1. Introduction

Climate sensitivity is one of the main descriptors of the climate system (e.g., Cubasch et al. 2001). Much attention has focused on the difference in climate sensitivity among climate models used to project future climate change. Climate sensitivity is usually defined as the model-simulated equilibrium change in global surface temperature due to a doubling of carbon dioxide. For practical reasons, an equilibrium solution is obtainable by using a slab ocean model, rather than a fully interactive ocean model. It is important to point out that this is only one measure of climate sensitivity, which has definite limitations given the global long-time average nature of the metric (e.g., Boer and Yu 2003). Models may agree in terms of the global climate sensitivity but have very different regional responses to increased greenhouse gases.

This study explores and documents the climate sensitivity of the most recent version of the Community Climate System Model version 3 (CCSM3; Collins et al. 2006a). The CCSM3 is a fully coupled atmosphere, ocean, land, and sea ice climate system model. A version of the CCSM3 employing the same atmosphere, land, and thermodynamic components coupled to a slab ocean model is used to obtain the equilibrium climate sensitivity. CCSM3 and its atmospheric component CAM3 have been designed to perform realistically at a variety of resolutions. The CCSM3 at the high resolution (T85) were used for the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4).

The CCSM has evolved over the past decade with major improvements to virtually all aspects of the climate system components (Boville and Gent 1998; Kiehl and Gent 2004; Collins et al. 2006a). The first version of the model, Climate System Model version 1 (CSM1), was released to the community in 1994 and had a climate sensitivity of 2.01°C (Boville et al. 2001). The next version of the model, CCSM2 (Kiehl and Gent 2004), was released to the community in 2003 and had a climate sensitivity, which was slightly higher than that of CSM1, that is, 2.27°C. The current T42 version of the model, CCSM3, has an even higher climate sensitivity of 2.47°C. Reasons for this progressive increase in climate sensitivity in the evolution of CCSM versions is discussed in this study, and not surprisingly much of the change is related to the treatment of clouds in the various model versions.

The present study is organized as follows. A brief description of the model and simulations is given in section 2. Section 3 provides an analysis of the climate sensitivity, and section 4 describes the dependence of climate sensitivity on horizontal resolution. Finally, section 5 summarizes the findings of the study.

2. Model description and simulations

A comprehensive description of the CCSM3 is given in Collins et al. (2006a). Only a brief description of aspects of the model germane to climate sensitivity is provided here. The atmospheric component of the CCSM3 is the Community Atmosphere Model version 3 (CAM3; Collins et al. 2006b). This version of the CAM includes extensive changes to the cloud parameterizations including a new formulation of the prognostic cloud water scheme that explicitly accounts for the cloud ice phase and solid precipitate (Boville et al. 2006). A reformulation for the diagnoses of convective cloud fraction in terms of the cumulus mass flux leads to a marked change in midlevel cloud fraction. In CAM3 the detrainment of cloud water from shallow convection acts as a source of condensate for the large-scale prognostic cloud scheme. In the radiation scheme, the longwave parameterization for the water vapor continuum has been modified to fit the latest understanding of continuum absorption. The shortwave scheme in CAM3 accounts for the realistic spatial distribution of aerosols. The ocean component of CCSM3 (Large and Danabasoglu 2006) is the Parallel Ocean Program (POP) with the physical parameterizations developed at the National Center for Atmospheric Research (NCAR), including a revision to the boundary layer scheme and inclusion of solar absorption by a prescribed climatology of ocean color. The Community Land Model (Dickinson et al. 2006) is a completely new land model that includes a new formulation for the determination of snow-covered vegetation.

There are three versions of the CCSM3 defined by the horizontal resolution of the atmosphere and ocean models. The lowest-resolution version of the CCSM (Yeager et al. 2006) employs a T31 (3.75° × 3.75°) spectral truncation for the CAM3 and a nominal 3° resolution for the ocean model. The moderate-resolution version of CCSM3 employs a T42 (2.8° × 2.8°) spectral truncation for the CAM3 and a nominal 1° ocean model, while the high-resolution version of CCSM3 used for the IPCC scenario simulations employs a T85 (1.4° × 1.4°) spectral truncation in CAM3 and the 1° ocean model. The three versions of the CAM3 (T31, T42, and T85) have been coupled to slab ocean models, where the horizontal resolution of the slab ocean is determined by the atmospheric resolution, and control and doubled CO₂ simulations have been carried out.

