The loss of mountain glaciers contributes to sea-level rise1, influences water resources in periods of drought or low precipitation2 and induces natural hazards3,4. Glaciers are projected to continue losing mass through the twenty-first century5 and beyond6, with the extent and pace of mass loss mainly driven by the increase in surface temperature. Global temperatures are rising at a record rate7 and, as a result of past inaction and insufficient emissions reductions, today, the risk of exceeding a global mean temperature level of +1.5 °C above pre-industrial levels as early as in the next decade is increasing1,8. In light of such an overshoot, substantial carbon removals will be required to reverse the warming to back below 1.5 °C over the long term9. The consequences of such an overshoot on the climate, ecosystems and societies10, as well as on hydrology and water resources11, are not well understood. Even if the global mean temperature returns to the target temperature, regional temperatures may not10,12,13. The magnitude and duration of the overshoot also matters. For example, the time spent above 1.5 °C is quasi-linearly related to projected sea-level rise in 230014 and will affect future emissions from other systems that respond more slowly (for example, permafrost and peatlands)12. Furthermore, global warming overshoots increase the risks of crossing potential tipping point thresholds15,16,17,18, which may lead to irreversible glacier changes19. However, if the local climate and characteristics are favourable, glaciers can regrow. Notable examples include the recent Little Ice Age period in Central Europe20 or the present-day regrowth of Crater Glacier, United States, following the eruption of Mount St. Helens21. Given that glaciers respond slowly to climate change and that regrowth takes longer than retreat22, the impacts of overshoot scenarios are likely to be long-lasting.

One challenge with quantifying overshoot impacts is a lack of suitable climate projections. Only a few climate models that participated in the sixth phase of the Coupled Model Intercomparison Project (CMIP6) have run global temperature overshoot simulations23 (Supplementary Fig. 1). Moreover, these simulations end in 2300, which is too early to capture glacier regrowth. Furthermore, these simulations lack corresponding reference temperature stabilization scenarios for comparisons because regional temperatures may evolve even under stabilized global mean temperatures. Climate simulations from the GFDL-ESM2M Earth System Model (ESM)24,25 address this gap by using an adaptive emissions reduction approach26 to modelling different stabilization and temporary overshoot scenarios for 1.5 °C through 250013.

Here we conducted a systematic analysis of irreversible glacier changes and the associated impacts for all (more than 200,000) mountain glaciers globally under temperature stabilization and overshoot scenarios. Forcing from both idealized experiments and ESM-generated stabilization and overshoot scenarios through 2500 was used in conjunction with the Open Global Glacier Model (OGGM)27 to reveal key drivers of change and characteristic time scales that vary regionally. We found that the potential regrowth of glaciers after a temperature overshoot causes a temporary depletion in glacier runoff, which we refer to as ‘trough water’. We highlight the regional variability and the potential future impacts of trough water for select glaciated basins. Overall, our projections reveal that long-term irreversible glacier mass loss is highly sensitive to the magnitude of the local temperature overshoot, and that avoiding overshoot results in a reduced glacier contribution to sea-level rise and smaller changes to water availability.

Glacier characteristics strongly shape their response under overshoot. We illustrate this by comparing the idealized temperature overshoot and stabilization experiments for two glaciers with different climatic settings and response times—the fast-responding Aletsch Glacier and the slow-responding Wykeham Glacier South. The pace and timing of the temperature increase, as well as the occurrence of an overshoot, result in considerably different glacier mass and runoff projections (Fig. 1). The rate of temperature increase drives the mass loss rate, although the relative changes vary considerably between the fast- and slow-responding glacier (Fig. 1a–c). The temperature overshoot causes both glaciers to lose substantially more mass during the overshoot phase, which the fast-responding glacier takes decades to centuries to recover from (Fig. 1b). The slow-responding glacier does not recover by the end of the idealized 500-year simulation (Fig. 1c), but the experiments converge to the same steady state after millennia (Extended Data Fig. 1).

a–e, Applied temperature change (a), glacier mass relative to the initial state (b,c) and glacier runoff relative to the initial state (d,e) for a typical fast-responding glacier in the mid-latitudes (Aletsch Glacier) (b,d) and a slower-responding glacier in the high latitudes (Wykeham Glacier South) (c,e). The overshoot phase refers to when the temperature exceeds the stabilization temperature, and the legacy phase refers to the period of time after the temperature stabilizes until the glacier reaches its new steady state. Precipitation is kept constant. Dry-melt-season runoff refers to the three driest months over the melt season and is shown relative to the annual initial glacier runoff. The runoff y-axis scales in d and e are different. Idealized changes for the largest glaciers in five other regions are shown in Extended Data Fig. 1. See Methods for additional detail on the idealized experiments.

