Weather Modification

Established Deliberate weather modification is the oldest intervention humans make in the atmosphere on purpose, and it is the only one practised continuously and at scale for eighty years. It began in a General Electric laboratory in 1946, when Vincent Schaefer found that dry ice dropped into a supercooled cloud converted it to ice; it is now an operational industry on every inhabited continent, with China alone reporting more than US$2 billion of investment since 2014 and roughly 48,000 people employed. Established The chemistry divides in two. Glaciogenic seeding introduces ice nuclei — silver iodide, dry ice, liquid propane — into clouds holding liquid water below freezing, so that ice crystals form and grow at the expense of droplets. Hygroscopic seeding introduces large salt particles into warm convective clouds to broaden the droplet spectrum and accelerate coalescence.

Established The efficacy gradient across the whole subject follows one physical principle: weather modification works cleanly exactly where the atmosphere is already metastable and the intervention only has to trigger a transition it was going to make anyway. Supercooled fog is thermodynamically unstable, and clearing it is the field's one uncontested success. Winter orographic clouds are marginally metastable, and the measured effects are small and hard to see. Hail-bearing convection is not, and the sign of the intervention is disputed. Hurricanes are energetically absurd targets, and the programme that tried closed after twenty-one years.

Frontier The spine of this brief is not whether seeding works. It is that after eight decades nobody can attribute a stated quantity of water at the ground to a seeding operation to a precision that would let a water right be transferred or a downwind complaint be heard. Climate Engineering and Geoengineering Governance both name detection and attribution as their binding constraint. Weather modification is the same problem, one order of magnitude easier, four orders of magnitude smaller, repeated thousands of times a year in dozens of countries — and it is still open. Established That makes this brief the empirical base rate for the whole geoengineering map.

Established The physical chain from a flare to a snowflake is real and has been observed end to end. Silver iodide's lattice is close enough to ice to nucleate well above homogeneous freezing; once a crystal exists in a cloud holding supercooled liquid, the vapour-pressure difference between ice and water grows it at the droplets' expense until it falls out. That is the Bergeron–Findeisen process, and glaciogenic seeding tries to start it early, or start more of it. Frontier The size distribution that actually nucleates from a commercial flare was not measured until 2024, when Chen and colleagues at ETH Zürich found flare aerosol nucleating comparably to pure silver iodide in the 200–400 nm range. Eighty years of operations preceded the characterisation of the deployed material.

Established The most rigorous randomised trial the field has ever run returned a null, and the number the field quotes from it comes from elsewhere in the same paper. The Wyoming Weather Modification Pilot Project ran a six-year randomised crossover from 2008 to 2013 over the Medicine Bow and Sierra Madre ranges. Rasmussen and colleagues reported it in 2018 in the Journal of Applied Meteorology and Climatology, and the abstract states the outcome in its own words: “insufficient statistical evidence (p value of 0.28) to reject the null hypothesis that there was no effect.” Established The same abstract reports a separate arm — an ensemble modelling evaluation — giving roughly 5% enhancement from seedable clouds, about 1.5% additional annual precipitation across the ranges. Established Two instruments, two answers, reported honestly together. Citation kept the model number and dropped the null.

Established The null was predicted in advance, by the project's own design paper, and that is what makes it interpretable. Breed and colleagues published the WWMPP design in 2014 and reported that with refined precipitation data the experiment would need 236 randomised cases to detect a 10% change, or 110 cases to detect 15%. The experiment ran six winters. Established A design that needs 236 cases to see 10% cannot see 5% at all. Frontier The honest statement is therefore neither “seeding failed” nor “seeding works”: it is a power result, not a physics result, and the field's most-cited efficacy number is a model output the same paper's randomised arm could not confirm.

Established The denominator is almost never carried with the headline. Ritzman and colleagues asked how much of the target area's winter precipitation was even eligible under the project's own criteria: they were met on roughly 26% of winter days, and about 27–30% of winter precipitation fell from clouds meeting them. Established A 10% enhancement applied to under a third of the season is a low-single-digit change in basin water — the arithmetic that turns an impressive percentage into a quantity a water manager would struggle to see against natural variability. Established Geerts and colleagues, on the same project, put it plainly: the level of noise in natural systems compared with the magnitude of the signal makes verification extremely difficult.

Established The strongest physical result in the field's history is also its most over-read. The SNOWIE campaign, reported in PNAS in 2020, tracked three January 2017 cases over Idaho's Payette basin and isolated precipitation unambiguously attributable to seeding, persisting 25–160 minutes within the radar domain. Established Total water generated across the three cases: 123,220 to 339,540 cubic metres, uncertainties 20–45%. Established The cases were selected because natural precipitation was below 1 mm per hour — because nature was doing almost nothing. Frontier The upper figure is about 275 acre-feet, roughly one afternoon's flow of a modest river: a mechanism demonstration of the highest quality and a seasonal-yield measurement of no quality at all, routinely cited as both.

