Scenarios, schmenarios…

The fantasy version of the normal updating of scenarios for a new round of CMIP simulations doing the rounds is bad faith BS.

As climate folk will know, the community is currently embarking on a new round of climate model simulations to support analyses and projections for the next IPCC report (due in 2028/9). This new effort has been dubbed CMIP7, because it is the sixth iteration of the CMIP effort (IYKYK), that started in the late 1990s. For each of these iterations, a new set of projections has been formulated for the modeling groups to use and the ones for this round were just published (van Vuuren et al., 2026). So far, so totally normal.

Why do scenarios need updating? Why can’t we use the three scenarios that Hansen et al. (1988) first came up with in the early 1980s? Three reasons. First the scenarios need to be continuous with the trajectories of the observed changes. The ‘join’ point was 1984 for Hansen’s original scenarios, then 2000 for the CMIP3, 2005 for CMIP5, 2014 for CMIP6, and it will be 2023 for CMIP7. As you can imagine, things have changed over the last 40 years (the Montreal Protocol, the Clean Air Acts, renewable energy price falls, fracking, the Paris Agreement, actual climate policies, reversal of climate policies, etc.). All of these things are the result of humans behaving in ways that humans behave and which are not easily predictable ahead of time. This is why future simulations have long been described as ‘projections’ and not predictions.

Second, the rationale for future scenarios has shifted in light of what we (as a society) are doing. At the beginning it made sense to think about a spread of baseline scenarios where no climate policy was enacted: “Business as usual” so to speak. But now? we have already done things and so ‘business’ is no longer ‘usual’. Now, ranges based on ‘current policies’, ‘current aspirations’ and ‘possible backsliding’ are perhaps more useful. Additionally, we are now much closer to 2100 than we used to be (also obvious, but often forgotten) and so scenarios need to be extended out further.

Third, what we are making scenarios for has expanded enormously. In 1984 there were only concentrations of well-mixed greenhouse gases, the solar cycle and the occasional volcano to project, but now, we have emissions of GHGs including CO₂, plus all of the halogenated gases, the short lived climate forcers (CH4, aerosols, NOx, SO₂/SO₄), land use change (deforestation, irrigation, agricultural shifts), possible anthropogenic impacts on dust and fire, and freshwater inputs from melting glaciers and ice sheets that are not otherwise represented in models. Did I mention nitrogen inputs, solar particle fluxes, and volcanic emissions of water vapor as well as sulfates? Keeping this all coherent and up-to-date is an enormous undertaking.

This all means that, duh, of course the scenarios would be updated for CMIP7.

People (hi Roger!) acting as if the publication of new CMIP7 scenarios is some huge policy shift or an admission that previous scenarios are no longer ‘official’ are just bull-shitting. This is something that was planned for and expected for literally years. Previous scenario sets were used in previous rounds, a new set will be used for the new round – that is all there is to it.

Oh noes!

The supposed focus of the ire are the high end scenarios of RCP85 (CMIP5) and SSP5-85 (in CMIP6). The reasons why these were set up in the first place (back in 2007!) was that IPCC wanted to span what had appeared in the literature before then – going beyond (in sophistication) what the (CMIP3) SRES projections had done. But the IAM folks involved decided (correctly) that they didn’t have the time to start from scratch, and so they decided to split up the task – come up with a spread that covered 99% of published scenarios with ‘representative’ concentration pathways (RCPs – gettit?) for the climate models to use quickly (in CMIP5), and back-fill plausible socio-economic pathways later (to be used in CMIP6). Thus the CMIP5 models (which were run in 2007-9 or so) used the RCPs (including RCP85 – which was so-called because it reached 8.5 W/m² of direct radiative forcing from GHGs in 2100).

For climate modellers, the reasons why the pathways are the way they are is a secondary concern – if they were only to be given CO₂ levels (and other GHGs etc.), the basic need is just for a low, middle and high scenario that encompass our most ambitious climate policy pathways, a worst case scenario (‘Burn it all!’), and something in the middle. When it came time for CMIP6 (2016-2019 or so), the SSPs (that had been promised a decade earlier) were ready, and so they were used. But for the purposes of the climate modellers, the drivers underlying the SSPs were not really that relevant. A climate model really doesn’t care how cooperative or antagonistic regional economic blocks are – it just responds to the resulting emissions. That there is a need for high end scenarios should be obvious – where are the tipping points in the system? what are the impacts of a 2ºC warmer world? what about 3ºC or 4ºC? Are these worth avoiding perhaps? Having seen these results, the answer (IMO) is definitely yes!

In the last few years, a number of people have pointed out that assumptions underlying the highest SSPs don’t look as plausible as they used to seem (this is also true of the lowest projections, but people seem less bothered by that). Note this is many years after all the models that were ever going to use them were run. But this is less of an issue that some people portray. Climate impacts rely on a chain of calculations – a specific set of emissions, a resulting concentration pathway, and a modelled sensitivity. Similar impacts can arise with lower emissions, but greater carbon cycle feedbacks and higher sensitivities, and given that each of these steps are quite uncertain, it is not really worthwhile for climate models (or modelers) to get too attached the specific storylines the IAM folks put together. Hence the collective shrug from climate modelers around the RCP85 ‘dialogue’ in the last couple of years.

