Megafaunal Climate Engineering

Speculative If winter snow across Arctic ground can be held at the density a dense herd produces — which no landscape-scale measurement has established, and 7 says what would — the prize is about 398 Mha of permafrost soil, 4.0 million km², that the same model loses without the animals. The condition goes first because it is the whole argument. In the one pan-Arctic simulation that exists, permafrost area falls from 1,037 Mha to 406 Mha by the 2090s under RCP 8.5; with snow compacted as a dense herd compacts it, the loss is 233 Mha instead of 631 Mha, and the permafrost that survives stays below −4 °C rather than warming to −2.9 °C.

Frontier The prize is stated in hectares and degrees, not tonnes, and that gap is the honest limit of the case. The anchor study reports permafrost area and ground temperature; it does not report avoided carbon emissions, so the quantity a climate ministry would need — gigatonnes not released — is not a number this field has published. What can be said is the size of the pool it would act on: permafrost carbon shrinks the remaining budget for 2 °C by 8 per cent (1–25) where negative emissions are assumed feasible and by 25 per cent (5–63) after a 1 °C overshoot. A lever on that pool is worth building even at low expected value; a lever whose output nobody has converted into tonnes is not yet a mitigation option.

Established The costs are local, the benefits are global, and the one published costing shows revenue covering about a tenth of spending. Three experimental areas of 1,000 animals each were costed at US$114 million over ten years — US$3.8 million per area per year — against carbon revenues put at US$360,000 per area per year. The land is reindeer-herding land in Fennoscandia and Siberia and the labour is pastoral labour; the avoided warming accrues to everyone. That asymmetry is not a detail of implementation, it is the reason the intervention has no owner.

Handwave The upside is real only if a density of 114 animals per km² is reachable over millions of square kilometres, and today’s Arctic average is 5. That is a factor of 23, applied to a region the size of the circumpolar north, sustained for eighty years. The bottleneck list in 5 says why nobody has a route to it, and 15 says what the 80 per cent headline does and does not mean.

Established Megafaunal climate engineering is the deliberate use of large grazing animals as instruments against a geophysical target: ground temperature, snow structure and surface albedo, rather than species counts. The target is what separates it from ecological restoration as usually practised, which Biodiversity Restoration covers on its own terms. A restoration programme succeeds when the animals are there and persisting. This programme succeeds only if a physical quantity moves — frozen ground that would otherwise thaw stays frozen — and it can put the animals back, keep them alive, and still fail.

Established The primary mechanism is snow, not grass. Dry Arctic snow is one of the better insulators in the natural world: a deep, low-density winter snowpack decouples the ground from air temperatures that fall below −40 °C, so the soil never gets the winter cold it would otherwise receive. Animals walking, digging and lying on that snow break its structure and pack it. A thinner, denser pack conducts heat better, lets winter cold reach the soil, and drives the annual mean ground temperature down. Everything else in the proposal is secondary to this.

Frontier The secondary mechanism is a biome shift from moss-and-shrub tundra to grass-and-forb steppe, which would raise summer albedo and strip out the insulating moss layer. Sustained heavy grazing suppresses woody shrubs and mosses and favours graminoids; the resulting surface is brighter in summer, drier, and has less organic insulation above the mineral soil. This is the mechanism named in the Pleistocene Park programme’s own rationale, and it is also the mechanism that carries the most contested evidence, because the summer consequences of drying and warming the topsoil run in the opposite direction to the winter consequences of compacting the snow.

Frontier The third mechanism is nutrient and moisture cycling: dung, urine, trampling and the conversion of waterlogged tussock tundra into drained meadow. Drying a wet tundra surface cuts methane production, because methane in these systems comes from saturated anoxic soils. It also raises aerobic respiration of the carbon already in the topsoil, so the sign of the net greenhouse-gas effect is not obvious from the mechanism alone, and 3 shows the one flux campaign at the flagship site measuring both at once.

Established The flagship demonstration is Pleistocene Park near Chersky in northeastern Siberia: an enclosed area of 20 km² carrying nine major herbivore species, with reintroduction experiments begun in 1988. The programme lists reindeer, Yakutian horse, moose, bison, musk ox, Kalmykian cattle, sheep, camels and goats. Four of those nine are domestic livestock rather than Pleistocene fauna, which matters when the project is described as reconstructing an extinct ecosystem: what is being built is a grazing pressure, assembled from whatever animals survive cold and can be obtained.

Established Snow’s thermal conductivity rises steeply with its density, so compacting a snowpack cools the ground beneath it. This is not in dispute and it is not specific to herbivores — it is the same physics that makes a ploughed road verge freeze deeper than an undisturbed field. The anchor modelling study implements it by two parameter changes: the snow compaction rate is raised from −0.002 to −0.003 h⁻¹ and the maximum snow density from 300 to 450 kg m⁻³. Those two numbers carry the entire result.

Frontier At Pleistocene Park, grazed ground is 1.9 °C colder than ungrazed ground at 90 cm depth in the annual mean, with snow depth reduced by about half. This is the measurement the whole field rests on and it is the right sign and a large magnitude. It is also a comparison between one grazed area and one reference area at one site, accumulated over decades in an unreplicated design, which is why this brief flags it frontier rather than established: the effect is measured, the attribution to grazing rather than to any other difference between two places is not independently reproduced.

Frontier The pan-Arctic model result: permafrost area falls from 1,037 Mha to 406 Mha by 2090–2099 under RCP 8.5 in the control, a loss of 631 Mha; under the grazed snow treatment the loss is 233 Mha. The JSBACH land-surface scheme was run at 0.5 degree resolution with five snow layers, a bryophyte and lichen layer, seven soil layers and thermal depths reaching 53 m. In the control the surviving permafrost warms from −6.7 °C to −2.9 °C, a rise of 3.8 °C; in the treatment it stays below −4 °C on average, and at 4–10 m depth the treatment run is 2.1 °C colder than the control.

