1. Introduction

The surface equilibrium temperature change in response to a doubling of the atmospheric CO₂ concentration is at the center of the climate change studies. Current global climate models (GCMs) give a wide range of global mean equilibrium surface temperature responses from 2 to 4.5 K for a doubling of CO₂ (Solomon et al. 2007). The spread in model climate sensitivity is a major source of uncertainty for climate change projection of GCMs. It is understood that climate sensitivity is largely determined by internal feedback processes that amplify or dampen the influence of radiative forcing of CO₂ on climate. Therefore, analyzing the strength and spatial distribution of climate feedbacks, quantifying their contributions to the global and regional temperature changes, and identifying related physical process parameterizations in GCMs are critically important for improving our understanding of climate sensitivity and reducing the uncertainty in future climate projection.

The climate feedback studies have mostly concentrated on quantifying the strength of global mean radiative feedbacks such as water vapor, cloud, surface albedo, and temperature lapse rate feedbacks (e.g., Colman 2003; Soden and Held 2006; Bony et al. 2006). However, it is difficult to identify the key physical processes behind the feedback only based on global mean climate feedback and response analysis (Taylor et al. 2011). Studies show that both climate changes and feedbacks have pronounced geographic variability. For example, observations show that the Arctic surface temperatures have warmed at approximately twice the rate of the global mean in recent decades (Comiso and Parkinson 2004; Bekryaev et al. 2010), a phenomenon commonly known as Arctic amplification. The surface albedo feedback is largest in regions with snow and ice cover and very small over tropical oceans (Winton 2006). A number of studies have looked at the geographic distribution of climate feedbacks. Colman (2002) evaluated the strength and spatial distribution of the top-of-the-atmosphere (TOA) radiative perturbations attributable to radiative feedbacks in a low-resolution GCM using the partial radiative perturbation (PRP) method (Wetherald and Manabe 1988). Taylor et al. (2011) applied the PRP method with the Monte Carlo independent column approximation technique to model output from the National Center for Atmospheric Research (NCAR) Community Climate System Model, version 3 (CCSM3), to investigate the geographical distributions of radiative feedbacks. They showed that the shortwave cloud feedback exhibits the largest spatial variability, whereas the lapse rate feedback, which has the second largest spatial variability, explains the most spatial variance of the surface temperature response. However, these studies did not provide a direct estimate of the surface temperature change caused by an individual feedback. Crook et al. (2011) evaluated the spatial pattern of climate feedback and the corresponding surface temperature response using the regression method of Gregory et al. (2004) for eight climate models from phase 3 of the Coupled Model Intercomparison Project (CMIP3). They derived the clear-sky and cloud feedback parameters by performing a linear regression of the clear-sky radiative flux and cloud radiative forcing against the surface temperature, respectively, for the years before equilibrium is reached. The Planck feedback is obtained by the PRP method under clear-sky conditions. The water vapor plus lapse rate feedback is then determined by the difference between the longwave clear-sky feedback and Planck feedback. Because the cloud masking effects on the longwave radiative flux cannot be determined, the longwave cloud feedback is overestimated in their analysis (Crook et al. 2011). They evaluated the performance of the linear regression method at different spatial scales and found that the regression analysis is only applicable to zonal means. In agreement with previous studies, they found that the greatest intermodel differences are in tropical cloud feedbacks. However, the evaluation of partial surface temperature response shows that the greatest intermodel spread in the equilibrium temperature response comes from the water vapor plus lapse rate feedback rather than cloud feedback. This indicates that the contributions of climate feedbacks to the TOA radiative budget and to surface temperature may be different. Therefore, quantifying the contribution of feedback to the global surface warming is important.

Every climate variable that responds to global surface temperature change and directly or indirectly affects the earth’s radiation budget can constitute a climate feedback (Bony et al. 2006). Traditionally the climate feedback analysis focuses on the radiative feedbacks associated with climate variables that directly affect the TOA radiation budget, namely, water vapor, cloud, albedo, and atmospheric temperature. The spatial pattern and strength of the feedback are measured by feedback parameters, defined as the ratio of the radiative perturbation at the TOA resulting from a specific feedback to the total change of the surface temperature. Several approaches have been proposed to diagnose radiative feedback perturbation at the TOA in GCMs. One widely used approach is the PRP method (Wetherald and Manabe 1988), which uses offline radiation calculations to compute the change in radiative fluxes at the TOA that results from substituting one feedback variable at a time from the perturbed climate state into the control climate. This procedure can be computationally expensive since it requires the radiative transfer model to be run for each feedback. Another one is the radiative kernel method proposed by Soden et al. (2008). While it facilitates an understanding of the spatial characteristics of the feedbacks, it cannot evaluate cloud feedbacks directly because of strong nonlinearities. It should be noted that both these two feedback analysis methods measure the climate feedback with radiative perturbation at the TOA or feedback parameter. They do not provide a direct estimate of the surface temperature change caused by an individual feedback. Therefore, it is difficult to directly use these two methods to quantify the contribution of climate feedback to surface warming. Although the online feedback suppression method, which evaluates the climate changes caused by suppressing one particular feedback in the model, can provide an estimate of the partial surface temperature change attributable to a specific feedback, this partial temperature change derived by suppressing one feedback in the model includes the compensating effects from other feedbacks (Cai and Lu 2009).

Recently, Lu and Cai (2009) formulated a coupled atmosphere–surface climate feedback–response analysis method (CFRAM) for quantifying the contributions of climate feedback to climate change. It directly isolates the partial temperature change attributable to an individual feedback by requiring the infrared radiation induced by the temperature change alone to balance the energy flux perturbation resulting from the feedback process under consideration. In the CFRAM, every physical and dynamic process in the model that responds to temperature change and that directly or indirectly affects the earth’s energy budget is considered a climate feedback process. Therefore, in the CFRAM climate feedbacks include not only the radiative processes associated with climate variables that directly affect the TOA radiation budget (e.g., water vapor, cloud, and albedo), but also any dynamic and physical processes that redistribute energy horizontally and vertically with no net radiative effect at the TOA (e.g., convection, surface sensible and latent heat fluxes, dynamical advection, and boundary layer process). This process-based feedback decomposition enables us to understand the roles of model physical and dynamic processes in climate change and climate sensitivity spread between the models. Note that because the CFRAM considers temperature changes both in the atmosphere and at the surface together as a coupled climate response, the atmospheric temperature or lapse rate feedback defined in the TOA-based feedback analysis methods (e.g., PRP and radiative kernel) no longer exists in the CFRAM.

