Cloud Responses to Abrupt Solar and CO2 Forcing: 1. Temperature Mediated Cloud Feedbacks

There are many uncertainties in future climate, including how the Earth may react to different types of radiative forcing, such as CO2, aerosols, and even geoengineered changes in the amount of sunlight absorbed by Earth's surface. Here, we analyze model simulations where the climate system is subjected to an abrupt change of the solar constant by ±4%, and where the atmospheric CO2 concentration is abruptly changed to quadruple and half its preindustrial value. Using these experiments, we examine how clouds respond to changes in solar forcing, compared to CO2, and feedback on global surface temperature. The total cloud response can be decomposed into those responses driven by changes in global surface temperature, called the temperature mediated cloud feedbacks, and responses driven directly by the forcing that are independent of the global surface temperature. In this paper, we study the temperature mediated cloud changes to answer two primary questions: (a) How do temperature mediated cloud feedbacks differ in response to abrupt changes in CO2 and solar forcing? And (b) Are there symmetrical (equal and opposite) temperature mediated cloud feedbacks during global warming and global cooling? We find that temperature mediated cloud feedbacks are similar in response to increasing solar and increasing CO2 forcing, and we provide a short review of recent literature regarding the physical mechanisms responsible for these feedbacks. We also find that cloud responses to warming and cooling are not symmetric, due largely to non‐linearity introduced by phase changes in mid‐to‐high latitude low clouds and sea ice loss/formation.


Introduction
As the climate warms due to the radiative forcing created by increasing CO 2 and other heat trapping gasses, one anticipates that many aspects of the climate system will experience change.Some of these changes will further impact the Earth radiation balance, creating feedback loops.Radiative feedbacks related to changes in cloud properties and cloud amount have been identified as the largest source of uncertainty (spread) in projections of future climate (e.g., Sherwood et al., 2020;Zelinka et al., 2020).To better understand cloud responses to forcing, in this paper we examine cloud feedbacks which occur following an abrupt increase and decrease in solar radiation and contrast these solar-driven changes with those caused by abrupt changes in CO 2 concentrations in several climate models.Studying the clouds' response to warming and cooling from abrupt solar and CO 2 forcing is an idealized framework intended to create a strong cloud response.This nominally makes it easier to recognize and understand the underlying mechanisms that contribute to the cloud response.This includes mechanisms that are important for realistic climate futures, such as the effects of aerosol forcing and some proposed geoengineering schemes which intend to diminish the amount of sunlight absorbed by the earth through efforts such as stratospheric aerosol injection or marine cloud brightening (e.g., Hulme, 2012;Keith et al., 2016;Kravitz et al., 2021;Niemeier et al., 2013;Shepard et al., 2009;Visioni et al., 2021).Using idealized changes in solar forcing isolates the adjustments and feedbacks due to insolation changes from other effects that can occur with more sophisticated forcing schemes such as, for example, changes in atmospheric chemistry.
Radiative feedbacks (including cloud feedbacks) are often quantified in climate model simulations by the relationship between the top-of-atmosphere radiative flux and global-mean surface temperature change (compared to a simulation of the pre-industrial climate), and this relationship is often approximated as a linear response (Gregory et al., 2004).The linear model separates the total change into a temperature mediated change (the cloud change per degree of global mean temperature anomaly) and an adjustment that occurs directly due to the forcing agent (in our case from changes in insolation or atmospheric CO 2 concentration).In this paper we focus on the temperature mediated component of the cloud response to solar and CO 2 forcing, while in a companion paper (Part II, Aerenson et al., 2023), we focus on the cloud changes that are a direct adjustment to the forcing agents themselves and are nominally independent of global mean surface temperature change.
Often, the temperature mediated changes in top-of-atmosphere radiative flux are simply called radiative feedbacks, or when they are due to clouds, simply cloud feedbacks.Naturally, cloud feedbacks that occur from CO 2 increase have become a widely studied topic (e.g., Andrews & Ringer, 2014;Dufresne & Bony, 2008;Sherwood et al., 2020;Taylor et al., 2007;Zelinka et al., 2020).Here, we analyze cloud feedbacks in model simulations produced as a part of the third phase of the Cloud Feedback Model Intercomparison Project (CFMIP3; Webb et al., 2017) which is a part of the sixth phase of the Coupled Model Intercomparison Project (CMIP6; Eyring et al., 2016).Specifically, in CFMIP3 a pair of model simulations were performed in fully coupled climate models initialized from the pre-industrial climate, and then perturbed by suddenly increasing or decreasing the insolation by 4% (hereafter solp4p and solm4p respectively).We compare these two abrupt-solar experiments with simulations in which there is an abrupt quadrupling of the CO 2 concentration (hereafter 4xCO2) and halving of CO 2 (hereafter 0p5xCO2) that were also produced as a part of the CMIP6 experiments (Eyring et al., 2016).An increase of the solar constant by 4% is designed to (loosely) match the radiative forcing of a quadrupling of atmospheric CO 2 , and as we will see, the experiments do produce a similar change in the mean global temperature.In contrast, a reduction of the solar constant by 4% does not match closely with the radiative forcing from 0p5xCO2, and again, as we will see, there are differences in the feedbacks between the solm4p and 0p5xCO2 related to the amplitude of the forcing.
In this article, we focus on the temperature mediated cloud changes to answer two primary questions: (a) How do temperature mediated cloud feedbacks differ in response to abrupt changes in CO 2 and solar forcing?And (b) Are there symmetrical (equal and opposite) temperature mediated cloud feedbacks to an increase and a decrease of radiative forcing?We also review recent literature on the physical mechanisms responsible for the temperature mediated cloud changes, as well as examine changes in atmospheric circulations and several atmosphere and surface variables that influence cloud occurrence and radiative properties, sometimes called cloud controlling factors, in the four model experiments.
Cloud feedbacks resulting from CO 2 increase have been more widely studied than those from solar forcing (e.g., Andrews & Ringer, 2014;Dufresne & Bony, 2008;Sherwood et al., 2020;Taylor et al., 2007;Zelinka et al., 2020).Recently, Kaur et al. (2023) performed coupled model experiments of an abrupt doubling of CO 2 and a 2% increase of the solar constant with a single model.They found differences in the temperature mediated feedbacks were caused by the differences in geographic distributions of each forcing.CO 2 increase has an instantaneous radiative forcing that is homogeneously distributed across the globe and is equal during all seasons.Solar forcing differs in that it is strongest in regions with more incident sunlight, so the forcing is greatest in the tropics and during the summer season.This causes a difference in the warming patterns from solar and CO 2 forcing, where the tropics are warmer, and the poles are cooler following a 2% increase in solar forcing compared with a doubling of CO 2 .They also find differences in the cloud feedbacks in their experiments, such as finding greater decrease of high cloud and lesser decrease in low clouds in the solar forcing experiment than CO 2 .This study was based on the output of a single climate model, and cloud feedbacks have notoriously large inter-model spread (Sherwood et al., 2020;Soden & Held, 2006;Zelinka et al., 2020).Our study expands upon this analysis by using several models and focusing specifically on the cloud changes from solar and CO 2 forcing.
Feedbacks following solar forcing have received comparatively less attention than those following CO 2 due to the importance of CO 2 in anthropogenic climate change.Feedbacks following solar forcing have received comparatively less attention than those following CO 2 due to the importance of CO 2 in anthropogenic climate change.However, due to the fact that solar and CO 2 forcing have different spatial distributions (solar forcing is stronger in the tropics and CO 2 forcing is globally homogeneous) there are similarities between the work presented here and previous studies comparing the climate's response to forcing agents with different spatial distributions.Salvi et al. (2022) used model simulations forced with historical greenhouse gas, and historical aerosol forcing, and found that (a) historical aerosol forcing (which is most concentrated in the midlatitudes) has a greater impact on low clouds than CO 2 forcing (which is globally uniform) and (b) the geographic distribution of aerosol forcing is important in understanding historical cloud feedbacks.On a related note, Rose et al. (2014) used climate model simulations forced by ocean heat uptake in the tropics, ocean heat uptake in the extratropics, and 4xCO2, and found that forcing applied in the extratropics yields more positive cloud feedbacks than 4xCO2, and 4xCO2 yields more positive cloud feedbacks than forcing applied to the tropics.So, both Salvi et al. (2022) and Rose et al. (2014) found that cloud feedbacks are more positive when the forcing is concentrated in the extratropics, which suggests that we should expect more positive feedbacks from 4xCO2 than solp4p.However, as we will see, this is opposite of what we find in Section 3.2.
The second question we address is on the differences in cloud feedbacks from a warming and cooling climate, which has received far less attention than characterizing the effects of warming the climate.Understanding the climate's response to abrupt negative forcing is an essential step to understanding more realistic forcing that cools the climate, such as volcanic aerosol forcing, and various methods of geoengineering that aim to reduce the amount of sunlight absorbed by the Earth through techniques such as stratospheric aerosol injection or marine cloud brightening (e.g., Hulme, 2012;Keith et al., 2016;Niemeier et al., 2013;Shepard et al., 2009).
There has been some previous research on the climate's response to warming and cooling.Chalmers et al. (2022) studied the climate's response to abruptly increasing and decreasing CO 2 and found that patterns in Tropical Pacific sea-surface temperature changes differ notably between increasing and decreasing CO 2 (warming and cooling) experiments and have a large impact on atmospheric circulation and teleconnections across the globe.Additionally, they found two differences between warming and cooling in the high latitude oceans.First, there is liquification and glaciation of mixed-phase clouds from warming and cooling respectively occurring at different latitudes, and second, there is sea-ice reduction and growth in the warming and cooling occurring at different latitudes which is likewise amplified by cloud feedbacks.There has also been extensive research on the differences in temperature mediated feedbacks from different amounts of global warming, such as studies by both Bloch-Johnson et al. (2021) and Zhu and Poulsen (2020).Similar to Chalmers et al. (2022), they find that cloud feedbacks change depending on the amount of abrupt forcing due to the mixed-phase cloud feedback and sea-icerelated feedbacks because both of these feedbacks occur at different latitudes depending on the global mean temperature change.Additionally, Zhu and Poulsen (2020) identified a nonlinear change in shortwave cloud feedbacks at low latitudes, due to the nonlinear change in the moisture gradient between the boundary layer and free troposphere that enhances low cloud thinning through dry air entrainment from the free troposphere.As such, we anticipate that the temperature mediated cloud changes may differ from cooling and warming, especially in regions where there are changes in sea-ice, mixed-phase clouds, and low-latitude low clouds.This paper is organized as follows: the model output and method for calculating the temperature mediated cloud changes and the associated radiative effect are summarized in Section 2. Results are presented in Section 3 and are split into six subsections.The first subsection examines the cloud responses in the solp4p, 4xCO2, solm4p, and 0p5xCO2 experiments and the second subsection examines the radiative effect of the cloud changes.The remaining four subsections are dedicated to exploring differences in cloud controlling factors.In Section 4, we review recent literature regarding the physical mechanisms related to the cloud controlling factors examined in Section 3.This is followed by additional discussion and conclusions in Section 5.