Established An independent chemical route to attribution exists, and it returns a much smaller number. Warburton and colleagues measured silver and indium in Sierra Nevada snowfall to separate seeded from natural crystals. In one eight-hour seeded storm at one site in the winter of 1992–93, the mean minimum enhancement in the number of crystals deposited was 16%; the associated precipitation enhancement was about 3%. Established Crystals are not water: the crystal-number enhancement is five times the precipitation enhancement, and quoting the crystal figure as a water figure inflates a claim fivefold. Speculative It is also the most promising unexploited attribution route in the subject, because it converts a statistical problem into an analytical-chemistry one.

Established The field has a long reanalysis tradition and it has generally run against the original claims. Rangno's 2025 review of the Colorado River Basin Pilot Project — the nation's most expensive randomised orographic experiment, run 1970–75 — gives four reasons it failed to replicate earlier claimed snowfall increases, among them statistical errors and seeding-agent transport problems. Frontier Rangno is the field's most persistent critic, his reanalyses have generally held up, and a reader should know the body of work has a consistent direction. Established The Israeli experiments are the field's most-cited positive randomised result and the subject of its most thorough demolition; the records for Gagin & Neumann 1981, Rangno & Hobbs 1995 and the 1982 Science report of a failed confirmation are verified, and none of their full texts was obtainable for this brief. The shape of the dispute is solid; no effect size from any of them is stated here.

Frontier Warm-cloud hygroscopic seeding may have a better modern evidence base than the glaciogenic work that dominates Western discussion. India's CAIPEEX experiment, reported in 2023, ran randomised hygroscopic seeding of convective clouds with 150 seeded and 122 unseeded samples — the largest modern randomised convective sample in the literature, and barely cited in Western efficacy debate. Established The reason is sociological: winter orographic glaciogenic seeding is what western United States water utilities buy. The geography of the evidence follows the geography of the customers.

Established Hail suppression is where the practice most exceeds the evidence, and the best available reanalysis says the sign may be wrong. Grossversuch IV, run in central Switzerland from 1977 to 1982 over roughly 1,300 square kilometres, is the best-designed randomised hail experiment ever conducted: 83 experimental days (37 seeded, 46 control), 253 cells (113 seeded, 140 not), silver iodide by Soviet Oblako rockets. Frontier In 2021 Auf der Maur and Germann re-evaluated the raw data in Atmosphere using permutation and bootstrap methods and found hail energy higher by approximately a factor of three when seeding, at p < 0.01. Frontier Their criticism of the 1986 original is specific: it replaced actual seeding-coverage values with a binary treatment indicator, log-transformed hail energy, and included unreliable meteorological predictors that worked against the finding. Established And the 1986 evaluation had itself observed that a majority of its evaluations suggested some trend to larger seeded hail energy and area — attributing that trend to the multiplicity of tests it ran.

Frontier The limits of that result matter as much as the result. It is one reanalysis of one experiment by one group, it has not been independently replicated, and this brief found no published rebuttal four years on — a fact about the literature, not an endorsement of the finding. Speculative A single reanalysis does not establish that hail seeding increases hail. What it establishes is that the sign of the most widely practised convective intervention on Earth is not known, and that the one experiment designed well enough to settle it points the wrong way when its raw data are analysed without transformation. Established China's 2021–2025 plan alone puts hail suppression over 580,000 square kilometres. Established The field's foundational 1977 review volume contains a full chapter on how to measure hail at the surface: the dependent variable was never well measured.

Established Cold-fog dispersal is the one unambiguous win, and it is unambiguous for a physical reason. Beckwith reported in 1965 that dissipating supercooled fog by aircraft seeding had been recognised since Schaefer's 1946 work, and that United Air Lines' dry-ice operations achieved over 80% success, enabling about 200 flights. Established At the 1984 Sarajevo Winter Olympics, eleven liquid-propane seedings produced nine successes, defined as raising visibility above the required minimum. Established Warm fog is the counter-case: thermodynamically stable, so clearing it means removing water or heating air, which is why wartime FIDO installations burned fuel along runways instead of seeding. Frontier Whether cold-fog dispersal is still practised at any airport today could not be established here.

Established Hurricane modification was tried seriously for twenty-one years and closed without demonstrating an effect. Project Cirrus seeded a hurricane off Cape Sable in 1947; the storm turned toward Savannah, and General Electric's resulting legal exposure is part of why the government took the programme over. Project Stormfury ran from 1962 to 1983, described in its own 1969 account as a project of Commerce and Defense for experimenting at modification of hurricanes, examining both cloud seeding and the inhibition of ocean evaporation. Established This brief could not obtain the full text of Willoughby and colleagues' 1985 chronicle, and states none of the numbers most often repeated from it — not the Hurricane Debbie wind reductions, not the supercooled-water finding, not the eyewall-replacement explanation. What the record supports alone is the shape: twenty-one years, two agencies, a close-out paper, no demonstrated effect.