In this study, we apply the CFRAM to the NCAR CCSM3 using the slab ocean model (CCSM3-SOM) to determine the global surface warming attributable to a doubling of CO₂ concentration. The spatial pattern of surface temperature changes will be examined and the contributions from the external forcing (the doubling of CO₂) and individual feedbacks will be quantified. Because of the computational limit, past climate feedback analysis is typically based on monthly-mean data from the model simulations. However, since atmospheric states have significant diurnal and intramonthly variations, the feedback analysis with monthly-mean data may result in large biases. For example, Taylor et al. (2011) evaluated the geographical distributions of the radiative feedbacks by applying the PRP method to monthly-mean model output of the CCSM3. In their calculation, the sum of TOA radiative perturbations from all individual feedbacks derived with the offline PRP method is smaller than the model-simulated TOA radiative response in most areas by up to 5 W m⁻². Shell et al. (2008) diagnosed the climate feedbacks in the NCAR Community Atmospheric Model, version 3 (CAM3), with the radiative kernel method. They compared the TOA clear-sky radiative flux change resulting from a doubling of CO₂ in the CAM3 to the sum of partial radiative perturbations derived by radiative kernel technique and found that the magnitude of shortwave change is underestimated by 23% for global mean in the kernel technique when monthly-mean data are used. The global mean bias in TOA shortwave flux is reduced to 10% when 3-hourly data are used. Taylor et al. (2013) applied the CFRAM to 20-yr mean data to diagnose temperature changes in the NCAR Community Climate System Model, version 4 (CCSM4). Their residual term, which is mainly attributable to subannual variability, is comparable in magnitude to individual feedbacks. These suggest that high temporal resolution data are required in order to quantify the contributions of climate feedback accurately. In this study, we apply the CFRAM method to 10-yr hourly model output to estimate the contribution of climate feedback to global and regional temperature changes.

Cloud feedback has been confirmed to be the primary source of climate sensitivity spread among climate models (Randall et al. 2007; Soden and Held 2006). Therefore, the cloud feedback analysis is particularly important for understanding climate sensitivity. Since clouds are nonlinear, primarily as a result of their complicated overlap patterns, the cloud feedback analysis using time-mean cloud information inevitably results in large biases. By using hourly data, the offline radiation calculation of cloud feedback uses exactly the same cloud information as that in the GCM simulation, ensuring the accuracy of cloud feedback in our analysis.

The paper is organized as follows: Section 2 briefly describes the model and experiment design. Section 3 describes the procedure to apply the CFRAM to the CCSM3-SOM simulations. Section 4 examines the spatial pattern of the surface warming in the CCSM3-SOM in response to the doubling of CO₂ and quantifies the contributions from external forcing and individual feedbacks to the surface warming. Section 5 presents the regional contrasts of the surface warming between low and high latitudes and between land and ocean. A summary and conclusions are given in section 6.

2. Model and experimental setup

The NCAR CCSM3 (Collins et al. 2006a) is a coupled global climate model consisting of four component models (atmosphere, ocean, land, and sea ice) linked by a central coupler. The atmospheric model, CAM3 (Collins et al. 2006b), is a global atmospheric general circulation model with spectral T42 truncation (approximately 2.8° × 2.8° latitude–longitude) in the horizontal and 26 levels in the vertical from the surface to 2.917 hPa. The land surface model is the Community Land Model (CLM), which uses the same horizontal grids as the atmospheric model. The ocean model used in this study is a slab ocean model (SOM) for the upper ocean, in which the ocean temperature is a prognostic variable, fully interacting with the surface energy flux. Because of the lack of ocean dynamics (current), an internal ocean mixed layer heat flux (Q flux) is specified, emulating the horizontal ocean heat transport and deep-water heat exchange. The sea ice model, the Community Sea Ice Model (CSIM), shares a common horizontal grid with the ocean model. In the slab ocean model version of CCSM3, only the thermodynamic sea ice component is employed. Because the ocean–sea ice fluxes are not well constrained because of inadequate observations, it is difficult to simulate a reasonable sea ice mass balance (Bitz et al. 2012). To ensure that the model simulates reasonable sea ice distribution and rate of change, the SOM employs an adjustment to the Q flux in the presence of sea ice. A further global adjustment is made to the Q flux to maintain zero adjustment on the global mean.

We conducted two 50-yr equilibrium simulations. In the present-day climate simulation (referred to as the 1×CO₂ run) the 1990 CO₂ value of 355 ppmv is used, whereas in the perturbation climate simulation (referred to as the 2×CO₂ run) the CO₂ concentration is instantaneously doubled to 710 ppmv. The spatially varying ocean mixed layer depth is specified from Levitus et al. (1998). The monthly-mean distribution of ocean heat transport is obtained from the net ocean surface energy budget of 10-yr (1981–90) Atmospheric Model Intercomparison Project (AMIP)-type simulation of the CAM3. Both simulations start from the same initial conditions obtained from the NCAR repository. The time series of global and annual mean surface temperatures from 1×CO₂ and 2×CO₂ runs suggest that the model attains equilibrium with the doubled CO₂ concentration after 25 yr (not shown). Therefore, the climate statistics and climate feedback analysis presented in this paper are based on hourly output from year 30 to 39 of the simulations.

3. Application of CFRAM

From Lu and Cai (2009), the difference of the energy conservation equation of CCSM3-SOM between two equilibrium states in an atmosphere–surface column can be written as

View Expanded

where stands for the difference between the two equilibrium states. Here is the vector of vertical convergence of radiative energy flux from the surface to the top level of the model atmosphere. The superscript T represents the transpose. Also, is the vector of nonradiative energy flux convergence, where the subscript pd is substituted with conv for atmospheric convection, adv for dynamical advection, ocn for ocean processes, pbl for vertical diffusion and boundary layer processes, lhf for surface latent heat flux, shf for surface sensible heat flux, cond for large-scale condensation, and gwd for gravity wave drag. Note that small energy changes resulting from model numerics in the dynamic core such as time filtering, energy correction, and numerical diffusion have been included in the dynamical advection term. For the sake of brevity, hereafter we will refer to the changes of energy flux convergence simply as “energy perturbations.”

