How long does a carbon dioxide emission go on warming the planet?

Why the question matters

If the warming from an emission decayed away, emissions would be a flow problem — reduce the rate and the temperature would settle at a lower level. Because the warming persists instead, emissions are a stock problem, and any continuing positive emission rate means continuing warming rather than a stable warmer climate.

The distinction decides what a mitigation target should even be. Stabilising the concentration of greenhouse gases and stabilising the temperature are not the same objective, and the first does not deliver the second.

It also sets the terms for what is reversible. Warming that persists for centuries after the emission is, for practical purposes, a permanent commitment rather than a consequence that later generations can undo by emitting less.

What our research finds

• Aviation's contrail cirrus forcing is confined to a narrow band of northern mid-latitudes, but the temperature response it produces is polar-amplified rather than local (Thomas et al., 2025).
• Under emission projections for policies already in place the rate of global warming declines from about 0.21 °C per decade around 2025 to about 0.15 °C per decade around 2050 — slower warming, but warming that does not stop (Duan and Caldeira, 2024).
• Emitting carbon and then removing exactly as much leaves surface temperature 0 to 1 °C above preindustrial and atmospheric CO2 12 to 29 ppm below it at the moment net emissions reach zero, with both offsets growing as the pulses get larger and longer (Jayakrishnan et al., 2024).
• Hydrogen leakage warms less than the same quantity of methane, and because its forcing tracks the emission rate rather than cumulative emissions the Earth cools rapidly once it stops — unlike after CO2 (Duan and Caldeira, 2023).
• A thousand years after releasing 600 PgC, fossil-fuel emissions leave 20 per cent of the atmospheric perturbation and about 1 °C of warming, while the same quantity from deforestation with regrowth returns CO2 close to preindustrial (Jayakrishnan et al., 2022).
• LongRunMIP assembles 50 millennial-length simulations from 15 complex climate models; under an abrupt quadrupling of CO2 in the model CESM104, the 150 years that standard intercomparison protocols simulate capture only 75% of the eventual surface warming, with 88% reached after 1,000 years (Rugenstein et al., 2019).
• The global climate feedback parameter weakens as the pattern of ocean heat uptake becomes high-latitude-enhanced, so sensitivity inferred from short records understates the long-run response (Rugenstein et al., 2016).
• Burning the attainable fossil-fuel resource would leave Antarctica almost ice-free, and West Antarctica destabilises after only 600 to 800 GtC of further emissions (Winkelmann et al., 2015).
• A linear regression model combining each variable's sensitivity to CO2 forcing, to solar forcing, and to global mean surface temperature change reproduces HadCM3L-simulated transient changes in surface temperature, precipitation, runoff, soil moisture, cloudiness, and radiative fluxes (Cao et al., 2015).
• Sea ice response accounts for about 20% of climate sensitivity in CESM 1.0.4, and its effect on the longwave climate feedback parameter is nearly half as large as its effect on the shortwave one (Caldeira and Cvijanovic, 2014).
• Fits to the CMIP5 models deliver 38-61% of equilibrium warming within a decade of a CO2 increase but leave 14-40% of it to arrive more than a century later, so models disagree about the timing of warming as much as about its size (Caldeira and Myhrvold, 2013).
• Shifting one watt per square metre of surface heat flux from sensible to latent form, with no change in the energy content of the climate system, cooled global mean surface air temperature by 0.54 K — evaporative cooling is a global effect, not only a local one, because the added evaporation raises low cloud cover and reflects more sunlight (Ban-Weiss et al., 2011).
• Instantaneously removing a quadrupled-CO2 forcing in the HadCM3L coupled model raised global mean precipitation by 5.1% within one year, whereas removing an equal-magnitude solar forcing made precipitation fall immediately — the transient increase reflects the fast tropospheric adjustment to CO2 rather than reduced radiative forcing in general (Cao et al., 2011).
• Doubling CO2 warms land 2.86 K radiatively and a further 0.42 K purely by closing plant stomata, and that physiological effect raises global runoff more than the radiative effect does — 8.4 per cent against 5.2 per cent (Cao et al., 2010).
• The apparent difference in hydrological sensitivity between solar and carbon dioxide forcing originates entirely in the fast response; once fast responses are excluded, hydrological sensitivity is the same for both forcings (Bala et al., 2010).
• Land plants stop amplifying silicate-rock weathering once atmospheric CO2 falls near 150 to 250 ppm, and that failure of the biological sink explains why CO2 has not dropped below about 200 to 250 ppm in 24 million years (Pagani et al., 2009).
• A single pulse of carbon raises globally averaged surface temperature by an amount that remains approximately constant for several centuries, so holding temperature constant requires near-zero future emissions (Matthews and Caldeira, 2008).