Established The operational industry is far larger than the research base that would justify it, and China's is larger than everyone else's combined. Simon, McDonald and Brent's 2023 analysis compiles the public figures: 13.267 billion yuan invested since 2014, about US$400 million of it central; roughly 48,000 people employed; coverage of more than a third of China's land area; operations over 2.83 million square kilometres in 2014–2020; and a 2021–2025 plan extending rain and snow enhancement past 5.5 million. Established The northwestern programme deploys 208 kilograms of silver iodide a year; the same paper reports the United States and Canada combined dispersing about three tonnes annually. Frontier Its estimate of downwind reach is that seeding agents can affect clouds up to 200 kilometres from the target area, and the northwestern environmental impact statement estimated a 1% precipitation increase over non-target areas.

Established The operator with the most transparent published economics declines to state a seasonal yield. The United Arab Emirates flew 390 seeding sorties in 2020, averaging about an hour each at up to US$8,000 per flight hour, into a country with mean annual rainfall below 80 mm. Established Wehbe and colleagues — all at the UAE National Center of Meteorology or its rain-enhancement programme — nonetheless write that “there are still too many multiscale and region-specific unknowns to report a specific percentage for seasonal (and annual) increases in precipitation from cloud seeding operations.” Frontier A statement running that hard against the authors' interest carries weight. The operators are more careful than their advocates.

Frontier The single most consequential unanswered question in the subject is whether the 2021 Swiss hail reanalysis is right. Four years on it is unreplicated and unrebutted — no independent reanalysis of the Grossversuch IV raw data has appeared, and no published response defends the original evaluation against the permutation critique. Speculative A physical mechanism for the wrong sign is available: beneficial competition requires seeding to produce many small hail embryos that compete for the same supercooled water, and under-seeding a vigorously updrafted cell produces fewer and larger stones instead. Frontier That mechanism is textbook; what is missing is any modern randomised experiment with radar-derived hail kinetic energy as the endpoint.

Frontier The models and the trials disagree, and nobody has reconciled them. Mehdizadeh and colleagues modelled five Lake Tahoe events in 2024 with a snow-growth model driven by reanalysis and found seeded snowfall rates 24% higher on average, ranging from +6% to +37%. Established That is a model, not a trial, and it sits an order of magnitude away from the randomised record. Frontier The reconciliation test is specific and nobody has run it: a seeding model earns the right to be used as a counterfactual when it reproduces the null results, not only the positive ones. No published model has been asked to predict the Wyoming p-value.

Frontier Dose is a variable no operational programme optimises against measured output. Zhu and colleagues report a 2025 WRF simulation in which increasing the silver iodide dose raises rainfall and then lowers it — a non-monotonic response in which over-seeding suppresses precipitation. Speculative The venue is low-impact and the result is a single simulation, but the structure is the same non-monotonicity that appears in cirrus modification, and it implies that a programme can be seeding past its own optimum without any way of knowing.

Frontier The most interesting minority claim in the field is that the nulls are contaminated. Bigg and Turton reported in 1986 and 1988 that Australian seeding produced apparent effects lasting not only for many days but for many months; Long's 2001 review reports effects lasting hours to days and discusses the hypothesis that deposited silver iodide modifies, or causes, bacteria on plants to loft into the atmosphere as persistent ice nuclei. Established Rosinski published a formal objection in 1987 and it stands in the record alongside the claim. Frontier The reason this belongs at the frontier rather than the fringe is structural: if seeding has multi-day or multi-month persistence, every randomised crossover design in the field is contaminated, because control days follow seeded days at the same site, treatment leaks into control, and every estimate is biased toward zero. Speculative That is the enthusiast case in its only rigorous form, it makes a falsifiable prediction, and it has never been tested with modern ice-nucleating-particle instrumentation. It was not testable in 1988. It is testable now.

Frontier China's transboundary exposure has been analysed in international law for the first time, and the finding is procedural. Simon, McDonald and Brent conclude that China's domestic arrangements do not satisfy procedural obligations under customary international law: no transboundary environmental impact assessment, no notification or consultation with neighbours, no cumulative impact evaluation. Established The object of that analysis has no analogue — the Sky River project on the Qinghai–Tibetan Plateau is designed to move up to 5 billion cubic metres of water a year from the Yangtze system, has more than 500 ground generators installed, and carries a proposal to extend to tens of thousands over up to 1.6 million square kilometres. Frontier Domestic precedent exists: in the 2000s, provinces in Henan, Guangxi and Shaanxi accused one another of stealing precipitation.

Frontier Materials substitution is being driven by commodity prices rather than by atmospheric science. Finnegan and Ates propose aluminium oxide as a non-silver seeding nucleus, motivated explicitly by escalating silver prices. Speculative A change in the deployed nucleating material would invalidate what little dose–response characterisation exists, and there is no regulatory step at which that would be noticed.

Speculative Drone delivery is changing the cost structure without changing the evidence base. No peer-reviewed evaluation of drone-delivered seeding was obtainable for this brief, and the claim to be careful with is that cheaper delivery improves efficacy. It improves coverage. Established Coverage has never been the binding constraint.

