Study pack · Statistics and applied mathematics

Warming and drought severity

Climate science · as it stood on 13 May 2026 · run unseen-2026-11-07/drought-severity · Claude Opus 5.5

Has a serious holeNo break

The index measures demand for water, not proven shortage of it.

claims: 6
holes: 6
lenses: 3
records cited: 10 of 13
retracted precedents: 2

What happened, 2 September 2026. Its authors retract it over calculation errors, and say the conclusion survives. The authors retracted the paper after being told of errors and finding more: study-wide averages were not area-weighted, one figure used an older dataset, a timestepping error affected regional drought areas, and the masking was not clearly explained. They say the percentages change but the conclusion does not, and that the corrected figure will be slightly lower. The revised paper has yet to pass peer review, so this one is not scored either way. Asked first, the model said it didn’t know how it ended.

The idea, as it stood

A June 2025 paper in Nature develops an ensemble of high-resolution global drought datasets for 1901 to 2022 and finds an increasing trend in drought severity worldwide. It concludes that atmospheric evaporative demand has increased drought severity by an average of 40% globally. It reports that in 2018 to 2022 the area in drought expanded by 74% on average compared with 1981 to 2017, with evaporative demand contributing 58% of that increase, and that in 2022, 30% of global land was in moderate to extreme drought, 42% of it attributed to increased evaporative demand.

An ensemble of high-resolution global drought datasets for 1901 to 2022 shows an increasing trend in drought severity worldwide.empirical · supported
Increased atmospheric evaporative demand has raised drought severity by an average of 40% globally.causal · contested
In 2018 to 2022 the area in drought expanded by 74% on average compared with 1981 to 2017.empirical · unsupported
Evaporative demand contributed 58% of that 2018 to 2022 increase in drought area.causal · contested
In 2022, 30% of global land was in moderate to extreme drought, and 42% of that area is attributed to increased evaporative demand.causal · contested
Implied premise: drought severity as measured by these datasets, which count evaporative demand, reflects real water shortage on land rather than only a greater potential for water loss.definitional · contested

The holes, and who found them

More than one lens found a hole in claim 2, each on its own. That is a stronger signal than one lens alone.

#	Hole	Lens	How serious	Claim	Status
1	The index measures demand for water, not proven shortage of it	Richard Feynman	serious	6	Open
2	Evaporative demand is partly an effect of the drought it is credited with	Norbert Wiener	serious	4	Open
3	The demand share is an accounting convention, not a measurement	Claude Shannon	serious	5	Open
4	The 40% inherits the formula and data behind evaporative demand	Claude Shannon	serious	2	Open
5	A percentage rise in severity depends on where the drought line is drawn	Richard Feynman	minor	2	Open
6	One five-year window against a 37-year average	Claude Shannon	minor	3	Open

What it rests on

Land–atmosphere feedbacks exacerbate concurrent soil drought and atmospheric aridityZhou et al. · 2019 · Proceedings of the National Academy of Sciences · cited on claim 4, 6 · doi:10.1073/pnas.1904955116
Uncertainties in historical changes and future projections of drought. Part I: estimates of historical drought changesDai & Zhao · 2016 · Climatic Change · cited on claim 2 · doi:10.1007/s10584-016-1705-2
Global drought trends and future projectionsVicente-Serrano et al. · 2022 · Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences · cited on claim 6 · doi:10.1098/rsta.2021.0285
Soil moisture–atmosphere feedback dominates land carbon uptake variabilityHumphrey et al. · 2021 · Nature · cited on claim 4 · doi:10.1038/s41586-021-03325-5
Comment on “Characteristics and trends in various forms of the Palmer Drought Severity Index (PDSI) during 1900–2008” by Aiguo DaiVicente-Serrano et al. · 2011 · Journal of Geophysical Research Atmospheres · cited on claim 6 · doi:10.1029/2011jd016410
Drought Assessment Based on Data Fusion and Deep LearningLi et al. · 2022 · Computational Intelligence and Neuroscience · retracted · cited on claim 2 · doi:10.1155/2022/4429286
Flash Drought Response to Precipitation and Atmospheric Evaporative Demand in SpainNoguera et al. · 2021 · Atmosphere · cited on claim 6 · doi:10.3390/atmos12020165
Assessing Potential Evapotranspiration Methods in Future Drought Projections across CanadaTam et al. · 2023 · ATMOSPHERE-OCEAN · cited on claim 2 · doi:10.1080/07055900.2023.2288632
Thermodynamic controls on vapor pressure deficit during droughtsGhausi et al. · 2026 · cited on claim 4 · doi:10.5194/egusphere-egu26-3487
Drought Severity in a Changing ClimateVicente-Serrano et al. · 2017 · CRC Press eBooks · cited on claim 2, 6 · doi:10.1201/9781315404219-16
Weakening vegetation control on global terrestrial evapotranspiration in a warmer worldLi et al. · 2026 · Communications Earth & Environment · retrieved, not cited · doi:10.1038/s43247-026-03372-8
Global Changes in Drought Conditions Under Different Levels of WarmingNaumann et al. · 2018 · Geophysical Research Letters · retrieved, not cited · doi:10.1002/2017gl076521
Land use land cover change as a casual factor for climate variability and trends in the Bilate River Basin, EthiopiaBikeko & Venkatesham · 2024 · PLoS ONE · retracted · retrieved, not cited · doi:10.1371/journal.pone.0311961