The radiative energy perturbation can be decomposed into perturbations resulting from the doubling of CO₂, temperature change (T), water vapor (wv), albedo, and cloud (cld) feedbacks based on the linear approximation of radiative feedback and then further into longwave and shortwave components:

View Expanded

where is the vector of vertical convergence of net solar radiation fluxes, is vector of vertical divergence of net longwave radiation fluxes, and is vector of temperature from the surface to the top level of the atmospheric model (TOM). The term is the Planck feedback matrix. Note that the impact of changes of aerosol radiative properties associated with relative humidity change is combined into water vapor feedback. Err is the residual error resulting from various approximations involved in the decomposition calculation.

If the residual error term is negligible, substituting Eq. (2) into Eq. (1), we obtain

View Expanded

where is the inverse of the Planck feedback matrix. The subscript pr is substituted with wv, cld, and albedo for water vapor, cloud, and albedo feedbacks, respectively. From Eq. (3), the partial temperature changes attributable to the doubling of CO₂ and each radiative and nonradiative feedback can be obtained by solving

View Expanded

This means that if the analysis is accurate enough (i.e., the Err term is small) the total equilibrium temperature change from the CCSM3-SOM model simulation output can be approximated by the sum of partial temperature changes resulting from each feedback and doubled CO₂ forcing calculated from Eq. (4). Thus, by comparing this sum with temperature changes from CCSM3-SOM output, we can validate the accuracy of our analysis. More importantly, we can use Eqs. (3) and (4) to identify the contribution from doubled CO₂ forcing and each subsequent feedback process to global surface warming.

The element of nonradiative energy flux convergence resulting from any nonradiative feedback process (conv, pbl, adv, etc.) for each layer at a grid point is derived from the total energy [] change before and after the calculation of process as follows:

View Expanded

where is the velocity vector, is pressure depth of the layer, and g is gravitational acceleration. The represents the time interval for the leapfrog time integration scheme. The value of is calculated at each time step during the model integration and output directly from the CCSM3-SOM.

The radiative energy perturbation resulting from each radiative feedback process (wv, cld, and albedo) is determined by offline calculations using the partial radiative perturbation method. Different from the standard PRP method, the perturbation of radiative energy flux convergence at each model level and the surface rather than only at the TOA is used in the CFRAM analysis. In previous work applying the CFRAM approach (Cai and Tung 2012; Taylor et al. 2013), long-term mean data, such as temperature, cloud, water vapor, and albedo, were used as input for offline radiative transfer calculations. This led to large errors [the Err term in Eq. (2)], comparable in magnitude to the actual feedbacks. A unique feature of this study is the use of high-frequency output from the CCSM3-SOM runs to obtain a more accurate estimate of the contribution from each feedback to surface warming from the doubling of CO₂. We use 10 yr of hourly output from the 1×CO₂ and 2×CO₂ runs. In previous studies (e.g., Kiehl et al. 2006) equilibrium climate sensitivity is usually estimated from 20-yr or longer simulations. However, we note that there is no significant difference between 10-yr (years 30–39) and 20-yr (years 30–49) averages for nonradiative feedbacks (nonradiative feedback terms are saved every time step and averaged to monthly files, therefore they do not depend on whether monthly or hourly data are used), and we expect the same for radiative feedbacks. Note that the PRP method assumes that all feedback variables are temporally uncorrelated with each other. Because some feedback variables are generally highly correlated (e.g., cloud and water vapor) at short time scales, the standard PRP method may produce substantial bias in the radiative perturbation calculation when hourly model data are used (Soden et al. 2008). To remedy this problem, we perform a two-sided PRP calculation for each feedback as suggested by Colman and McAvaney (1997), Colman (2002), and Soden et al. (2008). First, the radiative energy perturbation resulting from feedback in the control climate (1×CO₂) is determined by the offline radiation calculation with all input hourly variables from 1×CO₂ run, but substituting with that from the 2×CO₂ run. Then, the radiative energy perturbation resulting from feedback in the perturbed climate (2×CO₂) is determined by the offline radiation calculation with all input variables from 2×CO₂ run, but substituting with that from the 1×CO₂ run hour by hour. The average of the radiative energy perturbations resulting from feedback in the perturbed climate and control climate gives the final radiative energy perturbation resulting from feedback . The radiative energy perturbation attributable to external forcing (a doubling of CO₂) is determined similarly.

To obtain partial temperature changes attributable to individual feedback agents, we need to determine the Planck feedback matrix in Eq. (2) at every longitude–latitude grid point. Because temperature change in one layer of the atmosphere–surface column affects the longwave radiative flux at every level of the column, the jth (j = 1, … , 27) column of the Planck feedback matrix represents the vertical profile of the change rate of longwave radiation flux divergence with temperature changes in the jth layer. It is determined by a two-sided PRP calculation with 1-K temperature perturbation in the jth layer using the hourly output from 1×CO₂ and 2×CO₂ runs. Thus, at a given grid point and time the offline radiation calculation needs to be run 2N + 2 times, where N = 26 is the number of atmospheric model levels. This procedure is computationally expensive when hourly data are used. Therefore, hourly offline radiation calculation for 1-K temperature perturbation from each layer is only performed for 1 day per month (the fifteenth day) for year 36. The average of these 576 (=24 × 12 × 2) samples of longwave radiation energy perturbation yields the mean Planck feedback matrix , which is then inverted to obtain needed in Eqs. (3) and (4). Note that subsampling is only applied to because of the high computational toll. All other terms of perturbation energy in Eq. (2) use the 10-yr average of hourly data.