Frontier Meanwhile the number is escaping into the tertiary literature as a fact. A 2026 review-style paper states that cloud seeding can augment precipitation under optimal conditions, often by 5–25% in targeted regions. Established That range originates as what operational aircraft programmes report about themselves, relayed through World Meteorological Organization compilations. Frontier Within three years it has become a stated efficacy in secondary sources, and this brief cites the 2026 paper only as evidence of the propagation, not as evidence about seeding.

Established The target is not “does seeding work”. It is a measurement standard that can attribute a stated quantity of water at the ground, over a season, to a seeding operation, to a precision sufficient for a water right to be transferred or a downwind claim to be adjudicated. Nothing like that exists for any weather-modification technique anywhere in the world.

Established The chain runs in this order. One: a validated seedable-window climatology for each operating basin — what fraction of precipitation is even eligible — published before the programme starts. Established This is the least-blocked link; the method exists and is cheap. Two: a dose–response curve for delivered nuclei, which requires characterising the in-plume size distribution of the operational generator and flare inventory. Frontier That is blocked industrially rather than scientifically: operational flares are commercial products with proprietary formulations and nobody has characterised what is actually deployed. Three: a closed physical chain from flare to gauge under ordinary conditions rather than ideal ones. Frontier SNOWIE worked because natural precipitation was below 1 mm per hour; repeating it in a real storm needs a discriminating tracer, not just radar.

Frontier Four: a chemically discriminating tracer at gauge-level detection limits — the 1990s silver-and-indium snowpack chemistry redone with current inductively coupled plasma mass spectrometry, a dense sampling network and a pre-registered analysis. Speculative Five: resolution of the persistence question, which is upstream of everything, because if persistence is real then links six and seven have to be redesigned. Six: a trial design immune to persistence contamination — paired-basin randomisation with spatial rather than temporal separation, or block randomisation at seasonal granularity.

Established Seven binds, and it is not a scientific problem. The requirement is a multi-basin randomised crossover accumulating at least 236 cases: four to six basins, twenty to thirty years, a common protocol, pre-registration, and a commitment to publish a null. Established The design has been public since 2014. Frontier The obstacle is that the only parties who fund seeding are the parties who buy it, and a well-powered trial is a mechanism for terminating one's own appropriation. No jurisdiction has funded one. Wyoming came closest and got p = 0.28. Established Downstream of that, link eight is whole-domain moisture-budget closure — addition versus redistribution — which is absent from the entire literature this brief could reach.

Established One result is upstream of the entire chain and it costs a season of instrumentation. If the Bigg and Turton persistence effect is real at even a fraction of its claimed duration, then the crossover randomisation used by every modern trial systematically leaks treatment into control, and the field's nulls are artefacts of its own design. Frontier Nothing further can be designed properly until that is settled, and settling it requires modern ice-nucleating-particle measurements downwind of seeded storms at one day, one week and one month against matched unseeded controls.

Frontier The second dependency is a model that has earned counterfactual status. Storm-resolving models are the only route to a seasonal counterfactual that does not require decades of randomisation, and they currently produce effects several times larger than the trials. Established The admissibility test is reproduction of the nulls. Until a model predicts a p = 0.28 outcome from the Wyoming configuration, its 24% is a hypothesis about clouds, not a measurement of them.

Frontier The third is the deployed inventory. Dose–response cannot be established for a material nobody has characterised, and the 2024 result that flare aerosol nucleates in the 200–400 nm range is the first measurement of its kind. Speculative A move to non-silver nuclei on price grounds would reset even that, and there is no regulatory step at which the substitution would be noticed, because no regulator asks what is in the flare.

Established Fourth and cheapest: a per-basin seedable-window climatology. Without it, any percentage a programme claims is a percentage of an unstated denominator, and the Wyoming figures show how large the correction is — roughly a quarter of days and under a third of winter precipitation eligible. Frontier Fifth, and the one dependency that is not technical: a pre-registration norm. Speculative Every reanalysis in this field has turned on analytical choices made after the data were in hand — which transform, which covariates, which cells count — and the 2021 Swiss result is precisely a demonstration that those choices can invert a conclusion. Until analysis plans are fixed before seeding starts, no future trial will settle anything either.

Established The cheapest high-value experiment available today is the persistence test, and nobody is running it. Instrument ice-nucleating-particle concentrations downwind of seeded storms at intervals of one day, one week and one month, against matched unseeded control sites, for one season. Frontier It is a direct replication of a 1988 result with instrumentation that did not exist then, and it either validates or invalidates the design assumption sitting under every trial in the field.

Frontier Second: SNOWIE without its selection criterion. Repeat the radar-and-microphysics attribution chain in storms producing ordinary precipitation rather than in storms chosen for producing almost none. Established This is the experiment that would show whether the physical chain is detectable when it matters, and it needs a discriminating tracer because radar alone cannot separate seeded from natural echo in a working storm.

Speculative Third: the tracer network. Revive the silver-and-indium snowpack chemistry at current detection limits, with a sampling network dense enough to integrate over a basin, and validate whether the crystal-number-to-water conversion — 16% to 3% in the 1996 measurement — is stable across conditions. Frontier If it is, attribution stops being a statistical problem.

