Response of surface shortwave cloud radiative effect to greenhouse gases and aerosols and its impact on summer maximum temperature

Shortwave cloud radiative effects (SWCREs), defined as the difference of the shortwave radiative flux between all-sky and clear-sky conditions at the surface, have been reported to play an important role in influencing the Earth’s energy budget and temperature extremes. In this study, we employed a set of global climate models to examine the SWCRE responses to CO2, black carbon (BC) aerosols, and sulfate aerosols in boreal summer over the Northern Hemisphere. We found that CO2 causes positive SWCRE changes over most of the NH, and BC causes similar positive responses over North America, Europe, and eastern China but negative SWCRE over India and tropical Africa. When normalized by effective radiative forcing, the SWCRE from BC is roughly 3–5 times larger than that from CO2. SWCRE change is mainly due to cloud cover changes resulting from changes in relative humidity (RH) and, to a lesser extent, changes in cloud liquid water, circulation, dynamics, and stability. The SWCRE response to sulfate aerosols, however, is negligible compared to that for CO2 and BC because part of the radiation scattered by clouds under all-sky conditions will also be scattered by aerosols under clear-sky conditions. Using a multilinear regression model, it is found that mean daily maximum temperature (Tmax) increases by 0.15 and 0.13 K per watt per square meter (Wm−2) increase in local SWCRE under the CO2 and BC experiment, respectively. When domain-averaged, the contribution of SWCRE change to summer mean Tmax changes was 10 %–30 % under CO2 forcing and 30 %–50 % under BC forcing, varying by region, which can have important implications for extreme climatic events and socioeconomic activities.

3 events over Europe by enhancing solar heating (Rowell & Jones, 2006;Vautard et al., 2007;Zampieri et al., 2009;Chiriaco et al., 2014;Myers et al., 2018). This has influenced the environment, ecosystems and the economy through affecting the frequency and intensity of forest fires, power cuts, transport restrictions, crop failure and loss of life (De Bono et al., 2004;Ciais et al., 2005;Robine et al., 2008). For example, Wetherald and Manabe (1995) reported that in the summer for mid-60 latitude continents, higher temperature enhances evaporation in the spring and then evaporation decreases in the summer due to depleted soil moisture. Combined with higher temperature, this summertime evaporation reduction leads to lower relative humidity (RH), which reduces cloud cover and thereby invigorates solar heating. Cheruy et al. (2014) revealed that the intermodel spread of summer temperature projections in Northern mid-latitudes in CMIP5 (Climate Model Inter-comparison Project Phase 5) models is greatly influenced by SWCRE. 65 All the above studies suggest that the SWCRE plays an important role in influencing the surface energy budget and extreme temperature. Well-mixed greenhouse gases (WMGHGs) and aerosols are currently the two largest anthropogenic forcings (Myhre et al., 2013b). A better understanding on the climate response to these individual forcing agents is increasingly needed, considering their different trends across the globe and opposite impacts on climate (Shindell & Faluvegi, 2009). Due to the 70 difficulty of separating the forced climate signal of a single agent within observational records, these studies are generally based on model simulations, such as the widely used quadrupling of CO 2 experiments (Andrews et al., 2012). Many attempts have also been made to explore the aerosol impact on clouds and Earth's energy balance (Lohmann & Feichter, 2005;Chung & Soden, 2017), mean temperature (Ruckstuhl et al., 2008;Philipona et al., 2009), as well as extreme temperature (Sillmann et al., 2013;Xu et al., 2018). However, all these studies treated aerosols as a whole and the individual impacts from absorbing 75 and scattering aerosols are still less understood. Though some studies investigated the impact from individual aerosol species (Williams et al., 2001;Chuang et al., 2002;Koch & Del Genio, 2010), they generally used only a single model, and the results may be subject to model biases (Flato et al., 2013). Moreover, due to the continuing increase in the likelihood of hot temperature extremes (Seneviratne et al., 2014), as well as their serious consequences (De Bono et al., 2004), it is imperative to have a better understanding on the role of SWCRE from individual forcing agents in hot extremes. However, a multi-model 80 study on the cloud response to individual aerosol species and the impact of that response on Tmax is still lacking. Given these knowledge gaps, here we investigate the changes of SWCRE to CO 2 , BC and sulfate aerosols individually and explore its potential impact on Tmax by using a set of state-of-the-art global climate models. CO 2 is the most dominant WMGHG while the latter two represent absorbing and scattering aerosols respectively. This paper will proceed as follows: data and methods are described in Section 2. Results are presented in Section 3, discussions and summary are given in Section 4. 85 2 Data and Methods

