Irreversible climate changes driven by degree-years of temperature overshoot Authors: Mitchell Dickau, Kirsten Zickfeld, H. Damon Matthews Published: 2026 ABSTRACT It is becoming increasingly likely that meeting the long-term temperature goal of the Paris Agreement will involve a period of temperature overshoot. Understanding the effect of overshoot on climate outcomes requires an assessment of the reversibility of climate changes caused by overshooting global temperature levels. Here, using an intermediate-complexity Earth system model, we quantify the short-term reversibility of a suite of climate variables across an ensemble of 42 pairs of overshoot and baseline scenarios. For highly irreversible climate variables like permafrost and ocean changes, we show that the outcome of overshoot is linearly related to the time-integrated overshoot magnitude, which we define here as the degree-years of temperature overshoot. Our results show that degree-years of overshoot can be used to predict the changes of a range of irreversible ocean and permafrost variables in overshoot scenarios relative to a baseline scenario, therefore offering important insights into the risks of overshooting climate targets. Introduction Current climate policies have global emissions on track to push global warming well beyond 2.0 °C1, and as a result, it is becoming increasingly likely that if we are to meet the 1.5 °C to well-below 2.0 °C temperature goal of the Paris agreement, there will first be a period of prolonged temperature overshoot (OS). The increasing likelihood of OS highlights the importance of understanding the climate outcomes of overshooting global temperature targets. A key question is how the severity of climate changes will differ as a function of the duration and magnitude of OS across a range of policy-relevant 1.5 to 2.0 °C scenarios. Previous work has shown that OS leads to additional climate changes and associated impacts due to hysteresis (or irreversibility) in the climate system2,3. These changes can be broadly categorized into two groups: those affecting slow-responding variables, which exhibit hysteresis or are sensitive to tipping points, and those affecting fast-responding variables, which tend to exhibit less or no hysteresis. Slow-responding climate variables that show pronounced hysteresis following OS include permafrost carbon loss, ocean temperature, ocean oxygen content, ocean pH, thermosteric sea level rise (SLR), and ice sheet loss, along with the associated SLR3–13. These variables exhibit hysteresis because they involve large thermal or physical inertia. For instance, once permafrost thaws during OS, the carbon loss is irreversible from a policy perspective since carbon buildup in soils and vegetation and subsequent refreezing would take millennia12. Ocean temperature and thermosteric SLR are governed by the slow penetration of heat into the deep ocean, which takes centuries to equilibrate even if surface temperatures are reduced9,14,15. Similarly, changes in ocean oxygen content and pH that occur during the period of OS can become effectively locked in when surface waters are downwelled, transporting these altered conditions into the deep ocean10,11. Once subducted, this water is isolated from the atmosphere for centuries, preventing rapid re-equilibration and delaying recovery of the deep ocean for centuries after surface temperatures have stabilized. Ice sheet loss is marked by threshold behavior and self-reinforcing feedbacks, such as the ice-albedo effect, which make ice sheet recovery impossible on timescales of centuries after substantial loss has occurred12,16. Fast-responding climate variables that exhibit little to no hysteresis following OS include surface air temperature, precipitation, Arctic sea ice, and sea surface pH3,8,13,17. Surface air temperature is closely tied to radiative forcing—as CO2 concentrations decline, temperatures begin to decrease rapidly due to the relatively low thermal inertia of the atmosphere and upper ocean2,18. Precipitation also adjusts relatively quickly as surface temperatures change, with reversal occurring slightly after changes in temperature13,19. However, for both temperature and precipitation, even though global mean values show high reversibility, there is evidence that the regional response following OS can exhibit more path dependence19–22. Arctic sea ice, while sensitive to warming, can regrow rapidly during cooler periods because it is driven by short-term surface energy fluxes13. Sea surface pH, which reflects the balance of CO2 between the surface ocean and atmosphere, also responds relatively quickly as well—when atmospheric CO2 levels decline, the surface ocean can begin to outgas CO2 and partially recover its pH due to rapid gas exchange and mixing in the upper ocean layers3,13. These variables tend to lack the deep memory, physical inertia, or strong reinforcing feedback that characterize slower components of the climate system, allowing them to more closely track changes in forcing and exhibit minimal hysteresis. Finally, some