Established Fourth: the trial that binds. Four to six basins, twenty to thirty years, common protocol, pre-registered analysis plan, a fixed commitment to publish a null, and at least 236 randomised cases. Frontier Fifth: whole-domain moisture-budget closure. Instrument the full inflow and outflow boundary of a seeded domain — aircraft, radiosonde, GNSS precipitable water, satellite — during a randomised campaign, with net basin water as the endpoint rather than target-area gauge totals. Speculative That is the experiment that answers the downwind question — addition or redistribution — and this brief found no attempt at it anywhere in the reachable literature. Frontier Sixth: a modern randomised hail experiment with radar-derived hail kinetic energy as the pre-registered endpoint, which is the only way to settle a sign that 580,000 square kilometres of operations currently assume. Speculative A cheaper interim step exists and would be worth doing first: an independent reanalysis of the Grossversuch IV raw data by a second group, with the analysis plan published before the data are opened. Established The dataset exists, the method is public, and the cost is a statistician's year.

Established Delivery is the mature part of the system and it is not where the difficulty lives. Ground-based acetone-and-silver-iodide generators, aircraft-mounted burn-in-place and ejectable flares, dry-ice pellet dispensers, liquid-propane dispensers for fog, and rockets for hail — Grossversuch IV used Soviet Oblako rockets — are all decades-proven hardware. Frontier Uncrewed aircraft have lowered cost per targeted cloud further, and the honest description of that gain is that it buys coverage, not efficacy.

Frontier The binding engineering problem is transport and dispersion between release and the supercooled zone. Super named it in 1990 as the key challenge in winter orographic seeding and it has not been retired: a generator on a valley floor is useless if the plume does not reach the layer where supercooled liquid exists, at the right time, at a useful concentration. Established The 2024 flare-size result matters here because it defines what fraction of the emitted mass is even capable of nucleating on arrival.

Speculative The measurement side is where the engineering investment has never been made. An attribution-capable programme needs, at minimum: dual-polarisation radar with hail kinetic-energy retrieval; a snowpack chemistry laboratory operating at parts-per-trillion silver and indium detection; a gauge and core-sampling network dense enough to integrate a basin; ice-nucleating-particle counters at control and downwind sites; and GNSS precipitable-water and radiosonde coverage of a domain boundary. Frontier None of that is exotic and none of it needs invention. Established All of it costs more than the seeding it would evaluate, which is the whole reason it has never been built. Frontier The ratio is the finding. A seeding season for a western basin is a few hundred thousand dollars; an instrument suite capable of attributing its output is an order of magnitude more, runs for decades, and produces a number the purchaser may not want. Speculative Any serious programme of weather modification is therefore an evaluation programme with a small seeding component attached, and no procurement anywhere is written that way.

Established The nearest neighbour is solar radiation modification, and the relationship runs in an unexpected direction. Climate Engineering holds the stratospheric aerosol case and identifies detection and attribution as its binding constraint; Geoengineering Governance holds the instruments and reaches the same conclusion. Frontier Weather modification supplies the base rate neither can supply for itself: a smaller, cheaper, more frequently repeated intervention, unattributed after eighty years.

Established Inadvertent modification is larger than deliberate modification and better documented. Changnon's urban precipitation work and the METROMEX tradition established that cities modify their own downwind rainfall without anyone intending it; ship tracks and aviation-induced cirrus do the same over oceans and flight corridors. Frontier Atmospheric Management owns the tropospheric-chemistry framing of that.

Speculative And there is an exotic use for the operational programmes that nobody has taken up. Aerosol–cloud interaction is the largest single uncertainty in the radiative forcing budget, and deliberate seeding is the only randomised aerosol–precipitation perturbation humans perform anywhere. Handwave Organised as a globally distributed randomised experiment on aerosol–cloud interaction, operational seeding would be scientifically valuable even if its water yield were exactly zero. Nobody organises them this way, and the data are not pooled, standardised or archived to a common protocol. Speculative It is plausibly the most under-exploited dataset in atmospheric science. Established Downstream, the water-supply neighbours are Water Infrastructure Megaprojects and Desert Greening, both facing the same question of what a marginal cubic metre is worth against storage, leak reduction and demand pricing — all of which have measurable outputs that seeding does not. Frontier The honest framing of the procurement case is that seeding is an option-value purchase, not an evidence purchase, and it should be argued that way: cheap in absolute terms, plausibly positive in expectation, and unmeasurable. Speculative Stated like that it can still be rational for a utility. Stated as a demonstrated yield, it is not.

Established The United States has known for fifty-three years who is modifying the weather and has never once required them to say whether it worked. The Weather Modification Reporting Act of 1972 requires notification to the National Oceanic and Atmospheric Administration at least ten days before activities including the release of gases, dusts, liquids or aerosols into the atmosphere. Established It creates a list. It creates no obligation to evaluate, no standard for evaluation, no data-quality requirement, and no consequence for a programme that never measures its effect.

