Climate Overshoot and Lock-In

Overshoot is the plan of record. Almost every published pathway that ends the century at 1.5 °C passes above it first and comes back down, and coming back down requires removing carbon dioxide from the air at a scale that does not currently exist. Established The word “temporary” is doing the load-bearing work in that sentence, and it is a claim about five different systems that reverse at five different rates. Global mean temperature is the fastest and the only one that tracks emissions closely. Sea level is slower by orders of magnitude. Ice sheets exhibit hysteresis, meaning the path back is not the path out. Ecosystems reverse partially or not at all. Infrastructure and social arrangements have their own inertia, and nothing in a climate model describes it.

Frontier This brief owns reversibility and nothing else. The Institute’s existing coverage handles the interventions: Climate Engineering owns solar radiation modification and its physics, Geoengineering Governance owns the treaty record, Carbon Capture at Scale owns removal technology and its measured deployment, and Climate Migration Planning owns what happens to people who move. What none of them owns is the joint question those four presuppose: when the temperature comes back down, what comes back with it?

Frontier The answer this brief lands is that temperature reversal is the only reversal anyone has evidence for, and it is being used as a proxy for four others where the evidence runs the other way. A pathway that peaks at 1.7 °C and returns to 1.4 °C has not returned anything else to 1.4 °C conditions: sea level is still rising, a crossed ice-sheet threshold does not uncross, a reef that bleached out is a different reef, and a relocated community does not relocate back. Frontier The overshoot literature has begun to say this out loud, and the policy architecture has not caught up.

Established A note on sourcing. This brief was commissioned in September 2026 from the Institute’s research base. Reading-list entries without links are cited from the bibliographic record rather than re-fetched, and claims are dated no later than early 2026 unless carried by a linked source.

Established Peak warming is set by cumulative carbon dioxide emissions, and that relation is why overshoot is arithmetically possible at all. The transient climate response to cumulative emissions is close to linear across the policy-relevant range, so warming is approximately a running total rather than a rate. Established Its corollary is the one that makes overshoot a plan rather than an accident: if warming tracks a total, then subtracting from the total should subtract from the warming. Everything contested in this brief is about how far that corollary holds once the subtraction starts.

Established Temperature stops rising when emissions stop, and this is the strongest reversibility result in the field. The zero-emissions commitment intercomparison ran eighteen models through an abrupt cessation of emissions and found the temperature change over the following fifty years centred near zero, with individual models spanning roughly a third of a degree of cooling to a third of a degree of further warming. Established That result retired the older idea of a large warming already in the pipeline. It also does less than it is usually asked to do: it establishes that warming halts, not that it reverses, and the models disagree on sign.

Frontier The carbon cycle is asymmetric, so a tonne removed does not undo a tonne emitted. Modelling of positive and negative emission phases finds that land and ocean sinks, having absorbed carbon on the way up, give some of it back on the way down: removal from the atmosphere is partly offset by outgassing, so the airborne fraction of a removal is smaller than the airborne fraction of an emission. Frontier The magnitude is model-dependent and the sign is not in dispute. The operational consequence is that returning atmospheric concentration to a prior level requires removing more than was added, which means every removal budget quoted as a simple subtraction is an underestimate of unstated size.

Frontier The overshoot literature’s own recent verdict is that the field has been overconfident. A 2024 Nature analysis argued directly that returning below a temperature threshold after exceeding it is far less controllable than the scenario framing implies: peak warming carries deep uncertainty, the net-negative emissions needed to come back down run to hundreds of gigatonnes of carbon dioxide, and a portion of the impacts incurred at the peak does not reverse when the temperature does. Frontier That paper is the closest thing this field has to a self-correction, and its arithmetic is contested only at the margins.

Established Against that requirement, deployed removal is three orders of magnitude short. The independent assessment of the sector puts conventional removal — overwhelmingly forestry and land management — at about 2,200 MtCO₂ a year, and novel removal at approximately 2.0 MtCO₂ a year in 2025, with direct air capture at 0.0015. Established The same assessment records a historical delivery rate of roughly 20 per cent against projections and 8.4 MtCO₂ a year by 2030 from projects under construction. National pledges reach 2.5, 2.7 and 3.6 GtCO₂ a year in 2030, 2035 and 2050 against Paris-consistent medians of 2.9, 3.9 and 8.8, so the gap widens by a factor of seventeen across twenty years.

