Cloud Responses to Abrupt Solar and CO2 Forcing: 2. Adjustment to Forcing in Coupled Models

In this paper, we examine differences in cloud adjustments (often called rapid adjustments) that occur as a direct result of abruptly increasing the solar constant by 4% or abruptly quadrupling of atmospheric CO2. In doing so, we devise a novel method for calculating the cloud adjustments for the abrupt solar forcing simulations that uses differences between coupled model simulations with abrupt solar and CO2 forcing, in combination with uncoupled, atmosphere‐only, abrupt CO2 forced experiments that have prescribed sea‐surface temperature. Our main findings are that (a) there are substantial differences in the responses of stratocumulus and cumulus clouds to solar and CO2 forcing, which follow the differences in the direct radiative effect that solar and CO2 forcing have at cloud top, and (b) there are differences in the adjustment of the average optical depth of high clouds to solar and CO2 forcing that we speculate are driven by the differences in the vertical profile of radiative heating and differences in the pattern of sea‐surface temperature change (for a fixed global mean temperature). These cloud adjustments contribute significantly to the total net cloud radiative effect, even after 150 years of simulation.


Introduction
The climate is changing due to anthropogenic emissions of heat-trapping gasses, and our ability to predict the amount of surface warming that will occur depends critically on knowing how cloud optical depth, cloud-top height and cloud amount will change (Sherwood et al., 2020;Zelinka et al., 2020).How clouds will change can be decomposed into the sum of two components, a surface-temperature mediated change (the cloud change that is a function of global mean temperature anomaly and sea-surface temperature and sea ice change pattern) and a cloud adjustment that occurs directly due to the forcing agent, in our case from changes in insolation or atmospheric CO 2 concentration (Sherwood et al., 2015).In this paper we focus on the cloud adjustments, while in a companion paper (Aerenson & Marchand, 2023; hereafter Part I), we focus on the temperature mediated component.
As detailed in Part I, we analyze cloud changes 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).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 experiments respectively).In this paper, and Part I, we compare and contrast 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 CMIP6 (Eyring et al., 2016).We also use experiments from the Atmospheric Model Intercomparison Project (hereafter AMIP), in which atmosphere-only model configurations are run with the sea-surface temperatures prescribed to match reanalysis (Gates et al., 1999).Specifically, we use simulations where the atmospheric CO 2 is quadrupled without allowing for the sea-surface temperatures or sea-ice to adjust to the forcing.These simulations nominally allow us to estimate the adjustment that occurs directly from CO 2 increase independent of sea-surface temperature increase.There are however limitations to this method, in that the land temperature is allowed to warm, which introduces land-ocean temperature gradients (and associated monsoonal circulations) to the model simulation, which are not in-fact a direct response of the atmosphere to the forcing mechanism (Andrews, Smith, et al., 2021).In addition to the CMIP6 experiments, we use independently performed model simulations of the solp4p and 4xCO2 experiments from both coupled and atmosphere-only (AMIP-style) model integrations of the Community Earth System Model (CESM1), as well as simulations from CMIP5 generation models with 2xCO2 and a 2% increase of the solar constant in both the fully coupled and atmosphere-only (AMIP style) model configurations.These latter simulations serve as a testbed for the method we have developed to calculate the adjustment to solar forcing, without atmosphere-only integrations of solp4p from the CMIP6 models, which is described in detail in Section 2 of this paper.
Through this analysis, we seek to understand how cloud adjustments caused by solar and CO 2 forcing differ, and the underlying physical mechanisms.Studying the clouds' responses to abrupt solar and CO 2 forcing is an idealized framework that 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).Adjustments to CO 2 increase have been studied with a hierarchy of model simulations (e.g., Andrews et al., 2012;Chung & Soden, 2015;Gregory & Webb, 2008;Kamae & Watanabe, 2012;Larson & Portmann, 2016;Lutsko et al., 2022;Raghuraman et al., 2023;Schneider et al., 2019;Zelinka et al., 2013), where it was identified that cloud adjustments, and their radiative effect substantially impacts the equilibrium climate sensitivity of models (Andrews et al., 2012;Gregory & Webb, 2008), and even though there are some robust responses to CO 2 increase, such as a reduction in total cloud amount and positive cloud radiative effect anomaly, the adjustment component is a large source of spread among CMIP5 and CMIP6 projections (Kamae & Watanabe, 2012;Lutsko et al., 2022).Hence, it remains important to quantify and understand the ways that clouds adjust to forcing.There is also some recent observational evidence for cloud adjustments causing positive trends in shortwave cloud radiative effect.Raghuraman et al. (2023) use observations from Clouds and Earth's Radiatiant Energy System (CERES) Energy Balance and Filled (EBAF) satellite observational product to show that the current near-zero trend in net cloud radiative effect is due to the compensation of a positive shortwave effect from cloud adjustments, and a negative longwave effect from greenhouse gasses masking the effects of clouds in the longwave portion of the spectrum.
There are also a few previous studies which examine abrupt changes in solar forcing, and the differences in adjustments to different forcing agents.Smith et al. (2018) studied the adjustment to various forcing agents (including increasing the solar constant by 2% and doubling CO 2 ) using atmosphere only integrations of an ensemble of climate models (similar to the model configurations of the AMIP simulations).This allowed them to diagnose the adjustments from the various forcing changes.They found that the global mean adjustment of top-ofatmosphere radiation that results from cloud adjustments to solar forcing is of opposite sign from the cloud adjustment to CO 2 forcing.That is, after an increase in CO 2, cloud adjustments created a positive (warming) radiative forcing while increasing the solar constant produced a cloud response that contributed a negative (cooling) radiative forcing.Because of this difference in cloud adjustments, the top-of-atmosphere radiation imbalance is greater following CO 2 forcing than solar forcing.In our analysis, using a different method to diagnose cloud radiative effect, we find that the radiative adjustments to cloud following CO 2 and solar forcing are both positive (warming effect on the climate), however the adjustment is greater following the CO 2 forcing.We discuss this difference further in Section 4 of this article.Salvi et al. (2021) similarly studied the adjustment to various forcing agents in a model with prescribed seasurface temperature and sea-ice.They used offline radiative transfer calculations to find the expected change in the vertical profiles of radiative heating anomaly from each forcing agent and found that there were differences in the adjustment to solar and CO 2 forcing due to the radiative effect of CO 2 forcing being largest in the lower troposphere, while the radiative impact of solar forcing is nearly vertically uniform throughout the troposphere.Although it was not explicitly shown by Salvi et al. (2021), one expects the differences in the heating rate of the upper troposphere to impact the formation and lifetime of high clouds (Dinh et al., 2010;Gasparini et al., 2019;Seeley et al., 2019), as well as the static stability of the lower troposphere.
There are also adjustments to CO 2 and solar forcing over land which have received some attention.Evapotranspiration is an important moisture source over land, and the associated evaporative cooling is important for setting the climatological land temperature.Upon CO 2 increase, plant stomata do not open as wide, which reduces evapotranspiration rates (e.g., Betts et al., 1997;Cox et al., 1999;Field et al., 1995).In contrast, upon solar forcing increase one expects the increase in total SW radiation reaching the surface to increase photosynthesis (and evapotranspiration) rates (Mercado et al., 2009).In a comparison of experiments with CO 2 doubling and solar constant increase of 2.25% where the plant physiological effects of CO 2 are isolated from the radiative effects on the atmosphere Modak et al. (2016) found that the effect of CO 2 on evapotranspiration increases land surface warming on as short of timescales as 7-day following forcing (when little sea-surface temperature change has occurred).They find that the reduced evapotranspiration rate from CO 2 forcing causes less cloud occurrence over land after CO 2 forcing compared with solar forcing.
Additionally, there has been work done studying the effects of simultaneous solar and CO 2 forcing by Russotto and Ackerman (2018), who analyzed cloud changes in the Geoengineering Model Intercomparison Project (GeoMIP) G1 experiment, in which the CO 2 concentration is abruptly quadrupled while simultaneously the solar constant is decreased by an amount tuned so that the top-of-atmosphere radiation budget of each participating model has zero net radiative forcing (Kravitz et al., 2015).This required a decrease in the solar constant between 3.2% and 5.0% depending on the model.Russotto and Ackerman (2018) found that the immediate adjustments of clouds following the abrupt forcing was a vital component to determining how much solar forcing is required to balance the CO 2 forcing in each model.They found numerous cloud changes in the G1 experiment that contribute to the top-of-atmosphere radiation balance, such as a reduction of stratocumulus clouds associated with a decrease in inversion strength, and an increase of high clouds along the ITCZ and SPCZ.They did however recognize that understanding the underlying physical mechanisms responsible for the cloud changes would require simulations that perturb the CO 2 concentration and solar constant independently, as we do here.This paper is organized as follows: Section 2 contains a description of the model data, and methods used in this study, including a description of the method we use to calculate adjustment from coupled model simulations, and how we relate cloud changes to radiative flux using cloud radiative kernels.Then in Section 3 we present the results, including the cloud adjustment to solp4p and 4xCO2, the impact the cloud changes have on top-ofatmosphere radiative flux, and additional results which help interpret the physical mechanisms responsible for the adjustment difference between solp4p and 4xCO2.The results are discussed in the broader context of the existing literature in Section 4, and the main conclusions of this paper, and Part I are synthesized in Section 5.

