Nuclear Winter Modelling

Frontier If a regional nuclear exchange puts about 5 Tg of black carbon into the stratosphere — a figure the source models dispute by more than an order of magnitude — the modelling says the resulting famine would kill people at roughly a hundred times the rate of the explosions, and knowing which way that dispute resolves is the whole prize here. The direct blast and fallout casualties of a hundred-weapon India–Pakistan exchange are counted in the tens of millions; the climate-and-crop models put global famine deaths from the same event at more than two billion (Xia and colleagues, 2022). That multiplier, not the explosion, is what a decisive nuclear-winter model would fix in place.

Frontier The quantity in dispute is small, specific, and cheap to constrain relative to what rides on it. One thing — the mass, the black-carbon fraction and the injection height of the smoke that actually reaches the stratosphere from burning cities — separates the Rutgers projection of about 5 Tg and multidecadal cooling from the Los Alamos calculation of far less soot and a minor, short-lived perturbation. The climate physics downstream of that number is comparatively settled: a 5 Tg injection cools global average surface temperature by about 1.6 K for more than a decade and shortens growing seasons by 10 to 40 days a year (Mills and colleagues, 2014). The uncertainty lives almost entirely in the first step.

Established Nature has already run about one five-hundredth of the experiment, unplanned. The August 2017 Pacific Northwest pyrocumulonimbus event lofted roughly 0.3 Tg of smoke into the stratosphere, where it self-lofted from 12 to 23 km and stayed detectable for more than eight months — the first clean observation of the mechanism the entire theory turns on, at about 1/500 the mass of the 150 Tg superpower scenario.

Speculative The gain is distributed oddly: it accrues to everyone and to no institution. A sharper model changes deterrence reasoning, grain-reserve policy and civil-resilience planning for states that would never be struck, because a famine seeded by a distant war is a global event. Whether that value is ever realised turns on the source-term work in section 7, which no government or laboratory has funded as a coordinated programme.

Established Nuclear winter is a claim about smoke, not about explosions. The proposition, sketched by Crutzen and Birks in 1982 and modelled end-to-end the next year by Turco, Toon, Ackerman, Pollack and Sagan — the TTAPS study — is that mass fires ignited across cities and industrial areas would loft soot, meaning black carbon, into the stratosphere, where no rain falls to wash it out; the soot absorbs sunlight, warms the stratosphere, cools the surface beneath, and drives agriculture into failure across a hemisphere. This brief assesses the atmospheric physics of that chain. The arms-control and deterrence response to it belongs to Existential Risk Governance; the deliberate-injection mirror image, putting aerosol into the stratosphere on purpose, belongs to Climate Engineering.

Established The causal chain has five links, and they are not equally sound. A weapon starts fires (well understood); the fires in dense fuel merge into a firestorm that generates its own convective column (understood, and the point where the source studies diverge); the column carries soot above the tropopause (observed in wildfires, disputed in magnitude for cities); the black carbon absorbs sunlight and self-lofts higher, extending its lifetime from weeks to years (observed at small scale); and the reduced surface sunlight collapses crop yields (modelled, with the crop step adding its own uncertainty on top). Frontier A brief that treats the whole chain as one confidence level is the failure mode this subject invites, and most public summaries commit it.

Established Self-lofting is the mechanism that makes stratospheric smoke a climate agent rather than a nuisance. Ordinary aerosol injected at 12 km would sediment and mix out within weeks. Black carbon is different: it absorbs solar radiation, heats the air parcel around it, and rises — a soot layer becomes its own hot-air balloon. In the 2017 event this carried a plume from 12 to 23 km over about two months (Yu and colleagues, 2019). Frontier It is this positive feedback, not the initial injection height, that turns a transient smoke cloud into a multi-year veil, and it is why the nuclear-winter argument is fundamentally about black carbon specifically and not smoke in general; a smoke that is mostly organic behaves quite differently.

Frontier The scenarios are labelled by soot mass, and the labels are model inputs, not measurements. The literature runs a ladder: about 5 Tg of black carbon for a hundred-weapon India–Pakistan exchange, rising through 16, 27 and 47 Tg for larger regional wars, to 150 Tg for a full United States–Russia exchange (Toon, Xia and colleagues). Every one of those numbers is the product of assumptions about how much burnable material a city holds, what fraction becomes airborne soot, and how high it goes. The climate models take the soot mass as given; the argument this brief is about happens before the climate model is ever switched on. Established By convention the scenarios are quoted as pure black carbon, which is why the natural pyroCb events — only about 2 per cent black carbon — validate the transport physics far better than they validate the radiative forcing per unit mass.