Full size image

In steady state, the glacier mass and glacier runoff (comprising the meltwater plus liquid precipitation from the initially glacierised area) are constant (Fig. 1d,e). When the temperature increases, the glaciers begin losing mass, and annual as well as dry-melt-season runoff increases. As the glaciers retreat and the mass loss rates decrease, a maximum annual and dry-melt-season runoff is reached, referred to as ‘peak water’28. After decades to centuries, the annual runoff returns to the same level as in the initial state, although the seasonal distribution of runoff may change.

The differences between the experiments over the first century show that the magnitude of the peak water is considerably smaller when the temperature and glacier mass changes are slower. For faster-responding glaciers, the overshoot experiment highlights a unique phenomenon in which annual glacier runoff temporarily decreases below the initial level as the glacier regrows during the temperature decline phase (Fig. 1d). As a response to glacier regrowth, water available from the initially glacierised area is reduced over a certain period—trough water (as opposed to peak water)28,29.

The influence of a temperature overshoot on a given glacier depends on its response time (Fig. 1 and Extended Data Fig. 1), which depends on local characteristics, such as local temperature change, the glacier surface slope and/or the mass-balance gradient22,30. These model results under idealized climate scenarios suggest glacier change is reversible for a typical mountain glacier if the glacier is given enough time. In reality, precipitation also changes with increasing temperatures (Supplementary Fig. 2). Thus, trough water can occur in the absence of glacier regrowth. Glaciers are also not in a steady state. Hence, ESM simulations for overshoot and corresponding stabilization scenarios are needed to quantify the time needed for them to regrow or converge and the associated consequences for water availability.

Stabilization and overshoot scenarios from the GFDL-ESM2M were used to force the OGGM and quantify the mass changes in all glaciers globally from 2000 to 2500 (Fig. 2). Although our applied scenarios resemble our idealized temperature change experiments, the global projections explicitly account for local and regional variations in temperature and precipitation. Specifically, the local projected signal differs from the global temperature overshoot signal because the glacier-covered areas are, on average, subject to greater temperature change than the global mean. For example, the glacier-area weighted temperature is up to 1 °C warmer than the global mean for the 3.0 °C stabilization scenario, consistent with previous findings1.

a, Global mean surface air temperature change ΔT relative to the 2000–2019 average (left y axis) and above pre-industrial levels, as defined by the IPCC AR6 (ref. 1) (right y axis). b, Global glacier-area weighted near-glacier temperature change relative to 2000–2019. c,d, Global glacier mass remaining (c) and sea-level rise contributed by glaciers (d) for the same scenarios. Time series in a and b are smoothed with a 21-year centred rolling mean. Past changes from 2000–2019 are shown in black. The global precipitation changes are provided in Supplementary Fig. 9.

Full size image

Even if global warming ceased today at 1.2 °C above pre-industrial levels7, the world’s glaciers would still lose 30% of their mass by 2500, relative to 2020 (Fig. 2c), corresponding to an additional 90-mm sea-level equivalent (SLE) (Fig. 2d). Under the 3.0 → 1.5 °C overshoot, glaciers temporarily lose up to 16% more mass (50-mm SLE) than with 1.5 °C stabilization. Mass loss occurs more quickly in the overshoot scenario. After 2250, glacier mass loss slows and almost halts in the 3.0 → 1.5 °C overshoot, whereas long-term mass loss continues in the 1.5 °C stabilization. Although the influence of the overshoot diminishes over time, the 3.0 → 1.5 °C overshoot still results in 11% more global glacier mass loss in 2500 (34-mm SLE). The glacier contribution to sea-level rise is never negative, even in the 3.0 → 1.5 °C overshoot scenario. Beyond 2500, glaciers would continue to lose an additional 6% mass under 1.5 °C stabilization (Extended Data Fig. 2). Under the 3.0 → 1.5 °C overshoot scenario, glaciers stabilize faster because enhanced mass loss occurs due to the stronger temporary forcing during overshoot. These results confirm that overshoot-induced global glacier mass loss, and the corresponding contribution to sea-level rise, is irreversible for centuries to millennia post-overshoot.

