO₂ fluxes were calculated using the air–sea CO₂ partial pressure difference (ΔpCO₂) and a bulk gas transfer formulation incorporating windspeed. The dataset contains monthly 1°×1° NetCDF files of the AOML-ET outputs, along with the predictor variables. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0298989.html (last access: 7 January 2026) (Wanninkhof et al., 2024).

ULB-SOM-FFN-Coastalv2.1: Roobaert et al. (2024) present high-resolution (0.25°×0.25° grid) monthly maps showing the distribution of sea surface pCO₂ across the global coastal ocean, spanning from 1982 to 2020. This product (ULB-SOM-FFN-coastalv2.1) builds upon the work by Laruelle et al. (2017), incorporating a two-step methodology that utilizes Self Organizing Maps (SOM) and Feed Forward Networks (FFN). This updated product now captures temporal variability, enabling the assessment of interannual variability and long-term trends in coastal air–sea CO₂ exchange, unlike the product by Laruelle et al. (2017), which only offers a climatology for a short period (1998–2015). The enhancements include additional environmental predictors and an expanded dataset for training and validation, featuring approximately 18 million direct coastal observations from the SOCAT database, specifically the SOCATv2022 release (Bakker et al., 2016). The product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0279118.html (last access: 7 January 2026) (Roobaert et al., 2023).

RFR-LME: Sharp et al. (2024a) developed a data product delineating the temporal trends of OA indicators mapped on a 0.25°×0.25° spatial grid, across eleven US Large Marine Ecosystems (LMEs), with monthly coverage from 1998–2023. These indicators, which include the pCO₂, pH, Ωarag, DIC, TA, Revelle Factors, among others, were derived from SOCATv2023, along with other oceanographic properties, e.g., SST, SSS, SSH, and MLD. The methodology combined Gaussian Mixture Models to categorize the data into environmentally similar subregions, Random Forest Regressions for the spatial and temporal extrapolation of observational fCO₂ data, and regressions to estimate TA (Carter et al., 2021) to provide a second carbonate system constraint. The resulting maps are available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0287551.html (last access: 7 January 2026) (Sharp et al., 2024b), while an online portal at https://ecowatch.noaa.gov/thematic/ocean-acidification (last access: 7 January 2026) presents regionally averaged time-series for three key indicators: pCO₂, Ωarag, and pH.

ReCAD-NAACOM-pCO₂: Wu et al. (2025) developed a reconstructed pCO₂ product for the North American Atlantic Coastal Ocean Margins (NAACOM), spanning from the Gulf of Mexico/Gulf of America to the Grand Banks, called the Reconstructed Coastal Acidification Database-pCO₂ (ReCAD-NAACOM-pCO₂). This product employed a two-step approach combining random forest regression and linear regression to generate monthly pCO₂ data at 0.25° spatial resolution from 1993–2021. The model was trained using SOCAT v2023 observations as ground-truth values, incorporating various satellite-derived and reanalysis environmental variables known to influence sea surface pCO₂. The ReCAD-NAACOM-pCO₂ dataset is publicly accessible (10.5281/zenodo.11500974, Wu et al., 2024) and will be updated regularly.

Gridded Surface OA Indicators in the Northern Caribbean Sea: this dataset contains a high-quality dataset of derived products from over a million observations of surface water partial pressure/fugacity of carbon dioxide (pCO2w/fCO2w), for the Caribbean Sea, Gulf of Mexico/Gulf of America and North–West Atlantic Ocean covering the timespan from 1 January 2002 to 30 December 2019. The derived quantities include TA, acidity (pH), Ωarag and air–sea CO₂ flux (Wanninkhof et al., 2020). This data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0207749.html (last access: 7 January 2026) (Wanninkhof et al., 2019).