Speculative The word “winter” oversells and misdirects. Even the largest scenarios do not produce a literal frozen planet; they produce a cooling of a few kelvin in the global average, concentrated in growing-season temperature drops and sunlight loss that are lethal to harvests without being lethal to standing outside. The human catastrophe in the models is famine, not frost, and the distinction matters because it is the crop response, not the temperature, that carries most of the projected death toll. Frontier That framing also relocates the deepest uncertainty from the climate model to the agricultural model bolted onto it.

Established The 2017 Pacific Northwest pyroCb is the anchor observation, and it verifies the commission’s figure. Yu and colleagues (Science, 2019) found the August 2017 British Columbia firestorm injected about 0.3 Tg of carbonaceous aerosol — roughly 2 per cent black carbon and 98 per cent organic material — at 12 to 13 km, that the plume then self-lofted to 23 km within about two months, descended afterward at roughly 0.5 km per month, and remained detectable by satellite for more than eight months. Crucially, the observed stratospheric lifetime was about 40 per cent shorter than a standard model predicted, because reactive oxygen in the stratosphere degraded the organic fraction faster than expected — a first correction the natural experiment handed back to the modellers. Established That the observed lifetime came in shorter, not longer, is itself informative: it bounds how long a comparable natural layer can last, and it is the kind of number no pre-event model got exactly right.

Established The 2019–2020 Australian “Black Summer” is the larger analogue and points the same way. Peterson and colleagues (npj Climate and Atmospheric Science, 2021) documented an outbreak of 38 pyroCb pulses in 18 sub-events; more than half injected smoke directly into the stratosphere; the plume reached 34 km within 40 days and left stratospheric aerosol extinction about 150 per cent above background for at least 15 months. Cumulative stratospheric injection was 0.3 to 1.1 Tg — two to three times the 2017 event. Two independent natural firestorms, on two continents, both self-lofted and both persisted; the mechanism is no longer in dispute. Frontier What they do not settle is scale: both were an order of magnitude or more below even the smallest war scenario, and self-lofting behaviour need not extrapolate linearly to a far blacker, far larger injection.

Established The radiative and chemical physics downstream of stratospheric black carbon is also measured, not just modelled. The 2017 plume warmed the stratosphere locally and was associated with ozone losses of up to 50 per cent over parts of Canada; the Australian plume perturbed stratospheric chemistry hemisphere-wide. These are the same physical processes the nuclear-winter models invoke, observed at small scale. Where the observations bite hardest is on lifetime and self-lofting; where they say least is on the radiative forcing of a pure-soot layer, because wildfire smoke is mostly organic, not black, carbon.

Frontier The climate response to a specified soot mass is the best-agreed part of the modelling. Mills and colleagues (2014), using the Community Earth System Model with WACCM, put a 5 Tg black-carbon injection at a maximum global-average cooling of 1.6 K in year five, precipitation down about 6 per cent in year one and 9 per cent by year five, surface temperatures depressed for more than 25 years, ozone loss of 20 to 50 per cent over populated areas, and growing seasons shortened by 10 to 40 days a year for five years. Coupe and colleagues (2019) ran the 150 Tg United States–Russia case in the same model family with broadly consistent, larger effects. Established Different modern models given the same soot agree to within a factor that is small next to the disagreement over the soot itself — which is the single most important fact for a reader deciding where the real uncertainty lives.

Frontier The crop-and-famine step is where the largest death tolls are produced, and it is a second uncertain model stacked on the first. Xia and colleagues (Nature Food, 2022) coupled the climate output to the CLM5crop agricultural model and the BOATS fisheries model. A 5 Tg India–Pakistan injection cut global average crop calories about 7 per cent in years one to five; a 47 Tg injection cut them roughly 50 per cent; the 150 Tg case cut them about 90 per cent three to four years out. Translated into deaths from starvation, more than two billion people could die after the smaller regional war and more than five billion after the superpower war. The authors state their own central caveat plainly: the projections rest on one Earth system model, with only a single ensemble member for every case above 5 Tg.