Data
This study employs the model output from groups participating in the Precipitation Driver and Response Model Intercomparison Project (PDRMIP), utilizing simulations examining the climate responses to individual climate drivers . The nine models used in this study are CanESM2, HadGEM2,HadGEM3,MIROC, CAM4, CESM-CAM5, NorESM and IPSL-CM5A. The versions of most models used in the PDRMIP are essentially the same as their CMIP5 versions. The configurations and basic settings are listed in Table 1. In these simulations, global-scale perturbations were applied to all the models: a doubling of CO 2 concentration (CO 2 ×2), a tenfold increase of present-day black carbon concentration/emission (BC×10), and a fivefold increase of present-day SO 4 concentration/emission (SO 4 ×5). All perturbations were abrupt. Each perturbation was run in two parallel configurations, a 15-year fixed sea surface temperature 95 (fsst) simulation and a 100-year coupled simulation. One model (CESM-CAM4) used a slab ocean setup for the coupled simulation whereas the others used a full dynamic ocean. CO 2 was applied relative to the models' baseline values. For aerosol perturbations, monthly year 2000 concentrations were derived from the AeroCom Phase II initiative (Myhre et al., 2013a) and multiplied by the stated factors in concentration-driven models. Some models were unable to perform simulations with prescribed concentrations. These models multiplied emissions by these factors instead (Table 1). The aerosol loadings in the 100 CanESM2 model for the two aerosol perturbations are shown in Fig. 1 for illustrative purpose; the spatial patterns are similar for other models. In the BC experiment, the concentration is highest in East China (E. China), followed by India and tropical Africa. For the SO 4 simulations, the aerosols are mainly restricted to the Northern Hemisphere (NH), with the highest loading observed in E. China, followed by India and Europe. The eastern US also has moderately high concentrations. It is noted that only three of the nine models include aerosol-cloud interactions while the remaining ones only have aerosol-radiation 105 interactions. However, this does not impact our main conclusions (see section 4). More detailed descriptions of PDRMIP and its initial findings are given in Samset et al. (2016), Myhre et al. (2017), Liu et al. (2018) andTang et al. (2018).

Methods
In this study, we focus on the SWCRE at the surface in the low and mid-latitudes during boreal summer months (June-July-August, JJA hereafter), which is calculated as the difference in the SW radiative flux at the surface between all-sky and clear-110 sky conditions (Ramanathan et al., 1989). The base state of SWCRE in each model is shown in Fig. S1, with a multi-model mean (MMM) value of -57.9±1.8 W m -2 (MMM±1 standard error). The spatial patterns are fairly consistent across the models, with strong SWCRE in tropical regions and mid-to-high latitudes and weaker SWCRE in subtropics, regions generally with less clouds. Changes in SWCRE are obtained by subtracting the control simulations from the perturbations using the data of the last 20 years in each coupled simulation. The changes are then normalized by the effective radiative forcing (ERF) in the 115 corresponding experiments to obtain the changes per unit global forcing for comparison. Previous studies demonstrated that climate changes linearly with climate forcing for various forcing agents, including BC (Hansen et al., 2005;Mahajan et al., 5 2013). The ERF values for each model are obtained from Tang et al. (2019), which diagnosed those from the data for years 6-15 of the fsst simulations of each perturbation by calculating the radiative flux changes at the top-of-the-atmosphere . The MMM ERF values are 3.65±0.09 W m -2 (CO 2 ×2 ), 1.16±0.25 W m -2 (BC×10), and -3.52±0.63 W m -2 (SO 4 ×5 ) 120 for indicated experiments, respectively (MMM±1 standard error). Then the MMM changes are estimated by averaging all the nine models' results, giving the same weighting factor to each model. A two-sided student t-test is used to examine whether the MMM results are significantly different from zero. The same process was also repeated to other variables analyzed (i.e., temperature and humidity).

