Blind Spots in the Big Blue: The Challenge of Measuring the Ocean's Carbon Sink

Introduction to Ocean Carbon (CO2) Sequestration

The global ocean operates as a primary regulatory mechanism within the Earth's climate system. Since the dawn of the industrial era, the marine environment has absorbed approximately one-quarter of all anthropogenic carbon dioxide emissions, an amount equating to an estimated 180 plus or minus 35 Petagrams of carbon between 1850 and 2023¹. This marine carbon sink significantly attenuates the accumulation of greenhouse gases in the atmosphere, thereby modulating the pace of global temperature increases. However, the exact magnitude, spatial distribution, and temporal variability of this sink are influenced by complex physical and biogeochemical drivers, including sea surface temperature variations, ocean circulation patterns, and biological productivity. Accurately quantifying the flux of carbon dioxide across the air-sea interface requires extensive, highly accurate, and continuous in situ observations across all oceanic regions.

To address this global scientific necessity, the international marine carbon research community established the Surface Ocean CO2 Atlas (SOCAT), a synthesis activity designed to collate, standardize, and quality-control surface ocean carbon data⁴. Released in June 2026, the Surface Ocean CO2 Atlas version 2026 (SOCATv2026) provides the most detailed historical record of marine carbon data to date, summarizing its findings in an annual release poster⁵. While the dataset highlights significant methodological advancements in quantifying the ocean carbon sink, it simultaneously documents an escalating crisis within the global climate monitoring infrastructure due to a persistent decline in physical observations. This report provides a detailed analysis of the SOCATv2026 dataset, exploring its structural architecture, recent analytical corrections regarding temperature biases, the statistical impacts of declining observational coverage, and the integration of novel autonomous platforms intended to sustain future ocean carbon monitoring.

1. The Architecture and Scale of SOCAT Version 2026

SOCAT is an open-access, community-led synthesis product that focuses primarily on the fugacity of carbon dioxide (fCO2) in surface seawater⁵. Fugacity is a metric closely related to partial pressure but includes corrections for the non-ideal behavior of the gas under specific environmental conditions, providing a more thermodynamically accurate representation of the gas exchange potential between the ocean and the atmosphere.

1.1 Data Volume and Accuracy Standards

The foundational strength of the SOCAT database lies in its rigorous standardization and scale. SOCATv2026 contains 44 million individual observations collected between 1957 and 2025⁵. These physical measurements are aggregated from a diverse array of observational platforms, including dedicated research vessels, commercial ships of opportunity, fixed moorings, uncrewed surface vehicles, and sailing yachts⁴.

To maintain the integrity required for high-level climate modeling, SOCAT categorizes its data based on stringent accuracy thresholds. To be included in the primary SOCAT synthesis and gridded products—designated by quality control flags ranging from A to D—observations must achieve an estimated accuracy of better than plus or minus 5 micro-atmospheres⁵. These core observations serve as the primary baseline for calculating global carbon fluxes.

In addition to this primary dataset, SOCATv2026 includes 8.4 million calibrated sensor observations that possess a slightly lower estimated accuracy of plus or minus 5 to 10 micro-atmospheres, designated as Flag E data⁵. Historically, Flag E data were excluded from the most rigorous global flux estimates. However, as spatial data gaps have widened in recent years, the modeling community is increasingly evaluating methods to incorporate this supplementary data to constrain flux estimates in traditionally under-sampled regions¹⁰.

1.2 Governance and Quality Control Mechanisms

The operational maintenance of the SOCAT database relies heavily on the voluntary efforts of more than one hundred contributing scientists worldwide¹. The governance structure is highly decentralized. It consists of a Global Group responsible for overarching strategic coordination, supported by specific technical hubs and Regional Working Groups that execute the secondary quality control of data within defined geographic domains¹¹. This regional distribution ensures that raw data sets are evaluated by researchers who possess specialized local expertise regarding unique regional oceanographic dynamics, such as seasonal upwelling or sea-ice melt.