Global mass change is dominated by the regions with the most glacier mass, which have slow response times (Fig. 3 and Extended Data Fig. 3). For example, glaciers in many high-latitude regions (Alaska, Greenland periphery, Russian Arctic, Svalbard and Jan Mayen, Subantarctic and Antarctic islands) account for 66% of the global glacier mass and all respond slowly to climate forcing (Fig. 3c, ‘slow response’ cluster). Mass loss continues through 2500 in these regions, and no regrowth occurs in the overshoot scenario because these relatively flat glaciers are too slow to respond (Fig. 3f–h).

a, Regional mass differences between the 3.0 → 1.5 °C overshoot and the 1.5 °C stabilization scenario for all 19 glacier regions and the sum over the three clusters. Clusters were selected from these mass differences and represent regions with similar overshoot influence on mass change (Methods). b, Nineteen glacier regions colour-coded by the three clusters, with dots showing glacier locations. c–e, Glacier mass projections for the sum of each cluster with a slow response (c), a small overshoot influence (d) and a fast response (e). The percent refers to the cluster’s fraction of the 2020 global mass. f,g, Regrowth after 3.0 → 1.5 °C overshoot described by the year 2500 minus the minimum glacier mass versus the glacier mass in 2500 under 1.5 °C stabilization (f) and the regional glacier-area weighted glacier surface slope (g). h, Mass difference in 2500 between 1.5 °C stabilization and 3.0 → 1.5 °C overshoot versus the maximum regional temperature difference of the same scenarios. Numbers are given as a 21-year centred rolling mean in percent relative to 2020. Past changes from 2000 to 2019 are shown in black in c–e. All scenarios and regions are shown individually with their 2020 glacier mass in Extended Data Fig. 3. Basemap in b from Natural Earth (https://www.naturalearthdata.com/).

Full size image

Other mid- to high-latitude regions show moderate to no influence of climate overshoot (western Canada and United States, Arctic Canada, Iceland, Scandinavia and Southern Andes, with 30% of the global glacier mass) (Fig. 3d, ‘small overshoot influence’ cluster). The similarity between the scenarios is due to the limited remaining glacier mass in some regions, the dominant signal from present-day committed warming (especially in Arctic Canada South, with its substantial ice caps) and regional temperature change projections being similar between the overshoot and stabilization scenarios (Iceland).

Finally, several lower-latitude glacier regions in steep mountain areas (Central Europe, all Asian regions, low latitudes and New Zealand, with 5% of the global glacier mass) have faster response times than the idealized example of a fast-responding glacier (Fig. 1d). Their glaciers shrink considerably under the 3.0 → 1.5 °C overshoot until around 2350, but almost regrow to the 1.5 °C stabilization levels by 2500 as temperatures cool back down to the target level (Fig. 3e,g,h, ‘fast response’ cluster). Within this cluster, the rates of glacier regrowth depend on local geographic and climatic factors, and glacier regions show a wide range of responses (Fig. 3f).

The projected glacier runoff was aggregated to water-resource-relevant basins instead of glacier regions. We projected that glacier runoff would peak and then decline before 2100 in many basins and across all scenarios (Supplementary Figs. 3–5), consistent with previous studies28,31,32. However, our study assessed large-scale glacier runoff beyond 2100, where glacier runoff is projected to be permanently lower, relative to 2000–2020, in 75% of the major glaciated river basins across all scenarios because the 2000–2020 period is often near peak water. Interdecadal change in runoff varies between basins due to different glacier response times, initial states (pre- or post-peak water) and regional climate responses. The ‘fast response’ cluster (Fig. 3c) represents a small portion of the global mass, but is concentrated in mid-latitudes, in heavily populated mountain basins where glaciers are crucial to the water supply33. In these basins, trough water (Fig. 1c) may be a consequential temperature overshoot risk.