climate variables do not fit neatly into the categories of fast- or slow-responding. One such example is the Atlantic meridional overturning circulation (AMOC), which is sensitive to the rate of temperature change during OS, and can substantially influence both global and regional climate patterns, particularly in the North Atlantic and surrounding areas23. In some models, the AMOC recovers within two to three centuries after the peak of OS, suggesting it exhibits less hysteresis than slower-responding variables like permafrost carbon release or deep ocean heat content3. At the same time, this recovery time is slower and more path-dependent than that of fast-responding variables, such as surface temperature or precipitation. As such, the AMOC occupies an intermediate position—demonstrating a moderate level of hysteresis that depends on both the magnitude and duration of OS, as well as the rate of temperature change23,24. Existing modelling studies on the climate impacts of OS have used: (1) idealized CO2 only scenarios that are concentration driven or driven by emission pulses3,8,13,17,18,20,25–27, (2) a limited number of idealized multi-gas or SSP-based OS scenarios10,21,28–31, or (3) an ensemble of realistic scenarios that are simulated in a climate emulator rather than a more complex and spatially explicit Earth system model (ESM)32. There has yet to be a study that investigates climate outcomes of OS in relation to the magnitude and duration of OS across an ensemble of policy-relevant (i.e., 1.5° to 2.0 °C) multi-gas scenarios in a spatially explicit ESM. In this study, we investigate the climate outcomes of OS occurring this century in the University of Victoria intermediate complexity Earth System Climate Model (UVic-ESCM). We simulate 42 pairs of scenarios, each comprising one baseline reference scenario and one OS scenario with peak temperatures ranging from ~1.5 to ~2.0 °C (Fig. 1A). To explore how the duration and magnitude of OS affect the climate outcomes, we calculate the degree-years of OS, which we define here as the time integrated difference in global surface air temperature (GSAT) between an OS scenario and its baseline pair. We then quantify the short-term reversibility of a suite of different climate variables and show that the degree-years of OS are a robust predictor of the changes that occur across a range of highly irreversible climate variables.Fig. 1Global surface air temperature (GSAT) changes, overshoot magnitude, and degree-years of OS across the ensemble.Paired OS and baseline scenarios used in this study, which achieve a long-term temperature target of less than 2 °C with peak temperatures in the OS scenarios ranging from 1.5 to 2.0 °C. A Shows average global surface air temperature (GSAT) anomaly in OS (red lines) and their paired baseline (blue lines) scenarios with respect to 1850–1900. B Shows the magnitude of OS relative to the paired baseline scenarios. C Shows the degree-years of OS (i.e., the time-integrated difference in temperature change between an OS and baseline scenario pair). Global surface air temperature (GSAT) changes, overshoot magnitude, and degree-years of OS across the ensemble. Paired OS and baseline scenarios used in this study, which achieve a long-term temperature target of less than 2 °C with peak temperatures in the OS scenarios ranging from 1.5 to 2.0 °C. A Shows average global surface air temperature (GSAT) anomaly in OS (red lines) and their paired baseline (blue lines) scenarios with respect to 1850–1900. B Shows the magnitude of OS relative to the paired baseline scenarios. C Shows the degree-years of OS (i.e., the time-integrated difference in temperature change between an OS and baseline scenario pair). Results The climate outcomes of OS are often defined in terms of reversibility/irreversibility3,11,13,21. For a given climate variable, if there is a difference between the OS and baseline scenario values during the period of OS, and if there is no difference between the scenarios after the period of OS, then this climate variable would be considered fully reversible, and OS would be considered to have no impact at the timescale in question. Conversely, if there is still a difference in a given variable following OS, then the variable would be considered to have some degree of irreversibility. Here, we quantify the short-term reversibility of a suite of climate variables relative to that of global mean temperature in pairs of OS and baseline scenarios. These pairs of scenarios were developed using integrated assessment models for the ENGAGE project33. In each pair, one scenario (the OS scenario) was constrained by an end-of-century cumulative CO2 emissions budget, while the other scenario (the baseline scenario) was constrained by the cumulative CO2 emissions budget throughout the century. This framework allows for a temporary exceedance of the cumulative emissions budget followed by subsequent net-negative CO2 emissions to return to a given budget in the OS scenario, whereas the paired baseline scenario never exceeds the cumulative