Inadvertent Geoengineering

Frontier The prize is a measured value for the largest single uncertainty in the human forcing of climate, from an experiment that has already been run and already been paid for. The 2025 annual indicators update puts total aerosol effective radiative forcing over 1750–2025 at −0.96 W/m², range −1.58 to −0.40 — a spread of 1.18 W/m² that no other forcing agent matches. On 1 January 2020 the International Maritime Organization cut the sulphur content of marine fuel from 3.50% to 0.50% by mass, and shipping sulphur dioxide emissions fell from 10.4 million tonnes in 2019 to 2.0 million tonnes in 2020: a dated, near-instantaneous, 8.4 Mt-per-year step change in an aerosol precursor emitted almost entirely over ocean, into the marine low clouds whose response the models disagree about.

Frontier What the world does instead costs more and settles less. The alternative constraint on aerosol–cloud forcing is model ensembles disagreeing with each other: five models put the forcing of this same regulation at 0.073 W/m², the same model run with different emitted-aerosol assumptions puts it as high as 0.407 W/m², and the published central estimates across the field span 0.057 to about 0.5 W/m² — a factor of nine, from the same emissions inventory. Narrowing that to the width of a single study’s error bar would do more for near-term climate projection than any model development now funded, because the shipping step is the cleanest available lever on the cloud-adjustment term that dominates the aerosol spread.

Established The benefits already banked are health, not knowledge, and they accrue to coastal populations rather than to the people arguing about the forcing. The regulation was written to avoid an estimated 266,300 premature deaths a year and to cut childhood asthma cases attributable to shipping from about 14 million to 6.4 million a year. Those gains are the reason the cap exists and the bar any reversal argument has to clear. The radiative information is a by-product that a monitoring programme costing a rounding error against either number would have secured, and that nobody arranged.

Speculative The condition is that the signal is separable from weather at all, and on current evidence that is not yet true. An observational analysis of the three main shipping regions puts the cloud radiative response at 0.074 ± 0.005 W/m² while stating that the event itself has low detectability against natural cloud variability; an 18-member climate-model ensemble finds the resulting warming consistent with zero throughout 2020–2040. Section 5 is the list of reasons that separation is hard, and section 7 names the one differencing strategy that has already partly done it.

Established The regulation is simple and its date is exact. IMO resolution MEPC.305(73) amended regulation 14 of MARPOL Annex VI so that “the sulphur content of fuel oil used or carried for use on board a ship shall not exceed 0.50% m/m”, against a previous global limit of 3.50% m/m, with 0.10% m/m inside designated emission control areas. The use limit applied from 1 January 2020; the prohibition on merely carrying non-compliant fuel entered into force on 1 March 2020.

Established An exactly dated step is the single most useful property this experiment has. Almost every other constraint on aerosol–cloud forcing comes from a gradual trend confounded with everything else that changed over the same decades. Here the onset is a calendar date and the magnitude is inventoried.

Established Sulphur dioxide from ships is an aerosol precursor, and the clouds it feeds are the ones that matter most for radiation. Oxidised to sulphate, it supplies cloud condensation nuclei to marine stratocumulus and trade cumulus over dark ocean, where a small change in cloud brightness produces a large change in reflected sunlight. More nuclei means more, smaller droplets at the same liquid water content: the Twomey effect, which raises albedo.

Frontier What happens after the Twomey step is where the argument lives. Whether the cloud then holds more or less water, and whether it covers more or less sky, are the adjustments. They are larger than the Twomey term in several recent analyses, they are not agreed in sign, and they are the reason this brief has a frontier section.

Established The field has a name for this class of evidence and a catalogue of instances. Christensen and thirty-three co-authors set out the taxonomy of opportunistic experiments in 2022: injections of aerosol into an environmental regime whose unperturbed state is to some extent known. Their list runs through ship tracks, shipping corridors, industrial point sources such as Norilsk, volcanic eruptions, biomass-burning plumes, weekly emission cycles, socioeconomic shocks including COVID-19, and policy changes — naming the 2020 IMO regulation and the regional emission control areas explicitly.

Frontier The framing that makes this brief distinct is termination shock with the sign reversed. Deliberate marine cloud brightening adds aerosol to marine low cloud to raise its albedo; the sulphur cap subtracted aerosol from the same clouds, abruptly, globally, on a known date. Yuan and colleagues call it “an inadvertent geoengineering termination shock” and the result “a reverse MCB”. The physics under test is identical, at a scale no research programme would be permitted to attempt. Deliberate intervention belongs to Climate Engineering; this brief is about what the accident measured.

Established Two observable quantities carry the signal, and they are not the same thing. The first is the ship track: a narrow, visible, individually attributable cloud line behind one vessel, countable in satellite imagery. The second is the corridor-scale perturbation: a statistical brightening across a lane, averaged over many ships and many days, which is what appears in the radiation budget.

Frontier Ship tracks are the easier measurement and the worse estimator. Visible tracks form only under particular boundary-layer conditions and are therefore a biased sample of the cases that matter. Section 3 shows both quantities, and shows them disagreeing.

Frontier The estimates in circulation are of three incompatible kinds. Some are model differences between runs with 2019 and 2020 shipping emissions. Some are satellite observations differenced against a machine-learned counterfactual built from pre-2020 meteorology. Some are inferred from a corridor signal divided by an independent measure of fuel burned. A converged field would force these into agreement, and the field has not converged.

Established The emissions cut happened, it was close to the size predicted, and it is the least disputed fact in the subject. The 2025 indicators update records that “SO2 emissions from international shipping declined by 8.4 TgSO2 from 10.4 TgSO2 in 2019 to 2.0 TgSO2 in 2020, which is close to the expected 8.5 TgSO2 reduction”. That is an 81% cut in one year. The multi-model study of the same event used an 80% cut relative to 2019, equivalent to 8.7 Tg SO2 per year. The inventories agree to within a few per cent, which is why the disagreement downstream cannot be blamed on the emissions.

Established Not every study uses the same cut, and the differences are large enough to matter. Quaglia and Visioni imposed a 90% reduction from 1 January 2020, averaging 4.2 Tg of sulphur per year over the decade relative to 2000–2007; Sofiev and colleagues modelled about 75%. Forcing scales roughly with emissions removed, so that choice alone is a 20% spread before any cloud physics is invoked. Part of the published disagreement is bookkeeping, not science.