Methods and Theory
When an abrupt forcing is imposed on the climate, the cloud changes are often decomposed into two components: those driven by changes in global mean surface temperature (which are called temperature mediated change) and those that are independent of the global mean surface temperature (which are called the adjustments).This is described by Equation 1, where C(θ,ϕ,t) represents the cloud amount anomaly at a given latitude, longitude, and time in the simulation, A(θ,ϕ) is the adjustment to the forcing change at a certain latitude and longitude, ∆T(t) is the global mean surface temperature anomaly at a given time, M(θ,ϕ,T(t))∆T(t) is the temperature mediated component of the cloud anomaly, and ε represents internal variability which causes cloud changes which are due to neither the global mean temperature change nor adjustments.In this paper, we are concerned with calculating the temperature mediated cloud changes M(θ,ϕ,T(t)).

C(θ,ϕ,t)
We follow a frequently used framework that approximates the temperature mediated response by a linear relationship to global mean surface temperature, such that M is a constant in temperature (and time) M(θ,ϕ,T(t)) ≈ M (θ,ϕ) (for more details see Gregory et al., 2004).The temperature mediated component of cloud changes (M(θ,ϕ)) is calculated via a least squares linear regression of annual mean changes in cloud amount onto the annual and global mean surface temperature anomaly, using years 10-150 following the abrupt forcing.In this work we often do such to calculate temperature mediated changes of specific cloud types or categories, and at specific locations or globally averaged.The first 9 years are excluded because shortly following the forcing the linear model does not fit well to the simulated data due to a combination of the internal variability and the inherently non-linear nature of the climate's response to abrupt forcing (e.g., Andrews & Ringer, 2014;Armour et al., 2013;Williams et al., 2008).By using 140-year regressions and excluding the first 9 years of simulation we expect that internal variability has little contribution to the temperature mediated cloud changes we calculate.As such Equation 1 is simply reduced to a linear model where cloud changes depend on the adjustment to forcing (A), surface temperature change (∆T ), and the temperature mediated term (M).
In this paper, we focus on the temperature mediated term calculated following abrupt solar and CO 2 forcing, as well as the temperature mediated response to warming and cooling.In Part II, we focus on the cloud adjustments to solar and CO 2 forcing.
To perform a comparison of cloud changes across models this study also makes extensive use of the International Satellite Cloud Climatology Project (ISCCP) satellite simulator embedded into the climate models (Klein & Jakob, 1999;Webb et al., 2001).Specifically, the ISCCP simulator produces cloud-top-pressure (CTP) and optical depth joint histograms of cloud occurrence that are directly comparable across models, and consistent with the radiation scheme within each model (Bodas-Salcedo et al., 2011).Zelinka et al. (2012a) have created cloud radiative kernels to compute longwave (hereafter LW) and shortwave (hereafter SW) fluxes from the ISCCP histograms.Using the radiative kernels, Zelinka et al. (2013) have examined cloud adjustments and temperature mediated responses to 4xCO2 simulations from a collection of CMIP5 models.The findings from Zelinka et al. (2013) include a temperature mediated increase in cloud-topheight for high-altitude clouds at nearly all locations that leads to a strong LW temperature mediated feedback, a decrease in low-altitude clouds equatorward of 60°latitude causing a positive SW feedback, and an increase in high latitude low-altitude cloud optical depth leading to negative SW cloud feedback.Here we undertake a similar analysis but applied to the solar forcing experiments in addition to the 4xCO2 and 0p5xCO2 experiments performed by the current generation of climate models used in CMIP6.In order to understand the radiative impact that temperature mediated changes of cloud cover fraction, cloud-top-pressure (CTP), and cloud optical depth (τ) have on top-of-atmosphere radiation balance (feedbacks), we perform a decomposition of the kernel-derived radiative effect into the radiative anomalies caused by each type of cloud change (as well as a small residual).This is done by first calculating the mean or base-state cloudiness of each model by averaging its ISCCP simulator CTP-OD histogram output from its piControl simulation over a long period of time, and then subtracting the base-state histogram from the ISSCP simulator CTP-OD histogram output of each abrupt forcing experiment in each year to obtain histograms of cloud anomalies.The histograms of cloud anomalies are then reweighted so the cloud anomalies have the same distribution of heights and/or optical depths as clouds in the base-state histogram (see equations 1 through 5 in Zelinka et al. (2012b)).Finally, the reweighted cloud anomalies are multiplied by the SW and LW radiative kernels to determine the radiative effect of the cloud anomalies (with distribution of heights and/or optical depths forced to match the base-state).

Model Experiments
In total five modeling centers performed the solp4p experiment, and four performed the solm4p.The models are listed in Table 1, along with the experiments we make use of here, and a primary citation for each model.CESM2 did perform all the simulations, however there is no ISCCP simulator output for the 4xCO2 simulation, so CESM2 is excluded from the portion of the analysis where the ISCCP simulator output is compared between the solar and CO 2 forced experiments.

Results
In this section we present the results showing the temperature mediated cloud changes in model simulations with solar and CO 2 forcing and examine the cloud radiative effect the temperature mediated cloud changes have on top-of-atmosphere radiation using cloud radiative kernels.We additionally include in this section an examination of changes in atmospheric circulations and several atmosphere and surface variables that influence cloud occurrence and radiative properties.

Temperature Mediated Response in Cloud Properties to Solar and CO 2 Forcing
The primary focus of this section is to examine how cloud properties respond to global temperature change that is forced by increases and decreases of the solar constant and CO 2 concentration.Figure 1 shows the change in global total cloud amount (summed over all optical depth and CTP bins of the ISCCP simulator) that occurs in each simulation plotted against the global mean temperature anomaly from the simulation of the pre-industrial climate.Colors denote the individual models.

Abrupt-solp4p
Abrupt-solm4p Abrupt-4xCO2 Abrupt-0p5xCO2 Note.CESM2 did not produce ISCCP simulator output from the 4xCO2 and piControl experiments, so it is generally excluded from the analysis of cloud attributes, but the model is used to characterize other physical responses to the forcing.All ECS values are provided by Meehl et al. (2020).
Figures 1a and 1b show that in the 4xCO2 and solp4p simulations there is a reduction of cloudiness as the climate warms in all but the MRI-ESM2-0 simulations (which shows near-zero temperature mediated change).Similarly, Figures 1c and 1d show that there is an increase of cloudiness as the climate cools in all but the MRI-ESM2-0 simulations.The slopes of each model in Figure 1 (and the multi-model mean) are listed in Table 2, and the intercepts are available in Supporting Information S1.While there is substantial spread in the response of different models, the temperature mediated response for each individual model (in the global mean) is quite similar in both the 4xCO2 and solp4p experiments.However, the solp4p temperature mediated change in cloud amount is persistently slightly more negative than the 4xCO2.
The largest source of spread in equilibrium climate sensitivity (ECS) is associated with cloud feedbacks (Sherwood et al., 2020;Zelinka et al., 2020), so it is worth a brief note to compare the cloud fraction changes with each models' ECS.In Table 1 we list the ECS of each model, and we find that ECS appears to have a monotonic relationship with changes in cloud amount, where the lowest ECS is associated with the least amount of cloud reduction, and the highest ECS is associated with the largest amount of cloud reduction.CanESM5 is the exception, which has fairly little cloud reduction with warming, but also the highest ECS.We note that one does not expect a direct linear relationship between ECS and temperature mediated changes in cloud fraction, because there are numerous other factors that contribute to ECS and there can be differences in the base state of different models.Nonetheless, one expects that models with a larger temperature mediated change in cloud fraction will tend to have larger ECS.  2. Later in this paper we examine the cloud radiative feedbacks in each model experiment, and will show that in the case of CanESM5 it is the changes in cloud-to-pressure that cause it to have strongly positive cloud feedbacks (and thus higher ECS), even though this model experiences less change in cloud fraction than others.
In the solm4p there is one fewer model than the other experiments.Yet even on a model-to-model basis, it is apparent that there are greater differences in the magnitude of the global mean temperature mediated response between cooling and warming than between solar or CO 2 forcing.For example, we can see from Table 2 that in CanESM5, the global mean temperature mediated cloud change (slope of the line in Figure 1) in solm4p is more than double that of either 4xCO2 or solp4p, and in MRI-ESM2, the temperature mediated cloud change is of opposite sign in solm4p and 0p5xCO2 from the solp4p and 4xCO2 experiments.The 0p5xCO2 also exhibits differences from the solm4p (which may be due in part to the smaller forcing that halving CO 2 imposes on the climate as compared with a 4% reduction of the solar constant).
Across all four experiments HadGEM3-GC31-LL produces the greatest temperature mediated cloud reduction, IPSL-CM6A-LR has the second most cloud reduction, followed by CanESM5, and MRI-ESM2-0 either has the smallest temperature mediated cloud reduction (in the case of solp4p and 4xCO2) or a temperature mediated cloud increase (in the case of solm4p and 0p5xCO2).
The geographic distributions of the temperature mediated cloud responses are shown in Figures 2 and 3 for the 4xCO2 and solp4p experiments respectively (the two warming experiments).In this figure, the temperature mediated response of cloud fraction is calculated at each grid cell, for nine pressure and optical depth categories using the ISCCP simulator.Specifically the cloud optical depth is broken into three ranges: optically thin (τ ≤ 3.6), medium (3.6 < τ ≤ 23), and thick (τ > 23) clouds, and the CTP is likewise broken into three CTP ranges: low (CTP ≥ 680 hPa), mid-level (680 hPa > CTP ≥ 440 hPa), and high (CTP < 440 hPa) cloud.The multi-model means for the 4xCO2 experiment is given in the top nine panels, and those for the solp4p experiment are given in the lower nine panels.Stippling indicates grid cells where three out of the four models agree on the sign of the change.The data are plotted on a 1-degree grid (sampled for each model using a linear interpolation from the models' innate grid).