The equilibrium simulations used for the analysis of climate sensitivity are based on slab ocean model (SOM) versions of CSM1 (Meehl et al. 2000; Dai et al. 2001), CCSM2 (Kiehl and Gent 2004), and CCSM3. The slab ocean model specifies the spatial, monthly mean distribution of ocean heat transport, Q_flx, which is obtained from the net ocean surface energy budget, F_net, of a control integration of the atmospheric model, the climatological annual mean ocean mixed layer depths, h, and observed sea surface temperatures, SST, using the following expression,

View Expanded

where ρ_o is density of water and c_p is the ocean heat capacity. The sea ice model is purely thermodynamic and neglects sea ice dynamics. The CAM3 version of the slab ocean model employs the thermodynamic sea ice component of the CCSM3 model, where an ocean heat to the ice is specified for each hemisphere to ensure a reasonable total area of sea ice amount. The slab ocean version of the CCSM is run in a control configuration to simulate the present climate and then run with twice the control CO₂ concentration (355 ppmv) until the model asymptotes to a steady state, which typically requires about 40–50 yr of model integration. Climate statistics from the SOM simulations are based on 20-yr averages from the end of the simulations.

Gregory et al. (2004) have shown that the forcing due to increased CO₂ in these types of models can be obtained by plotting the global annual mean top of model net energy flux versus the change in global annual mean surface temperature for each year of the SOM integration. Figure 1 shows this type of scatterplot for the T42 SOM version of CAM3 for the first 20 yr of a 50-yr integration. Extrapolating the linear regression of the net flux dependence on change in surface temperature to ΔT_s = 0 yields the 2 × CO₂ forcing estimate. For CAM3 this forcing is 3.58 W m⁻² for a doubling of CO₂ and is in good agreement with the forcing value obtained by Gregory et al. (2004), that is, 3.5 ± 0.2 W m⁻².

The transient climate sensitivity is based on fully coupled simulations of CSM1.4, CCSM2, and CCSM3 where a 1% yr⁻¹ increase in CO₂ mixing ratio is initiated at some point in a long (i.e., multiple centuries) control integration. Given this rate of increase the CO₂ mixing ratio, for any year T (years), is given by

View Expanded

where CO₂(0) is the control CO₂ concentration. Equation (2) implies that a doubling of CO₂ occurs at year 70 and a quadrupling of CO₂ occurs at year 140. Note that the CSM1 was not run to a point of quadrupling, but both CCSM2 and CCSM3 have been run past the point of 4 × CO₂(0). Climate statistics at the point of doubling for this study are based on 20-yr averages centered at year 70 from the transient simulations. All model output used in this study are available to the public (see www.ccsm.ucar.edu).

3. Analysis of climate sensitivity

To eliminate effects of model resolution on climate sensitivity we first compare the response of the three versions of the CCSM that employ a T42 resolution for the atmospheric component. We explore the effects of resolution on climate sensitivity in section 4.

The time evolution of surface warming from the 1% CSM1.4, CCSM2, and the T42 CCSM3 simulations is shown in Fig. 2. The increase in surface air temperature is quite linear up to the point of doubling at year 70. The warming in CSM1.4 and CCSM3 agree closely with one another, while the CCSM2 warming is less than either of these two models. The lower warming response of CCSM2 is linked to shortwave cloud response in the CAM2. The time evolution of the changes in longwave and shortwave cloud radiative forcing from the three coupled model simulations is shown in Fig. 3, where cloud radiative forcing is the difference between the clear-sky and all-sky top-of-atmosphere radiative fluxes. Note that the changes in shortwave cloud forcing are in general 2 to 3 times larger in magnitude than the changes in longwave cloud forcing. Thus, the net effect of clouds on the simulated climate system is to reflect more energy to space as the climate system warms. This feedback in the system is strongest in CCSM2, which suppresses surface warming more in this version of the CCSM compared to either the CSM1 or CCSM3. To gain better insight into this process the change in tropical vertical cloud structure is shown in Fig. 4 for the coupled models. There are significant differences in the vertical cloud changes among these three models. The cloud changes in CCSM2 indicate large increases in cloud fraction between the surface and 700 hPa compared to either CSM1 or CCSM3. These changes in clouds lead to a larger increase in shortwave cloud forcing compared to the other models. Note the large differences in cloud structure between CCSM2 and CCSM3 in the middle and upper troposphere, where CCSM3 shows a more complex layering structure in cloud response to the CO₂ warming. Associated with these changes in cloud cover, there is also more structure to the specific and relative humidity (not shown), which is due to the changes in the role of evaporation to the moisture budget in CAM3 compared to CAM2. CAM3 includes prognostic equations for solid condensate and precipitate that alters the water vapor budget in the upper troposphere compared to the simpler formulation in CAM2.