Established Ship tracks fell, immediately and sharply. Yuan and colleagues built a machine-learning catalogue of ship tracks from satellite imagery and reported a global reduction after the regulation took effect. Working from that dataset, Diamond and Boss record that in the southeast Atlantic box 8–18 °S, 13 °W to 8 °E during September–October, “a typical year from the 2003–2019 period had 116 ± 61 detected ship tracks” and “59 were detected in 2020”. Watson-Parris and colleagues reported the same direction in 2022 under the title Shipping regulations lead to large reduction in cloud perturbations.

Established The corridor-scale microphysical change is also detected, at high confidence, in the one region clean enough to see it. Diamond compared the 2002–2019 climatology with the three post-regulation years 2020–2022 in the southeast Atlantic corridor and found the cloud droplet effective radius perturbation reduced, “detected at greater than 99 % confidence in the austral spring and at greater than 95 % confidence in the annual mean”. He put the local instantaneous forcing from aerosol–cloud interactions at order 1 W/m² in the corridor in austral spring and roughly 0.5 W/m² annually. That is a large local number and a small global one, and the translation between them is the whole problem.

Frontier The single most informative observation since the cap came from a war, not a research programme. Houthi attacks on Red Sea shipping from November 2023 pushed cargo traffic around the Cape of Good Hope, cutting transits through one corridor and raising them through another. Diamond and Boss used that as the differencing lever the original experiment lacked: the southeast Atlantic corridor got a large traffic increase in 2024 while the fuel stayed compliant. The droplet number perturbation there was 5%–10% before 2020, “indistinguishable from zero” over 2021–2023, and about 6% again in 2024 — with a 95% confidence interval of 3%–9% — once the diverted ships arrived.

Frontier Ratioing the cloud change against an independent tracer of fuel burned gives the first observational constraint on the sensitivity itself. Because nitrogen dioxide from ship exhaust is unaffected by fuel sulphur, Diamond and Boss used the NO2 perturbation as a traffic proxy and derived “a 66 % (25 %–88 %) reduction in the ability of ship emissions to influence cloud microphysics following IMO 2020”. Their summary: an “~80 % reduction in sulfur emissions leads to a ~67 % reduction in the increase in cloud droplet number concentration per unit marine fuel oil burned” — close to proportional, and “more linear in terms of cloud changes per sulfur decline than reported estimates from ship track detections”. The error bar is wide, one corridor deep and four years long.

Frontier The global forcing estimates span a factor of nine and the spread has widened rather than narrowed since 2024. Ordered by magnitude: Skeie and colleagues, running four global climate models and one chemical transport model, give 0.073 W/m² with model means from 0.057 to 0.089. Zhang and colleagues, using neural-network ensembles trained on 2003–2018 satellite observations, give 0.074 ± 0.005 W/m². Gettelman and nine co-authors report values tabulated elsewhere at 0.11, 0.14 and 0.15. Yoshioka and colleagues, in HadGEM3-GC3.1-LL, give 0.13. Jordan and Henry, in UKESM1, give 0.139. Quaglia and Visioni, in CESM2, give “+0.14±0.07 W m−2”. Yuan and colleagues give “+0.2±0.11 Wm−2 averaged over the global ocean”.

Frontier Nothing in that ordering is a quality ranking, and the methods do not sort cleanly by answer. The two lowest values come from independent methods — a five-model ensemble and a satellite machine-learning counterfactual — and the two highest come from a single model varied deliberately and from an energy-budget argument. Between them sit four coupled models that agree with each other to within 0.01 W/m² and share most of their aerosol assumptions.

Established One widely repeated discrepancy in that list is an area convention, not a disagreement. Yuan’s headline 0.2 W/m² is explicitly a global-ocean average and the paper gives no whole-globe figure; the 2026 preprint that tabulates the literature lists Yuan at 0.14 W/m², which is what 0.2 over 71% of the planet becomes when spread over all of it. Commentary contrasting Yuan’s “0.2” with Yoshioka’s “0.13” is comparing two denominators. The real spread is large enough without that.

Frontier The high end is held by one preprint and one veteran modeller, and they do not agree with each other either. Smith, Henry, Yoshioka, Johnson and Haywood, in a preprint under open discussion since May 2026, argue that the apparent convergence around 0.07–0.15 W/m² is an artefact of a shared modelling assumption. Their UKESM1.1 base configuration reproduces the consensus at 0.10 W/m²; reallocating emitted sulphate mass from the accumulation and coarse modes to the Aitken mode raises it to 0.187 at 60 nm, 0.256 at 88 nm, 0.309 at 44 nm, and 0.407 W/m² when the primary sulphate fraction is cut to 10%. Separately, Hansen and Kharecha assign about 0.5 W/m² of a 1.7 W/m² darkening of Earth to ship aerosols, writing that the IPCC estimate for the indirect aerosol forcing change over the past twenty-five years is “only about +0.1 W/m2” against their “+0.5 W/m2, due to reduced aerosol emissions from ships”.

Established Two competent groups predicted this forcing before it happened and differed by a factor of five. Partanen and colleagues in 2013 modelled exactly the scenario later adopted — 0.1% sulphur coastal, 0.5% open ocean — and put shipping aerosol forcing at −0.06 W/m² against −0.39 for the then-current 2.7% fuel, a change of about +0.33 W/m². Sofiev and colleagues in 2018 found that “cleaner fuels reduce radiative cooling from ship aerosols by ~80% (71 mW m−2)”. The disagreement predates the experiment by seven years and has survived it. Lauer and colleagues in 2007 had put the total indirect effect of all shipping at −0.19 to −0.60 W/m².

Frontier The temperature consequence is smaller than the forcing spread implies and is contested down to zero. Watson-Parris and thirteen co-authors ran an 18-member CESM2 ensemble sampling ENSO and Atlantic Multidecadal Oscillation phases and found a “20-year average global mean warming for 2020–2040 is +0.03 °C, with a 5 %–95 % confidence range of [-0.09,0.19]”, concluding the response “is unlikely to have been discernible above internal variability by the end of 2023 and is in fact consistent with zero throughout the 2020–2040 period”. Yoshioka and colleagues simulate 0.04 K averaged over 2020–2049, rising to 0.04–0.05 K and becoming statistically significant in the 2030s and 2040s.