Temperature Mediated Response of Low Clouds to solp4p and 4xCO2
The greatest decrease in cloud fraction with warming occurs in the optically-medium low clouds, where there is large reduction in cloud amount over most oceanic regions equatorward of about 60°latitude.The reduction is especially large in regions that are characterized by relatively cool sea-surface temperatures and large scale subsidence that supports the formation of stratocumulus clouds (Wood, 2012; see also Figures 7 and 8 of this paper where stratocumulus regimes are either marked, or can be seen in climatological subsidence rates).In the global mean there is a greater reduction of optically-medium low clouds in the solp4p than the 4xCO2 by 0.05%/ K, which is the largest difference between the two forcings of any cloud type, and is the primary contributor to the difference noted in Figure 1 and Table 2.   Poleward of 60°latitude, on the other hand, optically medium and thick low-level clouds increase, which is opposite of the change that occurs closer to the equator.This is accompanied by a decrease in optically thin lowlevel clouds.Reduction of stratocumulus clouds with warming and increases in the optical depth of mid-to-high latitude low cloud has been a well-documented response to increasing CO 2 in both global climate models and process models (Bjordal et al., 2020;Sherwood et al., 2020;Zelinka et al., 2013), and we discuss the associated physical mechanisms in more detail in Section 4.

Temperature Mediated Response of Mid-Level Clouds to solp4p and 4xCO2
The temperature mediated cloud changes in the mid-level cloud category are generally more muted, but they are often more consistent globally such that the global mean changes are comparable to the low and high-level cloud categories.There is nonetheless reduction of midlatitude mid-level clouds that is consistent across models.This reduction is most pronounced in the thin cloud category at mid to high latitudes, however there is a weak reduction in optically medium and thick mid-level clouds that is consistent between models in some regions (including in the Southern Hemisphere midlatitudes), but there is poor model agreement in most regions.There is also a slight increase in optically medium mid-level cloud in the Peruvian and Californian stratocumulus regions of some models, albeit with poor model agreement, which suggests that in some models there is a rising of the stratocumulus layer with increase in global mean temperature.The role of mid-level clouds in climate feedbacks has received less attention than that of high or low-level clouds (Sherwood et al., 2020), and we will return to this in Section 4.

Temperature Mediated Response of High Clouds to solp4p and 4xCO2
In the global mean there is a decrease in optically thin high-level cloud and increase in optically medium and thick high-level cloud in both the solp4p and 4xCO2 (the two warming scenarios).This signifies a global mean thickening of high clouds.This high-cloud thickening is especially pertinent over the Southern Ocean.The mechanisms associated with the optical depth change in the Southern Ocean are addressed in Section 4.3.In the equatorial Pacific, the temperature mediated response of high clouds in both the solp4p and 4xCO2 experiments form a clear dipole, with increasing high cloudiness over the central and eastern portion of the equatorial pacific and a decrease in high cloud over the western pacific and maritime continent.This pattern is consistent with a weakening Walker circulation.There is also a northeastward shift of the South Pacific Convergence Zone (SPCZ), and a reduction of high cloud over the Amazon and central America, which are responses that have been observed to be correlated with the phase of the El Nino Southern Oscillation (ENSO) (Adames & Wallace, 2017).Sea-surface temperature variability of ENSO, and the atmospheric variability of the Walker circulation are strongly coupled phenomena (e.g., Battisti et al., 2019;Bjerknes, 1969), as such we find these high cloud changes in the Tropical Pacific to be linked with the change in SST pattern with warming (see Section 3.4).

Comparing the Temperature Mediated Response of Clouds to solm4p and 0p5xCO2
While the cooling patterns from solar and CO 2 forcing share some similarity, it is noteworthy that there are larger differences in the cloud response between the two cooling experiments (0p5xCO2 and solm4p) than between the two warming experiments.Specifically, the global mean response for optically medium mid-level and low-level clouds is about half as strong in the 0p5xCO2 experiment ( 0.04% and 0.05%/K respectively) as compared with the solm4p ( 0.1% and 0.11%/K).In the 0p5xCO2 there is a strong increase of low clouds across the Equatorial Pacific with weak changes in the adjacent subtropics, while in the solm4p, there is a weak response in the tropics with poorer model agreement (note less stippling in the solm4p response).There is also an increase in low cloudiness in the Northern Pacific of the solm4p experiment that does not occur to the same extent in the 0p5xCO2 experiment.As previously noted, the solm4p and 0p5xCO2 experiments have different amounts of forcing, and thus different amounts of temperature change.There have been studies on differences in temperature mediated cloud changes from various amounts of CO 2 forcing, such as Bloch-Johnson et al. ( 2021) who found that models with the greatest nonlinearity in their response to different amounts of forcing, have the greatest differences in the temperature mediated cloud changes with different amounts of warming (i.e., M(θ,ϕ) from Equation 2 is different depending on ∆T ).Additionally, sub-polar low clouds contribute to this non-linearity through the saturation of mixed-phase clouds with warming, or in our case with cooling (Bjordal et al., 2020).Thus, we cannot easily separate the differences between the two cooling experiments that arise due to the different forcing mechanism (solar vs. CO 2 ) and the differences in the amount of cooling.For this reason, the comparison between the effects of solar and CO 2 forcing focuses on the warming experiments (solp4p and 4xCO2), while the cooling experiments (solm4p and 0p5xCO2) are used to compare between the temperature mediated response to warming versus cooling.

Comparing the Temperature Mediated Response of Clouds to Warming and Cooling
While similar in some respects, the pattern of low-cloud response is not quite the same between the warming and cooling experiments.In the warming experiments there is a transition between negative (orange) and positive (purple) temperature mediated cloud response in optically medium and optically thick low-clouds near 60°l atitude (in both hemispheres) while in the cooling experiment the transition occurs near 40°latitude; and likewise, the increase in low-clouds over the northern portions of North America and Eurasia is stronger in the cooling experiments.The reduction of optically thin low-level cloud is also about 20°closer to the poles in the warming experiments than the cooling experiments.In fact, in the cooling experiments, the response in optically thin midlevel cloud is the largest of the nine categories, whereas in the warming experiments it is the optically medium low clouds which have the largest change.Turning our attention to the high-level clouds, in the Tropical Pacific there is a very different response in the cooling experiments than occurs in the warming experiments.In the cooling experiments, there is a decrease in high clouds (a positive temperature mediated response) throughout the equatorial pacific, and an increase in high clouds in the subtropics (negative temperature mediated response).This pattern is consistent with a strengthening of the Hadley circulation and the associated intertropical convergence zone (ITCZ) and differs from the warming experiments which show a pattern of change more consistent with a change in the Walker circulation.The circulation changes are further discussed in Section 3.3.

