Interpreting Differences in Radiative Feedbacks From Aerosols Versus Greenhouse Gases

Experiments with seven Coupled Model Intercomparison Project phase 6 models were used to assess the climate feedback parameter for net historical, historical greenhouse gas (GHG) and anthropogenic aerosol forcings. The net radiative feedback is found to be more amplifying (higher effective climate sensitivity) for aerosol than GHG forcing, and hence also less amplifying for net historical (GHG + aerosol) than GHG only. We demonstrate that this difference is consistent with their different latitudinal distributions. Historical aerosol forcing is most pronounced in northern extratropics, where the boundary layer is decoupled from the free troposphere, so the consequent temperature change is confined to low altitude and causes low‐level cloud changes. This is caused by change in stability, which also affects upper‐tropospheric clear‐sky emission, affecting both shortwave and longwave radiative feedbacks. This response is a feature of extratropical forcing generally, regardless of its sign or hemisphere.

• Effective climate sensitivity is larger (feedback more amplifying) for historical anthropogenic aerosol than greenhouse-gas (GHG) forcing in Coupled Model Intercomparison Project phase 6 • The key difference is that GHG forcing is global and aerosols are mainly extratropical (and aerosol hemispheric contrast unimportant) • Extratropical forcing causes a shallower temperature response than tropical forcing, hence more amplifying cloud and lapse-rate feedbacks

Supporting Information:
Supporting Information may be found in the online version of this article.
SALVI ET AL.
10.1029/2022GL097766 2 of 9 In the recently released data of the Coupled Model Intercomparison Project phase 6 (CMIP6), historical single-forcing experiments across several models have become available for analysis. These new experiments allow us to obtain the effective forcings for different agents, allowing us to accurately calculate the corresponding radiative feedbacks over the historical period. Previous work has found a dependence of these feedbacks on SST patterns (Andrews et al., 2015;Ceppi & Gregory, 2019;Zhou et al., 2017), which affect the radiation budget via changes in stability and clouds . Following this reasoning, this work considers how historical greenhouse gas and aerosol forcing affect SSTs and stability, and how these in turn modulate the radiative responses.

Data and Methods
Data for several historical forcing experiments was obtained from the ESGF CEDA archive (esgf-index1.ceda. ac.uk) for seven models: CanESM5, CNRM-CM6-1, GISS-E2-1-G, HadGEM3-GC31-LL, IPSL-CM6A-LR, MIROC6, and NorESM2-LM. These experiments include a control (piControl) with constant preindustrial forcing agents as well as experiments both with coupled atmosphere-ocean models (AOGCMs) and with atmosphere models (AGCMs) given prescribed sea surface conditions, for historical GHGs, aerosols, and all historical forcings together. The experiments, models and variants used are detailed in Table S1 in Supporting Information S1. We analyze the upward TOA radiative response R, the stability response S as measured by the estimated inversion strength (EIS; Wood & Bretherton, 2006), and the cloud-radiative effect (CRE), measured as the difference between all-sky and clear-sky downward fluxes. Prior to analysis, all fields were regridded to a common T42 grid (corresponding to a grid resolution of approximately 2.8° in longitude and latitude), using conservative remapping for radiative fluxes, and bilinear interpolation for other fields. Monthly fields were aggregated into annual averages.
In order to separate out the surface warming (or cooling) driven feedbacks from the forcing and associated rapid adjustments (Hansen et al., 1997;Sherwood et al., 2015), we use the results from the AGCM experiments with fixed sea surface temperatures through the following equation: where X response represents the response of the variable of interest X. Radiative feedbacks and other responses per unit global warming were calculated through linear least-squares regression of the desired variable against globally averaged surface air temperature. The regressions are performed across the full length of the historical period (years 1850-2014) for all experiments, despite the possible time-dependence of α (Gregory & Andrews, 2016;Gregory et al., 2020) which we do not study in this work. Note that the impact of short-lived volcanic forcing should be minimal on the historical feedbacks calculated over this long time period.
Global averages in this paper refer to area-weighted averages of the variables of interest, except for stability S where the average is taken over ocean regions 50°S-50°N. This choice follows Ceppi and Gregory (2019), who found EIS equatorward of 50° to be well correlated with the global radiative response R in observations and in two GCMs.
The confidence intervals for results derived from regressions combine two aspects of uncertainty. The first is the variability among different ensemble members, which we have calculated for each model as the variance across members of the historical experiment. This experiment was chosen since it generally had the most ensemble members, with the assumption that the magnitude of unforced variability is representative of other forcing scenarios. The second is the estimated error in the regression slope of the ensemble-averaged data. The combined error is calculated as the square root of the sum of variances from these two aspects of uncertainty and given as ±1σ of the probability distribution.

