User Tools

Site Tools


Last Interglacial CMIP6/PMIP6 proposed experiments

The PMIP4 experimental design available on the PMIP3 wiki has been superseded by the experimental design detailed on the PMIP4 web site


The Summary for Policymakers for both the IPCC WG1 AR4 and AR5 included statements on the Last Interglacial (LIG):

AR5: There is very high confidence that maximum global mean sea level during the last interglacial period (129,000 to 116,000 years ago) was, for several thousand years, at least 5 m higher than present, and high confidence that it did not exceed 10 m above present. During the last interglacial period, the Greenland ice sheet very likely contributed between 1.4 and 4.3 m to the higher global mean sea level, implying with medium confidence an additional contribution from the Antarctic ice sheet. This change in sea level occurred in the context of different orbital forcing and with high-latitude surface temperature, averaged over several thousand years, at least 2°C warmer than present (high confidence).

Yet the AR4 and AR5 had no coordinated simulations for the LIG to assess the interplay of polar amplification of temperature, seasonal memory of sea ice, and precipitation/storm track changes on the stability of the Greenland ice sheet and its contribution to the sea level high stand nor the interplay of oceanic and atmospheric temperatures and circulation on the stability of the Antarctic ice sheet. Climate model simulations for the LIG assessed in the AR5, although completed by many modeling groups, varied in their forcings and often were not made with the same model/same resolution as the CMIP5 future projections, thus providing a useful but incomplete means for assessment (IPCC AR5 WGI Chapter 5; Lunt et al., Climate of the Past, 2013). Similarly, Greenland ice sheet simulations assessed in the AR5 used offline models with a variety of climate forcing setups, not then allowing feedbacks among the Earth system components (Chapter 5). No simulations were available to assess the Antarctic ice sheet (particularly, the West Antarctic Ice Sheet) contribution to the LIG sea level high stand.

We propose two CMIP6 experiments for the LIG to determine the interplay of warmer atmospheric and oceanic temperatures, changed precipitation, and changed surface energy balance on ice sheet thermodynamics and dynamics during this period. Still uncertain are how well ice sheet-climate models can predict the stability of the ice sheets and if thresholds may be passed this century. A LIG simulation will be of high societal relevance because of implications for sea level changes as well as sea ice and monsoons. The LIG simulations will also provide an‘out-of-sample’ evaluation of new features of CMIP6 models: coupled climate-ice sheet models. The LIG is the most suitable of the warm interglacials for a CMIP6 assessment because of the wealth of data including: ice cores providing measurements of well-mixed greenhouse gases, aerosols including dust and sea salt, and stable water isotopes as a proxy for temperature, as well as for Greenland, ice sheet elevation and extent; marine records for ocean temperatures and geotracers that can be interpreted in terms of water masses and overturning strength; speleothems that provide indication of monsoon strength; and terrestrial records that indicate temperature and vegetation. As well, new records are refining our knowledge of sea ice extent, fire, and biodiversity.

The proposed CMIP6 simulations for the LIG are particularly relevant to the WCRP Grand Challenges: Changes in Cryosphere and Regional Sea-level Rise, but also to Regional Climate Information and Clouds, Circulation and Climate Sensitivity because of the large forcings and thus large regional responses as recorded in the data. It addresses well the broad scientific questions: 1. How does the Earth System respond to forcing? and 2. What are the origins and consequences of systematic model biases (especially at high latitudes and relevant to the stability of the ice sheets)? As part of PMIP, some groups will additionally perform transient coupled ice sheet-climate simulations that will provide rates of change for sea level, including regional sea level if offline GIA models applied, as well as a measure of the capability of these models to initiate the next glacial inception.

The CMIP6 experiments will evaluate systematic biases and their origins among models on their ability to simulate Arctic warmth and sensitivity of Greenland ice sheet to this warmth, and ocean warming and transmission of subsurface warming from North Atlantic to Southern Ocean, with implications for basal melting of West Antarctic Ice Sheet.

Design of LIG Experiments


  • Evaluating climate models for warm period, high sea-level stand.
  • Impacts of smaller ice-sheets/higher sea-level on climate
  • Comparison to climate responses (monsoons, temperature, sea ice, etc) to mid-Holocene simulation with smaller orbital forcing.


128ka simulations - large orbital forcing, large responses.