Frontier The source term — how much soot, how black, how high — is the genuinely contested link, and the disagreement is between named groups. The Rutgers–Colorado group (Robock, Toon and colleagues) defends roughly 5 Tg for the hundred-weapon case; a Los Alamos group (Reisner and colleagues, 2018) recomputed the source with a detailed fire model and found far less — on the order of a few Tg at most reaching altitude, with much lower stratospheric retention and, they argue, little of the self-lofting that extends lifetime. The two camps have exchanged a formal comment and reply in the Journal of Geophysical Research (2019). Frontier A secondary synthesis puts Reisner’s worst-case regional figure near 3.2 Tg against the Rutgers 5 Tg for the comparable case — but that near-agreement on the top-line number hides a far larger split on stratospheric retention and lifetime, which is where the climate effect is actually made or unmade. Everything the headline famine numbers depend on is decided in this exchange, and it is not settled.

Established Set the human cost down plainly and sourced. The figures in circulation — more than two billion dead from a regional war, about five billion from a global one (Xia and colleagues, 2022; amplified by the IPPNW/ICAN Nuclear Famine report, 2022) — are projections from a contested model chain, not measured outcomes, and this brief flags them as such throughout. They are reported here because a reader is owed the stakes; they are not reported as settled fact, and the interested parties who quote them without their error bars are named in section 10.

Frontier How much burnable material does a modern city actually hold, and how much of it becomes stratospheric soot? This is the master question, because it sets the soot mass every other number multiplies. Robock and colleagues (2019) calculate fuel loadings of 12.6 to 94.5 g/cm² for top Indian and Pakistani urban targets; Reisner and colleagues chose a target the critics compute at 0.14 g/cm² — 15 to 110 times smaller. The firestorm threshold is often placed near 4 g/cm², so the two groups are not disagreeing about physics so much as about which city, and which part of it, is burning. Speculative No published fuel census resolves the gap to the standard the dispute needs, so both numbers remain defensible and the field cannot adjudicate between them on evidence.

Frontier Does a nuclear-ignited fire in a modern city reliably form a firestorm at all? Reisner and colleagues argue that for the smaller device size used in the regional-war scenarios, the blast-damaged zone overlaps the fire zone, scattering fuel and venting heat before a self-organising convective column can form. Critics counter that this reasoning weakens as fire radius outgrows blast radius, and that it depends on the low fuel loading already in dispute. Whether firestorms form is partly an empirical question about 20th-century mass fires and partly a modelling question nobody has resolved on an agreed target. Established The 20th-century mass fires that did form — Hamburg, Dresden, Hiroshima — are real and documented, but they burned denser, more flammable cities than the sparse target Reisner modelled, which is the critics’ core point.

Frontier Is dry convection enough to reach the stratosphere, or is latent heat required? This is the frontier question the natural pyroCb events have most sharpened. Tarshish and Romps (2022) found that condensing water vapour — latent heating — is required for firestorm plumes to reach the stratosphere, and that a dry simulation systematically under-lofts. Reisner’s source calculation was dry; Robock and colleagues (2019) make exactly this objection, that omitting pyrocumulonimbus latent heat “eliminated a major source of buoyancy.” The 2017 British Columbia plume reached the stratosphere despite precipitation in the cloud, which the Rutgers group reads as direct evidence against high rainout removal. Frontier This is the strongest single way the natural pyroCb events bear on the dispute: they are wet convective columns that lofted soot to the stratosphere anyway, which is exactly what a dry, high-rainout source model says should not happen.

Frontier How much soot rains out before it reaches altitude? Reisner’s lower stratospheric injection comes partly from higher modelled washout; Robock and colleagues respond that “there is no evidence for that assumption” and point again to the pyroCb observations, where wet convective columns delivered soot to the stratosphere anyway. Rainout is a single tunable process with an outsized effect on the answer, and no observation yet pins it down for a city-scale fire.

Frontier What is the black-carbon fraction of the smoke, and does it survive? Nuclear-winter scenarios treat the injected mass as soot — strongly absorbing black carbon. The natural pyroCb plumes were only about 2 per cent black carbon; the rest was weakly absorbing organic material that the 2017 observation showed degrades faster than models assumed. A city fire burning plastics, fuels and asphalt may be blacker than a forest fire, but by how much is unmeasured, and it drives both the self-lofting rate and the radiative forcing per Tg. Speculative The assumption that urban soot is essentially pure black carbon is plausible and untested, and it is one of the assumptions most favourable to a large climate effect.