Established Sea level is the clearest case of a variable that does not follow temperature back down. Thermal expansion and ice loss respond to integrated heat over centuries to millennia, not to annual mean temperature, so sea level continues rising after temperature stabilises and continues rising after it falls. Established The multi-millennial commitment literature puts the eventual rise at metres per degree of peak warming, on timescales of thousands of years, and no proposed intervention reverses it on a human timescale. The exposure arithmetic compounds the point: revised elevation data tripled the estimated global population on land below projected coastal flood levels, which means the impact of a given rise is itself larger than the earlier consensus.

Frontier Ice sheets exhibit hysteresis, and hysteresis is the technical content of “lock-in”. Modelling of the Greenland ice sheet finds a threshold in the low single degrees above pre-industrial beyond which sustained loss continues, and finds that restoring the prior temperature does not restore the prior ice, because a lowered ice surface sits in warmer air. Frontier Antarctic modelling finds the same shape with larger numbers and a marine-ice-sheet instability mechanism in the West Antarctic that, once engaged, is not obviously arrestable by cooling. The thresholds are model results, not observations, and their central estimates have moved more than once; what is robust is the asymmetry, not the number.

Frontier The Atlantic overturning circulation is the most-cited tipping element and the most contested. The last major assessment judged a decline over this century likely and an abrupt collapse before 2100 not likely, at medium confidence. Frontier Since then one statistical analysis of sea-surface-temperature fingerprints put a collapse in the middle of this century with wide bounds and drew immediate methodological objections; a full-complexity model run found a tipping event under sustained freshwater forcing and offered a physics-based early-warning indicator based on freshwater transport at the southern boundary; and further work argued that collapse this century remains unlikely. The disagreement is genuine and it is about the observational record being too short to constrain a slow mode, not about the physics being unknown.

Established Ecosystem reversibility is the category where the assessment language is bluntest. Warm-water coral reefs are assessed to decline by 70–90 per cent at 1.5 °C and by more than 99 per cent at 2 °C, and reef recovery after severe bleaching takes a decade or more under conditions that overshoot removes. Extinction is definitionally irreversible. A pathway that spends thirty years above 1.7 °C and returns to 1.4 °C therefore returns to a biosphere that the 1.4 °C label does not describe, and no scenario database reports that distinction as an output.

Frontier The organising open question is whether peak warming can be controlled at all, as distinct from whether end-of-century warming can be reported. Peak warming depends on climate sensitivity, on the carbon-cycle response and on near-term emissions, all of which carry wide uncertainty; end-of-century warming in a scenario depends additionally on removal that has not been built. A pathway can be robust in its endpoint and almost unconstrained at its peak, and it is the peak that crosses thresholds.

Frontier The second question is how much of the impact integrates over the whole trajectory rather than sampling the endpoint. Ice loss, permafrost carbon, species loss and sea level are integrators: they respond to time spent above a threshold, not to where the trajectory finishes. If most consequential impacts are integrators, then overshoot depth and duration are the policy-relevant variables and end-of-century temperature is close to irrelevant as a target — which would invert the entire architecture of climate goal-setting.

Frontier The third is whether the moral-hazard objection to overshoot planning has any empirical support, and the available evidence says less than its proponents assume. Established The nearest measured analogue is solar geoengineering: randomised message experiments found that presenting balanced information about it does not reduce average support for emissions reductions, and a separate study found little effect on mitigation engagement. Those results test public attitudes, not the behaviour of finance ministries or oil companies, which is where the mechanism is alleged to operate; the objection may be right for reasons those experiments cannot reach.

Frontier The fifth is the one nobody has a method for: social and institutional reversibility. A relocated community, a defaulted sovereign, an abandoned settlement and a collapsed insurance market are all states that a falling thermometer does not undo. There is no accepted formalism for institutional hysteresis and no scenario framework that carries one, so the most plausible irreversibilities in the whole system are the ones with no representation in the models used to argue that overshoot is recoverable.