of 9
Two idealized extratropical forcing experiments were run to investigate the impact of forcing localized away from the regions of deep convection on radiative feedbacks and tropospheric stability. These experiments are denoted as nh_extrop and sh_extrop for northern (30°N-90°N) and southern (30°S-90°S) hemisphere forcing, respectively. A uniform "ghost" radiative forcing of −4.5 W m −2 was imposed as an extra term in the surface energy budget in the relevant regions for each experiment. The imposed instantaneous radiative flux values were chosen so that the global average forcing would be comparable to that of hist-aer over years 1995-2014 (around − 1.13 W m −2 ). Although the nh_extrop forcing is zonally uniform unlike the forcing for hist-GHG and hist-aer (Figures 1a and 1b), the nh_extrop setup was chosen to capture the skew of the northern extratropical forcing that is seen in hist-aer relative to the more homogeneous hist-GHG ( Figure 1c). The idealized experiments were performed with the Hadley Centre Atmospheric Model version 3 (Pope et al., 2000) in both atmosphere-only (HadAM3) and slab ocean (HadSM3) configurations. The horizontal resolution is 2.5° × 3.75°; there are 19 levels in the atmosphere and the slab ocean has a thickness of 50 m. This model is one of the two used by Ceppi and Gregory (2019), and these experiments are the same as their − UNIF ET experiment for a single hemisphere at a time, except with a different forcing magnitude.
We also make use of tropical-only forcing experiments in HadSM3 and HadAM3 denoted as "tropical" in later figures. This is the same as the +UNIF T experiment in Ceppi and Gregory (2019), which involves a uniform forcing of 7 W m −2 over the tropics (30°S-30°N).