Frontier The other half of the temperature literature puts the response at several times that and already visible. Quaglia and Visioni attribute 0.08 K ± 0.03 K to the regulation by 2023, emerging about three years after the emissions change. Yuan and colleagues state that the forcing “could lead to a doubling (or more) of the warming rate in the 2020 s compared with the rate since 1980”, quoting 0.24 K/decade against 0.19 K/decade. Jordan and Henry title their paper with the claim that the regulations accelerate global warming by up to three years. The gap between 0.03 K and a doubled warming rate is not a measurement disagreement; it is the forcing spread propagated forward.

Established The thing the forcing is being asked to explain is real and is mostly not explained. Gavin Schmidt, in Nature in March 2024, put the 2023 anomaly at about 0.2 °C above what the models predicted, and accounted for roughly 0.02 °C from greenhouse gas growth since 2022, a few hundredths from Hunga Tonga and the solar cycle together, and a few hundredths from the shipping regulations — leaving about 0.2 °C unexplained. Goessling, Rackow and Jung independently find that “previous best-guess estimates of known drivers including anthropogenic warming and the El Nino onset fall short by about 0.2K”, attributing the gap to a record-low planetary albedo “caused largely by a reduced low-cloud cover in the northern mid-latitudes and tropics”.

Frontier Marine cloud reflectivity really is falling, and most of the fall is attributed to aerosol. Von Salzen and colleagues analysed 2003–2022 and report that “marine cloud reflectivity decreased on average by 2.8 ± 1.2% per decade”, a cloud radiative effect change of 1.04 ± 0.45 W/m² per decade over the North Atlantic and Northeast Pacific, with droplet number falling 6.7 ± 1.5 cm−3 per decade in the former and 2.9 ± 1.7 in the latter. They attribute “69% (range 55−85%)” of the decrease to sulphur dioxide and other precursor reductions.

Frontier That result is the largest observed aerosol–cloud signal in the marine record and it is not the shipping signal. The period is dominated by Chinese and North American emission controls, and the paper does not separate the IMO step out of it. It establishes that the mechanism operates at the scale claimed; it does not apportion.

Frontier The most recent observational assessments argue that none of this requires revising the aerosol numbers. Zelinka and nine co-authors, writing in 2026, examined low-cloud radiative effects over global ocean from 2003 to 2024 including the 2023 extreme and concluded that the observations “do not indicate an emerging stronger cloud feedback or underestimated future warming”. Their aerosol–cloud forcing of −1.20 ± 0.61 W/m² and low-cloud feedback of +0.21 ± 0.23 W/m² per kelvin sit inside the assessed ranges, with sensitivity near 3 °C.

Frontier An independent decomposition caps how much of the recent cloud trend any aerosol change can own. Ceppi and colleagues split the July 2003 to June 2024 low-cloud trend into a cloud feedback of 0.09 ± 0.02 W/m² per decade, a greenhouse-gas adjustment of 0.05 ± 0.03, an aerosol adjustment of 0.03 ± 0.03, and a natural-variability term of 0.01 ± 0.08, with forced responses jointly accounting for 74% of the trend. An aerosol adjustment of 0.03 W/m² per decade for all aerosol sources leaves little room for a shipping forcing near 0.4 W/m².

Established The background against which all of this is measured is itself moving fast. The 2025 indicators update reports Earth’s energy imbalance trending at 0.30 ± 0.1 W/m² per decade over 2001–2025 in the assessed record and 0.44 ± 0.13 in the satellite record, having more than doubled since 1976–1995, with human-induced warming increasing at “0.27 [0.2–0.4] °C per decade over 2016–2025”.

Frontier That observed rate is the one Hansen and colleagues predicted in 2023, and it is weaker evidence than it looks. They wrote that the “decline of aerosol emissions since 2010 should increase the 1970–2010 global warming rate of 0.18°C per decade to a post-2010 rate of at least 0.27°C per decade”. The rate can be right because the aerosol forcing is large, or because climate sensitivity is high, or both, and the observation does not separate them.

Frontier The live question is not how much sulphur was removed but how many particles it would have made, and that turns on a size distribution nobody has measured post-cap. Smith and colleagues identify the mechanism precisely: models emit ship sulphate as mass into prescribed size modes, and reallocating it from the accumulation and coarse modes into the Aitken mode raises emitted particle number by about an order of magnitude for the same sulphur mass. Number, not mass, activates droplets. Their conclusion — that a forcing two to four times the consensus “would explain a larger portion of Earth’s energy imbalance since 2020 and of the recent global temperature surge” — rests entirely on that reallocation being right.

Frontier Three referees have already said in public what is wrong with it, and the objections are the ones a reader should hold. Referee 1 objects that a single atmosphere-only model with fixed 2005–2014 sea surface temperatures is “structurally blind to alternative physical explanations”, that “a direct bottom-up comparison with aircraft data is completely missing”, and that the favoured configuration overshoots the regional observational constraint of about 0.5 W/m² by more than a factor of two. Referee 2 objects that the CERES cloud radiative effect used as the target is not a function of aerosol alone. Robert Wood, reviewing under his own name, calls the work “short, interesting, and useful” and still asks the hardest question in the subject.

Frontier Wood’s question is the one this brief exists to pose: why would commercial shipping induce large cloud adjustments when deliberate marine cloud brightening barely induces any? He notes that ship plumes spread at about 2 km per hour, so emitting into a gridbox is not the same physical problem as a concentrated plume in which coagulation is nonlinear, and that Diamond and co-authors found significant negative liquid water path changes in ship tracks, “directly opposite to the ESM results”. If the same aerosol gives a large forcing from a ship and a small one from a sprayer, one of those measurements is wrong, and both feed live policy arguments.