Established The binding constraint on any return path is removal capacity, and it is short by a factor of about a thousand. Novel removal at roughly 2 MtCO₂ a year against scenario requirements in the gigatonnes is not a scaling problem of the kind that solar photovoltaics solved; the accompanying assessment framing is explicit that closing the gap would require removal to scale faster than solar or electric vehicles did. Growth of 36 per cent a year, which the sector currently posts, is fast and still arrives late by decades.

Frontier The second constraint is verification, which decides whether a removal counts. Carbon Capture at Scale owns the measured record of underperformance at operating capture plants and the accounting disputes in the credit market. Independent reporting has found deployed direct air capture failing to cover its own emissions, and forest-credit evaluations have found claimed reductions far above delivered ones. An overshoot plan denominated in credits rather than in measured atmospheric removal is a plan whose central quantity is unaudited.

Frontier The third is observational: the systems whose reversibility matters most are the ones we monitor worst. Continuous overturning observations begin in 2004 at one latitude and 2014 in the subpolar gyre, which is a record too short to separate a trend from decadal variability in a circulation with centennial modes. Deep-ocean heat, Antarctic grounding-line retreat and permafrost carbon flux are similarly under-sampled relative to the timescales on which they lock in.

Frontier The fourth is that the rate of temperature decline is itself capped, and the cap is rarely stated. A return from a peak is limited by how fast net-negative emissions can be delivered, and even generous scenarios decline at a small fraction of a degree per decade. That means an overshoot of a few tenths of a degree implies decades spent above the threshold, so “temporary” in this literature routinely denotes a period longer than a typical infrastructure lifetime and longer than a coral recovery interval.

Frontier The result this brief most needs is a constrained estimate of the carbon-cycle asymmetry under realistic removal trajectories rather than idealised ramps. The asymmetry is established in principle from model intercomparison; its magnitude under a plausible century-scale path with a modest overshoot is not. Until it is, every stated removal requirement carries an unquantified multiplier, and the difference between a multiplier of 1.1 and 1.5 is hundreds of gigatonnes.

Frontier Second, ice-sheet thresholds constrained by observation rather than by model spread. Current threshold estimates come from model ensembles with limited palaeo constraint, and they have moved substantially between studies. Grounding-line observations, palaeo sea-level highstands and improved basal-melt physics are the three inputs that would narrow them, and all three are active fields with no scheduled convergence.

Frontier Third, an overturning early-warning indicator that has been validated against something other than the model that produced it. The freshwater-transport fingerprint is physically motivated and its observational estimate is sparse and sign-uncertain; the statistical fingerprint methods are sensitive to the choice of proxy and detrending. An indicator that disagrees with itself across methods is a research programme, not a warning system.

Frontier Fourth, impact functions that take a trajectory as input rather than a temperature. Most published damage and risk relations are functions of warming level, which by construction cannot distinguish a monotone approach to 1.5 °C from an overshoot that peaks at 1.9 °C and returns. Without path-dependent impact functions, the case for overshoot cannot be evaluated even in principle, because the quantity it trades away is invisible to the instrument.

Frontier The decisive observation is already running and does not need new hardware: sustained measurement of the Atlantic overturning circulation together with the South Atlantic freshwater-transport fingerprint is the one result that would most change this brief’s assessment. Frontier If the overturning arrays show a decline that is coherent across latitudes and the fingerprint moves toward the model-identified tipping regime, the reversibility assumption underneath overshoot planning fails for the largest single element in the system, and it fails while the temperature is still rising. The arrays exist, the record is roughly two decades long, and the limiting factor is continuity of funding rather than instrumentation.

Frontier The second is the one that would settle the carbon-cycle question and nobody has scheduled it: a coordinated model intercomparison of realistic overshoot-and-return trajectories, reporting the removal multiplier explicitly. Speculative Existing intercomparisons use idealised ramps that no policy will follow; the number policymakers need is how many tonnes must be removed to lower concentration by one part per million, along a path with a peak in the 2040s. This is a compute-and-coordination problem, not a science problem, and the reason it has not been run is that no body owns the question.