emissions budget. Because variations in non-CO2 assumptions in integrated assessment models can prevent temperature convergence within pairs of scenarios, we applied a strict filter to include only the 42 pairs where the global mean temperature in the OS scenario returns to the baseline level before 2100. Using only scenario pairs with temperature convergence occurring before 2100, we quantified reversibility for each climate variable at the return year, which is defined as the year that the global mean temperature in the OS scenario returns to the same temperature as in the paired baseline scenario. While this definition of reversibility captures the immediate impacts of OS at the point of temperature convergence—the return year—it is important to note that it does not reflect the long-term equilibration of climate variables. In the scenarios used here, the return years fall between 2086 and 2100. Despite each OS and baseline scenario pair having equal cumulative CO2 emissions by the end of the century, return years do not necessarily align with the year when cumulative CO2 emissions become equal because non-CO2 forcing in all OS scenarios falls below levels found in the baseline scenarios after 2090 (Supplementary Fig. 1). To calculate short-term reversibility, we first calculated the maximum difference (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{diff}}^{\max }$$\end{document}diffmax) for each variable between the OS and baseline value during the period of OS. Second, we calculated the difference between the OS and baseline values of a variable at the return year (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{diff}}^{{return\; year}}$$\end{document}diffreturnyear). Using the \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{diff}}^{\max }$$\end{document}diffmax and \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${{diff}}^{{return\; year}}$$\end{document}diffreturnyear we define the reversibility of a given climate variable at the return year as:1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${reversibility}=\frac{{{diff}}^{\max }-{{diff}}^{{return}\,{year}}}{{{diff}}^{\max }}$$\end{document}reversibility=diffmax−diffreturnyeardiffmax We calculated the reversibility of 30 globally averaged climate variables simulated by the UVic-ESCM, in addition to the spatial pattern of surface air temperature. Given our representation of reversibility (illustrated in Fig. 2A), global mean temperature itself is by definition 100% reversible. In addition to the magnitude of temperature OS, we further calculated the time-integral of the global mean temperature OS, which we define here as the degree-years of OS (Fig. 2B).Fig. 2Illustrative representations of reversibility and degree-years of OS. A Illustrates how reversibility is defined for climate variables. B Illustrates degree-years of OS, which is the time-integrated difference in global surface air temperature (GSAT) between an OS and baseline pair of scenarios. The return year is defined as the point at which GSAT in the OS scenario returns to the same level as the baseline scenario. Note that in many simulations, the OS scenario may cool below the baseline level after the return year due to differences in non-CO2 forcing pathways. Illustrative representations of reversibility and degree-years of OS. A Illustrates how reversibility is defined for climate variables. B Illustrates degree-years of OS, which is the time-integrated difference in global surface air temperature (GSAT) between an OS and baseline pair of scenarios. The return year is defined as the point at which GSAT in the OS scenario returns to the same level as the baseline scenario. Note that in many simulations, the OS scenario may cool below the baseline level after the return year due to differences in non-CO2 forcing pathways. Reversibility of climate outcomes Across the 30 globally averaged climate variables simulated by the UVic ESCM, the median reversibility across scenarios ranged from 0% (fully irreversible) to 120% (more reversible than global average temperature) (Fig. 3). For a discussion on the uncertainty in the reversibility of these variables across our ensemble, see Supplementary Note 1. The variables demonstrating the highest degree of irreversibility (i.e., reversibility ≤25%) in our ensemble are those characterized by substantial physical or thermal inertia, specifically ocean-related and permafrost carbon variables. For the permafrost carbon pool, total permafrost region soil carbon, and maximum AMOC strength (boxplots 1–3, Fig. 3), median reversibility at the return year is near 0%. This indicates that the impacts of the OS remain fully present at the return year, even when GSAT has returned to baseline levels. For thermosteric SLR, ocean temperature, and ocean oxygen (boxplots 4–6, Fig. 3), reversibility lies between 20 and 25%, confirming that most of the impact from the OS persists immediately following the return to baseline temperatures.Fig. 3Reversibility of variables in the UVic-ESCM.Median reversibility across all pairs of scenarios