Frontier The liquid water adjustment could move the answer most and its sign is disputed. Manshausen and colleagues tracked about two million equivalent ship paths over six years and found a positive liquid water response in clouds with no visible track against weakly negative responses in visible ones, the droplet response peaking at 3–4 hours and persisting up to 20 hours. They put the liquid water path adjustment at “−0.76 (−1.03, −0.49) W m−2” against “the estimate of 0.2 ± 0.2 W m−2 given in the latest Intergovernmental Panel on Climate Change report” — opposite in sign and roughly four times the magnitude. Glassmeier and colleagues argue the reverse, that tracks are sampled before the adjustment has run its course.

Frontier The response saturates, which breaks the power-law scaling that most forcing calculations assume. Manshausen and colleagues found in 2023 that droplet number responds linearly to emission rate over 1–10 kg per hour before saturating, while liquid water anomalies in raining clouds stay roughly constant across the whole observed range — a threshold behaviour in which even a small droplet enhancement shuts down drizzle. They state the consequence: this matters for “calculations of aerosol forcing, which are currently performed using power law relations”. If the adjustment saturates and the Twomey term does not, removing 80% of the sulphur does not remove 80% of the effect, and the sign of the error depends on where on the curve the fleet was sitting.

Frontier Cloud fraction, not droplet size, may be carrying most of the forcing, which would relocate the whole argument. Chen and colleagues, using a random-forest surrogate and the 2014 Holuhraun eruption as a near-pristine natural experiment, found aerosols increased cloud cover by about 10% and that this, not brightening, led the forcing. Their 2024 follow-up puts the cloud-cover share of aerosol–cloud forcing at 65% in June 2008 and 84–87% in 2018 against 25–34% for the Twomey term, with regional effects between −2 and −10.5 W/m². A cloud-cover-dominated forcing is much harder to detect in a corridor, because cloud cover is the noisiest field in the problem.

Speculative Hansen’s position is that the whole aerosol budget is wrong and the shipping step is the proof. Hansen and Kharecha evaluate a 0.5% albedo decrease as a 1.7 W/m² increase in absorbed solar energy, assign about 0.5 W/m² of it to ship aerosols and 0.15 W/m² to sea-ice albedo with the remainder cloud feedback, and conclude that “climate sensitivity for doubled CO2 is 4.5°C ± 0.5°C”. Their 2023 paper put Charney sensitivity at “1.2 ± 0.3°C (2σ) per W/m2, which is 4.8°C ± 1.2°C for doubled CO2”.

Frontier That position is contested on the record by named people, which is the useful form of disagreement. Karsten Haustein, responding through the Science Media Centre, called the 0.5 W/m² figure “far higher than the current estimates of 0.05-0.15 W/m²”; William Collins said the paper’s aerosol cooling sits “outside most previous assessments” and contradicts observed 1980s temperatures.

Frontier The high-profile result at the top of the published range was contested the week it appeared. Responding to Yuan and colleagues through the Science Media Centre in 2024, Laura Wilcox said the temperature claims “seem difficult to justify”, identified possible double counting and objected to the geoengineering framing; Joel Hirschi said the paper “likely overestimates the effect on global temperatures”; Edward Gryspeerdt, declaring a shared co-authorship, agreed it “suggests a stronger warming effect than other similar studies”.

Frontier Declare the tie: the field does not know whether the IMO forcing is 0.07 or 0.4 W/m², and the two camps are not separated by competence. On the low side sit five models converging, an observational machine-learning estimate agreeing with them to three decimal places, an 18-member ensemble finding no detectable temperature response, and a 2026 reassessment of the cloud record finding nothing that needs explaining. On the high side sit a microphysical mechanism with a name, a referee of standing calling the result of major potential importance, an ex-ante model from 2013 that predicted +0.33 W/m², and an energy-budget argument. Agreement among models sharing an assumption is not evidence about the assumption.

Established The binding constraint is that nobody measured the particle size distribution in a low-sulphur ship plume at sea, before or after the switch, at the sampling density the question needs. This is the quantity Smith and colleagues vary by hand to move the forcing from 0.10 to 0.407 W/m², and the one Referee 1 says is missing: “a direct bottom-up comparison with aircraft data is completely missing”. Laboratory work exists — Santos and colleagues measured how marine fuel regulations change engine emission physicochemistry and cloud activity — but a laboratory engine is not a plume dispersing at 2 km per hour into a marine boundary layer.

Established Cloud droplet number is a retrieval, not an observation, and it degrades exactly where the signal is. It is inferred from satellite-retrieved effective radius and optical depth under adiabaticity and homogeneity assumptions that fail in broken cloud, thin cloud and at high solar zenith angle. The southeast Atlantic is the workhorse region precisely because it has extensive, seasonally reliable stratocumulus and almost no continental aerosol — so the one place the retrieval behaves is a special case, and generalising from it is an extrapolation.

Frontier The radiative signal is below the noise floor of the instrument record for a period measured in years, and the field knows the number. Zhang and colleagues report low detectability of the cloud radiative effect “attributed to strong natural variability in cloud albedo and cloud cover”, and cite prior work finding “at least 5 to 6 years of observations are needed to derive significant signals” over the southeast Atlantic — against which their own study examined three. The cap took effect in 2020; six years of post-cap record existed as of 2026, and it is only now long enough for the albedo question to be asked.

Established There is no counterfactual ocean, and every method for manufacturing one carries its own failure mode. Model differencing needs the model’s aerosol scheme to be right, which is the thing under test. Machine-learned counterfactuals trained on pre-2020 meteorology need the post-2020 atmosphere to stay inside the training distribution, which 2023 conspicuously did not. Corridor-versus-background differencing needs a background shipping does not reach, which the North Atlantic does not have. The traffic-disruption method needs no counterfactual at all, and it needed a war.

Frontier Sub-gridscale plume treatment is a known defect with no accepted fix in any model used to estimate this forcing. A gridbox is tens to hundreds of kilometres across; a fresh ship plume is metres to hundreds of metres across, and coagulation losses inside it scale with the square of number concentration. Emitting sulphate directly into the gridbox skips the stage where most small particles are destroyed or preserved. Wood raises this against the high-forcing result; it applies with equal force to the low-forcing results, which use the same shortcut.