Speculative The paper’s own words for the headline are “theoretically 80% of current permafrost-affected soils (<10 m) is projected to remain until 2100”, and the Pleistocene Park density is described as a theoretical upper end. The arithmetic is consistent: 1,037 minus 233 leaves 804 Mha, which is 77.5 per cent of the starting area. The word doing the work is “theoretically”, and the assumption behind it is the subject of the next three paragraphs.

Speculative The grazed-snow treatment was applied to the entire pan-Arctic domain with no spatial restriction — every permafrost grid cell was treated as stocked at Pleistocene Park density for eighty years. A second fetch of the paper with that question put narrowly returned no sentence limiting the treatment to a subset of cells. This is a legitimate way to run a sensitivity experiment and an illegitimate way to read a policy number: the result is the ceiling of the mechanism, not a projection of any programme. Nobody claims to have a route to stocking 1,037 Mha.

Established The densities in play: Pleistocene Park runs at 114 individuals per km²; the present Arctic average is 5 per km². The paper’s two Swedish reference points are Vassijaure at 483 reindeer per km² during a winter feeding period, described by the authors as very high and possibly not reachable at a pan-Arctic scale, and Holmoen at 15 reindeer per km². The highest density in the source literature was achieved with supplementary winter feeding, which means it is not a pasture stocking rate at all — it is a feedlot rate, and the feed comes from outside the system.

Established The model included no grazing-induced vegetation change, no albedo change and no methane. The authors tested a change in moss turnover rate and found it negligible, and they state plainly that such first model experiments do not take into consideration all interactions between ecological and physical processes, naming changing vegetation type and cover as impacts on surface albedo and evapotranspiration that were left out. Every secondary mechanism the field advertises — brighter steppe, drier soils, altered methane — is absent from the number that advertises them.

Established The authors name their own two open questions: what the snow-depth-to-herbivore-density relationship is at landscape scale, and which herbivore density is reachable and useful. That is an unusually clean statement of where a result stops. The first question is the transfer function between an intervention and its physical effect; the second is the feasibility of the intervention. Neither has an answer in the literature obtained for this brief, six years after the paper.

Frontier The one growing-season flux campaign at Pleistocene Park found grazed ground with higher photosynthesis, higher respiration, near-identical net CO₂ exchange, and methane emissions about 44 per cent lower. Chamber measurements over 7–22 July 2019 at 68.51 N, 161.50 E gave gross primary productivity of 1.67–2.57 against 0.80–1.44 µmol C m⁻² s⁻¹ and ecosystem respiration of 0.95–1.20 against 0.49–0.72, leaving net ecosystem exchange similar across the two sites. Methane averaged 0.10 ± 0.028 at the grazed site against 0.18 ± 0.015 µmol C–CH₄ m⁻² s⁻¹ at the reference.

Established The same campaign found the grazed topsoil significantly warmer and an order of magnitude faster to track air temperature: a 4.3-hour lag against 40.7–86.7 hours at the ungrazed site. This is the summer penalty. Removing the moss and tussock layer couples the soil to the atmosphere in both directions, and in July the atmosphere is warm. The winter mechanism and the summer mechanism work against each other in the annual mean, and the annual mean is the only quantity that matters for permafrost.

Established That campaign covered 17 days, three plots at the grazed site within a six-metre radius and two plots at the reference about 50 m apart, with no pre-treatment baseline and no autumn or winter data. The authors say so themselves: the findings are qualitative trends for the peak growing season, absolute differences carry elevated uncertainty, and no inference can be made about the annual net ecosystem exchange or methane budget. A 44 per cent July methane reduction is not an annual methane reduction, and the paper does not claim it is.

Frontier The soil-carbon evidence from the same site is a single core per treatment. A pilot study of five sites across a grazing-intensity gradient in the thermokarst landscape near Chersky reported mean total organic carbon in the top 38 cm six times higher at the intensively grazed basin site than at the non-grazed basin site, and about twice as high in the upland minimum active layer. The design is one soil core per site. The authors state they cannot unambiguously distinguish previous conditions at their sites from the herbivory impact, and that missing core replication combined with expected spatial heterogeneity means the interpretation needs further data for validation.

Established An earlier submission of that soil-carbon work to Biogeosciences was not accepted after review, with reviewers identifying insufficient replication. The preprint record remains open and states that the revision was not accepted; the work was subsequently published in Frontiers in Environmental Science in 2022. This is worth stating precisely because it is the kind of fact that vanishes from citation chains: the central soil-carbon result of megafaunal climate engineering failed peer review at one journal on exactly the ground that 7 says has to be fixed.

Frontier A 2024 biomarker study across 23 Arctic sites found lipid signatures consistent with less-decomposed organic matter under heavier grazing, and reported them as trends that are not statistically well supported. The design sampled a grazing-intensity gradient from exclosure to intensive pasture in northeastern Siberia and northern Finland, yielding 58 subsamples from 23 sites, with no replicate samples. The authors conclude that significant effects may require prolonged, intensive grazing pressure.

Frontier The strongest counter-result is from Fennoscandian peatlands, where higher reindeer density is associated with less permafrost, not more. A study of 17 peatlands at 69–70 N compared 25 low-density and 25 high-density transects, with 24 hummocks examined at each of four detailed sites. Hummocks retaining permafrost were a significantly lower proportion at high density, mean hummock area was 2 m² at high-density sites against 65 m² at low-density sites, snow was deeper at high-density sites particularly at peatland edges, and standing surface water tended to increase. The densities compared were 1.6–2.0 against 2.2–2.6 reindeer per km² regionally, two orders of magnitude below Pleistocene Park, and the adverse effect appeared anyway. The authors’ conclusion is that in the lowlands they studied the cascading effects of summer trampling may well offset any climate protection gained by browsing, and they recommend reducing rather than increasing densities there. This is not a marginal case: permafrost peatlands hold about 185 ± 66 Pg of soil carbon and make up roughly half of the 3.7 ± 0.5 million km² of northern peatlands.