Model Experiments
In CMIP6 a total of five modeling centers performed the solp4p and 4xCO2 experiments, as well as the AMIP experiment with abrupt quadrupling of CO 2 (hereafter referred to as AMIP-4xCO2).Details on the CMIP6 models are available in Part I.
Additionally, we use a set of independently performed simulations with the Community Earth System Model 1.2.1 Community Atmosphere Model 5.3 (hereafter referred to as CESM1) run at 1.9°latitude × 2.5°longitude resolution (Neale et al., 2012).From these simulations we have results of both the 4xCO2 and solp4p experiments, using both the fully coupled and atmosphere only (AMIP-style) simulations with prescribed sea-surface temperature and sea-ice.The addition of AMIP-style runs with solar and CO 2 forcing allow us to compare several techniques to calculate the cloud adjustments from the fully coupled CMIP6 solp4p simulations.In this independent set of CESM1 simulations there are three ensemble members of the 4xCO2 experiments, and single simulations for the other experiments.Data from these simulations were first published by Zhou et al. (2023) and are available for download at https://doi.org/10.5281/zenodo.7193943.
Lastly, we use output from model experiments that were requested for the Precipitation Drivers Response Model Intercomparison Project (PDRMIP), in which there are simulations of 2xCO2, and solp2p (abrupt doubling of CO 2 and 2% increase of the solar constant respectively) that were performed in both the fully coupled, and fixed sea-surface temperature and sea ice configurations.We also use these PDRMIP experiments as a testbed for the method we ultimately use to estimate the adjustment from the solp4p in the CMIP6 coupled model simulations.However, these experiments do not include the ISCCP satellite simulator outputs.A full description of the PDRMIP data is provided by Andrews, Boucher, et al. (2021) and Myhre et al. (2022), and these data are available at https://cicero.oslo.no/en/projects/pdrmip/pdrmip-data-access.

Methods and Theory of Adjustment Calculation
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 changes), and those that are independent of the global mean surface temperature (which are called the adjustments), as described by Equation 1, where C F (θ, ϕ, t) represents the cloud amount anomaly at a given latitude, longitude, and time in the simulation, A F (θ, ϕ) is the adjustment to the forcing change (F) at a certain latitude and longitude, ∆T (t) is the global mean surface temperature anomaly at a given time, M F (θ, ϕ, ∆T F (t)) is the temperature mediated component and is some function of the global mean temperature anomaly, and ε(t) represents internal variability which causes cloud changes which are due to neither the global mean temperature change or adjustments.In this paper, we are concerned with calculating the adjustment term A F (θ, ϕ) from a quadrupling of CO 2 and an increase of the solar constant by 4% (hereafter A CO2 (θ, ϕ) and A sol (θ, ϕ) respectively).
The temperature mediated changes are often approximated by a linear relationship to global mean surface temperature, such that M can be written as M(θ, ϕ, ∆T(t)) ≈ M(θ, ϕ)∆T(t) (e.g., Ceppi et al., 2017;Gregory et al., 2004;Zelinka et al., 2013) as is done in Part I, and ideally A F (θ, ϕ) is simply the intercept that is found by fitting a line to the simulated cloud anomaly as a function of ∆T.However, in truth this system is not completely linear, and for example, cloud amount depends on not-only the global mean temperature, but also the surface temperature pattern and associated dynamical circulations (Andrews et al., 2015(Andrews et al., , 2022;;Williams et al., 2008).
Although the linear model fits the total global mean cloud response well after the first couple decades following the abrupt forcing there can be large deviations from linearity (especially at the local grid-cell level) which make it problematic to obtain A F (θ, ϕ) as the intercept, or to interpret the intercept as the true adjustment (for discussion of this subject see Supporting Information S1).
We are not the first to recognize this problem, and to avoid reliance on a linear model, as well as to avoid errors that might result from internal variability (especially with variability on longer-than-annual timescales), the adjustment has often been calculated using model experiments that impose an abrupt forcing with sea-surface temperatures (SST) prescribed to match reanalysis such that the atmospheric adjustment is isolated from the effects of SST change, hereafter referred to as fixed-SST experiments (e.g., Colman & McAvaney, 2011;Forster et al., 2016;Gregory & Webb, 2008;Smith et al., 2018;Zelinka et al., 2013).Specifically, the adjustment term is calculated as the difference in cloud amount or radiative effect between fixed-SST experiment with and without the addition of forcing (i.e., AMIP-4xCO2 minus AMIP) over periods long enough to make the effect of internal variability small, typically 20-30 years.
As was noted in Section 1, this fixed-SST approach is not perfect because the land-surface is allowed to warm, which likely changes clouds over land, but also creates land-sea temperature contrasts that change atmospheric circulation patterns.Andrews, Smith, et al. (2021) compared the adjustments using fixed-SST experiments to those calculated in model experiments where all surface temperature is held constant during CO 2 quadrupling to understand the effect that land warming has on adjustment calculations.The impact of the fixed-SST approach on our results are discussed in Section 4.4.
At a practical level, our set of the solp4p simulations, contains no fixed-SST version, hence we derive a new method of calculating the adjustment to solp4p using a combination of coupled simulations of 4xCO2 and solp4p with the AMIP-4xCO2.In Supporting Information S1 we test a variety of methods to calculate the adjustment without an abrupt solar forcing experiment with fixed-SST based on a set of simulations from CESM1, for which we have simulations with and without fixed-SSTs for both the 4xCO2 and solp4p experiments.These methods include approximating the temperature mediated component of the cloud change to be linear such that the adjustment is found via (a) a simple intercept of a linear regression between cloud change and global mean surface temperature (as was done in Chung & Soden, 2015), or (b) where the difference between the adjustment to solar and CO 2 forcing is calculated by removing the linear-approximation temperature mediated cloud change from the total difference between the solar and CO 2 forced cloud changes.Both such approaches do not work better than the simple approach used here (as benchmarked against fixed-SST simulations).In Table S3 in Supporting Information S1 we provide a summary table that briefly describes each method and their limitations, as well as the root-mean-squared error of each benchmarked against the fixed-SST method.
The first step in our new method is to calculate the difference in cloud amount (and/or cloud radiative effect) from the long-term average (years 10-150 following abrupt forcing) between the 4xCO2 and solp4p coupled model simulations following Equation 2. Using a long climatology (such as 140 years) diminishes the impact of internal variability on the calculation.Hereafter we refer to this quantity (∆A sol CO2 (θ, ϕ)) as the adjustment difference.
Then, to estimate the adjustment solely due to solp4p we simply add the adjustment to 4xCO2 (calculated as the difference between the AMIP-4xCO2 experiment and AMIP experiment) to the adjustment difference, as shown in Equation 3. Hereafter, we will refer to this as the solp4p estimated adjustment or simply the solp4p adjustment.
One advantage of this method is that it does not require any assumption of the functional form of the temperature mediated cloud changes, and it can work even in the case that M(θ, ϕ, ∆T(t)) is not linear so long as the temperature response to both the 4xCO2 and solp4p forcings is the same, which in Part I we do show to be the case for the simulations analyzed here.
To demonstrate the effectiveness of Equation 3 in capturing the adjustment to solp4p Figure 1 shows the estimated adjustment of cloud amount to solp4p derived with Equation 3, and the fixed-SST derived adjustment of total cloud amount in CESM1 (top row), as well as cloud radiative effect (derived from top-of-atmosphere fluxes) adjustment in the solp2p experiments from both methods applied to the PDRMIP multi-model mean (lower three rows).We show the adjustment in total cloud amount in CESM1 to highlight the effectiveness of this method on estimating the adjustment in cloud amount, which is central to the following analysis of the CMIP6 model simulations.However, because different models define cloud differently and there are eight different models in the PDRMIP ensemble, we show instead the multi-model mean cloud radiative effect, where the cloud radiative effect has been calculated directly from each models' top of atmosphere fluxes.We stress that the solp2p results from PDRMIP are not comparable to the solp4p experiment, and are only used here to validate the estimated adjustment method.Comparing the panels in the left column of Figure 1, which are based on Equation 3 with the right column that are based on AMIP-style experiments demonstrates that the estimated adjustment obtained via Equation 3 captures many of the patterns which occur in the AMIP-derived adjustment.For example, in the CESM1 simulations, there is a reduction of cloud amount over the Indian Ocean, and Tropical Pacific, which is consistent across the two methods.Additionally, there is a decrease in cloud amount over North America, and an increase in cloud amount over Amazonia, Africa, and Southeast Asia.Likewise, the estimated adjustment is in good agreement with the fixed-SST derived adjustment in for the PDRMIP solp2p experiments (consisting of eight participating models).
Ultimately, we find Equation 3 to be superior to several other approaches that we tested for estimating the solp4p adjustment, qualitatively we find this method to be most effective at matching the pattens of the fixed-SST derived adjustment, and this method yields the lowest root mean squared error when compared to the fixed-SST adjustment.Details on all of the methods tested can be found in Supporting Information S1.We do of course also find that the adjustment difference between the 4xCO2 and solp4p in the coupled model simulations produces a difference pattern that is remarkably consistent with the difference in adjustment calculated using the fixed-SST simulations.These results are also provided in Supporting Information S1.
It is worth noting that this approach is not without its limitations.First, it hinges upon the global mean temperature change being nearly equal in the solp4p and 4xCO2 experiments.If the temperature difference were large, then there would be considerable temperature mediated changes aliased with the adjustment difference such that it would be quite difficult to untangle the two.Fortunately, solp4p has roughly the same amount of radiative forcing as 4xCO2, which produces nearly the same change in global mean surface temperature.In Part I we show that the solp4p and 4xCO2 experiments have roughly the same amount of global mean surface temperature change, similar warming patterns, and quite similar temperature mediated cloud changes during the 10-150 years timescale following the abrupt forcing change.So the error introduced to our adjustment difference from differences in the temperature mediated cloud changes are quite small when considering the solp4p and 4xCO2 experiments.In Supporting Information S1 we calculate the adjustment difference between simulations with an abrupt reduction of the solar constant by 4% and an abrupt halving of atmospheric CO 2 (referred to as solm4p and 0p5xCO2 respectively).Details on these simulations are available in Part I.There is roughly two times more cooling in the solm4p than the 0p5xCO2 simulations, because reducing the solar constant by 4% results in twice as much reduction in radiative forcing as halving atmospheric CO 2 .We do not have fixed-SST versions of these simulations, so we cannot test if our new method works for these simulations, however upon inspection it is apparent that many of the patterns seen in the adjustment difference between solm4p and 0p5xCO2 are in fact due to the temperature mediated changes.
Second, even the adjustment differences calculated from our method are not entirely independent of SST change.
Although the global mean temperature response in solp4p and 4xCO2 are quite similar, there is some difference in the warming pattern.We will see some adjustment differences in cloud that are likely due to differences in SST patterns in the solp4p and 4xCO2 adjustments.We consider this an important nuance of our method because this makes our method conceptually somewhat different from fixed-SST methods where the SST pattern is not allowed to change.
In the following sections, both the adjustment difference and the estimated adjustment to solar forcing are presented.Each metric has its own utility, as the adjustment difference highlights the ways in which the cloud responses to the two forcing mechanisms differ, while the solp4p estimated adjustment shows how clouds change only because of increase in the solar constant.