The focus up to this point has been on changes in clouds from the different versions of the CCSM. Changes in surface properties also contribute to climate sensitivity. In high latitudes, reduction in sea ice area due to greenhouse forcing leads to enhanced warming, which is the sea ice albedo feedback. Over high-latitude land regions, reduction in snow cover also leads to an analogous positive feedback in the climate system. Meehl et al. (2004) argue that the sea ice feedback dominates in the CCSM models.

One measure of the strength of surface feedback is to consider the change in surface albedo for the 1% transient simulations, which is shown in Fig. 5. Now we see more disagreement among all three models, with CCSM3 showing the strongest surface albedo feedback compared to either CSM1.4 or CCSM2. The reason for this difference appears in Fig. 6, which shows the change in sea ice area for the three model simulations for the Northern and Southern Hemispheres. The CCSM3-simulated Northern Hemisphere sea ice change is greater than changes simulated by the other versions of CCSM. The differences in the changes are quite large even in the global mean. The impact of these changes on the clear-sky radiative budget is quantified in the following feedback analysis.

Finally, a measure of the response of the hydrologic cycle to global warming is provided by the hydrological sensitivity or the change in the global mean rate of precipitation at doubling divided by the change in surface temperature at doubling, which is 1.5% °C⁻¹ for CSM1.4, 1.1% °C⁻¹ for CCSM2, and 1.6% °C⁻¹ for CCSM3, indicating again that CCSM2 has a lower hydrological sensitivity than either CSM1.4 or CCSM3.

To investigate the causal mechanisms that contribute to the climate response of these three versions of the CCSM, we employ the approach of effective climate sensitivity proposed by Gregory and Mitchell (1997); Senior and Mitchell (2000), and Raper et al. (2002). This approach is based on the following expression,

View Expanded

where ΔQ is the forcing due to increased CO₂, Λ is the climate feedback parameter, ΔT is the change in surface air temperature, and ΔF is the change in net surface flux, which is equivalent, on annual or longer time scales, to the change in the top-of-atmosphere net radiative flux. This expression states that the CO₂ radiative forcing of the climate system is balanced by emission to space by warming the system and by energy flowing from the ocean mixed layer to the deep ocean. We evaluate the terms ΔQ, ΔT, and ΔF from the 1% transient simulations from CSM1.4, CCSM2, and the T42 CCSM3, which enables us to determine the climate feedback factor. We use a forcing of 3.58 W m⁻² for ΔQ at the time of doubling, which was derived from the regression method (see Fig. 1). We use the same forcing value for all three models, since the CO₂ radiative parameterization is identical for all three versions. Gregory and Mitchell (1997) further decompose the total feedback parameter as

View Expanded

where

View Expanded

and

View Expanded

where ΔF_x is the change in the top-of-atmosphere longwave clear-sky (lclr) flux, longwave cloud forcing (lwcf), clear-sky shortwave (sclr) flux, and shortwave cloud (swcf) forcing. Note that the smaller (larger) the magnitude of Λ, the larger (smaller) the climate sensitivity, that is, the greater (smaller) the magnitude of ΔT. The ΔT at the time of doubling is the transient climate response (TCR; e.g., Cubasch et al. 2001). Finally, following Gregory and Mitchell (1997) we define the “ocean heat uptake efficiency” as

View Expanded

which according to (3.1) leads to

View Expanded

This relation can be used to explore the relative magnitude of atmospheric versus oceanic processes in determining the response of the climate system to an applied forcing. Note that there are small differences between these values and those of Meehl et al. (2004), where the differences arise from using different averaging periods for the calculations.