Established The general vegetation physics favours the herbivore argument: tall shrubs trap snow and warm the ground in winter more than their summer shading cools it, so suppressing shrubs should cool soil on the annual mean. A review of tundra vegetation change and permafrost states that increases in vegetation cover and height mitigate thaw in summer through shading but raise annual soil temperatures through snow-related winter warming, with the winter effect of increasing shrub height larger than the summer effect. Grazers suppress shrubs. That is the cleanest theoretical support the intervention has, and it is independent of the Pleistocene Park measurements.

Frontier The systematic-review evidence on what reindeer actually do to Arctic vegetation is thinner and more equivocal than the rewilding literature implies. A systematic review screening over 6,000 records found 40 articles containing 41 independent studies, about two-thirds from Fennoscandia. Effects on lichens were consistently negative; effects on forbs, graminoids, woody species and bryophytes were weak or non-significant. The reviewers name inadequate replication, pseudoreplication, confounding and short duration among the study weaknesses.

Established A fifty-year natural experiment across the Finland–Norway reindeer fence found summer grazing shortened willow and favoured sedges, with little effect on total bryophyte cover, and concluded that tundra wetlands can withstand moderately high grazing pressure sustained over several decades. The contrast is year-round grazing on the Finnish side against winter-only grazing on the Norwegian side, with four-year exclosures added. It is the longest grazing contrast in the Arctic literature obtained here, and it is a boundary rather than a randomised treatment.

Established Arctic field science is spatially concentrated to a degree that makes any pan-Arctic extrapolation fragile: 31 per cent of study citations across 1,840 Arctic field studies come from sites within 50 km of just two research stations, Toolik Lake and Abisko. The Canadian high-Arctic archipelago and the Russian Arctic coastline are named as major understudied zones, and the authors warn that the sampling and citation pattern may bias the scientific consensuses that underpin attempts to predict and mitigate climate change in the region. A model applied to 1,037 Mha is being evaluated against plots that cluster in two places.

Frontier Whether herbivores made the mammoth steppe, or the steppe made the herbivores, is unsettled in the palaeo record, and the two sides are named. Sedimentary ancient DNA from central Yukon shows megafaunal signals highest between 30,000 and 15,000 years ago, mammoth DNA declining from about 20,000 years ago, woody shrubs rising after about 13,500 years ago and ecosystem turnover between 13,500 and 10,000 years ago — an ordering in which herbivore loss precedes shrub expansion. Monteath and colleagues argue the reverse for eastern Beringia. The premise that a grazing regime can hold a biome in a grass state depends on which ordering is right, and this brief could not open the second paper.

Established The permafrost being defended is degrading on schedule: active-layer thickness increased significantly at 55 per cent of Arctic monitoring sites at 0.8 ± 0.8 cm per year over 2000–2024. That analysis covers 156 monitoring sites with at least ten years of record, and reports much faster deepening in European mountains at 10.3 ± 9.9 cm per year. Thawing degree-days is the primary driver, and sites with thick organic layers and high ground ice show smaller increases. Published in August 2026, it is the most current measurement of the problem this intervention addresses.

Established The southern permafrost margin is on a shorter clock than the intervention is. Projections of climate space for permafrost peatlands put all of Fennoscandia beyond climatic suitability by 2040 under moderate-to-high warming, with 37.0–39.5 Gt of carbon in the affected region. Palsa mires in northern Sweden lost 30, 35 and 54 per cent of their lateral extent over 61 years, at decay rates of −0.58 to −1.25 per cent per year, with the 1994–2016 rate twice the 1955–1994 rate and winter precipitation the strongest single predictor.

Frontier That last finding cuts both ways and is the most interesting number in this brief. If winter precipitation is the strongest predictor of palsa collapse, the winter snowpack really is the controlling variable at the margin, which is precisely the variable herbivores act on. It is also the margin where the peatland counter-result was measured. The same physics that makes the mechanism plausible makes the place where it would matter most the place where trampling does the most damage.

Frontier The first open question is the one the anchor paper asks: what is the snow-depth-to-herbivore-density relationship at landscape scale? Everything downstream is a transfer function. At plot scale, Pleistocene Park reports roughly half the snow depth at 114 animals per km². Nobody has published the curve between 5 and 114 animals per km² over a landscape, so nobody can say whether the effect is linear, threshold-like, or saturating at a density well below the one modelled. If it saturates at 30 per km² the programme becomes far cheaper; if it is threshold-like at 100 per km², it becomes impossible.

Speculative The second is whether any achievable density does anything at all outside enclosures. Free-ranging herds do not distribute themselves uniformly; they concentrate on preferred forage, along waterways, and away from predators. A landscape mean of 30 animals per km² with a strongly clustered distribution may compact snow over five per cent of the area and leave the rest untouched, while the model assumes the treatment everywhere. No published work obtained here measures the spatial coverage of snow compaction under free-ranging herds.

Frontier The third is the sign of the annual mean where summer warming and winter cooling both occur. The July campaign at Pleistocene Park measured warmer, faster-responding topsoil on grazed ground; the annual measurement at the same site records 1.9 °C colder ground at 90 cm. Those are consistent — the winter term can dominate — but they are from different depths and different instruments, and there is no published year-round energy budget that decomposes the annual difference into its seasonal parts at any site.

