This paper investigates whether aerosol forcing is the reason why the CMIP6 ensemble is cooler than observations in the middle of the 20^th century, as opposed to the CMIP5 ensemble that warms roughly in line with observations. This is a topic that continues to intrigue researchers and one that it is important to try to understand, in order to potentially correct model biases in the future. Unfortunately, the authors could not reach a definitive conclusion around what causes the mid-century cool period in CMIP6 models, finding it is not solely due to aerosol forcing which is the obvious candidate, but aerosol forcing is likely to be one factor of many. In this regard, they reach similar conclusions to Smith & Forster (2021). Despite a null result, this is a useful contribution to the literature and will hopefully motivate other researchers to continue to study the topic.

Comments are mostly minor, but please note for the ERFari values reported in this study, unfortunately the method I used in Smith et al. (2020) was slightly flawed! See Zelinka et al. (2023), https://egusphere.copernicus.org/preprints/2023/egusphere-2023-689/egusphere-2023-689.pdf which explains and corrects this. Updated values are in table 2 of that paper. I trust that using corrected values will not change the results of this paper substantially.

Minor comments:

4: “observed anomaly” – which period are we talking about here?

13: “encouraging” – why? Either that there is some consistency that hints at a constraint, or that weak aerosol forcing is good in the sense that it implies a smaller committed warming (Watson-Parris & Smith 2022, https://www.nature.com/articles/s41558-022-01516-0)?

15-17: Fully agree with this statement

31: Bellouin et al. (2020); also Forster et al. (2021), the AR6 WG1 Chapter 7 assessment, came to a very likely range for 1750 to 2005-14 of -2.0 to -0.6 W m-2, again from multiple lines of evidence; the Bellouin et al. paper put in much of the foundations for this work.

38-39: Smith et al. (2021; https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2020JD033622) also found the Stevens model to be overly simplistic and could not capture the diversity of historical aerosol forcing, so proposed two modifications: the addition of additional species, and relaxing the constraint that the aerosol indirect effect depends on natural emissions and replacing this with a generalised shape factor, allowing forcing to scale logarithmically (as proposed by Stevens) or approximately linearly (as proposed by Booth and Kretzschmar) with emissions, depending on parameter choices. With this model it was easily possible to obtain stronger aerosol forcing than -1.0 W m-2 that was consistent with historical warming.

74: following comment above, best to take results from Zelinka et al. (2023).

91: is it necessary to exclude 1963-66? Since the simulated climate projections from CMIP6 should have included volcanic forcing too and hence the contribution from Agung is present in both the observations and the models.

115: Would it be better to use piClim-anthro? The sum of piClim-ghg and piClim-aer excludes contributions from land use change and ozone. It also excludes natural forcings, though there is not a time slice in RFMIP available to estimate it and it’s fair to assume there wasn’t a big change from 1850 to 2014.

119: I’m not sure I understand the drift correction method in the piClim-ghg and piClim-aer experiments. The piClim-control is only 30 years for most models so I’m not sure there are many branch points for piClim-ghg and piClim-aer. As the ERF calculation uses fixed SSTs this should also remove the need for drift correction. For some forcings there is a relaxation time where the atmospheric response to a forcing is not instant; fig. 2 of Smith et al. (2020) shows this in action (CNRM-ESM2-1 aside).

184: Clear sky flux change is not the same as ARI. However, they are highly correlated, so I suppose you can use clear sky flux change as a proxy for ARI. Section 4.3 in Zelinka et al. (2014) gives a good discussion (https://agupubs.onlinelibrary.wiley.com/doi/full/10.1002/2014JD021710). I did calculate ERFari from 13 CMIP6 models (using the correct version of the APRP code) here, if you want to use it: https://github.com/chrisroadmap/cmip6-aerosol-forcing/tree/main/output

204: this is true in the ensemble of opportunity that CMIP6 models provide: a sample size of 36 models, only about half of which can give you an estimate of present-day aerosol forcing, even fewer give you an estimate of the aerosol forcing during the period of interest. Although not stated, I’m uncomfortable in claiming this to be a true result in the real world, as we showed in Smith et al. (2021).

243: no fault of the authors, but some of the results in the paper suffer from a lack of participating models in each Block, showing again how important that models run the ERF experiments from RFMIP.

252: this suggests that >a low< greenhouse gas forcing…

282: it remains a mystery. Would the pattern effect have anything to do with it? I’m not sure how this study would evaluate this. Smith et al. (2021) included the effect of a forced pattern effect from the increasing climate sensitivity over time as simulated by an ensemble of energy balance model simulations trained on CMIP6 models, but did not evaluate this effect either. I could see how a strong aerosol forcing could be consistent with a virtually non-existent historical pattern effect, or a weak aerosol forcing masking a strong pattern effect. AMIP experiments (Andrews et al. 2018, https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2018GL078887) point towards a strong historical pattern effect, maybe adding weight to the suggestion that aerosol forcing may be on the weak side.