125
In order to investigate the impact of circulation changes on specific humidity, following Banacos and Schultz (2005), the horizontal moisture flux convergence (MFC) is calculated as: 130 In Eq (1), q is specific humidity in g kg -1 , and V is horizontal wind including both zonal and meridional components. All variables have a monthly temporal resolution. Equation (1) could be further written as: 135 In which and are zonal and meridional wind components in m s -1 .

SWCRE Change
Figure 2a-c show the SWCRE changes in response to abrupt changes in CO 2 , BC and SO 4 . CO 2 causes positive changes in SWCRE over most areas in the NH, indicating that more SW radiation reaches the surface. BC causes similar changes, but 140 with enhanced (ERF-normalized) magnitude, especially in North America (N. America), Europe and East Asia (E. Asia). In some source regions of BC aerosols (tropical Africa and India), however, the SWCRE changes are negative, which means more SW was reflected. These changes are all statistically significant and are unlikely to be caused by natural variability.
When it comes to individual model response ( Fig. S2-S3), these patterns are also consistent across at least eight of the nine models and are not very sensitive to the model setup (emission-based or concentration-based). For SO 4 , the SWCRE changes 145 are relatively small compared with the other two forcings and few significant changes are found over low-to-mid latitude regions. When domain averaged (green boxes in Fig. 2), the MMM SWCRE from CO 2 forcing is, 1.7 W m -2 (N. America), 2.0 W m -2 (Europe) and 1.5 W m -2 (E. China) respectively for the indicated regions. The SWCRE of BC forcing is 7.0 W m -2 (N. 6 America), 9.0 W m -2 (Europe) and 9.4 W m -2 (E. China) respectively, which is roughly 3 to 5 times larger than that from CO 2 forcing whereas sulfate aerosols induced 1.2 W m -2 over E. China and near-zero impact in N. America and Europe, with even 150 the sign of change being uncertain ( Fig. 3 and Fig. S4). Such SWCRE changes could be largely explained by the changes of cloud cover (Fig. 2d-f). Low-level cloud cover decreased significantly in regions where SWCRE is positive for CO 2 and BC forcing, with a stronger decrease from the latter, indicating that the cloud response is more sensitive to BC forcing than to WMGHGs. The sulfate aerosols caused increased cloud cover over mid-latitudes (Fig. 2f). The cloud cover in other levels show similar patterns of change (Fig. S5). In order to better understand these cloud responses, we will explore a set of potential 155 mechanisms driving such changes.