Frontier The confounders that arrived with the cap are not small and are not separable without extra information. Between 2020 and 2024 the marine boundary layer also received a COVID-19 traffic collapse, the Hunga Tonga eruption, a solar maximum, a strong El Nino, the Red Sea diversion, and a continuing decline in Chinese sulphur emissions. Schmidt’s accounting gives shipping a few hundredths of a degree and leaves 0.2 °C unexplained, so any method inferring the shipping forcing from global temperature is fitting a small term inside a large residual.

Established Compliance is heterogeneous in a way the inventories smooth over. Ships may comply with distillate or very low sulphur fuel oil, or by burning high-sulphur residual behind an exhaust gas cleaning system. The inventory agreement — 8.4 Tg observed against 8.5 Tg expected — holds at the global total and says nothing about which lanes, vessel classes or seasons the residual 2.0 Tg sits in. Lane-by-lane differencing needs that resolution and does not have it.

Established The first dependency is an aerosol-physics measurement nobody has made. Every high-forcing result rests on the emitted size distribution of sulphate from compliant marine fuel: aircraft transects through identified plumes with size-resolving instrumentation and a cloud condensation nucleus counter, cross-referenced to the emitting vessel. Until that exists, the difference between 0.10 and 0.407 W/m² is a modelling choice.

Established The second is a ship-level inventory that resolves fuel type, not just fuel mass. The inventories descend from automatic identification system tracking plus engine models and reproduce the global total well. The differencing experiments need sulphur actually burned per lane per month, which requires bunker delivery notes or fuel sampling linked to vessel identity — administrative data held by flag states and port authorities and not published in a differenceable form.

Frontier The third is a way to remove the sea-surface-temperature pattern effect from the residual. Referee 1 objects that a substantial part of the post-2020 surge may be a response to changing sea surface temperature patterns rather than to forcing at all. Samset and colleagues showed that filtering pattern effects with a Green’s function brings a forced signal out of the noise five years earlier in their central case and up to nine years earlier overall. Applying that to the shipping step is tractable and unpublished.

Frontier The fourth is a cloud-controlling-factor decomposition trusted enough to arbitrate. Ceppi and colleagues attribute 74% of the observed low-cloud trend to forced responses with the aerosol adjustment at 0.03 ± 0.03 W/m² per decade. If that holds it caps how much of the post-2020 albedo change shipping can own; if not, the cap is illusory. The shipping literature does not settle it, and the dependency runs the other way.

Established The fifth is continuity of the instruments. Every result rests on CERES for radiation and MODIS or its successors for cloud microphysics, with records starting around 2000–2003, and the detection argument is explicitly about record length. A gap in either record does not delay the answer; it destroys the pre-cap baseline that makes differencing possible.

Frontier The decisive experiment is ship-track- and corridor-resolved satellite retrieval of cloud droplet number and albedo before and after the cap, differenced against shipping lanes whose traffic changed for reasons unrelated to fuel — and its first instance has already been run. Diamond and Boss used the post-November-2023 Red Sea diversion to supply the missing arm: one corridor lost traffic, another gained it, fuel sulphur was held constant by regulation, and exhaust nitrogen dioxide served as a tracer of fuel burned that the cap does not touch. Dividing the droplet number perturbation by the NO2 perturbation yields a sensitivity per unit fuel that depends on no model, no counterfactual and no assumption about emitted size distribution. This is the single result this brief would most want repeated, because it is the only published number in the subject that is a measurement rather than an inference.

Frontier What it returned, and what would make it decisive rather than suggestive. The published constraint is a 66% reduction, range 25% to 88%, in the ability of ship emissions to influence cloud microphysics after the cap — one corridor, one season, one year of post-diversion data, and an interval wide enough to accommodate most of the estimates in section 3. Narrowing it requires the same analysis across the northeast and southeast Pacific and the North Atlantic, across all seasons, with more years of post-2024 record and the NO2 tracer validated against vessel-level fuel data. None of that needs a new instrument, a new satellite or a permission.

Established The second is the aircraft campaign that would close the size-distribution question outright. Fly identified vessels burning documented compliant fuel; sample the plume at several ages; measure the number-size distribution and the cloud condensation nucleus spectrum at several supersaturations; compare against a scrubber-equipped vessel burning high-sulphur residual. That is a bounded, conventional campaign, it would adjudicate directly between the 0.10 and 0.407 W/m² configurations of a single model, and it has not been funded.

Speculative The third would settle the matter fastest and will not happen. A designed perturbation — a fleet at 3.5% sulphur on a randomly assigned open-ocean corridor for two seasons against a compliant control corridor — would deliver the sensitivity with a clean control in about three years. It also means deliberately emitting a pollutant the same regulation removed to avoid hundreds of thousands of deaths a year, which makes it a non-starter under the constraints set out in Geoengineering Governance. Naming it marks the ceiling of what observation must substitute for.

Frontier The fourth is simply waiting, and the waiting has a stated end. Five to six years is the record length at which the southeast Atlantic albedo signal should emerge; the cap is six years old, and the 2026 preprint already compares CERES anomalies over 2020–2025 against a 2002–2019 baseline. Each additional year both lengthens the post-cap sample and contaminates it, because the Asian aerosol trend continues underneath.

Frontier The fifth is the emission control area analogue, which supplies a spatial discontinuity rather than a temporal one. The 0.10% m/m limit inside those areas came into force earlier and covers bounded, heavily trafficked regions with dense surface observation. A step in fuel sulphur across a fixed boundary at fixed meteorology is a stronger identification strategy than a step in time. Christensen and colleagues list these areas among the available opportunistic experiments; published exploitation of them is thin.

Speculative The sixth is the reconciliation test between commercial shipping and deliberate spraying. Wood’s question is answerable by running both in the same air mass on the same days with the same retrieval. If the answers differ, the difference is in the aerosol rather than the cloud, and that is diagnosable. If they agree, one of the two literatures carries a systematic error large enough to explain the whole spread in section 3.

Frontier What would change this brief’s assessment, ranked. A multi-corridor, multi-season repeat of the traffic-differencing analysis returning a sensitivity reduction tightly bounded near 66% would put the forcing near the low convergence and retire the high-end preprint. The same analysis returning a much smaller reduction — cloud response barely changed despite an 80% sulphur cut — would mean the droplet response is saturated and the forcing small for a reason nobody has modelled. Aircraft size distributions falling in the Aitken mode would move the field to the high end in one paper.