Frontier The fourth is the annual greenhouse-gas balance, which nobody has measured. Peak-season methane was 44 per cent lower on grazed ground and net CO₂ exchange was similar, but shoulder-season and winter fluxes were not measured and the authors explicitly decline to infer an annual budget. Winter CO₂ efflux from Arctic soils is a large term, and a grazed surface with a thinner snowpack and a colder, but also more thermally coupled, soil could plausibly move it in either direction.

Frontier The fifth is the upland-lowland divergence, which currently splits the evidence base cleanly in two. Dry upland thermokarst sites near Chersky give the positive results; wet Fennoscandian peatlands give the negative ones. Both are real measurements. If the divergence is genuine and terrain-controlled, then megafaunal climate engineering is a tool for a subset of the permafrost domain and the pan-Arctic number is meaningless as stated. Nobody has mapped the fraction of permafrost ground on each side of that line.

Speculative The sixth is whether the biome shift is attainable at all on a human timescale. The rewilding-carbon literature that quotes the 80 per cent figure concedes that scaling it up holds large uncertainties because it requires rapid population build-up and an ecosystem regime shift from shrub and tree dominance to steppe vegetation. Regime shifts in slow-growing cold systems are not reliably reversible on command, and the palaeo disagreement in 3 is precisely a disagreement about whether grazing can force one.

Speculative The seventh is albedo, which is quoted as a mechanism and quantified almost nowhere. The claim that reindeer grazing raises summer albedo by reducing shrub abundance rests on a small number of studies — Te Beest and colleagues in 2016 and Cohen and colleagues in 2013 are the ones repeatedly cited — that this brief could not open. The anchor model contains no albedo response to grazing at all. An albedo term would add to the case; its absence from the headline number means the headline number is not inflated by it.

Handwave The eighth is whether the intervention is stable without permanent management. A herd at 114 animals per km² is above any natural carrying capacity documented in the modern Arctic; the highest reference density in the anchor paper was sustained by winter feeding. If the density cannot be held without inputs, the permafrost benefit is a function of a continuing subsidy, and the intervention is closer to a perpetual public works programme than to an ecosystem.

Established The binding bottleneck is animals: there is no source of the number of large herbivores the mechanism needs, at any price. Getting from 5 to 114 animals per km² across even a tenth of a 1,037 Mha domain is a demand in the tens of millions of head. The published costing in this field covers three experimental areas of 1,000 animals each. The gap between the demonstration unit and the deployment unit is four to five orders of magnitude, and unlike an industrial technology the unit cannot be made faster by building a bigger factory — it is rate-limited by gestation and by winter survival.

Frontier Winter feeding is the hidden input that makes the highest quoted densities possible, and it inverts the carbon accounting. The 483 reindeer per km² figure was recorded during a winter feeding period. Feed has to be grown, harvested and hauled to sites without roads, and no published assessment of megafaunal climate engineering obtained here nets off the emissions cost of doing so. A permafrost-protection scheme that runs on trucked fodder is not obviously a net sink.

Established Methane from the animals is a real subtraction and it has been quantified once, at the level of the whole rewilding proposition. Restoring large herbivores to a 1500 AD baseline is estimated to add 11.7–13.4 MtCH₄ per year, or 292–335 MtCO₂e per year, to current emissions. That estimate is for global rewilding, not for an Arctic programme, and the Arctic share would be smaller; but it is the right order of magnitude to matter against an Arctic benefit that has never been expressed in tonnes at all.

Frontier Measurement in winter is the technical bottleneck that keeps the central question open. The quantity that decides the case is the annual mean ground temperature and the annual greenhouse-gas budget, and the season that carries the mechanism is the season when chambers freeze shut, eddy-covariance flux is small and noisy relative to instrument drift, snow pits destroy the pack they measure, and sites are unreachable for months. Every flux number in the Arctic grazing literature obtained for this brief comes from the growing season.

Established Snow density itself is poorly observed at the scale the model needs. The model changes maximum snow density from 300 to 450 kg m⁻³ to represent grazing; validating that change requires landscape-scale snow density fields under known stocking, which requires either dense in-situ profiling or a remote-sensing retrieval of snow density that does not currently exist at the required accuracy. This is the measurement bottleneck behind the transfer function in 4.

Frontier Verification for payment is a separate and harder bottleneck than measurement for science. The product here is an avoided emission from a soil carbon pool whose baseline release rate is itself uncertain, over a century, in a location nobody visits. Carbon Removal Verification covers the general problem; the Arctic version is worse, because the counterfactual is a model output and the crediting period outlasts every institution that would issue the credit.

Established The scale mismatch between the evidence and the claim is itself a bottleneck. The claim is made on a 0.5 degree grid; the evidence is chambers a metre across, cores a few centimetres wide, and one enclosure of 20 km². There is no intermediate-scale observation — nothing at the scale of a herding district with instrumentation adequate to close an energy budget — and building one is what 7 asks for.

Established The field depends on snow physics it does not do itself. The whole intervention is a claim about the thermal conductivity of a seasonal snowpack as a function of density, layering and depth hoar. Land-surface schemes represent this with a small number of parameters; the anchor result is the difference between two settings of two of them. Improved snow microstructure physics, and observational constraint on it under trampling, would change the answer more than any ecological result would.

Frontier It depends on Earth-system models learning to represent fauna at all. The anchor authors call for the integration of fauna dynamics into complex Earth system models, and that is not a small ask: it means a herbivore population module coupled to forage availability, snow, predation and management, with two-way feedback onto vegetation and the surface energy balance. Nothing in the coupled-model literature obtained here does this, which is why the headline number came from a prescribed parameter change rather than a simulated herd.