The quality control workflow requires comprehensive metadata submission. Data providers utilize an online metadata editor formatted to the latest requirements outlined in the 2024 SOCAT Quality Control Cookbook, which standardizes reporting formats to ensure interoperability¹¹. To manage the continuous influx of data, the SOCAT community actively maintains data fidelity through organized 24-hour quality control hackathons, such as those held in January and February 2026¹³. These events allow experienced reviewers across multiple time zones to collaboratively resolve data discrepancies prior to the annual data freeze.

To illustrate the distributed nature of this scientific effort, the regional groups responsible for the evaluation of SOCATv2026 data are detailed in the table below.

Regional Working Group	Geographic Scope	Key Oceanographic Focus Areas
Coastal and Marginal Seas	Global coastal zones	Complex hydrodynamics, terrestrial carbon inputs, high temporal variability.
Arctic Ocean	Arctic Ocean including coastal waters	Sea-ice dynamics, seasonal meltwater impacts, high-latitude gas solubility.
North Atlantic	30 degrees North to Fram Strait	Deep water formation zones, the Gulf Stream, significant anthropogenic sink.
North Pacific	North of 30 degrees North	Subpolar gyres, ocean acidification impacts on regional ecosystems.
Tropical Atlantic & Pacific	30 degrees North to 30 degrees South	Equatorial upwelling zones representing strong natural carbon dioxide sources.
Indian Ocean	North of 30 degrees South	Monsoonal forcing, historical under-sampling.
Southern Ocean	South of 30 degrees South	The largest single oceanic carbon sink, driven by intense winds, highly data-sparse during austral winter.

Data sourced from the SOCATv2026 Regional Group Coordination Directory¹⁶.

2. Methodological Advancements: Resolving the Warm Bias

A critical scientific evolution applied to the SOCATv2026 dataset involves the systematic recalculation of historical surface water data to correct for temperature anomalies inherent in ship-based measurements. This physical correction, often referred to as resolving the "warm bias," fundamentally refines the baseline estimation of the global ocean carbon sink.

2.1 The Physics of the Ship Intake Problem

Historically, the vast majority of surface ocean carbon observations have been collected via flow-through systems installed on commercial ships of opportunity and research vessels. These systems measure the temperature and carbon dioxide concentration of seawater pumped through intake valves typically located several meters below the ocean surface¹⁷. However, the actual air-sea exchange of trace gases occurs at the immediate surface boundary—the sub-skin layer¹⁷.

Because seawater temperature exerts a profound influence on the solubility of gases and the speciation of carbonate ions, even minor temperature variations between the intake depth and the surface sub-skin (approximately 0.2 meters deep) can skew gas measurements. Furthermore, seawater traveling through a ship's internal plumbing often warms slightly before reaching the analytical sensors. Because warmer water holds less dissolved gas, artificially warmed water will yield a higher partial pressure of carbon dioxide than what actually exists at the ocean surface interface.

Previous analyses of this warm bias suggested the global mean temperature anomaly ranged from 0.02 degrees Celsius to 0.09 degrees Celsius, with notable latitudinal variability, ranging from a cooling effect of 0.05 degrees Celsius in certain subtropical regions to a warming effect of 0.15 degrees Celsius at higher latitudes¹⁷. While seemingly minute, these variations alter the calculated gas transfer velocity and the concentration gradients that drive carbon fluxes.

2.2 Recalculating to a Climate Data Record

To correct this structural discrepancy, researchers systematically recalculated the entire SOCAT database with an estimated uncertainty of less than 5 micro-atmospheres⁴. Because the recalculation process does not assume isochemical conditions, it accurately captures temperature-driven shifts in carbonate speciation⁴.

The methodology pairs every in-water gas measurement with a highly stable sea surface temperature climate data record. Specifically, researchers utilized the European Space Agency Climate Change Initiative Sea Surface Temperature (ESA CCI SST) product⁴. By standardizing the temperature reference to a consistent depth of 0.2 meters, the artificial warm bias introduced by ship infrastructure is effectively removed, producing a dataset that represents the true thermodynamic conditions at the air-sea boundary⁴.