Speculative Would a real war even resemble the idealised scenarios? The models inject a clean pulse of soot on 1 January, uniformly over target regions, as pre-aged hydrophilic black carbon. A real exchange would be seasonally timed, spatially clumped, and chemically messy, and the recent Cheung and colleagues (2025) run in the Canadian model found an unexpected transient summer land warming — a “whiplash” — after a winter injection of 12 to 24 Tg, a reminder that even the downstream climate response holds surprises when a new model is asked the question.

Established There is no shared, fully specified urban target set, and this is the field’s central bottleneck. The Rutgers and Los Alamos groups reached order-of-magnitude-different soot masses partly because they modelled different cities with different fuel assumptions. Until one target — a named city, a fuel inventory, a device placement drawn from the open literature — is agreed and handed to every group, the disagreement cannot be adjudicated, only restated. Frontier This is a coordination failure, not a physics wall, which is what makes it both frustrating and, in principle, cheap to fix.

Frontier The source models and the climate models are run by different communities with different tools. High-resolution fire and plume dynamics live in combustion and cloud-resolving codes; the multi-year climate response lives in Earth system models; the crop response lives in agricultural models. Each hand-off passes a number, not a distribution, so uncertainty is truncated at every stage and the final famine figure carries error bars far tighter than the inputs deserve. Established Propagating a distribution rather than a point estimate across the three hand-offs is standard practice elsewhere and has never been done end-to-end here.

Established The famine projection rests on a single climate model and, above 5 Tg, a single ensemble member. Xia and colleagues say so themselves. Internal climate variability alone can move regional crop outcomes; running one realisation of one model means the spread between plausible worlds is unmeasured for exactly the scenarios that produce the largest death tolls. This is the cheapest gap to close and it has not been closed.

Frontier In-situ measurement of stratospheric smoke barely exists. The pyroCb observations are overwhelmingly remote-sensed — satellite extinction, lidar, balloon. Direct aircraft sampling of a fresh stratospheric smoke layer, which would return the black-carbon fraction, particle size distribution and optical properties the models must assume, has essentially never been done for a major pyroCb, because the aircraft and the tasking authority are not pre-positioned when one erupts. Frontier Each pyroCb that passes unsampled is a validation opportunity permanently lost, and there have been several since 2017.

Established The historical record is thin and cannot be replayed. The one real data point — the mass fires of the Second World War, Hamburg, Dresden, Hiroshima — predates instrumentation, and Robock and colleagues concede it is “not possible to formally detect a cooling signal from World War II smoke.” The field therefore leans on wildfire analogues that are wetter, less black, and smaller than the scenarios of concern, and on models that cannot be validated end-to-end against any event that has happened. Speculative This is an unusual epistemic position for a quantitative science: the thing being predicted has, by design, never been observed, and never can be without the catastrophe itself.

Frontier Every result here depends on wildfire pyroCb science, which is advancing for reasons unrelated to nuclear war. The natural analogues are supplied by an intensifying fire regime; the same 2023 Canadian season that drove the observations discussed in Wildfire Systems is what gives this field its validation data. Established The dependency runs one way: nuclear-winter modelling now waits on fire and plume observations it does not itself generate, and a warming climate is, grimly, improving the supply of natural analogues.

Established The climate step depends on the stratospheric-aerosol modelling stack shared with solar geoengineering. The radiative transfer, microphysics and transport used to age and spread a soot layer are the same tools studied deliberately under Climate Engineering; improvements there propagate directly here, and the two fields are studying, from opposite motives, the same physical object: an absorbing or scattering layer placed above the weather.

Frontier The crop projection depends on agricultural models validated against a world that has never gone dark. CLM5crop and its peers are calibrated on observed growing conditions; a sudden multi-kelvin cooling with sunlight loss and shortened seasons is an extrapolation far outside their training range, and the trade-and-storage response of a panicked world is not in them at all. Speculative Human responses — hoarding, export bans, rationing, substitution — could push the death toll either way, and no model in this literature represents them; the famine number inherits every weakness of both the climate model above it and the crop model within it.