Established It depends on permafrost carbon inventories that remain the largest source of uncertainty in the payoff. The rewilding-carbon literature quotes roughly 500 Gt of carbon in the Yedoma domain; permafrost peatlands alone are put at 185 ± 66 Pg. A benefit expressed as avoided release inherits that uncertainty before any modelling error is added, and the anchor study sidesteps it by reporting area and temperature instead.

Established It depends on a pan-Arctic observing network that does not exist outside two clusters. With 31 per cent of Arctic study citations drawn from within 50 km of Toolik Lake and Abisko, and the Russian Arctic coastline named as a major gap, the intervention’s evidence base and its deployment domain barely overlap. The flagship grazing site is in northeastern Siberia, and the most recent primary grazing flux data from it obtained for this brief are from July 2019.

Frontier It depends on palaeoecology to establish that a grazing-maintained grass biome is a stable state and not a transient. Ancient environmental DNA and pollen records are the only evidence available on whether heavy grazing can hold a cold biome out of shrub dominance for millennia. The central Yukon sedimentary DNA record is consistent with it; the eastern Beringia shrub-first argument is not. This is a dependency on a literature that is actively disagreeing with itself, and the 2025 synthesis of late Quaternary environmental change in eastern Beringia could not be opened for this brief.

Speculative The decisive experiment is a replicated, randomised, multi-site grazing exclosure network across a permafrost gradient, instrumented for ground temperature, snow density and year-round greenhouse-gas flux, run for a decade. It is decisive because every competing claim in this brief is a claim about the annual mean at landscape scale, and every existing measurement is either a single site, a single core, a 17-day campaign, or a model. Nothing else on the list of possible studies moves the assessment as far.

Frontier Exclosures rather than stocking is the design choice that makes it affordable and randomisable. Introducing herds costs animals, fencing, transport and consent at every site and takes years before densities are reached; fencing animals out of blocks within existing herding districts uses grazing pressure that is already there, can be allocated at random, and starts producing a contrast in the first winter. Its cost is fence and instruments, a different order of magnitude from the US$114 million costed for three 1,000-animal areas.

Frontier The gradient has to span terrain, not just latitude, because the terrain split is where the evidence currently contradicts itself. A defensible network needs site clusters in continuous, discontinuous and sporadic permafrost, and within each cluster both dry upland and wet peatland or thermokarst terrain, because the positive results come from the former and the negative from the latter. Five or more randomised paired blocks per cluster, each at hectare scale, is the minimum that survives the pseudoreplication criticism levelled at the existing literature.

Established The instrumentation is standard and none of it needs inventing. Thermistor strings to 10 m give the ground temperature profile at the depths the model reports; snow density profiles through the winter give the state variable the mechanism acts on; automated chambers plus eddy covariance give CO₂ and CH₄ through the shoulder seasons and winter, which is where every existing dataset stops. The problem is power, servicing and instrument survival, not measurement principle.

Speculative A stocking-density arm is what turns the network from a yes-or-no test into the transfer function the field actually needs. Running blocks at several densities across the range from the current Arctic average of 5 per km² to the Pleistocene Park value of 114 per km² would estimate the snow-depth-versus-density relationship at landscape scale that the anchor authors name as their first open question. Without that curve, no stocking target can be derived from any climate objective.

Established A decade is the minimum duration because that is the response time of the quantity being measured. Ground temperature at several metres integrates over years; active-layer trends are detected in records averaging 20 years; the existing Pleistocene Park difference accumulated over decades. A three-year project would measure snow and summer fluxes and would not measure what the intervention is for.

Frontier The result that would most change this brief’s assessment is a multi-site annual mean ground-temperature benefit that survives in peatland as well as upland, with a year-round greenhouse-gas budget that does not cancel it. That would move the central claim from speculative to frontier and make the pan-Arctic model something other than a sensitivity experiment. The symmetric result — benefit confined to dry uplands, cancelled or reversed in wet terrain — would retire the pan-Arctic framing and leave a defensible local tool.

Established Two natural experiments are already running and neither is instrumented for this question. The Fennoscandian herding districts provide a standing contrast in reindeer density that has already produced the peatland counter-result, and the Finland–Norway fence provides a fifty-year contrast between year-round and winter-only grazing. Both are cheap to instrument relative to a new network, and adding thermistor strings and winter flux to existing density contrasts is the fastest partial answer available.

Speculative A before-after-control-impact installation at a new site would resolve the counterfactual problem that the flagship site cannot resolve retrospectively. The flux study at Pleistocene Park states that quantitative pre-treatment information for the grazed ecosystem is lacking, which no amount of later measurement can repair. Instrumenting ground that is about to be stocked, for two years before stocking, costs almost nothing and is the single cheapest methodological improvement available to this field.

Handwave Nobody has funded any of it, and no costed proposal for such a network appears in the literature obtained for this brief. The money in this space is venture capital attached to de-extinction and philanthropy attached to rewilding, neither of which buys thermistor strings in Siberian peatland for ten years. That is an institutional fact rather than a scientific one, and 10 says why it binds.

Established The hardware of this technology is animals, and that changes every engineering requirement. A herd cannot be commissioned, tested to specification, switched off, or shipped in a container. Its performance depends on forage, weather, disease and behaviour, and it degrades to zero if management stops for one winter. Every other approach in Climate Engineering has a failure mode you can write a maintenance schedule against; this one has a mortality rate.

Frontier Winter logistics dominate the cost structure and are the reason the demonstration unit is 1,000 animals rather than 100,000. Transporting large herbivores to roadless Arctic sites, holding them through a first winter in unfamiliar terrain, and supplying fodder if the range cannot yet carry them are the operations that the US$114 million for three 1,000-animal areas over ten years is mostly buying. Arctic Engineering carries the general problem of building and operating in this environment.

