The Coupled Model Intercomparison Project (CMIP) and Other MIPs

As you will have seen from our pages on climate model fundamentals and discussion of emissions scenarios, the process of creating projections of future climate is quite complicated! One difficulty that researchers have encountered is that when you create many projections with many different climate models, you get slightly different answers: although models agree on overall trends (such as the increase in global temperature due to human-caused greenhouse gas emissions), they often disagree on the details (such as exactly how much precipitation may change in a given portion of, say, the western United States).

So what does this mean for our ability to create and interpret consistent future projections, that we can use to plan for the impacts of climate change? Well, we have to be able to understand why the models are telling us slightly different things!

The Problem:

There can be many reasons why one might expect slightly different answers when running multiple simulations with climate models.

The major reasons for this are:

• Structural Differences: Models will often represent physical processes in the climate system slightly differently, which can have varying effects on the overall climate.
• Scenario Differences: If two simulations are run with different assumptions about future greenhouse gas emissions or other aspects of human activity, then they will give very different pictures of future climate change! 
• Internal Variability: There is a lot of random noise in the climate system (think “butterfly effect”). That means that even using the same model, and the same scenario of climate change, changing the initial conditions at the start of two simulations can cause them to differ.

(Part Of) The Solution: The Coupled Model Intercomparison Project (CMIP)

Researchers have been aware of the difficulties with interpreting differences among model simulations for quite some time. As a result, the Coupled Model Intercomparison Project, or CMIP, was created in 1995. The goal of CMIP is to provide protocols, or sets of instructions on how to set up a model simulation, that multiple modeling centers [link to modeling center page] can use to ensure that everyone is running their models the same way. That way, when you’re looking at a given set of simulations, you can be sure that you’ve ruled out scenario differences as a cause of disagreement among them! (You still have to worry about structural differences and internal variability – more on that later.)

The details of the CMIP protocol have changed over time – there have been five different phases of CMIP over the course of its history, and we are now on CMIP6. (Yes, I know, 5 and 6 are different numbers… they skipped CMIP4 in order to align the numbering of CMIP phases with the IPCC reports.) But in every phase of CMIP, the idea has been the same: to agree on a set of model experiments that every participating modeling center will perform, the output of which will then be placed on a central server and provided for use by the international climate research community.

The Gist: CMIP is a coordinated set of experiments that different climate models around the world run the same way. In order to participate in CMIP, modeling centers must all complete a minimum required set of these experiments!

Here the most recent two generations of CMIP experiments are described, since those are the output which is most commonly used by climate and environmental scientists right now: these are CMIP5 and CMIP6. 

Phases of CMIP: CMIP5

The CMIP5 simulations were performed in (roughly) 2010-2012, in preparation for the Fifth Assessment Report of the IPCC. They were configured such that there was a “core” set of simulations (the ones that all modeling centers were required to perform), and a recommended set of simulations (“Tier 1” and “Tier 2”) that would be nice to have but not absolutely required. CMIP5 also identified two main science focus areas for their simulations:

• “Near-term” decadal experiments, looking at predictability of climate 
• “Long-term” experiments focusing on the full 20th and 21st centuries

For our purposes here, the “long-term” experiments are more relevant, so we’ll focus on those. More detailed descriptions on these and all the other simulations are available here:

Taylor et al. (2009): A Summary of the CMIP5 Experimental Design

Taylor et al. (2012): An Overview of CMIP5 and the Experimental Design

The long-term experiments can be summarized in the below diagram, with concentric circles showing them in order from highest priority (center) to lowest (outer circle):

What’s in the Core?

There are a few different types of simulations considered “core” to CMIP5. The major types are:

1. A historical Atmospheric Model Intercomparison Project (AMIP) simulation.
   
   Basically, this is a simulation where you impose historical (observed) changes in external forcing (greenhouse gas emissions, air pollution, land-use changes, etc) – but ONLY to the atmosphere. In an AMIP simulation, the sea surface temperature is *not* taken from an ocean model; instead, you just use SST information from observations. This allows you to isolate any differences across simulations to differences in the atmospheric physics, rather than changes in either the ocean below or the way in which the atmosphere or the ocean communicate with one another.
   
