CMIP5 offers natural-forcings-only simulations over 1850-2005, which allows comparison with the historical simulations, which allows us to show how anthropogenic forcings have affected the climate over that time.

CMIP5 also offers projections over the 21st century, with a range of RCPs allowing for different emissions scenarios.

However, CMIP5 doesn't include a 21st century natural forcing for comparison, even though this would be useful. Do such simulations exist anywhere else?

I realise that these would not be perfectly possible, since data for natural forcings like volcanoes are not available. However, these might be reasonably simulated using a stochastic model. Other natural forcings, such as orbital forcings, can be calculated fairly easily over the next century. Solar irradiance probably falls somewhere in between the two.

If such simulations do not exist, would it be unreasonable to use the 1850-2005 natural forcings simulations as the basis for a statistical model of the 21st century? Or are some of the natural forcings likely to change substantially (e.g. are 20th century volcanoes unrepresentative of the background, or is the orbital forcing likely to change substantially over the coming century?).

  • $\begingroup$ Is there even a meaningful way to generate such a simulation, since the current climate system already has long-term anthropogenic forcings baked into it via CO2 and ocean heat content? ("Due to its large heat capacity, the ocean is the likely source of natural long-term climate variability on interdecadal time scales." pnas.org/content/106/38/16120) $\endgroup$ Commented Dec 11, 2018 at 16:55
  • 1
    $\begingroup$ @jeffronicus: Ocean heat content isn't a driver, it's a storage pool. The pre-industrial-CO2-level simulations are already run for 1850-2005 (CMIP5), it wouldn't be technically hard to continue running them with the same CO2 levels. Adding in Volcanoes would be a bit more complex, I'm not sure how reasonable a stochastic model would be for volcanoes, or how much impact different eruptions would have over the long term. $\endgroup$
    – naught101
    Commented Dec 12, 2018 at 1:20

1 Answer 1


You’re right that there were no future natural forcings experiments defined in the CMIP5 protocol. The reason being simply that the modeling groups thought that other experiments were more deserving of their limited computing time. Compare this with the much stronger call for the historicalNat and historicalGHG experiments, which were required for detection and attribution studies. Individual modeling centers may have run future natural forcings experiments, but that will be hit and miss, and there wasn’t an experiment name to lodge the outputs to in the CMIP5 database. Incidentally, the situation with the RCP (all forcings) experiments themselves was actually a bit imperfect, because the implementations of future natural forcings were left to the discretion of each modeling group:

  1. … It is recommended that either volcanic aerosols should be omitted entirely from the control and future runs, or, alternatively, the same background aerosol should be prescribed in both runs.
  2. It is recommended that some representation of the solar cycle be included in the 20th and 21st century simulations, though that is left up to the discretion of the modeling groups.

Source: Taylor et al (2011). A summary of the CMIP5 experiment design. (pdf)

There was a further recommendation for how future solar irradiance forcing should be represented:

In CMIP5, climate projections were based on a stationary-Sun scenario, obtained by simply repeating solar cycle 23, which ran from April 1996 to June 2008 (Lean and Rind, 2009).

Source: Matthes et al (2017). Solar forcing for CMIP6 (v3.2), Geosci. Model Dev., 10, 2247-2302, https://doi.org/10.5194/gmd-10-2247-2017

But, in practice, groups did slightly different things from each other, so you usually have to check the output file metadata and any associated journal papers for individual model runs to get an idea of the details.

This is clearly not a crazy question though, because in CMIP6 some future natural forcings experiments have been defined, albeit in the lower priority tiers. Specifically, the Detection and Attribution MIP (DAMIP) includes ssp245-GHG and ssp245-nat experiments, and the volcano MIP (VolMIP) has the volc-cluster-21C experiment. If you dig down into those you’ll find that future solar variability is derived more or less the way you describe, i.e., projecting forward based on a set of statistical time series models trained on past variability. Note, however, that Matthes et al are careful to point out that this is a projection not a prediction:

As of today, predicting solar activity up to 2300 is very challenging, if not impossible… The unusually long solar cycle number 23 that ended in 2009, and the weak one (no. 24) that followed came as a surprise, and manifested our evident lack of understanding of the solar cycle.

Conversely, future volcanic forcing in ssp245-nat will be prescribed as a near-constant background forcing:

Volcanic forcing will be ramped up from the value at the end of the historical simulation period (2015) over 10 years to the same constant value prescribed for the piControl simulations in the DECK, and then will be kept fixed.

Source: O'Neill et al (2016). The Scenario Model Intercomparison Project (ScenarioMIP) for CMIP6. Geosci. Model Dev., 9, 3461-3482, http://doi.org/10.5194/gmd-9-3461-2016

So why not a statistical model for volcanic emissions? This forcing is episodic rather than cyclic and is prescribed to the CMIP models as several variables on latitude-height-time grids. As you hint at in your question, this is too many degrees of freedom to pin down satisfactorily from just the 1850-2005 record. Even if it could be done, you then have the problem of how to use the statistical model, given that each climate model will only run a few realisations of a particular projection. Do you generate one scenario to be shared by all models, or do you let each group generate its own scenarios using the statistical model? Either way, you can’t explore enough of the parameter space to avoid ending up with a CMIP6 ensemble containing decadal variability that is strongly influenced by forcing scenarios that are extremely unlikely to happen.

As I understand it, the current assumption is that volcanic forcing is stationary and, in the absence of knowing the future sequence of eruptions, its effect on slowly varying climate components like ocean heat content is best represented by a constant mean forcing.

  • $\begingroup$ Solid answer, thank you. Your points about the too large degrees of freedom make a lot of sense, especially given the high resolution of many models these days. I guess what would be nice would be a lot of low-resolution models with as broad a sample of the input uncertainty as possible. But that's really a separate question to what I was asking. $\endgroup$
    – naught101
    Commented Jan 29, 2019 at 23:43

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.