OA Data in the Gulf of Mexico/Gulf of America and Wider Caribbean: the Acidification, Climate, and Coral Reef Ecosystems Team (ACCRETE) Lab within AOML's Ocean Chemistry and Ecosystems Division (OCED) developed a data product for tracking OA in the Caribbean and Gulf of Mexico/Gulf of America from 2014 to 2020 (van Hooidonk, 2022). Utilizing satellite imagery and a data-assimilative hybrid model, the tool maps key indicators of the water's carbonate system, including pCO₂, TA, pH, Ωarag, and Ωcalc. This innovation builds upon an update to the experimental OA Product Suite (OAPS) developed by NOAA's Coral Reef Watch. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0245950.html (last access: 7 January 2026) (van Hooidonk, 2022).

pCO₂ Climatology of the Baltic Sea: Bittig et al. (2024) used biogeochemical model output to inform the mapping of sea surface pCO₂ observations in the Baltic Sea and to build a mean monthly climatology for the period 2003 to 2021, with spatial resolutions of 0.10°×0.05° (approximately 3 nautical miles in both directions). In a first step, spatial patterns of variability were extracted from 20 years of model surface pCO₂ data by an EOF analysis. These spatial patterns were then used to map surface pCO₂ observations from SOCAT (see above No. 1) (Bakker et al., 2014) onto the Baltic Sea domain. By using an ensemble approach with varying number of EOF patterns, the spatial scales of the mapping were locally adjusted based on the observation's data density. Mapped monthly fields of pCO₂ from 2003–2021 were combined for the product into a mean monthly climatology and a spatially-resolved linear trend. The climatology product is available at PANGAEA: 10.1594/PANGAEA.961119 (Bittig et al., 2023).

INCOIS-ReML: the Indian National Centre for Ocean Information Services-Regional Machine Learning model (INCOIS-ReML) pCO₂ data product offers machine learning based monthly climatological sea surface pCO₂ and the corresponding air–sea CO₂ flux for the Bay of Bengal (Joshi et al., 2024). This data product integrates publicly available open-ocean observations with data from the Indian Exclusive Economic Zone. This high-resolution (0.083°×0.083°) monthly climatological pCO₂ data product is available from the INCOIS Portal: https://las.incois.gov.in (last access: 7 January 2026), and from OCADS: https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0307627.html (last access: 7 January 2026) (Joshi et al., 2025a).

INCOIS_TA: the Indian National Centre for Ocean Information Services-Total Alkalinity (INCOIS_TA) data product offers a machine learning based monthly interannual surface TA from 1993–2020 for the North Indian Ocean (Joshi et al., 2025b). This data product integrates publicly available open-ocean observations with data collected during Indian scientific expeditions and from the Indian Exclusive Economic Zone. This high-resolution (0.083°×0.083°) long-term monthly TA data product is available from the INCOIS Portal: https://las.incois.gov.in (last access: 7 January 2026, and from OCADS: https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0307789.html (last access: 7 January 2026) (Joshi et al., 2025c).

Aragonite Saturation State Climatology: Jiang et al. (2015) developed a global interior-ocean Ωarag climatology (referenced to 2000), on a 1°×1° grid at 9 standardized depth levels from the surface down to 4000 m. This was accomplished by integrating data from the first version of GLODAP (Key et al., 2004), CARINA (Key et al., 2010), and PACIFICA (Suzuki et al., 2013), along with additional recent cruise datasets up to 2012. Temporal adjustments were made to a reference year of 2000, accounting for an annual fCO₂ increase of 1.6 µatm in the surface mixed layer (SML), with a rate that decreases linearly to 0 µatm yr⁻¹ from the bottom of the SML to a depth of 1000 m (Sabine et al., 2008). The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0139360.html (last access: 7 January 2026) (Jiang and Feely, 2015).

Mapped Observation-Based Oceanic Dissolved Inorganic Carbon (MOBO-DIC): a.

MOBO-DIC (Version 2020): Keppler et al. (2020a) produced a global interior ocean DIC monthly climatology (average climatological values for January through December) on a 1°×1° grid at 33 standardized depth levels from the surface to 2000 m. The MOBO-DIC mapping method adapts and extends the SOM-FFN technique originally introduced by Landschützer et al. (2013). It starts by categorizing the ocean into clusters with comparable physical and biogeochemical characteristics using self-organizing maps (SOM). Subsequently, within each SOM-defined cluster, a feed-forward network (FFN) is employed to estimate and enforce the statistical correlation between the targeted DIC data and the predictor data available in globally mapped fields. The product uses data from January 2004 to December 2017, and is thus centered around the years 2010/2011. The data product is available at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/oceans/ndp_104/ndp104.html (last access: 7 January 2026).