Established Fencing at deployment scale is not a scaled-up version of fencing at park scale. The 20 km² enclosure at Pleistocene Park is a manageable perimeter; a herding district is not, and a fenced Arctic at continental scale would obstruct caribou and reindeer migration, which is the ecological objection in 11 rather than an engineering one. Unfenced deployment solves that and reintroduces the spatial-coverage problem in 4.

Frontier Instrument survival is the engineering requirement that decides whether the decisive experiment can be run. Autochambers and eddy-covariance towers must operate unattended at temperatures below −40 °C, through months of darkness, on power that is either solar-impossible in winter or fuel that has to be flown in, with telemetry over satellite, in the presence of the large animals whose effect is being measured. These are solved problems individually and an unsolved integration problem at network scale.

Established Animal husbandry at these densities is the discipline the programme most obviously lacks and least obviously acknowledges. Mixed herds of bison, horse, musk ox, reindeer and domestic cattle at 114 animals per km² raise parasite load, disease transmission and forage competition questions that are routine in temperate ranching and unstudied here. The programme’s own species list — which includes camels, sheep and goats — is evidence that composition has so far been driven by availability.

Frontier De-extinction is the most visible adjacent technology and the least load-bearing one. De-Extinction Technologies covers the biology. The relevant point here is that the mechanism in 2 needs mass and hooves, not taxonomy: a Yakutian horse compacts snow as well as a mammoth would and is available now. A de-extinction programme is adjacent to megafaunal climate engineering as a source of attention and capital, not as a source of the intervention.

Speculative Colossal Biosciences states that it is editing 65 genes to make a cold-adapted elephant cell line and implanting embryos into Asian or African elephant surrogates, and gives no date for first calves and no release numbers. (vendor) The company’s mammoth page states that restoring mammoths could help protect the Arctic’s permafrost, and cites 1,600 billion metric tonnes of carbon in permafrost. The pages obtained for this brief contain no statement of how many animals would be released, where, or on what timetable. Section 15 does the arithmetic on why that matters.

Speculative Genetic engineering of extant species is the more plausible biological adjacency and nobody is doing it for this purpose. Cold tolerance, forage efficiency and parasite resistance in bison or horses would relieve the winter-survival constraint in 5 far more directly than recreating an extinct species would. Genetic Engineering covers the tools; no programme obtained here applies them to Arctic grazers.

Frontier The intervention is a member of the albedo family as much as the carbon family, which puts it adjacent to solar climate intervention. A brighter summer steppe and a less insulating winter pack are surface energy-balance changes, and the governance questions they raise are the ones in Geoengineering Governance rather than the ones in conservation policy. Inadvertent Geoengineering is the mirror case: land-use change already moves Arctic albedo without anyone intending it.

Frontier Methane is the adjacency that could flip the sign. The July measurement at Pleistocene Park had grazed ground emitting about 44 per cent less methane than the reference, while the animals themselves emit methane enterically. Atmospheric Methane Removal treats the sink side of this; for this brief the point is that a grazing programme is simultaneously a methane source and a methane-sink enhancement, and no study obtained here has netted the two at the same site.

Established The carbon market is the adjacency that would decide deployment, and it has no methodology for this. Carbon Capture at Scale and Carbon Removal Verification between them cover what a creditable tonne requires. Avoided permafrost emission from a modelled counterfactual over a century is at the far end of everything those briefs treat as difficult.

Established The single most important institutional fact is that the land is already occupied and used, mostly by reindeer-herding peoples. Fennoscandian herding districts and Siberian herding territories are the terrain this intervention would be conducted on, and the animals it would use are in many cases already theirs. The peatland counter-result was measured across exactly such districts. Any programme is therefore a change to an existing pastoral economy, negotiated with its holders, rather than a deployment onto empty ground.

Frontier Free, prior and informed consent is a requirement here in a stronger form than in most climate interventions, because the intervention is a stocking-rate instruction. Telling herders to run more animals than their range and their traditions support, for a benefit that accrues globally, is a proposition with a long and unhappy history in Arctic administration. No consent framework for climate-motivated stocking changes appears in the material obtained for this brief.

Established Most of the Yedoma domain and the flagship experiment are in Russia, and the observational record reflects the access situation. The most recent primary grazing-flux campaign at Pleistocene Park obtained for this brief is from July 2019, and the Russian Arctic coastline is named in the sampling-bias literature as one of the two major understudied zones. A research programme whose decisive experiment needs a decade of instrumented sites across the permafrost gradient has to solve an access problem before it solves a science problem.

Frontier No crediting methodology exists for avoided permafrost emissions, which means there is no revenue mechanism and therefore no private-sector route. The one published costing pairs US$3.8 million per area per year of expenditure with US$360,000 per area per year of carbon revenue. Even taking that revenue figure at face value, the programme is about nine-tenths grant-funded at the demonstration scale, and the deployment scale has no costing at all.

Speculative The institutional home is genuinely ambiguous, and that ambiguity is doing real damage. A conservation ministry will not fund an intervention whose objective is ground temperature; a climate ministry will not fund one whose evidence is ecological; an Arctic council has no mandate over stocking rates. The proposal therefore sits between Biodiversity Restoration and Climate Engineering and is claimed by neither budget.

Established The rewilding-as-climate-solution literature is itself moving faster than its measurement base. A 2025 commentary presenting the 12,500 km European Green Belt across 24 countries as a prototype for grounding nature-based solutions in rewilding contains no quantified climate outcome. That is typical rather than exceptional, and it is why this brief separates the snow physics, which is measured, from the climate claim, which is modelled.

Established The clearest ethical cost is animal welfare at densities chosen for a physical effect rather than for the animals. A stocking rate of 114 individuals per km² is selected because it compacts snow, not because it is what the range supports; the highest density in the source literature was sustained by winter feeding. Deliberately holding large mammals above carrying capacity in an Arctic winter is a welfare decision, and it is being made for a climate objective.