ISCCP Simulator and Cloud Radiative Kernels
To perform a comparison of cloud changes across models we make extensive use of the International Satellite Cloud Climatology Project (ISCCP) satellite simulator, which is part of the CFMIP Observation Simulator Package and has been embedded into many climate models (Bodas-Salcedo et al., 2011).The ISCCP simulator is designed to imitate the results of ISCCP retrievals of cloud-top-pressure (CTP) and cloud optical depth based on visible and infrared images collected by geostationary weather satellites.The actual observational data have been collected into an ongoing global cloud data sets that has been operational since 1983 (Rossow & Schiffer, 1991).
The ISCCP simulator parses total cloud fraction into CTP and cloud optical depth joint histograms, just as the ISCCP retrieval algorithm does.This allows for comparison of model clouds with observations, but also a comparison between models that is independent of each models' internal definition of "cloudiness".Zelinka et al. (2012a) calculated cloud radiative kernels to compute longwave (hereafter LW) and shortwave (hereafter SW) top-of-atmosphere radiative fluxes associated with cloud effects from the ISCCP histograms.
Using the radiative kernels, Zelinka et al. (2013) have examined cloud adjustment and temperature mediated response to 4xCO2 simulations from a collection of CMIP5 models.Here we undertake a similar analysis of the adjustment to solar and CO 2 forcing, and in order to understand the radiative impact that changes of cloud cover fraction (CF), cloud-top-height (CTH), and cloud optical depth (τ) have on top-of-atmosphere radiation balance, we perform a decomposition of the kernel-derived radiative effect into the radiative anomalies caused various cloud changes (as well as a small residual), following the method of Zelinka et al. (2012bZelinka et al. ( , 2013)).

Results
In this section we present the results showing how cloud properties adjust to solar and CO 2 forcing and briefly examine the cloud radiative effect the adjustments have on top-of-atmosphere radiation using cloud radiative kernels.In Section 4 we discuss the physical mechanisms that likely contribute to the adjustments, and as a prelude to that discussion we close this section with an examination of changes in several other atmosphere and surface variables (such as 500 hPa vertical velocity, and estimated inversion strength).

Adjustment of Cloud Properties to Solar and CO 2 Forcing
Figures 2 and 3 show the adjustment difference, and the adjustment to quadrupling CO 2 calculated from the AMIP-4xCO2 experiment (top nine panels of Figure 3), and the estimated adjustment to solp4p calculated following the approach described in Section 2.2 (lower nine panels of Figure 3).Following Zelinka et al. (2013), we calculated the 4xCO2 rapid adjustment as the anomaly in cloud amount of each category averaged over the duration of the simulations, from years 5-36, using fixed SST simulations; and the estimated adjustment to solp4p is calculated following Equation 3. We separate the cloud changes into nine categories of Cloud Top Pressure (CTP) and optical depth.Specifically the cloud optical depth is broken into three ranges: optically thin (τ ≤ 3.6), optically medium (3.6 < τ ≤ 23), and optically 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.We show the cloud changes as a multi-model mean, where stippling indicates regions where at least three out of four participating models agree on the sign of the cloud adjustment.Results from individual models are available in Supporting Information S1.The 4xCO2 adjustments shown here are directly comparable to the results of Zelinka et al. (2013), and any differences are due primarily to differences between the set of CMIP5 models used by Zelinka et al. (2013) and the CMIP6 models used here (with a small contribution from internal variability).In the following paragraphs we describe the high, mid-level, and low cloud changes in sequence.

High Clouds
The top row of Figure 2 shows the adjustment difference of high clouds, where it is apparent that in most areas and in the global mean, there is less optically thin high cloud in the solp4p than the 4xCO2 adjustment (orange color) and more optically thick high cloud (purple color).The reduction in optically thin clouds is greater than the increase in optically medium and optically thick clouds such that there is less total high cloud in the solp4p experiment (Note: adjustment differences summed across all optical depth ranges are shown in Supporting Information S1).
For optically thin high cloud the adjustment difference (Figure 2) is greatest in the Tropics, especially over the Indian Ocean, Tropical West Pacific, and along the Pacific and Atlantic ITCZ.This difference in optically thin high cloud occurs across all individual models (see Supporting Information S1 for individual model results).In Figure 3, we can see that this difference is largely due to a stronger reduction of optically thin-high cloud in the solp4p estimated adjustment.In the AMIP-4xCO2 adjustment there is an increase in global mean cloud in this category, with noteworthy increases over the Central Pacific and over Africa.There is reduction in the 4xCO2 adjustment Indian and Atlantic Oceans, South America, and the Eastern Pacific, but in all of these regions the reduction in solp4p estimated adjustment is much greater, and in general, there are very few regions where the solp4p estimated adjustment shows an increase in optically thin high cloud.
For optically thick high cloud, the adjustment difference (Figure 2) is greatest in the Tropical West Pacific, but there is also more optically thick high cloud (purple color) and good model agreement (stippling) in other regions, including over the Eastern Indian Ocean, and several midlatitude locations (especially along the Southern Ocean storm track between 40°and 60°S).In the Eastern Equatorial Pacific, and a part of the Equatorial Atlantic, there is more optically thick high cloud in the 4xCO2 than solp4p (orange color), although there is poor model agreement as regards this feature.The overall adjustments in optically thick high clouds for the individual solp4p and 4xCO2 adjustments (Figure 3) are similar.In the Tropical West Pacific and Indian Ocean, where the optically thick high cloud adjustment difference is greatest, both the 4xCO2 adjustment and solp4p estimated adjustment show a decrease in optically thick high cloud.So, both forcing mechanisms cause a decrease of cloud in this category, but the change is greater from CO 2 than solar forcing.Likewise, in the midlatitudes, both adjustments show a reduction of optically thick high cloud over much of the midlatitudes.Thus, we again find that the greatest adjustment difference occurs when both experiments yield a decline in cloud occurrence, but there is greater decrease of optically thick high cloud in 4xCO2 than solp4p.For optically medium high clouds, the adjustment difference (Figure 2) has fewer regions with good model agreement than the optically thin or thick high cloud categories.However, there is notably less cloud in this category in the solp4p than the 4xCO2 (orange color) in a small handful of regions including over the Indian Ocean, Atlantic ITCZ, Eastern Pacific portion of the ITCZ, Gulf Stream, and Kuroshio current.All of these regions have relatively warm SST (compared with surrounding regions), and are locations where deep convection is common.As was the case for high-thick clouds, the overall pattern of adjustments in optically medium clouds for the individual solp4p and 4xCO2 adjustments (Figure 3) are similar, but in this case, it is the reduction in solp4p that is generally larger and results in a negative adjustment difference in the Indian and Atlantic Oceans.Over the Kuroshio current, near the coast of Japan there is poor model agreement on the individual adjustment terms, and this is one of the few places where the adjustment difference appears to be more robust than the individual adjustment terms.

Mid-Level Clouds
Perhaps the most striking adjustment difference occurs in the mid-level cloud category (middle row of Figure 2), where there is positive difference (blue colors) in all optical thickness categories.This positive difference is widespread throughout the subtropics, midlatitudes and in the arctic.This difference is especially strong at medium optical depths, and in regions occupied by extensive stratocumulus off the western boundaries of continents.In the global mean (mean values are listed at the top of each panel), the adjustment difference is strongly positive in all three mid-level categories, with the change in thin cloud being stronger at higher latitudes.Looking at the estimated adjustments for the individual forcings (Figure 3), we see that in all three optical depth categories there is a substantial reduction in mid-level clouds associated with 4xCO2, that is especially strong in the stratocumulus regions, while the solp4p estimated adjustment shows an increase in the mid-level cloud over most oceans (and only weak decreases over land with little model agreement).

Low-Level Clouds
In contrast with mid-level clouds, optically thin and medium low clouds show a negative adjustment difference between solp4p and 4xCO2 (Figure 2, orange color) with good model agreement in the midlatitude oceans (at least for optically thin clouds) and marine stratocumulus regimes including much of the Southern Ocean.Looking at the 4xCO2 adjustment (Figure 3, third row) there is a ubiquitous decrease in optically thin low cloud over subtropical and midlatitude ocean, and in the optically medium category there is cloud loss over most ocean regions.In combination, the mid-and low-level changes suggest that the tops of clouds residing in and near the boundary layer over subtropical and midlatitude ocean lower and that these clouds become optically thinner in response to increased CO 2 .This marine cloud lowering and thinning is not present in the simulations with increased solar forcing, which show an increase in mid-level clouds over most ocean, and if anything, suggest a lifting and thickening of mid-and low-level clouds in some regions.Over most land on the other hand, there is decrease in low cloud in response to increased CO 2 in optically thin and medium cloud categories, at least in locations with good agreement between models (shown by stippling).The pattern of response to increases in solar forcing is similar to CO 2 over land, but with decreases in low-and mid-level cloud being stronger in the adjustment to CO 2 , resulting in a positive adjustment difference over land.We will discuss the mechanisms responsible for the cloud adjustment described here in Section 4, but before doing so we turn our attention to the radiative impact of these adjustments.