Because the energy perturbations resulting from nonradiative feedbacks are directly calculated by the CCSM3-SOM, the bias in the CFRAM analysis is mainly from the calculation of the Planck feedback matrix and to some extent from the linear decomposition of radiative energy perturbation as well. As stated earlier, the partial radiative energy perturbation attributable to external forcing and radiative feedbacks (water vapor, cloud, and albedo) are determined by the two-sided PRP method, while that attributable to temperature changes can be estimated by the product of CCSM3-SOM temperature change and the Planck feedback matrix. Thus, we can compare the sum of partial radiative perturbations attributable to the external forcing, radiative feedbacks, and temperature change with the total radiative energy change obtained directly from the CCSM3-SOM output. The difference between the two represents the aforementioned two biases, which reflect the accuracy of the CFRAM analysis using hourly data. Figure 1a shows the latitude–pressure cross section of the zonal annual mean total radiative energy change attributable to the doubling of CO₂ from the CCSM3-SOM simulation output () and the sum of the partial radiative energy perturbations attributable to external forcing, each feedback process, and temperature change . The sum of partial radiative energy perturbations derived from the offline radiation calculation is visually indistinguishable from the total radiative energy change output from the CCSM3-SOM, demonstrating that the Planck feedback matrix and the decomposition of radiative energy change in the atmosphere using hourly data are very accurate. Figure 1b shows the geographical distribution of the annual mean surface radiative energy change attributable to the doubling of CO₂ from the CCSM3-SOM simulation output and the sum of the partial radiative energy perturbations. The two are again in excellent agreement in the mid- and low latitudes, although there are some noticeable differences in high latitudes. The global mean bias in the sum of partial radiative energy change derived with CFRAM analysis is −0.03 W m⁻² at the surface, compared to the radiative energy change of 2.81 W m⁻² directly from the CCSM3-SOM simulations, indicating that the decomposition of the surface radiative energy change is also reliable.

View Full Size

Annual mean of total radiative energy perturbations from CCSM3-SOM (shaded, W m⁻²) and the sum of partial radiative energy perturbations (contours) (a) in the atmosphere and (b) at the surface. The global mean value shown above (b) is for CCSM3-SOM, and the mean bias and RMSE are with respect to CCSM3-SOM output.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

4. Global surface warming pattern

Figure 2 presents the annual mean atmospheric and surface temperature change attributable to the doubling of CO₂ from the CCSM3-SOM (shadings) and the sum of partial temperature changes attributable to the external forcing and each feedback process [, contours] to further demonstrate the high accuracy of the CFRAM analysis using hourly data. The two fields largely overlap with each other, indicating that the CFRAM is appropriate for quantifying contributions of climate feedbacks to atmospheric warming projected by CCSM3-SOM. The rest of this paper will focus on the analysis of global surface warming.

View Full Size

Annual mean of total temperature change from CCSM3-SOM (shaded, K) and the sum of partial temperature changes attributable to all processes (contours) (a) in the atmosphere and (b) at the surface. The global mean value shown above (b) is for CCSM3-SOM, and the mean bias and RMSE are with respect to CCSM3-SOM output.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

The global mean surface warming from the 10-yr CCSM3-SOM simulation is 3.16 K. The CFRAM analysis almost exactly reproduces the surface temperature change of the CCSM3-SOM in the mid- and low latitudes, with some noticeable cold biases in high latitudes. The spatial pattern correlation between the two is 0.94. The global mean bias in the sum of the partial surface temperature change derived with CFRAM analysis is −0.07 K, which is about 2% of the mean surface temperature change from the CCSM3-SOM simulations. Note that the relative bias in the estimate of surface temperature change using CFRAM is higher than that in radiative energy change (1%) because of some inaccuracy in the calculation of the Planck matrix.

Similar to the multimodel projection from the Intergovernmental Panel on Climate Change (IPCC) Fourth Assessment Report (AR4; Meehl et al. 2007), the global surface warming in response to a doubling of CO₂ projected by CCSM3-SOM displays significant spatial variation. The most striking feature is that the warming in high latitudes is more than twice that in the tropics, with a maximum warming in high latitudes of the Northern Hemisphere. In addition, there is a tendency for much stronger warming over land than over oceans.

The CFRAM analysis enables us to investigate the strength and spatial pattern of the partial surface temperature changes attributable to external forcing and each climate feedback process. Figure 3 presents the partial surface temperature changes attributable to the doubling of CO₂ as external forcing and each radiative feedback. The global mean and corresponding standard deviation for each process are provided above each frame for reference. The changes of the basic fields, such as cloud fraction, water vapor, and precipitation are presented in Fig. 4 for ease of discussion. The partial surface temperature changes attributable to external forcing [; Fig. 3a] exhibit the smallest spatial variation, with a standard deviation of 0.24 K. The global mean contribution is 1.31 K, with minimum warming in the tropics. The water vapor feedback [; Fig. 3b] contributes the most (2.47 K) to the global mean surface warming. However, its spatial variation is relatively small. The spatial pattern of water vapor feedback is similar to that of column-integrated water vapor change from the 1×CO₂ to the 2×CO₂ simulation (Fig. 4a), with maxima in tropical convective regions. Noting that changes in column-integrated water vapor are mainly determined by the water vapor changes in the lower troposphere, the results indicate that the lower-tropospheric water vapor plays an important role in determining the water vapor feedback in the surface warming. This is because the CFRAM analysis here focuses on the surface energy budget, and the downward longwave radiation at the surface is more strongly affected by water vapor change in the lower troposphere than the upper troposphere. In contrast, the upper-tropospheric water vapor is more important in the traditional water vapor feedback analysis, which focuses on the radiation budget at the TOA.

View Full Size

Partial surface temperature changes attributable to radiative processes: (a) CO₂ forcing, (b) water vapor feedback, (c) albedo feedback, (d) cloud feedback, (e) longwave cloud feedback, and (f) shortwave cloud feedback. The global mean and corresponding standard deviation above each frame show the warming amplitude and spatial variability from each process.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Annual mean difference between 2×CO₂ and 1×CO₂ runs: (a) precipitable water, (b) sea ice fraction, (c) cloud fraction, (d) cloud liquid water path, (e) large-scale snow rate, (f) convective precipitation rate, (g) sensible heat flux, and (h) latent heat flux.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

The surface temperature changes attributable to albedo feedback are concentrated in narrow sea ice zones with very strong warming and are small in mid- and low latitudes. As a result, it has the largest spatial variation with a standard deviation as high as 3.14 K. The global mean contribution of albedo feedback to the surface warming is 1.23 K; however, it can be as high as 10 K in the high-latitude sea ice zone. The warming attributable to albedo feedback in the Northern Hemisphere land can be attributed to snow melting in warm climate, while the warming over the sea ice zone is a result of the decrease in the sea ice fraction (Fig. 4b).

