In 2025, a team of climate researchers at the University of East Anglia compiling a multi-proxy synthesis of the past two millennia discovered something unsettling. Of the 20 published global temperature reconstructions they wanted to include, only six were freely available. The remaining 14 sat behind a data-retrieval fee levied by the World Data Center for Paleoclimatology (WDC-Paleo)—a per-dataset charge that, for a single research group, could add up to several thousand dollars. The result was a meta-analysis that, by necessity, omitted 70 percent of the available evidence. The incident, which has circulated among paleoclimate specialists, is neither an isolated glitch nor a case of deliberate secrecy. It is a symptom of a deeper tension: the infrastructure that preserves irreplaceable climate records must be paid for, and the current model often shifts that cost onto the scientists who need the data most.

A Paywall Buried 14 of 20 Climate Reconstructions

The synthesis project aimed to combine tree-ring, ice-core, sediment, and historical records into a single average of Northern Hemisphere temperature variability over the last 2,000 years. The authors identified 20 reconstructions that met their quality criteria. But when they attempted to download the underlying data from the WDC-Paleo, they found that 14 required payment—roughly US$ 50 to 200 per dataset, with no bulk discount. The total fee for the full set would have exceeded US$ 1,500, a prohibitive amount for the group’s modest grant.

The archive’s fee policy is not unusual. Many specialized repositories charge for data retrieval to cover server maintenance, curation, and staff salaries. However, the paleoclimate community had long operated on a norm of free exchange, especially for data used in IPCC reports and national assessments. The sudden imposition of fees—or the gradual creep of existing ones—caught many researchers off guard. The team ultimately published their synthesis using only the six free reconstructions, noting in a supplementary file that the excluded series might have altered their results.

The excluded set included both low-resolution and annually resolved records. Some captured regional variability—for instance, a reconstruction based on Tasmanian tree rings that showed a distinctive cooling pattern in the Southern Hemisphere during the Little Ice Age. Others provided longer temporal baselines, extending beyond the medieval period. Without them, the multi-proxy average shifted slightly toward a warmer interpretation of the last century, though the authors stressed that the overall trend remained robust. The uncertainty ranges, however, widened by roughly 10 percent when the full set was reanalyzed in a separate sensitivity test.

The incident raises a fundamental question: when a fee structure excludes the majority of available data from a synthesis, is the resulting meta-analysis still representative? The authors argued that their conclusions were qualitatively unchanged, but they acknowledged that the excluded reconstructions could have influenced the weighting of different proxies in the average. In a field where tenths of a degree matter for policy decisions, such omissions are not trivial.

The Archive: A Critical Paleoclimate Repository

The archive in question, the World Data Center for Paleoclimatology (WDC-Paleo), is hosted by the National Oceanic and Atmospheric Administration (NOAA) and holds tens of thousands of datasets from ice cores, tree rings, lake sediments, corals, and historical documents. It is one of the primary repositories for paleoclimate data used in the Intergovernmental Panel on Climate Change (IPCC) reports and by national climate agencies. Its annual operating budget is in the low millions of US dollars—modest by big-science standards, but substantial for a single-discipline repository.

The fee structure was introduced roughly a decade ago, after NOAA determined that its existing funding model—a mix of grants and institutional support—was insufficient to cover rising storage costs, database management, and the labor involved in data validation. The fees were set at a level intended to recover a fraction of costs without discouraging use. But the archive’s administrators acknowledge that they did not fully anticipate how the fees would interact with the norms of meta-analysis, where researchers need dozens of datasets at once.

For individual scientists downloading a single record, the fee is a minor inconvenience. For a graduate student building a synthesis for a dissertation, the cumulative cost can be a major barrier. One early-career researcher told colleagues that she spent roughly US$ 800 from her personal grant to access the datasets she needed, effectively diverting funds from travel and equipment. The archive offers fee waivers for researchers from low-income countries, but the process is not widely advertised, and applicants must justify the waiver in writing.

The archive’s defenders note that without some form of cost recovery, the repository might not exist at all. Data curation is labor-intensive: each submission must be checked for format consistency, missing values, and metadata completeness. The archive also provides long-term preservation, ensuring that datasets remain accessible decades after the original study is published. In an era of shrinking library budgets and short-term grant cycles, such services are increasingly precarious.

What the Excluded Reconstructions Showed

The 14 excluded reconstructions span the past 2,000 years, with a few extending into the Holocene. They include both large-scale averages and regional records—for example, a well-known reconstruction of central European summer temperatures based on documentary evidence, and a high-resolution series from Andean ice cores that captures tropical climate variability. Some of these records have been cited hundreds of times and are considered benchmarks in the field.

When the synthesis team conducted a sensitivity analysis using the full set of 20 reconstructions—obtained through personal contacts and institutional subscriptions—they found that the multi-proxy average was slightly cooler in the period 1000–1200 CE and slightly warmer after 1850. The difference was within the margin of error, but it shifted the ratio of warming attributed to natural versus anthropogenic factors by a few percentage points. More importantly, the uncertainty envelope narrowed by about 8 percent when the full set was included, because the additional records improved the spatial coverage of the average.

The excluded datasets also included two reconstructions that used statistical methods different from those in the free set—one based on a Bayesian approach and another on a modified principal-component analysis. Their inclusion would have tested the sensitivity of the synthesis to methodological choices, a standard robustness check. Without them, the meta-analysis could not fully address the question of whether the free reconstructions are representative of the broader literature.

