An important positive aspect of the TIPMIP protocol is its simplicity, with only CO2 emissions (positive, zero, and negative) needing to be prescribed. Using emission rates derived from each model's TCRE, we aim to control the GMSAT warming rate to be common across models during the ramp-up, and ensure models switch to zero-emissions at the same global warming level, relative to their pre-industrial temperatures, after the same rate and duration of global warming. These assumptions rest on an assumed linearity of TCRE across warming levels (and increasing cumulative CO2 emissions) in the ramp-up. Put another way, we expect a unit emission of CO2 (X GtC) to cause the same unit increase in global mean surface air temperature (Y °C), whether the unit emission happens early in a ramp-up run (when accumulated CO2 emissions and induced warming are relatively small) or later in the same run (when accumulated CO2 emissions are larger and induced warming is also larger). This assumption of linearity appears accurate at lower warming levels (modest accumulated CO2 emissions) but becomes less accurate at higher warming levels (greater accumulated CO2 emissions), resulting in a slight deviation in some models in the rate of warming away from the target 2 °C per century at higher GWLs (Krasting et al., 2014). The degree to which this deviation complicates our analysis will be one important outcome of the first round of simulations and will inform further development of the protocol. While any global warming rate could, in principle, be realized, we chose 2 °C per century as this is the approximate rate of observed warming over the past ∼50 years.

We do not explicitly control the evolution of GMSAT in the zero-emission runs. Our a priori expectation is that changes in GMSAT during these phases (the Zero Emissions Commitment (ZEC), Jones et al., 2019) will be close to zero (MacDougall et al., 2020). That said, each model's ZEC response is an emergent property of that model – which may depend on global warming level and even on the length of the zero-emission run. While we force the ramp-down runs to have a negative, mirror image emission of the ramp-up, this does not guarantee each model will cool at -2 °C per century. The realized cooling rate in each model will depend on the amount of excess heat sequestered during the ramp-up and zero-emission phases, as well as how the carbon cycle, atmospheric CO2 concentrations, and climate, respond in each model during the negative emission phase. We might expect a near mirror image ramp-down global cooling, but this too will be an emergent property of each model, likely related to its ZEC (Koven et al., 2023). It is important to reiterate our experiment protocol aims to control the evolution of GMSAT across models, acknowledging this may only be approximated in the zero and negative emission phases. In doing this, the cumulative CO2 emissions in each model will differ significantly. For example, while models all branch into zero-emission runs at the same GWL after the same duration of warming, the cumulative emitted CO2 and the atmospheric CO2 concentration at this branch point will differ across models. This is a direct result of controlling for GMSAT at the expense of cumulative CO2 emissions and needs to be considered in analysis of the resulting ensemble.

We choose a negative emission rate in the ramp-down that is a mirror image of the ramp-up to investigate the degree of symmetry in global warming and cooling under opposite and equivalent positive and negative emissions. This level of negative emissions (ranging from -8 to -18 GtCyr-1 depending on each model's TCRE) is clearly far greater than anything technically feasible today and probably also in the future (Gidden et al., 2025). We therefore recommend in the potential extensions to the phase 1 experiments (Sect. 3.2) that groups repeat the ramp-down phase using smaller negative emission rates (e.g. 50 %, 25 %, 12.5 % of their original rate).

While the simplicity of our protocol is a strong positive, it is also its main weakness. This is notably because we ignore other important anthropogenic forcings. In particular, emissions of aerosol and aerosol precursors. The protocol does include natural aerosols (as simulated in each individual model esm_piControl), including any response of natural aerosols to warming/cooling along the simulation pathway. Anthropogenic aerosols have provided an important (time varying, spatially heterogeneous) negative radiative forcing during the historical period (1850 to near present-day), balancing a significant fraction of the historical warming due to greenhouse gas (GHG) emissions (Gillett et al., 2021; Smith and Forster, 2021). In recent decades anthropogenic aerosol emissions have decreased, impacting the Earths top of atmosphere (TOA) Energy Imbalance (Quaas et al., 2022; Yuan et al., 2026; Loeb et al., 2025), potentially contributing to an acceleration of global warming (Foster and Rahmstorf, 2026) Aerosol emissions also vary significantly across different future emission scenarios, such as the CMIP6 Shared Socioeconomic Pathways (SSPs) (Gidden et al., 2019). Due to their short atmospheric lifetime and strong interaction with low-level clouds (Williams and Igel, 2021; Fons et al., 2023) anthropogenic aerosol forcing has a specific geographical structure (Wilcox et al., 2023). Over the historical past, aerosol induced cooling is far greater in the Northern Hemisphere (NH) than in the Southern Hemisphere (SH), while GHG forcing is more homogeneous. As a result, the total historical anthropogenic radiative forcing at the top of atmosphere (TOA) has a distinct NH to SH gradient. This gradient, as well as the detailed spatial structure of aerosol forcing, will not be captured in our protocol and may have important consequences for simulated AMOC changes (Menary et al., 2020).