MOBO-DIC (Version 2023): Keppler et al. (2023a) extended the temporal resolution of MOBO-DIC to resolve monthly fields from January 2004 to December 2019, as opposed to the average climatological values in Keppler et al. (2020a). This data product is on a 1°×1° grid at 28 depth levels from the surface to 1500 m. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0277099.html (last access: 7 January 2026) (Keppler et al., 2023b).

Monthly Interior Ocean TA Climatology: Broullón et al. (2019) developed a monthly global interior ocean TA climatology using a feed-forward neural network approach. This dataset offers a spatial resolution of 1°×1° in the horizontal, spans 102 depth levels (ranging from 0–5500 m) in the vertical dimension, and features a temporal resolution that varies from monthly (0–1500 m) to annual (1550–5500 m). The development of this climatology was based on the analysis of TA in relation to several key predictor variables, including temperature, salinity, nutrients (phosphate, nitrate, and silicate), DO, and sampling position (coordinates and depth), as outlined in Velo et al. (2013). Both TA and these predictor variables were sourced from GLODAPv2 (version 2016) (Olsen et al., 2016). The global interior ocean TA climatology was constructed by leveraging the established relationships between TA and the predictor variables, as well as the monthly climatologies of temperature, salinity, and DO from the WOA 2013 (Locarnini et al., 2013; Zweng et al., 2013; Garcia et al., 2014), and nutrients data that were obtained through the CANYON-B neural network process, applied to the previously mentioned fields. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0222470.html (last access: 7 January 2026) (Broullón et al., 2020b).

Monthly Interior Ocean DIC Climatology: Broullón et al. (2020a) employed a feed-forward neural network approach to create a monthly global interior ocean DIC climatology, centered around the year 1995. This dataset offers a 1°×1° spatial resolution in the horizontal domain, encompassing 102 depth levels ranging from 0–5500 m vertically. The temporal resolution varies, ranging from monthly (0–1500 m) to annual (1550–5500 m). In contrast to their previous work on TA (Broullón et al., 2019), this analysis includes the variable “year” to account for anthropogenic DIC pool changes. It also incorporates data from the LDEO pCO₂ database (Takahashi et al., 2017) alongside GLODAPv2.2019 (Olsen et al., 2019) to establish relationships between DIC and its input variables: temperature, salinity, DO, as well as location, pressure, and time. The DIC climatology was derived using these relationships, along with monthly climatological data for temperature, salinity, and DO from WOA 2013 (Locarnini et al., 2013; Zweng et al., 2013; Garcia et al., 2014), as well as phosphate, nitrate, and silicate values computed from the CANYON-B neural network fed with the aforementioned fields. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0222469.html (last access: 7 January 2026) (Broullón et al., 2020c).

Acidification Metrics in the Ocean Interior: Fassbender et al. (2023) generated estimates of global interior ocean changes to pH, [H⁺], Ωarag, pCO₂, and the Revelle sensitivity factor driven by the accumulation of anthropogenic carbon (C_ant) from the preindustrial period to 2002, and quantified the component of these changes caused by carbonate system nonlinearities. For each OA metric, the dataset includes year 2002 values and quasi-preindustrial values, which were estimated by subtracting C_ant from the year 2002 carbonate chemistry information and recomputing each OA metric without considering any warming, circulation, or biological changes that may have occurred since the preindustrial era. Data from the upper 2000 m of the GLODAPv2 Climatology (No. 46, Lauvset et al., 2016) and from the preformed properties product of Carter et al. (2021) were used to make these estimates on the 1°×1° GLODAPv2 Climatology grid for 26 depth levels from the surface to 2000 m. The provided uncertainties were estimated using a 1000-iteration Monte Carlo simulation. Calculation details are described in Fassbender et al. (2023). Year 2002 Ωarag and pH values, and their uncertainties, are reproduced from the GLODAPv2 Climatology and are provided in this dataset for user convenience with the permission of the original data producer. This data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0290073.html (last access: 7 January 2026) (Fassbender, 2024).