Top of Atmosphere Cloud Radiative Feedback
Thus far we have examined changes in clouds that occur in models forced with abrupt changes of insolation and CO 2 concentration.The cloud changes previously described alter the Earth radiation budget, and thereby feedback on the climate to enhance or diminish the impact of the forcing.The cloud radiative anomaly can be calculated in many ways such as directly from top-of-atmosphere radiation output as the cloud radiative effect (Su et al., 2010), Partial Radiative Perturbation (Taylor et al., 2007), or cloud radiative kernels (Zelinka et al., 2012a).
Here we use the latter because it provides the most direct link between cloud changes and radiation.Note however that cloud radiative anomaly from radiative kernels is calculated directly from changes in the underlying cloud distribution, thus it is independent of cloud masking (see Zelinka et al., 2013).To account for how solar forcing impacts the amount of SW radiation reflected by a cloud (even if the clouds themselves are unchanged), the shortwave kernels are multiplied by 1.04 and 0.96 for the solp4p and solm4p simulations respectively.The effect of this adjustment is small and has no impact on the conclusions.The global mean cloud radiative anomaly change due to temperature mediated cloud changes (often referred to as the cloud feedback) is shown in Figure 6, and to further connect the cloud changes to the induced cloud feedback, we separate the cloud radiative feedback into three categories: those due to changes in cloud amount, CTP, and cloud optical depth using the kernel decomposition method developed by Zelinka et al. (2012b).In the Supporting Information S1 we show and discuss in detail the geographic pattern of cloud feedbacks in both the multi-model mean, and for each simulation.
Figure 6 shows that the LW temperature mediated changes are similar between the 4xCO2 and solp4p experiments in the global mean (which is not surprising given the strong similarity in the clouds responses shown in Figure 2).In the warming experiments all models exhibit negative LW feedbacks associated with changes in cloud fraction (which is indicative of cloud fraction reduction allowing more LW flux emitted from the surface to reach space in the global mean), and stronger positive LW feedbacks associated with changes in CTP (which is indicative of clouds forming higher in the atmosphere).In fact, all models performing all four experiments produce a positive LW CTP feedback, which is expected given that rising cloud tops with warming (causing a positive feedback) is a consistent response across models and is fairly well constrained by theory (Hartmann & Larson, 2002;Zelinka & Hartmann, 2010).The LW feedback associated with cloud fraction is consistently more negative across models in the solp4p experiment than the 4xCO2.The differences are relatively small compared to the intermodel spread of the feedbacks, but they are consistent across all models.
The cloud fraction changes are (by a wide margin) the largest contribution to SW temperature mediated feedbacks in both warming experiments.All models have positive SW cloud fraction feedbacks, which is consistent with the overall reduction of cloud amount that occurs in all models in Figure 1, because less cloud allows for more sunlight to reach Earth's surface.All models have positive total SW cloud feedbacks in both warming experiments.Similar to the LW cloud fraction feedback, in the SW there is consistently stronger feedback in the solp4p than the 4xCO2.Although the feedback patterns are quite similar in the two experiments, the total temperature mediated low-cloud reduction is slightly larger in the solp4p experiment which allows for more insolation to reach Earth's surface in solp4p than 4xCO2, this point is discussed further in Section 4.
The lower panel of Figure 6 shows the cloud feedback decomposition for the solm4p and 0p5xCO2 experiments (the two cooling scenarios).There are relatively large differences between the two cooling experiments, as compared to the warming experiments.The 0p5xCO2 experiment produces less feedback associated with change in cloud fraction in the global multi-model mean when compared with either the warming experiments or the solm4p.In the solm4p experiment the SW cloud fraction component is of similar amplitude to the SW cloud fraction component of the 4xCO2 and solp4p experiments.In both cooling experiments there is a relatively strong negative SW feedback from cloud optical depth change (meaning clouds become thinner with cooling in the global mean), which contrasts the warming experiments which have positive SW cloud optical depth feedback (meaning clouds also become thinner with warming in the global mean).Figure 5 shows that in the solm4p experiment the negative SW optical depth feedback is largely due to mid-to-high latitude cloud changes (which we later suggest are related to cloud phase partitioning), and in the warming experiments the positive optical depth feedback is somewhat lesser in magnitude, and mostly occurs in the tropics and Southern Hemisphere midlatitudes, where there is a decrease in low and mid-level clouds, especially in the optically medium and optically thick categories.Such a difference results in less positive total SW cloud feedbacks in the cooling experiments than the warming experiments.

Surface Temperature
Figures 7a-7d contain plots of the temperature mediated change in surface temperature.The zonal temperature gradient across the equatorial Pacific is a strong predictor of the strength of the Walker circulation, because a strong temperature gradient supports easterlies due to the associated pressure gradient.This circulation is fundamentally coupled to the ocean circulation, where the wind stress forces a thermocline gradient, which further supports a sea-surface temperature gradient through upwelling in the East Pacific (Battisti et al., 2019;Bjerknes, 1969).Therefore, we expect that changes in the surface temperature pattern may help explain the zonal structure of the high cloud changes shown in Figures 2-5.
In the equatorial pacific of the 4xCO2 and solp4p experiments there is enhanced warming (greater warming than the global mean) in the East, and less warming in the West Pacific.In the 0p5xCO2 and solm4p experiments there is some cooling in the East Pacific relative to the West Pacific, however the change in the temperature gradient across the equatorial Pacific is weaker than in the 4xCO2 and solp4p experiments (and has poor model agreement in the cooling experiments).This difference in sea-surface temperature patterns likely explain much of the differences in the high cloud response to warming and cooling.
On land there is generally greater warming than in the adjacent oceans in the 4xCO2 and solp4p experiments (Figures 7a and 7b).Enhanced warming over land is a well-documented response in models and is constrained by the differences in lapse rate over land and ocean, owing to the greater moisture availability over ocean.Further details on such mechanisms are described by Byrne and O'Gorman (2013) and Joshi et al. (2008).Interestingly, in the 0p5xCO2 and solm4p there is little difference between the ocean and land cooling, especially in the Southern Hemisphere.This suggests that the oceanic and continental lapse rate adjustments to cooling exhibit less difference than their adjustments to warming.
In the 0p5xCO2 and solm4p experiments there is some cooling in the East Pacific relative to the West Pacific, however the change in the temperature gradient across the equatorial Pacific is weaker than in the 4xCO2 and solp4p experiments.In southern subtropical Pacific there is generally a smaller temperature change than in the global mean.The relatively warm subtropical surface air temperatures are collocated with an increase in high cloud amount and contributes to the South-Western shift of the SPCZ.Narsey et al. (2022) found that in CMIP5 and CMIP6 models the SPCZ shifts towards the region with relatively high surface temperatures under warming due to the associated shift of the midlatitude jet.Thus, we expect that when cooling occurs the same mechanisms apply, and the SPCZ (and the associated high clouds) shifts towards regions with relatively less cooling.

Estimated Inversion Strength
To better understand low cloud changes, we show in Figures 7e-7h maps of temperature mediated changes in Estimated Inversion Strength (EIS).Stratocumulus cloud occurrence in particular is typically very well correlated with EIS on monthly or longer timescales.This is especially true in the Tropics and Subtropics, where stronger inversions reduce the entrainment rate of free-tropospheric dry-air into the boundary layer, allowing for more and thicker clouds to form (Bretherton, 2015;Wood & Bretherton, 2006).In Figures 7e-7h we have highlighted a few subtropical regions with red boxes where there are persistent stratocumulus cloud in the modeled piControl climatology, and observations (Klein & Hartmann, 1993;Qu et al., 2014Qu et al., , 2015)).But stratocumulus are also common over colder waters at mid and high latitudes, especially in the winter (Wood, 2012).
In the warming experiments there is increasing EIS in the northern hemisphere tropics and parts of the subtropics, especially from 0°to 20°in the Central and Eastern Pacific, and Atlantic oceans.Additionally, there is an increase of EIS in the Eastern Atlantic, off the west coast of Northern Africa and Europe.In the southern hemisphere increases in EIS are less widespread.There is narrow region of EIS increase primarily along a line connecting Indonesia to the Peruvian stratocumulus deck (denoted as the red box off the west coast of South America).In the midlatitudes of both hemispheres (poleward of 40°) and along the equator in the Eastern and Central Pacific, there is also notable decrease of EIS.In the stratocumulus regimes (marked with red boxes) there are inconsistent changes in EIS, where some stratocumulus regimes experience strengthening inversions, while others experience weakening with increasing temperature.For example, there is decreasing EIS over most of the Californian, Peruvian, and Australian stratocumulus regions, and increasing EIS in the African and North Atlantic Stratocumulus.We note that in all these regions there is a decrease in low cloud amount (see Figures 2 and 3), the role of EIS in contributing to such changes are further discussed in Section 4. The temperature mediated EIS changes shown here broadly agree with the late stage temperature mediated changes of EIS in CMIP5 model simulations of 4xCO2 found by Qu et al. (2015), and the differences that occur are attributable to the different set of models used in each study.
In the cooling experiments there are broadly similar patterns of temperature mediated EIS change to the warming experiments, however there are some key differences.In the tropical Atlantic in contrast to the response to warming, there is little EIS change in the solm4p and 0p5xCO2.In the Southern hemisphere there is decrease of EIS in the subtropical Pacific, Atlantic and Indian Oceans with decreasing temperature (green color), which is more widespread than the EIS changes in the southern hemisphere caused by warming.Like the warming experiments, in the stratocumulus regimes the EIS is not consistently increasing or decreasing.

Relative Humidity
Recent research has shown that in climate models, trade cumulus occurrence is mediated by the moisture fluxes into and out of the boundary layer (Vogel et al., 2022).Drying of the free-troposphere can increase the rate of boundary-layer drying through convective mixing, which desiccates the cloud layer.In Figures 7i-7l we show the temperature mediated changes in relative humidity at 700 hPa (RH 700 ), which indicates the dryness of the lower portion of the free troposphere.
In the warming experiments there is increased RH 700 in the Tropical East Pacific and decreased in the Tropical West Pacific, while in the cooling experiments there is an increase in relative humidity (negative temperature mediated change) around 20°north and south of the equator, and a decrease in relative humidity along the equatorial Pacific, Atlantic, and Indian oceans.In the subtropical stratocumulus regions, there is an increase in RH 700 with warming (green colors), and a more mixed response to cooling.For example, in the solm4p there is increased RH 700 in the Californian, Peruvian and Australian stratocumulus (purple colors), and in the 0p5xCO2 experiment there are relatively weak decreases with poor model agreement.In the midlatitudes of the warming experiments, there is relatively little change in the Northern Hemisphere between 40°and 60°N and a general drying pattern (purple colors) in the Southern Hemisphere between 40°and 60°S.In the cooling experiments there is also drying (green colors) in various regions of the Southern Ocean.These temperature mediated changes in cloud controlling factors shown in Figure 7 will be used in Section 4.2 to determine which mechanisms are responsible for the cloud changes shown in Figures 2 and 3.

