The 2025 Nature paper by Gebrechorkos et al. on warming and global drought severity was retracted in early September 2026 due to calculation and coding errors.

The paper is: Gebrechorkos, Solomon H., Justin Sheffield, Sergio M. Vicente-Serrano, Chris Funk, Diego G. Miralles, Jian Peng, Ellen Dyer, et al. (2025). “Warming accelerates global drought severity.”

Nature 642(8068): 628- 635 (published online 4 June 2025; DOI 10.1038/s41586-025-09047-2). It has been formally retracted, with a Retraction Note published 2 September 2026.

Using high- resolution drought datasets (1901- 2022) that incorporated atmospheric evaporative demand (AED, sometimes called “atmospheric thirst”), the authors reported an increasing trend in drought severity.

Key quantitative claims included that AED had increased drought severity by an average of ~40% globally; that drought- affected areas expanded substantially in recent years (e.g., +74% on average in 2018- 2022 vs. 1981- 2017, with AED contributing a large share); and that 2022 was particularly extreme.

They linked the rise in AED to warming and argued this role would likely grow under further warming. The work received substantial media coverage and citations while active.

The authors retracted it after being notified of errors and identifying additional issues themselves. The main problems were:

Area- weighted grid cells were not used when calculating study- wide averages. This affected numerical values in the main text, Fig. 1b, and some Extended Data figures.

Results based on an earlier version of a dataset (GLEAM v3 instead of GLEAM4) were used in one Extended Data figure; the datasets are largely consistent for most of the period, but the proper weighting changed the 2022 drought– affected area figure.

A related coding and time stepping error affected date handling when subsetting data.

The authors stated that the errors changed the reported percentages (the corrected figures are expected to be somewhat lower) but that the overall conclusions were not overturned. They indicated they intend to submit a fully revised and corrected version for peer review. Retraction Watch coverage quotes the lead author describing it as a coding error unrelated to the underlying data or core methodology.

Lead author Solomon H. Gebrechorkos and co- authors including Ellen Dyer and Simon J. Dadson are (or were) affiliated with the University of Oxford’s School of Geography and the Environment, related units. Other co- authors are from institutions including the University of Southampton, Spanish and Belgian research groups, US institutions (e.g., UCSB), and others. It is an international collaboration, not solely an Oxford “pack,” though Oxford researchers played prominent roles and the university publicized the original findings.

Retractions of high- profile climate- related papers for methodological or calculation errors do occur and are part of the normal scientific process of correction (especially for complex global datasets and derived indices).

The paper’s high citation count before retraction means some of the original quantitative claims circulated widely; corrected numbers, once published, will need to be checked against those earlier reports.

The authors’ stated position is that the directional findings on AED’s role hold, albeit with adjusted magnitudes.

Independent verification of any revised version will be the proper next step.

The specific quantitative claims you listed from the 2025 Gebrechorkos et al. Nature paper are the ones that were reported and widely disseminated, and those published numbers are unreliable due to the calculation and coding errors that led to the retraction.

They cannot be treated as established results.

The paper’s abstract and key findings stated (among other points):

• An increasing trend in drought severity worldwide since roughly the early 1980s.
• Atmospheric evaporative demand (AED, the “thirst of the atmosphere”) increased drought severity by an average of ~40% globally.
• In 2018- 2022 versus 1981- 2017, drought- affected areas expanded by 74% on average, with AED contributing 58% of that increase.
• In 2022, ~30% of global land area experienced moderate to extreme drought, with a substantial fraction attributed to increased AED.
• The role of AED was expected to grow with further warming.

These figures were the ones circulated in media coverage and the authors’ own communications.

The official Retraction Note (2 September 2026) and related statements identify these problems:

• Area- weighted grid cells were not used when calculating study- wide averages. This directly affected numerical values in the main text, Figure 1b, and Extended Data Figures 4 and 8.
• Results based on an earlier dataset version (GLEAM v3 instead of GLEAM4) were used in one figure; after proper area weighting the 2022 drought- affected area figure was reduced.
• A time stepping and coding error meant dates were not properly parsed when sub setting data for the 1981- 2017 versus 2018–2022 periods. This misclassified time steps and affected regional averages.
• The application of an arid- regions mask (areas with mean annual precipitation < 180 mm) was not clearly explained and was applied selectively.

These are concrete computational and data- handling mistakes in producing the global and regional averages and percentage attributions. The authors themselves describe them as changing the reported percentages (they expect the corrected numbers to be “slightly lower”) while asserting that the broader directional conclusion, that rising AED linked to warming has exacerbated drought severity, still holds.

They withdrew the paper because the scope of corrections required full revision and new peer review rather than a simple correction notice.

The published numerical results (the 40%, 74%, 58%, 30% figures, etc.) are invalidated by the errors and should not be cited as accurate.

That is why the paper was retracted. However, the authors and the retraction notice do not state that the entire analysis or the qualitative finding is false or that the methodology was fundamentally invalid.

They frame the issues as correctable calculation and coding problems (including failure to area- weight properly and a date- parsing bug) and plan to resubmit a revised version. Independent evaluation of any corrected analysis will be needed to judge the revised magnitudes and robustness.

The colorful description of “comparing apples at the equator with dry ice in the Arctic” overstates what the documented errors were.

The documented failures were in area- weighting of grid cells, dataset version, time- period subsetting via code, and clarity of masking, not a wholesale mixing of incompatible physical regimes in the manner suggested. Area- weighting errors on a global grid can distort averages (especially if high- latitude or other regions are mishandled), which is why proper weighting is standard practice; that is precisely one of the errors acknowledged.

In short: the headline quantitative claims that were disseminated are not reliable. The paper is retracted for those reasons. Whether a corrected analysis will support substantially the same qualitative conclusions (with lower percentages) remains to be seen once the revised work appears and is scrutinized. Until then, those specific 2025 numbers should be disregarded.