for each variable is represented by red lines. Each box represents the 1st and 3rd quartiles of reversibility for a given variable. The whiskers represent the 5th and 95th percentiles of reversibility. Reversibility of variables in the UVic-ESCM. Median reversibility across all pairs of scenarios for each variable is represented by red lines. Each box represents the 1st and 3rd quartiles of reversibility for a given variable. The whiskers represent the 5th and 95th percentiles of reversibility. In contrast to the slow-responding ocean and permafrost variables, several sea surface and atmospheric variables exhibit median reversibility exceeding 100%. These include the atmospheric concentration of CO2, surface ocean carbon cycle variables that follow atmospheric CO2 levels closely, and global average ocean alkalinity (boxplots 27–30 Fig. 3). For atmospheric CO2, sea surface pH and sea surface dissolved inorganic carbon, greater than 100% median reversibility reflects the understanding that, due to carbon cycle inertia, the ocean continues to take up CO2 following peak OS as temperatures decline, resulting in lower atmospheric CO2 at the return year relative to non-OS scenarios18. As a result, temperature reversal lags changes to atmospheric CO2 concentration in OS scenarios, so achieving 100% reversible temperature requires more than a 100% reversal of atmospheric CO2 concentration. The globally averaged temperature is, by definition, 100% reversible in our simulation setup. However, small differences in reversibility for hemispheric and land vs. ocean temperature suggest spatial differences in the reversibility of regional temperature changes. A grid-cell by grid-cell analysis of surface air temperature reversibility shows that regional temperatures vary in their level of short-term reversibility from about 80% at high latitudes to about 110% across tropical continental regions (Fig. 4). In other words, OS scenarios lead to warmer high latitudes and cooler equatorial and middle latitudes compared to their respective baseline scenarios at the same global warming level. Increased warming in high latitudes is consistent with spatial changes in SAT following OS in idealized simulations23,25, as well as with long-term averages in some CMIP6 models simulating SSP5-3.4OS and SSP1-1.9 scenarios21. In agreement with previous work 20,21,31, our results show that the lower reversibility at high latitudes is slightly more pronounced in the Southern Hemisphere, where hemispheric average temperature is slightly less than 100% reversible, compared to the Northern Hemisphere, where temperature reversibility is slightly above 100% on average (boxplots 21 and 26 Fig. 3). Irreversible warming at high latitudes is related to polar feedbacks, such as the ice albedo feedback, as well as to ocean dynamics, such as changes meridional overturning31,34. Based on our definition of short-term temperature reversibility, irreversible warming at the poles necessitates greater than 100% reversible changes to temperature in the middle and lower latitudes. Further details regarding the range of uncertainty in these spatial SAT reversibility patterns are provided in Supplementary Note 2.Fig. 4Ensemble mean reversibility of the spatial pattern of surface air temperature (SAT).This figure shows the ensemble mean reversibility of SAT at each grid cell at the return year. The equatorial latitudes and the middle latitudes show greater than 100% reversibility, which demonstrates that, following OS, temperature is lower in those regions relative to baseline scenarios. At the poles, temperature is less than 100% reversible, which demonstrates that, following OS, temperatures remain higher in these regions relative to baseline scenarios. The solid black contour line demarcates the 100% reversibility threshold, illustrating the spatial division between regions that achieve full temperature recovery and those exhibiting a persistent warming legacy at the return year. Ensemble mean reversibility of the spatial pattern of surface air temperature (SAT). This figure shows the ensemble mean reversibility of SAT at each grid cell at the return year. The equatorial latitudes and the middle latitudes show greater than 100% reversibility, which demonstrates that, following OS, temperature is lower in those regions relative to baseline scenarios. At the poles, temperature is less than 100% reversible, which demonstrates that, following OS, temperatures remain higher in these regions relative to baseline scenarios. The solid black contour line demarcates the 100% reversibility threshold, illustrating the spatial division between regions that achieve full temperature recovery and those exhibiting a persistent warming legacy at the return year. Given the high degree of irreversibility and path-dependency observed in the slow-responding variables examined here, identifying a robust predictor of the impact of OS is essential for effective climate risk assessment. While global warming levels are commonly used to communicate climate impacts, our results suggest