Top of Atmosphere Cloud Radiative Adjustments
The cloud adjustments previously described alter the Earth radiation budget, and thereby enhance or diminish the effective radiative forcing (depending on the change).The impact of clouds on the top-of-atmosphere radiation balance can be calculated in many ways such as directly from top-of-atmosphere radiation output (e.g., Su et al., 2010), Partial Radiative Perturbation (Taylor et al., 2007), or cloud radiative kernels (Zelinka et al., 2012a).
Here we use cloud radiative kernels because they are simple to use and provide a direct link with cloud changes described in Section 3.1.The cloud radiative anomaly from cloud radiative kernels are calculated directly from changes in the underlying cloud distribution and are independent of changes in non-cloud variables that impact top-of-atmosphere radiation and how it is impacted by cloud (Zelinka et al., 2013).On a minor note, we have multiplied the shortwave kernels by a factor of 1.04 in the solp4p calculations to account for the increased insolation.The effect of this adjustment is small and has no impact on the conclusions drawn from the application of cloud radiative kernels to these experiments as the differences noted between the cloud radiative anomaly changes in the different experiments are greater than this 4% change.
In Figure 4, we show the global mean radiative effect of the cloud adjustment calculated from the fixed-SST experiment for 4xCO2, and following Equation 3 for solp4p (hereafter referred to as cloud radiative adjustment), separated into the LW, SW, and NET radiative anomaly resulting from adjustments in CTP, optical depth, and Cloud Fraction following the Zelinka et al. (2012b) decomposition.In Supporting Information S1 we partition the cloud radiative adjustment into the contributions from low, mid-level, and high-cloud adjustments.
The NET adjustment is simply the sum of the LW and SW components.Additionally, in Supporting Information S1 we show a version of Figure 4 which has been calculated using a different method of estimating the adjustment, where the effects of differences in the global mean temperature change between the 4xCO2 and solp4p is removed using the average of the temperature mediated cloud feedback parameter in each experiment.
We provide this figure to show that in the global mean, this alternative method yields qualitatively similar results to the estimated adjustment method primarily used in this paper.
In the shortwave component (top row of Figure 4) the cloud radiative anomaly of both the solp4p estimated adjustment and AMIP-4xCO2 adjustment are positive across all models except for the adjustment of MRI-ESM2-0 to solp4p (in which there is a strong negative SW adjustment from mid-level clouds).In the multi-model mean (blue and orange bars) there is a greater positive SW radiative adjustment from 4xCO2 than solp4p.This is due to differences in the optical depth component of the SW radiative adjustment, where there is a positive adjustment to 4xCO2 that is consistent across models, and a negative SW adjustment in the solp4p.The difference is especially notable in the high cloud category (see Supporting Information S1).There is also considerable difference in the radiative effect of CF adjustments, where there is a more positive SW adjustment to solp4p than 4xCO2 for all models but MRI-ESM2-0 such that the multi-model mean SW adjustment is more positive to solp4p than 4xCO2.
The difference in CF cloud radiative adjustment is quite pronounced in both the low and high-cloud categories.Not surprisingly, CTP changes have little effect on the SW, and there is very little SW radiative adjustment from CTP.
In the longwave component (middle row of Figure 4) the 4xCO2 total cloud radiative adjustment is negative across all models, meaning that the LW cloud adjustment has a cooling effect on the climate.In the multi-model mean the total LW adjustment to solp4p is also negative (albeit less so than the adjustment to 4xCO2), however in HadGEM3-GC31-LL there is a positive total adjustment (which is mostly due to the CTP adjustment of that model to solp4p being far more positive than the other models', signifying that it has a net decrease in CTP or rising cloud tops), and in MRI-ESM2-0 there is a small but positive total LW adjustment, because in this model the adjustment to solp4p includes a global mean increase of low and mid-level cloud fraction.All models aside from MRI-ESM2-0 produce a more negative LW CF adjustment to solp4p than 4xCO2 (which, similar to the SW component, is due to CF adjustments in the low and high-cloud categories).There is also a difference in the LW optical depth adjustment to solp4p and 4xCO2.There is near-zero LW optical depth adjustment to 4xCO2 in all models, and a positive LW optical depth adjustment to solp4p (which we show in Supporting Information S1 is mostly due to the high-cloud changes).
In the bottom row of Figure 4, we show the shortwave and longwave components summed together as the NET cloud radiative adjustment.The total NET adjustment is positive in all models for 4xCO2, and all but MRI-ESM2-0 for solp4p.In the multi-model mean the total NET adjustment to 4xCO2 is more positive than to solp4p.However, we note that the inter-model spread is greater in the adjustment to solp4p, such that in IPSL-CM6A-LR and CanESM5, the total NET adjustment to 4xCO2 is in fact smaller than to solp4p.We find that the largest difference in the total NET adjustment comes from the optical depth adjustment of high clouds, where there is thickening due to solp4p, and thinning due to 4xCO2.
In Figure 5, we show spatial maps of the SW, LW, and NET total cloud radiative adjustment to solp4p and 4xCO2 and the cloud radiative adjustment difference to highlight some locations where the cloud radiative adjustment is noteworthy.In Supporting Information S1 we provide additional figures showing the radiative anomaly from CF, CTP, and optical depth adjustments broken down by low, mid-level, and high clouds.Beginning with the estimated adjustment to solp4p, there is positive adjustment over the Indian Ocean, the Tropical Atlantic, East Pacific, North America, most of Eurasia, the Southern Ocean, the Northern Pacific and Atlantic due to decrease in cloud fraction.In Supporting Information S1 we show that the positive cloud radiative adjustment to solp4p over the Indian, Atlantic, and East Pacific oceans is due to reduction in high cloud, we likewise find a negative LW adjustment in these regions (as is expected from high-cloud reduction).In contrast, the positive adjustments over Northern Hemisphere continents, and the North Atlantic and Pacific, and Southern Ocean are due to low-cloud reduction, in these locations there is little LW response, because low-clouds have similar emission temperature at cloud top as the surface, so they cause little LW radiative anomaly.Thus, when combined, the NET cloud radiative adjustment to solp4p is positive, and is strongest in regions with low-cloud reduction (over Southern Ocean, and Northern Midaltitude ocean and continents), with a negative contribution in the Tropical Atlantic and East Pacific (from CTP reduction).We also point out that in the NET there is some positive cloud radiative adjustment to solp4p over the Peruvian and Californian Stratocumulus where there is reduction of low-level cloud.
In the 4xCO2 many of the patterns of cloud radiative adjustment are similar to the solp4p (as is expected from the similarity in cloud adjustments shown in Figure 3), for instance, there is a positive SW and negative LW radiative adjustment over the Indian Ocean and Tropical Atlantic due to high-cloud CF change such that they sum to nearly zero NET radiative effect.There is also a positive SW response in the Southern Ocean and Stratocumulus regimes due to change in low-cloud CF.There is however, a positive NET radiative adjustment in stratocumulus regimes which is greater and more widespread than the adjustment to solp4p, and in contrast to the solp4p (where the stratocumulus adjustment is mostly from low-clouds), in the 4xCO2, this is due mostly to a reduction of mid-level cloud.
There are a handful of other key differences in the cloud radiative adjustments to solp4p and 4xCO2, which are shown in the bottom row of Figure 5.For instance, in the North Pacific, the radiative anomaly from low-cloud adjustment to solp4p is greater than that from 4xCO2, such that the adjustment difference is positive in the SW and the NET.There is additionally, a large negative SW and positive LW adjustment difference in the Tropical Pacific, which we show in Supporting Information S1 is due to high-cloud optical depth adjustment.In the Tropical Atlantic and Indian Oceans, there is a positive SW and negative LW adjustment difference, due to the greater high-cloud CF reduction that occurs in solp4p than 4xCO2.There is also a significant adjustment difference in stratocumulus regimes, where there is negative SW and positive LW adjustment difference due to a combination of low and mid-level CF adjustments.Over land surfaces, the adjustment difference varies by location, and has sparse model agreement.One region however with good model agreement is Northern Europe, where there is a negative SW adjustment difference, and a weak LW adjustment difference, such that the NET is negative.In Supporting Information S1, we show that this change is mostly due to mid-level and low clouds.
There is also good agreement on the adjustment difference over Northern Africa, where there is a weak positive SW adjustment difference, and a negative LW, such that the NET is negative due to changes in CF of high clouds (see Supporting Information S1).