Established The whole enterprise depends on Earth-system modelling capacity documented under Earth-System Digital Twins. A proper multi-model, multi-member intercomparison is compute-limited and coordination-limited, not idea-limited; the ensembles that would put honest error bars on the famine projection are the same infrastructure the climate community is building for other reasons.

Speculative Interpretation depends on scenario inputs no scientist owns. Which cities, how many weapons, what season — these are strategic assumptions, and a modeller who chooses them is making a policy claim in the guise of a boundary condition. The dependency the field least likes to name is on the plausibility of the war it is modelling.

Frontier The decisive experiment is a model-intercomparison on one agreed, fully specified urban target set, in which soot mass, black-carbon fraction and injection height are treated as outputs rather than tuned inputs, paired with aircraft sampling of the next large pyrocumulonimbus. Hand every group the same city, the same fuel inventory drawn from the open literature, and the same device placement, and require each to compute — not assume — how much soot of what blackness reaches what altitude; then run the resulting source terms through a common multi-model, multi-member climate-and-crop ensemble. This is the one result the brief judges most able to change its assessment, because it converts a decade of restated disagreement into a single falsifiable number with an honest spread. Frontier The design is unglamorous and the tools already exist; what is missing is a body willing to commission it and bind the groups to a common target.

Frontier The observational half is a natural experiment that recurs, needing instrumentation rather than construction. Large pyroCbs now occur most years. A pre-positioned rapid-response aircraft with black-carbon, size-distribution and optical-property instruments, cleared to sample a fresh stratospheric smoke layer within days, would return exactly the quantities the source models must otherwise guess: how black the smoke is, how fast it self-lofts, and how quickly it degrades. Nature keeps offering the experiment; no agency has funded the aircraft or the tasking authority to take it. Established A single well-sampled large pyroCb would do more to constrain the black-carbon fraction and self-lofting rate than another decade of dueling source models.

Frontier A resolved fuel-loading census would settle the fight beneath the fight. The 0.14 versus 12.6–94.5 g/cm² gap between the two groups is, at bottom, a disagreement about how much combustible material a real city block contains. An open, auditable urban fuel inventory for a representative target — buildings, contents, vehicles, fuel stores — is a measurement anyone could make and no one has published to the standard the dispute requires.

Frontier A wet-versus-dry firestorm plume test is now specifiable. Tarshish and Romps showed latent heating is required for plumes to reach the stratosphere; the experiment that follows is a cloud-resolving intercomparison of the same firestorm with and without moist convection, benchmarked against the 2017 and 2019–2020 plume rise rates, to quantify how badly a dry source model under-lofts. This needs only compute and the observational benchmarks already in hand.

Frontier A single-model, single-member famine projection can be upgraded immediately. Repeating Xia and colleagues across several Earth system models and several ensemble members per scenario — standard practice everywhere else in climate science — would put real error bars on the two-billion and five-billion figures. Nothing prevents this but funding and coordination; the models exist and the scenarios are published.

Speculative What cannot be run is the whole chain end-to-end. No one will detonate weapons over a city, and the Second World War firestorms left no usable climate record. The field is therefore permanently dependent on decomposing the problem into steps each of which can be tested against something — a wildfire, a fuel census, a model benchmark — and on being honest that the sum of well-tested steps is still an untested whole.

Established The natural-experiment half is already producing corrections. The 2017 plume’s 40-per-cent-shorter-than-modelled lifetime, and the ozone perturbations from both events, are results the wildfires handed to the modellers without anyone scheduling them; the decisive programme above is the deliberate version of what nature is doing by accident.

Established The instruments to run the decisive experiment mostly exist. Cloud-resolving fire-plume models, Earth system models with interactive stratospheric aerosol, and airborne black-carbon photometers and optical particle counters are all mature and in use for other problems. The engineering task is integration and tasking, not invention: a common source-term protocol, a shared target specification, and a standing agreement to fly instruments into the next stratospheric smoke plume. Frontier None of these is a research breakthrough; each is a decision someone has to make and pay for.

Frontier Rapid-response stratospheric sampling is the one genuine hardware gap. Reaching a fresh smoke layer at 15 to 30 km within days of a pyroCb needs a high-altitude aircraft or balloon fleet on standby with pre-cleared flight corridors — the same class of capability discussed under Climate Engineering for aerosol monitoring. It is buildable now; it is simply not funded or pre-positioned, so every pyroCb since 2017 has been observed mainly from orbit.