Ocean Interior Acidification Over the Industrial Era: building on the total anthropogenic carbon estimates for 1994 from Sabine et al. (2005) and the decadal changes between 1994 and 2014 reconstructed by Müller et al. (2023a), Müller and Gruber (2024a) quantified ocean interior acidification over the industrial era. To convert the increasing anthropogenic carbon concentrations into acidification estimates, their approach relied on time-invariant climatologies of ocean interior DIC, TA, temperature, salinity, and other relevant variables to determine the background state of the marine carbonate system. Hence, their estimates resolve exclusively the acidification driven by the anthropogenic carbon accumulation. In contrast to direct observations of acidification variables, such as those collected at time-series stations, this approach does not account for changes in the natural carbon cycle or the displacement of water masses. The approach by Müller and Gruber (2024a) is conceptually similar to that of Fassbender et al. (2023), but provides temporally resolved estimates, enabling the tracking of both the spatial distribution and temporal evolution of ocean interior acidification. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0298993.html (last access: 7 January 2026) (Müller and Gruber, 2024b).

Decadal changes of anthropogenic CO₂: a.

Anthropogenic CO₂ from 1994 to 2007: Gruber et al. (2019a) estimated the decadal time-scale changes in the oceanic content of anthropogenic CO₂ (ΔC_ant) between 1994 to 2007. The results were derived from the GLODAPv2.2016 product (Olsen et al., 2016), utilizing the eMLR(C^*) methodology pioneered by Clement and Gruber (2018). The product is combined with the estimated amount of C_ant for 1994 derived by Sabine et al. (2004) from GLODAPv1 to infer C_ant for 2007. All estimates are geospatially distributed on a horizontal grid with a resolution of 1°×1°. Two primary files are available: one providing the complete three-dimensional distribution of ΔC_ant, and the other containing vertically integrated values, i.e., the column inventories. This data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0186034.html (last access: 7 January 2026) (Gruber et al., 2019b).

Decadal Trends in Anthropogenic CO₂ from 1994 to 2014: Müller et al. (2023a) extended the analysis by Gruber et al. (2019a) to reconstruct decadal trends in the oceanic storage of ΔC_ant in the global ocean interior from mid-year 1994 to mid-year 2004, and further to mid-year 2014. They applied the extended multiple linear regression (eMLR) method (Clement and Gruber, 2018) to ship-borne observations of DIC and other biogeochemical variables from GLODAPv2.2021 (Lauvset et al., 2021). All estimates are provided on a 1°×1° horizontal grid. Two principal data files are provided: one featuring the comprehensive three-dimensional distribution of ΔC_ant for the two time periods, and the other presenting the vertically integrated quantities, i.e., the column inventories. The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0279447.html (last access: 7 January 2026) (Müller et al., 2023b).

Tracer-based rapid anthropogenic carbon estimation from 1750 to 2500: Carter et al. (2025) developed a method for estimating C_ant based on a machine learning translation of ocean circulation information inferred from transient tracer distributions. They applied it to the gridded GLODAPv2 climatology to obtain estimates of the past and projected C_ant distribution between 1750 and 2500. Projections are made using a range of simple assumptions and shared socioeconomic pathway projections. Estimates are provided on 1°×1° spatial grids at 33 standard depth levels in micromoles C_ant per kg of seawater. This data product is available at 10.5281/zenodo.15003059 (Carter, 2025).

Preformed TA and other biogeochemical properties: Carter et al. (2020) estimated preformed seawater TA, nitrate, silicate, phosphate, and oxygen using empirical seawater property estimation routines (Carter et al., 2017) with ocean circulation pathway information from ocean circulation transport matrices (John et al., 2020). Preformed properties are estimated property contents that seawater had when it last left contact with the atmosphere, and are used as an aid in interpretation of measured ocean property distributions. This data product is available at 10.5281/zenodo.3745002 (BRCScienceProducts, 2020).

Monthly Interior Ocean pH Climatology: Zhong et al. (2025) developed a monthly 1°×1° gridded global seawater pH (total scale) climatology from 1992 to 2020 at in situ temperature, derived using a machine learning algorithm trained on pH observations from GLODAPv2 (Lauvset et al., 2024). The product spans from 1992 to 2020 and covers depths from the surface to 2000 m across 41 vertical levels. Its development involved a three-step machine-learning approach: (1) regional division using a self-organizing map neural network, (2) predictor selection via stepwise regression, which iteratively adds or removes variables based on their impact on reconstruction error, and (3) nonlinear regression using feedforward neural networks (FFNNs). The developed data product is available at 10.12157/IOCAS.20230720.001 (Zhong et al., 2023).