Frontier The fourth is an ecological reversibility trial that is being run without consent: the sequence of mass coral bleaching events now occurring at intervals shorter than reef recovery time. Recovery intervals are measured and bleaching return intervals are measured, and the comparison is the cleanest natural test of whether a thermal excursion is recoverable. If reef assemblages shift composition permanently under repeated excursions that individually end, overshoot reversibility fails at the level of an entire biome while global temperature does exactly what the scenario said.

Frontier The fifth is institutional and cheap: pre-register what would count as failure. No overshoot pathway published to date states the observation that would retire it — a removal deployment rate below some threshold by some year, a measured ice-sheet retreat rate, an overturning index value. A plan with no stated falsifier is not a plan being tested, and the absence of falsifiers across an entire scenario literature is the strongest evidence that overshoot is functioning as an accounting device.

Established The engineering requirement for a return path is a removal industry the size of the present oil industry, built in half the time. Gigatonne-scale removal means gigatonne-scale carbon dioxide handling: capture, compression, pipelines, injection wells, and monitoring for permanence measured in centuries. The capture step attracts the attention and the storage step carries the liability, and the assessed energy requirement of 5–6 per cent of global energy demand at gigatonne scale is a demand on a grid that is simultaneously being asked to decarbonise everything else.

Frontier The third is observational infrastructure for the lock-in variables, which is the cheapest item on this list and the least funded. Sustained overturning arrays, grounding-line radar, deep Argo and permafrost borehole networks are all in the tens of millions of dollars a year against removal programmes in the billions. Earth-System Digital Twins owns the modelling-and-decision chain that would consume those observations, and its central finding — that the chain from observation to decision is where these systems break — applies directly here.

Speculative The fourth is the intervention nobody wants to design and several people are already designing: cooling to hold a peak down while removal catches up. Climate Engineering owns the physics and the cost, and the relevant point for this brief is structural rather than technical — a cooling intervention deployed to shave a peak must be sustained until the removal it was buying time for arrives, or the avoided warming returns quickly. Frontier The termination-risk literature is the analysis of exactly that failure, and it is the reason peak-shaving is a commitment rather than a stopgap.

Established The nearest neighbour is the intervention literature, and this brief is its missing denominator. Climate Engineering establishes that the physics of cooling is the least contested part, the cost is low enough to be destabilising, and regional consequences are the live dispute; Geoengineering Governance establishes that two multilateral attempts failed and the only binding prohibitions in force are two US state statutes. Both presuppose an answer to the question this brief owns: whether the thing being bought time for is recoverable.

Frontier On the removal side the dependency is direct. Carbon Capture at Scale owns deployment, cost and the verification record, and supplies the numbers that make the return path either a schedule or a wish. Nothing in this brief argues against removal; the argument is that removal restores a temperature and is being credited with restoring a world.

Frontier On the impact side, the irreversible categories already have owners. Climate Migration Planning owns planned relocation and records that it is a real practice with perhaps a few dozen documented cases, no legal status and no funding line — which is what social irreversibility looks like operationally. Compound Climate Hazards owns the interaction of simultaneous extremes, Infrastructure Resilience owns the built-asset response, and Climate Health Adaptation owns exposure to heat and disease across a trajectory rather than at a level.

Established The Paris Agreement is written in endpoint language, and endpoint language cannot express an overshoot constraint. Holding warming “well below” a level and pursuing efforts to limit it to another is a statement about a state, not about a path, and nothing in the stocktake architecture distinguishes a pathway that peaks low from one that peaks high and returns. A regime that cannot express peak warming as an obligation cannot allocate responsibility for crossing a threshold, which is precisely the harm that turns out to be irreversible.

Frontier The second institutional gap is that removal obligations are not owned by anyone in particular. Scenario pathways assign net-negative emissions to the second half of the century, which is to say to governments that do not yet exist, on the basis of commitments that are not binding and against a delivery record of roughly 20 per cent against projections. An intertemporal transfer with no enforcement mechanism and no counterparty is the structure of a default, and the default cannot be observed until the date arrives.