Frontier The biodiversity cost is real and measured: lichens decline consistently under reindeer grazing. The systematic review found lichen effects consistently negative while effects on forbs, graminoids, woody species and bryophytes were weak or non-significant. Lichen is winter forage and a defining component of Arctic ground communities. A programme justified as ecological restoration that reliably reduces one of the region’s signature groups is not doing restoration, and should not be described as if it were.

Established The intervention introduces species that were never part of the system it claims to restore. The flagship site’s own list includes Kalmykian cattle, sheep, camels and goats. These are chosen for hardiness and availability. That is a defensible engineering decision and an indefensible restoration claim, and the two framings appear in the same programme literature.

Frontier Indigenous rights are not a side constraint here; they are the governing consideration on most of the candidate land. Reindeer husbandry is the livelihood and in many cases the legal basis of land use across Fennoscandia and much of Siberia. The peatland study that produced the strongest negative result recommends reducing densities for climate reasons, which is a stocking instruction in the opposite direction from the rewilding proposal — and herders would receive both.

Speculative De-extinction attaches an ethical load to this field that the field’s own mechanism does not require. Creating a gene-edited elephant and gestating it in an Asian or African elephant surrogate raises questions about the surrogate, the founder animal and the welfare of a social mammal without conspecifics. None of that is necessary for snow compaction. A brief that lets the two be conflated is doing the reader a disservice in both directions.

Speculative The honest ethical summary is that the intervention asks a small number of Arctic communities to change how they live for a benefit that is global, unquantified in tonnes, and would be verified by outsiders. That arrangement has failed often enough in conservation that the burden of proof sits with the proposers. The strongest version of the programme is one designed with herders as principals rather than as implementers, and no such design appears in the literature obtained here.

Speculative If it works, this is the first case of a biological population deployed as an instrument against a geophysical variable at continental scale. That is a different category from farming, forestry or conservation: the animals would be infrastructure, with a performance specification, a monitoring regime and a maintenance budget. Whether a civilisation can hold a herd to a specification for a century is an open institutional question with no precedent in either conservation or engineering.

Frontier The commitment is perpetual, and that is the property that distinguishes it from most climate interventions. A direct-air-capture plant that shuts down stops removing carbon; a herd that is not maintained disperses, crashes or is eaten, and the snow deepens again. The permafrost preserved is only preserved while the density is held, which means the intervention creates an obligation that outlasts every political cycle that could authorise it.

Established The pool being defended is large enough to matter to the whole emissions arithmetic. Permafrost carbon release reduces the remaining budget for 2 °C by 8 per cent (1–25) where negative emissions are assumed feasible, 13 per cent (2–34) where they are not, and 25 per cent (5–63) after a 1 °C overshoot; for 1.5 °C the reduction ranges from about 10 per cent to more than 100. Any lever on that term is worth an experiment even at low probability of success.

Frontier There is a version of the outcome where the science is vindicated and the deployment never happens, and it is the most likely one. The mechanism can be real, the annual-mean benefit can survive replication, and the programme can still fail on animals, consent and money — the three constraints in 5 and 10 that no experimental result addresses. That outcome would still be worth the experiment, because it would settle whether Arctic land management generally should weight winter snow structure as heavily as this argument implies.

These horizons track what could be measured and what could be built, separately, because in this subject the two diverge sharply: the science is tractable on a decade and the deployment is not tractable on any horizon anyone has costed.

• 10 yr: Frontier Expect additional single-site and gradient studies of the Pleistocene Park type, and expect them to continue reporting trends rather than tested differences unless the replicated network in 7 is funded. Frontier Expect at least one year-round greenhouse-gas budget from a grazed Arctic site, because the instrumentation exists and the gap is now named in every paper. Speculative Expect the snow-depth-versus-density transfer function to remain unpublished at landscape scale, because nobody owns the question. Established Expect permafrost degradation to continue at the rates in 3 regardless. Speculative Expect no mammoth calf, and expect the de-extinction programme to remain the most-reported part of this subject.
• 25 yr: Speculative If the replicated network is funded in the next few years, this is the horizon on which the upland-lowland question in 4 is settled and the pan-Arctic framing either survives or is retired. Speculative A crediting methodology for avoided permafrost emission is conceivable on this horizon and would be the single largest change to the field’s prospects. Handwave Deployment beyond demonstration areas on this horizon requires an animal supply nobody has described.
• 50 yr: Handwave Any scenario in which megafaunal climate engineering is protecting permafrost at climatically significant scale requires a herbivore population increase of more than an order of magnitude over a region where the southern permafrost margin has already gone, which reduces the addressable area as the programme scales. Speculative The more plausible 50-year outcome is regional: grazing management written into Arctic land-use policy as one variable among several, with ground temperature monitored as an outcome alongside the species targets that Biodiversity Restoration tracks.
• 100 / 250+ yr: Handwave On this horizon the question is whether a maintained grazing regime is a stable state of the Arctic land surface, which is the palaeoecological disagreement in 3 projected forward and cannot be resolved by any experiment a present institution could run. Handwave If it is stable, the obligation created in 12 is permanent; if it is not, the intervention is a holding action whose benefit ends when management does.