2.3 Implications for the Global Carbon Budget

The implications of this recalibration are substantial. By robustly accounting for the observed warm bias, calculations indicate that the surface exchange layer is slightly cooler than previously recorded by ship intakes. Consequently, the ocean's theoretical capacity to absorb and hold carbon dioxide is higher than historical estimates suggested. Removing this temperature artifact via recalculation results in an approximate 12 percent increase in the calculated ocean carbon sink for the year 2024, equating to an additional 0.4 Petagrams of Carbon per year, bringing the total estimated sink to 3.4 Petagrams of Carbon per year⁴.

Because of the profound impact this correction has on atmospheric climate models and inventory closures, these recalculated data products are now deemed essential for annual carbon assessments⁴. This adjustment played a pivotal role in the Global Carbon Budget 2025 assessment, where the ocean carbon sink was revised upward, contributing to a more accurate closure of the global carbon inventory¹⁵.

3. Data Extrapolation and Machine Learning Reconstructions

Despite containing 44 million observations, the physical measurements within the SOCAT database cover only a fraction of the world's oceans. At a spatial resolution of 1-degree latitude by 1-degree longitude, SOCAT observations cover approximately 2 percent of the global ocean surface in any given month between 1982 and 2024¹⁹. To translate these sparse data points into a globally comprehensive map of air-sea carbon exchange, researchers rely on sophisticated observation-based data products that reconstruct surface ocean fugacity across unmeasured space and time¹⁹.

3.1 Neural Networks and the fCO2-Residual Method

These data products utilize statistical interpolation and machine learning techniques, specifically training neural network ensembles on the available SOCAT datasets alongside secondary environmental predictor variables. For example, methods such as the fCO2-Residual approach extrapolate carbon fugacity globally by finding statistical relationships between the sparse carbon measurements and ubiquitous satellite data, such as sea surface temperature, salinity, and chlorophyll-a concentrations¹⁹.

By learning the biogeochemical relationships in regions where data exist, the models can predict carbon dioxide concentrations in regions where ships have not traveled. However, the fidelity of these neural networks is entirely dependent on the spatial and temporal diversity of the training data provided by SOCAT.

3.2 Evaluating Reconstruction Error via Testbeds

To assess the accuracy of these machine learning reconstructions, researchers employ a "testbed" approach using outputs from global ocean biogeochemistry models, such as those from the Coupled Model Intercomparison Project Phase 6 (CMIP6)¹⁹. Instead of using actual SOCAT observations, the algorithmic training set is sub-sampled from the testbed to match the exact spatial and temporal locations where real ships have traveled.

Because the testbed provides a complete, known "truth" for the entire simulated ocean, researchers can compare the neural network's reconstructed map against the true simulated map. This process reveals the inherent errors and biases of the reconstruction methods, demonstrating that data sparsity—particularly in harsh, unnavigable regions during winter months—is the primary driver of uncertainty in global flux estimates¹⁹.

4. The Crisis of the Declining Observational Network

While the analytical methods and machine learning algorithms applied to SOCAT data have reached unprecedented levels of sophistication, the physical collection of the underlying data is experiencing a severe and sustained crisis. Despite the release of new observations in version 2026, the overall rate of data acquisition has been contracting significantly for nearly a decade, threatening the reliability of future climate assessments⁵.

4.1 The Trajectory of the Decline

The zenith of open-ocean carbon dioxide observing efforts occurred in 2017⁷. Since that peak, the number of measurements contributed to the annual SOCAT updates has decreased by almost half²². In practical terms, the geographic area of the global ocean featuring monthly measurements has shrunk from roughly 18,000 grid cells in 2017 to approximately 10,000 grid cells globally²². The current data density mirrors the limited observing capabilities seen a decade ago⁷.