2. A pre-industrial control simulation.
   
   This simulation is run in the fully coupled configuration, where all of the sub-models for the different components of the Earth system are turned on and talking to one another. However, what makes it a control simulation is that none of the external forcing factors are changing – instead, they are fixed at values that approximate their values before human activity became a big deal, or ‘pre-industrial’. By keeping all of the external influences on climate constant, and allowing the model to evolve freely, you can determine the equilibrium state of the model, which is likely to be similar but not identical to the average climate of the real Earth before human activities began to matter.
   (For more on control simulations and model bias, see “Climate Model Overview”).
   
3. A historical simulation.
   
   The historical simulation is a coupled simulation run over the “historical” period; for CMIP5, the standard period was 1850-2005. This experiment is fully coupled including all model components, and is forced with a standardized set of observational estimates for all the influences of human activity: namely, greenhouse gas emissions, emissions of other anthropogenic gases (air pollution, smog, etc), and changes in land-use. The historical simulation is generally used as the ‘baseline’ estimate for comparisons with future projections, to assess the importance of future climate change on any given quantity of interest.
   
4. Future projection simulations run with RCP4.5 and RCP8.5.
   
   The future projections are some of the most commonly used CMIP simulations – these are coupled simulations which are initialized from the end points of the historical simulations, and run through the end of the 21st century. For CMIP5, the standard time period for future projections was 2006-2100.
   The forcings for future projections are more complicated than for the historical runs, since obviously we don’t know exactly what human activities are going to be like in the future. To get a sense for the range of possibilities, scenarios for the important anthropogenic influences were constructed; for more on how that was done, see “Emissions Scenarios”.
   
   In CMIP5, the scenario families that were used are the Representative Concentration Pathways (RCPs) – the number after RCP refers to the top-of-atmosphere radiative imbalance, or net amount that humans are warming the planet (bigger numbers = more climate change). There are four main RCPs used in CMIP5, but RCP4.5 and RCP8.5 were chosen to represent the core since RCP8.5 was thought to be the most “business as usual”-like pathway, while RCP4.5 represented a more optimistic but potentially achievable target pathway.
   
5. A coupled simulation forced by an abrupt quadrupling of CO2.
   
   This is exactly what it sounds like – starting from the control simulation, the concentration of atmospheric CO2 is increased suddenly by a factor of 4.
   
   Why 4x CO2? Because that roughly approximates the difference between the pre-industrial concentration of CO2 (about 250 ppm) and what is currently thought to be a “worst case scenario” for the end-of-century values (about 1000 ppm).
   
   Why an abrupt increase? Because the climate system can take a while to respond to CO2 increases, and that transient response can depend on many different aspects of the system, and be affected by differences among models. By using a very simple (and fast) increase in CO2, any differences in model adjustment seen in these simulations have to be due to real physical differences between the models.
   
6. A coupled simulation forced by a 1 percent/year increase in CO2.
   
   This is another idealized CO2 increase simulation, run with the fully-coupled model. However, rather than a single abrupt increase, CO2 is increased continuously throughout the course of this simulation, by a constant fraction (1%/year).
   
   Why 1%/year? Because this roughly corresponds to the magnitude of increase that has been taking place due to human activities. By keeping the rate of increase constant, we can also avoid any complications due to changing rates of warming that might arise in the more complex RCP scenarios.
   
7. Emissions-driven simulations.
   
   In all the simulations described above, when greenhouse gases and aerosol emissions are varied, technically the way this is done is by directly changing the concentration of those gases in the atmosphere as a function of time – the simulations are therefore considered to be “concentration-driven”. Of course, this is not how things work in the real world: in reality, CO2 in the atmosphere is increasing because those gases are being emitted by power plants and other fossil fuel-burning activities. But most CMIP5 models did not include an active carbon cycle, so it was not possible in those models to simulate the process of emitting greenhouse gases, then having them be absorbed by the ocean or land surface, and so on.
   
   I said *most* CMIP5 models, since at this point some models were starting to include these capabilities: the “emissions-driven” simulations refer to using those models which *could* represent the movement of carbon in the Earth system, to conduct experiments where the emissions, rather than the concentration, of greenhouse gases was imposed. This allows for models’ individual representations of land and ocean processes to adjust the amount of carbon taken up from the atmosphere – so the concentration of CO2 is no longer specified exactly, but the total amount added to the atmosphere is instead. 

What about the other tiers?