Cloud Controlling Factors
In this subsection we apply the same formalism as before (Equation 3) to non-cloud variables that previous studies have shown influence clouds.Specifically, we calculated the adjustment difference as in Equation 4, where X is some non-cloud variable; and we calculate the solp4p estimated adjustment of X, as in Equation 5 by adding the adjustment calculated for 4xCO2 based on differencing averages of the AMIP and AMIP-4xCO2 experiments.As with the cloud changes, this method relies upon the fact that the temperature mediated changes of the non-cloud variables are the same from solar and CO 2 forcing.In Part I we show that this is indeed the case, and the temperature mediated changes of many variables are roughly the same in the solp4p and 4xCO2 experiments.Additionally, in Supporting Information S1 we provide side-by-side comparison of adjustments in the non-cloud variables that we show in Figures 6 and 7 calculated via Equation 5 and using a fixed-SST experiment from HadGEM3 that was performed as a part of PDRMIP.These results validate that many of the broad patterns of adjustment are similar when calculated via Equation 5 and fixed-SST experiments.However, there are some differences and the estimated adjustment method is likely not equally effective for all variables.We show the adjustments of these other non-cloud variables to aide in the interpretation of the cloud adjustments shown in Figures 2 and 3, but we recognize that there will likely be some differences between the adjustments shown in Figures 6 and 7 and similar adjustments calculated via fixed-SST experiments, and we comment in the following text where the method is likely a significant factor.
In Figure 6, we show the adjustment to solp4p and 4xCO2 in the surface temperature and 500 hPa vertical velocity.Not surprisingly, the 4xCO2 surface temperature adjustment shows significant increases over land and sea ice, and near zero temperature change over ocean (which is equivalent to sea-surface temperature and will hereafter be referred to as such).This result is inherent to the experimental design where 4xCO2 adjustment is calculated from fixed-SST simulations in which the sea-ice surface temperatures are fixed.The surface temperature adjustment difference, on the other hand, is not constrained to be zero at any location, and the adjustment difference includes differences in temperature pattern (which are not mediated by global mean temperature); and consequently, our solp4p estimated adjustment likewise includes the effects of deviations in the surface temperature from the 4xCO2 pattern.In particular, the adjustment difference plot in the bottom left panel of Figure 6 shows that in the solp4p there is more warming in the tropics and less warming in the poles as compared with 4xCO2.One can certainly interpret this change in surface temperatures as a limitation (or error) in our estimated solp4p adjustment technique, but in some respects, there is a philosophical question regarding what should or should not be considered an adjustment.We address this point further near the end of Section 4. Regardless, the point remains that increase in CO 2 and insolation result in slightly different patterns of surface warming.
The 500 hPa vertical velocity indicates changes in large scale circulation, where positive anomalies (green) indicate regions with diminished ascent or enhanced subsidence.The pattern of changes in the estimated adjustment to solp4p and 4xCO2 are similar, with (a) strong upward anomalies (purple colors) over most land equatorward of 40°latitude, (b) downward anomalies (green colors) over most tropical and subtropical oceanic regions of ascent (dashed contours) including the Tropical Atlantic and Indian Oceans, indicative of diminished ascent and (c) upward anomalies over most tropical and subtropical oceanic regions of descent (solid contours) including the subtropical Pacific (20°-40°latitude), and the subtropical Indian Ocean ( 20°to 40°latitude), indicative of diminished subsidence.There is mixture of upward and downward anomalies over the midlatitude oceans (latitudes poleward of 40°) depending on the location; with perhaps the strongest and most significant feature being downward anomalies over much of the eastern North Pacific and North Atlantic.
Figure 7 shows the adjustment of Estimated Inversion Strength (hereafter referred to as EIS) in the left column, and relative humidity at 700 hPa in the right column.Over ocean, EIS is well correlated with the global low cloud occurrence in observations and models (Qu et al., 2014;Wood & Bretherton, 2006).The 4xCO2 EIS adjustment (middle panel) shows increasing inversion strength over most ocean areas, especially at mid-latitudes.This result is expected because EIS depends on the difference in potential temperature between the surface and 700 hPa, and thus heating the troposphere while fixing SST will inherently increase the inversion strength.Figure 6 shows there is more warming in the tropics and subtropics in the solp4p experiment at the surface, and at 700 hPa the greater warming in the solp4p experiments extends to 60°N and S (see Supporting Information S1 for the 700 hPa potential temperature adjustment).Over land and sea ice, the surface temperature is not fixed in the AMIP-4xCO2 simulations and there is much greater surface warming over land in the Northern Hemisphere Midlatitudes in the 4xCO2 experiment than solp4p which is associated with reductions in EIS.We stress that the increase in heating over land (where surface temperatures are not held fixed) also affects EIS over neighboring ocean.Kamae et al. (2019), for example, used model experiments where SST and land warming were held fixed to isolate the atmospheric adjustments from the effects of land warming.They find that atmospheric adjustments to 4xCO2 over ocean are significantly influenced by the land warming (and vary seasonally).One might argue as to what degree land warming or the pattern of surface heating should be included in the adjustment, and we discuss this further in Section 4.4.
Turning attention to the relative humidity at 700 hPa (hereafter RH_700), the right panels of Figure 7 indicate the moisture availability of the free troposphere.The moisture difference between the surface and the free troposphere has a large impact on the efficiency of turbulent entrainment-driven drying of the boundary layer and has a large effect on the occurrence of low and mid-level clouds.The right panels of Figure 7 show that in both the adjustment to solp4p and 4xCO2 there is a reduction of RH_700 in the midlatitude and polar regions (poleward of 30°latitude).There is also an increase of RH_700 over Central Africa, the Eastern Equatorial Pacific, and parts of the subtropical Pacific (especially in the northern hemisphere).While the patterns are similar, there is a positive adjustment difference (bottom panel) in nearly all regions further than 10°from the equator, with especially large difference in the subtropics, meaning that the free troposphere is less dry in solp4p than 4xCO2 in the subtropics in particular.Looking at the 500 hPa vertical velocity adjustment difference (right column of Figure 6) there is broadly similar pattern with slower descent over most of the subtropics in solp4p than 4xCO2.Hence, we speculate that the slower mean-descent in solp4p results in less drying via subsidence than 4xCO2.Changes in EIS are having little role in the difference drying between solp4p and 4xCO2 over most ocean areas, and especially in the subtropics where there is very little EIS adjustment difference.

Tropical Temperature Profile
The large difference in the adjustment of optically thin high-level cloud between the solp4p and 4xCO2 simulations draws attention to potential differences in the temperature profiles in the tropical pacific.Figure 8 shows vertical profiles of temperature and equivalent potential temperature in the Indian Ocean and Tropical West Pacific (60-180°longitude and 15 to 15°latitude) for the solp4p and 4xCO2 experiments, as well as the adjustment difference between the equivalent potential temperature from the two experiments.The equivalent potential temperature is a particularly useful metric here because it is conserved under moist adiabatic processes, such that when looking at the equivalent potential temperature differences in surface temperature between models and experiments are not amplified aloft simply due to changes in the moist adiabatic lapse rate.We have isolated the Indian Ocean and West Pacific because it is a large region of mean-state ascent, and in Figure 2 we show that there is a large difference in the occurrence of optically thin high cloud between the solp4p and 4xCO2 in the Indian and Western Pacific Oceans.Additionally, gravity waves caused by deep convection homogenize the temperatures aloft such that the temperature aloft throughout the tropics is set by the temperature profile in regions of ascent (Bretherton & Smolarkiewicz, 1989;Mapes, 1993;Sobel & Bretherton, 2000).In regions of ascent, the temperature profile is typically near that of a moist adiabat rising from the surface, such that temperature variations at the surface tend to be amplified aloft.As can be seen in the right-hand panel of Figure 8 the upper atmosphere is warmer in the solp4p than the 4xCO2 in the multi-model mean, and the difference is larger in the upper atmosphere than at the surface.In IPSL-CM6A-LR the surface is in fact cooler in the solp4p than 4xCO2 (albeit only slightly), but the upper-atmosphere is warmer.The middle plot of Figure 8 shows that in all experiments, the portion of the troposphere above about 300 hPa is warmer than would be expected from the moist adiabatic lapse rate beginning at the surface (meaning that the temperature profile is warmer than a constant equivalent potential temperature beginning at the surface).In the solp4p simulations, this portion of the temperature profile is warmer than in the 4xCO2 simulations, so in the solp4p experiment the upper-troposphere in tropical regions of ascent is further from a moist adiabat starting at the surface than in the 4xCO2 experiment.
There are two aspects of the solar and CO 2 forcing mechanisms that likely contribute to the differences in the tropical temperature profile.(a) As previously noted, solar forcing is most effective at low latitudes, where insolation is strongest, so even when the global mean surface temperature change is the same, solar forcing causes greater tropical temperature increase than CO 2 (Kaur et al., 2023), and (b) CO 2 forcing preferentially heats the Journal of Geophysical Research: Atmospheres 10.1029/2023JD040297 lower troposphere, while solar forcing induces anomalous heating which is homogenous through the troposphere (Salvi et al., 2021), hence solar forcing warms the upper atmosphere more efficiently than CO 2 forcing.

Discussion
In this section we discuss the cloud adjustment to solp4p and 4xCO2 in the context of previous literature and the adjustment of cloud controlling factors shown in Sections 3.3 and 3.4.In Sections 4.1 and 4.2 we focus on high clouds and low and mid-level clouds respectively and hypothesize on the mechanisms contributing to the cloud adjustments.Then in Section 4.3 we discuss our findings in the context of previous studies on cloud adjustment to solar and CO 2 forcing.Finally, in Section 4.4 we discuss the possible limitations of the methods used in our study, and the ways that future work on this topic could reduce the uncertainty in their estimations of adjustment.

High Clouds
In the high cloud adjustments to both 4xCO2 and solp4p there is a decrease in high cloud fraction at all optical depths in the Tropical West Pacific, Tropical Atlantic, midlatitude oceans (between 20°and 60°latitude), and the eastern portion of Amazonia.There is increase in high cloud over the central Pacific (especially for medium and high optical thickness), and over tropical land masses such as Africa and Southeast Asia.The patterns of change are quite similar between the adjustment to 4xCO2 and solp4p (for at least optically medium and optically thick high cloud), and as one might expect, there is a strong correspondence of these changes with adjustments in the 500 hPa vertical velocity described in Section 3.3 (Figure 6).In short, there is a decrease in high cloud fraction where there are positive anomalies in the 500 hPa vertical pressure velocity (either increased downward motion or decreased upward motion) indicative of regions with diminished convection or enhanced subsidence and vice versa for negative anomalies.
Despite the similarity in the overall pattern, there are distinct differences between the cloud adjustments to solp4p and 4xCO2, and we focus the remaining discussion in this section on these differences.Specifically, we focus on the adjustment difference of optically thin, and optically medium and thick high clouds respectively.
Optically thin: There are fewer optically thin high clouds in the solp4p than in the 4xCO2 experiment.Many optically thin high clouds form via horizontal detrainment from deep cumulonimbus convective clouds, where moisture detrains horizontally in anvils that either directly form thin cirrus clouds or deliver moisture to the upper atmosphere that can form clouds in response to lifting by a variety of dynamical mechanisms including gravity and kelvin waves (Immler et al., 2008;Spichtinger et al., 2005).Cirrus clouds can exist in the upper-troposphere for a long time because the cold temperatures maintain slow sublimations rates of cloud particles (Seeley et al., 2019), and a circulation induced by differential radiative heating at cloud base and cloud top advects water vapor into the cloud, maintaining ice-crystal growth even in the presence of radiative heating (Dinh et al., 2010).Seeley et al. (2019) use cloud resolving simulations to show that, to first order, cloud lifetime in the upper troposphere is determined by the lifetime of condensate, and thus, the upper-tropospheric temperature.Solar forcing can be expected to heat the upper troposphere more than CO 2 (Salvi et al., 2021), and indeed we find the upper troposphere is warmer in the solp4p experiment than in the 4xCO2 experiment (Figure 8).We hypothesize this diminishes the saturation deficit through diabatic heating of the upper-troposphere without any additional supply of water vapor and the Clausius-Clapeyron relationship.This then leads to a higher sublimation rate of high-cloud particles, and consequently shorter-lived anvil clouds and less thin cloud in solp4p than in the 4xCO2 experiment.Increased LW heating at cloud base in the 4xCO2 experiment could increase turbulent mixing and in principle might also prolong high cloud lifetime in this experiment (consistent with a smaller loss of high thin cloud), but such turbulent mixing occurs on spatial scales that are not resolved by climate models, and so is not a factor in the present results.There is also the possibility that dynamical effects contribute to the optically-thin high altitude cloud adjustment difference.In the Tropical West Pacific there is actually a greater ascent rate in solp4p than 4xCO2 (see Figure 6), and less optically-thin high cloud.If dynamical effects were a leading cause in the adjustment difference of optically thin high clouds, one would expect there to be more cloud in the experiment with a greater ascent rate.Seeing as this is not the case, we conclude that the large scale dynamics are less important than the effects of local heating of the upper-troposphere for determining the optically-thin high cloud adjustment difference.