Established The compute for a proper ensemble is ordinary by climate-science standards. A multi-model, multi-member intercomparison of a handful of soot scenarios is a routine ask for a national modelling centre; the reason it has not happened is that no centre has been commissioned to do it, not that the machines are lacking. The engineering constraint here is institutional, and it is picked up in section 10 and in the timelines under section 13.

Frontier An open, auditable urban fuel inventory is a data-engineering deliverable. Building-by-building material and contents estimates for a representative city, assembled from public construction, census and infrastructure data, would be laborious but unglamorous and entirely feasible — and it would remove the single assumption that most divides the two groups. Speculative That such a census does not yet exist to publication standard is the clearest sign the field is under-resourced rather than stuck.

Established The nearest physical sibling is deliberate stratospheric aerosol injection. Climate Engineering studies the same object — an aerosol layer above the weather — from the opposite intent, and shares the radiative-transfer, microphysics and transport tools; the 2017 and 2020 pyroCb events are treated in both fields as the best natural analogues yet observed. Geoengineering Governance owns the question of who is allowed to change the stratosphere on purpose, which is the mirror of who is accountable for changing it by war.

Frontier Wildfire science is now the supplier, not merely a cousin. Wildfire Systems owns the pyroCb source events this brief depends on for validation; the intensifying fire regime described there is why the natural-experiment data now exist at all. Compound Climate Hazards carries the framework for the simultaneous, cascading failures a global harvest collapse would represent. Speculative The unusual export from this brief to wildfire science is a warning: if a warming world keeps generating larger pyroCbs, the natural analogues will eventually approach the smaller war scenarios in mass, an uncomfortable convergence neither field sought.

Established The impact-winter physics generalises to other sudden-darkening events. Planetary Defense models the impact winter from an asteroid strike using the same soot-and-dust radiative logic; a large-eruption volcanic winter is the historical natural case. Established The 1815 Tambora eruption and its 1816 “year without a summer” are the closest real, instrumented-adjacent analogue for a sudden sunlight-and-temperature shock to agriculture, and they killed by famine, not frost. Nuclear winter is one member of a family of abrupt-cooling scenarios, and the crop and famine machinery is shared across all of them.

Frontier The consequence side connects to resilience and strategy briefs. Civilization Resilience Planning owns the grain-reserve and food-shock response that a credible famine projection would inform, and Military AI and Strategic Stability carries the deterrence context into which any hardened climate number would land.

Established The two poles of the science sit in institutions with visible interests, and the reader should weigh both. The concern case is led from Rutgers and the National Center for Atmospheric Research (Robock, Toon and colleagues), a group that has argued for decades that nuclear winter strengthens the case for disarmament. The low-estimate case comes from Los Alamos National Laboratory (Reisner and colleagues), a United States nuclear-weapons laboratory. Neither affiliation makes its science wrong, and a weapons laboratory producing a lower climate estimate is not the obvious conflict it might first appear; but a reader owes each result the same scrutiny, and the brief names the affiliations rather than hiding them. Frontier The healthiest reading is that an adversarial pair of well-resourced groups is exactly what a contested source term needs, provided a neutral third party eventually adjudicates on a shared target.

Frontier The famine numbers are amplified by advocacy organisations that drop the error bars. The IPPNW/ICAN Nuclear Famine report (2022) presents the Xia projections — more than two billion dead regionally, about five billion globally — in service of an explicit case for the Treaty on the Prohibition of Nuclear Weapons, emphasising that a regional war uses “less than 3 per cent of the global nuclear arsenal.” The underlying paper states its single-model, single-member caveat; the advocacy restatement generally does not. Established This is not a claim that the advocates are wrong about the danger; it is a claim that a contested projection quoted as a settled body count is a different object from the projection the scientists published. Flagging that gap is exactly the product this site sells.

Established No institution is funded to adjudicate the dispute. There is no standing body charged with maintaining an agreed nuclear-winter source term, no intercomparison project of the kind that exists for climate sensitivity, and no agency tasked to instrument the next pyroCb. The science advances through individual grants and the accident of wildfire seasons, which is why a factor-of-ten disagreement has persisted for years without resolution.