that for stock variables, such as ocean heat and permafrost carbon, the climate state at the end of OS—when temperatures return to baseline levels—is dependent on the preceding temperature pathway. Consequently, we propose that the total exposure to elevated temperatures, captured by the degree-years of OS, provides a more physically grounded framework for quantifying these irreversible outcomes. In the following section, we evaluate the predictive power of this framework by comparing the magnitude of change in irreversible variables to the degree-years metric across our ensemble of 42 scenario pairs. Proportionality between degree-years and overshoot climate outcomes Simulated changes in the most irreversible variables in our study can be well characterized by the time-integrated temperature difference—the degree-years of OS associated with each OS, and baseline scenario pair. In the pairs of scenarios used in this study, degree-years of OS range from approximately 1 to 9 °C-yr (Fig. 1C). We investigated the relationship between the impact of OS on climate variables and the degree-years of OS by performing a linear regression between degree-years of OS and the difference of a given variable between an OS and baseline scenario pair at the return year. We found robust relationships (R2 ~ ≥0.9, see Fig. 5) between degree-years of OS and the impact of OS for the highly irreversible permafrost and ocean variables we identified in the previous section. Interestingly, the relationship appears to be fairly independent of the peak temperature achieved in the OS scenario, particularly for ocean temperature, oxygen, and thermosteric SLR. This is demonstrated by the tight clustering of scenarios with different peak temperatures along regression lines for any given degree-year value (Fig. 5). This relative independence from peak temperature demonstrates that degree-years of OS is a good predictor of the short-term effect of OS on highly irreversible climate variables across a range of OS scenarios compatible with the Paris Agreement (Fig. 5).Fig. 5Proportionality between the degree-years of OS and the impact of OS in irreversible variables.The difference between ocean and permafrost variables in the OS and baseline pairs of scenarios plotted against degree-years of OS for A global total permafrost carbon pool, B permafrost region total soil carbon, C relative sea level height, D maximum meridional overturning streamfunction, E global average ocean oxygen, and F global average ocean temperature. The difference between the OS and baseline of a given variable at the return year is represented by the colored points, in which the colors indicate the peak temperature anomaly reached in the OS scenario. Gray lines represent the relationship between the OS-baseline difference of a given variable and degree-years of OS from 2015 until the return year. R2 values (red) denote the strength of the linear relationship between the climate outcome at the return year and degree-years of overshoot across the 42-scenario ensemble. Proportionality between the degree-years of OS and the impact of OS in irreversible variables. The difference between ocean and permafrost variables in the OS and baseline pairs of scenarios plotted against degree-years of OS for A global total permafrost carbon pool, B permafrost region total soil carbon, C relative sea level height, D maximum meridional overturning streamfunction, E global average ocean oxygen, and F global average ocean temperature. The difference between the OS and baseline of a given variable at the return year is represented by the colored points, in which the colors indicate the peak temperature anomaly reached in the OS scenario. Gray lines represent the relationship between the OS-baseline difference of a given variable and degree-years of OS from 2015 until the return year. R2 values (red) denote the strength of the linear relationship between the climate outcome at the return year and degree-years of overshoot across the 42-scenario ensemble. For globally averaged ocean variables—ocean oxygen, ocean temperature, and relative sea level height—the difference in absolute values between OS and baseline scenarios grows before partially reversing as the return year is approached (Fig. 5C, E, F). This trajectory reflects the slow response of the deep ocean to net-negative emissions and cooling3,10,11. Earlier studies demonstrate that heat sequestered in the ocean interior remains largely isolated from surface cooling, driving irreversible changes9–11. Modelling studies have also demonstrated similar impacts from OS on biogeochemical ocean variables because of the slow response of the deep ocean3,10,11. The persistent sub-surface heat and biogeochemical changes to the ocean lead to irreversible change at the timescales examined here, driving the characteristic looping behavior observed in the transient pathways of globally averaged ocean variables. In contrast to the transient looping paths observed in the ocean variables, maximum meridional overturning, permafrost carbon pool, and