CODAP-NA Climatology: Jiang et al. (2024) developed a coastal OA indicators climatology on a 1°×1° grid, covering North American ocean margins from the surface to 500 m at 14 standardized depth levels. This product includes 10 key oceanographic variables: fCO₂, pH, [H⁺]_total, free hydrogen ion content ([H⁺]free), carbonate ion content ([CO32-]), Ωarag, Ωcalc, DIC, TA, and Revelle Factor (RF), as well as temperature and salinity. The climatology was produced with the WOA gridding technologies of the NOAA National Centers for Environmental Information (NCEI), based on the recently released Coastal Ocean Data Analysis Product in North America (CODAP-NA) (Jiang et al., 2021), along with GLODAPv2.2022 (Lauvset et al., 2022). The relevant variables were adjusted to the year of 2010 before the gridding. The first-guess fields for this analysis were calculated using ESPERs (Carter et al., 2021), based on the WOA (Version 2018) climatologies for salinity (Zweng et al., 2019), temperature (Locarnini et al., 2019) and DO (Garcia et al., 2018b). The data product is available in NetCDF at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0270962.html (last access: 7 January 2026) (Jiang et al., 2022b). Additionally, maps of these indicators are available in jpeg at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/synthesis/nacoastal.html (last access: 7 January 2026) (Jiang et al., 2022b).

RECCAP2: in the context of the second iteration of the project REgional Carbon Cycle Assessment and Processes (RECCAP2), the ocean carbon community compiled, quality-controlled, and harmonized (in the sense of providing output on the same regular grid at the same spatial and temporal resolution) 12 GOBMs simulations, 11 pCO₂ products, one ocean interior DIC product, and three data assimilation models to constrain the ocean carbon sink between 1985 and 2018. The RECCAP2 synthesis effort stands as a distinct but complementary resource to the GCB project (Friedlingstein et al., 2025), which primarily focuses on anthropogenically perturbed surface CO₂ fluxes from a global budgeting perspective. The individual chapters of RECCAP2 were published in this special issue of Global Biogeochemical Cycles: https://agupubs.onlinelibrary.wiley.com/doi/toc/10.1002/(ISSN)2169-8961.RECCAP2 (last access: 7 January 2026). The data products of this assessment are available on a 1°×1° horizontal grid, with monthly resolution for surface ocean variables such as air–sea CO₂ fluxes, and annual resolution for interior ocean variables, such as DIC content. The data compilation, which is described in detail in DeVries et al. (2023) and evaluated in Terhaar et al. (2024b), is available at 10.5281/zenodo.7990823 (Müller, 2023).

Global Carbon Budget: the GCB collects annually updated estimates of the ocean carbon sink from currently nine fCO₂-products and ten GOBMs for the period 1959 to the past calendar year (https://globalcarbonbudget.org/gcb-2025, last access: 7 January 2026, Friedlingstein et al., 2025). In contrast to Earth System Models (ESMs), the GOBMs are here forced with atmospheric reanalysis that ingested atmosphere and ocean observations and are thus thought to be closer to the observed climate. Gridded fields are provided on a 1°×1° horizontal grid and monthly resolution. In addition, globally and regionally integrated air–sea CO₂ fluxes from the native model grids are provided. Globally integrated time-series are adjusted for full ocean coverage and model bias and drift and are available for each individual fCO₂-product and GOBM (https://globalcarbonbudget.org/download/1442/?tmstv=1731323337, last access: 7 January 2026). The model data goes well beyond surface fluxes and includes data to analyze drivers of carbon fluxes, including several 3D variables. The model data request has been updated since RECCAP2 and also provides, for example, monthly interior ocean data of DIC, TA, nutrients and DO. The GOBM data request was also updated to have all variables available that are needed to serve as a testbed for fCO₂-products (e.g., sea surface height). Gridded surface data of sea surface fugacity and air–sea CO₂ flux of all fCO₂-products and GOBMs as used in the latest release of GCB (Version 2024) are published on Zenodo (10.5281/zenodo.14639761, Hauck et al., 2025). All other GOBM output is available via https://globalcarbonbudgetdata.org/closed-access-requests.html (last access: 7 January 2026).

