the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 30 Aug 2024
Constraining Aerosol-Cloud Adjustments by Uniting Surface Observations with a Perturbed Parameter Ensemble
Abstract. Aerosol-cloud interactions (aci) are the largest source of uncertainty in inferring the magnitude of future warming consistent with the observational record. The effective radiative forcing due to aci (ERFaci) is dominated by liquid clouds and is composed of two terms: the change in cloud albedo due to redistributing liquid over a larger number of cloud droplets (Nd) and the change in cloud macrophysical properties due to changes in cloud microphysics. These terms are respectively referred to as the radiative forcing due to aci (RFaci) and aerosol-cloud adjustments. While the magnitude of RFaci is uncertain, its sign is confidently negative and results in a cooling in the historical record. In contrast, the adjustment of cloud liquid water path (LWP) to enhanced Nd and associated radiative forcing is uncertain in sign. Increased LWP in response to increased Nd is consistent with precipitation suppression while decreased LWP in response to increased Nd is consistent with enhanced evaporation from cloud top. Observational constraints of these processes are poor in part because of causal ambiguity in the relationship between Nd and LWP. To better understand this relationship, precipitation (P), Nd, and LWP surface observations from the Eastern North Atlantic (ENA) atmospheric observatory are combined with the output from a perturbed parameter ensemble (PPE) hosted in the Community Atmosphere Model version 6 (CAM6). This allows causal interpretation of observed covariability. Observations of precipitation and cloud from ENA constrain the range of possible LWP aerosol-cloud adjustments relative to the prior from the PPE by 15 %, resulting in a global value that is confidently positive (a historical cooling) ranging from 2.1 to 6.9 g/m2. It is found that observed covariability between Nd and LWP is dominated by coalescence scavenging and that this observed covariability is not strongly related to aerosol-cloud adjustments.
- Preprint
(6134 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2024-2158', Anonymous Referee #1, 20 Sep 2024
Please find the referee comment in the attached document.
-
AC1: 'Reply on RC1', August Mikkelsen, 06 Jan 2025
Reviewer questions are in italics, author responses are not.
Referee #1
Figure 2 caption: specify that Figure 2a is from the model.
Thanks for the catch – this has been updated in the manuscript.
P8, L196: What was the approach in Eidhammer et al. 2024 designed to sample the uncertainty in and how does that correspond to the parameters that are likely important for ACI?
The approach in Eidhammer et al., 2024, was designed to investigate uncertainties in subgrid scale processes relating to cloud microphysics, turbulence, convection and aerosols. These processes are important to aerosol-cloud interactions through their intertwined effects on cloud formation, aerosol activation, aerosol emissions, cloud formation, and precipitation formation. This has been expanded upon in Section 2.2
P9, L205-210: Some more information on the model set is needed to improve this section. How long are the simulations for PD and PI? Do the simulations have all anthropogenic and emissions set to PD and PI respectively, or is it just the anthropogenic aerosol emissions that change between the two simulations? Why was the year 2000 used for PD when the observations are from a different year? How useful is a constraint trend on years with different emissions? What is used for prescribing SSTs, sea ice and land cover?
Thanks for pointing this out. 1995-2005 anthropogenic emissions averages have been used, set as such to be consistent with Eidhammer et al., 2024 and AeroCom standards. While we acknowledge that there are differences in aerosol emissions between this average and the modeled period, we don’t think they are significant enough that the constrain would be affected. SST/sea ice are fixed to 1995-2005 averages. Land cover is run through the coupled CESM land model. Clarifications to this point are added in text.
P9, L209: What model levels are nudged?
The model is nudged at all levels. This has been clarified in the manuscript.
P9, L213: It the model output analysed in the gridbox for the ENA surface site is in or is it interpolated in some way?
The model output is analyzed in the gridbox for the ENA surface site. This has been clarified in text.
P10, L218: Does aerosol impact convective precipitation in this model?
While we did not examine convective precipitation specifically in this study, the convection scheme in this model (ZM) is not directly impacted by drop number or activation. Results from Eidhammer et al., 2024, (see figure 8) which has the same PPE setup, shows this succinctly.
P20, L360: Could you remind the reader what criteria is used to rule out an emulate again here, I had to go back to the methods to look it up. Or it might be clearer to reference the constraint process in its own subsection in the methods?
Thanks for pointing this out- we’ve added a quick refresher (and internal reference to the relevant section) in the part of the paper describing the constraint results.