A second important omission is human land-use. Human land use has provided a negative forcing of the Earth system over the historical period. For example, in the HadGEM3-GC3.1 CMIP6 historical run, human land-use forcing was -0.11 Wm-2, ∼6% of the total historical anthropogenic forcing in that model (Andrews et al., 2019). At regional scales, particularly with respect to tropical forests and their resilience, human land use (i.e. deforestation) plays a critical role (Flores et al., 2024), potentially of similar importance to global climate change (da Cruz et al., 2021, Boulton et al., 2022). For potential tipping of forest ecosystems (particularly the Amazon) our experiment protocol therefore neglects an important driver. Introduction of human land-use and anthropogenic aerosol are high priorities for a tier 2 set of experiments, but require careful testing before they can be included, as our aim will be to retain as much commonality as possible in warming pathways across models.

The protocol does not consider non-CO2 anthropogenic GHG emissions, such as methane (CH4), ozone (O3), nitrous oxide (N2O) or chloroflurocarbons (CFCs and HCFCs). While these are important forcers, for example when combined giving an historical radiative forcing ∼75 % of that due to anthropogenic CO2 emissions (Chap. 7, IPCC, 2023a), their lifetime, horizontal distribution, and impact primarily on the longwave part of Earth's radiation budget, suggest to a first order their forcing can be mimicked by an equivalent CO2 emission. Omission of anthropogenic CFCs and HCFCs also means there will be no stratospheric ozone depletion (Solomon, 2019) in our experiments.

Because our protocol is forced solely by CO2 emissions, with no aerosol or other well-mixed greenhouse gas (WGHG) emissions, it is likely that when the idealized runs reach GWL=2 and 4 °C (the start of the zero-emission runs) they will have experienced a larger cumulative emission of CO2 than equivalent multi-gas and land-use scenarios at the same GWL. This is primarily because the positive radiative forcing arising from non-CO2 WGHGs (e.g. methane, ozone, nitrous oxide etc) in the multi-gas scenarios is being achieved (in a warming sense) by CO2 emissions in our idealized runs. This is partially offset by not including aerosol (negative radiative forcing). As a result, at common GWLs, models will have different atmospheric CO2 concentrations and will have absorbed different amounts of emitted carbon into the ocean and terrestrial biosphere, impacting responses such as plant photosynthesis or ocean acidification. These caveats should be considered when analysing ecosystem responses.

The Tier 1 protocol requests only one member per model. This is to maximise the number of models contributing. In some cases, tipping events are expected to be rare occurrences. Hence, a large ensemble would help in sampling such events, including potential stochastic forcing of tipping (Romanou et al., 2023), while also helping identify when, and under what conditions, a tipping event is more likely to occur. This can be partially addressed by the multi-model aspect of the protocol, with 11 ESMs already running these experiments. Nevertheless, we recognize the need for more ensemble members. For instance, a bifurcation in the AMOC has been shown to exist in one ESM and could therefore have implications for the timing and reversibility of AMOC-related tipping (Romanou et al., 2023). We therefore encourage groups to consider running additional ensemble members (started from different time points in their esm_piControl) in the potential extensions to Tier 1. Similarly, the Tier 1 protocol asks for the zero-emission runs to be a minimum of 300 years in length. This is sufficient for investigating many Earth system responses, but for some of the slower Earth system components (e.g. ice sheets or the deep ocean) longer runs are required. We therefore encourage groups to extend their zero-emission runs to 500 years or longer, in the potential extensions to the Tier 1 protocol.

There are a number of highly desirable and technically straightforward extensions to the Tier 1 protocol, that we encourage modelling groups to consider running. These include: 1.

Extending the three zero-emission runs shown in Fig. 1 to 500 years each (or as long as possible).