To estimate these feedback terms, we use 20-yr averages centered about the time of doubling (year 70) in the 1% transient simulations. Using the above relations the various response and feedback measures are shown in Table 1. The TCR for CSM1.4 and CCSM3 are very similar to one another, 1.44°C compared to 1.48°C, while CCSM2 has a much lower climate response of 1.09°C. Much of the analysis in this section will explore the reasons for this lower climate response in this version of the CCSM. The effective climate sensitivity defined by Murphy (1995) is the expected equilibrium warming due to a doubling of CO₂ assuming the climate feedbacks in the coupled model remains fixed at and beyond the time of doubling. Again we see that the CCSM2 has a lower effective climate sensitivity than either CSM1.4 or CCSM3. The equilibrium climate sensitivity, ΔT_eq, is the warming due to a doubling of CO₂ obtained from coupling the atmospheric model to the slab ocean component, and running the system to a steady state, where ΔF = 0.

To better understand the reasons for the differences in climate response we show the net, clear-sky, and net cloud feedback parameters in Table 2. The net clear-sky feedback decreases by 89% from CSM1.4 to CCSM3, a significant change implying a warmer response in CCSM3 compared to either CSM1.4 or CCSM2 due to clear-sky feedbacks. According to (3.6) the fact that the transient climate response in CSM1.4 is similar to that of CCSM3 implies that the clear-sky feedbacks are being offset by cloud feedbacks and/or the efficiency in ocean mixing. This is evident in that the cloud feedback factor is greater in CCSM3 than in CSM1.4 by 34%, and that the ocean mixing efficiency is 38% greater in CCSM3 compared to CSM1.4. Thus, a significant amount of the decrease in clear-sky feedback effect between CSM1.4 and CCSM3 is offset by the combined increases in cloud feedback effect and the efficiency of ocean mixing (see Table 3). The picture is very different when we consider CCSM2 compared to CCSM3.

A further breakdown of the feedback effects is given in Table 3, where the clear-sky and cloud feedback effects are separated into shortwave and longwave components. Note that the net cloud feedback is negative, indicating that clouds act to reduce the climate response.

Let us first consider the clear-sky feedbacks in the three models. The longwave clear-sky feedback effect monotonically decreases with each version of the CCSM. This decrease is due to two factors: changes in the vertical distribution of water vapor in the three models, and improvements to the longwave water vapor continuum in the later versions of the CCSM. The percent change in specific humidity centered about the time of CO₂ doubling from the three 1% simulations is shown in Fig. 7. The CSM1.4 shows a large increase in stratospheric water vapor, which will not have a large impact on Λ_lclr, since absolute amount of water is so small, and the thermal gradient is opposite to that of the troposphere. The significantly larger increase in upper-tropospheric water vapor in CCSM3 will lead to a greater trapping of outgoing longwave radiation, which leads to a smaller value of Λ_lclr. This change in upper-tropospheric water vapor between CCSM3 and CCSM2 is due the changes in the prognostic cloud water scheme and associated evaporation processes in CAM3 compared to CAM2. The atmospheric components of CCSM2 and CCSM3 also incorporate a new parameterization of the water vapor continuum, which leads to enhanced absorption of upwelling longwave radiation in these models, compared to CSM1.4.

The shortwave clear-sky feedback effects increase from CSM1.4 to CCSM3, with the largest increase (32%) occurring between CCSM2 to CCSM3. The major change that occurred in the sea ice simulation between these models was a reduction in the simulated sea ice thickness in the CCSM3, which was a significant improvement in CCSM3, since sea ice was too thick compared to observations in CCSM2. The stronger sea ice response of CCSM3 was noted in Fig. 6. The net clear-sky feedback effect is the difference between Λ_lclr and Λ_sclr, which is dominated by the decrease in the longwave clear-sky processes.