  • Orbital parameters set to 128ka (Figure 1).
  • Greenhouse gas concentrations well-known from ice cores (Figure 2), [CO2 275ppm; CH4 709ppb; N2O 266 ppb].
  • Experiment 1 (Tier 1): AOGCM; modern geography, ice sheets, and vegetation; length: ≥ 100 years after spinup. Initialize from CMIP6 preindustrial DECK simulation. Save appropriate output to allow forcing of offline ice-sheet models.
  • Experiment 2 (Tier 2; collaborative with ISMIP6): Coupled AOGCM-ISM; predictive Greenland ice sheet models, asynchronous coupling with 10 ISM years per 1 AOGCM years; length: AOGCM 300 years, ISM 3000 years. Initialize from Experiment 1.

Click to get a larger version


  • 125ka, 122ka time slices.
  • Plus transient simulation: 128-118ka (possibly should be started earlier, then would need one more snapshot to initialize, ice sheet and meltwater reconstructions for remnant North American and Eurasian ice sheets).

Extra notes

  • Relevant topics: climate, ice sheets and sea level, sea ice, monsoons, carbon cycle.
  • LIG CMIP6 experiments, collaborative between PMIP6 and ISMIP6.
  • The experiments should be well established for Tier 1 and at least tested for Tier 2 and beyond.
  • A sufficient number of modelling centers (~8) are a part of the MIP and committed to simulating all of the MIP‘s Tier 1 experiments for a particular science question.

[ PMIP3 Wiki Home ] - [ Help! ] - [ Wiki syntax ]


Jean-Yves Peterschmitt, 2014/12/17 16:05

Use this Discussion section to submit your comments

Gavin Schmidt, 2014/12/29 22:52, 2014/12/29 22:56
  1. Exp 1 does not allow for any particular useful comparisons to observed data, and so does not provide a good evaluation of model responses. Specifically, using a modern ice sheet (GR and WAIS) is a waste of time. I strongly suggest that the group use any pre-calculated LIG ice sheet as a first cut, and allow (and encourage) different variations as and when they become available through either Exp 2 or other efforts. These can be seen as forcing variations, and not new experiments.
  2. Having extra diagnostics just for this is not sensible. Instead, lobby and make sure that sufficient diagnostics are included within the basic CMIP6 specification. This is not the only experiment where people might want to run off-line ISMs.
  3. Specifying coupling strategies is too prescriptive for exp. 2. Absent any good scientific rationale, groups should do what they can here.
Anders Carlson, 2014/12/30 17:54

To add on to Gavin's comments, the likely minimum in the Greenland ice sheet was ~120 ka so the CMIP exp. 1 and 2 really won't catch that interval. The PMIP approach would. It is likely that the GIS was ~similar to its present extent (at the GCM scale) around 128 ka and so modern topography could be okay.

Bette Otto-Bliesner, 2015/01/19 16:04

When did the Greenland ice sheet start retreating? When did GMSL get above present?

Eric Wolff, 2015/01/07 15:16

I already made this comment offline to some of you, but Bette encouraged me to add this to the wiki. It adds to what Gavin has raised. For the CMIP 128 ka experiment, even with correct ice sheets, there is a problem comparing to the data. For sure, we can (for example using the data syntheses in Capron et al 2014) create a 128 ka timeslice. However, remember that 128 ka is very close to the time when things change rapidly: the fast methane jump in TII centres at 128.6 on AICC2012 (very close to the fast speleo jump in Cheng et al (2009)). The implication is that northern hemisphere temperature and tropical hydrology jumped rapidly somewhere around that point. The problem is that it remains impossible to synchronise records precisely enough at this period, so we would regard any time slice as being at best a 2ka wide window. That means we run the danger that, if we create a 128 ka slice, some records are from before the rapid jump and some are after, making a highly inhomogeneous data target: much more so than if the timeslice was firmly before or after 128 ka.

Of course this also highlights the question of why there was a rapid change in NH climate around this time, which many of us think is likely to be due to a bipolar seesaw response. That then raises the question of what AMOC will be doing in your model runs. Presumably the default would be as Holocene, in which case a data slice firmly after the rapid climate changes would be the more appropriate data target.

I'm afraid there isn't an easy answer to this. 128 ka will give the strongest astronomical/GHG forcing, but you'll definitely have a cleaner data target if you compare to something a little later (or earlier but then you may need to worry about doing something to force AMOC).

Anders Carlson, 2015/01/16 02:34, 2015/01/19 13:37

Eric's comment is a very poignant deal for proxy comparison. This whole LIG kinda brings up the idea that with a bit more forcing, things get 'weird'. I would suggest contacting Jeremy Hoffman Oregon State ( who has a full paleoclimate database through this interval to say when is the best time to assess pure orbital effects (this would include all uncertainties).