The tests that settle it

The deciding test · weeks of work

Check the attributed drought against water on the ground. For 2018–2022, sort land cells into three groups: in drought under the index both with observed evaporative demand and with demand held at the reference level used for the attribution (rain-driven drought); in drought only with observed demand (demand-attributed drought); and in drought under neither. Compare independently measured anomalies across the groups: soil moisture from in-situ networks and satellites, streamflow at near-natural gauges, terrestrial water storage, and vegetation greenness, kept separate so that plant stress from dry air is not mistaken for water shortage.

It fails if: The demand-only group is no drier in soil moisture and streamflow than the no-drought group (difference within measurement and ensemble uncertainty), meaning the new drought area credited to evaporative demand corresponds to no extra water shortage on land.

For hole	Test	Result
1	For each land cell over 1981–2022, compare the demand-driven part of the index trend with trends in independently measured soil moisture (in situ and satellite), streamflow at near-natural gauges and terrestrial water storage where available. Count against the premise if, in most cells where the index attributes most of its drying to demand, those measures show no drying trend (a trend indistinguishable from zero, or wetting).	Proposed
2	Recompute the demand share of the 2018–2022 increase using demand that cannot respond to the land: from model runs in which soil moisture is not allowed to feed back on the air (land–atmosphere coupling experiments of the kind named in E3), or after removing the land-surface part of vapour-pressure deficit with a thermodynamic split like the one E12 proposes. Count against C4 if the feedback-free share falls below 50%, which would make rainfall, not demand, the larger contributor.	Proposed
3	Compute the demand share of the 2022 drought area and of the 2018–2022 increase three ways: demand held at its reference with observed rain; rain held at its reference with observed demand; and the joint part split evenly between them. Count against the 42% and 58% as findings if the first two differ by more than the ensemble's own spread, or if the even split leaves demand with less than half of the 2018–2022 increase.	Proposed
4	Report the 40% separately for each evaporative-demand formula in the ensemble (or recompute with a second, physically based formula if only one was used), and separately for 1981–2022 versus 1901–2022. Count against the figure if members using radiation, wind and humidity give less than half the effect of temperature-based members, or if the spread across members is wider than the 40% itself.	Proposed
5	Recompute the demand-driven severity increase with the drought line moved across the index's moderate, severe and extreme boundaries, and with frequency and duration as alternative severity measures. Count against 40% as a general effect size if it moves by more than 10 percentage points across those choices; the honest statement is then a range tied to its threshold.	Proposed
6	Using the index with the demand trend removed, compute for every five-year window since 1938 the change in drought area against the 37 years before it. Count against reading C3 as a trend if a 74% rise sits within the central 90% of those swings.	Proposed

Proposed tests that haven’t run are the agenda for what comes next. Download them as a draft pre-registration, ready to edit and register on OSF.

Two answers, matched blind

The same model answered twice: once plainly, as a standard AI, and once through Feynman. Both answers’ findings were matched blind, without the ending.

The lenses found, the standard AI missed

The demand share is an accounting convention, not a measurement Shannon's question of which distinctions survive transmission led the lens to argue that an all-or-nothing threshold crossing loses which input caused it. It cites no records. The standard AI traced the split's overlap to demand rising partly because of drought, and said the percentages look bigger and more comparable than they are.

The standard AI found, the lenses missed

Very dry land can score as drought with little water left to lose The standard AI drew on cited SPEI and evaporative-demand papers, reasoning that in very dry cells standardising turns a small steady demand rise into large anomalies, and arid land is about a fifth of all land. Feynman instead said the index measures demand, not proven shortage, and percentage rises depend on where the drought line sits.

The finding

The datasets very likely do show worsening drought as they define it, but the 40%, 58% and 42% rest on a step from evaporative demand to water shortage that is not shown to have been checked, on a demand term that drought itself pushes up, and on unstated accounting choices.