This decline exhibits a severe geographic bias. The Northern Hemisphere retains moderate coverage due to the high density of trans-Atlantic and trans-Pacific commercial shipping routes that host automated sensors. However, the global south is drastically under-sampled. Massive data voids now exist in the subtropical and tropical regions of the Northern Hemisphere (south of 20 degrees North latitude) and across nearly the entirety of the Southern Hemisphere⁷.

4.2 Compounding Vulnerabilities: Funding and Infrastructure

The initial catalyst for this decline was a structural funding shortfall for sustained monitoring programs. Observational networks are often viewed by funding agencies as operational overhead rather than novel scientific research, making long-term financial support difficult to secure². This financial fragility was acutely exacerbated by the COVID-19 pandemic starting in 2020, which grounded research vessels and severely disrupted the commercial shipping networks upon which the Surface Ocean CO2 Reference Observing Network relies².

Furthermore, the digital data management architecture of SOCAT itself is highly vulnerable. The system relies heavily on the voluntary efforts of a small international team of scientists. Following the closure of its European data management hub in 2022, the database has critically relied on limited support from the Pacific Marine Environmental Laboratory of the United States National Oceanic and Atmospheric Administration⁹. This limited support has kept basic data ingestion functioning but has prevented necessary updates to the core data architecture, leaving the entire system brittle and susceptible to external organizational shocks²⁵.

4.3 The Statistical Cost of Missing Data

The absence of raw observations actively degrades the mathematical models used to estimate global climate health. When the foundational training data diminishes, the neural networks lose their predictive fidelity. A comprehensive analysis demonstrated that the decline in data availability from 2017 to 2021 caused a 65 percent increase in the uncertainty of SOCAT-based ocean carbon flux estimates, raising the standard deviation of the flux from 0.15 to 0.25 Petagrams of Carbon per year¹⁰.

Moreover, reducing data availability to historical levels artificially introduces substantial bias into long-term flux trends, causing deviations of up to 50 percent in certain modeled scenarios¹⁰. This divergence is particularly pronounced in highly dynamic regions such as the Southern Ocean, upwelling zones, and the subpolar Pacific, where seasonal extremes require dense, year-round sampling to accurately capture the temporal shift between carbon dioxide outgassing and absorption¹⁰.

Observational Metric	Peak Era (circa 2017)	Current Era	Impact on Global Modeling
Active 1-degree Grid Cells	~18,000 cells	~10,000 cells	Loss of spatial resolution in dynamic upwelling zones and coastal margins.
Flux Estimate Uncertainty	Baseline (0.15 Pg C yr-1)	+65% Increase (0.25 Pg C yr-1)	Reduced confidence in Global Carbon Budget calculations.
Long-term Trend Bias	Minimal	Up to 50% bias introduced	Skewed understanding of the ocean's response to rising anthropogenic emissions.
Southern Hemisphere Data	Moderate	Severely Degraded	Inability to accurately map the Earth's largest marine carbon sink.

Table 1: Metrics synthesizing the decline in observational coverage and its modeled impacts on climate data products¹⁰.

5. Bridging the Gap: Autonomous Platforms and Sensor Biases

Faced with a shrinking fleet of commercial ships of opportunity and the high operational costs of dedicated research cruises, the oceanographic community is rapidly pivoting toward autonomous technology to sustain the observational requirements of the SOCAT database.

5.1 Uncrewed Surface Vehicles

To combat the geographic biases inherent in commercial shipping, agencies are expanding the deployment of autonomous uncrewed surface vehicles (USVs), such as Saildrones. These wind- and solar-powered vessels can be equipped with high-accuracy carbon dioxide sensors and navigated into hostile or remote environments that are rarely transited by traditional vessels. For example, targeted deployments during the Atlantic hurricane season allow scientists to continuously monitor fluxes during extreme weather events—periods characterized by massive, rapid gas exchange that historically went unrecorded due to the severe danger posed to crewed ships⁷.