Dry-melt-season runoff was evaluated for seven relatively arid and heavily glaciated basins in Asia, South America and Europe that experience regular periods of low precipitation such that potential trough water may have the greatest impact (Fig. 4). The selected basins all experienced a regional temperature overshoot in the global overshoot scenarios, regaining most of their glacier mass by 2500 and forming trough water in the dry-melt season. In the most extreme case, the Rapel Basin is projected to have more than 300 years of trough water and, at its minimum, at least 50% less seasonal runoff from initially glacierised areas. Dry-melt-season trough water is more intense and persistent than annual-runoff trough water (Supplementary Figs. 3–5). For three of the basins (Indus, Ysyk-Kol and Tarim), the projected precipitation increases with rising temperature in the dry months of the melt season, which explains why the dry-melt-season runoff is greater for the 3.0 °C stabilization scenario than the 1.5 °C stabilization or 3.0 → 1.5 °C overshoot scenarios (Fig. 4b), with the annual meltwater from snow and ice melt being similar (Supplementary Fig. 6).

a, Glacier mass relative to 2020 and glacier-area weighted basin-wide temperature changes to 2000–2019 (both 21-year averaged). b, Glacier runoff relative to 2000–2050 and respective precipitation relative to 2000–2019 (both from the three driest months in the melt season and 51-year averaged). For each basin, we give the glaciated area and the number of years with estimated trough water with the maximum difference (Δmax). We give the driest melt months in Supplementary Fig. 8. We only show here three scenarios to allow for a better comparison. The same illustration for only the melt components is in Supplementary Fig. 6. The map shows the outlines of the seven selected basins. All basins have an aridity index <2 (according to ref. 2) and are initially above 0.9% glaciated. The projected annual glacier runoff of all 60 glaciated basins is in Supplementary Figs. 3–5. Map from Natural Earth (https://www.naturalearthdata.com/).

Full size image

Overall, annual trough water occurs in 31 out of the 60 glaciated basins after 2170 (Supplementary Fig. 7). Trough water in 24 of those basins is caused by glacier mass regrowth after the temperature overshoot. For basins where precipitation decreases with warmer temperatures, longer and more intense trough water becomes more likely (consistent with the idealized experiments) (Supplementary Fig. 2).

Our results show that a global temperature overshoot results in substantially more glacier mass loss for decades to centuries once the cooling phase begins. The degree of temporal irreversibility is dependent on the magnitude of the overshoot, but even small overshoots will have long-lasting impacts on glacier contributions to sea-level rise (Extended Data Fig. 3). Glaciers respond variably to warming due to differences in their geometry and sensitivity, with committed mass losses dependent on the glaciers’ response times. To understand the regional water availability of glacier runoff, it is crucial to consider whether the 2000–2020 period is near peak water (for example, the basins in Central Europe) or if peak water may occur only until the twenty-second century (for example, the Tarim Basin) (Supplementary Figs. 3–5). Our results show that steeper glacier regions respond faster to overshoots (Fig. 3g), experiencing quicker mass loss initially and partial regrowth later, which aligns with findings from response-time studies (for example, ref. 22). Slower-responding glaciers strongly dampen the overshoot signal and reduce, or completely prevent, trough water (Fig. 1 and Extended Data Fig. 1). Furthermore, some glacier regions experience substantial warming from the overshoot (for example, the Russian Arctic, Arctic Canada North, Svalbard and Jan Mayen), while others see weaker (for example, the Southern Andes) to no (Iceland) overshoot, or ongoing regional warming despite global stabilization (for example, the Greenland periphery, Iceland, Subantarctic and Antarctic islands) (Fig. 3h and Extended Data Fig. 3). The stabilization and overshoot scenarios from the only available ESM (GFDL-ESM2M13) enable us to disentangle overshoot effects from changes occurring during temperature stabilization. Although idealized approaches have been developed to investigate committed and overshoot responses34,35,36, the GFDL-ESM2M simulations consist of, to our knowledge, the only approach that constructs both temperature-focused stabilization and overshoot scenarios with a historical period and realistic carbon dioxide (CO2), aerosol and non-CO2 forcings, and extends until 2500. While the physical response of the GFDL-ESM2M, particularly for temperature and precipitation patterns, falls well within the range of multi-model means37,38, our study shows only one potential outcome of a regional overshoot glacier response. Regional temperature and precipitation responses during overshoot and stabilization phases can differ between ESMs12. Consequently, the regional glacier response presented in this study may also vary depending on the ESMs used, which highlights the need for additional ESM overshoot simulations to quantify the uncertainties with our estimates. Nevertheless, other ESM stabilization studies have shown similar regional warming trends around the Subantarctic and Antarctic islands during global stabilization phases39,40,41, underscoring the potential robustness of our results.