total permafrost region carbon all exhibit more linear patterns of irreversibility as it relates to degree-years of OS (Fig. 5A, B, D). For these variables, the impact of OS at the return year appears to be more directly related to the integrated exposure to elevated temperatures as captured by degree-years of OS, with less evidence of the partial recovery seen in ocean variables. This divergence in behavior suggests that while the degree-years metric is a robust predictor for both categories, the underlying physical sensitivity of permafrost and maximum meridional overturning strength to the integrated heat anomaly is more direct than that of the partitioned ocean response. Physical basis for proportionality with degree-years The robust correlations observed between the short-term climate outcomes of OS in irreversible variables and the degree-years metric are rooted in the Earth’s energy balance. Our analysis confirms that the degree-years of OS serve as a high-fidelity physical proxy for cumulative TOA net radiation throughout the OS period. Across the 42-scenario ensemble, we identify a near-perfect linear relationship between the time-integral of the global temperature anomaly and cumulative TOA net radiation, exhibiting an R2 of 0.997 at the return year (Supplementary Fig. 2). Such linearity between integrated forcing and integrated warming is physically consistent with the near-linear relationship between radiative forcing and temperature change observed across many ESMs, particularly for the low-to-moderate warming levels relevant to our study35,36. This proportionality demonstrates that the impact of OS on the irreversible variables examined here is proportional to the total energy added to the system, providing a physically grounded justification for using a time-integrated temperature metric rather than peak warming levels. The proportionality between degree-years of OS and the impact of OS on irreversible variables is primarily due to the nature of the climate components most affected by OS. Most of the highly irreversible variables identified in this study are slow-responding stock variables, such as deep-ocean heat and permafrost carbon, which possess thermal or physical inertia. This inertia prevents these components from equilibrating rapidly with surface temperature conditions. This distinguishes them from faster-responding stock variables, like sea ice volume (boxplot 18, Fig. 3), which can achieve higher levels of short-term reversibility through rapid exchanges with the atmosphere. Among the highly irreversible variables examined, the maximum meridional overturning stands out as a unique flow variable—representing a rate of volume transport in the North Atlantic—that nonetheless exhibits profound irreversibility in the short-term. Its path-dependency arises from slow internal feedbacks that prevent the circulation from recovering to baseline levels as global temperature parity is reached37. Because these slower components effectively remember the preceding temperature trajectory, the degree-years metric serves as a robust predictor for their state at the return year. The vertical distribution of ocean temperature and oxygen demonstrates the physical mechanisms driving high irreversibility in ocean variables. Consistent with previous work, a pronounced reversibility asymmetry exists between the ocean surface and its interior in our simulations, as illustrated in Supplementary Fig. 3. Sea surface temperatures show variable ensemble average anomalies relative to baseline levels at the return year—middle and low latitudes show no difference or slightly cooler sea surface temperatures, while the high latitude sea surface temperatures are generally higher except where there is substantial meridional overturning like in the North Atlantic (Supplementary Fig. 3, Panel C). In contrast, the ocean interior—particularly between 200 and 1000 m—retains a substantial thermal legacy (Supplementary Fig. 3, Panel C). During the OS period, excess heat absorbed at the surface is transported to deeper layers through downwelling. Once sequestered in the interior, this heat can no longer freely exchange with the atmosphere, ensuring a persistent impact on thermosteric SLR and interior temperatures regardless of surface cooling. This physical heat retention is mirrored by a corresponding displacement of ocean oxygen, where disrupted ventilation patterns leave a biogeochemical footprint in the deep ocean following OS (Supplementary Fig. 3, Panel D). The ensemble average differences between the surface and sub-surface between OS and baseline scenarios (Supplementary Fig. 3, Panels E, F) show that, particularly for ocean temperature, the surface ocean sees a reversal to the baseline by the end of the century, while changes to the deep ocean persist. The reversal of surface-level impacts explains the characteristic looping behavior of the global ocean trajectories in Fig. 5. While surface anomalies drive the initial divergence between scenarios, their subsequent recovery during the cooling phase causes the