Optically medium and thick high clouds:
We find more optically medium and thick high clouds occur in solp4p than 4xCO2, especially in regions of ascent such as the Tropical West Pacific, ITCZ and SPCZ (positive Journal of Geophysical Research: Atmospheres 10.1029/2023JD040297 adjustment difference in Figure 2).This difference is largely due to a smaller loss of optically medium and thick clouds in the solp4p experiment compared with the 4xCO2 experiment in the Atlantic and Tropical West Pacific, however in the ITCZ and SPCZ, the change in optically medium and thick high clouds has poor agreement among models in the adjustment to solp4p and 4xCO2, yet there is good agreement on the adjustment difference of high medium and thick clouds in these regions.
We show in Figure 8 that there is a positive adjustment difference of sea-surface temperature in the Tropical West Pacific, ITCZ and SPCZ.Higher sea-surface temperature potentially provides more latent and sensible heat release into the lower atmosphere of solp4p, destabilizing the atmosphere.Certainly, the adjustment difference in mean 500 hPa vertical velocity (Figure 6 bottom panels), shows a smaller reduction in vertical velocities in the Tropical West Pacific, ITCZ and SPCZ ascent regions, and much of the mid-latitudes in the solp4p than in the 4xCO2 experiment.This suggests that the difference in optically medium and thick high cloud is linked to differences in the strength of the circulation response-a dynamical difference likely resulting from differences in the pattern of surface heating (rather than a direct radiative response).

Low and Mid-Level Clouds
Perhaps the most striking difference between the solp4p and 4xCO2 adjustments is the large and widespread decrease in mid-level clouds in the 4xCO2 adjustment as compared to the increase in mid-level clouds in the solp4p adjustment.This is perhaps most clear in the plot of the adjustment difference.Figure 2 shows that this difference in mid-level cloud response is widespread and occurs over both land and ocean.The mid-level adjustment difference is largest for optically medium clouds and is especially large over regions occupied by marine stratocumulus clouds; and there is a corresponding adjustment difference of low-level clouds (at least for optically medium and thin low-level clouds) that is of the opposite sign over land and most oceanic areas.We will return to clouds over land momentarily, and first focus on marine cloud.
Although marine stratocumulus and cumulus clouds are often thought of as low clouds, the tops of these clouds sometimes reach altitudes where the pressure is measured below (at a higher altitude than) 680 hPa in both models and observations (Tselioudis et al., 2021;Zelinka et al., 2022).As such, dynamic and thermodynamic changes to stratocumulus and cumulus clouds can have apparent impacts on both the low and mid-level cloud category.Taken as a whole, we interpret the combination of ISCCP low and mid-level cloud changes to mean that there is a net increase in the cloud-top-height (CTH) in boundary layer marine clouds in the solp4p adjustment and the opposite (a reduction of CTH) in most marine clouds in the 4xCO2 adjustment.
In trying to understand these low and mid-level cloud changes we follow the framework of Bretherton (2015), who attribute changes in boundary layer clouds (especially stratocumulus) to four primary mechanisms: (a) The radiative effect of water vapor and CO 2 in the free troposphere, where increases in water vapor or CO 2 warms cloud-tops and results in a lowering of cloud-top and a thinning or reduction in cloud amount.(b) The dynamic effect related to changes in subsidence where a decrease in subsidence results in rising cloud-tops and a thickening or increase in cloud amount.(c) the thermodynamic effect related to changes in surface temperature and free tropospheric relative humidity where an increase in surface temperature or decrease in free tropospheric relative humidity results in thinning or decrease in cloud amount.And finally (d) the stability effect related to changes in inversion strength where strengthening inversions result in a lowering of cloud-top and thickening or increase in cloud amount.In the following paragraphs we first discuss the cloud adjustments to solp4p and 4xCO2 individually, before turning attention to the adjustment differences.
Marine Clouds Solp4p: In the low and mid-level cloud adjustment to solp4p (Figure 3) there is a lifting of cloud top and a net increase in the low and mid-level cloud amount in stratocumulus regimes.We find in Figure 6 that there is also subsidence decrease in stratocumulus regimes (upward vertical velocity adjustments in regions with climatological subsidence).The dynamic effect predicts an increase in cloud-top-height and cloud amount with decreasing subsidence.Hence, we find that the reduction in low cloud and increase in mid-level cloud adjustment in stratocumulus regions is consistent with decreases in subsidence rate to solp4p.This effect appears to play a critical role in the total cloud response, as none of the other Bretherton (2015) effects explain the increase in CTH.There is a reduction in 700 hPa relative humidity (top right panel of Figure 7) which is expected to thin or reduce cloud amount via the thermodynamic effect, though in general, the reduction in relative humidity is not strong in stratocumulus regions.The increase in solar flux will only slightly warm cloud-tops such that the radiative effect of solp4p is likely to be small and would also be expected to cause cloud thinning or decrease in cloud amount and a lowering of cloud tops.There is a strong increase in EIS in stratocumulus and trade-wind regions, which is consistent with the net increase in low and mid-level cloud fraction in these regions, however the stability effect also predicts a decrease in CTH, which is again opposite what we find in the stratocumulus regions in the solp4p experiment.Of course, one expects that all the effects described by Bretherton (2015) occur to varying degrees; nonetheless it appears that the dynamic effect is having a large impact in the subtropical stratocumulus dominated regions in the cloud adjustment to solp4p.
In parts of the midlatitudes such as the Southern Indian Ocean, the North Pacific, and the North Atlantic, there are increases in both low-and mid-level cloud amount (Figure 3).For example, in the North Pacific, the low cloud adjustment (averaged from 150°to 220°longitude, and 40°-60°latitude) is 0.89% and the mid-level cloud adjustment is 0.30%, so the increase in low clouds is greater than mid-level clouds in these locations.In combination with the large positive adjustment in EIS, this suggests that the stability effect is playing a stronger role in the solp4p cloud adjustment at these higher latitudes, though a smaller (offsetting) contribution from the thermodynamic effect (as there is more surface warming in the adjustment at lower latitudes) is also likely a factor in the different response in solp4p between mid-latitudes and the subtropics.
Marine Clouds 4xCO2: In the adjustment to 4xCO2 there is widespread decrease in mid-level cloud and increase in low-level cloud in most oceanic regions such that when the low-and mid-level cloud is combined there is net decrease in CTH and reduction in cloud amount.These cloud changes occur over effectively all ocean surfaces but are greatest over stratocumulus regimes.Of the four mechanisms from Bretherton (2015), only the radiative effect predicts that with increasing CO 2 there will be a lowering and thinning or reduction of boundary layer clouds consistent with the cloud adjustment in stratocumulus regions.And the radiative effect from CO 2 increase has been studied using high-resolution large-eddy simulating models to show that there is in fact a certain CO 2 threshold (for a fixed subsidence rate) that when surpassed causes stratocumulus decks to dissipate into opencumuli (Schneider et al., 2019), resulting in decreased cloud fraction and cloud-top-height.There is widespread decrease in relative humidity at 700 hPa, and no change in sea-surface temperature (by experimental design).The thermodynamic effect predicts thinning or decrease of cloud with free-tropospheric drying (when there is no sea-surface temperature change).There is also increase in EIS over midlatitude oceans and marinestratocumulus regions, and the stability effect leads to thickening or increase in cloud amount and CTH reduction with increasing EIS.We in fact, do find that the adjustment to 4xCO2 includes a decrease in CTH of stratocumulus and cumulus clouds, and in the trade-wind regions there is increase of medium and thin clouds (when summed together).Also, like the solp4p, the adjustment to 4xCO2 includes weakening of subsidence (upward anomalies in regions of mean-state subsidence in Figure 6) over the Californian and Australian stratocumulus regimes, so in these locations the dynamic effect is counter-acting the radiative effect and is likely damping the thinning and decreasing CTH of stratocumulus cloud shown in Figure 3.As in the solp4p, one expects that each of the four mechanisms contribute to the total cloud changes in different locations simultaneously.We find that the cloud adjustment to 4xCO2 in stratocumulus regions is quite consistent with that expected from the radiative effect due to the decrease in medium optical depth mid-level cloud and (lesser) increase in optically thin low-level cloud.However, we expect that there are counter-acting effects from the stability and dynamic effects, and some contribution to the cloud thinning and decrease from the thermodynamic effect.In the trade-wind regions the cloud changes are most consistent with the stability effect due to the decrease in CTH and increase in cloud amount, which is likely damped by the thermodynamic effect.
In both the adjustment to solp4p and 4xCO2, there is widespread increase in EIS.This is an expected result, because (as was previously mentioned) EIS depends on the difference in potential temperature between the surface and 700 hPa, so when SST is fixed, any atmospheric heating will inherently increase the inversion strength.Kamae et al. (2019) used model experiments where SST and land warming were held fixed to isolate the atmospheric adjustments from the effects of land warming.They find that atmospheric adjustments to 4xCO2 (without land warming) cause increased summertime EIS over midlatitude oceans and increased wintertime EIS in stratocumulus regions.They additionally find that land warming increases summertime EIS over midlatitude ocean but has little impact on EIS in stratocumulus regions.Hence, we expect that the EIS increase we see in both solp4p and 4xCO2 are due to a combination of land warming, and the adjustment of the atmosphere with fixed-SST.
Marine Clouds Adjustment Difference: As described at the beginning of this section, there is striking and widespread adjustment difference between solp4p and 4xCO2 experiments, with more mid-level cloud in most marine areas (including the stratocumulus regimes, mid-latitude and polar oceans) in the solp4p experiment relative to the 4xCO2 experiment, and a corresponding adjustment difference of low-level clouds (at least optically medium and thin low-level clouds) of the opposite sign.We expect that the radiative effect of CO 2 is likely the primary contributor to the adjustment difference in low and mid-level cloud for two reasons.First, unlike CO 2 , solar forcing has little impact on cloud top radiative cooling.While there is likely some cloud adjustment to solp4p originating from the diabatic solar heating of clouds, this will be small compared to the effect of quadrupling CO 2 , which will significantly reduce cloud top longwave radiative cooling.So, in short, one expects the radiative effect of 4xCO 2 to be much larger.Second, the changes in the other cloud controlling factors do not match the broad pattern of adjustment difference.Specifically, the patterns of 500 hPa vertical velocity and EIS adjustments are broadly similar between the two forcing experiments.So, while there are small adjustment differences in the 500 hPa vertical velocity and EIS that almost certainly contribute somewhat to the cloud adjustment difference in some regions, the associated dynamic and stability effects seem unlikely to explain the broad pattern of the adjustment difference.Similarly, changes in SST, which are possible in the adjustment difference because of the approach we use (and which one might consider an error or limitation of the approachsee Section 4.4 for discussion of this point), are small in the midlatitudes and do not match the pattern of the cloud adjustment differences.
Land: To this point our discussion has focused on mid and low-level marine clouds for which the Bretherton (2015) framework is applicable.We now shift our focus to the cloud adjustments over land surfaces.Land warming in fixed-SST experiments certainly has a large influence on cloud adjustments over land via thermally induced circulations caused by the land-sea temperature gradient (Andrews, Smith, et al., 2021).We view this as a limitation of the fixed-SST approach, and as such, focus our discussion of land adjustments on the adjustment difference which does not rely on fixed-SST methods.In contrast to the marine cloud changes, over most land areas there are positive difference in cloud amounts (meaning more low cloud occurring following solp4p than 4xCO2-see purple colors in Figure 2) in all optical depth and CTP categories, except for optically thin high cloud.The larger reduction in optically thin high clouds occurs over both land and ocean and appears (we speculate) to be a response that is not specific to land (see Section 4.1).In general, the high and mid-level cloud adjustments are broadly similar over land and ocean.This contrasts with the case of optically thin and medium low-level clouds, where there is persistently more low cloud over land in the solp4p than 4xCO2 experiment and the opposite (fewer low clouds) over ocean.Admittedly, there is poor model agreement on low cloud changes over land, but the delineation between land and ocean is distinct.
Over land, one of the primary sources of moisture is the latent heat fluxed from the biosphere into the atmosphere via evapotranspiration.In plant physiology there is a well-established effect of CO 2 increase where plant stomata do not open as wide, reducing the transfer of moisture and energy to the atmosphere through evapotranspiration, which we hereafter refer to as the plant physiological effect (e.g., Betts et al., 1997;Cox et al., 1999;Field et al., 1995).There is also the effect of solar forcing on evapotranspiration; increase in the amount of total SW radiation reaching the surface causes photosynthesis (and evapotranspiration) rates to increase (Mercado et al., 2009).In Table 1, we show the adjustment difference in the global mean latent heat release from land surface.There is greater latent heat release in the solar forcing experiments than the 4xCO2, which we speculate contributes to the greater amount of low and mid-level cloud simulated over most land areas in solp4p compared with 4xCO2 via the plant physiological effect.Chadwick et al. (2019) performed model simulations which separate the influences of 4xCO2 on land precipitation into the component that is due to land warming (in a fixed-SST experiment) and the plant physiological effect.They find that the plant physiological effect decreases precipitation over most land areas because of its impact on moisture availability.Along the equator in portions of Africa and South America (in the only two locations coincident with a negative adjustment difference of low-level cloud) they find an increase in precipitation, because of how the plant physiological effect impacts local surface temperature and causes surface convergence.Thus, we speculate that the cloud adjustment difference shown in Figure 2 is likely a combined effect of CO 2 and solar forcing having opposite effects on evapotranspiration rates (and thus moisture availability) over most land areas, and the impact the plant physiological effect to CO 2 can have on dynamics in the tropics (specifically causing surface convergence).