Circulation Changes
In this section we examine metrics related to the atmospheric circulation to contextualize the cloud changes described in Section 3.1.In Figure 8   pressure velocities (decrease in upward motion) per degree of warming (green colors) over Indonesia and the Maritime continent, along the northern edge of the ITCZ (and to a lesser degree the adjacent subtropics), and along the southwestern edge of the SPCZ.In at least three models the positive pressure velocity changes over the Maritime region extend well into the Indian Ocean.
The cooling experiments differ notably from the warming experiments, and have relatively little change (less purple, less stippling) in the East Pacific and along the Pacific Cold Tongue, as well as over the Peruvian stratocumulus zone.
The positive pressure velocity changes along the northern edge of the ITCZ and southwestern edge of the SPCZ are indicative of the ITCZ and SPCZ shifting equatorward and to the east with warming (and opposite direction with cooling).The shift of the ITCZ and SPCZ is perhaps more easily seen in the zonally averaged 500 hPa vertical velocity (bottom panel of Figure 8.The latitude where zonal mean upward motion is maximized (there is a minimum in pressure velocity) in each hemisphere is noted by vertical dashed lines for each model experiment.In the warming experiments there is an equatorward shift of the latitude of maximum updraft (minimum in pressure velocity) in the Pacific in both the northern and southern hemisphere.However, the distance shifted is far greater in the southern hemisphere.
In the solm4p and 0p5xCO2 experiments the opposite shift in ascent does occur compared to the warming experiments, however it is not of equal magnitude.In the southern hemisphere the latitude of maximum ascent (minimum in pressure velocity) shifts poleward with cooling by the same amount in both cooling experiments (solm4p and 0p5xCO2), however the shift is of much shorter distance than the equatorward shift in the two warming experiments.In the northern hemisphere the tropical ascent shifts poleward in the solm4p, but there is no change in the location of maximum ascent in the 0p5xCO2 experiment.
Much of the upward motion in the tropics is related to the large-scale dynamical circulations that occur in both the zonal and meridional directions.Figure 9 shows the zonal mean 500 hPa meridional stream function, which has been widely used to diagnose the strength and width of the zonally overturning Hadley circulation (e.g., Chemke, 2022;Frierson et al., 2007;Oort & Yienger, 1996;Staten & Reichler, 2014).We have separated the stream function into the DJF and JJA seasons, in Figures 9a and 9b respectively, because the Hadley circulation, in fact, only occurs in the Winter Hemisphere, so annual mean plots can misleadingly depict the Northern and Southern node of the Hadley circulation simultaneously.In both seasons and at most latitudes, the 500 hPa meridional streamfunction in the 4xCO2 and solp4p experiments are nearly the same, so much so that the lines in Figure 9 overlap.The strength of the Hadley circulation is often quantified as the maximum absolute value meridional stream function at 500 hPa (e.g., Oort & Yienger, 1996).Figure 9 shows that the Hadley circulation weakens in both the solp4p and 4xCO2 and strengthens in the solm4p in both DJF and JJA.In the 0p5xCO2 experiment there is a small increase in the Hadley circulation strength in DJF, however there is little change in JJA.As such, the high cloud changes in the Southern Hemisphere of the 0p5xCO2 experiment are apparently unrelated to changes in the strength of the Hadley circulation.
The Hadley circulation width can be characterized using the latitude where the 500 hPa stream function crosses the zero line (e.g., Chemke, 2022;Frierson et al., 2007).By this metric, the Hadley cell widens in the warming experiments and shrinks equatorward in the cooling experiments.Because the width changes can be small and difficult to see in Figure 9, the Hadley cell width in each season (and associated hemisphere) calculated from the last 30 years of each simulation is shown in Tables 3 and 4. In all models there is a widening of the Hadley circulation in both hemispheres and in both the 4xCO2 and solp4p.Neither of the warming simulations has a Hadley circulation that is persistently wider than the other across all models.
Although it is not the focus of this paper, we do point out that in more complex models with interactive stratospheric ozone solar and CO 2 forcing may cause quite different circulation responses owing to how they heat the stratosphere differently.CO 2 increase cools the stratosphere by blocking upwelling LW radiation from the surface reaching the stratosphere (Goessling & Bathiany, 2016).On the other hand, insolation increase causes stratospheric warming via ozone absorption, and Bednarz et al. (2022) have used simulations from the GeoMIP ensemble to show that the stratospheric warming and temperature gradients incurred by solar forcing have notable impacts on tropospheric circulations, including on the position of the tropospheric jet and Hadley circulation.Thus, we speculate that if our experiments were run on models with interactive stratospheric ozone we might find more notable differences in how the Hadley cell (and associated high-clouds) change in the 4xCO2 and solp4p experiments.

Cloud Phase Feedbacks and Sea Ice-Cloud Interaction
To better understand the cloud changes at mid-to-high latitudes, in Figures 10a-10d we show the changes in whole-column cloud ice mass fraction, which is the vertically integrated atmospheric ice-mass content divided by the combined mass of ice and liquid water.There is an increase in cloud ice mass fraction in the solm4p and 0p5xCO2 that extends through the midlatitudes down to 30°in the solm4p and to 40°in the 0p5xCO2.In the solp4p and 4xCO2 (the warming experiments) there is a reduction in cloud ice mass fraction poleward of 50°l atitude.The change in cloud ice fraction in each of the four experiments occurs at roughly the same latitude as the change in low-cloud optical depth shown in Figures 2-5.
In Figures 10e 10h we show the change in sea-ice extent as a 30-year average deviation from the pre-industrial climatology.In the cooling experiments (solm4p and 0p5xCO2) the sea-ice reaches much lower latitudes than in the pre-industrial climate, where there is sea-ice growth past 50°in both hemispheres of the solm4p and growth past 55°in the 0p5xCO2.In the warming experiments there is a large reduction of sea-ice that reaches Antarctica and the North Pole such that the arctic is nearly ice-free.The effects of both the cloud phase and sea-ice on clouds and their radiative feedback are discussed in Section 4.  In summary of Section 3, we find that (a) for most cloud types and cloud controlling factors, the temperature mediated response from solar and CO 2 forcing is quite similar, with perhaps the most notable difference being a slightly greater loss in low-cloud amount in solp4p as compared with 4xCO2 and (b) there are numerous differences between the response to cooling and warming.In the following section we hypothesize on the mechanisms responsible for these differences based on the changes in atmospheric circulation and cloud controlling factors.

Discussion of Physical Mechanisms
In this section we discuss further the simulated cloud changes and likely mechanisms in the context of previous studies.In turn, we focus on high clouds, low and mid-level clouds, and lastly high latitude clouds.We also briefly discuss some limitations of the radiative kernel approach at the end of this section.