To maximize the number of models contributing to TIPMIP, we also propose a reduced set of Tier 1 experiments. While we encourage groups to run the full set of experiments whenever possible, groups will be able to participate in Tier 1 by realising the reduced set of runs:

An esm_piControl for 300 years.

The TIPMIP ramp-up and zero-emission simulations have similarities with the constant emission, zero-, and negative- emission simulations planned in CMIP7 flat10MIP (Sanderson et al., 2025). A key difference between the two sets of experiments is that the TIPMIP protocol aims to enforce approximate commonality across models in the temporal evolution of GMSAT; i.e. similar warming rates and duration of warming in the positive emission runs before zero-emission runs are branched at common GWLs. The zero and negative emission runs are not guaranteed to exhibit common temperature evolution across each model, these being an emergent property of each model. The protocol accepts the amount of CO2 emitted into each model will necessarily be different to realize this. In flat10MIP, commonality in CO2 emissions is enforced across models, while inter-model variability in the evolution of GMSAT is accepted. While there are similarities between the two sets of experiments, joint analysis of model responses in common-GMSAT versus common-CO2 emission space will help understand differences seen between the two protocols when run by the same model. The flat10 set of experiments allow TCRE to be diagnosed for each ESM under a common emission rate of +10 GtCyr-1, which can be used to inform the positive emission rates in TIPMIP. TIPMIP simulations are more extensive and naturally follow on from flat10. They are designed to explore in more detail model responses, following close-to-uniform warming rates and sampling a number of common global warming levels at which net zero CO2 emission runs are started. The state-dependence of ZEC and the response to negative emissions is a particularly useful and unique analysis opportunity offered by the TIPMIP simulations

Earth system change and regional patterns of change seen in the TIPMIP experiments can also be compared with results from ScenarioMIP (O'Neill et al., 2016; van Vuuren et al., 2026) and CMIP7 DECK (diagnostic, evaluation and characterization of klima) experiments (e.g. 1pctCO2 and abrupt 4×CO2 runs, Dunne et al., 2025) at similar transient global warming levels. This will provide information on the impact of staying long-term at, or close to, a given GWL (in the TIPMIP zero-emission runs) compared to transiently passing through it. Attention will need to be paid to the fact the TIPMIP simulations are forced solely by variable CO2 emissions. ScenarioMIP forcing is a multi-gas mix, including time varying aerosol emissions and human land-use. The latter, while more realistic, is more difficult to achieve commonality of GMSAT behaviour across models, which is a primary aim of the TIPMIP protocol.

A recent multi-model initiative, AERA-MIP (Silvy et al., 2024), provides 1.5 and 2.0 °C stabilization simulations that account for all anthropogenic drivers. This approach utilizes the adaptive emission reduction approach (AERA, Terhaar et al., 2022) to produce temperature stabilization simulations for two GWLs, following realistic historical forcing trajectories.

TIPMIP simulations can be compared with AERA-MIP results to explore, for example, the impact of non-CO2 radiative forcing on the spatial patterns of climate change in the AERA-MIP runs stabilized at GWL=2°C and the TIPMIP net zero-emission runs initialized at GWL=2°C. This analysis will need to pay careful attention to the different experiment set ups in the two runs, particularly the inclusion of non-CO2 forcers in AERA-MIP. A difference between the two protocols is the commonality of model warming rates in the TIPMIP ramp-up runs, as well as the commonality of GWL at which net-zero CO2 emission runs start. We intend to compare tipping point behaviour in common model realizations of the two protocols at common GWLs.

In CMIP6, the Carbon Dioxide Removal Model Intercomparison Project (CDRMIP, Keller et al., 2018) explored the potential for, and impacts of, CDR deployment. In CMIP7, CDRMIP will address questions concerning the response of the Earth system to atmospheric CO2 removal (negative CO2 emissions), the efficacy of different CDR approaches, such as afforestation/reforestation and ocean alkalinization, and begin to assess active CDR as simulated in a number of ESMs, contrasting these with equivalent representations in Integrated Assessment Models (IAMs).