Like Eric says, the 128 ka would be like using a 9 or 8 ka simulation for the Holocene, or not but that timing is undatable.

Allegra, 2015/01/16 14:14

I think that experiment two is the pathway forward. Since, a good chunk of the polar amplification will come from the changes in topography, I'd really like to include these.

For experiment one, couldn't we just do more Holocene time slices… and maybe then single forcing Holocene instead? (Better proxy coverage.) Although, I feel like another mid-Holocene experiment will be very marginal in the amount of extra info we can get that we didn't already understand from the previous PMIPs.

 Eric -- what period of time during the Eemian do you find to be the most promising.

To do experiment two well, we need strong input from the ISMIP folks. I emailed Sophie Nowicki to see if she would chime in here.

One thing I remember from their meeting last summer is (as usual), everyone wants/needs higher, and higher resolutions from the GCMs. Have we thought about how we will do the down-scaling?

Lev Tarasov, 2015/01/17 00:50

Not sure what you expect to get out of an accelerated setup such as experiment 2 (or are you also accelerating orbital forcing, GHGs,..)? Results would also depend on the spinup/initialization of the ice sheet model.

I second that climate downscaling to ISM scale needs attention.

For a devil's advocate” instead of focussing on transient response during a period with little paleo-data coverage, why not transient response during H1 or YD onset or 8.2 ka for CMIP AR6? Sure the Eemian has little ice, but 128 ka ice and ocean state will have a strong imprint from a poorly constrained penultimate deglaciation.

Bette Otto-Bliesner, 2015/01/19 16:01

The rationale for experiment 2 is to simulate the coupled responses of the Greenland ice sheet and climate for a time period with strong insolation forcing at NH high latitudes.

A start date of 128ka is chosen as it captures the strong late spring-summer insolation anomalies and is after the overshoot in Antarctic temperatures and CO2. Based on Eric's comments, possibly this experiment should start a bit later, 127ka?

This is not a true transient experiment, but rather a simulation to investigate the 3000-year response of the Greenland ice sheet to large insolation anomalies. Some of the CMIP models may not have the capability to change the orbital parameters during a run. 128ka insolation anomalies at NH high latitudes are a reasonable approximation to those from 128-125ka and GHG are relatively stable also from 128-125ka.

There is now the capability at many of the modeling centers to run simulations with their AOGCM at a relatively coarse resolution coupled to their Greenland ice sheet model at much finer resolution. These models can also accelerate the ice sheet model relative to the AOGCM. This is important as an AOGCM run of 3000 years with the next generation of CMIP models will not be generally feasible.

Jeremy Hoffman, 2015/01/21 01:42

As Eric noted above, a first order comparison between model simulation/data around Greenland can now be done with data presented in Capron et al. (2014) but do note that their relevant time slices are centered on 130- and 125- ka, with little explicit information about the total uncertainty for time slop in their estimates. This suggests that comparison of a 126- or 127-ka model slice may be just as valid with respect to the data realm, if we suspect a modest uncertainty around +/- 2ka for the compilation. However, the Capron et al. tuning scheme uses the final TII methane jump (~128.6 ka) as one of its characteristic tie-points for developing the age models of N. Atlantic marine records (assigned a relative uncertainty of +/- of 0.5 ka), which would actually reduce the uncertainty of what N. Atl SSTs were doing at/around that time, as long as we agree that SST-based stratigraphies are well-founded. As Anders/Eric/Lev noted full age model uncertainty does need to be addressed by the data folks before truly meaningful comparisons can be made, but I feel that a comparison of the ~transient simulations (exp 2) to the data compilations would/could be useful and at least improve uncertainties for AR6, if the other issues (downscaling et al.) are sufficiently addressed.

Pepijn Bakker, 2015/01/28 16:39

In line with what Bette noted, PMIP3 Last Interglacial equilibrium and transient experiments show that it does not make too much of a difference whether one picks 128ka or 125ka, as long as you're after the early CO2 maximum. Other differences like fixed/dynamic vegetation and the prescribed changes in ice sheet configuration and freshwater fluxes are of larger impact. Perhaps an argument in favor of 125ka is that many more groups performed a 125ka simulation as part of PMIP3 (~10) while only ~3 groups did a 128ka experiment. As such, the proposed experiment 1 could be a continuation of the PMIP3 125ka experiment.