ECCO-Darwin: Carroll et al. (2022) used the Estimating the Circulation and Climate of the Ocean-Darwin (ECCO-Darwin) global-ocean biogeochemistry state estimate to generate a data-constrained DIC budget and investigate how spatiotemporal variability in advection and mixing, air–sea CO₂ flux, and the biological pump have modulated the ocean sink for 1995–2018. ECCO-Darwin assimilates ocean circulation and physical tracers, including temperature, salinity, and sea ice, derived from the Estimating the Circulation and Climate of the Ocean (ECCO) LLC270 global-ocean and sea-ice data synthesis (Zhang et al., 2018). Additionally, it assimilates biogeochemical observations encompassing the cycling of carbon, nitrogen, phosphorus (PO₄), iron (Fe), silica (SiO₂), DO, and TA. This inclusive approach enhances the model's fidelity by aligning it with a diverse array of observations. All ECCO-Darwin model output is available on the ECCO Data Portal: https://data.nas.nasa.gov/ecco/ (last access: 7 January 2026). The model code and platform-independent instructions for running ECCO-Darwin simulations can be found at https://github.com/MITgcm-contrib/ecco_darwin (last access: 7 January 2026).

Global Surface Ocean Acidification Indicators: a.

Surface pH and Revelle Factor: Jiang et al. (2019a) produced a high-resolution (1°×1°) data product delineating a regionally varying view of global surface ocean pH, acidity, and Revelle Factor (RF) from 1770 to 2100 by amalgamating recent observational seawater CO₂ data from the SOCAT database (Version 6) (Bakker et al., 2016) and temporal trends at individual locations of the global surface ocean from an Earth System Model, i.e., GFDL-ESM2M (Dunne et al., 2013). The calculations were conducted under historical atmospheric CO₂ levels (pre-2005) and four Representative Concentrations Pathways (post-2005) corresponding to the Intergovernmental Panel on Climate Change (IPCC)'s 5th Assessment Report, specifically RCP2.6, RCP4.5, RCP6.0, and RCP8.5. Surface ocean TA was calculated from SSS and SST using the updated locally interpolated alkalinity regression (LIARv2) method (Carter et al., 2017). Surface ocean pH, acidity, and RF were then calculated using a MATLAB version of the CO2SYS program (Orr et al., 2015). The data product is available at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0206289.html (last access: 7 January 2026) (Jiang et al., 2019b).

Surface OA Indicators: Jiang et al. (2023b) developed a comprehensive model-data fusion product that delineates the trajectory of 10 OA indicators: fCO₂, pH, [H⁺]_total, [H⁺]_free, [CO32-], Ωarag, Ωcalc, DIC, TA, and RF, as well as temperature and salinity at all locations of the global surface ocean from 1750 to 2100. This product marks a significant improvement in OA forecasting by refining temporal trends with data from 14 ESMs within CMIP6, and by applying bias and drift corrections using three updated observational ocean carbonate system data products: SOCAT (Version 2022) (Bakker et al., 2016), GLODAPv2.2022 (Lauvset et al., 2022), and CODAP-NA (Jiang et al., 2021). This dataset offers 10-year averages on a 1°×1° global surface ocean grid, capturing trends from preindustrial times (1750) through historical conditions (1850–2010), and projects future conditions to 2100 across five Shared Socioeconomic Pathways: SSP1-1.9, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5. The gridded data product is available in NetCDF at https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0259391.html (last access: 7 January 2026) (Jiang et al., 2022c), and global maps of these indicators are available in JPEG at https://www.ncei.noaa.gov/access/ocean-carbon-acidification-data-system/synthesis/surface-oa-indicators.html (last access: 7 January 2026) (Jiang et al., 2022c).