Investigating Radiative Feedback Differences in Terms of Stability Differences
The radiative feedbacks for historical aerosol and all historical forcings, relative to those from GHGs, are shown in Figure 2a, for each model analyzed and for the MMM. Colored bars show the all-sky net feedback parameter α, whilst markers show the CRE feedback parameter α CRE (minuses) and clear-sky feedback parameter α CS (pluses). In all cases, positive numbers mean less amplifying feedbacks, that is, a relatively larger upward radiative flux perturbation for positive T. The findings here show more amplifying feedbacks for hist-aer than hist-GHG in the MMM. On a model-by-model basis, hist-aer shows either significantly more amplifying or very similar feedbacks to hist-GHG (except for CNRM-CM6-1, the only model with a substantially less amplifying feedback for hist-aer). Neither α CS (correlation of 0.80 with allsky α, Figure S1b in Supporting Information S1) nor α CRE (correlation of 0.85 with allsky α, Figure S1d in Supporting Information S1) entirely explains the differences of all-sky α for aerosol (or the net historical forcing) compared to GHG forcing. Both correlations are highly significant (p < 0.001), which indicates that both clear-sky and CRE feedbacks are important in explaining the spread in α as neither α CS nor α CRE dominates the spread.
We propose that differences in radiative feedback across forcing agents may be explained in terms of different tropospheric stability responses and their impact on cloud and lapse-rate feedbacks. Figure 2b shows that hist-aer causes lower stability responses than hist-GHG across all models (except for CNRM-CM6-1). A greater increase in stability (as in GHG compared with aerosol) means more warming in the upper troposphere than at the surface, and hence a positive (less amplifying) lapse-rate feedback (Andrews & Webb, 2018;Ceppi & Gregory, 2019), with an additional minor contribution from relative humidity changes ( Figure 6 in Ceppi & Gregory, 2019). Furthermore, tropospheric stability is a key variable for cloud formation, with higher stability encouraging the formation of low boundary-layer clouds over marine regions (Andrews & Webb, 2018;Ceppi & Gregory, 2019;Ceppi et al., 2017;Zhou et al., 2016). Low clouds have little impact on outgoing longwave radiation due to their temperatures being similar to those at the surface. Since they reflect incoming solar radiation, however, low clouds have an overall positive upwards (cooling) effect on radiation (Hartmann, 1994). Positive forcings that increase stability will thus tend to promote low-level cloudiness and give a positive upwards radiative feedback that opposes the forcing. The feedbacks from increased low cloud, combined with clear-sky feedbacks, are why we expect a positive correlation between net α and the stability response S per unit global warming, which we refer to as dS/dT. Figure 2 (through its parts c-g) supports this. Considering all models and experiments together, there is a positive correlation between net α and dS/dT (0.59 for the AOGCM experiments, Figure 2c). Much of the spread in α among this set of models is related to stability, despite the implication of previous work that intermodel differences in climate feedback are dominated by cloud responses to mean SST warming (Ringer et al., 2014). This is still the case when feedbacks are broken down into α CS (Figure 2e) and, although to a lesser extent, α CRE (Figure 2g). We interpret the correlation in Figure 2e as being primarily driven by the linkage between stability and lapse-rate feedbacks (Andrews & Webb, 2018;Ceppi & Gregory, 2019).
By considering the difference in α between hist-aer and hist-GHG for each model, we remove part of the model spread in α , which affects both forcing agents equally, revealing the positive correlation (Figure 2d) between α and stability change in response to different forcing agents. Although the correlation across models is only moderate (0.52), it is nevertheless significant (p = 0.010), and the difference is also significant in the MMM according to estimated error bars. The lack of correlation in Figure 2h, both across models and in the MMM, despite such correlation in Figure 2f, suggests that the impacts of stability on clear-sky feedbacks are more robust than the impacts on α CRE for explaining differences in α between historical aerosols and GHGs. This may be because the relationship between S and CRE is not consistently simulated among climate models. Alternatively, it is possible that, contrary to the findings of Ceppi and Gregory (2019), global stability changes are not strongly physically linked to CRE in some of the models, and that regional changes in S would be a better explanatory factor.
The historical all-forcing case is dominated by responses to GHGs and aerosols (Smith et al., 2020). Therefore, differences between historical and hist-GHG experiments are mostly due to differences between hist-aer and hist-GHG. Both the feedback parameter ( Figure 2a) and the stability response ( Figure 2b) are greater for historical than for hist-GHG in the MMM. This results from combining hist-GHG and hist-aer responses, given that aerosols and GHGs forcing are of opposite sign (Appendix B in the online supporting information of Gregory & Andrews, 2016). A visual explanation of it can be found in Figure S2 in Supporting Information S1.
The results here are in agreement with findings from previous studies of a greater transient climate response (indicative of a less positive α) from aerosols (Marvel et al., 2016) and extratropical forcing (Rose & Rayborn, 2016;Rose et al., 2014;Shindell, 2014) compared to forcing from well-mixed GHGs. By contrast, Richardson et al. (2019) found no significant differences in feedback between the two kinds of aerosols (SO 4 and BC) and GHGs, regardless of whether they calculated ERF as in the present paper, or additionally correcting for the impact of land surface temperature adjustments (Andrews et al., 2021). There are several possible explanations for our disagreement, including the following. (a) Despite its statistical significance, the difference we find between feedbacks to the aerosol and GHG forcing may be specific to our selection of models, which is smaller than theirs (7, vs. 11 in PDRMIP models). (b) Historical aerosol and the 5 × SO4 forcing in PDRMIP might have important 6 of 9 differences in feedback, both because of different SO 4 concentrations and because of contributions of aerosols other than SO 4 (Myhre et al., 2014). (c) The step forcing experiments analyzed by Richardson et al. (2019) will be more representative of equilibrium responses than the historical simulations used here, which may affect the extent to which the feedbacks depend on the forcing agent.