• Depends on This brief waits on results that other briefs on this map are closer to than it is. It waits on Carbon Removal Verification for the question of how an avoided emission from a modelled counterfactual could ever be credited, because without that there is no revenue and the costing in 10 stands. It waits on Atmospheric Methane Removal for the methane accounting that decides the sign of a grazing programme. It waits on Arctic Engineering for the unattended winter instrumentation that the decisive experiment needs. It does not wait on De-Extinction Technologies at all, which is the most useful thing this brief can say about that relationship.
• Requires (not on this map) Five things sit outside what any brief here can deliver. The first is the curve relating snow depth to herbivore density over a landscape rather than a plot — the anchor authors’ own first open question, unanswered six years on, and the quantity that converts a climate objective into a stocking target. The second is a crediting methodology for an avoided permafrost emission, without which the programme has no revenue line and remains about nine-tenths grant-funded even at the demonstration scale. The third is a supply of large herbivores and the winter transport to place them, which is the constraint that keeps demonstration units at 1,000 animals. The fourth is a consent framework for asking herding peoples to change stocking rates for a global benefit, which exists in no form obtained here. The fifth is money that recurs for ten years rather than three, because the quantity being measured integrates over years and every existing dataset stops before it.
• Enables A validated landscape-scale snow-density control on ground temperature would be usable well beyond grazing: it would apply to any Arctic land-use decision that changes winter snow structure, including infrastructure siting, vegetation management and the shrub-encroachment question. It would also give Climate Engineering a rare worked example of a surface-energy intervention with a measurable local outcome and no stratospheric component.
• Adjacent Biodiversity Restoration shares the instruments and not the objective, and the divergence is sharpest over lichens. Synthetic Ecosystems shares the problem of specifying a community for a function. Geoengineering Governance and Inadvertent Geoengineering share the surface-albedo framing. Genetic Engineering is the adjacency that would matter if anyone applied it to extant Arctic grazers rather than to extinct ones.

Speculative “Bringing back mammoths will save the permafrost.” The mechanism does not need mammoths and the timeline does not permit them. Snow compaction is a function of mass and hoof pressure, and Yakutian horses, bison, musk ox and reindeer supply it today. Colossal’s own pages (vendor) describe editing 65 genes into an elephant cell line and using Asian or African elephant surrogates, and state no date for first calves and no release numbers. An animal produced one at a time through elephant surrogacy cannot reach the population scale in 5 this century, whatever the biology does.

Established “Eighty per cent of permafrost can be preserved.” That is a model output under an assumption nobody claims is achievable, and the paper says so. The figure comes from applying a grazed-snow parameterisation to every permafrost grid cell in a 1,037 Mha pan-Arctic domain for eighty years at a density of 114 animals per km², against a present Arctic average of 5. The paper’s own phrasing is “theoretically 80%” and it describes the Pleistocene Park density as a theoretical upper end. Quoting it without the density is the single most common error in this subject.

Frontier “Grazing cools Arctic ground.” It cools some Arctic ground and there is a well-designed study showing the opposite in peatlands. Across 17 Fennoscandian peatlands, high reindeer density was associated with a significantly lower proportion of hummocks retaining permafrost, hummocks of 2 m² against 65 m², deeper snow at peatland edges and more standing water, leading the authors to recommend reducing densities there. The same July campaign that supports the winter mechanism measured grazed topsoil significantly warmer in summer.

Frontier “Pleistocene Park proves the concept.” Pleistocene Park is one unreplicated site of 20 km², and its central soil-carbon result was not accepted at the first journal it was submitted to, on replication grounds. The preprint record states the revision was not accepted; the work appeared in 2022 in another journal, with one soil core per site across five sites and an explicit statement that previous site conditions cannot be distinguished from the herbivory impact. The park is a valuable demonstration and it is not a controlled experiment.

Established “Herbivores are net carbon sinks.” Enteric methane is a real subtraction with a published number. Restoring large herbivores to a 1500 AD baseline is estimated to add 11.7–13.4 MtCH₄ per year, or 292–335 MtCO₂e per year. The offsetting Arctic effect, a 44 per cent reduction in July soil methane at one site, is a peak-season plot measurement that the authors explicitly refuse to extrapolate to an annual budget.

Established “Rewilding delivers 6.4 gigatonnes of CO₂ a year.” That total is dominated by one marine term with an uncertainty almost as large as itself. The 6.44 GtCO₂ per year figure across ten animal species and groups includes marine fish at 5.50 ± 4.40 Gt per year. The Arctic terrestrial entries are muskox at 0.037 ± 0.015 and bison at 0.011 ± 0.002. Anyone citing the total in support of Arctic grazing is citing fish.

Frontier “This is ecological restoration, so the co-benefits are automatic.” The measured biodiversity effect of heavy reindeer grazing on lichens is consistently negative. The systematic review covering 41 independent studies found that result. A programme whose stocking rate is chosen for snow density will not deliver the community a conservation objective would choose, and the flagship site’s own species list includes cattle, sheep, camels and goats.

Speculative “It is cheap because animals are free.” The one published costing puts three 1,000-animal areas at US$114 million over ten years against carbon revenues of US$360,000 per area per year. That is US$3.8 million per area per year of cost against US$0.36 million of revenue, roughly ten to one, at the demonstration scale. Those figures come from a university press release rather than from a source this brief could open, which is itself a comment on how thin the economics of this subject are.

Frontier “The mammoth steppe was made by its animals, so animals can remake it.” The palaeo record is genuinely contested. Sedimentary ancient DNA from central Yukon has megafaunal signals highest from 30,000 to 15,000 years ago and woody shrubs rising after about 13,500 years ago, an ordering consistent with herbivore loss enabling shrub expansion. Monteath and colleagues argue for eastern Beringia that shrub expansion came first. The proposal’s central premise sits on the unresolved side of that question.

Handwave “It is a natural solution, so it is safe.” The intervention is a deliberate, maintained, continental-scale change to surface energy balance, which is the definition the governance literature uses for climate engineering. Being made of animals does not change the category; it changes the failure modes, from mechanical to biological, and it adds a welfare dimension that engineered interventions do not have.