We have dubbed the TIPMIP Tier 1 protocol “see what happens” experiments. By this we mean the runs do not include any external interventions or forcings to ensure a tipping event occurs. There is a widespread feeling that ESMs may be designed (explicitly or unconsciously) to be overly stable with respect to tipping events (Rahmstorf, 2024; Valdes, 2011). This may result in only a small number of tipping events occurring in our Tier 1 ensemble. A natural extension to the TIPMIP protocol, in particular using the zero-emission runs started at GWL=2 and 4 °C, is to deliberately induce a tipping event in these simulations. This has the advantage of knowing, in advance, the climate state at which the tip occurs and automatically having a non-tipping counterfactual run sampling the same climate state as the tipping run. Tips can be induced by prescribing external forcing terms or by making targeted modifications to key model parameterizations to increase the sensitivity of a given system to climate forcing. Such forced tip “make it happen” experiments will be developed in collaboration with the domain-specific activities in TIPMIP, as well as TIPMIP-WHATIF (Winkelmann et al., 2025) to ensure widespread utility of the resulting experiments across projects.

The data request for TIPMIP starts from the data request for the esm-historical experiment in CMIP6. In a first step the number of requested variables was reduced to only include variables that 3 sample ESMs were able to produce, assuming these provide a reasonable subset of CMIP6 models. The list was further reduced by excluding most variables at sub-daily frequencies. TIPMIP experiments cover several centuries, and the amount of data would become impractical if sub-daily data, especially on multiple levels, were saved. The reduced list was then passed to TIPMIP domain experts for inspection, which resulted in some additional variables that were considered important for analysing domain specific tipping points. TIPMIP ESM output will be used as forcing for offline models in TIPMIP. We also anchored our diagnostic list with the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) (Frieler et al., 2024) to ensure data can be used to force sectorial impact models. All diagnostics follow CMIP6plus CMOR (climate model output rewriter) protocols. CMIP7 protocols will be used once they are fully established.

TIPMIP comprises a number of zero-emission (ZE) simulations at specific GWLs, as well as positive CO2-emission (ramp-up) and negative CO2-emission (ramp-down) experiments that link one warming level with another. A ZE experiment branching directly off a ramp-up experiment will therefore be different from another ZE experiment, at the same GWL, branching off a ramp-down experiment itself started from a higher GWL value. To distinguish between these two experiments, we propose a naming convention that allows us to construct experiment names by using short blocks describing each experiment phase and then adding these blocks together to form unique experiment_ids for each possible experiment. The building blocks and example experiment file names are listed below in Tables 2 and 3.

The building blocks are combined to identify the full history of a given experiment, forming part of the output filename for a given model. By using these building blocks, it is possible to distinguish ZE experiments at different GWLs that have started directly off a ramp-up run, from those that have started after a ramp-up, ZE run at a higher GWL, and ramp-down back to the GWL of the ZE run in question, or distinguish between experiments with different length ZE runs before a ramp-down run is started. Table 3 lists the experiment_id and experiment metadata for the Tier 1 experiments.

Tipping of Earth system elements is often considered a low-likelihood, high-risk event that may result in severe consequences for ecosystems, biodiversity, and society. In recent years there has been increasing evidence that several parts of the Earth system are already undergoing rapid, potentially irreversible change (Lenton et al., 2008, 2019), which may lead to crossing of tipping points. Components at risk of tipping include the West Antarctic and Greenland ice sheets, Amazon and some boreal forests, savanna and dryland ecosystems, coral reefs, the Atlantic Meridional Overturning Circulation (AMOC), the North Atlantic Subpolar Gyre (SPG), and ground permafrost. (e.g. Levermann and Winkelmann, 2016; Sgubin et al., 2017; Staal et al., 2020; Swingedouw et al., 2020, 2021; Chap. 9, IPCC, 2023a; Armstrong McKay et al., 2022). Recent studies estimate that widespread, at least partly irreversible, mass loss from the West Antarctic ice sheet may be triggered when global warming levels range from 1–3 °C above pre-industrial (Golledge et al., 2017; Garbe et al., 2020; Reese et al., 2023). For the Greenland ice sheet, a critical threshold is estimated in the range 0.8–3 °C of global warming, with a best estimate of about 1.5 °C (van Breedam et al., 2020; Noël et al., 2021; Höning et al., 2023). The recent Global Tipping Points Reports (Lenton et al., 2023, 2025) conclude that some Earth system tipping events are no longer low-likelihood, high-impact events, rather they are rapidly becoming high-likelihood, high-impact events.