The partial surface temperature changes attributable to cloud feedback have large spatial variability. However, the global mean contribution is only 0.02 K. This is consistent with recent studies (Shell et al. 2008; Soden et al. 2008) that show that cloud feedback is slightly positive or near neutral when the effect (or contamination) of noncloud factors on cloud radiative forcing is considered. Note that the cloud feedback derived directly from modeled cloud radiative forcing is often negative (Shell et al. 2008; Soden et al. 2008). Recent observational analysis using the Cloud and the Earth’s Radiant Energy System (CERES) data also find the short-term cloud feedback to be fairly small, but most likely positive (Dessler 2010). The temperature changes attributable to cloud feedback are mostly negative in the tropical oceans except in the eastern Pacific, the South China Sea, and the Bay of Bengal, but positive in midlatitude oceans. The contribution of cloud feedback can be further decomposed into shortwave and longwave components, which are in general out of phase (Figs. 3e,f). In addition, in the mid- and low latitudes the spatial pattern of cloud feedback is dominated by shortwave cloud feedback, while in the polar regions it is dominated by longwave cloud feedback. The longwave cloud feedback shows a good positive correlation with cloud fraction changes (Fig. 4c). There is longwave surface warming where the cloud fraction is increased because of the cloud greenhouse effect. The strong warming (by up to 5 K) attributable to the longwave cloud feedback in the polar regions is mainly related to the increase of high-level cloud fraction, whereas the cooling in Europe, the South China Sea, the eastern tropical Pacific, the northeastern Pacific, the northern Atlantic, and the southern subtropical ocean regions is owing to the decrease of cloud fraction. The global mean contribution of longwave cloud feedback to surface temperature change is 0.44 K. The spatial pattern of shortwave cloud feedback is negatively correlated with changes in the cloud liquid path as well as changes in cloud fraction (Figs. 4c,d). There is surface cooling where the cloud liquid water path or cloud fraction is increased because of the corresponding increase in cloud albedo and consequently the decrease of absorbed solar radiation at the surface (cloud albedo feedback). The longwave warming and shortwave cooling from clouds in the tropical Pacific and southern Indian Ocean is attributable to increased high-level clouds associated with ITCZ convection. The spatial variation of shortwave cloud feedback is very large, with a standard deviation as high as 2.41 K. With dominant contribution coming from water vapor and albedo feedbacks, the global mean impact of the sum of radiative feedbacks (water vapor, cloud, and albedo feedbacks) is a net surface warming of 5.0 K.

In contrast, the nonradiative feedbacks (Fig. 5) tend to dampen the warming at the surface on global average except for the surface sensible heat exchange and dynamical advection process, which tend to warm the surface slightly. The largest cooling contribution to surface temperature comes from convection , which contributes to the global mean surface temperature change by −0.97 K. The spatial distribution shows strong cooling over the ITCZ and South Pacific convergence zone (SPCZ) regions and strong warming flanking it. There is also cooling in the midlatitude storm tracks. The pattern is well correlated with the change of convection intensity, as indicated by the change of convective precipitation rate (Fig. 4f). There is surface cooling (warming) where convective precipitation is increased (decreased). The enhanced convection over the ITCZ region transports more energy from the lower troposphere to the upper troposphere, resulting in a reduced temperature lapse rate. This leads to less downward infrared radiation at the surface and therefore lower surface temperature. Convection is suppressed on both flanks of the enhanced ITCZ convection by the compensating subsidence outside the ITCZ convection. The suppressed convection transports less energy from the lower troposphere to the upper troposphere, resulting in an enhanced temperature lapse rate. This leads to less outgoing longwave radiation and more downward infrared radiation at the surface and therefore the surface warming in regions of suppressed convection. The spatial pattern of the surface temperature change attributable to dynamical advection is largely opposite to that from convection in the mid- and low latitudes, indicating that the dynamical advection tends to offset the impact of convection. The global mean energy perturbation resulting from dynamical advection averaged between 1000 and 600 hPa is 0.07 W m⁻², while it is −0.13 W m⁻² for the average between 600 and 10 hPa. This indicates that the dynamical advection tends to increase the temperature lapse rate slightly in the global mean, which results in less outgoing longwave radiation to the space and more downward infrared radiation into the surface and therefore slight warming at the surface. Although the global mean contribution from the dynamical advection process to the surface temperature change is nearly zero, it has large spatial variability, with a standard deviation of 1.88 K. It shows strong cooling north of the Pacific ITCZ and strong warming over the Pacific ITCZ and SPCZ and the southern Indian Ocean. Warming in the storm track regions by dynamical transport in both hemispheres is also apparent.

View Full Size

Partial surface temperature changes attributable to nonradiative feedbacks: (a) convection, (b) dynamical advection, (c) ocean process, (d) PBL process, (e) latent heat flux, (f) sensible heat flux, (g) large-scale condensation, and (h) gravity wave drag. The global mean and corresponding standard deviation above each frame (except for the ocean process) show the warming amplitude and spatial variability from each process. The statistics for the ocean process are only applied to ocean grid points.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

The ocean process produces a −0.49-K mean surface cooling over the ocean grid cells (ocean fraction >0.001), with large spatial variation. This contributes to a global mean cooling of −0.38 K. It consists of two components, the internal energy change and the adjustment of the Q flux. As mentioned earlier, an adjustment to the Q flux is applied in the SOM in the presence of sea ice to ensure that the model simulates reasonable sea ice distribution and change rate. A further global adjustment is made to the Q flux to maintain zero adjustment on the global mean. Therefore, different adjustments of the Q flux in the 1×CO₂ and 2×CO₂ runs contribute to the ocean process feedback. The change of internal energy of the ocean mixed layer is zero when the mixed layer model reaches a true equilibrium. However, in the current study, the ocean mixed layer is in a quasi-equilibrium state. It could be nonzero locally and acts as an ocean process that contributes to the SST change. A decomposition of partial surface temperature change into components associated with internal energy change and the adjustment of the Q-flux term (not shown) finds that the global mean cooling of −0.34 K comes from the ocean Q-flux adjustment, while the contribution from the ocean internal energy change is only −0.04 K. The strong cooling in high-latitude sea ice zones associated with ice melting in a warmer climate is attributable to the adjustment of the Q-flux term. The warming in the North Pacific and cooling in the North Atlantic are both caused by the ocean internal energy change term. In the absence of a dynamic ocean model, they cannot be related to any specific dynamic or physical process in the ocean.