Established The requirement is observational, not industrial, and the shortfall is in three instrument classes. First, a broadband radiation budget instrument with calibration continuity across the 2019-to-2020 boundary, because the detection argument is a difference against a baseline. Second, an imager retrieving effective radius and optical depth at a kilometre or better, fine enough to resolve individual tracks. Third, geostationary coverage of the major corridors, because the droplet response peaks 3–4 hours after emission and persists up to 20 hours, which a twice-daily polar overpass samples badly.

Frontier The aircraft requirement is the binding one and it is modest by the standards of climate observation. One instrumented aircraft carrying a scanning mobility particle sizer, an optical particle counter, a cloud condensation nucleus counter and a cloud probe, flying from a coastal base near a major lane for two seasons, would produce the plume size distributions the high-forcing argument rests on. The same instrument class is argued over for deliberate intervention in Climate Engineering; here the perturbation already exists and only the measurement is missing.

Established The data-engineering requirement is joining three systems built for different purposes. Automatic identification system tracks give vessel position and identity; bunker delivery notes give fuel sulphur per vessel per voyage; satellite retrievals give cloud state on a fixed grid. Joining them yields, for each cloud pixel, the fuel burned upwind and its sulphur content — a database problem with a jurisdictional obstacle, of the same class as the integration described in Earth-System Digital Twins and Future Ports and Shipping.

Frontier The modelling requirement is a plume parameterisation that survives contact with the plume. Every model estimate in section 3 emits ship sulphate directly into a gridbox tens to hundreds of kilometres across. The alternative is a sub-gridscale plume model, calibrated against large-eddy simulation and against the aircraft data that does not yet exist, reporting how many particles survive coagulation and dilution to activation. Until then, agreement at 0.073 W/m² is agreement about a shared shortcut.

Speculative None of this needs new industrial capacity, which is unusual for this corpus. The satellites are flying, the aircraft exist, the ships are sailing and the data are being generated whether or not anyone analyses them. What is missing is a funded programme with the authority to join four datasets and fly one aircraft, which makes the gap an institutional fact rather than an engineering one — the subject of section 10.

Frontier Marine cloud brightening is the closest adjacent technology and this event is the largest dataset it will ever get. A brightening scheme injects sea-salt aerosol into marine stratocumulus to raise albedo; the cap removed sulphate from the same clouds. The sensitivity is one quantity measured with the sign reversed, and the ratio between ship-derived and sprayer-derived answers is the discrepancy Wood flags. Deployment and governance belong to Climate Engineering; this brief supplies the measurement, and the measurement does not take a side.

Established Detecting a deliberate intervention is the same statistical problem, and this is its dress rehearsal. Whether anyone could tell a solar radiation modification programme was working is the question of whether anyone can tell the sulphur cap worked — known forcing, known start date, a long instrument record. That the answer is so far a confidence interval spanning zero to twice the central estimate bears directly on the monitoring requirements in Geoengineering Governance.

Established Air-quality policy elsewhere is the same experiment with worse identification. Chinese, North American and European sulphur controls removed aerosol over the same decades, but over land, gradually and confounded with everything else. They carry the largest observed marine cloud signal and are a blunter instrument than the shipping step.

Frontier Machine-learning emulators are on both sides of this problem, as the analysis tool and as the thing being tested. The 0.074 ± 0.005 W/m² estimate comes from neural networks trained on 2003–2018 satellite data to produce a business-as-usual counterfactual; the cloud-cover results come from random forests on twenty years of retrievals. The same out-of-distribution failure mode appears in AI Weather Prediction: a model trained on a past atmosphere asked about an atmosphere outside its training range, in the year the atmosphere left that range.

Frontier The shipping decarbonisation agenda will repeat this problem. Ammonia, methanol and liquefied natural gas pathways each change the aerosol emitted as well as the carbon, and the fleet-scale transition described in Future Ports and Shipping will deliver another uncontrolled aerosol perturbation over ocean within two decades. Whether anyone measures that one in advance tests whether this lesson was learned.

Established The institution that ran the experiment is a shipping regulator and had no reason to measure the climate outcome. The International Maritime Organization adopted the 0.50% m/m limit as an air-quality and public-health measure under MARPOL Annex VI. Its mandate covers pollution from ships, its committee structure runs on flag-state consensus, and nothing in it obliges or funds radiative monitoring. The climate consequence was known in advance — Sofiev and colleagues in 2018, Partanen and colleagues in 2013 — and no observing programme was attached to the switch.

Established The miss is the cheapest institutional lesson available anywhere in climate policy. An aircraft campaign and a data-join effort across the 2019/2020 boundary would have cost a small fraction of the health benefit the same regulation was projected to deliver, and would have converted a factor-of-nine disagreement into a measurement. No body had both the mandate and the money, and the window for a pre-perturbation baseline closed on 31 December 2019.

Frontier The same institution is now mid-way through a larger decision and the politics have hardened. At an extraordinary session of the Marine Environment Protection Committee in October 2025, adoption of the IMO Net-Zero Framework was deferred: “A majority of 57 member states voted to adjourn the meeting for one year, while 49 voted to continue”, with the meeting resuming in October 2026 and entry into force on 1 March 2028 at the earliest. The framework would set greenhouse-gas fuel-intensity requirements with a pricing and reward mechanism, at an implied carbon price put in the literature at roughly USD 380 per tonne. A 57-to-49 split is not a body that will readily fund an observing programme it is not obliged to fund.

Frontier The next fuel transition is being designed now with the same blind spot. Published analysis of the Net-Zero Framework treats it as a governance precedent reaching beyond shipping, with no discussion of aerosol or sulphur co-effects and no reference to the 2020 cap. That omission is the institutional finding of this brief: the body about to change marine fuel chemistry again is not, in its own framing documents, treating the last time it did so as an event that produced data.

Established The scientific institutions that could act have organised the argument well and the measurement badly. Open discussion at Copernicus put three referee reports on the high-forcing preprint into public view within a month of submission, and the Science Media Centre put named disagreement on record in 2024 and 2025 with declared interests. None of that produces an aircraft. The reviewing machinery is in better shape than the observing machinery, which is the wrong way round for a subject whose problem is a missing measurement.