Comparison With Previous Literature
As mentioned in the introduction, there are a handful of recent studies which have posed relevant questions to this study.
First, in the GeoMIP G1 experiment CO 2 is abruptly quadrupled, and solar forcing is reduced such that the net global radiative forcing is zero, thus there is no global mean temperature change, so the total cloud response is equivalent to an adjustment to simultaneous solar and CO 2 forcing.Russotto and Ackerman (2018) examined the cloud changes in these experiments and found a reduction of stratocumulus clouds associated with a decrease in inversion strength, and an increase of high clouds along the ITCZ and SPCZ, and in the Indian Ocean.We similarly find a reduction of stratocumulus clouds from 4xCO2 that is not matched by the solp4p.However, we conclude that the role of EIS is in fact secondary in this adjustment difference and the direct radiative effect of solar and CO 2 forcing (where CO 2 more efficiently warms cloud tops and reduces LW cooling) is the primary driver.Concerning high clouds, we find that there is a negative adjustment difference between solp4p and 4xCO2 of optically thin high cloud, and positive adjustment difference of optically medium and thick high cloud, such that when combined, there is a negative adjustment difference of all high clouds that is largest in the Indian Ocean, ITCZ and SPCZ (see Supporting Information S1 for combined figure).Thus, the cloud adjustments we find are consistent with the findings of Russotto and Ackerman (2018) for high clouds.
There is also the work of Salvi et al. (2021), which examines the adjustment to vertically localized heating experiments, to understand how the vertical heating profile of various forcing mechanisms (including solar and CO 2 forcing) impact cloud adjustment.They find that solar forcing (which is more top-heavy than CO 2 ) increases the amount of low cloud by increasing the strength of the boundary layer inversion.The effect is evident in the experiment analyzed here.Indeed, the solar forcing is more effective in warming the free troposphere, capping the moisture in the boundary layer.This effect is shown as the strengthening inversions in the solp4p experiment (Figure 8).
Through their study of the adjustment to a range of forcing agents simulated in PDRMIP Smith et al. (2018) found that cloud adjustments to CO 2 increase contributes a global mean positive net radiative forcing, while the cloud adjustment to solar forcing contributes a global mean negative net radiative forcing.Using the cloud radiative kernels, however, we find that there is a global mean net cloud radiative adjustment that is positive (0.37 and 0.86 W/m 2 respectively) for both solp4p and 4xCO2.Our finding that both forcing mechanism cause positive cloud radiative adjustments contrasts the Smith et al. (2018) result of opposite sign adjustments.It is unclear if the discrepancy between our results and those of Smith et al. (2018) are due to the different set of models used in each study, differences in the response to 2xCO2 and 2% solar forcing versus 4xCO2 and 4% increase in solar forcing, or if it is related to the method we use to calculate the estimated adjustment of solp4p.Regardless, this discrepancy highlights the importance of further constraining cloud adjustments (and not focusing only on the temperature mediated feedbacks), documenting cloud adjustments, and understanding of the underlying physical mechanisms.The latter of which, is especially important if we are to relate the results of process and regional models to the total cloud response.And more generally, it seems likely that differences between models are (to some extent) likely due to bias in the models' initial state.For example, two models with roughly the same response of stratocumulus to the radiative impact of CO 2 will have very different global mean response if one model has twice the stratocumulus as the other.
Regarding the differences in cloud adjustments to solar and CO 2 forcing over land, Modak et al. (2016) find that the plant physiological effect of CO 2 causes there to be more clouds in the adjustment to solar forcing than CO 2 .We indeed, find a similar result in the adjustment difference over land, where there is a positive adjustment difference in low-cloud.

Limitations of Estimated Adjustment Method
In this paper, we have relied on forced fixed-SST simulations to calculate the adjustments to the forcing in the 4xCO2 experiment and subsequently to estimate the adjustments for the solp4p experiment.This fixed-SST method has been widely used in previous studies (e.g., Colman & McAvaney, 2011;Gregory & Webb, 2008;Smith et al., 2018;Zelinka et al., 2013), but is limited in that, while the sea surface temperature and the location of sea ice are fixed, the land surface is allowed to warm.A warming land surface does of course cause changes in atmospheric circulations to occur, and the global mean surface temperature is not constrained to be zero.Andrews, Smith, et al. (2021) compared AMIP-4xCO2 experiments with 4xCO2 experiments where both land and sea-surface temperatures were fixed.Many of the cloud adjustments in Figure 3 are consistent with the adjustment that Andrews, Smith, et al. (2021) found are due to land warming.This includes a decrease in high cloud amount over the Atlantic and Indian Oceans, an increase in high cloud over land masses along the equator (especially over Central Africa), an increase in low cloud over midlatitude oceans (North Atlantic, North Pacific, and Southern Indian and Pacific Oceans), and a decrease in low-level cloud over continents especially in the optically medium low cloud category.
Our calculated adjustment difference does not rely on fixed-SST simulations; however, the adjustment difference is impacted by differences in warming pattern between solp4p and 4xCO2.More broadly, we stress that our estimated adjustment does not work for all model experiments, and in fact is only effective because (a) in solp4p there is a similar amount of warming as 4xCO2, and (b) the temperature mediated changes are quite similar from solar and CO 2 forcing.We also stress, that in Equation 1, we wrote the total cloud change as the sum of the temperature mediated change and adjustment (as well as some contribution from internal variability), but in fact the total cloud change (at any point in time) is not given by the sum of the adjustment (calculated with fixed-SST simulations or our estimated adjustment method), and the temperature mediated cloud changes shown in Part I (calculated as a linear fit between cloud changes and global mean surface temperature after year 10).This is because the cloud response is not a linear function of global mean surface temperature (especially in the first 10 years) and consequently the intercept (obtained when calculating the temperature mediated slope) is not the same as the adjustment (see Supporting Information S1).Hence, neither the temperature mediated changes presented in Part I nor the adjustments presented here in Part II characterize the non-linear cloud changes that occur (especially during the first 10 years).
One might argue that cloud adjustments should be defined as the changes in clouds that are a direct result of the forcing agent on the atmosphere with no change in the surface temperature, including changes in surface temperature pattern (not just the global mean temperature), perhaps following Andrews, Smith, et al. (2021).And in this sense, one can simply view the difference between the fixed-surface temperature ("Fixed-Ts") and "fixed-SST" (or our estimated adjustment) as an error or limitation in the calculation of the adjustment (due to land warming)-and at a practical level that is what we have done in this article.But if so, this still leaves us with the problem of characterizing and understanding the non-linear changes that occur as the surface and oceans warm at different rates.From a radiative perspective and at least on the global scale, one can view the situation as one in which there is time-varying radiative feedback (e.g., Knutti & Rugenstein, 2015;Rugenstein & Armour, 2021;Williams et al., 2008) or following the arguments of Rugenstein et al. (2016), a time-varying forcing.Given a sufficiently large ensemble of simulations (which would be used to mitigate the impact of internal variability), it might be possible to extend this to local cloud response.One could use a piecewise linear model to approximate the cloud response to global mean surface temperature such that the slope in the first 10 years can differ from that between years 10-150, and we leave such as a possible area of future research.But it seems likely to us that, much as SST patterns have been found to influence the slope of the temperature mediated response on long time scales (e.g., Andrews et al., 2015;Armour, 2017;Rugenstein et al., 2020), variations in both land and sea-surface temperature patterns are likely to have a large effect on the cloud response at shorter time scales; suggesting that it might be better to focus on a unified approach which characterizes the evolution of land and sea-surface temperature, and the impact the patterns of land and sea-surface temperature have on clouds.