Since there are many different simulations also recommended by CMIP5, we aren’t going to detail absolutely all of them here; however, some of the main types of Tier 1 and Tier 2 experiments are:

• RCP Extensions
  
  Many aspects of climate change impacts take a long time to be felt: this is particularly true for processes involving the ocean and the melting of sea ice. Some modeling centers chose to run “RCP Extensions”, where the RCP simulations are continued out past 2100 to 2300 without additional changes to forcing. This allows one to get a sense of the amount of “committed” climate change built into the system.
  
• Additional RCPs
  
  RCP4.5 and RCP8.5 are not the only possible future scenarios; other RCPs designed for use in CMIP6 are RCP2.6 (the most optimistic mitigation scenario) and RCP6 (an intermediate scenario). Fewer modeling centers ran these simulations, but you’ll see quite a few available if you search on the CMIP5 website.
  
• Mid-Holocene and Last Glacial Maximum
  
  CMIP5 considered selected past time periods as part of Tier 1. The two periods considered here are the mid-Holocene (roughly 4,000 to 6,000 years ago) and the Last Glacial Maximum (roughly 21,000 years ago).
• Why these two periods? During the mid-Holocene, the climate of Earth was similar to today, but the seasonal distribution of sunlight was different due to changes in the Earth’s orbit. In the Last Glacial Maximum, or LGM, things were much colder than today – much of North America was covered in glaciers, and sea level was significantly lower since so much water was stored as ice on land. This makes these two time periods interesting comparisons with present-day climate!
  
• Last Millennium
  
  A slightly lower priority for CMIP5 was simulating the climate of the entire last millennium. These simulations typically started in the year 850AD, and ran through the end of the CMIP5 historical period (2005).
  
  Why the last millennium? Compared with the mid-Holocene or LGM, the climate of the last millennium is much more similar to present-day (at least, it would be in the absence of human activities). But doing these longer simulations tells us more about the slower (lower-frequency) variations in climate that aren’t well sampled by just simulating the last 150 years. The effects of volcanic eruptions are also extremely noticeable when looking at the last millennium!
  For more on simulations of past climate (paleoclimate), see the discussion of the Paleoclimate Model Intercomparison Project (PMIP) below in the MIP section [link to sub-MIP].
  
  
• Individual forcing
  
  Identifying the effects of climate change is a complicated problem – and it’s important to remember that “climate change” doesn’t just mean greenhouse gas emissions. The effects of human-caused air pollution emissions (‘aerosols’) and changes in patterns of land use are also extremely important, and can interact with the effects of GHG increases. In CMIP5, some modeling centers tried to address this by creating simulations which did not include all human influences at once: instead, they used only one at a time, or sometimes a selected combination of things. 

Phases of CMIP: CMIP6

The CMIP6 simulations were performed in (roughly) 2018-2021, in preparation for the Sixth Assessment Report of the IPCC. By this point, there was a large and growing community of researchers using output from the CMIP simulations, and people were starting to do a LOT of different things with the results. People had also realized that the experimental design of CMIP doesn’t always let you figure out why the models differ from each other very easily (more on that below, when we get to sub-MIPs). For that reason, the eventual setup of CMIP6 was a bit different from CMIP5 – although many things did remain quite similar. 

In order to participate in CMIP6, modeling centers were required to submit a standard set of simulations, same as for CMIP5. These are collectively named the DECK, which stands for “Diagnostic, Evaluation, and Characterization of Klima” (Klima = German for climate). BUT the new thing about CMIP6 was that the DECK is pretty small: in other words, to participate in CMIP6, you don’t have to run too many simulations. Of course, many modeling centers did do more complicated things – those were just collected in smaller projects dedicated to those topics (again, we’ll get to that in a second).

If you want the full technical description of how CMIP6 was set up, it’s available here:

Eyring et al. (2016): Overview of the CMIP6 Experimental Design

What’s in the DECK?

There are four basic types of simulations in the CMIP6 DECK, all of which also appeared in the CMIP5 core:

1. A historical Atmospheric Model Intercomparison Project (AMIP) simulation (amip).
   
2. A coupled pre-industrial control simulation (piControl).
   
3. A coupled simulation forced by an abrupt quadrupling of CO2 (abrupt-4xCO2)
   
4. A coupled simulation forced by a 1 percent/year increase in CO2 (1pctCO2)

Note that none of those simulations contain estimates of any kind of “realistic” future human activities! That is all left to another sub-project, called “ScenarioMIP”. More details on ScenarioMIP can be found in the section on MIPs below, and also on the Emissions Scenario page.