Figure 12 and discussion: In Section 3.1 it is noted how the observable covariation between Nd and LWP in PD may not be predictive of PI-PD LWP changes driven by Nd, and the differences in the predictability of delta Nd for delta LWP between the ENA and globally. These limitations are not really mentioned around Figure 12 or in the discussion. I.e. can you really be confident that constraint of model parameters based on PD observations from one site are representative of PI-PD global changes?
Thanks for bringing this up, it warrants further discussion. One way to interpret the result from Figure 12 is that the PD observations from this one site are ultimately not very strongly predictive of ∆LWP. After constraining to the 6 observations from ENA, the spread of ∆LWPgl is only reduced by ~15% - This can potentially be attributed to the inherent limitations in constraining to a single surface observatory as opposed to broader climatology. While ENA (and other similar surface observatories) provide a unique and useful venue for observing fine-scale processes, it should not be expected that ENA alone will be able to tightly constrain ∆LWPgl. For clues as to the path forward, it is useful to compare the results of this study to Song et al., 2024, which constrains ∆LWPgl in the same PPE to satellite measurements of globally-averaged LWP, Nd, and upwelling top-of-atmosphere shortwave radiation. When comparing the constrained ∆LWPgl ranges between these two studies, while neither constraint is especially strong on its own, each constraint rules out different extremes in ∆LWPgl. This work shows a larger minimum constrained ∆LWPgl (2.08 g/m2); Song et al., 2024, shows a smaller maximum constrained ∆LWPgl (4.33 g/m2). While this should not be considered a strict scientific comparison, it suggests that a more robust constraint can be found by constraining on the fine-scale (e.g., surface-observed precipitation) and the broader climate-scale (e.g. satellite-observed average global upwelling shortwave radiation) simultaneously. This is an idea we will be exploring in future work. This additional discussion has been added to the manuscript in the Discussions section.
P24, L415: Typo – figure 14 is repeated a few times.
Fixed, thank you for catching that.
Citation: https://doi.org/10.5194/egusphere-2024-2158-AC1
-
AC1: 'Reply on RC1', August Mikkelsen, 06 Jan 2025
-
RC2: 'Comment on egusphere-2024-2158', Anonymous Referee #2, 03 Oct 2024
The comment was uploaded in the form of a supplement: https://egusphere.copernicus.org/preprints/2024/egusphere-2024-2158/egusphere-2024-2158-RC2-supplement.pdf
-
AC2: 'Reply on RC2', August Mikkelsen, 06 Jan 2025
Reviewer questions are in italics, author responses are not.
Referee #2
Section 1 (95-105) and Section 2.1 (125-130) both contain discussions about the utility of surface observations to constrain aerosol-cloud adjustments. Suggest re-arranging and/or merging these comments and keeping 2.1 focused on the features of the ENA observatory.
Thanks for the feedback, we have re-arranged and combined these sections for clarity and conciseness.
While the paper focuses largely on the LWP response to aerosol perturbations, no mention is made of the cloud fraction (CF) adjustment. Typically, SW ERFaci is decomposed into the Twomey effect, LWP , and CF adjustments. The emphasis on cloud water changes is clear throughout the paper, but the CF adjustment might merit a mention in the Introduction to fully address the scope of what “aerosol-cloud adjustments” implies to readers.
Thank you for the feedback, we have added some additional clarification to this point in the Introduction.
Section 2.1.2 mentions that the observations of surface rain rate may miss those in the “drizzle domain,” and Figure 6 shows that the ENA observations approach the upper range of PPE values for mean rain rate. How might a more accurate precipitation observation (that captures drizzle) impact this result, and can you be certain that differences can be attributed to sampling (grid cell mean vs station)? If the precipitation observations were out of the range of the PPE, how would that change your analysis? Given the importance placed on mean-state precipitation as a constraint (line 370), this may bear further discussion.
We are not certain whether the difference in precipitation between models and observations stems from sampling vs. instrument capability. If we had a more accurate precipitation observation, though, we could at least eliminate one of these issues, allowing us to home in on process disagreements. If this caused us to get observations that were outside of the range of the PPE, we would need to expand our PPE to a broader range of parameter values. However, we would expect a more accurate observation to bring the mean precipitation down towards the bulk of the PPE distribution, because it would see a larger number of light rain observations. Because we cannot adequately separate these uncertainties at this time, however, the observation stands as-is. Additional clarification to these points has been added to the Discussion section of the manuscript.
How does the 15% reduction in spread of global delta LWP compare to the results in Song et al. (2024)? Would you expect the constraint found here to be weaker due to the use of surface observations as a constraint? The use of an identical PPE could be cause for assessing the constraint found here.