Frontier Third, the one body convened specifically on this question produced recommendations and no authority. An independent overshoot commission reported on the combination of mitigation, adaptation, removal and the question of whether to research solar intervention, and its output is a report rather than an instrument. That is the same institutional shape found across the geoengineering record: assessment without enforcement.

Frontier Fourth, the legal position is moving faster than the political one. The International Court of Justice took the question of states’ climate obligations and delivered an advisory opinion in 2025, and advisory opinions change the litigation landscape even where they bind nobody directly. If a duty of prevention attaches to peak warming rather than to end-of-century warming, overshoot planning acquires a legal exposure that no current pathway has priced.

Frontier Overshoot is an intergenerational and interregional transfer, and it is the kind that cannot be repaid. The benefit — a slower, cheaper transition — accrues to the present; the cost falls on those exposed to the peak, and the compensation offered is a temperature reduction that arrives after their exposure. For low-lying states the asymmetry is total, because sea level does not come back down within any horizon in which their territory exists.

Frontier The moral-hazard objection deserves to be stated at full strength and then measured rather than assumed. The strong form is that a credible return path licenses delay, and delay raises the peak, which is the only variable that produces irreversible harm. The nearest experimental evidence — on balanced messaging about solar geoengineering — found no reduction in average support for emissions reductions. That evidence bears on publics, not on the actors who set capital allocation, and this brief treats the objection as unresolved rather than refuted.

Speculative There is a mirror-image hazard that gets less attention: refusing to plan for overshoot when overshoot is the most likely outcome. If peak warming above 1.5 °C is close to certain on current trajectories, a discourse that treats overshoot as unthinkable prevents preparation for the conditions that will actually occur. The honest position is that both hazards are real and only one of them is currently being argued about.

Frontier Removal at scenario scale has distributional content of its own. Land-based removal competes with food production and with the territory of people who did not consent to the trade, and the constrained potential estimates for bioenergy with capture are constrained precisely by that competition. A return path built on land acquisition in low-income countries reproduces the extraction pattern that produced the problem, and no existing pathway reports land tenure as an output variable.

Frontier The civilizational question is whether humanity will have acquired, for the first time, a deliberate and reversible hand on a planetary variable — and the answer this brief reaches is that reversibility is the part that fails. Cooling can be applied and removed. Carbon can be added and, expensively, subtracted. Ice sheets, sea level, species and settlements do not have that property, so the control problem is not a thermostat but a ratchet with a slow return spring.

Speculative The most consequential long-run outcome is a commitment that outlives the institutions that made it. A peak in the 2040s commits sea-level rise over centuries and ice loss over millennia, which means decisions taken by present governments determine coastlines for societies with no continuity of law, language or record with ours. The palaeo record supplies the scale: the eventual equilibrium response to a few degrees is metres, not centimetres.

Speculative The optimistic reading is real and should be stated. The zero-emissions commitment result means that stopping works: the temperature does not keep climbing on its own, which is a genuinely better world than the one the pipeline framing described. If removal reaches gigatonne scale and the carbon-cycle multiplier is near the low end, a shallow overshoot of a few tenths of a degree over a few decades is a survivable trajectory whose worst permanent legacy is coastal.

These horizons track what becomes knowable about reversibility, not what becomes achievable.

• 10 yr: Frontier Peak-warming uncertainty narrows as the record lengthens and the overturning arrays reach a three-decade record long enough to separate trend from decadal variability. Novel removal is in the tens of megatonnes a year at best against gigatonne requirements, and the first credible national net-negative claim is tested against atmospheric inversion. Expect path-dependent impact functions to appear in the literature and not yet in the scenario databases.
• 25 yr: Speculative Peak warming is either passed or imminent, and the question becomes empirical rather than projected: whether removal is delivering at the rate the pathways assumed, and whether any ice-sheet threshold has demonstrably been crossed. This is the decade in which the overshoot plan either becomes a schedule or is retired.
• 50 yr: Speculative If the return path worked, global temperature is falling at a small fraction of a degree per decade while sea level continues to rise, which is the clearest possible demonstration of the asymmetry this brief is about. If it did not, the world is holding at a peak with a removal industry an order of magnitude short.
• 100 / 250+ yr: Handwave Claims about restoring pre-industrial conditions, regrowing lost ice or recovering lost species on these horizons are coherent only as statements about equilibrium physics; they assume institutional continuity of removal and monitoring across a period longer than any state has maintained a continuous technical programme.