What is the cause of the twofold difference in shortwave cloud feedback between CCSM2 and CCSM3? Results from the three simulations indicate that in the CCSM2 low cloud generally increases, while there is a smaller increase or even a decrease in low-cloud amount in the other versions of the CCSM. There were a number of changes in moist physics parameterizations between CCSM2 and CCSM3, but one in particular affected the diagnosis of convective low-cloud amount. In CSM1.4 and CCSM3 the convective low-cloud amount was parameterized in terms of the cloud mass flux, while in CCSM2, this parameterization was changed to one where fraction of these clouds was diagnosed as a function of relative humidity. To test whether this single change in the diagnosis of convective low-cloud amount caused the twofold change in shortwave cloud feedback, we have carried out two pairs of simulations with the CAM2 model with prescribed sea surface temperatures (SSTs). We first used the standard CAM2 with the relative humidity dependence for low convective cloud and carried out a control simulation with climatological SSTs and a simulation with a structured perturbation in SSTs. The perturbation SSTs were obtained from the monthly mean 1% simulation of CCSM2 at the time of doubling. Thus, we are “forcing” the atmosphere model with a change in SST that matches that simulated by the coupled model. These two simulations then produce a change in radiative fluxes. We then carried out a parallel set of simulations where we turned off the relative humidity parameterization and employed the convective low-cloud fraction parameterization that uses cloud mass flux as the determinant. The change in the cloud parameterization explains ∼85% of the difference in shortwave cloud forcing, which indicates that the larger Λ_swcf in CAM2 is due mainly to this single change in the cloud fraction parameterization for low-level convective cloud fraction.

This approach also exhibits the power of using structured SST perturbation simulations in atmospheric models to diagnose feedback processes. The advantage of this approach is that prescribed SST simulations only need to be run for a few years, and thus are computationally far more affordable than running the fully coupled model for multiple decades.

4. Resolution dependence

As discussed in section 2, there are three versions of the CCSM3, which are defined in terms of the horizontal resolution of the atmosphere and ocean components. The low-resolution version of the CCSM3 employs a T31 version of CAM3 and a 3° version of the POP ocean component. The moderate-resolution model employs a T42 version of CAM3 and a 1° version of POP. The high-resolution version of CCSM3, used in the IPCC simulations, employs a T85 version of CAM3 and the 1° version of POP. In this section we comment on the climate sensitivity of these three versions of CCSM3.

As shown in Table 4, it is apparent that the equilibrium climate sensitivity of CCSM3 monotonically increases with increased horizontal resolution, where the high-resolution (T85) version is 17% greater than the low-resolution (T31) version. The transient climate response of the high-resolution (T85) version is only 5% larger than the low-resolution (T31) CCSM3. The response in the hydrological sensitivity across resolution are, in general, smaller than the changes that have occurred in the evolution of the CSM1.4, CCSM2 to the CCSM3.

Table 5 provides a further breakdown to the climate feedback parameter, Λ, in terms of the clear-sky and cloudy-sky components. The increase in the longwave clear-sky feedback parameter with increased resolution is due to changes in the vertical distribution of water vapor. The lower shortwave clear-sky feedback parameter for the T31 version of CCSM3 is due to an overestimate of simulated sea ice volume for this version compared to the T42 or T85 versions of the model. This larger volume of sea ice responds less to the warming due to increased CO₂. This result is counter to the concept that models that are colder and have more extensive sea ice have higher climate sensitivity. Meehl et al. (2004) comment further on the reason for this difference, which is related to the complexity of the sea ice model.

In terms of cloud feedbacks, the largest difference occurs for the change in the shortwave cloud feedback factor between the T42 and T85 versions of the CCSM3, where the models differ by 33%. The temporal evolution of the change in longwave and shortwave cloud forcing for the three 1% simulations is shown in Fig. 8. It is clear from this figure that the change in shortwave cloud forcing in the T85 resolution version of the model is systematically lower than the other two resolutions. What is the major reason for this difference? We have looked at the changes in global and tropical low-, middle-, and high-level cloud fraction and have found that this lower-cloud climate sensitivity is due to the behavior of tropical low-cloud amount in the T85 version of the CCSM3. Figure 9 shows the temporal evolution of tropical low-cloud fraction from the 1% simulations of low-, moderate-, and high-resolution (T85) versions of the CCSM3. Note that the T85 change in tropical cloud fraction is a factor of 2 lower than either the T31 or T42 versions of the model. Beyond the time of doubling, year 70, the change in low cloud for the high-resolution (T85) model remains around 5%, while the change in cloud fraction for the T31 and T42 models continues to monotonically increase.