Let me give two examples why high end scenarios are important: Impacts on ice sheets are a very important part of the climate change and have yet to be fully integrated into the standard climate models. So independent efforts with ice sheet models were set up using the output from the CMIP5 and CMIP6 models – they used two scenarios, RCP85 and RCP26 to bracket possibilities in ISMIP6 (Seroussi et al, 2020). Interestingly, they found that, particularly for Greenland, that none of the models had melt rates as high as observed even with RCP85 forcing. Thus for a situation where the (ice sheet) models are insufficiently sensitive, a higher than expected forcing might give you a more likely outcome.

A second example is the use of ‘global warming levels’ in the last IPCC assessment. These were averages of the models when they reached particular temperature levels (2ºC, 3ºC, 4ºC etc.), but for that to work, enough models had to reach those temperatures in order to make an average – and in practice for 3ºC and 4ºC, this was only possible with SSP5-85 scenarios. Other assessments used the higher signal-to-noise ratios in the high end scenarios to estimate sensitivities across many systems that would have been noisier and more uncertain if that was not available. How the models got there is basically irrelevant. The new high end runs will also be used for this (note that H gets to 8.5 W/m² only about 20 years after RCP85).

Note that even the harshest critics of RCP85 will admit (in academic circles at least) that these are legitimate uses. However, some of the more stupid commentaries equate the mere mention of RCP85/SSP585 with scientific misconduct, claiming that counting the number of times the ‘naughty’ words of RCP85 appear in publications or assessments is a damning indictment of the entire field’s integrity. This is so dumb and lazy that I find it hard to credit.

But wait!

The funny thing is that there are real issues with the way this whole endeavor has grown up. First, because CMIP is the only (serious) climate projection game in town, as climate change has become more salient, CMIP projections have been used to inform a far wider array of science than was imagined back in 2007. Not all of those uses are optimal. For instance, the ERA5 reanalysis still uses CMIP5 projections of solar forcing from 2008 which didn’t turn out to be so good at matching what actually happened. Consultants and banks have used CMIP6 projections as if they were real predictions, and the adaptation community have often assumed that specific CMIP6 pathways are the most likely outcomes.

All of these misuses are compounded by the fact that it appears to need a decade to update these pathways in the light of new science and societal decisions and changes. This is way too long – annual updates of the process should be achievable if funders prioritized it.

And finally, there are of course far more scenarios that would be interesting to explore in climate models (policy specific scenarios, delta scenarios (where only one thing changes at a time), in-between scenarios, annually updated scenarios etc.) than can possibly be performed given existing computational capacity. This is (right now) prohibitive, but at the rate that faster, more efficient machine-learning emulators are advancing, it might not be for much longer.

A serious critique of the climate modeling enterprise would be focused on these issues (for instance), rather than tilting at RCPs.

References

1. D. van Vuuren, B. O'Neill, C. Tebaldi, L. Chini, P. Friedlingstein, T. Hasegawa, K. Riahi, B. Sanderson, B. Govindasamy, N. Bauer, V. Eyring, C. Fall, K. Frieler, M. Gidden, L. Gohar, A. Jones, A. King, R. Knutti, E. Kriegler, P. Lawrence, C. Lennard, J. Lowe, C. Mathison, S. Mehmood, L. Prado, Q. Zhang, S. Rose, A. Ruane, C. Schleussner, R. Seferian, J. Sillmann, C. Smith, A. Sörensson, S. Panickal, K. Tachiiri, N. Vaughan, S. Vishwanathan, T. Yokohata, and T. Ziehn, "The Scenario Model Intercomparison Project for CMIP7 (ScenarioMIP-CMIP7)  ", 2025. http://dx.doi.org/10.5194/egusphere-2024-3765
2. J. Hansen, I. Fung, A. Lacis, D. Rind, S. Lebedeff, R. Ruedy, G. Russell, and P. Stone, "Global climate changes as forecast by Goddard Institute for Space Studies three‐dimensional model", Journal of Geophysical Research: Atmospheres, vol. 93, pp. 9341-9364, 1988. http://dx.doi.org/10.1029/JD093iD08p09341
3. H. Seroussi, S. Nowicki, A.J. Payne, H. Goelzer, W.H. Lipscomb, A. Abe-Ouchi, C. Agosta, T. Albrecht, X. Asay-Davis, A. Barthel, R. Calov, R. Cullather, C. Dumas, B.K. Galton-Fenzi, R. Gladstone, N.R. Golledge, J.M. Gregory, R. Greve, T. Hattermann, M.J. Hoffman, A. Humbert, P. Huybrechts, N.C. Jourdain, T. Kleiner, E. Larour, G.R. Leguy, D.P. Lowry, C.M. Little, M. Morlighem, F. Pattyn, T. Pelle, S.F. Price, A. Quiquet, R. Reese, N. Schlegel, A. Shepherd, E. Simon, R.S. Smith, F. Straneo, S. Sun, L.D. Trusel, J. Van Breedam, R.S.W. van de Wal, R. Winkelmann, C. Zhao, T. Zhang, and T. Zwinger, "ISMIP6 Antarctica: a multi-model ensemble of the Antarctic ice sheet evolution over the 21st century", The Cryosphere, vol. 14, pp. 3033-3070, 2020. http://dx.doi.org/10.5194/tc-14-3033-2020