Simulated and Constrained Global and Southern Ocean Carbon Sink: these two datasets include spatially-integrated and annually averaged values for the ocean carbon sink from 1850 to 2100 for different scenarios over the 21st century for the global ocean (Terhaar et al., 2022a, b) and the Southern Ocean (Terhaar et al., 2021b, c). All results are based on CMIP5 and CMIP6 models. For the global ocean carbon sink, values are available for SSP1-2.6, SSP2-4.5, and SSP5-8.5. For the Southern Ocean, values are also available for SSP1-2.6, SSP2-4.5, and SSP5-8.5 and additionally for RCP2.6, RCP4.5, and RCP8.5. In addition, to the raw simulated values, constrained estimates of the annually averaged ocean carbon sink estimates are available. These constrained estimates adjusted the simulated carbon sink estimates for biases on the ocean's circulation and surface carbonate chemistry (see Terhaar et al., 2021b, 2022a for details). It is recommended to use the constrained estimates. The datasets are available at 10.17882/103934 (Terhaar et al., 2022b) and 10.17882/103938 (Terhaar et al., 2021c).

Composite model-based estimate of the ocean carbon sink from 1959 to 2022: this data product, developed by Terhaar (2025), presents an estimate of the global ocean carbon sink by combining forced hindcast simulations and simulations made by coupled ESMs. Hindcast models manage to adequately simulate the short-term variability of the ocean, but struggle to simulate the long-term climate change trend (Huguenin et al., 2022; Takano et al., 2023; Hollitzer et al., 2024). ESMs cannot simulate the observed short-term variability by definition, but accurately simulate long-term trends (Takano et al., 2023; Hollitzer et al., 2024). The composite model-based estimate combines the simulated short-term variability from hindcast simulations and the long-term trend from ESMs. The output is supplied with the associated study (10.5194/bg-22-1631-2025) (Terhaar, 2025).

pCIBR_Clim and pCIBR_Int: a machine learning (ML) model is employed to correct biases in surface pCO₂ simulations generated by the INCOIS-BIO-ROMS model (pCO₂model) over the period 1980–2019. The ML model is trained using the differences between observed (pCO₂obs) and modeled pCO₂ to estimate the spatio-temporal deviations (pCO₂obs-pCO₂model). These interannually and climatologically varying deviations are then added back to the original model output, resulting in two improved data products: pCIBR_Int and pCIBR_Clim (Ghoshal et al., 2025a). Evaluation against independent datasets, including moored observations (BOBOA), the gridded SOCAT product, and other ML-based pCO₂ products (such as CMEMS-LSCEv2 and OceanSODA), demonstrates a significant improvement of approximately 40 % ± 3.31 % in RMSE compared to the original model. This high-resolution (0.083°×0.083°), long-term monthly pCO₂ data product is available from the INCOIS Portal (https://las.incois.gov.in, last access; 7 January 2026) and from OCADS: https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0307788.html (last access: 7 January 2026) (Ghoshal et al., 2025b).

INCOIS-BIO-ROMS Simulated Surface pCO₂ and pH for the Indian Ocean: this data product presents a comprehensive assessment of OA trends across the Indian Ocean and its sub-regions from 1980 to 2019, leveraging outputs from a regional, high-resolution coupled ocean-ecosystem model (INCOIS-BIO-ROMS), an offline biogeochemical (BGC) model, and two machine learning-based products (Chakraborty et al., 2024). INCOIS-BIO-ROMS, configured at 1/12° resolution for the Indian Ocean, was developed in accordance with the “RECCAP-2: Ocean Modeling Protocol” for regional oceans. The INCOIS-BIO-ROMS simulated surface pCO₂ and pH data product is available from the INCOIS Portal (https://las.incois.gov.in, last access: 7 January 2026) and from OCADS: https://www.ncei.noaa.gov/data/oceans/ncei/ocads/metadata/0307663.html (last access: 7 January 2026) (Chakraborty et al., 2025).

Ocean Circulation Inverse Model (OCIM): DeVries (2022b) utilized a two-step procedure to estimate anthropogenic carbon in the ocean interior using an ocean inverse model. First, a steady-state ocean circulation inverse model (OCIM) was fit to observations of physical circulation tracers such as temperature, salinity, radiocarbon, and CFCs (Holzer et al., 2021). Then, the circulation model was coupled to an abiotic carbon cycle model, spun up to equilibrium in 1780 and then forced by observed atmospheric CO₂ time history from 1781–2020. Simulations were run with and without historical changes in sea surface temperatures. The difference between preindustrial and transient simulations represents the anthropogenic carbon accumulation in the ocean. Results are provided on a regular grid with a nominal resolution of 2° in the horizontal with 48 depth levels. The output is available at 10.6084/m9.figshare.19341974.v2 (DeVries, 2022c).