High Clouds
In the solp4p and 4xCO2 experiments, there is a high cloud change that is indicative of a weakening walker circulation, a weakening and widening of the Hadley circulation, as well as a shift of the ITCZ toward the equator and a Northeastward shift of the SPCZ.The cloud changes occur in the same locations as changes in the vertical velocity, indicating that the tropical high cloud changes are in fact circulation driven.A similar circulation pattern occurs during the El Nino phase of ENSO (Adames & Wallace, 2017).In the warming experiments, the increase in tropical surface temperature is largest in the Tropical Central Pacific (see Figures 7a-7d), similar to the pattern associated with El Nino.
There are a handful of mechanisms which have been proposed to mediate the zonal temperature gradient and associated Walker circulation (Held & Soden, 2006;Knutson & Manabe, 1995;Williams et al., 2023).Each of which predict walker circulation changes consistent with those seen in the warming and cooling experiments analyzed here.Knutson and Manabe (1995) proposed a pair of contributing mechanisms, the first of which is that higher static stability occurs in the warming experiments, which slows the ascent in the West Pacific and decreases subsidence in the East Pacific, causing less wind stress along the equatorial pacific, bringing less cool water from the Eastern Pacific into the Central Pacific.This creates a weaker surface temperature gradient between the Central and Western Pacific.We do find higher static stability in the West Pacific (more below on this) and decreased subsidence in the East Pacific (see Figure 8) in the warming experiments.
The opposite does not occur (at least not to the same extent) following the abrupt reduction of the solar constant or CO 2 , where there is not a clear shift toward a La Nina-like state in the solm4p or 0p5xCO2 experiments.In the solm4p experiments the static stability decreases in the West Pacific, but the change is weaker than the increase in the solp4p and 4xCO2 experiments.The static stability anomaly averaged over the final 30 years of simulation in the Tropical West Pacific from 600 to 200 hPa shown in Table 5, which shows that there is less change in static stability in the Tropical West Pacific from cooling than warming.
This can be understood as a consequence of the moist adiabatic lapse rate, which effectively sets the lapse rate in convective regimes near quasi-equilibrium.The moist adiabatic lapse rate has a nonlinear relationship with surface temperature (due in part to the dependence of the saturation vapor pressure on temperature, and the Clausius-Clapeyron relationship), such that the lapse rate increases more rapidly with increasing temperature.This creates the asymmetric static stability changes in the warming and cooling experiments, and following Knutson and Manabe (1995) causes a greater slowing of the Walker circulation in the warming experiments than there is hastening of the Walker circulation in the cooling experiments.
The other contribution to Walker circulation changes identified by Knutson and Manabe (1995) is the role of the longwave cooling profile, due to changes in specific humidity of the upper troposphere.In a warmer climate the upper tropospheric cooling rate over the West Pacific increases more than is balanced by convective heating.The net radiative cooling over the West Pacific weakens the pressure gradient aloft that circulates air from the convective West to subsiding East Pacific, weakening the Walker circulation.Only two of the models we have examined here saved output of vertically resolved radiative flux, however both models with radiative flux outputs do show a larger increase in LW cooling from the 4xCO2 and solp4p than the reduction in LW cooling in the ).This suggests that the asymmetrical response in the tropical pacific to warming and cooling is consistent with both mechanisms presented by Knutson and Manabe (1995).
In the framework of Held and Soden (2006), a change in convective moisture flux is viewed through the lens of the hydrological budget.Moisture flux (M) is equal to the product of boundary layer specific humidity (q BL ; in our case averaged from the surface to 850 hPa), and precipitation (P) such that a change in convective moisture flux is the difference between a change in boundary layer specific humidity, and precipitation. 1 The temperature mediated change in convective moisture flux ( dM dT ) over the Tropical West Pacific (120-140°l ongitude, 15-15°latitude) is 6.0%/K for solp4p, 6.9%/K for 4xCO2, 1.6%/K for solm4p, and 3.6%/K for 0p5xCO2.This is again consistent with a stronger dampening of the Walker circulation and reduced high cloud in the Tropical West Pacific in the warming experiments.A table showing the individual model results is available in Supporting Information S1.
Our findings of a different response in the Walker circulation to warming and cooling is also consistent with the results of Williams et al. (2023), who performed model experiments with warming and cooling patches in the Tropical West Pacific.They use the tropical moist static energy budget to find that nonlinear response to tropical warming and cooling is a direct result of quasi-equilibrium in the ascending portion of the tropical atmosphere, and relatively weak temperature gradients in the tropical free troposphere.They find that cooling reduces the amount of deep convection in the West Pacific, which causes weaker coupling between the Western Pacific boundary layer, and free tropospheric temperature.Such decoupling means that cooling the west Pacific has less impact on the east Pacific than warming, hence there is a weaker Walker circulation change from cooling than warming.
In addition to the zonal (Walker) circulation in the equatorial Pacific, there are also changes in the meridional (Hadley) circulation in the tropics and subtropics which impacts high cloudiness.In the solp4p and 4xCO2 there is a weakening and widening of the Hadley circulation (shown in Figure 9), that is similar across the two warming experiments.In the solm4p there is a characteristically similar effect in the opposite direction, where in both the Northern and Southern Hadley Cells the circulation strengthens, and narrows.Interestingly, however, in the 0p5xCO2 there is only a change in the Hadley cell strength in the Northern Hemisphere.In the Southern Hemisphere, the Hadley circulation strength and width remains nearly the same as during the pre-industrial climate simulations.This suggests that greater forcing may be necessary to change the Southern Hemisphere Hadley circulation than the Northern Hemisphere circulation.In the 0p5xCO2 there is a narrowing of the Hadley circulation, indicating that even though the strength does not change, the static stability decrease in the subtropics still moves the threshold of baroclinic instability closer to the equator, which drives the Hadley cell boundary (Lu et al., 2007).
There is also a high cloud shift associated with changing locations and strength of the ITCZ and SPCZ.In the subtropical Southern Pacific, the temperature change is smaller than the global mean change.The relatively warm subtropical surface temperatures are collocated with an increase in high cloud amount and contributes to the South-Western shift of the SPCZ.Narsey et al. (2022) found that in CMIP5 and CMIP6 models, the SPCZ shifts toward the region with relatively high surface temperatures under warming.Similarly, in our experiments we find that the SPCZ shifts toward the regions with greater warming in the solp4p and 4xCO2, and toward regions less cooling in the solm4p and 0p5xCO2.
There is positive feedback in the tropics of all experiments associated with the change in CTP with warming and cooling, consistent with the Fixed Anvil Temperature (FAT) hypothesis (Hartmann & Larson, 2002;Zelinka & Hartmann, 2010).Under FAT, the maximum height of deep convection is set by the height at which longwave cooling of the clear-sky atmosphere is no longer efficient.The longwave cooling of the upper atmosphere is predominately due to water vapor, and the spectral properties of the water vapor molecule cause strong radiative cooling to space throughout the troposphere with temperatures above 220 K, and little cooling to space at temperatures below 220 K (Jeevanjee & Fueglistaler, 2020).Thus, when the climate warms deep convective clouds rise in altitude (and vice-versa for cooling climates) keeping convective cloud tops near 220 K.This creates a positive LW cloud feedback because the relative cooling of the cloud top emission temperature remains Journal of Geophysical Research: Atmospheres 10.1029/2023JD040296 nearly constant as the surface temperature increases (Hartmann & Larson, 2002;Zelinka & Hartmann, 2010).There is observational support for this feedback, including a recent study we have published on rising cloud tops of high clouds based on stereo-imaging observations from the NASA Multiangle Imaging Spectro-Radiometer (MISR) (Aerenson et al., 2022;Norris et al., 2016).We also find in Section 3.2 that the feedback associated with decreasing CTP is slightly stronger in solp4p than 4xCO2.The difference is small but is consistent with solp4p causing more concentrated warming in the tropics, where deep convection is more common.

Low and Mid-Level Clouds
All of the abrupt forcing experiments produce significant temperature mediated changes in low and midlevel clouds.In the solp4p and 4xCO2 there is a reduction of optically medium low cloud between 40°S and 40°N, especially along the Eastern Pacific cold tongue, and in regions with persistent stratocumulus decks (marked by the red boxes in Figures 7e-7h).Stratocumulus clouds form when there is supply of moisture at the surface, the boundary layer is well mixed and has sufficient instability to lift surface air to the condensation level, and the free troposphere is stable enough to cap the instability at a relatively low altitude.Bretherton (2015) identified four feedback mechanisms that cause stratocumulus clouds to change with increased CO 2 : (a) the radiative effect where an increase in water vapor or CO 2 in the free troposphere inhibits cloud top cooling (which stabilizes the boundary layer causing stratocumulus to lower and thin with an increasingly emissive free troposphere), (b) the dynamic effect where decreases in subsidence with warming (for a fixed inversion strength) results in an increase in the boundary layer thickness and thicker stratocumulus (assuming there is sufficient mixing within the boundary layer to maintain coupling between the cloud and surface), (c) the thermodynamic effect where a warmer sea-surface temperature or drier free troposphere results in a larger gradient in the specific humidity (across the inversion) that promotes more efficient turbulent entrainment-driven drying of the boundary layer and thins stratocumulus, and finally (d) the stability effect where a stronger inversion (larger EIS) results in less entrainment that lowers and thickens stratocumulus.
The radiative effect (which differ substantially between the solp4p and 4xCO2) is part of the adjustment to the solar and CO 2 forcing rather than the temperature mediated response, and so is discussed in detail in Part II which specifically addresses cloud adjustments (Aerenson et al., 2023).The other three mechanisms are nominally all present in the temperature mediated response.Mechanism 2 suggests that slowing circulations might result in an increase in stratocumulus with warming, while mechanisms 3 and 4 would decrease stratocumulus.As shown in Figure 8 there is a reduction of subsidence with warming in stratocumulus regimes.However, no regions experience a net increase in low clouds, and so mechanism 2 is clearly not dominating the other response mechanisms.We note however, that a careful examination of Figure 2 shows that there is an increase (purple colors) in optically medium mid-level, and to a lesser degree optically thick mid-level clouds (though we note there is poor model agreement).This is consistent with rising cloud tops in some models and mechanism 2-the dynamic effect-is the only mechanism that causes rising stratocumulus cloud tops with warming.The remaining two mechanisms predict a decrease in cloudiness with increasing sea-surface temperature and EIS.There is lowcloud loss even in locations with decreasing EIS, which suggests that sea-surface temperature changes (the thermodynamic effect) likely play a dominant role in the changes of marine low-clouds.This result is consistent with the review paper by Klein et al. (2017) who find that sea-surface temperature is the leading cloud controlling factor for subtropical stratocumulus followed by EIS, and with lesser contributions from subsidence, advection, and free tropospheric relative humidity.
In the midlatitudes and subtropics outside of the stratocumulus regimes, low level clouds are commonly cumulus clouds, often called trade wind cumulus.Such trade cumulus also experiences substantial temperature mediated change; however, they are often overlooked in favor of stratocumulus clouds, due to stratocumulus clouds' tendency to dominate the low-cloud radiative effect and feedbacks.They can be understood through a similar framework of boundary layer moisture availability and free tropospheric drying as stratocumulus, and feedbacks from trade cumulus clouds have been found to be tightly linked with boundary layer convective mixing (Sherwood et al., 2014;Vogel et al., 2022).We find a substantial decrease in low cloudiness in the warming experiments over Southern Hemisphere Ocean centered around 40°S that is larger (and more consistent between models) than that occurring in the Northern Hemisphere.Near this latitude in the Southern Hemisphere, there is relatively large sea-surface temperature increase (as compared with the global mean, or the Northern Hemisphere), a stronger temperature mediated decrease in EIS, and greater drying at 700 hPa, all of which may contribute to the decrease in low cloudiness.Zelinka et al. (2020) examined the cause of high cloud feedbacks in Journal of Geophysical Research: Atmospheres 10.1029/2023JD040296 4xCO2 between 60°and 30°S in CMIP5 and CMIP6 models.Through a multi-linear regression, they show that the decrease in low cloud amount in CMIP6 models is largely due to a combination of increasing EIS, and a drying of the free troposphere (reduction in 700 hPa relative humidity).The ensemble of models used by Zelinka et al. (2020) does include the models examined here, as such we expect the same mechanisms to be responsible.
Turning attention to the difference between the cloud response to 4xCO2 and solp4p, the largest difference occurs in the optically medium low cloud category, where there is about a 0.07% greater reduction of cloud with warming due to solar forcing than CO 2 forcing.This difference in low and mid-level cloud amount is consistent with a stronger thermodynamic effect in solp4p, as there is slightly greater warming in the tropics and subtropics in solp4p than occurs in the 4xCO2 (for an equivalent change in global mean temperature).The difference in warming pattern following solar and CO 2 forcing is further explored in Part II, which examines the cloud changes which are not mediated by global mean temperature change (including changing SST pattern effects).Zhou et al. (2023) find that warming patterns are a strong predictor of cloud feedbacks to a variety of forcing mechanisms such that a green's function approach can be used to reconstruct the feedback to a specific forcing agent if the warming pattern is known.In the solp4p experiment there is more warming in the tropics and subtropics than the 4xCO2 (shown in Part II), hence we speculate that the enhanced warming in the tropics of subtropics of solp4p (when compared with 4xCO2) causes a stronger cloud feedback via the thermodynamic effect.
In the cooling experiments, the positive feedback associated with oceanic low cloud increases (orange colors in Figures 4 and 5) is broadly similar, but somewhat weaker than in the warming experiments, and does not extend as far poleward.There is, however, a stronger (and more consistent) change in mid-level cloud over the Namibian, Australian, and Peruvian stratocumulus decks, associated with increases in subsidence rates and consistent with a relatively strong dynamic effect.