Established Oversight is currently decreasing in at least one state, for reasons unrelated to cloud seeding. Florida's 2025 statute prohibiting geoengineering repealed the state's existing weather-modification licensing framework outright, as recorded in Geoengineering Governance. Frontier Whatever one thinks of the prohibition, a state acting against geoengineering deleted the only permitting regime it had for the intervention that is actually practised. That is governance by accident.

Established There is no metrology of weather modification. No standards body certifies a yield. The World Meteorological Organization issues statements on weather modification; it does not certify effects, and its compilations of what operators report about themselves have been read back as findings. Frontier A standards body publishing an evaluation protocol with a stated uncertainty and an acceptance threshold is the institutional link the technical chain runs into, and no body currently holds the mandate.

Established The treaty that people cite does not apply, and the reason is not the one usually given. The Environmental Modification Convention prohibits hostile use; the Conference of the Committee on Disarmament's interpretive understandings define “widespread” as encompassing an area of the scale of several hundred square kilometres and “long-lasting” as lasting for a period of months, or approximately a season. Established Those are not planetary thresholds. Grossversuch IV's target area was about 1,300 square kilometres over five seasons; China's hail-suppression coverage is 580,000. Frontier Ordinary operational weather modification clears ENMOD's size and duration thresholds comfortably. What keeps it outside the treaty is Article I's hostile-use requirement, not its scale. The treaty itself is held by Geoengineering Governance.

Speculative And no forum has jurisdiction over a downwind claim. Customary international law's no-harm rule might reach one, but requires the harm to be demonstrated, which returns to the attribution problem. Established A United States Government Accountability Office review of federal and state cloud-seeding activity was published in December 2024 and is the natural source for the industry's size and spend; it could not be retrieved for this brief and nothing is stated from it.

Frontier The distributional question is the live one, and it is blocked by measurement rather than by law. If seeding moves water rather than adding it, the losing party has no way to prove a loss and no forum in which to allege one. Established The Chinese northwestern environmental impact statement estimates a 1% precipitation increase over non-target areas — an extra-area gain — but the mechanism that produces extra-area gains can equally produce extra-area losses depending on trajectory and depletion. Established Chinese provinces have already accused one another of stealing precipitation. This is the cleanest case on the map of a technology whose liability regime is blocked by a metrological problem rather than a legal one.

Established The epistemic damage runs in both directions, and that is the fair way to state it. When a flood follows a seeding sortie, the attribution machinery that cannot demonstrate a 10% benefit also cannot demonstrate a 1000% harm. Frontier A field that has spent eighty years unable to prove its own effect is poorly placed to be blamed for a catastrophe — and equally poorly placed to reassure anyone, because it cannot show its own footprint either. Speculative Public consent for an intervention nobody can measure is being sought on the basis of trust in the operator, which is not a stable arrangement.

Frontier A programme can be designed to be unfalsifiable without anyone intending fraud. Seed when conditions are favourable, report an estimated yield from a model calibrated on favourable conditions, and evaluate against a historical baseline including unfavourable years: the reported yield is then a function of the selection rule rather than of the seeding. Established That is what happens by default when nobody imposes a design, and naming the structure is more useful than accusing anyone. Speculative On silver iodide itself, both sides overstate: three tonnes a year across the United States and Canada dispersed over continental areas is a small quantity, but the honest position is a monitoring commitment rather than a settled question, and the interesting risk is not toxicity but ecological ice-nucleus modification, which nobody has looked for.

Speculative The trajectory that matters is not better seeding. It is seeding at continental scale in shared airsheds with no attribution standard and no forum. China's stated plan covers rain and snow enhancement over more than 5.5 million square kilometres and hail suppression over 580,000; the Sky River concept proposes moving up to 5 billion cubic metres of water a year across the Qinghai–Tibetan Plateau, upwind of the most densely populated river basins on Earth. Frontier No regional framework exists for South and Southeast Asia comparable to the river-basin commissions that manage surface water.

Frontier The deeper civilizational point is a precedent rather than a risk. Weather modification demonstrates that a technology can be practised globally, funded publicly, and regulated by notification for eighty years without anyone establishing whether it does what it claims. Established That is not a failure of anyone's integrity; it is what happens when the only party with an incentive to fund the measurement is the party the measurement could defund. Speculative Every argument for governing solar radiation modification through monitoring and attribution has to explain why the easier version of that problem is still open.

Handwave The optimistic version deserves stating too. If the tracer route closes — parts-per-trillion chemistry, dense sampling, pre-registered analysis — weather modification becomes the first atmospheric intervention with a real metrology, and that metrology is portable: the same chain of custody that lets a basin bill a neighbour for stolen rain is the chain that would let a state be held to account for a stratospheric injection. Speculative On that reading the cheapest route to governable planetary-scale geoengineering runs through a snowpack chemistry problem in Wyoming, and costs a rounding error against any deployment budget. Handwave Nobody has proposed it in those terms, and it is the most under-priced item on this map.