Large uncertainty still exists with respect to tipping point risks. Many studies use a range of observational data including paleo-proxy records, in situ measurements, as well as atmosphere and ocean reanalyses. Satellite data, covering the past few decades, is also playing an increasing role (Swingedouw et al., 2020; Lenton et al., 2024), because of its global coverage at fine temporal and spatial resolution. In situ data is often spatially too coarse, and satellite data temporally too short, to reliably detect early warning signals of tipping events. ESMs offer global coverage at reasonable (but by no means sufficient) spatial resolution and can be integrated for multi-centuries. ESMs can therefore be used to investigate where, when, and how, tipping events are triggered (Romanou et al., 2023; van Westen et al., 2024a, b), as well as how different tipping systems interact. Studies to date have mostly been based on either a single model, and/or individual experiment design, with a focus on individual tipping element (e.g., the North Atlantic Hosing Model Intercomparison Project (NAHosMIP) and AMOC, Jackson et al., 2023).

A key aim of the TIPMIP ESM protocol is to investigate a range of tipping events using a common experiment protocol sampled by multiple ESMs. We aim to investigate the risk of tipping events under net zero emissions initiated at different levels of global warming, and if triggered, whether the ensuing changes are reversible (through global-scale cooling) on societally relevant timescales. Ritchie et al. (2021, 2025) stress both the peak global warming level (GWL), and the duration spent at a given GWL before cooling, are important determinants of the risk of triggering a tipping event. The TIPMIP Tier 1 protocol offers a (minimal) set of multi-model experiments to address such tipping risks (i) under net zero emissions started at two different GWLs, (ii) associated with the duration of net zero emissions, and (iii) the potential for reversibility as warming is reduced.

A number of studies have identified abrupt changes in ESMs (Drijfhout et al., 2015; Swingedouw et al., 2021), including stochastic collapse of the AMOC (Romanou et al., 2023), localized, rapid Amazon loss (Parry et al., 2022), and rapid, irreversible mass loss of the West Antarctic ice shelf (Naughten et al., 2023). In addition, van Westen et al. (2024a) show how a freshwater flux into the Atlantic in the CESM model can induce positive feedback driving a collapse of the AMOC. They further highlight such feedback is poorly represented in present-day models, due to large biases in simulated salinity fields. van Westen et al. (2024b) extend this work to derive a fingerprint for AMOC collapse (based on surface ocean buoyancy), which they apply to an ensemble of CMIP6 projections, finding a greater than 50 % likelihood the AMOC will pass a critical point for tipping during the 21st century following a middle-of-the-road emission scenario. With respect to permafrost, recent studies emphasize the need to incorporate frozen soil thaw dynamics and associated carbon release into models to accurately estimate available carbon budgets (Natali et al., 2021; Turetsky et al., 2020). CMIP6 ESMs are capable of simulating gradual permafrost thaw but lack the necessary resolution and process representation to capture the spatial heterogeneity of permafrost thaw dynamics. Quantifying the impact of missing processes is essential for advancing our ability to estimate future permafrost behaviour. Offline models representing high northern latitude terrestrial processes will be forced by TIPMIP ESM output to help refine estimates of annual permafrost thaw using the methodology of Burke et al. (2020), and improve predictions of potential rapid loss of permafrost volume.

For many of the phenomena at risk of tipping, ESMs exhibit biases in the representation of these phenomena (Li et al., 2021; McCarthy and Caesar, 2023) and in the climate drivers of the phenomena (Mecking et al., 2017; Robson et al., 2022; Jensen et al., 2024), including the representation of extreme events that might push a system beyond its resilience limits (Romanou et al., 2025). An important task is to assess the quality of the TIPMIP ESMs in simulating the phenomena of interest, the key climate controls of these phenomena, including those that control the risk (or not) of a phenomenon tipping. For this we will focus on established metrics and climate controls for each phenomenon, including statistical and physically based metrics (or fingerprints) that are considered robust indicators of a potential tip. In this respect, the development of robust observational constraint approaches (e.g. Portmann et al., 2025) is important, as is the need for a large multi-model ensemble.

Once a tipping event is identified, we will assess: (i) the mechanisms underpinning the event, comparing the key driving processes across the multi-model ensemble (MME); (ii) whether there are early warning signals of the tip and then search for the occurrence of similar indicators across the MME; (iii) the broader consequences of any tip for the rest of the Earth system, considering both remote impacts (Ritchie et al., 2020) and interactions between components of the Earth system (Klose et al., 2024). Systems at risk of tipping have the potential to interact (i.e. if one regional system tips, cascading teleconnections can influence the risk of a tip in another remote system Wunderling et al., 2024). An ESM dynamically couples different Earth system processes. The TIPMIP protocol allows us to investigate potential tipping point interactions across models, supporting quantification of any interactions and possible cascades.