High Latitude Clouds
In both warming experiments there is an increase in low and mid-level cloud optical depth poleward of about 60°, and the opposite response: a decrease in optical depth (poleward of about 40°) in both cooling experiments.This is evidenced in Figures 2-5 as a reduction in optically thin clouds and an increase in optically medium and thick cloud in the warming experiments, and vice-versa in the cooling experiments.
As demonstrated by comparing Figure 2-with Figures 10a-10d, this optical depth change is occurring at a similar latitude as a reduction in the relative fraction of cloud ice mass fraction in the warming experiments (and increase in the cooling experiments).Cloud ice crystals tend to be larger than liquid droplets, and as such for an equivalent amount of condensate mass, there are more particles in a liquid cloud than an ice cloud and this causes ice clouds to be less reflective of sunlight than liquid clouds (e.g., Cesana & Storelvmo, 2017;Rogers & Yau, 1989).Additionally, for an equivalent liquid/ice-water path (IWP) liquid clouds are less efficient at precipitating, so liquid clouds tend to contain more water than ice clouds, which may cause them to be optically thicker and have longer lifetime (McCoy et al., 2015;Mitchell et al., 1989;Mülmenstädt et al., 2021;Senior & Mitchell, 1993;Tsushima et al., 2006).These theoretical expectations are supported by observations, including ground based measurements by Terai et al. (2019), who found that at high latitudes clouds with a mean temperature less than 0°C are observed to have an increased optical depth with warming.
The cloud changes are consistent with this expected cloud optical depth-phase feedback.Zhu and Poulsen (2020) likewise identified the latitude at which this cloud phase feedback occurs to shift and create a non-linearity in the amount of temperature change that occurs from different amounts of forcing.However, we note that the cloud optical depth changes might also be related to sea-ice changes (at least at very high latitudes), because shrinking sea-ice (which is shown in Figures 10e 10h) allows for greater heat absorption in the Summer and release in the Autumn, which deepens the boundary layer and thickens low clouds following the mechanism described by Morrison et al. (2019).
While increases in optical thickness at high latitudes can be ascribed to phase changes for low and mid-level clouds, there is also significant thickening of high-clouds at mid-to-high latitudes, where even in the warmed climate it remains too cold for phase change to occur.This is perhaps strongest near 60°S.As mentioned by Zelinka and Hartmann (2012), observational studies have indicated that the total water content of clouds increases with increasing temperature at rates near the expected increase in water content following a moist adiabatic parcel (Somerville & Remer, 1984).An air parcel beginning at a warmer surface temperature will produce more Journal of Geophysical Research: Atmospheres 10.1029/2023JD040296 condensed water as it rises adiabatically through a cloud than a parcel beginning with a cooler surface temperature.Betts and Harshvardhan (1987) demonstrated this effect analytically and showed that one expects a greater change in condensed water at high latitudes than in the tropics.In Supporting Information S1 we show the temperature mediated changes in IWP.We find that in the warming experiments there is an increase in IWP at mid-to-high latitudes, even when the ice cloud fraction decreases (see Figures 10a and 10b).The patter in IWP change at mid-to-high latitudes matches well the pattern of increase in optically thick high-altitude cloud, with the largest increase happening near 60°S.

Limitations of the Cloud Radiative Effect From Kernels
Using the cloud radiative kernels, we find that the previously mentioned cloud optical depth change associated with phase partitioning changes constitutes a positive feedback in the solm4p and 0p5xCO2, due to the increased reflectivity of the low-cloud layer, and relatively little feedback in the 4xCO2 and solp4p.This kernel-derived cloud feedback illustrates a limitation with the cloud radiative kernel method.The radiative kernels isolate the radiative effect of cloud changes from the effect of changes below the cloud layer which are often referred to as cloud masking effects (Zelinka et al., 2013).In the forced experiments we use cloud radiative kernels which correspond to the local surface albedo in the models' base-state.So, in the solp4p and 4xCO2 experiments, the cloud thickening does not cause a strong cloud feedback using the cloud radiative kernels because in the initial state there is high-albedo sea-ice beneath the clouds.This diminishes the SW radiative effect of the clouds, and the kernel method does not account for the sea-ice reduction when calculating the radiative anomaly in the warmed climate, or the sea-ice growth in the cooled climate of solm4p and 0p5xCO2.

Conclusions
We began this paper by posing the following two questions: (a) How do cloud feedbacks differ in response to abrupt changes in CO 2 and solar forcing?And (b) Are there symmetrical (equal and opposite) cloud feedbacks to an increase and a decrease of radiative forcing?Overall, this paper has allowed us to parse through the effects of solar and CO 2 forcing to determine what types of temperature mediated cloud changes occur from each forcing agent, and how temperature mediated cloud changes are different in warming and cooling model experiments of both CO 2 and solar forcing.In short, the answer to the first question is that the temperature mediated cloud feedbacks are quite similar between solp4p and 4xCO2, however there are small differences in the feedbacks (which we discuss further below).And concerning the second question we find numerous differences between the response to increase and decrease of radiative forcing, most notably are those related to cloud phase feedbacks and changes in tropical circulation.
The primary value of this work is that it examines results for a small set of models in a consistent manner.This highlights similarities and differences between models.We note the stippling used to express where there is agreement in temperature mediated cloud responses between models only means agreement in the sign.There is a large variation in the magnitude of the temperature mediated cloud responses, and results for individual models are given in Supporting Information S1.In fact, one benefit of this multi-model analysis is that when comparing the results from 4xCO2 with solp4p we find that the spread across models is greater than the difference between the 4xCO2 and solp4p simulations.This means that the difference between the temperature mediated cloud changes from solar and CO 2 forcing do not exceed the inter-model variability.To a significant degree, the multimodel analysis presented here supports results of previous studies based on simulations using individual models.For example, consistent with Kaur et al. (2023), we do see some differences in the surface warming pattern that results from differences in the pattern of radiative forcing between the solp4p and 4xCO2 forcing experiments.In particular, there is slightly greater warming in the tropics, and less warming near the poles in the solp4p experiment than in the 4xCO2 that we speculate is due to the radiative forcing in the solar experiment being larger in the tropics.This greater temperature increase in the tropics drives a greater loss in low-cloud amount in solp4p as compared with 4xCO2.This does not have a great impact on high cloud changes, because in both the solp4p and 4xCO2 the tropical high cloud changes are dominated by a weakening of the walker circulation, which causes cloud changes of similar magnitude in both the solp4p and 4xCO2.
Both Rose et al. (2014) and Salvi et al. (2022) compared cloud feedbacks caused by spatially non-uniform forcing with those from CO 2 forcing.They both used forcings localized to the midlatitudes or tropics.This work relates to the present study because CO 2 forcing is spatially uniform across the globe, due to the long lifetime of CO 2 making it evenly mixed through the atmosphere, while solar forcing is strongest in the tropics, where the greatest insolation occurs.So, one might expect the cloud feedbacks from solar forcing to be more similar to those that occur following tropical forcing (and less similar to midlatitude forcing).However, the comparisons of midlatitude and tropical forcing with CO 2 forcing contain much larger differences in the geographical distribution of the forcing than our comparison of solar and CO 2 forcing.Such localized forcing may also create circulations and teleconnections that do not occur in our experiments.In fact, both Rose et al. (2014) and Salvi et al. (2022) found that forcing the midlatitudes causes more positive cloud feedbacks, yielding a less negative total feedback parameter, than CO 2 forcing, or forcing concentrated in the tropics.We find more positive feedbacks from solar than CO 2 forcing (Figure 6), which is opposite that suggested by Rose et al. (2014) and Salvi et al. (2022).Our results do not directly contradict their conclusions, but rather suggest that the cloud response is non-linear, and cannot be decomposed entirely into a sum of forcing applied to different regions.This also cautions against simply scaling and summing the abrupt CO 2 and abrupt solar forcing cloud feedbacks to understand solar geoengineering, as the forcing from geoengineering is not likely to be geographically uniform.While we expect that same physical mechanisms that drive cloud responses will be present, there are likely to be differences in the global circulation that will change the relative importance of these mechanisms.
Regarding the question of how climate warming compares to climate cooling, our results are largely consistent with Chalmers et al. (2022), in that we find key differences between warming and cooling to occur at high latitudes, where the latitude at which ice processes are active in the pervasive low-level clouds and sea-ice extend farther equatorward from cooling.The greater spatial coverage of the cloud phase and sea-ice transition, as well as the increase in insolation with decreasing latitude causes the associated SW feedback to be stronger in the cooling experiments than warming experiments.This result is also consistent with work that has been done on non-linear feedbacks to different amounts of warming by Bloch-Johnson et al. (2021) and Zhu and Poulsen (2020), who also identified the sea-ice and cloud phase transitions as locations where of non-linear feedbacks are prominent.
The results of the solm4p and 0p5xCO2 experiments also indicate the importance of the temperature pattern in the tropics for dictating cloud feedbacks.We find that the zonal temperature gradient across the equatorial Pacific weakens from global warming far more than it strengthens due to global cooling, such that there is a stronger Walker circulation change as a response to warming than cooling.Chalmers et al. (2022) found a similar result by comparing simulations of 2xCO2 and 0p5xCO2 in CESM1.Our results with a multi-model ensemble (albeit a small one) solidify the robustness of the differences between warming and cooling found by Chalmers et al. (2022).
As a caution, we note this analysis was performed with a relatively small subset of the CMIP6 models, and although the results discussed are consistent across our set of models, they may not be representative of a larger ensemble.Additionally, we use single realizations from each model for each experiment, which limits our ability to assess whether differences between experiments surpass internal variability.We try to overcome such limitations by using model experiments with relatively large abrupt forcing, such that the signal-to-noise ratio is large, and derive temperature mediated cloud changes from relatively long model simulations (150 years), to dampen the importance of internal variability.
In closing, we have focused in this study on the temperature mediated component of could changes.We have found that the magnitude and sign of the forcing does matter.Nonetheless, given a similar magnitude of forcing and change in global mean surface temperature, differences between the temperature mediated response to solar and CO 2 radiative forcing are subtle, meaning the temperature mediated cloud changes are fairly insensitive to the forcing mechanism.This supports the underlying premise of the feedback model, that cloud feedbacks can be understood as a combination of the response to global temperature and an adjustment that occurs directly due to the forcing agent.If the abrupt changes in solar and CO 2 radiative forcing had resulted in a substantially differing temperature mediated cloud response, this would indicate that the temperature mediated component were not necessarily driven by global temperature, and instead were specific to the forcing agent.This is not to suggest that there are no differences in the temperature mediated cloud responses.We do find small differences in the temperature mediated cloud response driven by differences in the pattern of sea-surface temperature.
To be clear, the temperature mediated cloud response is only part of the total cloud response.There are larger differences in the cloud adjustment component of the response to solar and CO 2 forcing (meaning the cloud changes which are not mediated by global mean temperature), which is examined in detail in the companion paper to this article (Part II Aerenson et al., 2023).In particular, there are substantial differences in the adjustment of stratocumulus and cumulus clouds to solar and CO 2 forcing, which follow from differences in the direct radiative effect that solar and CO 2 forcing have on heating at cloud top, and there are differences in the adjustment of high clouds to solar and CO 2 forcing that are driven by the differences in the vertical profile of radiative heating.These adjustments contribute significantly to the total cloud radiative effect.