An overshoot results in irreversible glacier change over many centuries. We use the term ‘irreversible’ as defined by the Global Tipping Points Report 2023—that is, “a change in a system that is not reversed under the same boundary conditions that triggered it, or that takes significantly longer to recover from than the time it took to reach”19. Our model setup using the OGGM did not include positive feedbacks, such as the climate-independent retreat of calving glaciers if destabilization thresholds are reached42,43,44, surface darkening by decreased snow cover, dust, black carbon, thin debris or glacier algae45,46,47, glacier-related slope failures or thermokarst processes48,49, nor negative feedbacks, such as thicker debris cover, potentially decreased ice flow50 or terrestrial uplifting from glacial isostatic adjustment51. However, our model setup accounted for the positive mass-balance elevation feedback and the negative glacier retreat to higher altitudes feedback52. With our model setup, we found no hysteresis in ice caps, as has been theorized for ice sheets16. One possible reason is OGGM’s flowline approach, which employs the shallow-ice approximation and only partially captures ice cap dynamics. However, this influence appears limited because we still projected the largest ice caps would melt away entirely under the current climate, given sufficient time53. Non-ice-cap-like glaciers—thus, typical mountain glaciers—can retreat to higher altitudes and regrow from there. Although the first studies including individual feedbacks in regional models indicated limited influence on regional projections (for example, debris cover4, calving54), we expect the cumulative impact of all non-accounted feedbacks to increase the temporal irreversibility of glacier mass loss beyond our current projections. Additionally, the ESM and glacier model are not coupled—that is, related feedbacks to the ocean or atmosphere are not considered. We only used one global glacier model, but qualitatively compared the model response to two other global glacier models using the projections from ref. 55 and one regional model56 under CMIP6 temperature overshoots. We observed regrowth in fast-responding regions, but the regrowth magnitude up to 2300 varies greatly across glacier models (Supplementary Fig. 1). The different temperature overshoots and lack of comparative stabilization scenarios in CMIP6 prevent the quantification of uncertainty from ESMs and further analysis of overshoot glacier impacts, such as sea-level rise or trough water (Methods).

We have introduced the concept of trough water for glacier runoff in temperature overshoot and stabilization scenarios. This projected reduced glacier runoff occurs specifically in steeper fast-responding non-Arctic basins when glaciers regrow. However, an overshoot scenario also impacts the adaptation planning of near-future water availability, with peak water being more intense and shorter under the overshoot scenario, even for basins without trough water. How relevant a potential glacier-runoff trough water might be for river basin discharge downstream in periods of drought or low precipitation depends on many aspects of the discharge regime. The different non-glacierised components of mountain catchments (for example, snow, vegetation, permafrost and groundwater storage) and non-accounted glacierised components (for example, debris cover) buffer runoff on different time scales, and glacier runoff is often only a small contributor to the entire runoff downstream57.

Future studies need to assess whether, and how far downstream, the signal of glacier-runoff trough water remains important when forcing temperature overshoot scenarios on coupled glacier–hydro large-scale modelling approaches. Coupling large-scale glacier models with hydrological models shows that glaciers alter future discharge32, so representing glaciers explicitly is important in hydrological models58. For that, absolute glacier runoff contributions are necessary, despite their large uncertainties31. Separating glacier-sourced runoff ablation components (here approximated by the glacier meltwater components) into a balanced (glacier mass compensated by accumulation on an annual basis) and imbalanced (present-day committed loss) term could give further insights57,59.

The world after an overshoot will be different from the world before the overshoot12. Glacier impacts on future water availability and sea-level rise are just two examples of the uncertain dangers resulting from delayed action. Surpassing 1.5 °C global warming necessitates returning to, or below, 1.5 °C to avoid long-term tipping points in the climate system15,18,19. Counting on climate mitigation by overshoot from cooling temperatures over the coming centuries is risky due to physical climate system feedbacks, uncertainties on the scalability of CO2 removal techniques, and the nonlinear dependence of glacier response on the temporal pattern of mitigation12,60. Paradoxically, having reached 3.0 °C global warming, some basins’ glacier runoff might be reduced more under global cooling than if the climate were held steady there because of glacier regrowth. Cooler temperatures might be locally detrimental, raising fears of trough water in glacier-dependent sectors. Trough water therefore puts local climate adaptation interests in tension with global climate mitigation interests. Relying on future carbon reductions rather than taking action in this decade will have unforeseen consequences, including local disputes over cooling impacts. Immediate emissions reductions are essential to avoid competing motivations in a post-overshoot world. This study underscores the urgency of near-term emissions reductions to limit peak global warming to the lowest possible level.