trajectory to curve back toward the baseline. Consequently, the remaining impact at the return year—which correlates with the degree-year metric—is fundamentally an expression of the sequestered sub-surface debt that surface-level equilibration cannot erase. The physical legacy of OS is equally apparent in the spatial distribution of permafrost carbon loss. As shown in Supplementary Fig. 4, the impact of OS is concentrated in high latitudes across the Arctic. These losses are driven by the slow, downward conduction of heat into the soil, which progressively thaws carbon-rich layers and makes more soil available for microbial decomposition38. Because the refreezing of thawed permafrost and the subsequent recovery of lost soil carbon occur on millennial timescales12, the impact of OS on permafrost carbon is effectively determined by the total thaw experienced during the OS. This is why permafrost carbon is largely irreversible, even though median permafrost extent exhibits some reversibility by the return year (Supplementary Fig. 4, Panel B). Permafrost carbon behaves as a one-way stock variable on these timescales; once the soil has thawed and carbon is released, carbon loss cannot immediately reverse upon the return to baseline temperatures, which is why OS leads to irreversible permafrost carbon loss4–7. This ensures that the total carbon loss is driven by the cumulative duration and magnitude of thermal exposure above thaw thresholds, making the degree-years metric a highly effective predictor of the permafrost carbon legacy (Fig. 5A). Discussion Due to the lack of progress on reducing emissions, it is increasingly important that we understand the climate effects of OS scenarios across a range of climate variables. An important question in furthering our understanding of the climate outcomes of OS is the question of how the severity of changes relates to the magnitude and duration of OS. Here, we identify variables that exhibit the highest degree of short-term irreversibility and confirm that there are nonnegligible climate effects of OS on these variables, leading to path-dependent climate outcomes. Our findings align with higher-complexity modelling studies, highlighting that there is a lasting climate cost of OS pathways. For instance, we see a substantial slowdown in the maximum meridional overturning as a result of OS that is not reversed in the short term, although it should be noted that in centuries-long simulations, models show varying amounts of recovery23,31. Similarly, we see cooling in the North Atlantic driven by this slowdown 23,31. Other regional temperature differences immediately following OS align with those in other studies—even those with centuries-long simulations allowing more time for equilibration—showing lower reversibility on land than over the ocean and lower reversibility in the Southern Hemisphere20,21,31. The impacts on the deep ocean in our simulations are also similar to OS modeling studies with longer equilibration times, with persistent changes to the deep ocean affecting thermosteric SLR, temperature, and biogeochemistry9–11,31. Our results also align with other studies in showing irreversible short-term permafrost carbon loss 4–7, even though there is partial recovery of permafrost extent23. We find that degree-years of OS are linearly related to the outcome of these slow-responding irreversible permafrost and ocean variables. Surprisingly, the relationship between degree-years of OS and the outcome of irreversible variables appears to be relatively independent of the peak temperature reached in the 42 pairs of below 2.0 °C scenarios simulated here. This finding demonstrates that degree-years of OS could potentially be used to capture the impact of OS for slow responding variables in a range of scenarios compatible with the Paris Agreement. We acknowledge that our definition of reversibility is based on decadal OS timescales that are shorter than the equilibration time for many climate variables. However, defining reversibility at a point in time this century, immediately following a period of OS, is relevant from climate impact and adaptation planning perspectives. This is particularly important for the practice of communicating climate impacts in terms of global warming levels 39, which is problematic in OS scenarios, given that the path to a given level of warming is an important determinant of climate outcomes. Our finding that degree-years of OS is a strong predictor of the effect of OS for several highly irreversible climate variables suggests that degree-years of OS could be used to predict the difference in climate outcomes between a stabilization scenario and an OS scenario for the most irreversible set of climate variables. Methods Scenarios We use 42 pairs of scenarios from the ENGAGE project. In the ENGAGE project, Riahi et al.33 developed pairs of scenarios using several integrated assessment models where one scenario was constrained by a remaining carbon budget (RCB) that cannot be exceeded at any point before 2100 (i.e., the baseline scenario) and