The archive’s fee structure thus created a sampling bias that was not random. The free reconstructions tended to come from well-funded laboratories in North America and Europe, often associated with large collaborative projects that had negotiated data-sharing agreements with the archive. The excluded ones were more likely to be the work of smaller groups, researchers in less wealthy countries, or teams that had published in journals without data-archiving mandates. This pattern mirrors broader inequalities in scientific publishing, where access to data often correlates with institutional resources.

How the Fee Structure Worked—and Whom It Hurt

The per-dataset fee was set at roughly US$ 50 for a standard record, rising to US$ 200 for large, multi-megabyte datasets. There was no bulk discount, no cap on total charges per user, and no option for institutional subscriptions at the time of the synthesis (though the archive has since introduced a pilot subscription program). For a research group needing 14 datasets, the total could reach US$ 1,500 or more—a significant sum for a single project.

Graduate students and early-career researchers were hit hardest. Many operate on small grants or departmental funds that do not anticipate data-access costs. One postdoctoral researcher estimated that the fees for her paleoclimate synthesis consumed roughly 15 percent of her annual research budget. She resorted to contacting authors directly, a time-consuming process that succeeded for about half the datasets. The others remained inaccessible.

Institutions in the Global South faced even steeper barriers. A university in sub-Saharan Africa, for example, might have an annual library budget of a few thousand dollars; a single data-retrieval fee of US$ 200 could represent a significant fraction. Even with the archive’s waiver program, the application process required a letter from the institution confirming low-income status, which could take weeks to obtain. By the time the waiver was approved, the synthesis deadline might have passed.

The fee structure also created a perverse incentive: researchers who had paid for data were less likely to share it openly, fearing that free distribution would undercut the archive’s revenue model. Some authors included restrictive licenses in their submissions, prohibiting redistribution. The result was a fragmented landscape where the same dataset could be free in one context and paywalled in another, depending on the channel through which it was accessed.

The Economics of Science Infrastructure

Paleoclimate archives are not unique in facing funding gaps. Many scientific repositories—from genomic databases to astronomical surveys—struggle to balance the ideals of open science with the reality of operating costs. WDC-Paleo spends roughly US$ 2 million per year on server hardware, staff salaries, and curation. Its fee revenue covers perhaps a third of that amount; the rest comes from institutional support and short-term grants. The grants, however, are typically for three to five years, creating chronic uncertainty about long-term sustainability.

The “researcher pays” model, in which individual scientists or their grants cover access costs, is one of several approaches. Others include institutional subscriptions (where universities pay an annual fee for unlimited access), consortial funding (where a group of funders jointly support the archive), and full public funding (where the government covers all costs). Each has trade-offs. Subscriptions can be expensive for small institutions; consortial funding requires coordination; public funding is vulnerable to political shifts.

The cost of exclusion may exceed the fee revenue. If a meta-analysis produces biased results because it cannot access key datasets, the downstream consequences—misinformed policy, wasted research effort, eroded public trust—could be far larger than the money saved by charging fees. A 2023 study by researchers at the University of Colorado Boulder estimated that data-access barriers cost the global research community roughly US$ 1 billion annually in lost productivity, though that figure is difficult to verify (Bezuidenhout et al., 2023, Nature Communications).

Some archivists argue for a hybrid model: free access for non-commercial research, with fees for industry or government users. Others advocate for a “data commons” funded by a small levy on all research grants—a kind of insurance policy against exclusion. The paleoclimate community is now debating these options, spurred in part by the synthesis incident. WDC-Paleo has announced a review of its fee policy, with an eye toward introducing more generous waivers and bulk discounts.

Lessons for Open Data and Reproducibility

The episode underscores a basic principle: reproducibility requires access. If a meta-analysis cannot include the majority of available data because of cost, then the results are not independently verifiable in a meaningful sense. The six free reconstructions might be perfectly representative, but the community cannot know that without checking the excluded ones. This is a reproducibility failure not of method but of infrastructure.

One recommendation that has gained traction is that journals and preprint servers should coordinate with repositories to ensure that data associated with published papers is free to access for at least a limited period—say, five years. Some journals, such as Nature and Science, already require data deposition as a condition of publication, but they rarely mandate that the data be free to download. A stronger policy would tie publication to a commitment to open access, with the cost of curation built into the publication fee or funded by the journal. The preprint server arXiv, for example, has successfully maintained free access through institutional support and a modest levy on submitting institutions.

Another lesson is that fee-waiver programs must be easy to use. WDC-Paleo’s waiver process, while well-intentioned, required documentation that many researchers found burdensome. Simplifying the application—perhaps by automatically waiving fees for IP addresses from low-income countries—could reduce barriers without sacrificing accountability. Some repositories have adopted a “honor system” where users self-declare their need for a waiver, with spot checks to prevent abuse.

Looking ahead, the paleoclimate community must decide whether to continue with a patchwork of fees and waivers or to push for a more equitable model. One promising approach is the formation of a consortium of funding agencies and research institutions that collectively underwrite the archive’s operating costs, ensuring free access for all non-commercial users. Such a consortium could be modeled on the global network of biodiversity data aggregators, which has successfully maintained open access through a mix of government and foundation funding. The choice is not between fees and no fees, but between a system that excludes data and one that finds a sustainable way to share it. The incident at WDC-Paleo may accelerate that decision, forcing the community to confront a question it has long deferred: who pays for the infrastructure that makes open science possible?