5.2 Biogeochemical Floats and the Bias Challenge

Another highly promising avenue is the integration of data from biogeochemical profiling floats, such as those utilized by the Global Ocean Biogeochemistry Array (GO-BGC) program²⁶. Unlike surface vehicles, these floats drift with ocean currents and can adjust their buoyancy to profile the water column, periodically surfacing to transmit data via satellite. Incorporating float-based observations is vital for the Southern Ocean, a region responsible for approximately 40 percent of the global oceanic uptake of anthropogenic carbon dioxide, yet remains vastly under-sampled during the harsh austral winter³.

However, the integration of autonomous float data into the SOCAT framework presents significant analytical challenges. While traditional shipboard measurements achieve an accuracy of better than 5 micro-atmospheres, float-based estimates are derived from proxy sensor readings and often carry random uncertainties of plus or minus 11 micro-atmospheres³. Advanced simulations have demonstrated that adding float observations to the global array significantly reduces the underestimation of the ocean carbon sink by filling spatial gaps.

Yet, this mathematical improvement holds true only if the sensor data are perfectly unbiased. If systematic bias exists within the float sensors—even a minor persistent offset—it drastically degrades the machine learning reconstructions. Research indicates that systematic bias in float observations can lead to an underestimation of the global ocean carbon sink by up to 0.32 Petagrams of Carbon per year³. Because the global mean air-sea disequilibrium driving gas exchange is extremely small (on the order of 5 to 8 micro-atmospheres), even slight sensor drift can alter the modeled output, shifting a region from indicating a carbon sink to a carbon source³. Consequently, rigorous, ongoing cross-calibration between autonomous platforms and high-accuracy shipboard data remains an absolute necessity for future dataset versions.

6. Policy Implications: The European Audit and Strategic Stewardship

The scientific vulnerabilities highlighted by the release of SOCATv2026 have profound geopolitical and policy implications. The ocean's capacity to absorb carbon dioxide directly dictates the remaining carbon budget compatible with limiting global warming to 1.5 degrees Celsius—a threshold that the Global Carbon Budget 2025 assessment notes is now virtually exhausted¹⁵. To provide context for these negotiations, preliminary data for 2025 indicated total anthropogenic carbon dioxide emissions of 11.6 plus or minus 0.9 Petagrams of Carbon per year²⁷. Of this, the atmospheric growth rate accounted for 7.9 Petagrams, the land sink accounted for 1.9 Petagrams, and the recalculated ocean sink accounted for 3.4 Petagrams²⁷. If the measurements dictating climate policy are based on a degrading observational network, international carbon reduction targets rest on highly unstable empirical foundations.

This precarious reality was the focal point of a comprehensive 2026 audit titled "State of Europe's surface ocean CO2 observations," published by the Joint Programming Initiative (JPI) Oceans⁹. The audit investigated Europe's capacity across the entire marine carbon value chain, aiming to strengthen regional monitoring capabilities⁹.

6.1 The Leadership Mismatch

The primary finding of the JPI Oceans audit was the identification of a severe "leadership mismatch" within the global carbon science community. European scientific institutions produce the gold-standard biogeochemical models and data products that drive the Global Carbon Budget and inform the Intergovernmental Panel on Climate Change (IPCC)⁹. However, this advanced modeling prowess is dangerously dependent on an observational foundation that is actively fracturing. The number of European nations actively funding and providing surface ocean observations is shrinking, and critical regional blind spots persist in the Mediterranean Sea, the Black Sea, and the North-East Atlantic⁹.

Furthermore, the audit highlighted the strategic risk of Europe relying almost entirely on non-European infrastructures for data stewardship, a vulnerability made acute following the closure of the SOCAT European regional hub⁹.

6.2 Strategic Recommendations for Network Resilience

To prevent the collapse of this critical climate monitoring infrastructure, the JPI Oceans audit outlined several urgent policy recommendations:

1. Institutionalize Funding: Transition ocean carbon monitoring from short-term, grant-based research projects to sustained, operational funding mechanisms. The audit recommends integrating these measurements as routine components of maritime operations rather than bespoke research activities⁹.