Mechanism of the Cloud Changes
Clouds form when air rises and cools to saturation, and are thus closely linked to changes in RH (Fig. 4a-c). The general pattern of RH changes corresponds well with cloud cover changes ( Fig. 2d-f). That is, the cloud cover decreases in regions where the RH drops and vice versa for most areas. A larger RH reduction due to BC compared with CO 2 also aligns with a larger cloud 160 cover decrease under BC forcing, especially over N. America and Europe. This spatial pattern is not surprising as it is easier for air masses to reach saturation in conditions with higher RH. By definition, RH depends on both specific humidity and saturation vapor pressure (which, in turn, depends on temperature). To probe which factor determines the RH changes, we further analyzed specific humidity changes ( Fig. 4d-f). Specific humidity increases ubiquitously under both CO 2 and BC scenarios, as a result of increased evaporation in a warmer climate. Thus, the main driver of the RH drop is the atmospheric 165 temperature that drives a faster increase of saturation vapor pressure. Figure 5 shows the changes of vapor pressure as a function of temperature change over Europe at 850 hPa. For example, the temperature increases by ~1.1 K under CO 2 forcing, accompanied by ~0.02 kPa vapor pressure increase. Such a vapor pressure increase, however, cannot keep pace with the rise in saturation vapor pressure, which is about 0.1 kPa. Consequently, the RH decreases in Europe and this is also the case for most other land areas. BC causes stronger temperature increases (and hence larger RH drop) in Europe and N. America, 170 explaining the larger cloud cover reductions compared with CO 2 . In the source regions of BC, such as India and tropical Africa, the RH increases because of stronger increases of specific humidity, combined with weak or no temperature changes (Fig. S6).
The response of cloud liquid water in the BC experiment could further support this conclusion (Fig. 4h). Liquid water decreases (increases) in regions with decreasing (increasing) cloud cover, following the pattern of RH. As cloud water content directly impacts cloud optical thickness and albedo, such a response may further impact SWCRE (i.e., enhance reflectance in regions 175 showing increasing liquid water and enhance transmittance in regions with decreasing liquid water). However, the liquid water responses under CO 2 and sulfate aerosols are much weaker, only significant in part of Asia and tropical Africa ( Fig. 4g and i).
Changes in moisture flux, dynamics and stability may also play a role in altering specific humidity and cloud formation (Bretherton, 2015). Here we analyze the changes of MFC, vertical velocity (omega), as well as lower tropospheric stability 180 (LTS), and find significant changes under the BC experiment again (Fig. 6). It is seen that more moisture is transported to tropical Africa and India (Fig. 6b), which could explain the abovementioned increases of specific humidity in these regions despite their lack of warming. A similar response was noted by Liu et al. (2018), which suggested that more moisture could be brought into monsoon regions due to BC forcing. Koch and Del Genio (2010) noted that BC particles could promote cloud cover in convergent regions as they enhance deep convection and low-level convergence when drawing in moisture from ocean 185 to land regions. This is also observed in our analyses, for example over Africa, North India, Pakistan and part of North China ( Fig. 6b and e), which is consistent with the dynamic cloud response mechanism noted by Myers and Norris (2013). However, these impacts may be further compounded by cloud type, circulation and the altitude of BC particles relative to the clouds (Koch & Del Genio, 2010;Samset & Myhre, 2015). The changes in moisture flux and dynamics in the CO 2 experiment are relatively weaker compared with those from BC, and most of the changes are only observed in low-latitude regions, possibly 190 due to the shift of Intertropical Convergence Zone (ITCZ) or monsoon circulations. The sulfate aerosols, on the other hand, generally show opposite changes to those from CO 2 and BC ( Fig. 4c and f), owing to sulfate's cooling effect. Another mechanism that has been reported to influence cloud cover is LTS, in which a stable boundary layer could trap more moisture, thereby permitting more low-level clouds (Wood & Bretherton, 2006;Bretherton, 2015). In order to investigate this mechanism, we further analyzed LTS, defined as the difference of potential temperature between 700 hPa and surface (Fig. 195 6g-i), in which positive anomalies indicate a stronger inversion or weaker lapse rate. The LTS response is again strongest in response to BC forcing (Fig. 6h), with a widespread increase in stability. A previously reported positive correlation between LTS and low-level cloud cover is, nonetheless, only observed in BC source regions (tropical Africa and India) and part of the central US (Fig. 6h). The LTS responses over land are much weaker in response to CO 2 and SO 4 forcing, with some responses in Africa and India in response to sulfate aerosols (weaker inversion and less cloud). Some other factors have also been 200 suggested to play a role in modifying low-level clouds, such as the diurnal cycle (Caldwell & Bretherton, 2009) and radiative effects of cirrus clouds (Christensen et al., 2013). Due to the limited model output, however, we acknowledge that it is impossible to examine these factors in the current study and it is beyond the scope of our study to probe all possible factors driving the cloud changes. In summary, the above analyses illustrate that the cloud cover changes we see can be primarily explained by RH changes and, to a lesser extent, changes of liquid water content, circulation, dynamics, and stability. 205