It partly held: Its authors retract it over calculation errors, and say the conclusion survives. What would overturn the review’s reading: The demand-only group is no drier in soil moisture and streamflow than the no-drought group (difference within measurement and ensemble uncertainty), meaning the new drought area credited to evaporative demand corresponds to no extra water shortage on land.

Cite it, disclose it, reproduce it

A paragraph for your methods section

We used Feynman (feynman.works), an AI research-review tool running on Claude Opus 5.5, to critique “Warming and drought severity” on 2026-11-07 (run unseen-2026-11-07/drought-severity). It reconstructed 6 claims, retrieved 13 published records dated no later than 2026-05-13, and raised 6 objections through reasoning frameworks drawn from the published work of Richard Feynman, Norbert Wiener and Claude Shannon. Each citation in its output was checked against the records it retrieved, and we read every source we relied on. Its objections informed our analysis; the conclusions, and any errors, are our own.

The lenses

Richard FeynmanPhysics & scientific method · 1948–1980What experiment would distinguish this explanation from a convincing story?
Norbert WienerCybernetics & automation · 1943–1964How does the output change the system that produced it?
Claude ShannonInformation theory · 1937–1961Which distinctions survive transmission, and which are lost?

BibTeX

@misc{feynman-unseen-2026-11-07-drought-severity,
  title = {{Feynman review: Warming and drought severity}},
  author = {{Feynman Works}},
  year = {2026},
  howpublished = {\url{https://www.feynman.works/pack/drought-severity}},
  note = {Run unseen-2026-11-07/drought-severity, Claude Opus 5.5, 2026-11-07}
}

@article{zhou2019land,
  title = {{Land–atmosphere feedbacks exacerbate concurrent soil drought and atmospheric aridity}},
  author = {Zhou and others},
  year = {2019},
  journal = {Proceedings of the National Academy of Sciences},
  doi = {10.1073/pnas.1904955116}
}

@article{dai2016uncertainties,
  title = {{Uncertainties in historical changes and future projections of drought. Part I: estimates of historical drought changes}},
  author = {Dai and Zhao},
  year = {2016},
  journal = {Climatic Change},
  doi = {10.1007/s10584-016-1705-2}
}

@article{vicente2022global,
  title = {{Global drought trends and future projections}},
  author = {Vicente-Serrano and others},
  year = {2022},
  journal = {Philosophical Transactions of the Royal Society A Mathematical Physical and Engineering Sciences},
  doi = {10.1098/rsta.2021.0285}
}

@article{humphrey2021soil,
  title = {{Soil moisture–atmosphere feedback dominates land carbon uptake variability}},
  author = {Humphrey and others},
  year = {2021},
  journal = {Nature},
  doi = {10.1038/s41586-021-03325-5}
}

@article{vicente2011comment,
  title = {{Comment on “Characteristics and trends in various forms of the Palmer Drought Severity Index (PDSI) during 1900–2008” by Aiguo Dai}},
  author = {Vicente-Serrano and others},
  year = {2011},
  journal = {Journal of Geophysical Research Atmospheres},
  doi = {10.1029/2011jd016410}
}

@article{li2022drought,
  title = {{Drought Assessment Based on Data Fusion and Deep Learning}},
  author = {Li and others},
  year = {2022},
  journal = {Computational Intelligence and Neuroscience},
  doi = {10.1155/2022/4429286}
}

@article{noguera2021flash,
  title = {{Flash Drought Response to Precipitation and Atmospheric Evaporative Demand in Spain}},
  author = {Noguera and others},
  year = {2021},
  journal = {Atmosphere},
  doi = {10.3390/atmos12020165}
}

@article{tam2023assessing,
  title = {{Assessing Potential Evapotranspiration Methods in Future Drought Projections across Canada}},
  author = {Tam and others},
  year = {2023},
  journal = {ATMOSPHERE-OCEAN},
  doi = {10.1080/07055900.2023.2288632}
}

@misc{ghausi2026thermodynamic,
  title = {{Thermodynamic controls on vapor pressure deficit during droughts}},
  author = {Ghausi and others},
  year = {2026},
  doi = {10.5194/egusphere-egu26-3487}
}

@article{vicente2017drought,
  title = {{Drought Severity in a Changing Climate}},
  author = {Vicente-Serrano and others},
  year = {2017},
  journal = {CRC Press eBooks},
  doi = {10.1201/9781315404219-16}
}

Run unseen-2026-11-07/drought-severity on Claude Opus 5.5. The review is reproducible from its record: the claims, the records retrieved before 13 May 2026, and each lens’s card. https://www.feynman.works/pack/drought-severity