Why is there so little in the DECK? The thought process was that as CMIP continues, people are going to think of more and more different types of experiments they’d like to run – and the models are going to get more and more complicated. That means that it might become hard to compare simulations across different generations of CMIP. 

Sub-MIPs, or CMIP-Endorsed MIPs

Although CMIP is a big advance over previous uncoordinated sets of model experiments, when you think about it more closely, it quickly becomes obvious that CMIP itself doesn’t let you explain everything. For instance, if you’re trying to determine why two simulations of, say, the historical period run with two different climate models are giving different answers, there could be many reasons! Some common things people wonder about:

• Differences in the representation of atmospheric processes (mainly: clouds)
• Differences in the physics of ocean circulation (think: small-scale ocean eddies! Waves! Effects of sea ice!)
• Differences in the way the land surface is represented (types of plants, how they interact with water/radiation/clouds etc)

As researchers started thinking about these issues more closely, they began to realize that in addition to CMIP, there is a need for smaller (“targeted”) comparisons between climate models, where you isolate only one portion of the model to figure out how much that contributes to differences in the answer you get when looking at the fully-coupled climate system. As you may imagine, there are a LOT of different things you can do with this approach!

The Gist: Like CMIP, Sub-MIPs are coordinated sets of experiments with different models – but they are set up in a simplified way, to isolate a particular aspect of things happening within the climate system.

There are dozens of MIPs which have been “endorsed” by the World Climate Research Programme (WCRP), the governing body for CMIP, as of CMIP6. Some of the more popular ones include:

• ScenarioMIP: the Scenario Model Intercomparison Project
  
  This is the big one for CMIP6, since as mentioned above, it’s where all of the future emissions scenario projections are organized. Dividing this into its own MIP makes it easier for people working on developing new scenarios to organize their experiments, and to identify multiple pathways that can get us to the same global warming target in different ways!
  
  The scenario families used to generate most of the future climate projections for CMIP6 are similar to the RCPs from CMIP5, but they have a new name: the Shared Socioeconomic Pathways, or SSPs. As for the RCPs, part of their names refer to the amount of radiative imbalance (= warming) experienced by the end of the century. The SSPs are also organized into “families” representing different visions of future socioeconomic development, which have their own numbers: for instance, the highest emissions pathway is SSP5, which has a top-of-atmosphere radiative imbalance of 8.5 W/m2, and its name is therefore SSP5-8.5.
  
  You can find lots more details on all of the SSPs on the Emissions Scenario page.
  
• OMIP: the Ocean Model Intercomparison Project
  
  We mentioned AMIP-style simulations in previous sections: OMIP is designed to be the ocean complement to AMIP, where experiments use only the ocean component of a model in order to isolate the effects of physical processes in the ocean.
  
  In all of the OMIP simulations, the atmosphere is prescribed such that all of the different ocean models experience the same conditions at the surface of the ocean. This means that things like surface winds, heat fluxes, rain, and evaporation are taken from observations and imposed on the ocean model: a standard set of observational products is used for this purpose.
  
• PMIP: the Paleoclimate Model Intercomparison Project
  
  Paleoclimates make a useful ‘test case’ for climate models, since they are generally designed to be good at simulating the present-day climate but we know that the Earth has gone through periods which were quite different from now. By running the models in very different climatic situations, we can identify places where things go wrong – or gain more confidence in how they perform, if they’re able to match up with our estimates of what the real past Earth was like during those periods. This is the purpose of PMIP: to standardize experiments that span a range of different past climates, so that we can test models in a consistent way.
  
  PMIP didn’t start at exactly the same time as CMIP, so the numbers of its phases aren’t quite lined up: the CMIP6-era models, run for past climates, are part of PMIP4. There are two main experiments that are required for a modeling center to participate in PMIP4: the mid-Holocene and Last Glacial Maximum, which are described above in the section on CMIP5. Other time periods that are also part of PMIP4: the Last Millennium, the Last Interglacial (roughly 127,000 years go), and the mid-Pliocene Warm Period (roughly 3.2 million years ago).
  