The contribution of the planetary boundary layer (PBL) process to surface temperature change is mostly negative, with a global mean contribution of −0.43 K. The PBL process does not directly contribute to the surface energy budget, but rather it affects the surface temperature by modifying the atmospheric energy budget and subsequently the downward infrared radiation. Therefore, the surface cooling induced by the PBL process can be attributed to the decrease of atmospheric energy in the lower part of PBL because of the enhanced upward transport by PBL turbulence in a warmer climate.

The contributions from the rest of the nonradiative feedbacks (sensible and latent heat fluxes, large-scale condensation, and gravity wave drag) to global mean surface temperature change are very small, with −0.16 K from latent heat flux and virtually zero from others. The increase in surface latent heat flux to the atmosphere in a warmer climate is responsible for the corresponding surface cooling. The slight surface warming in the global mean associated with sensible heat exchange is a result of the reduced sensible heat flux from the surface to the atmosphere, which is reduced from 19.77 to 18.35 W m⁻² in global mean in the 2×CO₂ run. The cooling contribution from large-scale condensation is concentrated in a narrow zone over the Southern Hemisphere sea ice region. The reduced freezing heating corresponding to less snow production in a warmer climate (Fig. 4e) results in less warming of the atmosphere, which leads to less downward infrared radiation. This in turn cools the surface. The contribution from the gravity wave drag (Fig. 5h) is negligible and will not be discussed in the rest of this paper.

Several atmospheric feedback processes, especially dynamical advection, display significant contribution to the surface temperature change in the Southern Ocean near 60°S. We further examine the partial energy perturbation in this region and find that (not shown) in a warmer climate the PBL turbulent transport is enhanced south of 63°S, which transports more energy from the lower PBL to the upper PBL. The moist physical processes (convection and large-scale condensation) are strengthened north of 63°S over the Southern Ocean, transporting more energy from the PBL to the free troposphere aloft. In addition, the enhanced surface latent heat flux provides more energy to the bottom layer of the PBL. The net effect of these physical processes is to increase the energy perturbation in the lower troposphere south of 63°S and decrease it north of 63°S. The dynamical advection tries to remove this energy perturbation by decreasing it in the lower troposphere south of 63°S and increasing it north of 63°S. As a result, the warming (cooling) in the lower troposphere north (south) of 63°S increases (decreases) downward longwave radiation at the surface, leading to a strong surface warming and cooling in respective regions over the Southern Ocean.

To quantify the roles of external forcing and each feedback process in the formation of the global surface warming pattern, we calculate the pattern correlation and the centered root-mean-square (RMS) difference of the partial surface temperature change attributable to external forcing and each feedback process with respect to the total surface temperature change from the CCSM3-SOM, as well as the ratio of variations (represented by standard deviation) between them. Figure 6 summarizes them in a Taylor diagram. The Taylor diagram concisely characterizes the statistical relationship between two fields, the partial and total surface temperature changes in this case. The pattern correlation between them is given by the azimuthal position. The normalized (divided by the standard deviation of total temperature change) standard deviation of partial surface temperature changes is proportional to the radial distance from the origin. The normalized RMS difference from the total temperature change is proportional to the distance from the point to the “REF” point on the x axis. The concentric half circles around REF indicate the RMS values. In general, if a feedback process plays an important role in producing the global surface warming pattern, the corresponding partial temperature change will have a relatively high pattern correlation and low RMS difference and lie close to the REF point. The pattern correlation and normalized RMS difference between the partial temperature change attributable to the external forcing and total temperature change are 0.36 and 0.95, respectively (point labeled 1), while those attributable to the total climate feedbacks are 0.93 and 0.36, respectively (point labeled 2), demonstrating that the climate feedbacks play a dominant role in the global surface warming pattern. Among the climate feedbacks, the longwave cloud feedback shows the largest pattern correlation (0.69) and smallest RMS difference (0.83) and therefore is the largest contributor to the model global warming pattern. The net cloud feedback (point labeled 3) is not as well correlated to the total warming because of the opposite contribution from shortwave cloud feedback. The albedo feedback shows the second largest pattern correlation (0.55) but with large RMS difference (1.78). All other feedback processes contribute to the global warming pattern to varying degrees, with convection and latent heat flux feedbacks contributing positively and water vapor feedback, shortwave cloud feedback, dynamical advection, large-scale condensation, sensible heat flux, and ocean process contributing negatively (not shown).

View Full Size

Taylor diagram quantifying the contribution of each feedback to the formation of global warming patterns. Cosine of the azimuth and radial distance from the origin represent the pattern correlation and the ratio of standard deviations of partial surface temperature change associated with each feedback to the total surface temperature change from the CCSM3-SOM. The distance to the REF point represents the normalized RMS difference from the total temperature change of CCSM3-SOM.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

5. Regional differences

a. High latitudes versus low latitudes

To gain more insight into the much stronger warming in high latitudes than in the tropics, the partial surface temperature changes attributable to external forcing and each feedback are averaged over high latitudes (60°–90°S and 60°–90°N) and the tropics (30°S–30°N), respectively, and are shown in Fig. 7. The surface temperature changes from CCSM3-SOM and the sum of partial surface temperature changes diagnosed using the CFRAM analysis are also shown in the first two columns in Fig. 7. There is a warming of 1.9 K in the tropics and 4.9 K in high latitudes, leading to a warming difference of 3.0 K between high and low latitudes in the CCSM3-SOM simulation. The CFRAM analysis successfully reproduces this latitudinal warming contrast with a slightly smaller magnitude (2.8 K). In the tropics, the major contributors to the warming are water vapor feedback (3.0 K) and CO₂ forcing (1.1 K). Convection (−1.0 K) and cloud shortwave feedback (−0.9 K) contribute negatively. In high latitudes, the main warming contributors are albedo (3.5 K), water vapor (2.2 K), cloud longwave feedback (2.1 K), and CO₂ forcing (1.4 K). The negative contributors are ocean processes (−1.5 K) and cloud shortwave feedback (−1.5 K). Since the warming attributable to albedo feedback is negligible in the tropics and is very large in high latitudes resulting from ice/snow melting, the albedo feedback itself contributes about 3.5 K to the warming difference, making it the largest contributor to the high- and low-latitude warming contrast. The second largest contribution comes from the longwave cloud feedback, which contributes 2.0 K to the warming difference. Because enhanced convection results in less downward infrared radiation and therefore relative cooling at the surface, the less increase in convection in high latitudes than in the tropics in a warmer climate leads to about a 0.5-K high–low latitude warming contrast. In addition, the surface latent heat flux and CO₂ forcing also have small positive contributions to the stronger warming in high latitudes. On the other hand, water vapor feedback, ocean processes, and shortwave cloud feedback reduce the latitudinal warming contrast by warming the high latitudes less (water vapor) or cooling the high latitudes more (ocean process and shortwave cloud feedback) than the tropics. All other feedback processes, that is, dynamical advection, large-scale condensation, PBL process, and surface sensible heat flux have small negative contribution to the stronger warming in high latitudes.