The Paris Agreement aims to limit global warming to below 2 °C (and pursue efforts to limit warming to 1.5 °C). GMSAT has already increased by almost 1.5 °C and cumulative GHG emissions continue to increase (Friedlingstein et al., 2025). It is therefore unclear when, and at what level, temperatures (and associated emissions) might be stabilized in the future. It is also unclear if, how, and how rapidly, global temperatures can be reduced, for example through net negative CO2 emissions. It is possible global temperatures remain significantly above present-day values for a long period before any significant cooling occurs. Furthermore, Gillett et al. (2011), Cassidy et al. (2024), and Chamberlain et al. (2024) show strong regional changes in climate, particularly over the Southern Ocean, following net-zero emissions even when global mean temperatures are approximately stable. Continued Southern Ocean warming, particularly at depth, is associated with a slowing of the overturning circulation at subpolar latitudes, decreasing the export of cold Antarctic Bottom Waters into the global ocean (Chamberlain et al., 2024). The amount of heat released from the Southern Ocean may also affect the dependence of ZEC on the level of (surface) warming at which zero emissions are realized. MacDougall et al. (2020) found in a limited number of ZECMIP models that higher levels of cumulative emissions may lead to greater ZEC values.

It is therefore urgent to understand how the Earth system will respond to net zero emissions at different GWLs (King et al., 2021, 2024; Chamberlain et al., 2024; Silvy et al., 2024). Such analysis requires long-term, zero-emission simulations at a range of GWLs. The TIPMIP protocol provides such a data set.

Slow processes, such as melting of ice sheets or deep ocean warming, will continue to evolve long after net-zero emissions are achieved. Paleo evidence suggests a committed global sea level rise of around two meters per degree of global warming on millennial time scales (Levermann et al., 2013; Pattyn et al., 2018). Continuous freshwater input to the ocean, from melting land ice, will also have long-term impacts on ocean circulation (Swingedouw et al., 2007; Hu et al., 2008). As long as the ocean is still adjusting, feedbacks onto other components of the Earth system, expressed through changes in marine heat and carbon fluxes (DeVries et al., 2017; Rugenstein et al., 2020), have the potential to impact patterns of regional change.

The potential for gradual warming under net-zero emissions may result in system thresholds being exceeded, with impacts on ecosystems and society, or even tipping of elements in these systems, such as abrupt shifts in vegetation (Wei et al., 2025), regime shift in marine primary productivity (Vasilakopoulos et al., 2017), biodiversity loss (Ureta et al., 2022) or permanent displacement of people (Defrance et al., 2017). Furthermore, gradual change in the mean climate can induce shifts in modes of regional climate variability (Horton et al., 2015; Tamarin-Brodsky et al., 2020; Kim and An, 2024), with consequences for extreme events (Fischer et al., 2021; Wehrli et al., 2022). The TIPMIP protocol offers the possibility to study both incremental climate change and associated regional changes in variability in long-term, zero-emission simulations.

While climate change is not yet one of the top drivers of biodiversity loss (Jaureguiberry et al., 2022), it is emerging as a significant threat (Soultan et al., 2022; Kubelka et al., 2022). Recurrent extreme temperature or precipitation events impact the ecology of different disease vectors (Liu-Helmersson et al., 2014) and put stress on agriculture and infrastructure (Sgubin et al., 2019). The TIPMIP ESM experiments will be used to drive impact models in ISIMIP (Frieler et al., 2024) to investigate such risks to ecology, biodiversity, and society.

Most studies to date looking at climate impacts at a given GWL have been based on transient projections, analysing periods where global warming falls in an interval around a given target level (Lennard et al., 2018; Tebaldi et al., 2021; Swaminathan et al., 2022). These studies suffer from the fact that the transient climate at a certain warming level differs from the climate after a longer period at the same level. In many regions, this leads to differences between the transient and the equilibrium climate at the same GWL (King et al., 2020, 2024; Lacroix et al., 2024). The TIPMIP zero-emission simulations offer the opportunity to assess the long-term response of the Earth system to net zero CO2 emissions at different GWLs. Key questions to address include:

What are the long-term regional changes under net-zero emissions and do these changes differ depending on the GWL at which net-zero emissions are realized?