2. Expand the Fleet: Increase the deployment of high-accuracy systems across the full spectrum of maritime platforms, providing funding and regulatory incentives to encourage commercial shipping fleets to host automated sampling infrastructure⁹.

3. Establish a European Hub: Implement a dedicated European hub for ocean carbon reference materials and rebuild regional data stewardship capacity. This would decentralize the risk currently shouldered primarily by United States agencies⁹.

4. Target Under-observed Regions: Mandate cost-sharing models at the European level to prioritize data collection in severe blind spots, specifically the Mediterranean Sea, Black Sea, and the climatically critical Southern Ocean⁹.

5. Annual Auditing: Establish the JPI Oceans audit as a yearly, transparent review of observational capacities to ensure the network evolves in tandem with strategic climate priorities⁹.

7. Conclusion

The release of the Surface Ocean CO2 Atlas version 2026 underscores the immense dedication of the international marine carbon research community. With 44 million meticulously quality-controlled observations spanning nearly seven decades, SOCAT remains the premier tool for understanding the thermodynamic and biogeochemical complexities of the ocean carbon sink. Methodological triumphs, such as the recalculation of the database to a standardized sea surface temperature climate data record, have successfully resolved historical warm biases, refining our understanding of air-sea gas exchange and closing critical gaps in the Global Carbon Budget.

However, the narrative surrounding SOCATv2026 is inherently paradoxical. At the exact moment when machine learning algorithms and climate models have achieved the sophistication required to map the global carbon cycle with unprecedented nuance, the physical hardware supplying these models with real-world validation is being allowed to atrophy. The continuous decline in raw observations since 2017—driven by pandemic aftershocks and chronic underfunding—has injected measurable levels of uncertainty into climate assessments. While autonomous uncrewed surface vehicles and biogeochemical floats offer a necessary lifeline to fill vast data voids in the Southern Hemisphere and extreme weather environments, their utility is entirely dependent on the maintenance of rigorous, unbiased calibration standards.

Ultimately, the ocean's role as a climate shock-absorber is altering as global waters warm and stratify. Tracking this shifting dynamic requires recognizing that marine carbon observation is not merely an academic exercise, but a critical planetary defense infrastructure. Securing the future of the Surface Ocean CO2 Atlas will require immediate, coordinated international policy interventions to transition ocean carbon monitoring from a voluntary scientific endeavor into a permanently funded, globally integrated operational network.

Works cited

1. Search datasets - Catalog - Data.gov, https://catalog.data.gov/?q=Surface+Ocean+CO2+Atlas+(SOCAT)

2. Partnership Progress 2021-06-30 | Department of Economic and Social Affairs, https://sdgs.un.org/partnership-progress/partnership-progress-2021-06-30

3. The importance of adding unbiased Argo observations to the ocean carbon observing system - PMC, https://pmc.ncbi.nlm.nih.gov/articles/PMC11347652/

4. (PDF) Recalculating the Surface Ocean CO2 Atlas (SOCAT) to a sea surface temperature climate data record - ResearchGate, https://www.researchgate.net/publication/406512124_Recalculating_the_Surface_Ocean_CO2_Atlas_SOCAT_to_a_sea_surface_temperature_climate_data_record

5. Surface Ocean CO₂ Atlas – A Collection of Surface Ocean CO₂ Observations Quality Controlled by the Science Community, https://socat.info/

6. https://socat.info/wp-content/uploads/2026/06/2026_Poster_SOCATv2026.pdf

7. Oceanic measurements of carbon dioxide continue to decrease, as reported in this year's ocean carbon data atlas, https://globalocean.noaa.gov/oceanic-measurements-of-carbon-dioxide-continue-to-decrease-as-reported-in-this-years-ocean-carbon-data-atlas/

8. NOAA hosts newly released surface ocean CO2 data, https://oceanacidification.noaa.gov/noaa-hosts-newly-released-surface-ocean-co2-data/

9. State of Europe's surface ocean CO2 observations - Audit 2026, https://www.jpi-oceans.eu/sites/jpi-oceans.eu/files/publications/state_of_europe_s_surface_ocean_co2_observations_audit.pdf