Explaining Stability Differences in hist-aer in Terms of Extratropical Forcing
Next, we interpret the distinct radiative and stability responses to aerosols and GHGs in terms of the latitudinal distribution of forcing. Ceppi and Gregory (2019) demonstrated that positive tropical forcing tends to increase global stability per unit global surface warming, while positive extratropical forcing has the opposite impact (and vice versa for negative forcing). To understand why, we recall that the tropics are generally well coupled to the free troposphere, with the lapse rate closely following a moist adiabat due to moist convection (Flannaghan et al., 2014;Sobel, 2002). Consequently, tropical warming has a relatively large impact on free-tropospheric temperature. Mixing by atmospheric motions propagates the warming signal to the extratropical free troposphere, stabilizing the atmosphere there (Figures 3b and 3d). Conversely, positive forcing in the extratropics is expected to decrease stability, since surface temperature in the extratropics is more weakly coupled to the free troposphere. The effects of extratropical surface forcing tend to be more confined near to the surface (Figures 3a and 3c), and since this forcing acts on a region that is (on average) climatologically stable, the stability response is similar to that found for warming in other stable regions such as in the tropical South-East Pacific (Andrews & Webb, 2018). This effect can be seen by comparing air temperature changes in the hist-aer and hist-GHG cases ( Figure 3e). (Note that aerosol forcing is negative and causes a surface cooling, but the patterns in Figure 3 are normalized by regression against global mean surface temperature change.) This reasoning could explain why the hist-aer case gives a less positive stability response per unit surface warming than the hist-GHG case. The skew of forcing toward the extratropics in the hist-aer case (blue line in Figure 1c) means that a relatively larger fraction of the surface temperature response is in vertically decoupled regions, leading to the smaller dS/dT than in hist-GHG. In support of this hypothesis, we note that the pattern of tropospheric temperature change in the HadSM3 nh_extrop experiment compared to the 2 × CO2 experiment (Figure 3g) is similar to the difference between hist-aer and hist-GHG (Figure 3e). The pattern from tropical-only forcing (Figure 3d) shows the propagation of warming to both the tropical and extratropical free tropospheres in accordance with an increase to stability as seen in Figures 2c-2g, although stability is further increased over the tropics than in the more homogeneous 2 × CO2 case (Figure 3h). Figure 2e shows that global-mean dS/dT is negative for both nh_extrop and sh_extrop, whereas it is positive in nearly every historical forcing experiment. This , and relative to hist-GHG (e-f). Also shown are the results from the HadSM3 NH extratropical forcing (c) and tropical-only forcing (d), and these relative to the HadSM3 2 × CO2 experiment (g-h). Note that the x-axis is scaled by geographical area. 7 of 9 difference is probably related to the absence of tropical forcing in the idealized cases. That the negative dS/dT occurs for forcing in both hemispheres suggests that the essential characteristic is that the forcing is extratropical, rather than hemispheric. The historical all-forcing case shows the opposite pattern to hist-aer when compared to hist-GHG (Figures 3e and 3f), indicating that the aerosol has a similar effect on stability whether applied independently or jointly with GHGs.
To corroborate this reasoning, we consider the MMM zonal means of feedbacks and climate responses in Figure 4. There is less meridional contrast in response to hist-GHG than hist-aer (top row). Subtracting the responses from hist-GHG (bottom row), we see a positive NH temperature response per unit global warming in hist-aer relative to hist-GHG, and the opposite in the SH. This anticorrelates with the stability response per unit surface warming, which in turn correlates (in low latitudes) with the radiative response. The radiative response is then finally well explained, in terms of pattern, by the combination of clear-sky and CRE feedbacks.
Just as the global average feedbacks for the historical case are more positive than those for hist-GHG, the zonal-mean curves for hist-GHG generally lie between those for the net historical and hist-aer cases (top row of Figure 4), again consistent with the expected effect of aerosols. Likewise, the difference between historical and hist-GHG responses (bottom row of Figure 4) can be interpreted as representing the effect of aerosols, but with the sign reversed (see Text S1 in Supporting Information S1).

