What is established about the climatic effects of nuclear war, what rests on one modelling group, and what is repeated in the press without a paper behind it.
A research dossier compiled for Grid/84, as the basis of the model at the years after and the validation table in its brief. It exists because a model of nuclear winter is only as good as the numbers it is set against, and those numbers are scattered across four decades of journals, several of which contradict each other.
Method. Primary papers were read as PDFs and grepped for the figures quoted here; where a number appears in a figure rather than a table, that is said. Press summaries were traced back to the paper they come from, and where the trail ends, that is said too.
Limits. The climate and crop results below come from a small number of modelling groups, and several of the cases above five teragrams rest on a single ensemble member of a single Earth system model. That is a property of the field, not of this summary, and it is stated wherever it applies.
DocumentedStated in a named paper at a locatable place: a table, an equation, or a sentence quoted verbatim.
InferredA reading of published figures, or a composite assembled from more than one paper.
ContestedPublished groups disagree, or the figure rests on assumptions the same field disputes.
UntraceableWidely quoted; could not be traced to a paper, or traced and found to be garbled.
How a burning city becomes a mass of soot, and how much of it there is. This is the weakest link in the chain and the one the argument is actually about.
DocumentedThe chain has four links, and Toon's group state a value for each. Area burned per weapon: 13 km² for 15 kt, taken from the Hiroshima firestorm, and scaled linearly with yield — "for different yields, we take the area subject to fire as proportional to the yield", which is not the conventional thermal scaling and is why higher-yield weapons give disproportionately more soot in their accounting. Fuel per person: 1.1 ± 0.4 × 107 g, about eleven tonnes, from Turco et al. 1990 for the developed world. Emission factor: 0.02 g of black carbon per gramme of fuel burned in Toon 2019, 0.016 in Toon 2007. Fraction burned: 1.0, which the paper itself calls an upper limit. Prompt rainout: 20%, also called an upper limit, with a further 10–15% lost before the smoke is above the weather.
DocumentedPopulation is the input, not area. The regression Toon 2007 uses is 0.0011 g/cm² of fuel for every person per square kilometre — the same eleven tonnes a person, seen from above. Their fifty-target averages: India 35 g/cm², Pakistan 28, China 50, the United States 12, with individual Indian and Pakistani targets spanning 12.6 to 94.5.
ContestedThe canonical "5 Tg from 100 weapons" is a rounded choice, not a computed result. Toon 2007 never computes it: its India and Pakistan rows sum to 6.57 Tg, cited later as 6.6. The paper's stated range for an attack of fifty 15 kt weapons on one country is "slightly less than 1 Tg to more than 5 Tg". Robock et al. 2007 then injected the top of that range. With 2016 population data the same hundred-weapon war gives 8.7 Tg. The figure is defensible as a case; it is not the paper's calculated value for that war.
Contested"150 Tg = 4,400 weapons of 100 kt" is a composite of two papers. Toon, Robock and Turco (Physics Today, 2008) computed 180 Tg from 4,400 hundred-kilotonne weapons on cities, about 440 Mt. Robock, Oman and Stenchikov (2007) independently chose 150 Tg, justified as "the entire current global nuclear arsenal". Coupe et al. 2019 joined them, rounding 180 down to 150 by doubling the assumed rainout. The pairing is later than either paper.
| Soot | Weapons | Yield | Source and note |
|---|---|---|---|
| 5 Tg | 100 | 15 kt | Robock et al., ACP 7 (2007) — the top of Toon's 1–5 Tg range, injected at 30°N 70°E on 15 May |
| 16 Tg | 250 | 15 kt | Toon et al., Science Advances 5 (2019), table S1 — India 100 weapons, Pakistan 150 |
| 27 Tg | 250 | 50 kt | Toon 2019 |
| 37 Tg | 250 | 100 kt | Toon 2019 |
| 47 Tg | 500 | 100 kt | Toon 2019 says 250 weapons for this case and for the 37 Tg case, which cannot both be right; Xia 2022 reads it as 250 against each country. Xia's reading is the only arithmetically consistent one. |
| 150 Tg | 4,400 | 100 kt | Toon et al., Physics Today 61 (2008) at 180 Tg; rounded to 150 by Coupe et al. 2019 |
Whether the smoke rises above the weather at all. This is the whole disagreement, and it is not where it is usually said to be.
DocumentedThe climate models are not in dispute. Robock, Toon and Bardeen record that when Reisner's group repeated their simulation with a five-teragram injection, "they reproduced the same climate response". Given the same soot, the same climate follows. The argument is entirely about how much soot there is and how high it goes.
DocumentedReisner's finding. Reisner et al., JGR Atmospheres 123 (2018), coupling a fire model to an Earth-system model, found that "the vast majority of the black carbon never reaches an altitude above weather systems (approximately 12 km)", and that significant global cooling from a regional exchange is "highly unlikely". Their 2019 Reply tabulates it: even at 72 g/cm² of fuel — the top of Toon's own range — they get 1.53 Tg above 12 km, about 6% of the black carbon produced.