Fast and Slow Responses
The above responses shown are total responses, which could be further split into fast responses (also called rapid adjustments) and slow responses (Andrews et al., 2010;Boucher et al., 2013). The fast responses generally occur within weeks to a few months with the global mean temperature unchanged, and also with the expectation of a small change over land, which could be obtained by fsst simulations. The slow response is mainly depending on global mean temperature change, which could be 210 estimated by the difference between coupled simulations and fsst simulations, assuming the total response is a linear combination of fast response and slow response Stjern et al., 2017). For the CO 2 experiment, fast responses dominated in E. US and Europe while both fast and slow responses influence Asia (Fig. 7). When it comes to BC, both fast and slow responses are important in these regions, and in some regions the fast and slow response even show opposite changes 8 (e.g., N. Europe). This is consistent with the findings of Stjern et al. (2017) that the response of cloud amount under BC forcing 215 typically consists of opposite rapid adjustments. Regarding sulfate aerosols, the SWCRE changes are much weaker, with both fast and slow responses influencing Asia and Africa. As discussed in Section 3.2, the slow responses in Asia is likely to be associated with circulation changes, as significant changes in MFC, omega and stability are observed in tropical regions and monsoon regions across all three experiments (Fig. 6). These circulation changes could be, but are not limited to, shifts in the monsoons or ITCZ and tropical expansion, and both greenhouse gases and aerosols have been reported to impact these 220 circulations (Menon et al., 2002;Wang, 2007;Meehl et al., 2008;Seidel et al., 2008;Allen et al., 2012;Turner & Annamalai, 2012).

SWCRE Response to Sulfate Aerosol
Another interesting phenomenon worth noting is the relatively small change in SWCRE induced by sulfate aerosols compared with CO 2 and BC. SWCRE at the surface is obtained as the difference of SW fluxes between all-sky and clear-sky conditions 225 ( Fig. 8). However, both clouds and aerosol particles scatter solar radiation, so that at least part of the radiation scattered by clouds under all-sky conditions will also be scattered by aerosols under clear-sky conditions (no clouds). This means the SW radiation change at the surface due to scattering may not be as sensitive to cloud fraction changes, which leads to reduced changes in their difference (SWCRE), at least in the source regions (Fig. 8). The SWCRE under sulfate aerosols will not be further discussed due to its small radiative impact at the surface. 230