To answer these questions robustly, an ensemble of simulations is required. Single model simulations can differ significantly from each other, examples include Arctic Sea ice (Notz et al., 2020) and AMOC (Weijer et al., 2020), and their subsequent evolution in future projections (e.g. Romanou et al., 2023). Regional climate change patterns are even more challenging as they strongly depend on large-scale circulation patterns and internal variability (Kjellström et al., 2018; Koenigk et al., 2020). Internal variability can be partially addressed by the long zero-emission simulations. However, as the mean state keeps evolving, late and early periods in these simulations may be different. In addition, ESMs simulate different regional and cascading changes across Earth system components at the same GWL (Evin et al., 2024). Dealing with these model-to-model differences calls for an ensemble of models running the same set of experiments. The Tier 1 protocol requests 300-year zero-emission simulations. This may not be sufficiently long to fully answer questions pertaining to the new equilibrated state at different GWLs. It is therefore important to link with activities looking at timescales exceeding a few hundred years (King et al., 2024; Rugenstein et al., 2020).

The IPCC Special Report on the Ocean and Cryosphere in a Changing Climate (SROCC) (IPCC, 2022a) defines reversibility as the ability of the Earth system to reverse anthropogenic-induced changes and return to a state equivalent to that prior to the external perturbation (e.g. a preindustrial condition), on a certain timescale. Irreversibility, in contrast, implies a suite of forcings and system responses that moves the system across some critical threshold, from one stable equilibrium to another (Lenton et al., 2008), with the original state unattainable under reasonable bounds of modified forcing or time periods. Irreversibility or hysteresis is often related to tipping events (van Nes et al., 2016; Armstrong McKay et al., 2022; Lenton et al., 2023), although irreversible change can also occur as a result of incremental warming or climate change.

Current literature suggests irreversible Earth system change may result either from nonlinear responses to an external forcing, or state-dependent responses to this forcing, or a combination of both ( Litzow and Hunsicker, 2016; Scheffer et al., 2015; Santana-Falcon et al., 2023; Heinze et al., 2023; Ritchie et al., 2023; Fröb et al., 2024). The question whether changes are reversible is therefore connected to the response of key processes to variable external forcing (Frölicher and Joos, 2010; Kug et al., 2022), where this forcing might be external to the Earth system (e.g. anthropogenic CO2 emissions or solar variability) or internal to the Earth system but external to a specific phenomenon (e.g. changed ENSO variability and its impact on Amazon resilience). Details of the external forcing matter, for example, changes in either the magnitude or duration of forcing, such as the rate of warming or the magnitude and duration of warming overshoot, are thought to be important factors in driving a system towards irreversible change (Schwinger and Tjiputra, 2018; Jeltsch-Thömmes et al., 2020; Ritchie et al., 2023, 2025).

The TIPMIP experiments allow us to address questions around the potential reversibility of Earth system change. A first question concerns the reversibility of global warming realized in the positive emission (ramp-up) phase. One might expect equal and opposite negative emissions to induce global cooling of a similar rate to the (ramp-up) warming. This is not necessarily the case for the TIPMIP runs, as each model's response during the 50 years of zero emissions will differ, resulting in different GMSAT and atmospheric CO2 values across models at the start of the negative emission run. The spread in these quantities at GWL=4°C may also be greater than at GWL=2°C. Hence, while the linearity of TCRE and the link between cumulative emissions and peak warming suggest TCRE and TCRR are symmetric, for the TIPMIP ensemble this needs to be established. While previous assessments (e.g. IPCC Special Report on Global Warming of 1.5 °C, IPCC, 2022b; IPCC 6th Assessment Report WGI, IPCC, 2023a) provide some insight into how changes differ between warming levels, the question of how reversible the changes are, particularly regional changes, remains uncertain.

Another important question relates to how changes in one location (or component) of the Earth system can cause changes elsewhere. Emerging literature (Armstrong McKay et al., 2022; Lenton et al., 2023; Wunderling et al., 2024) suggest some tipping elements, once triggered, may influence the stability of other parts of the Earth system. The suggestion is that an initial global-scale change can induce a major regional (abrupt) change that itself amplifies the global-scale signal or triggers independent regional abrupt changes elsewhere (Wunderling et al., 2022). Quantifying these risks requires an improved understanding of region-to-region interactions in the Earth system. Recent studies highlight how anthropogenic forcing has the potential to significantly alter prominent modes of climate variability, such as the Inter-Tropical Convergence Zone (Kug et al., 2022) or the El Nino Southern Oscillation (ENSO, Vaittinada Ayar et al., 2022; King et al., 2024; Cassidy et al., 2024), with the potential to induce further, potentially amplifying regional responses in remote Earth system domains.