DocumentedThe target Reisner modelled. Robock's group obtained his fuel map from the supplier and measured it: 0.14 g/cm² averaged over the 13 km² target area, and 0.91 over the 10 × 10 km domain — suburban Atlanta, "a golf course, playground, and individual houses with large yards". Glasstone and Dolan's firestorm threshold is 4 g/cm². Toon's Indian and Pakistani targets run 12.6 to 94.5.
ContestedThe headline soot masses appear to agree, and that agreement is an artefact. Reisner's ~3.7 Tg of black carbon looks comparable to Toon's 5 Tg. But Reisner converts all the fuel carbon to black carbon — about 1,000 g/kg — against Toon's 20 g/kg, and concedes in his own Reply that this makes the estimate "high by a factor of 10-100". A fuel loading roughly 10 to 70 times too low is combined with an emission factor roughly 50 times too high, and the two errors nearly cancel. The two numbers should not be read as corroborating each other.
DocumentedThe independent test, and it is the most useful paper in the argument. Wagman et al., JGR Atmospheres 125 (2020), from Livermore, swept the fuel loading and held everything else fixed:
| Fuel burned | Black carbon, 100 fires | Outcome |
|---|---|---|
| 1 g/cm² | 0.31 Tg | Removed from the troposphere. No global forcing at all. |
| 5 g/cm² | 1.56 Tg | Reaches the upper troposphere, then lofts itself |
| 10 g/cm² | 3.13 Tg | A fraction injected directly into the stratosphere |
| 16 g/cm² | 5.0 Tg | About 40% direct injection; 1 to 1.5 K of cooling by year three |
DocumentedThe moisture question, settled inside Reisner's own setup. Tarshish and Romps, JGR Atmospheres 127 (2022), found that a dry plume needs a temperature anomaly of 60 K or more at the top of the boundary layer to reach the cold-point tropopause, and firestorm plumes deliver less than half that. Re-running Reisner's case: the dry plume tops out at about 5 km; a moist plume at 70% relative humidity reaches the tropopause at 12 km. Their conclusion is that latent heating is "essential to plume rise", which makes a dry-atmosphere assumption not a conservative simplification but the thing that determines the result.
DocumentedThe observational analogues support self-lofting. The 2017 British Columbia pyrocumulonimbus put 0.3 Tg of carbonaceous aerosol into the stratosphere and it rose from 12 to 23 km in two months (Yu et al., Science 365, 2019). The 2019–20 Australian fires lofted smoke from 14–17 km to 34 km in about 40 days, with an e-folding of 120–150 days below 18 km and 250–350 days above 19 km (Peterson et al., npj Clim. Atmos. Sci. 4, 2021) — the altitude dependence being the cleanest observational support for the lofting-to-longevity chain.
ContestedThe National Academies declined to adjudicate. Its 2025 review found the fire models and the urban fuel data both inadequate to settle the question. Its own survey of American urban fuel loads — Bush et al. 1991 at 1.4 to 2.1 g/cm² — sits far closer to Reisner's figure than to Toon's, which neither camp emphasises. Hiroshima's own fuel loading, the one calibration point everyone uses, is published at 3.9, 10 and 16 g/cm² by three different authorities.
Not the total soot mass, and not primarily the atmospheric residence time. It is fuel loading multiplied by emission factor, which decides whether a firestorm forms, which decides whether the smoke rises. On that chain the independent work since 2020 — Wagman, Redfern, Tarshish and Romps — supports Toon and Robock quantitatively. On the fuel loading itself, the National Academies' own American survey supports neither.
What the general circulation models give for each soot mass, and where they disagree with each other.
| Quantity | 5 Tg | 150 Tg | Source |
|---|---|---|---|
| Peak global surface cooling | −1.25 °C | −7 to −8 °C | Robock, Oman & Stenchikov, JGR 112 (2007), ModelE |
| The same, in WACCM4 | −1.1 K at yr 1, −1.5 K peak | −9.5 K at 1 yr, minimum at 2 | Mills et al. 2014; Coupe et al., JGR Atmos. 124 (2019) |
| After ten years | −0.5 °C | −4 °C | Robock 2007 |
| Peak precipitation change | −10% (ModelE), −6% (WACCM) | −45% (ModelE), −58% (WACCM4) | Robock 2007; Coupe 2019 |
| Peak surface shortwave | −15 W/m² | −100 W/m² | Robock 2007 |
| Soot e-folding time | 6.0 yr | 4.6 yr | Robock 2007. 50 Tg gives 5.5 yr |
| Land against ocean | — | land ~−18 K, ocean ~−6 K | Toon et al. 2019, figs. S6 |
DocumentedThe counter-intuitive ordering of the lifetimes is published and deliberate. A smaller injection lasts longer — six years at 5 Tg against four and a half at 150 — because "the large soot amounts shade some of the lower soot", so only the top of a thick layer is heated enough to keep climbing. Against this, a volcanic sulfate layer has an e-folding of about one year, which the same model reproduces "in excellent agreement with observations".