These horizons track the attribution standard rather than the technology, because the technology is mature and the standard does not exist.

• 10 yr: Frontier The persistence question is answerable within one funded season and would be settled if anyone paid for it; the flare and generator inventory can be characterised in the same window. Speculative A modern randomised hail experiment with a radar kinetic-energy endpoint is technically ready and institutionally unlikely, because the countries running the largest hail programmes are not the countries running randomised trials. Frontier Expect operational expansion to continue faster than evaluation, and the 5–25% range to harden further in secondary literature. Established Expect also more national bans framed against geoengineering that incidentally remove what licensing exists, as Florida's did.
• 25 yr: Speculative A multi-basin randomised crossover started today would report inside this window and is the only route to a defensible seasonal number. Speculative Nothing currently funded resembles it. Frontier A more likely path is that trace-chemistry attribution matures first and makes the randomised trial partly unnecessary by converting attribution into measurement. Speculative Watch for the first insurance-industry evaluation of a hail-suppression programme: an insurer is the one plausible funder with an interest in a true answer rather than a favourable one, and hail is the one endpoint an insurer already prices.
• 50 yr: Speculative If an attribution standard exists by then, downwind claims become adjudicable and weather modification acquires a liability regime, which will do more to shape the industry than any efficacy result. Handwave If it does not, continental-scale programmes will have run for a century on self-reported yields, and the first serious interstate dispute over precipitation will be argued without evidence on either side.
• 100 / 250+ yr: Handwave The far case is not bigger seeding but a different object: an atmosphere instrumented densely enough that any deliberate perturbation carries a signature, deliberate or not. Handwave That capability would retire the weather-modification attribution problem, the solar-geoengineering detection problem and the treaty-verification problem in one move, and it is a sensing-and-computation problem rather than an atmospheric one.

• Depends on Ice-nucleation microphysics and the characterisation of deployed nucleating materials, which was first measured for a commercial flare in 2024; storm-resolving numerical weather prediction good enough to serve as a counterfactual, which requires reproducing the null results and not only the positive ones; trace-element analytical chemistry at parts-per-trillion detection limits for silver and indium, plus a sampling network dense enough to integrate a basin; dual-polarisation radar retrievals for precipitation and for hail kinetic energy, the endpoint hail suppression has never had; ice-nucleating-particle instrumentation capable of resolving the persistence question; GNSS precipitable water and radiosonde coverage sufficient to close a moisture budget around a seeded domain; and randomised experimental design with pre-registration, which is a statistical technology rather than an atmospheric one and is the piece the field has most consistently done without.
• Requires (not on this map) Two things block this subject and neither is a scientific result. The first is a written attribution standard — an evaluation protocol with a stated uncertainty and an acceptance threshold, issued by a body with the mandate to issue it — without which no downwind complaint can be filed, let alone adjudicated, across the shared airsheds where China, India and the Gulf states now operate at scale. No standards body currently holds that mandate, and the World Meteorological Organization compilations that get read as findings are collations of operator self-reports. The second is a funder willing to sustain a multi-basin randomised programme through two or three decades and at least 236 cases, including at least one null interim result. The design has been published since 2014 and no jurisdiction has bought it, because the only parties who fund seeding are the parties who buy it, and a well-powered trial is a mechanism for cancelling one's own appropriation.
• Enables An attribution standard for local atmospheric intervention, which is the precondition for a downwind liability regime, for transferable water rights derived from seeding, and for any monitoring-and-attribution regime over solar radiation modification — the harder problem that both neighbouring briefs name as their binding constraint; a validated aerosol-perturbation dataset that would constrain aerosol–cloud interaction, the largest single uncertainty in the radiative forcing budget, obtainable at near-zero marginal cost from programmes already running; and an evidence base against which continental-scale water-transfer proposals can be assessed rather than asserted.
• Adjacent Solar radiation modification, inadvertent urban and ship-track precipitation modification, contrail management, desalination and demand-side water management as competing supply options, agricultural hail insurance, and airport operations, where cold-fog dispersal remains the field's only uncontested product.

Established “The Wyoming randomised trial showed cloud seeding increases precipitation by about 5–10%.” This is the field's central citation error and correcting it is the most useful thing this brief does. The randomised arm returned p = 0.28 and did not reject the null; the 5% figure is the output of the ensemble modelling arm of the same paper, and the 1.5% annual figure is a derivation from that. Established Rasmussen and colleagues reported both arms honestly in one abstract. Citation kept the model number, dropped the null, and the number then grew.

Established “Cloud seeding produces 5–25% more precipitation — the WMO says so.” Wrong in attribution. The range is what operational aircraft programmes report about themselves, relayed via World Meteorological Organization compilations and quoted as such in the UAE operator's 2023 paper. Established Not a WMO finding, not a trial result, not a measured range. Frontier The same paper's authors, who run one of those programmes, decline to state a seasonal percentage at all.