Established The regulation was right on its own terms and nothing here argues otherwise. Sofiev and colleagues projected total premature mortality from shipping in the 2020 business-as-usual case at 403,300 per year, with 266,300 avoided per year under the low-sulphur standards, and childhood asthma cases falling from about 14 million to 6.4 million. Partanen and colleagues gave a smaller figure for their closest scenario, 48,200 deaths avoided per year at a 96% mortality reduction, on different baselines; the health literature has its own spread. Both point the same way and both dwarf any plausible monitoring cost.

Frontier The trade-off is real, was stated in advance, and is a genuine ethical object rather than a rhetorical one. A regulation saving on the order of a quarter of a million lives a year also warms the planet by between nothing detectable and 0.5 W/m² of forcing, and the two are not commensurable. The health gain is measured and near-certain while the warming is contested and deferred — which is exactly why the warming went unmeasured.

Speculative The argument that the sulphur should be put back is live, wrong on its own terms, and worth stating so it can be answered. If ship aerosol were worth 0.5 W/m², the reasoning runs, restoring it would be the cheapest cooling available. It fails on the health arithmetic above; it fails because emission at sea level is a poor delivery mechanism compared with anything a deliberate scheme would choose; and it fails because the case is strongest exactly where the evidence is weakest, at Hansen’s end of the range. Using a contested forcing estimate to argue for reversing a measured health benefit inverts the evidential weights.

Frontier The distributional facts run in opposite directions and both belong on the record. The avoided deaths concentrate in coastal and port populations, heavily in East and South Asia and in Europe. The warming, if real, is global and falls hardest on populations with the least adaptive capacity, as the exposure mapping in Compound Climate Hazards sets out. The people who gained clean air and the people who bear the forcing are not the same people, and no institution adjudicated between them because no institution was asked to.

Frontier There is a framing hazard here that cuts both ways. Advocates of solar radiation modification can read a large IMO forcing as proof that marine cloud modification works at scale; opponents can read a small one as proof it does not. Both smuggle a policy conclusion out of a sensitivity measurement. What the evidence bears on is how much radiative response a marine cloud gives per unit of aerosol, and that number is an input to the arguments in Climate Engineering and Geoengineering Governance, not a verdict on them.

Frontier A routine public-health rule changed the planet’s radiation budget by an amount comparable to two to three years of carbon dioxide growth, and nobody planned to find out by how much. Skeie and colleagues make the comparison explicitly: their computed forcing is “comparable to the increase in CO2 concentration over the last 2 to 3 years”. If the high-end estimates are right, the comparison is to a decade. Either way, a committee of shipping regulators moved a planetary forcing term and the geophysical community found out afterwards.

Frontier The capability this reveals is not the ability to cool the planet but the ability to change it without noticing. The marine aerosol column is thin, short-lived and coupled to a cloud field with high radiative leverage, so small changes in an industrial input produce forcings visible in the global budget. Any transition touching combustion over water — the fuel pathways in Future Ports and Shipping, aviation fuel changes as in Zero-Carbon Aviation — carries a side effect of the same class, and the default is that it is discovered late.

Speculative Deliberate intervention is not the first hard governance case; it is the second. Debates about solar radiation modification assume the hard question is authorising a deliberate act. The sulphur cap poses a prior one: attributing an intervention nobody authorised as climate policy, with instruments that cannot resolve it, after the baseline window has closed. If a state wished to claim another had altered the radiation budget, this is the precedent for how well such a claim could be substantiated — a confidence interval spanning a factor of nine.

Frontier The reversibility asymmetry is the part that generalises. The emissions change was instantaneous and is permanent in practice; measuring its effect needs five to six years of record and degrades as other forcings move. Aerosol interventions can be made faster than they can be evaluated — the structural feature the termination-shock literature in Climate Engineering worries about from the other direction, and which Climate Overshoot and Lock-In treats for slower systems.

Established The lesson available at no cost is procedural, and small enough to be adopted. Any regulation changing a global emission of a short-lived climate forcer by more than a stated threshold should carry a measurement annex: a baseline observing period before the switch, an identified instrument set, a funded analysis. The cost is a rounding error against 266,300 avoided deaths a year and a 1.18 W/m² aerosol uncertainty. Nothing in the 2025 Net-Zero Framework deliberations suggests it is being adopted.

These horizons track what becomes knowable about the forcing, not what the climate does.

• 10 yr: Frontier The post-cap satellite record passes ten years, well beyond the five-to-six-year detectability threshold, and the albedo question becomes answerable in the southeast Atlantic and probably the northeast Pacific. Expect the traffic-differencing method to be extended to more corridors and the constraint on the sensitivity reduction to tighten from 25%–88% to something a model can be falsified against. Whether an aircraft campaign measures the emitted size distribution in this window is a funding question, and it decides whether the 0.10-to-0.407 W/m² spread closes.
• 25 yr: Speculative The fuel transition under whatever succeeds the Net-Zero Framework delivers a second aerosol perturbation over the same ocean, with ammonia and methanol pathways changing nitrogen and particle emissions as well as carbon. If the measurement annex argued for in section 12 exists by then, that transition is a designed experiment; if not, it is a second accident contaminating the first one’s baseline. The aerosol forcing range should have narrowed from 1.18 W/m², and the shipping constraint will be one of the reasons if it has.
• 50 yr: Speculative Marine fuel sulphur approaches zero across the fleet and the natural experiment is exhausted — there is no further sulphur to remove and no pre-industrial marine aerosol baseline to recover. Whatever was learned by then is what will be known from this source. If deliberate marine cloud modification is ever attempted, it will be calibrated against numbers derived in the 2020s and 2030s from an experiment nobody designed.
• 100 / 250+ yr: Handwave The IMO 2020 step survives as a dated discontinuity in the marine aerosol column, and its usefulness depends entirely on whether the instrument calibration and shipping inventories of the 2010s and 2020s remain interpretable. Claims that a future civilisation will reconstruct aerosol–cloud sensitivity from proxy records of this event are assertion: no proxy resolves a six-year cloud albedo anomaly, and the satellite archive is the only witness.