Conclusions
A set of model experiments were requested by CFMIP to allow comparison between the climate response to changes in solar forcing and CO 2 concentrations.In Part I to this paper (Aerenson & Marchand, 2023), we examine the temperature mediated cloud changes from a 4% increase in solar intensity (solp4p) and quadrupling of CO 2 (4xCO2); and in Part II (this paper) we have focused on cloud adjustments-that is the changes in clouds that are a direct result of the forcing agent on the atmosphere which nominally have no influence from change in mean global surface temperature (or perhaps even the surface temperature pattern).Nonetheless, we calculated the 4xCO2 adjustments in the "standard way" using fixed SST simulations (which do allow land surface temperature to increase and do not hold global mean surface temperature fixed) and the calculation for the solp4p adjustment (our new approach, see Section 2) also allows a change in the global pattern of SST.We discuss this situation in more detail in Section 4.4 and raise this issue in this concluding section primarily to stress that the surface temperature changes do have a significant effect on the cloud adjustments presented here.For the remainder of this section, we discuss how temperature medicated and cloud adjustment differ, and address the question "How important are cloud adjustment relative to the temperature mediated feedback?" As regards the differences between solar and CO 2 cloud responses, in Part I we find that the only notable difference between the temperature mediated cloud changes in solp4p and 4xCO2 is the low cloud fraction change, where there is a greater reduction of low-clouds in the temperature mediated response to solp4p than 4xCO2.In Part II, we find that there are also noteworthy differences in the low and mid-level cloud adjustment to solar and CO 2 forcing.While a variety of mechanisms contribute to these low and mid-level cloud differences, two mechanisms appear to drive much of the differences between the two forcing experiments: (a) First, there is a large mid-level cloud reduction in the adjustment to 4xCO2 due to the radiative effect of CO 2 on cloud-top cooling, which is not present in the solp4p experiment (because the increase in solar forcing does not reduce cloud-top cooling to the same extent).(b) Second, there are differences in the pattern of surface temperature change.In the solp4p experiment there is more warming in the tropics and subtropics than in the 4xCO2 experiment in both the temperature mediated cloud response and in the adjustment.The enhanced warming in the tropics and subtropics of solp4p (as compared with 4xCO2) causes a stronger low cloud feedback via the thermodynamic effect.In Part I, we find the thermodynamic effect to be the most important mechanism driving the temperature mediated change of low clouds in the tropics and subtropics.Overall, the differences in adjustments (between solp4p and 4xCO2) have a larger radiative effect than the differences in the temperature mediated cloud changes.The NET cloud feedback parameters for solp4p and 4xCO2 are 0.87 and 0.82 W/m 2 /K respectively, which is well within 10% of one another.Meanwhile the NET cloud radiative adjustments (which can be thought of as the cloud contribution to the effective radiative forcing) have a much greater difference between the solp4p and 4xCO2: 0.37 and 0.86 W/m 2 for solp4p and 4xCO2, respectively, in the multi-model mean.
To demonstrate the relative importance of the adjustment and temperature mediated effect during the simulations we show in Table 2 the change in global mean total NET cloud radiative anomaly averaged over the first and last 20 years of the solp4p and 4xCO2 simulations (where the pre-industrial average has been subtracted), and in parentheses we show the ratio of the adjustment to the total cloud change at each time period, as Adjustment Total Anomaly .As previously mentioned in Section 3.2, MRI-ESM2-0 produces a strong negative radiative adjustment to solp4p associated with an increase in mid-level clouds (such that the mid-level CF component of the radiative adjustment Note.In parentheses is the ratio of the adjustment to the total anomaly averaged over each time.The adjustment to solp4p in MRI-ESM2-0 (denoted by asterisks) is negative, while the temperature mediated changes are positive resulting in the total radiative anomaly nearing (and crossing) zero during the simulation.As such the ratio for this simulation becomes spuriously large (and negative) so is excluded from the multi-model mean ratio calculation.
Journal of Geophysical Research: Atmospheres 10.1029/2023JD040297 is negative), which does not occur in the other models, where the adjustments and temperature mediated cloud changes both result in a positive NET radiative effect.As such, this model is excluded from the multi-model mean ratio calculation.
In the multi-model mean, the adjustment accounts for roughly 40% of the total cloud radiative anomaly in both the solp4p and 4xCO2 in the first 20 years following the abrupt forcing at the end of the simulations the ratio has reduced to 19% and 15% for the 4xCO2 and solp4p respectively.There is considerable inter-model spread in the ratio with the adjustment being most important (excluding the MRI-ESM2-0 solp4p simulation) in the 4xCO2 from MRI-ESM2-0, in large measure because the temperature mediated changes are small in this model.The cloud radiative adjustment is least important in the solp4p from HadGEM3-GC31-LL, which has a small global mean NET adjustment (see Figure 4).This is not to suggest that cloud adjustments are unimportant in this model, because this small global mean NET effect is due to significant LW and SW adjustments which are counter-acting such that the NET is small.As expected, comparing years 0-20 with years 130-150 there is significant reduction in the ratio during each model simulation (because cloud radiative effect of the temperature mediated changes becomes larger as the surface temperature increases).So at the end of the simulations the cloud adjustments are less important than the temperature mediated cloud changes, but remain a significant contributor to the overall radiative effect.
Clearly, if we hope to understand future warming both in the near and longer term, understanding and accurately simulating the adjustments, and more generally cloud responses in the first decade (or so) following forcing will be important.Based on our findings and the limitations of the methods used in this study, in our view the community needs to move toward some approach that does not focus only on adjustment (based on fixed-SST simulations) and temperature mediated (linear slope) response in later years, and perhaps characterizes the temporal evolution of land and sea-surface temperature changes and the associated cloud response over multiple timescales, including the first decade or so following the forcing change.

Figure 1 .
Figure 1.Top row: Estimated adjustment (following Equation 3, lefthand figure) and AMIP-solp4p adjustment (righthand figure) of total cloud amount from CESM1 simulations Note that we use the ensemble mean of the three available 4xCO2 simulations of CESM1, the inter-realization variability is discussed in Supporting Information S1.The numbers in the figure titles indicate the global mean cloud adjustment.Bottom three rows: Estimated adjustment to 2% increase of solar constant (solp2p), and AMIP-solp2p adjustment of longwave, shortwave, and net cloud radiative effect.Stippling indicates regions where at least 6/8 models agree on the sign of the adjustment.

Figure 2 .
Figure 2. Multi-model mean of the adjustment difference estimated following Equation 2. Stippling indicates regions where at least three out of four participating models agree on the sign of the adjustment difference.

Figure 3 .
Figure 3. Top 3 rows: multi-model mean adjustment of cloud amount in nine categories calculated from 30 years long averages of the amip-4xCO2 experiment, where the atmospheric CO 2 is quadrupled while the sea-surface temperature and sea-ice are held fixed.Bottom 3 rows: multi-model mean estimated adjustment of cloud amount in nine categories calculated following Equation 3.

Figure 4 .
Figure 4. Bar chart of the global mean cloud radiative adjustment to 4xCO2 (blue) and solp4p (orange) for the shortwave (top row), longwave (middle row) and net (bottom row) component of the cloud radiative adjustment calculated with cloud radiative kernels.Bars indicate the multi-model mean; black symbols indicate individual model values.

Figure 5 .
Figure 5. Maps of SW, LW, and NET total cloud radiative adjustment to solp4p (top row), 4xCO2 (middle row), and the adjustment difference (bottom row).As with previous figures, stippling indicates regions where at least 3 out of 4 models agree on the sign of the response.

Figure 6 .
Figure6.Top row: Adjustment to solp4p of surface temperature (Ts) and 500 hPa vertical velocity.Middle row: Adjustment to Ts and 500 hPa vertical velocity calculated using the AMIP-4xCO2 experiment.Bottom row: the adjustment difference between solp4p and 4xCO2 for Ts and 500 hPa vertical velocity.As previously, stippling indicates regions where at least 3 of 4 models agree on the sign of the response.In the right column, the contours represent the piControl climatology, where dashed contours are regions with mean-state upward motion and solid contours are mean-state downward motion.

Figure 7 .
Figure 7. Top row: Adjustment to solp4p of Estimated Inversion Strength (EIS) and 700 hPa relative humidity (RH_700).Middle row: Adjustment to Ts and RH_700 calculated using the AMIP-4xCO2 experiment.Bottom row: the adjustment difference between solp4p and 4xCO2 for Ts and RH_700.As previously, stippling indicates regions where at least 3 of 4 models agree on the sign of the response.

Figure 8 .
Figure 8. Temperature and Equivalent Potential Temperature vertical profile in the Indian Ocean and tropical west pacific.Which is defined as ocean area between 15°a nd 15°latitude and 60°-180°longitude.It is shown as the vertical profile of the multi-model mean from years 10-150 and the difference between the solp4p and 4xCO2 averages.

Table 1
Adjustment Difference of Latent Heat Release From Land Surface to the Atmosphere

Table 2
Global Mean NET Cloud Radiative Anomaly Averaged Over the First and Last 20 Years of the solp4p and 4xCO2 Simulations