Figure 1 .
Figure 1.Area weighted global mean annually averaged cloud fraction anomaly (as seen by International Satellite Cloud Climatology Project simulator) plotted against global mean annually averaged surface temperature change.Note that the scale is halved for the 0p5xCO2 simulation because the temperature change is smaller than in the other experiments.Colors denote the individual models.The temperature mediated change in global total cloud amount (for each model) is the slope of the fitted line which can be found in Table2.

Figure 2 .
Figure 2. Temperature mediated response of cloud from the 4xCO2 experiment.Colors show the % change in cloud amount in each category per unit of global temperature change, with the global (area weighted) mean change given in the title of each panel.The cloud optical depth is broken into three ranges: optically thin (τ ≤ 3.6), medium (3.6 < τ ≤ 23), and thick (τ > 23) clouds, and the CTP is broken into three CTP ranges: low (CTP ≥ 680 hPa), mid-level (680 hPa > CTP ≥ 440 hPa), and high (CTP < 440 hPa) cloud.The global mean total cloud change (summed over all optical depth and CTP ranges) is available in Table 2.In the figure titles we have included the total global mean value, which is found by summing the global mean values in each of the nine subplot titles, these values are identical to those in the bottom row of Table 2. Stippling indicates regions where at least 3 out of 4 models agree on the sign of the temperature mediated cloud change.
Overall, Figures2 and 3show that the pattern of temperature mediated cloud change is quite similar in both the solp4p and 4xCO2 experiments.In fact, the pattern of temperature mediated cloud change varies from model to model, but within an individual model the pattern is quite similar in both experiments (see Supporting Information S1 for individual model results).The global mean change is listed in the title of each panel.In the multimodel mean, there is a net reduction in global mean cloud amount in seven out of nine categories with only optically-thick high-level clouds and optically-medium high-level clouds having an increase.Over the next three subsections, we discuss the geographic structure of the low, mid, and high-level cloud responses to warming, respectively.After which we turn attention to the cooling experiments, Figures4 and 5show the temperature mediated cloud change in the solm4p and 0p5xCO2 experiments.The panels in Figures4 and 5follow the same format as Figure2, where the ISCCP simulator histograms have been separated into nine cloud categories.Note that these cloud changes are on a per temperature basis, so positive temperature mediated values in these cooling experiments correspond to cloud loss with global cooling, in contrast to positive temperature mediated values in Figures2 and 3which correspond to cloud increase with global warming.Sections 3.1.4and 3.1.5describe the differences between the cloud changes in solm4p and 0p5xCO2, and the differences in cloud changes occurring during warming and cooling, respectively.

Figure 3 .
Figure 3. Temperature mediated response of cloud from the solp4p experiment shown in the same form as Figure 2.

Figure 4 .
Figure 4. Temperature mediated response of cloud from the 0p5xCO2 experiment shown in the same form as Figure 2. Note that these cloud changes are on a per temperature basis, so positive temperature mediated values in these experiments corresponds to cloud loss during the simulations.

Figure 5 .
Figure 5. Temperature mediated response of cloud from the solm4p experiment shown in the same form as Figure 2 except that in the solm4p stippling indicates agreement from at least 2 out of 3 models (as opposed to 3 out of 4 for the other experiments).Note that these cloud changes are on a per temperature basis, so positive temperature mediated values in these experiments corresponds to cloud loss during the simulations.

Figure 6 .
Figure 6.Global mean temperature mediated cloud radiative feedbacks decomposed into the feedbacks due to three different types of cloud changes (and a residual term).Bars indicate the multi-model mean, black symbols mark the individual models.The top panel shows the solp4p and 4xCO2 experiments while the lower panel shows the solm4p and 0p5xCO2 experiments.In each plot the dotted line separates the SW components on the left from the LW components on the right.

Figure 7 .
Figure 7. (a-d) Temperature mediated change in surface temperature.Stippling indicates good model agreement on whether the change is above or below 1 K/K(global).(e-h) Temperature mediated changes in Estimated Inversion Strength (EIS), contours are the pre-industrial average EIS, red boxes denote known stratocumulus regimes.(i-l) Temperature mediated changes in relative humidity at 700 hPa.As with previous figures stippling indicates regions with good model agreement on the sign of the change of both EIS and RH 700 hPa.
we show maps of the temperature mediated change in 500 hPa vertical velocity (units are pressure change per day per kelvin), as well as the zonal mean 500 hPa vertical velocity averaged over years 10-150 of the simulations.Due to the per-temperature basis of the temperature mediated Journal of Geophysical Research: Atmospheres 10.1029/2023JD040296 changes the sign convention of the map plots of pressure velocity is such that in the warmed climate (solp4p and 4xCO2) regions with more upward motion or less subsidence are shown by negative values and purple colors, while in the cooled climate (solm4p and 0p5xCO2) the opposite is the case, more upward motion or less subsidence have positive values and green colors.Not surprisingly, the temperature mediated changes in 500 hPa vertical velocity correlates well with the high cloud changes shown in Figures2-5in the tropics and subtropics.In all experiments there is negative pressure velocity (increase in upward motion or decreased subsidence) per degree of warming (purple colors) in the central equatorial pacific (at least between 150 E to 150 W) and positive

Figure 8 .
Figure 8. (a-d) Multi-model means of 500 hPa vertical velocity.Contours are the pre-industrial climatology with contour intervals of 20 hPa/day.The coloration of these maps are the temperature mediated change in 500 hPa vertical velocity.(e)Pacific zonal mean climatologies of 500 hPa vertical velocity, averaged over ocean area from 120°W to 60°E and the final 30 years of each experiment.Shading indicates the 95% confidence interval found by bootstrap resampling the piControl simulations in 30-year chunks.Vertical dashed lines indicate the latitude of maximum mean upward velocity in each hemisphere.Note that in the Northern Hemisphere this occurs at the same latitude for the piControl and and for the solp4p and 4xCO2 simulations, and in the Southern Hemisphere this occurs at the same latitude for the solm4p and 0p5xCO2 simulations.

Figure 9 .
Figure 9. Zonal mean 500 hPa meridional streamfunction shown as averages over the final 30 years of each model experiment.We note that at most latitudes the lines representing the solp4p and 4xCO2 simulations overlap such that it appears only as one line.Shading represents 95% confidence intervals found by bootstrap resampling the piControl simulation in 30-year chunks.

Figure 10 .
Figure 10.(a-d) Multi-model mean change in cloud ice mass fraction shown as the average of the final 30 years of simulation subtracting the pre-industrial average.As in previous figures, stippling indicates regions where there is model agreement on the sign of the change.(e-h) Multi-model mean change in sea-ice fraction shown as the average of the final 30 years of simulation subtracting the pre-industrial average.

Table 1
Summary of Models and Data Used in This Analysis Along With Primary Citations for Each Model

Table 3
Hadley Circulation Width Metric in the Northern Hemisphere During DJF, Quantified as the Latitude Where the 500 hPa Meridional Stream Function Crosses Zero for the First Time in the Associated Hemisphere Note.These values are calculated by linearly interpolating the models' latitude grid to the point at which 500 hPa stream function of zero is crossed.Uncertainties are 95% confidence intervals found by bootstrap resampling the piControl simulation in 30-year chunks.

Table 4
Same as Table 3, but for the Southern Hemisphere and JJA Season

Table 5
Static Stability Averaged Over the Final 30 Years of Each Simulation in the Tropical WestFrom 600 to 200 hPa