• GeoMIP: the Geoengineering Model Intercomparison Project
  
  As climate change progresses, there are more and more discussions around potentially using technological methods to cool off the planet (“geoengineering”). The most commonly discussed methods for doing this, as of CMIP6, involve stratospheric aerosol injection (SAI), which is what it sounds like: injecting small particles (aerosols) into the upper atmosphere (stratosphere) where they can reflect incident sunlight and therefore cool off the Earth. However, there are all sorts of potential unintended consequences to doing this: to better understand how the details of possible geoengineering implementation might lead to different climate impacts, GeoMIP was born.
  
  The idea behind GeoMIP is to provide a standard set of protocols designed to compensate for different amounts of CO2-induced climate change. This is accomplished in a few different ways: some GeoMIP scenarios artificially reduce the brightness of the sun to approximate reflection by aerosol injection; in others, the sulfate particles associated with SAI are added to the model’s stratosphere. GeoMIP also contains some simulations where marine cloud brightening (addition of sea salt particles to low clouds to increase their reflectivity) is performed.

This is by no means the full extent of MIP activities! The full list of CMIP6-endorsed MIPs is available on the CMIP6 website.

Resources for Further Reading:

CarbonBrief explainer for CMIP6

CMIP6 overview from the World Climate Research Program   

List of approved (“endorsed”) MIPs for CMIP6

References

Eyring, V., Bony, S., Meehl, G.A., Senior, C.A., Stevens, B., Stouffer, R.J. and Taylor, K.E., 2016. Overview of the Coupled Model Intercomparison Project Phase 6 (CMIP6) experimental design and organization. Geoscientific Model Development, 9(5), pp.1937-1958.

Griffies, S. M., Danabasoglu, G., Durack, P. J., Adcroft, A. J., Balaji, V., Böning, C. W., Chassignet, E. P., Curchitser, E., Deshayes, J., Drange, H., Fox-Kemper, B., Gleckler, P. J., Gregory, J. M., Haak, H., Hallberg, R. W., Heimbach, P., Hewitt, H. T., Holland, D. M., Ilyina, T., Jungclaus, J. H., Komuro, Y., Krasting, J. P., Large, W. G., Marsland, S. J., Masina, S., McDougall, T. J., Nurser, A. J. G., Orr, J. C., Pirani, A., Qiao, F., Stouffer, R. J., Taylor, K. E., Treguier, A. M., Tsujino, H., Uotila, P., Valdivieso, M., Wang, Q., Winton, M., and Yeager, S. G.: OMIP contribution to CMIP6: experimental and diagnostic protocol for the physical component of the Ocean Model Intercomparison Project, Geosci. Model Dev., 9, 3231-3296, doi:10.5194/gmd-9-3231-2016, 2016.

Kageyama, M., Braconnot, P., Harrison, S.P., Haywood, A.M., Jungclaus, J.H., Otto-Bliesner, B.L., Peterschmitt, J.Y., Abe-Ouchi, A., Albani, S., Bartlein, P.J. and Brierley, C., 2018. The PMIP4 contribution to CMIP6–Part 1: Overview and over-arching analysis plan. Geoscientific Model Development, 11(3), pp.1033-1057.

Kravitz, B., Robock, A., Tilmes, S., Boucher, O., English, J. M., Irvine, P. J., Jones, A., Lawrence, M. G., MacCracken, M., Muri, H., Moore, J. C., Niemeier, U., Phipps, S. J., Sillmann, J., Storelvmo, T., Wang, H., and Watanabe, S.: The Geoengineering Model Intercomparison Project Phase 6 (GeoMIP6): simulation design and preliminary results, Geosci. Model Dev., 8, 3379-3392, doi:10.5194/gmd-8-3379-2015, 2015

O’Neill, B.C., Tebaldi, C., Van Vuuren, D.P., Eyring, V., Friedlingstein, P., Hurtt, G., Knutti, R., Kriegler, E., Lamarque, J.F., Lowe, J. and Meehl, G.A., 2016. The scenario model intercomparison project (ScenarioMIP) for CMIP6. Geoscientific Model Development, 9(9), pp.3461-3482.

Taylor, K.E., Stouffer, R.J. and Meehl, G.A., 2012. An overview of CMIP5 and the experiment design. Bulletin of the American meteorological Society, 93(4), pp.485-498.