Summary and Conclusions
Our analysis of AOGCM historical experiments from CMIP6 (including new experiments which allow forcing to be diagnosed) reveals that climate feedback is more strongly amplifying (greater climate sensitivity) in response to anthropogenic aerosol forcing than greenhouse-gas (GHG) forcing. This difference is shown and is statistically significant in the MMM, though only seven AOGCMs have so far provided the required historical experiments and variables for this analysis, so it would be useful to repeat it with more. Our finding is consistent with those from past studies that also found greater climate sensitivity to aerosols than GHGs (Marvel et al., 2016;Shindell, 2014;), but appears inconsistent with the recent study of Richardson et al. (2019). Further work is needed to explain the disagreement, which may relate to differences in the details of the prescribed aerosol forcing or the timescales of calculated feedbacks.
Furthermore, we find that the difference in (positive-stable) the net top-of-atmosphere radiative feedback parameter for aerosols and GHG forcing is positively correlated across AOGCMs with a difference in the response of tropospheric stability to the two kinds of forcing. We propose that the difference arises from the different latitudinal distributions of the forcing. An idealized slab model experiment with uniform surface forcing confined to the Northern extratropics qualitatively reproduces the near-surface extratropical temperature change that differentiates the historical aerosol experiment from the historical GHG experiment. The shallower extratropical temperature change in the former is explained by the lower proportion of forcing in the tropics, where the surface is relatively strongly coupled to the free troposphere by deep convection (Flannaghan et al., 2014;Sobel, 2002), compared to the higher proportion of forcing in the extratropics, where the coupling is weaker and the effect of forcing more confined to the surface.
Thus a positive extratropical forcing causes a near-surface warming, which reduces tropospheric stability, whereas a positive tropical forcing has less effect on stability. A reduction in stability tends to reduce low-level cloudiness, which gives an anomalously negative (more amplifying) shortwave cloud feedback, whilst it also induces an anomalously negative clear-sky feedback. In this way, the latitude of forcing is linked to the radiative feedback it produces. Historical aerosol forcing is negative, so the signs of temperature and stability change are reversed, but the feedback parameter, sensitivity of stability (change per unit warming), and hence the correlation with the feedback parameter have the same sign for either sign of forcing: extratropical forcing tends to give higher climate sensitivity. This link accords with previous studies that have highlighted the impact of forcing patterns on radiative feedbacks (Ceppi & Gregory, 2019;Rose & Rayborn, 2016;Rose et al., 2014;Winton et al., 2010).
Historical climate change is dominated by GHG forcing. Hence the net feedback simulated in the historical experiments with all forcings is nearer to that for GHG than for anthropogenic aerosols. The effective climate sensitivity for historical forcing is slightly smaller than for historical GHG forcing (the magnitude of α is larger), because combining negative hist-aer forcing with positive hist-GHG forcing reduces the warming by more as a proportion than it reduces the net forcing, due to the greater sensitivity to aerosol forcing. We find also that some of the spread across AOGCMs in the climate sensitivity to GHG forcing is also correlated with the response of tropospheric stability to forcing; this aspect is intriguing and requires further investigation.

Data Availability Statement
Data that support the findings of this study are available online. The data from Coupled Model Intercomparison Project phase 6 (CMIP6) experiments can be found on the ESGF CEDA node available at esgf-index1. ceda.ac.uk. The data from the HadSM3 and HadAM3 experiments can be found at https://doi.org/10.6084/ m9.figshare.17197748 and https://doi.org/10.6084/m9.figshare.17182907.