one scenario where the same RCB could be temporarily exceeded but must be returned to by 2100 (i.e., the OS scenario). The RCB constraints in the 42 scenarios used range from 450 Gt CO2 to 1600 Gt CO2. Forcing from non-CO2 emissions was determined by assumptions embedded in individual integrated assessment models; therefore, models simulated different temperature outcomes for the same RCBs. Pairs of scenarios in the ENGAGE ensemble were excluded from our analysis if: (1) annually averaged GSAT in the OS scenario failed to return to GSAT in the baseline scenario before 2100 due to differences in non-CO2 emissions, or (2) the OS scenario had less than one degree-year of OS. We did not control for internal climate variability using multi-year averages since our results are averaged across an ensemble of scenarios, and the intermediate complexity model we use does not exhibit much internal variability40,41. Model The University of Victoria Earth System Climate Model (UVic-ESCM) is an intermediate complexity climate model with a spatial resolution of 3.8° of longitude and 1.8° of latitude 40. The model couples atmospheric, oceanic, land surface, and sea ice components while maintaining relatively low computational demands compared to more complex ESMs 40. The atmospheric component uses a two-dimensional single-layer energy-moisture balance model, effectively simulating large-scale heat and moisture transport without detailed atmospheric circulation 40. Thus, surface wind and cloud albedo are both held static at prescribed values based off observations. In contrast to the atmosphere, the ocean representation is more comprehensive, featuring a three-dimensional general circulation model that resolves major ocean currents and includes dynamic sea ice and marine biogeochemistry modules 40. The land component incorporates vegetation dynamics by simulating competition between five plant functional types (PFTs)—shrubs, C3 and C4 grasses, and needleleaf and broadleaf trees—and their response to environmental changes 40. The land model is coupled to carbon cycle models representing the marine-terrestrial carbon cycles40,41. The interactive carbon cycle includes physical and biological ocean carbon cycling, sedimentary carbon cycling, and a representation of land carbon via the forementioned PFTs, permafrost carbon storage, and soil carbon storage41–44. The UVic-ESCM’s coarse resolution and simplified atmosphere enable computationally efficient century-long simulations of ensembles of scenarios, making it well-suited for this study. Other reasons the UVic-ESCM is well suited for this study is that, first, the model features a fully dynamic 3D ocean and a comprehensive interactive carbon cycle including permafrost and vegetation, capturing the physics of slow-responding stock variables. Second, its computational speed allowed for the simulation of ~100 initial scenario pairs (~200 total simulations), from which 42 pairs were retained after the temperature-convergence screening. This provides an ensemble size capable of robustly identifying the degree-years relationship across varied OS trajectories that would otherwise be computationally prohibitive for high-complexity ESMs. Third, while the simplified atmosphere may not simulate all circulation-dependent tipping points, the risk of exceeding such thresholds is substantially reduced at the relatively low warming levels (~≤2.0 °C) analyzed here45. However, two key limitations should be noted: (1) the non-dynamic atmosphere means changes in precipitation or atmospheric circulation are not well captured, and (2) only thermosteric SLR is represented. Therefore, changes in precipitation are ignored in this study, and all sea-level rise results in this study do not include SLR due to melting ice sheets. Simulations Future simulations in the UVic-ESCM were driven by fossil and land use change CO2 emissions as well as aggregated non-CO2 forcing from the ENGAGE project, the latter being estimated using MAGICC33. Given that the scenarios from the ENGAGE project do not include spatially explicit land cover change33, land cover in our simulations was held static at preindustrial values. Similarly, the scenarios used here do not include spatially explicit changes to aerosol emissions, so only the aggregate globally averaged forcing of aerosols is included. For consistency, the historical simulation used to calculate temperature anomalies in Fig. 1A was emissions-driven with static land cover and prescribed globally averaged aerosol forcing. The combined non-CO2 greenhouse gas and aerosol forcing for the historical simulation was harmonized with the future scenarios non-CO2 GHG and aerosol aggregated forcing in the year 2015. Solar forcing was held constant in both the historical and future simulations. Volcanic forcing was included in the historical simulations but was held constant at the average historical value after the year 2015. Supplementary information Transparent Peer Review File Supplementary Information File Transparent Peer Review File Supplementary Information File