Impact on Radiation and Tmax
From the energy perspective, the net incoming radiation (Rin) at the surface is the combination of downward SW radiation and downward longwave (LW) radiation minus the reflected SW radiation (Rin = ↓SW -↑SW + ↓LW). Rin represents the total energy available to maintain the surface temperature and to sustain the turbulent fluxes (Philipona et al., 2009). The surface responds to the imposed Rin by redistributing the altered energy content among the outgoing LW radiation and 235 nonradiative fluxes (ground heat flux and turbulent flux) (Wild et al., 2004). Because SW radiation is in effect only during daytime while LW radiation works both day and night, Rin is directly related to Tmax. In a perturbed climate, both SW and LW radiation will change, thereby changing Rin and Tmax. The net SW radiation change is further linearly decomposed into SW changes under clear-sky conditions and SWCRE changes. The changes of Rin and its individual components, as well as Tmax are shown in Fig. 9. For the CO 2 ×2 experiment, the SW under clear-sky conditions shows slight decreases over most of 240 land surfaces, mainly due to the absorption of SW radiation by enhanced water vapor, except for some high-latitude regions where albedo effect is important (Fig. 9a). Combined with the changes of SWCRE and ↓LW radiation, Rin shows significant increases over all land surfaces and thus, increasing Tmax ( Fig. 9g and i). The BC×10 experiment shows similar responses, with significantly negative SW radiation under clear-sky conditions due to SW absorption by BC particles (Fig. 9b) and enhanced ↓LW radiation resulted from atmospheric heating (Fig. 9f). The resulting Rin changes largely explained Tmax 245 changes on the first order, with cooling observed in source regions (India and tropical Africa) and warming elsewhere ( Fig. 9h and j). Nonetheless, some exceptions occurred (i.e., E. China), with decreased Rin but increased Tmax, possibly due to the atmospheric heat transport (Menon et al., 2002) and reduced turbulent fluxes (Wild et al., 2004).
In order to further determine the contributions in Tmax changes from each individual radiative component, a multilinear 250 regression model is applied by regressing Tmax changes to SW clear-sky, SWCRE and ↓LW radiation changes with zero intercept, obtaining the following models:  Table 2. Physically, Tmax increases in these regions are mainly due to the increased flux from SWCRE and ↓LW, and partially offset by the reduced flux from SW clear-sky (Table 2 & Fig. 9). Taking N. America under CO 2 ×2 experiment as an example, the warming in Tmax from SWCRE and ↓LW are 0.95 K and 3.24 K respectively, in which SWCRE contributed roughly by 23% to the total warming and the 270 remaining 77% is from the ↓LW radiation change. Such warming is offset by the 0.27 K cooling from SW changes under clear-sky conditions, leading to a net increase of 3.92 K in Tmax. The contributions of SWCRE in Tmax increases are 29% (Europe), 20% (E. China) and 9% (India) for the indicated regions under the CO 2 ×2 experiment. For the BC×10 experiment, the contributions from SWCRE are larger than those in the CO 2 experiment, i.e. 34% (N. America), 47% (Europe) and 34% (E. China) for each region. The response over India under the BC experiment is opposite, in which both SW components cause 275 cooling in Tmax due to reduced fluxes and such cooling is slightly offset by the warming from increased ↓LW radiation. In this case, the negative SWCRE change contributed 54% to the reduction in Tmax. It is noted that the radiation change might not explain all Tmax changes, as other factors may come into play. For instance, the temperature response would be different when surface is getting drier under a warmer climate. This is because more net radiation is realized as sensible heat instead of latent heat under drier conditions, which has been suggested to play an important role in recent European heatwaves 280 (Seneviratne et al., 2006;Fischer et al., 2007).

Discussion and Summary
Our study shows that cloud cover in the summer is reduced in a warming climate over most mid-latitude land regions. The reduction of clouds, at the same time, may also reduce the warming effect by reducing downwelling LW radiation (LWCRE, Fig. S7). Specifically, the LWCRE changes per unit CO 2 forcing, in MMM, are -1.1 W m -2 (N. America), -0.8 W m -2 (Europe) 285 and -1.0 W m -2 (E. China) respectively, resulting in net CRE (SWCRE+LWCRE) changes of 0.6 W m -2 (N. America), 1.2 W m -2 (Europe) and 0.5 W m -2 (E. China) at the surface. The LWCRE changes per unit BC forcing are -1.7 W m -2 (N. America), -2.1 W m -2 (Europe) and -1.5 W m -2 (E. China) respectively, leading to net CRE changes of 5.3 W m -2 (N. America), 6.9 W m -2 (Europe) and 7.9 W m -2 (E. China). The net CRE changes are positive under both forcings and work as a positive feedback in these areas. As SWCRE is only active during daytime, the CRE changes have an even more pronounced amplifying effect 290 on summer extreme temperature in these populated regions.  Figure 11. For the regions of interest in the current study, the positive SWCRE over N. America, Europe and E. China and negative SWCRE over India are still observed in the models including indirect effects, but with reduced magnitude. Thus, our main conclusions hold in both sets of models, since the responses do not qualitatively vary between those with indirect effects and models without those effects. Such effects are not likely to be a large source of uncertainty but merit future study. Secondly, the aerosol perturbations are idealized time-invariant 10× and 5× 310 present-day aerosol concentrations. Such simulations provide valuable physical insights into the effects of different forcings on a variety of aspects of the climate system. Aerosol concentrations, however, changed inhomogeneously during the historical period and in recent decades, both spatially and temporally. For example, aerosol concentrations have been decreasing in Europe and N. America since the 1980s and have been increasing in Asia since the 1950s (Smith et al., 2011). Future simulations may use aerosol forcing with realistic spatio-temporal changes. 315 In conclusion, our study shows that both CO 2 and BC could cause positive SWCRE changes over most regions in the NH, with a stronger response caused by BC, except over some key source regions of BC aerosols (e.g., India, tropical Africa) which show opposite changes. The SWCRE changes under sulfate aerosol forcing are, however, relatively small compared with the other two forcers. The SWCRE changes are mainly a consequence of RH changes and, to a lesser extent, liquid water, 320 circulation, dynamics and stability changes. The SWCRE changes may have contributed 10~50% of summer mean Tmax increases, depending on forcing agent and region, and contributed substantially to Tmax decreases in the source regions of India and Africa, which has important implications for extreme climatic events and socio-economic activities.