Frontier The policy relevance is real regardless of which estimate wins. Even Reisner’s lower numbers describe a regional catastrophe; even a heavily discounted famine projection reorders the expected cost of nuclear use. The institutional question — who commissions the decisive work — is taken up as a governance problem in Existential Risk Governance, which owns the arms-control machinery this brief deliberately does not.

Established Reporting the human cost honestly is itself an ethical constraint. The projections — billions of famine deaths, concentrated in countries far from any target — are among the largest numbers in any frontier-science brief, and they demand the opposite of sensationalism. The duty is to state them, source them to Xia and colleagues (2022), flag them as contested-input model output, and neither round them up for effect nor round them down for comfort. Frontier A number this large invites both amplification by those who want the weapons gone and dismissal by those who do not, and the flag system exists precisely to make the register visible instead of letting the reader’s priors fill it in.

Frontier The distribution of harm is the moral core: the dying are largely non-combatants in non-belligerent nations. A regional war between two states is projected to kill most of its victims by starvation elsewhere, in food-importing countries that had no part in it. This is why the modelling is a global-governance matter and not a bilateral one, and why the accuracy of the number is not an academic nicety — it is the evidentiary basis on which the rest of the world’s stake is argued.

Established Overstating the certainty has its own cost. If the famine projection is quoted without its error bars and later revised down, the correction can be weaponised to dismiss the whole concern — a dynamic the field has already seen. Honest flagging protects the finding as much as the reader; a contested number defended as settled is more fragile, not less.

Speculative There is a hard line this brief does not cross. Assessing the consequences of nuclear war requires saying nothing about how weapons are built, targeted for maximum effect, or made more incendiary. This brief discusses only fuel already in cities, smoke, and climate; it treats device parameters solely as the scenario labels the published climate literature already uses, and it carries no weapon-design content of any kind. That restraint is a condition of doing this work responsibly, not an omission.

Frontier If the higher estimates are right, a regional war is a civilisation-scale event by way of the harvest. A global calorie decline of tens of per cent for several years has no modern precedent; the models describe grain reserves exhausted, export bans cascading, and mortality driven by distribution collapse as much as by absolute shortage. Frontier Global grain reserves cover only a few months of consumption, so a multi-year yield shock has no buffer to draw on; that structural fragility is measured and real even where the soot input is not. The claim that a hundred-weapon exchange could kill more people than any event in human history rests entirely on the source-term dispute this brief is about.

Established The abrupt-darkening family is the general lesson. Nuclear winter, asteroid impact winter, and super-eruption volcanic winter share a physics and a food-system failure mode; a civilisation that can model and buffer one has largely learned to buffer the others. The grain reserves, seed banks and crop-diversity measures that would blunt a nuclear famine are the same ones that blunt an impact or eruption, which is the constructive reading of an otherwise grim subject.

Speculative The deterrence feedback is real but unquantifiable. A widely believed nuclear-winter result may itself lower the probability of use — the argument Robock and colleagues make in titling a 2022 essay “How Nuclear Winter Has Saved the World, So Far.” Whether belief in the theory has deterred anyone is untestable, and treating a scientific finding as a strategic instrument is a move this brief notes without endorsing.

Handwave At the extreme, the scenarios approach but do not reach human extinction. Even the 150 Tg case leaves large populations in the tropics and southern hemisphere with reduced but non-zero food production in the models; framing nuclear winter as guaranteed extinction goes beyond what any current model supports, and the honest civilisational claim is catastrophe on a scale that reorders the world, not the certain end of it.

These horizons track when the modelling uncertainty could be resolved, not when any war might occur.

• 10 yr: Frontier The observational half is essentially certain to advance — large pyroCbs will keep occurring and, if a rapid-response sampling capability is funded, will return the black-carbon fraction and self-lofting rate the source models need; a multi-model, multi-member famine ensemble is achievable within this window with existing tools and modest funding, as is a shared urban fuel census.
• 25 yr: Speculative A genuine source-term intercomparison on an agreed target could, if commissioned, collapse the factor-of-ten disagreement into a bounded number with honest error bars; whether it is commissioned depends on the institutional gap named under section 10 and on Existential Risk Governance more than on any technical obstacle.
• 50 yr: Speculative The abrupt-cooling modelling stack — shared with impact winter and volcanic winter — matures into a single validated framework for sudden-darkening events, with the crop-and-famine step benchmarked against whatever real food shocks the century supplies.
• 100 / 250+ yr: Handwave Any statement this far out assumes the weapons, the cities and the institutions still exist in recognisable form; the honest position is that the physics will be settled long before the political conditions that make the question urgent are, and that the model, not the world, is the tractable part.