A third question concerns long-term committed change under net zero emissions at different GWL, such as Greenland or Antarctic ice loss, or permafrost thaw, and its potential reversibility when negative emissions (and cooling) are initiated. Recent work (Schleussner et al., 2024) suggests reducing global temperatures can limit long-term climate risks compared with merely stabilizing warming (i.e. negative emissions compared to net-zero emissions). However, there are deep uncertainties in the long-term response of the Earth system, associated with the severity and duration of warming before cooling is realized, as well as how accurately models represent the long-term features of the coupled Earth system. Recent advances in Earth system models participating in TIPMIP will help address this question. For example, ESMs now include interactive treatment of phenomena such as continental ice sheets, permafrost, wildfires, and are increasingly running in emission mode for GHGs, allowing a more complete representation of interactions between physical climate change and the carbon cycle (Sanderson et al., 2025).

The magnitude of global warming scales approximately linearly with cumulative CO2 emissions (IPCC, 2023b). This underpins the expectation that global mean temperatures should approximately stabilize under net-zero CO2 emissions, and that reducing CO2-induced warming requires net negative CO2 emissions. The TIPMIP ESM simulations will also allow us to explore the efficacy and implications of reversing global temperature through net negative emissions.

The linearity of warming arises from the climate response to forcing – both the transient response and the rate of adjustment to constant forcing – as well as the radiative forcing from CO2 emissions (Allen et al., 2022), the latter controlled by the airborne fraction of emitted CO2. The resulting relationship – known as the transient climate response to cumulative carbon emissions (TCRE) – gives a monotonic and approximately path-independent peak-warming response to CO2 emissions. While some studies have shown this relationship may break down for both very large cumulative emissions and strong negative emissions (e.g. Zickfeld and Herrington, 2015; Zickfield et al., 2021), these results are based on single model studies. It remains unclear how symmetrical TCRE is (i.e. the extent to which TCRE under net negative emissions, also referred to as TCRR (Chimuka and Zickfeld, 2026), is equal and opposite to TCRE). Koven et al. (2022) show that deviations from reversibility of TCRE are related to the legacy response to previous emissions and correlate with the ZEC for each model – models with a positive ZEC show an overshoot of temperature under negative emissions while models with a negative ZEC may undershoot – i.e. they may reverse more quickly than the negative emissions alone would imply. Models with a ZEC close to zero are much closer to reversible along the same TCRE gradient. Models with a higher ZEC, and importantly if ZEC becomes increasingly positive at higher warming levels, may exhibit reduced sensitivity to negative emissions – i.e. they may cool slower than they warmed in the ramp-up phase. Extending the CMIP7 flat10MIP experiments (Sanderson et al., 2025), TIPMIP samples multiple GWLs, allowing a systematic analysis of multi-GWL responses to positive, zero, and net negative, emissions.

Carbon dynamics under negative emissions are qualitatively understood, although models differ in the quantitative magnitude of their response. Over the last century, CO2 emissions have caused a rapid rise in atmospheric CO2 concentration and natural carbon sinks – both land and ocean – while removing some of this, have not kept up and so increase their sink rates in step with the emissions (Raupach et al., 2014). When CO2 emissions stop increasing, atmospheric CO2 declines and natural sinks weaken, eventually saturating, in particular the land biosphere (Silvy et al., 2024). Under negative CO2 emissions, atmospheric CO2 concentration strongly declines, and the natural sinks eventually reverse as both ocean and land ecosystems outgas excess CO2 (Tokarska and Zickfeld, 2015; Jones et al., 2016; Koven et al., 2022). Thus, in the long term, the same processes which lead to natural sinks opposing positive emissions also lead to the natural carbon cycle opposing CO2 removal, i.e. for every ton of CO2 removed (e.g. by CDR), atmospheric CO2 concentrations will decline by less than one ton. Uncertainty in the carbon cycle response to negative emissions is likely dominated by land ecosystems in the same way as it is for the response to rising emissions (Jones and Friedlingstein, 2020), although on longer timescales the land sink will saturate more quickly leaving a larger role for ocean uptake (Randerson et al., 2015). The TIPMIP ESM experiments will enable quantification of the various contributions to the total carbon cycle response to negative emissions simulated in the latest generation of ESMs.