Established “SNOWIE proved cloud seeding works.” Overstated in a specific and correctable way. SNOWIE proved the physical chain from nuclei to crystals to snow at the ground beyond reasonable dispute, which is a real and important result. Established What it measured was 123,220 to 339,540 cubic metres across three storms in one January, selected for near-zero natural precipitation, with 20–45% uncertainty. Not a seasonal yield, not a basin yield, and never claimed as either by its authors.

Frontier “Hail suppression reduces hail damage.” Not established, and the best available reanalysis points the other way: Grossversuch IV reanalysed with permutation methods on raw data shows hail energy roughly three times higher on seeded days at p < 0.01, and the 1986 original saw the same direction and attributed it to test multiplicity. Speculative Stated precisely, this is one unreplicated reanalysis of one experiment and it does not establish that seeding increases hail. Frontier What it establishes is that the opposite claim has no randomised support either, and that an industry operating over hundreds of thousands of square kilometres does so without knowing the sign of its own effect.

Established “Cloud seeding is a drought response.” Backwards, and structurally so. Seeding requires clouds already carrying water and, for glaciogenic seeding, supercooled liquid at suitable temperatures; only about 26% of winter days met the Wyoming criteria in a normal year. Established Yield is highest in wet years and lowest in dry ones — anticorrelated with the need. That is the reason seeding cannot be the drought instrument it is procured as.

Established “China's programme adds 79 billion tonnes of water a year.” That figure, and the associated claim of about 15% of annual national water consumption, are programme self-reports relayed in the 2023 legal analysis, not measurements. Established Nothing in that paper, or anywhere this brief could reach, independently attributes that water. Frontier Given that the best-instrumented randomised trial in the world could not resolve 10% in one basin over six years, a national-scale attribution over 2.83 million square kilometres is not a measurement anyone knows how to make.

Frontier “Cloud seeding costs a cent a cubic metre, thirty times cheaper than desalination.” An enthusiast-side error in which the number is real. The US$0.01–0.04 per cubic metre figure is a cost per unit of harvestable water, conditional on three stacked assumptions the paper states plainly: a 5–25% enhancement taken from self-reported operational claims, a 50% target rate, and a further roughly 50% loss to evaporation and soil retention. Established Strip any one and the number moves by a factor of two to five, while desalination's US$0.31 is an unconditional cost of delivered water. The comparison is not like for like and the authors do not claim it is.

Established “ENMOD prohibits weather modification.” It prohibits hostile use, and the size thresholds are the opposite of the obstacle people assume: several hundred square kilometres, and months or approximately a season, are cleared easily by an ordinary operational programme. Established Anyone citing the convention as governing civil cloud seeding is citing a treaty that does not apply, for a reason that has nothing to do with scale.

Established “Weather warfare is a conspiracy theory.” The general chemtrail claim is false and is answered in detail in Climate Engineering. But the historical programme is documented, and the accurate account is stranger and more useful than the invented one. Established The United States conducted weather-modification activities in Southeast Asia — Operation Popeye — an unclassified Department of Defense account reached Senate Foreign Relations Committee records, and the resulting revulsion is the direct cause of the Environmental Modification Convention. Frontier The same contemporaneous legal analysis records that a Department of Defense consultant claimed similar operations were attempted against Cuba in 1969–70 to induce drought, and that Pentagon sources denied it; the denial travels with the claim, because the point is not to assert the Cuban operation but that the question was live in the treaty's own negotiating record. Established And the first multilateral instrument ever proposed for deliberate atmospheric intervention was about weather rather than climate, and it was Soviet: a draft Convention on Weather Modification tabled in 1974, opened by Gromyko's letter to the Secretary-General that August.

Speculative The interesting middle, which neither the conspiracy nor the debunking reaches: a state today could enhance precipitation upwind of a rival at a scale clearing ENMOD's thresholds and no instrument would apply, because the intent is not hostile and no method exists to demonstrate the effect. Frontier The gap is not concealment. The lawful version is unmeasurable and therefore unaccountable. A better story than the conspiracy, from the same facts.

Established “Cloud seeding caused a named flood.” It did not, and the physics is worth stating rather than asserting: a sortie delivers grams to kilograms of nuclei into one cloud over minutes and can act only on water already condensing, while a flood-producing mesoscale convective system processes water at rates many orders of magnitude larger, driven by synoptic moisture convergence that seeding does not touch. Frontier The narrower point belongs to this brief: the machinery that cannot demonstrate a 10% benefit cannot demonstrate a catastrophic harm either.

Frontier Finally, a misconception about the sources themselves, which this brief is obliged to record. Much of what is stated above rests on publisher-deposited abstracts rather than full texts, because most of this subject's primary literature sat behind unreachable publishers while it was written. Established Where a number lives only in a paper's body — Stormfury's findings, the Israeli effect sizes, the December 2024 GAO review — this brief names the source and states no number. Speculative That is not cosmetic. The bibliographic tooling behind this brief was observed returning wrong documents under correct identifiers and mis-scaling a unit by nine orders of magnitude, and this category's subjects are governed by assessment reports compiled exactly that way. A citation that cannot be opened is a claim about a document, not a claim about the world.