• Depends on the deployment and verification record in Carbon Capture at Scale, which supplies the removal rates any return path must use; the physics and cost findings in Climate Engineering, which determine whether a peak can be shaved at all; and the treaty record in Geoengineering Governance, which shows what institutional capacity actually exists to authorise anything planetary.
• Requires (not on this map) a constrained estimate of how much extra must be removed to undo a tonne emitted along a realistic path, without which every removal budget is an unquantified underestimate; ice-sheet thresholds constrained by observation rather than by model spread, without which lock-in cannot be dated; an overturning early-warning indicator validated outside the model that produced it, without which the largest tipping element has no monitoring; a legal obligation expressed as peak warming rather than as an endpoint, without which nobody is responsible for crossing a threshold; verified removal capacity in gigatonnes per year, which is three orders of magnitude above the present novel-removal rate; and an identified counterparty for the net-negative emissions that pathways assign to the second half of the century.
• Enables peak-constrained rather than endpoint-constrained climate targets, path-dependent damage functions that can price an overshoot, liability rules attaching to threshold crossing, adaptation planning calibrated to a trajectory, and an honest comparison between shaving a peak and living through it.
• Adjacent to Climate Migration Planning on social irreversibility, Compound Climate Hazards on simultaneous extremes at the peak, Infrastructure Resilience on built-asset commitment, Climate Health Adaptation on exposure across a path, and Earth-System Digital Twins on the observation-to-decision chain these results have to travel.

Handwave “Overshoot is temporary.” The word describes the temperature and nothing else. Sea level rises for centuries after the peak, a crossed ice-sheet threshold does not uncross when the air cools, extinction is permanent, and relocation is not undone by a falling index. The honest formulation is that overshoot is a temporary excursion in one variable and a permanent change in several others, and the scenario databases report only the first.

Frontier “There is warming already in the pipeline, so it does not matter when we stop.” This was the older view and the zero-emissions commitment intercomparison largely retired it: across eighteen models the fifty-year temperature change after cessation centres near zero. The correction cuts both ways — stopping works, which strengthens the case for stopping sooner and weakens the case for treating the peak as already determined.

Frontier “Removing a tonne undoes emitting a tonne.” The carbon cycle is asymmetric: sinks that absorbed carbon during the emission phase release some of it during the removal phase, so restoring a concentration requires removing more than was added. The multiplier is model-dependent and unconstrained for realistic paths, which makes it the most consequential missing number in overshoot accounting.

Frontier “The overturning circulation is collapsing / is definitely fine.” Both claims are made with more confidence than the evidence supports. The assessed position is a likely decline this century and an abrupt collapse before 2100 judged not likely at medium confidence; subsequent fingerprint analyses, a full-complexity tipping demonstration and rebuttals arguing against near-term collapse all postdate it. The limiting factor is a two-decade instrumental record against a centennial mode.

Speculative “Planning for overshoot causes overshoot.” The mechanism is plausible and the direct evidence is thin. Randomised messaging experiments on the closest analogue found balanced information about solar geoengineering did not reduce average support for emissions cuts. Those experiments measure publics, not capital allocation, so the objection survives as unresolved rather than refuted — and the symmetrical error, refusing to prepare for the most likely outcome, receives almost no attention.

Handwave “Solar radiation modification can hold the peak down until removal catches up.” Physically it can reduce temperature; the difficulty is that it must then be sustained until the removal arrives, and stopping early returns the avoided warming rapidly. The termination-risk literature is the study of that failure mode and the biodiversity consequences of abrupt cessation are its sharpest result. Peak-shaving is therefore a multi-decade commitment made by whoever starts it, which is a governance claim rather than a technical one.