• Depends on results mapped elsewhere in the corpus: the pyroCb source observations of Wildfire Systems, which supply this field’s only validation data; the stratospheric-aerosol radiative and microphysical modelling shared with Climate Engineering; and the ensemble-modelling capacity documented under Earth-System Digital Twins.
• Requires (not on this map) five things nobody currently supplies: an agreed, fully specified urban target set on which soot mass, blackness and injection height are computed as outputs rather than assumed; aircraft in-situ sampling of the next large pyrocumulonimbus to measure the black-carbon fraction and self-lofting rate directly; a multi-model, multi-member climate-and-crop ensemble to replace the single-model, single-member famine projection; an open, auditable urban fuel-loading census to settle the 0.14-versus-94.5 g/cm² dispute; and a standing institution charged with maintaining an agreed source term, of which none exists.
• Enables an evidence-grade input to nuclear-use expected-cost reasoning, a shared framework for the whole abrupt-darkening family (impact winter, volcanic winter), and grain-reserve and food-shock policy that is calibrated rather than guessed.
• Adjacent briefs carry the couplings: Climate Engineering (the deliberate-injection mirror), Planetary Defense (impact winter), Existential Risk Governance (who commissions the work), and Civilization Resilience Planning (the food-shock response).

Established “Nuclear winter was disproved in the 1990s.” The 1980s TTAPS work was criticised and revised, and early claims were softened to “nuclear autumn” by some critics; but the modern results (Robock, Toon, Mills, Coupe, Xia, 2007–2022) use full Earth system models the original authors did not have, and they find larger, longer effects for the superpower case, not smaller. The theory was revised, not refuted, and the live dispute is about the source term, not about whether stratospheric soot cools the surface.

Frontier “The wildfire observations prove the famine numbers.” They do not. The 2017 and 2020 pyroCbs validate self-lofting and multi-month lifetime — the physics of the middle of the chain — at about 1/500 the mass and roughly 2 per cent the blackness of the scenario soot. They constrain how stratospheric smoke behaves; they say almost nothing about how much soot a burning city makes, which is where the famine numbers are actually decided. Established Confusing validation of the transport mechanism with validation of the source term is the most common error in the popular coverage of this subject.

Frontier “Los Alamos showed nuclear winter is a myth.” Reisner and colleagues found much less soot for the specific target they modelled; the critics showed that target held 15 to 110 times less fuel than major cities and that the dry source calculation omitted the latent heat now known to be required for stratospheric lofting. The Los Alamos work is a serious challenge to the source term, not a refutation of the phenomenon, and its own numbers still describe a regional catastrophe.

Speculative “A regional war would only affect the region.” This is the misconception the whole field exists to test. If the soot reaches the stratosphere in quantity, the harvest failures are global and the death toll falls mainly outside the belligerent states; if it does not, the effect is regional. Which is true is unresolved — but “regional war, regional effect” is an assumption, not a finding, and the models put serious probability on the global outcome. Frontier The comfort of “it stays over there” is precisely the belief the science is built to test, and it should not be adopted by default.

Established “Nuclear winter means an ice age.” No model produces a frozen planet. The largest scenarios give a few kelvin of global-average cooling with severe growing-season and sunlight effects; the killer is crop failure, not glaciation, and calling it “winter” misleads about both the temperature and the mechanism.

Handwave “It would certainly cause human extinction.” Even the 150 Tg case leaves substantial surviving food production in the tropics and southern hemisphere in the models. The honest claim is a global catastrophe that could kill billions, not a certainty of extinction; overstating it to the ceiling is the same error, in the other direction, as dismissing it to the floor.

Frontier “The single-model famine study is just alarmism.” The Xia study names its own limitation — one model, one ensemble member above 5 Tg — in its own text; that is the mark of disclosed uncertainty, not alarmism. The correct response is the multi-model ensemble in section 7, not dismissal; a disclosed weakness is an invitation to do the harder study, which no one has yet funded.