10. Accuracy of Ocean CO2 Uptake Estimates at a Risk by a Reduction in the Data Collection - OceanRep, https://oceanrep.geomar.de/60266/1/Geophysical%20Research%20Letters%20-%202024%20-%20Dong%20-%20Accuracy%20of%20Ocean%20CO2%20Uptake%20Estimates%20at%20a%20Risk%20by%20a%20Reduction%20in%20the%20Data.pdf

11. Data Submission and Quality Control – Surface Ocean CO₂ Atlas - SOCAT, https://socat.info/index.php/data-upload-and-quality-control/

12. Governance – Surface Ocean CO₂ Atlas - SOCAT, https://socat.info/index.php/governance/

13. Steve Jones – Surface Ocean CO₂ Atlas - SOCAT, https://socat.info/index.php/author/zuj007/

14. QC Hackathon 22nd Jan 2026 – Surface Ocean CO₂ Atlas - SOCAT, https://socat.info/index.php/2025/11/03/qc-hackathon-feb-3rd-2026/

15. News – Surface Ocean CO₂ Atlas - SOCAT, https://socat.info/index.php/news/

16. Groups for SOCAT v2026 - Surface Ocean CO₂ Atlas, https://socat.info/index.php/groups-for-socat-v2026/

17. (PDF) Recalculating the Surface Ocean CO2 Atlas (SOCAT) to a sea surface temperature climate data record - ResearchGate, https://www.researchgate.net/publication/398280047_Recalculating_the_Surface_Ocean_CO2_Atlas_SOCAT_to_a_sea_surface_temperature_climate_data_record

18. Thomas Holding's research works | University of Exeter and other places - ResearchGate, https://www.researchgate.net/scientific-contributions/Thomas-Holding-2166151237

19. An update of the LDEO fCO2-Residual method: algorithmic choices improve ocean carbon - EarthArXiv, https://eartharxiv.org/repository/object/11157/download/20296/

20. Impact of Surface Ocean CO2 Atlas (SOCAT) observing network extensions on the quantification of global air-sea CO2 fluxes - PubMed, https://pubmed.ncbi.nlm.nih.gov/40857945/

21. The Alarming Decline in Ocean CO2 Observations - EGU Blogs, https://blogs.egu.eu/divisions/os/2024/10/14/decline-in-ocean-co2-observations/

22. Latest Ocean Carbon Data Atlas Shows a Significant Decline in Ocean CO2 Measurements, https://globalocean.noaa.gov/latest-ocean-carbon-data-atlas-shows-a-significant-decline-in-ocean-co2-measurements/

23. SOCAT version 2023 released - Scientific Committee on Oceanic Research, https://scor-int.org/2023/06/22/socat-version-2023-released/

24. https://globalocean.noaa.gov/latest-ocean-carbon-data-atlas-shows-a-significant-decline-in-ocean-co2-measurements/#:~:text=Global%20capacity%20for%20open%20ocean%20CO2%20observations%20is%20declining&text=SOCAT%20Version%202023%20revealed%20that,only%2010%2C000%20grid%20cells%20globally.

25. Surface Ocean CO2 Atlas Database Version 2025 (SOCATv2025) (NCEI Accession 0304549) - University of Otago, https://ourarchive.otago.ac.nz/esploro/outputs/dataset/Surface-Ocean-CO2-Atlas-Database-Version/9926747383901891

26. News - International Ocean Carbon Coordination Project (IOCCP), https://www.ioccp.org/news

27. Dorothee C. E. Bakker's research works | University of East Anglia and other places - ResearchGate, https://www.researchgate.net/scientific-contributions/Dorothee-C-E-Bakker-2145954404

28. State of Europe's Surface Ocean CO2 Observations - Plumtri, https://plumtri.org/plumtri.org/State_of_Europes_surface_ocean_CO2_observations

29. Ocean Carbon Capacities | JPI Oceans, https://www.jpi-oceans.eu/en/ocean-carbon-capacities