Thanks for the feedback on this. Generally, this result is in line with results of Song et al., 2024. Both constraints on ∆LWP are relatively weak and constrain CAM6 towards stronger adjustments. We would also agree with the idea that the ∆LWPgl constraint in Song et al., 2024, is stronger because they are using global observations rather than point-source observations. While we attempted to control for a larger regime with the state variables (e.g. median LWP) at ENA, ultimately controlling for observations from a single point on the planet, even one well-correlated with ∆LWPgl, may not be strong enough to significantly constrain uncertainty in this area. One interesting contrast between our ∆LWPgl constraints is that, with the constrained distributions taken together, we rule out different extremes. In this work, we see a larger minimum constrained value (2.08 g/m2); in Song et al., 2024, a smaller maximum constrained value is found (4.33 g/m2). While we would not put forth this exact comparison as a precise scientific result, we posit this is representative of the benefits that unifying these styles of analyses can have: by combining the benefits from constraining on the fine-scale (e.g., surface-observed precipitation) and the broader climate-scale (e.g. satellite-observed average global upwelling shortwave radiation) simultaneously, a more robust constraint can be found. Further discussion on this point has been added in the beginning of the Discussions section of the manuscript.
How relevant is a constraint on the spread of global delta LWP when the prior range is entirely positive, given that there is a well-documented tendency for GCMs to predict uniformly positive LWP responses to aerosol perturbations, which disagrees with conclusions from many observational studies? There is relatively little acknowledgement of this in the introduction and conclusion, but it merits consideration.
Results from many present-day observational studies regarding LWP responses to aerosol perturbations tend to occur in highly specific environments from which it is difficult to extrapolate PD-PI changes, and causality is not so easily observed (e.g. Christensen et al., 2017). Furthermore, it has been shown that PD variability in aerosol-cloud interactions that can create these sorts of “negative relationships” between Nd and LWP can be replicated in models with positive adjustments (Mülmenstädt et al., 2024). So, following Mülmenstädt et al., 2024, and the results from Figure 14, it is not considered problematic that this model uniformly produces positive ∆LWPgl because this result is ultimately not in contradiction of these observational studies. To help better convey this point, additional clarification has been added to the end of the Discussions section.
Can the findings in Figure 13 be utilized by CAM6 developers for parameter tuning the next version of CAM?
Yes- we think that if used in concert with Figure 10 from Eidhammer et al., 2024 (which is a similar collection of constrained histograms) CAM6 developers may be able to make some specific inferences about parameters that can be tuned towards better representation of aerosol-cloud interactions. We specify “in concert with Eidhammer et al., 2024” because, going off the thought process described in our response to “How does the 15% reduction in spread of global delta LWP compare to the results in Song et al. (2024)?” we think it is prudent for model developers to utilize responses from fine- and broad-scale constraints simultaneously. Tuning parameters where the constraints seem to agree strongly (e.g., clubb_C2rt) may be a good place to start. Additional commentary in service of this point has been added to the Discussions section.
And can any of your findings inform future PPE studies - are there parameter ranges that should be widened or narrowed?
From Figure 13, we would suggest that the prescribed parameter range for zmconv_capelmt should be limited to a smaller range in future PPEs for CAM6. We have added this recommendation to the Results section surrounding Figure 13. Following the discussion to the previous question, we cannot confidently recommend changes to any of these ranges otherwise.
Are there parameters that should be omitted due to showing little influence on ACI and related processes? Some of this is briefly discussed in line 510, but could be expanded to widen the implications of this study.
We think that would depend on the experiment. If we were to rerun a PPE for this experimental setup, then yes- we would remove all parameters that had a flat distribution, tighten the distributions of the remaining parameters towards the constraint, and rerun the ensemble. But for future work, as discussed in prior responses, we are interested in simultaneously constraining off fine- and broad- scale observations for a more holistic and robust climate-wide aci constraint. It’s possible that some of these other parameters, which didn’t show a strong response in this particular regime, may be important when constraining across these broader climate levels. So, in this case, we would be hesitant in removing them altogether. Commentary on these points has been added to the beginning of the Discussion section of the text.
Citation: https://doi.org/10.5194/egusphere-2024-2158-AC2
-
AC2: 'Reply on RC2', August Mikkelsen, 06 Jan 2025
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 271 | 83 | 159 | 513 | 6 | 8 |
- HTML: 271
- PDF: 83
- XML: 159
- Total: 513
- BibTeX: 6
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1