This behavior is reminiscent of low-cloud response that was observed in an intermediate version of the CAM3 at T42 resolution. This behavior was discovered when analyzing results from the 2 × CO₂ slab ocean simulations of this model. It was related to an overprediction of low-cloud amount in the lowest model level in the control version of the model (i.e., 1 × CO₂). Under the condition of doubled CO₂, there was a large decrease in the cloud fraction leading to an anomalous change in shortwave cloud forcing and tropical sea surface temperatures. The reason for the overprediction in low-level cloud was traced back to inefficient mixing of drier air into the lowest model level, which led to a near-saturated model level. This problem was addressed in this version of the CAM by bounding the depth of the boundary layer to ensure efficient mixing across this layer. It appears that this adjustment to the boundary layer parameterization is not sufficient to prevent this pathologic behavior to occur in the T85 version of CAM3.

To conclusively illustrate this behavior, Fig. 10 shows the geographic distribution of the change in the equilibrium annual mean surface temperature for the slab ocean T42 and T85 versions of CCSM3. The global mean change in surface temperatures agrees to within 11%, but there are large regional differences in response between these two versions of CCSM3. For example, in the eastern equatorial Pacific the T42 surface warming is 2°C, while in the T85 version the warming is more than twice this amount. The surface warming over many of the continental regions is larger at the higher horizontal resolution version of the CAM3. There is a strong correlation between these regional changes in surface temperature and the changes in low cloud (see Fig. 11). There is a ∼15% decrease in eastern tropical Pacific low-cloud cover in the T85 model compared to a ∼2% decrease in the T42 model. The larger decrease in low-cloud cover at T85 results in more solar radiation reaching the ocean surface and hence a larger surface warming than at T42. Similarly, the larger response in surface warming in western Africa is also associated with a larger cloud decrease in this region at the T85 horizontal resolution. An analogous anticorrelation between low cloud and surface warming occurs over western Australia. The large difference in cloud response in the deep Tropics is a signature of anomalous behavior seen in the intermediate version of the CAM. Thus, improvements are required in the boundary layer parameterization to address this low-cloud problem.

5. Summary

The present study has explored the evolution of the climate sensitivity of three versions of the CCSM: CSM1.4, CCSM2, and CCSM3. The CCSM community has developed these versions of the model over the past 10 yr. The equilibrium climate sensitivity of these models is: 2.01°, 2.27°, and 2.47°C, respectively. Where the atmospheric resolution is fixed across these versions, the transient climate response to a 1% compound increase in CO₂ in the fully coupled models is 1.44°, 1.09°, and 1.48°C, respectively. It is found that there are significant changes in both clear-sky and cloudy-sky feedbacks across these model versions. The longwave clear-sky feedback parameter monotonically decreases (i.e., an increase in sensitivity) from CSM1.4 to CCSM3, and is due to changes in the vertical distribution of water vapor and improvements to the parameterization of the water vapor continuum. The changes in the clear-sky feedback parameter are due to improvements in the simulated sea ice properties in the coupled model. Note that the similarity of the TCR between CSM1.4 and CCSM3 is due to compensation between the differences in the climate sensitivity and ocean heat uptake of these two models.

The lower climate response in the CCSM2 is related to a change in the way convective cloud fraction is diagnosed in the CAM2 compared to the CCM3 and CAM3 models. We carried out atmospheric model simulations employing structured SSTs to show that this single change in convective cloud fraction explains most of the difference in cloud response in CCSM2.

Exploration of the dependence of the equilibrium climate sensitivity on horizontal resolution of the Community Atmosphere Model version 3 (CAM3) indicates that the lowest-resolution version (T31) has the lowest sensitivity of 2.32°C, while the highest-resolution version (T85) has the highest sensitivity of 2.71°C. Examination of differences in the geographic response of clouds among the various resolutions indicates that a low-cloud response can have marked differences among the three resolution versions of CAM3. The transient model simulations exhibit a similar dependence of cloud response on horizontal resolution. The anomalous behavior in low-cloud response in the T85 version of the CCSM3 is due to deficiency in the control simulation in boundary layer mixing processes, where the lowest model level reaches near saturation conditions, allowing for an overestimate in cloud fraction. Under warming conditions, this low cloud is dissipated, leading to an enhanced low-cloud feedback in this version of the model.

The purpose of the present study is to document the climate sensitivity of the CCSM and compare this sensitivity to previous versions of the CCSM and various horizontal resolution versions of the model. Diagnosis of these simulations indicates that low-cloud processes in the various versions of the CAM contribute to differences in the equilibrium climate sensitivity and the transient climate response. We have also shown the value in using structured SST perturbations in the CAM to study regional cloud feedback processes, which provides a very computationally efficient method to understand model processes.