View Full Size

Regional average of partial surface temperature changes attributable to CO₂ forcing and each feedback over low (30°S–30°N) and high (60°S–90°S and 60°–90°N) latitudes. CCSM3 is from the model output and CFRAM is from the diagnostic analysis, with each contributing process shown in the x-axis label.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

b. Land versus ocean

The land surface is where human activities take place; its warming will directly affect food, economy, and the living environment of human society. Therefore, understanding the mechanism of land surface warming and improving its projection in response to the increase of greenhouse gases have more practical significance to the climate change community. Observations and climate model simulations consistently show that warming is greater over land than over ocean in response to increased greenhouse gases. However, the physical mechanism responsible for the land–ocean warming difference is still a matter of debate. Sutton et al. (2007) attributed the greater land warming to the relatively weaker surface latent heat flux, since comparatively dry land has much less potential to enhance surface evaporative cooling. However, they did not consider land–ocean interaction. Compo and Sardeshmukh (2009) studied the atmospheric model simulation forced with observed SST without considering changes of greenhouse gases. They simulated most of the observed land warming and therefore argued that the dominant mechanism for stronger land warming is the enhanced downward longwave radiation associated with increases of both upper-tropospheric humidity and temperature over land, which are induced by the ocean warming through changing the circulation and water vapor transport. Dong et al. (2009) investigated the mechanisms causing the initial development of the land–ocean warming contrast using an atmospheric general circulation model. They found that the land warming is enhanced because of increased downward shortwave radiation at the surface resulting from reduced cloud amount. Boer (2011) analyzed energy balance over land and ocean to investigate the land–ocean warming contrast in the climate change simulations of CMIP3 models. He found that the energy transport from the ocean to the land region enhances the land–ocean warming contrast.

To identify the major feedback processes that are responsible for the stronger warming over land than over ocean, the partial surface temperature changes attributable to the external forcing and each feedback averaged over global land and ocean are shown in Fig. 8a. Note that, in the following analysis, only the grid cells covered by 100% fraction of land (ocean) are counted as land (ocean) grids. The CFRAM analysis accurately reproduces the warming difference of 1.0 K between land and ocean in the CCSM3-SOM. The longwave cloud feedback produces close to 0.4-K warming over land and 0.4-K cooling over ocean, making it the largest contributor (0.8 K) to the land–ocean warming contrast. The difference in longwave cloud feedback between land and ocean can be attributed to changes of the vertical structure of cloud fraction. Clouds absorb upward longwave radiation from the surface and reemit infrared radiation at their local temperature. Although clouds at any height emit downward longwave radiation and warm the surface, low-level clouds are more efficient as a result of their closer proximity to the surface and higher temperature than high-level clouds. The net impact of longwave cloud feedback on the surface temperature depends on the relative changes of high-level and low-level clouds. Over land, there is a large increase of high-level cloud (1.99%) and a small decrease of low-level cloud (−0.14%), therefore producing a surface warming by longwave cloud feedback. Over oceans, there is a large decrease of low-level cloud (−0.79%), comparable to the increase of high-level cloud (1.21%), resulting in a surface cooling. The larger increase in convection over the tropical ocean results in greater decrease of lapse rate, which contributes about 0.6 K to the land–ocean warming contrast. In addition, the surface albedo feedback related to snow melting over land also contributes 0.4 K to the stronger land warming. Together these three processes account for 1.8-K warming contrast between land and ocean.

View Full Size

Regional average of partial surface temperature changes attributable to the CO₂ forcing and each feedback over land and ocean (a) between 90°S and 90°N and (b) between 50°S and 50°N. The x-axis label is the same as in Fig. 7.

Citation: Journal of Climate 27, 8; 10.1175/JCLI-D-13-00567.1

• Download Figure
• Download figure as PowerPoint slide

On the other hand, the PBL process produces more cooling over land than over ocean, contributing −0.4 K to the land–ocean warming contrast. The cooling difference attributable to the PBL process between land and ocean mainly reflects the regulation of convection. The PBL turbulence is usually enhanced in a warmer climate, which transports more energy from near the surface to the upper layer of the PBL and therefore reduces temperature lapse rate and downward longwave radiation, resulting in surface cooling. However, in the deep convection area, the enhanced convection tends to transport more energy from the PBL to the upper troposphere, resulting in a weakening of the PBL. The weakened PBL then increases the temperature lapse rate and downward longwave radiation, leading to a surface warming. This explains why the PBL process produces surface warming in the region where convective precipitation is increased most significantly in the tropics. Since the deep convection mainly occurs over tropical oceans, the average of surface warming attributable to the PBL process in convective area and cooling in other areas of the ocean results in a weak surface cooling over the ocean. In contrast, the PBL process produces much stronger surface cooling over the land. The dynamical advection and shortwave cloud feedback contribute −0.2 K each to the land–ocean warming contrast. The rest of the feedback processes have very small contributions. Of the factors believed important to the land–ocean warming contrast in previous studies (latent heat flux, longwave radiation from increased upper-tropospheric humidity and temperature, shortwave cloud feedback, and dynamic transport), none contribute positively to the land–ocean warming contrast in our case.