Data code availability
The PDRMIP model output used in this study are available to public through the Norwegian FEIDE data storage facility. For 325 more information, please see http://cicero.uio.no/en/PDRMIP. This study is performed by using Matlab R 2019a. The Matlab code is available upon reasonable request.

Competing interests
The authors declare no competing interests.
i) Response of cloud liquid water is added.
ii) Response of lower tropospheric stability is added.
iii) SWCRE response for individual models is added to the supporting material.
This paper investigates the response of shortwave cloud radiative effect and daily maximum temperature to greenhouse gases and aerosols (BC and sulfate). It is found that BC results in a stronger positive SWCRE change than CO2 when normalized by effective radiative forcing, but sulfate does not have much effect on SWCRE. It is also shown that the increase in SWCRE resulting from CO2 and BC leads to an increase in daily maximum temperature during the summer. The results are interesting and have some important implications, however a number of things need to be addressed before recommendation for publication.

Major
1. Most of the results are normalized by effective radiative forcing. What are the surface temperature responses to CO2 and BC, respectively? Could the difference in SWCRE be partly due to the difference in the temperature change (i.e., the efficacy of BC)? Response: the multi-model mean temperature changes for CO 2 and BC experiments are 2.5K and 0.7K respectively. The ratio of 2.5/0.7=3.6 is slightly larger than the ERF ratio (3.65/1.16=3.15), which means if SWCRE changes were normalized by dT, the difference of between CO 2 and BC would be slightly larger.
As the results would not change much, however, the efficacy of BC will not significantly influence our results in the bar plot (Fig. 3).
2. The SWCRE change is attributed to the change in cloud cover. I would be interested to see some discussion in the change in cloud liquid water content or liquid water path, which also plays an important role in determining SWCRE.
Response: added in Fig. 4 and section 3.2. We added the following discussion after line 172: "The response of cloud liquid water in the BC experiment could further support this conclusion (Fig. 4h). Liquid water decreases (increases) in regions with decreasing (increasing) cloud cover, following the pattern of RH. As cloud water content directly impacts cloud optical thickness and albedo, such a response may further impact SWCRE (i.e., enhance reflectance in regions showing increasing liquid water and enhance transmittance in regions with decreasing liquid water). However, the liquid water responses under CO2 and sulfate aerosols are much weaker, only significant in part of Asia and tropical Africa ( Fig. 4g and i)." 3. The change in cloud cover is explained by the change in RH. However, there are a lot of other factors affecting clouds (radiation, dynamics, thermodynamics, etc., see Bretherton (2015) and references therein), and I think a more detailed discussion would be helpful. The authors look at vertical velocity and suggest that the change in stability plays less of a role, but it is not clear to me how the conclusion is reached. The estimated inversion strength or lower troposphere stability may be a better predictor for stability. Response: accepted. We added lower troposphere stability in Fig. 6 and also kept the vertical velocity, as this is reported by some previous studies saying that subsidence could impact cloud cover (e.g., Myers and Norris, 2013). Thus, for the cloud cover changes, on top of humidity, we also discussed liquid water, moisture flux, dynamics and stability. Some discussions are also included and we also acknowledged that it is impossible to examine all the factors in the current study due to limited output.
We added the following discussion after line 192: "Another mechanism that has been reported to influence cloud cover is LTS, in which a stable boundary layer could trap more moisture, thereby permitting more low-level clouds (Wood & Bretherton, 2006;Bretherton, 2015). In order to investigate this mechanism, we further analyzed LTS, defined as the difference of potential temperature between 700 hPa and surface (Fig. 6g-i), in which positive anomalies indicate a stronger inversion or weaker lapse rate. The LTS response is again strongest in response to BC forcing (Fig. 6h), with a widespread increase in stability. A previously reported positive correlation between LTS and low-level cloud cover is, nonetheless, only observed in BC source regions (tropical Africa and India) and part of the central US (Fig. 6h). The LTS responses over land are much weaker in response to CO2 and SO4 forcing, with some responses in Africa and India in response to sulfate aerosols (weaker inversion and less cloud). Some other factors have also been suggested to play a role in modifying low-level clouds, such as the diurnal cycle (Caldwell & Bretherton, 2009) and radiative effects of cirrus clouds (Christensen et al., 2013). Due to the limited model output, however, we acknowledge that it is impossible to examine these factors in the current study and it is beyond the scope of our study to probe all possible factors driving the cloud changes. In summary, the above analyses illustrate that the cloud cover changes we see can be primarily explained by RH changes and, to a lesser extent, changes of liquid water content, circulation, dynamics, and stability." 4. Why not use the same colorbar in Fig. 2 for normalized forcing change and cloud fraction change to make comparison easier (of course range of values can be different)?
Response: changed to same colorbar.