DocumentedThe growing season, which is the figure a harvest actually cares about. Mills et al., Earth's Future 2 (2014), for the 5 Tg case: "Killing frosts would reduce growing seasons by 10–40 days per year for 5 years", on a definition of consecutive days with minimum temperatures above 0 °C. No equivalent day-count is published for the 50 or 150 Tg cases; those exist only as maps.
DocumentedRegional figures worth having. Robock 2007 for 150 Tg in June–August: cooling of more than 20 °C over large areas of North America and more than 30 °C over much of Eurasia, "including all agricultural regions". In Iowa and Ukraine minimum temperatures "rapidly plummet below freezing and stay there for more than a year"; in Ukraine, more than two.
ContestedThe soot's optical properties differ between the models that produced these numbers. Mass absorption is 7 m²/g in Turco 2000, 7.5 ± 1.2 measured by Bond and Bergstrom, 6.21 in ModelE and 5.48 in WACCM4. Coupe 2019, using coagulating fractal particles, gets a shorter 3.5-year e-folding than Robock's 4.6 while reporting that "the magnitude of the climate response is not reduced".
DocumentedOrganic carbon is excluded throughout. Toon 2007: only black carbon is emitted and simulated, and "our analysis of emissions ignores the other components of the smoke, they will augment the climate effects". Wagman found that co-emitting 15 Tg of particulate organic matter doubles the surface radiative perturbation. Every figure in this section is therefore a floor in one specific respect.
The published food-system results, and the conditions attached to the number everyone quotes.
DocumentedXia, Robock, Scherrer, Harrison, Bodirsky, Weindl, Jägermeyr, Bardeen, Toon and Heneghan, "Global food insecurity and famine from reduced crop, marine fishery and livestock production due to climate disruption from nuclear war soot injection", Nature Food 3 (2022), pp. 586–596. Open access, with the full tables in the supplementary material.
| Soot | Weapons | Yield | Direct fatalities | Without food, end of year 2 |
|---|---|---|---|---|
| 5 Tg | 100 | 15 kt | 27 M | 255 M |
| 16 Tg | 250 | 15 kt | 52 M | 926 M |
| 27 Tg | 250 | 50 kt | 97 M | 1,426 M |
| 37 Tg | 250 | 100 kt | 127 M | 2,081 M |
| 47 Tg | 500 | 100 kt | 164 M | 2,512 M |
| 150 Tg | 4,400 | 100 kt | 360 M | 5,341 M |
ContestedThe five-billion figure carries four conditions that are almost always dropped. It is the partial livestock, no international trade variant: survivors fed at 1,911 kcal per person per day, half the grain that would have gone to animals diverted to people, on the 2010 population of 6.70 billion. It is the lowest of the three no-trade variants. And the paper's own term is "people without food", not deaths — it is not the output of a demographic model.
DocumentedThe structurally important result is what trade does, and does not do. At 5 Tg, restoring trade takes the figure from 409 million to 17 million. At 150 Tg it moves 5,442 million to 5,423. The buffers decide everything at the small end and nothing at the large one.
DocumentedThe optimism ladder. At 5 Tg, with 2010 waste levels and equitable distribution, everyone eats. Cutting household waste from 20% to 10% covers 16 Tg; eliminating it covers 27 Tg; every crop calorie of livestock feed to humans, no household waste and equitable global distribution covers 47 Tg. Nothing covers 150 Tg.
ContestedEvery case above 5 Tg rests on one ensemble member. Xia's climate comes from Coupe 2019 and Toon 2019; three members were run for the control and the 5 Tg case, and one for everything above it. One crop model (CLM5crop), one fishery model (BOATS), one food-balance year (FAO 2010).
DocumentedWhat the paper says it does not model: radioactive contamination, ozone loss and ultraviolet, pollinators, farm adaptation, cropland shifting, inland fisheries, and any population feedback from the deaths it counts.
The effect that arrives late, and changes sign on the way.
DocumentedBardeen, Kinnison, Toon, Mills, Vitt, Xia, Jägermeyr, Lovenduski, Scherrer, Clyne and Robock, JGR Atmospheres 126 (2021). Peak global mean column ozone loss: 25% at 5 Tg and 75% at 150 Tg, with mid- and high-latitude losses above 75% and a reduction lasting fifteen years at the larger injection, twelve at the smaller.
DocumentedThe sign change, which matters for any timeline. At 150 Tg the soot initially compensates for the ozone loss and ultraviolet at the surface is reduced for the first few years. The index begins rising above the control after three years and peaks at eight to nine: above 35 in the tropics for four years, against a present-day map maximum of 26. For a regional war there is no such reprieve — too little soot to shade anything, enough to damage the ozone — so the ultraviolet rises from the start.
InferredThe structural finding worth carrying into any model. For a regional war, cold and ultraviolet damage arrive together; for a global war they arrive in sequence, dark and cold first and the burning years later.
Each traced to where it comes from, and what is wrong with it.
UntraceableAn explicit time-to-peak for the 150 Tg ozone loss. Bardeen 2021 gives "during the first few years" and nothing more precise. The 5 Tg peak timing, two to three years, is explicit.
ContestedToon 2007 carries no soot lifetime at all, deferring lofting and residence time entirely to Robock et al. 2007. It is nonetheless cited for e-folding times.