Sutton et al. (2007) show that the land–ocean warming contrast varies with latitude. To see if the role of each feedback to the land–ocean warming contrast varies with latitude, Fig. 8b shows the partial surface temperature changes attributable to the external forcing and each feedback averaged over land and ocean between 50°S and 50°N. Again the CFRAM analysis accurately reproduces the warming difference (0.7 K) between land and ocean in the mid- and low latitudes. Compared to Fig. 8a, the major difference is that shortwave feedback becomes the dominant contributor to the land–ocean contrast, whereas the contribution from the longwave feedback becomes insignificant. There is a warming of 0.34 K over land and a cooling of 0.12 K over ocean from the shortwave cloud feedback, making it the largest contributor (~0.46 K) to the land–ocean warming contrast in the mid- and low latitudes. This is obvious if we refer to Fig. 3f, which shows that shortwave cloud feedback over high-latitude land in both hemispheres is negative. Thus, excluding this cooling makes the land surface warming stand out. For longwave feedback, without the strong warming over high-latitude land, the land surface actually shows cooling (Fig. 3e). Excluding the large warming over the Southern Ocean near 60°S, the contribution of dynamical advection to land–ocean warming contrast changes from −0.2 K on the hemispheric scale to 0.3 K between 50°S and 50°N. The role of convective feedback is reduced from a 0.6-K contribution to a land–ocean contrast on the hemispheric scale to 0.1 K.

Both surface albedo feedback and the CO₂ forcing contribute to the land–ocean warming contrast by about 0.2 K. On the other hand, the PBL and ocean processes reduce the land–ocean warming contrast by about 0.4 and 0.2 K, respectively. In this mid- and low-latitude belt, shortwave cloud feedback and dynamic transport do seem to play a positive role in the land–ocean warming contrast, in agreement with previous studies. The role of latent heat flux seems to be insignificant in both the global and mid- to low-latitude land–ocean warming contrast.

6. Summary and conclusions

Climate feedbacks have been extensively investigated by the climate change community. Traditionally the climate feedback analysis mostly concentrated on the radiative feedbacks that directly affect the TOA radiation budget and did not provide a direct estimate of the partial surface temperature change caused by an individual feedback. Thus, it is difficult to quantify the contribution of climate feedback to surface warming. In this study, we apply a coupled atmosphere–surface climate feedback–response analysis method (CFRAM) to the slab ocean model version of NCAR CCSM3 to analyze the strength and spatial distribution of climate feedbacks and quantify their contributions to the global and regional surface temperature changes in response to a doubling of CO₂. The process-based decomposition of nonradiative feedback enables us to understand the roles of GCM physical and dynamic processes in climate change. The radiative feedback decomposition uses hourly model output, which greatly reduces the residual errors noted in previous radiative feedback analysis that use monthly or longer time-mean data (Taylor et al. 2011, 2013; Shell et al. 2008).

The global mean bias in the sum of partial radiative energy change derived with CFRAM analysis is about 1% of the radiative energy change computed directly from the CCSM3-SOM simulations (2.81 W m⁻²). The sum of partial surface temperature changes attributable to external forcing and climate feedbacks derived from the CFRAM analysis is also in very good agreement with the total temperature change directly obtained from the CCSM3-SOM, with a global mean bias of about 2%. The spatial pattern correlation between the two is about 0.94. This demonstrates that the CFRAM analysis with hourly model output is highly accurate for quantifying the role of climate feedbacks in the spatial distribution of global warming projected by the CCSM3-SOM.

Similar to the multimodel projection from the IPCC AR4 (Meehl et al. 2007), the global surface warming in response to a doubling of CO₂ projected by the CCSM3-SOM displays significant spatial variation. The warming in high latitudes is more than twice as strong as that in the tropics. The land warming is much stronger than that over oceans. For the global mean surface temperature change, the largest warming contributor is the water vapor feedback, followed by CO₂ forcing, albedo feedback, and longwave cloud feedback. The albedo feedback exhibits the largest spatial variation, followed by shortwave cloud feedback, ocean process, dynamical advection, and convection.

The pattern correlation, the centered RMS difference, and the ratio of variations between the total surface temperature change from the CCSM3-SOM and partial surface temperature change attributable to external forcing and each feedback process are examined to quantify the roles of climate feedbacks in the global warming pattern formation. The pattern correlation and normalized RMS difference show that the climate feedbacks dominate the global warming pattern formation. This result is consistent with the previous study of Boer and Yu (2003). Among the climate feedbacks, the longwave cloud feedback shows the largest pattern correlation and smallest RMS difference and therefore is the largest contributor to the formation of the model global warming pattern. The albedo feedback also plays an important role.

The analysis also shows that the albedo feedback is the largest contributor to the stronger warming in high latitudes than the tropics. The longwave cloud feedback further amplifies the high- to low-latitude warming contrast. In addition, convection, surface latent heat flux, and CO₂ forcing also have nonnegligible contributions. Both the land–ocean warming contrast and the contributions of climate feedbacks vary with latitude. In the mid- to low latitudes (50°S–50°N), the land–ocean warming difference is relatively small. The shortwave cloud feedback and dynamical advection are the two largest contributors. On the hemispheric scale, the land–ocean warming difference is larger, and the longwave cloud feedback is the largest contributor. The convection and albedo feedbacks also add to the stronger land warming.

This study is based on the NCAR CCSM3 using the slab ocean model. Therefore, ocean dynamical processes are excluded in current analysis. Danabasoglu and Gent (2009) examined the equilibrium climate sensitivities of NCAR CCSM3 in response to a doubling of CO₂ using both the slab ocean model version and full-depth ocean model version of CCSM3. They found that the slab ocean model version does give a good estimate of the equilibrium climate sensitivity of the full CCSM3. The spatial distributions of surface temperature change from the slab ocean and full-depth ocean simulations are very close over much of the globe, with relatively large differences in the high-latitude Southern Ocean and North Atlantic Ocean where the sea ice area is reduced more extensively in the full-depth ocean version than in the slab ocean version. Therefore, when the full-depth ocean model is used we expect no major change in our conclusions except that the ice–albedo feedback will be more effective over the high-latitude Southern Ocean and North Atlantic Ocean. Nevertheless, the current conclusions are based on one model (CCSM3-SOM). Different GCMs may exhibit different feedback characters. The comparison analysis with CFRAM will help us identify and understand the crucial physical processes that are responsible for the model spread of climate sensitivity. Improving the representation of these physical processes in GCMs should help reduce the uncertainty in climate change projection.