5.
By showing only MMM results and nothing about model spread we have no idea how much the models diverge in predictions. Not sure there is an easy way to convey that. Response: As these are spatial maps, we could not figure out a way of showing inter-model spread at this moment. So we just follow the traditional way by showing MMM results. In fact, the uncertainty bars in the bar plot (Fig. 3) could shed some light on the inter-model spread of the results. For CO2 and BC, the results are quite consistent across the models and for SO4, even the sign of change is uncertain and thus, a larger range is seen. These results are further illustrated by the individual model response, which has been included in the supporting material ( Fig. S2-S4).   6. I imagine the radiative treatment of aerosol differs widely among models. Not discussed. When you change emissions instead of concentrations directly, divergence is introduced too. Response: the readers could refer to the literature documenting each model in Table 1 for detailed radiative treatment of aerosols, as it is nearly impossible to discuss them one by one. We added the SWCRE changes for individual models into supporting material (Fig. S2-S4; see the response above). The main features are consistent across models and not sensitive to model setup (e.g., emission, concentration or radiative treatment), indicating that our results are fairly robust. We added these in section 3.1 line 144: "When it comes to individual model response (Fig. S2-S3), these patterns are also consistent across at least eight of the nine models and are not very sensitive to the model setup (emission-based or concentration-based)." 7. I also imagine that the base state of the models is quite different too. Care to comment? Response: The multi-model mean value of SWCRE in the base run is -57.9±1.8 W m -2 (MMM±1 standard error). The spatial patterns are fairly consistent across the models, with strong SWCRE in tropical regions and mid-to-high latitudes and weaker SWCRE in subtropics, regions generally with less clouds. We added this figure in the supporting material (Fig. S1) and also mentioned this in section 2.2 in line 111: "The base state of SWCRE in each model is shown in Fig. S1, with a multi-model mean (MMM) value of -57.9±1.8 W m -2 (MMM±1 standard error). The spatial patterns are fairly consistent across the models, with strong SWCRE in tropical regions and mid-to-high latitudes and weaker SWCRE in subtropics, regions generally with less clouds."