Andrea Dutton, 2015/02/03 15:15

I would like to add another twist into the rapid transition that is occurring at ~128.5 ka as others have mentioned above. There is evidence to suggest that sea level rose extremely rapidly at this time, surpassing present eustatic sea level and possibly reaching as high as 6 m above present at this time (according the the Seychelles data) or at least a few meters above present (according to Western Australia data). Either way, this early sea-level rise was most likely sourced from Antarctica if Greenland mass loss occurred gradually and peaked much later (see Anders' comment above). This is to say that in my mind there is a distinct possibility that some sector of the AIS collapsed – coincident with the onset of the sea level highstand. If the models start at some ice sheet configuration similar to present, then they may miss this scenario entirely. Is there a possibility to, for example, set it up such that the WAIS has collapses at 128.5 ka or start with no WAIS, or something that captures this possibility? Surely this will have a significant impact on how things play out from there.

Eric Wolff, 2015/02/04 12:50

I apologise for starting what seems to have been a blizzard, and I stood back a bit because I think the modelling folks need to decide what the experiment is for before they can decide what it is: I see a few motivations in the messages here and you can't study them all with 2 experiments.

One practical issue is to make sure you are aware that there is a new, more comprehensive compilation of CO2 data now available. (Bereiter, B., S. Eggleston, J. Schmitt, C. Nehrbass-Ahles, T. F. Stocker, H. Fischer, S. Kipfstuhl, and J. Chappellaz (2014), Revision of the EPICA Dome C CO2 record from 800 to 600 kyr before present, Geophys. Res. Lett., 2014GL061957, doi: 10.1002/2014GL061957.) While the main part of the paper is a small correction to values older than 600 ka, they also provide a file that includes and compiles all measured data for the LIG. I don't think it changes much for you (I would read 276 at 128 ka) but you should be aware especially if anyone plans transients, as the new data are much less noisy than the old Vostok data.

When I read the 3 purposes near the top of this page (just above where it says CMIP6), it seems to me that purpose 1 and 3 require roughly modern sea level, ice sheets and maximum insolation, and that this would be the point of expt 1. So why not do as you planned with 128 ka insolation and GHG, and modern sea level - that will give the strongest forcing for intermodel comparisons. BUT do data comparisons with a slightly later time (like 125 ka) when you expect that AMOC and sea level are at least close to modern, with no data straddling regimes. I am sure this kind of asynchronous data-model comparison hurts, but I don't see much alternative unless you simply want to use the weaker insolation forcing at 125 ka anyway.


Dan Lunt, 2015/02/18 10:04

Regarding the snapshot LIG experimental design, I don't think there is any merit in prescribing ice sheets which are different to modern. This is because: (a) we don't know what the LIG ice sheets looked like - we have some idea of total sea level change, but within error bars of at least +- 2 metres - and we have very little idea of how this sea level change was distributed between Greenland and Antarctica. (b) the impact of a smaller Greenland ice sheet on global climate is very limited - there is a strong surface air-temperature effect locally, due to lapse-rate and albedo, but outside of the ice sheet itself, the signal is quite small (see e.g. Toniazzo et al, Lunt et al, Ridley et al). Granted - removing parts of West Antarctica may have a larger non-local effect, but i woudl argue that given the uncertainties this shoudl be a subject of sensitivity studies, rather than in the core simulation. © given (a) and (b), and that changing the ice sheets may be a barrier to participation for some groups, I would conclude that we should keep modern ice sheets.

Lev Tarasov, 2015/02/18 14:18

Responding to Dan's point (and echoing some of Andrea's earlier comments), this will depend on what timeslice is chosen. If we go for Eemian minimum extent, then I would disagree. Subject to the unaccounted for uncertainties in current (eg Koop et al.) estimates for minimum Eemian sea-level, I see the D. Pollard et al. EarthandPlanetaryScienceLetters412(2015)112–121 results in their Figure 5a providing a maximum ice estimate for minimum Eemian Antarctic state (ie WAIS collapse). I would suggest all in this discussion compare that figure to their PD configuration in figure 3a and judge for themselves. A key issue in my mind is potential impacts on regional deep ocean convection. This does open up the changing land mask can of worms, but it's about time that GCMs are able to handle this.

Enter your comment. Wiki syntax is allowed:
pmip3/wg/qi/bc/lidisc.txt · Last modified: 2016/06/20 15:35 by jypeter