283: again unfortunately the ARI is wrong in Smith et al. (2020).

291: this should not be that surprising. Energy budget arguments can permit a present-day aerosol forcing as strong as -2 W m-2, as discussed in Bellouin et al. (2020) and AR6; Smith et al. get quite close with a 5^th percentile of -1.8 W m-2. The time evolution of the historical forcing matters, not just its present day value, and pattern effect probably matters too.

Review of “Strong aerosol cooling alone does not explain cold-biased mid-century temperatures in CMIP6 models” by Flynn et al.

This study presents an analysis of the mid-century (1940-1970) “suppressed warming” or cold temperature bias in the CMIP6 generation of models versus CMIP5 and observed warming. The authors partition the CMIP6 models into 3 blocks depending on whether the simulated mid-century temperature anomaly is above the CMIP5 mean; is between the CMIP5 and CMIP6 means or is below the CMIP6 mean anomaly. The relationship between these blocks of models and the effective radiative forcing from both aerosols and well-mixed greenhouse gases is examined. While block 1 models have the most realistic mid-century temperature anomaly and weakest aerosol forcing, there is no meaningful difference between the other 2 blocks. Block 3 in particular deviates from the expected temperature anomaly-aerosol ERF relationship simulated by sensitivity runs conducted with the MPI-ESM1.2-CR model. This points to other causes, outside of the aerosol ERF, of the cold biases in these models. In addition no meaningful correlation is found with the WMGHG ERF. While not the golden nugget in understanding the underlying cause(s) of the anomalous historical cooling that is a trait of many CMIP6 models, this is a useful addition to the expanding literature discussion on this topic. It is a topical, and relevant study and would be suitable for publication in ACP following consideration of minor comments below.

Minor comments:

My main over-arching concern regarding the main Conclusions of this paper is around the limited number of models in the CMIP6 database that have carried out these, one could argue, essential simulations. This number is further reduced if you break it down into models which interactively simulate the evolution of aerosol versus those that simply prescribe aerosol concentrations.  While the authors acknowledge this in their results it very much limits the significance of the findings, in particular wrt Block 1 which only includes 3 models.  The reader should be very clear on these limitations, it should be clearly mentioned/repeated in the Conclusions and the authors should highlight the requirement for more models to do these experiments.

L80 : What is the limitation of only using the first ensemble member of each models historical ensemble? This might not necessarily represent the ensemble mean temperature anomaly and so could impact the results. Have the authors examined this?

L85 : There is a more recent version of HadCRUT data (HadCRUT5) that is better to use (https://doi.org/10.1029/2019JD032361)

L116: I don’t understand why the models exhibit drift, these should be 30 year fixed SST timeslices? There should also be no branching necessarily (L119 in the perturbed forcing experiments as they are just parallel experiments?

Figure 3: It would be informative to include the Block 1/2/3 mean values here as well as the observed temperature anomaly for reference.

L193: misspelt aerosol

L202-204:  I don’t understand what is meant by this sentence, consider rephrasing.

L214: I think one of the findings / hypotheses of the Zhang paper was the role of process complexity (ie: earth system additional process such as fully interactive chemistry influencing aerosol composition and subsequent forcing versus physical climate models with lower complexity) and not just aerosol emissions.

Section 4.2: I find this section a little limited/simplistic as there could be many other sources of systematic differences in these models outside of whether it takes aerosol emissions as input or not. For instance, the level of complexity of the aerosol scheme, how the ACI and ARI are represented not to mention the role of other parts of the physical system which could influence surface temperatures, eg the role of the ocean. These should at the very least be mentioned.

L222: should read: a standardized anthropogenic aerosol emissions input data set. Not all natural emissions were standardized across the models.

L223-230 It might be better to visualise the prescribed concentration vs emissions-based models. Figure 3 could be repeated for instance changing the colours of data points to represent PC or E or PC/ice or E/ice, where ice = aerosol-ice interaction.

Why is Figure 5 only mentioned in the Conclusions? Should this not be part of Section 4.1?

Finally, consider citing the paper of Mulcahy et al. (2023) (https://doi.org/10.5194/gmd-16-1569-2023). This updated configuration of UKESM1 shows a significant improvement in the cold historical temperature bias despite only a small change in the present-day aerosol ERF. This directly supports this study and points to a more nuanced process driving surface temperature response in the fully coupled system.