• Global-scale deforestation cools the modelled climate, because albedo and evapotranspiration changes outweigh the warming from released CO2 — so counting only carbon gets the sign wrong outside the tropics (Bala et al., 2007).
• Explaining the Paleocene-Eocene warming requires either a carbon release of 5,400 to 112,000 PgC with no known source, or a climate sensitivity far above the conventional 1.5 to 4.5 °C per doubling (Pagani et al., 2006).
• Forests absorb more solar radiation than the ecosystems they replace and so warm the climate, especially outside the tropics, and that warming should itself raise silicate rock weathering rates — a climate-mediated pathway distinct from the soil-chemistry effects usually considered (Caldeira, 2006).
• Burning all currently estimated fossil fuel resources warms the world about 8 K by 2300 and takes atmospheric CO2 to 1,423 ppmv, with nearly half of cumulative emissions still airborne at that date (Bala et al., 2005).
• Ocean general circulation model simulations give surface ocean pH reductions of 0.3 to 0.5 units by 2100 under the IPCC SRES A1, A2, B1, and B2 emission pathways, while a cumulative emission of 5,000 petagrams of carbon produces a surface pH reduction of 0.8 units by 2300 (Caldeira and Wickett, 2005).
• A severe reduction of marine primary productivity can explain the abrupt end-Permian carbon- isotope excursion, and would have driven atmospheric CO2 from a Late Permian baseline of 850 ppm to about 2,500 ppm (Rampino and Caldeira, 2005).
• Whether leaky carbon storage is nearly as good as permanent storage or worth nothing depends entirely on the carbon price path — 97 per cent effective at 1,500 m ocean depth under constant prices, 0 per cent if prices rise at the discount rate forever (Herzog et al., 2003).
• In a three-dimensional ocean general circulation model, injection depth predicts retention of directly injected CO2 well and local radiocarbon age does not — the expected time for a water parcel to return to the surface is closely related to its depth, not in general to the time since it was last there (Caldeira et al., 2002).
• Reacting CO2-rich power-plant gas with seawater and then with limestone on site, releasing calcium and bicarbonate ions to the ocean, would greatly expand the ocean's capacity to store anthropogenic carbon while minimizing the impact of that carbon on ocean biota (Caldeira and Rau, 2000).
• Spreading the salt rejected during sea-ice formation through the upper 160 metres rather than a 25-metre surface layer lowers modelled global ocean CFC-11 uptake by about 30%, and Southern Ocean column inventories by up to 90% (Caldeira and Duffy, 1998).
• Atmospheric radiocarbon content is predicted to reach a minimum and then rise under business-as-usual emissions, even though fossil fuels contain no radiocarbon, because rising atmospheric CO2 drives a net flux of carbon-14 out of the ocean and land biosphere (Caldeira et al., 1998).
• Silicate-mineral dissolution rates are nearly independent of pH between 5 and 8 and deep- ocean pH is above 7.5, so low-temperature seafloor basalt alteration is unlikely to be an effective feedback on atmospheric CO2; terrestrial silicate-rock weathering is the primary long-term control (Caldeira, 1995).
• The warming effect of a CO2 emission is largely insensitive to the background scenario, because the falling radiative effect of each added molecule is offset by the rising atmospheric lifetime of CO2 as ocean uptake saturates (Caldeira and Kasting, 1993).
• Allowing for C4 photosynthesis, which persists below 10 ppm CO2, extends the life span of the biosphere from roughly 100 million years to at least 0.9 to 1.5 billion (Caldeira and Kasting, 1992).
• Sustained enhancement of Cenozoic chemical weathering requires a matching increase in CO2 supply, which the Jurassic shift of carbonate deposition to the deep sea floor — recycled through subduction zones — could have provided (Caldeira, 1992).
• Atmospheric CO2 responds over more than 100,000 years while sea ice and snow cover respond in under a year, so the silicate-weathering feedback could not buffer the early Earth against a rapid excursion in the ice line, and with highly reflective carbon dioxide clouds a transient global glaciation could have been irreversible (Caldeira and Kasting, 1992).
• A carbonate-silicate cycle model gives mid-Cretaceous atmospheric CO2 of 3.7 to 14.7 times the pre-industrial value of 285 ppm from super-plume tectonics, implying 2.8 to 7.7 degrees Celsius of warming, though CO2 released from oceanic plateaus alone is unlikely to account for more than 20% of that increase (Caldeira and Rampino, 1991).
• Partitioning carbonate burial between shallow-water and deep-water settings gives two stable steady states — a continental mode with low metamorphic CO2 flux and a pelagic mode with high flux — and Cenozoic burial patterns suggest the Earth is moving from the continental toward the pelagic mode over roughly 100 million years (Caldeira, 1991).
• A carbon release spread across several hundred thousand years is largely absorbed as it accumulates, which is why Deccan Traps volcanism produced under 1 °C of warming despite an enormous total release — rate matters, not only total (Caldeira and Rampino, 1990).
• Neither climate modulation nor altruism can have been the primary evolutionary driver of planktonic dimethylsulphide production, since altruistic production is not an evolutionarily stable strategy; local benefits such as osmoregulation can (Caldeira, 1989).