• Depends on the aerosol–cloud physics and marine cloud brightening record owned by Climate Engineering, which holds the deliberate-intervention side of the same sensitivity question; the detection-and-attribution machinery in Geoengineering Governance, which owns the question of whether an intervention can be seen at all; and the observation-to-model pipeline in Earth-System Digital Twins. Nothing else on the map blocks this topic, because what it waits on is measurement nobody is making rather than capability nobody has.
• Requires (not on this map) the number-size distribution of sulphate emitted by compliant marine fuel measured in a real plume at several ages, which is the quantity that moves a single model from 0.10 to 0.407 W/m²; a satellite droplet-number retrieval validated against in-situ aircraft sampling in broken and thin cloud, without which the corridor signal is an assumption about adiabaticity; a vessel-level record joining automatic identification system tracks to fuel sulphur actually burned, without which lanes cannot be differenced against each other; an uninterrupted broadband radiation calibration spanning the 2019-to-2020 boundary, because a discontinuity there is indistinguishable from the signal; a standing requirement that regulations changing a global short-lived forcer emission carry a baseline observing period, which no treaty body currently imposes; and a funder for that monitoring that is not the shipping industry, since the regulated party cannot credibly certify the climate consequence of its own compliance.
• Enables an observational constraint on the cloud-adjustment term that dominates the aerosol forcing uncertainty; a calibrated prior for what deliberate marine cloud modification would achieve per unit aerosol; a bounded correction to near-term warming projections that currently inherit the full 1.18 W/m² aerosol spread; and a worked precedent for attributing an unauthorised change in planetary forcing to its source.
• Adjacent to Climate Engineering on the sensitivity being measured, Geoengineering Governance on who could authorise or detect an intervention, Future Ports and Shipping on the fuel transition that will repeat this, AI Weather Prediction on the emulator methods used to build the counterfactuals, Atmospheric Management on short-lived forcer budgets, and Weather Modification on the older and harder problem of attributing a cloud intervention to its cause.

Frontier “IMO 2020 caused the 2023 heat record.” Not supported, and the paper most often cited for it does not say it. Zhang and colleagues write only that their result “positions our study to suggest that substantial warming from IMO2020 might have contributed to 2023 being the warmest year on record”, from a forcing of 0.074 W/m². Schmidt gives the shipping regulations a few hundredths of a degree against a 0.2 °C anomaly; Yoshioka and colleagues suggest shipping plus Hunga Tonga “could potentially combine to explain up to 0.08 K” against an anomaly of 1.45 ± 0.12 K above pre-industrial. Even the high-end estimates leave most of 2023 unexplained.

Established “The forcing is settled at about 0.07 W/m².” Also not supported. Skeie’s five models and Zhang’s neural network agree to within 0.001 W/m², which invites more confidence than it should: Skeie’s own paper states that “the full uncertainty in the ERF due to the new regulation is not quantified but will very likely be high”, and the models share an emitted-size-distribution assumption a 2026 preprint identifies as the thing holding the answer down. Two methods agreeing while sharing an input is not two independent measurements.

Speculative “Termination shock has now been observed.” Overclaimed. What has been observed is an emissions step, a halving of detected ship tracks in one box, a microphysical change in one corridor at greater than 99% confidence in austral spring, and model temperature responses whose 5%–95% range includes zero. Yuan and colleagues use the termination-shock framing as an analogy for abruptness, and it is apt; it is not a demonstration that abrupt cessation of a deliberate programme would do what the modelling literature says.

Frontier “This proves marine cloud brightening works.” It does not, and Wood’s referee question is why: the adjustments inferred from commercial shipping are much larger than those measured in deliberate brightening experiments, and a discrepancy that large means one is mismeasured. A large IMO forcing would raise the estimated sensitivity of cloud to aerosol, one input among several; delivery, persistence, regional response and governance are separate problems, treated in Climate Engineering.

Frontier “This proves marine cloud brightening does not work.” Equally unsupported and more common in policy commentary. A small measured forcing would be consistent with a saturated droplet response, with compensating adjustments, or with the emitted particles being the wrong size. A deliberate scheme chooses its particle size, altitude, location and timing; a container ship chooses none of them.

Frontier “Ship tracks are the forcing.” The commonest technical error in popular coverage. Visible tracks form only in particular boundary-layer conditions and are a biased sample: Glassmeier and colleagues argue that ship-track data overestimate aerosol–cloud cooling, Manshausen and colleagues found clouds with no visible track show a more positive liquid water response than those with one, and Diamond and Boss report that their per-unit-fuel estimate is more linear than ship-track detections imply. Counting tracks is a detection statistic, not a radiative measurement.

Handwave “The regulation was a mistake and should be reversed.” Assertion, and it works by leaving out the health arithmetic. The measured side of the ledger is 266,300 avoided premature deaths a year and asthma cases falling from 14 million to 6.4 million; the contested side is a forcing whose published estimates run from 0.057 to about 0.5 W/m². Reversal arguments quote the top of the contested range against none of the measured one.

Frontier “The models were tuned to hide it.” No evidence, and the actual failure mode is duller. What Smith and colleagues argue is that the models make the same defensible-looking choice about emitted sulphate size and mode, inherited from inventories rather than from tuning, and that the choice happens to suppress particle number by about an order of magnitude. Three referees contest even that. A shared unexamined assumption is not a conspiracy, and it is harder to fix.

Speculative “Hansen has been vindicated by the observed warming rate.” Partly, and less than the claim implies. The 2025 indicators update gives human-induced warming of 0.27 [0.2–0.4] °C per decade over 2016–2025, matching the at-least-0.27°C-per-decade figure published in 2023. The rate can be right for reasons other than a 0.5 W/m² ship aerosol forcing — high climate sensitivity, a larger total aerosol decline from Asian controls, a sea-surface-temperature pattern effect — and the observation does not distinguish them.

Frontier “The emissions data are uncertain.” Less than almost anything else here. The 2025 indicators update records a decline of 8.4 Tg SO2 against an expected 8.5 Tg, and the modelling studies use 75%, 80% and 90% cuts bracketing the observed 81%. The emissions are the best-constrained link in the chain; the disagreement is downstream, in how many particles that sulphur makes and what the cloud does with them.