Cloud feedbacks on greenhouse warming in the ... - Wiley Online Library

PUBLICATIONS Journal of Advances in Modeling Earth Systems RESEARCH ARTICLE 10.1002/2014MS000355 Key Points: The superparameterized SP-CCSM4 has moderate positive cloud feedbacks It simulates less low cloud over land as climate warms SP-CCSM4 rapid cloud adjustments associated with abrupt CO2 increase are weak

Correspondence to: C. S. Bretherton, [email protected]

Citation: Bretherton, C. S., P. N. Blossey, and C. Stan (2014), Cloud feedbacks on greenhouse warming in the superparameterized climate model SPCCSM4, J. Adv. Model. Earth Syst., 6, 1185–1204, doi:10.1002/ 2014MS000355. Received 18 JUN 2014 Accepted 27 OCT 2014 Accepted article online 31 OCT 2014 Published online 13 DEC 2014

Cloud feedbacks on greenhouse warming in the superparameterized climate model SP-CCSM4 Christopher S. Bretherton1, Peter N. Blossey1, and Cristiana Stan2 1

Department of Atmospheric Sciences, University of Washington, Seattle, Washington, USA, 2Department of Atmospheric, Oceanic, and Earth Sciences, George Mason University, Fairfax, Virginia, USA

Abstract Cloud feedbacks on greenhouse warming are studied in a superparameterized version of the Community Climate System Model (SP-CCSM4) in an atmospheric component SP-CAM4 that explicitly simulates cumulus convection. A 150 year simulation in an abrupt quadrupling of CO2 is branched from a control run. It develops moderate positive global cloud feedback and an implied climate sensitivity of 2.8 K comparable to the conventionally parameterized CCSM4 and the median of other modern climate models. All of SP-CCSM4’s positive shortwave cloud feedback is due to a striking decrease in low cloud over land, which is much more pronounced than in most other climate models, including CCSM4. Four other cloud responses – decreased midlevel cloud, more Arctic water and ice cloud, a slight poleward shift of midlatitude storm track cloud, and an upward shift of high clouds – are also typical of conventional global climate models. SPCCSM4 does not simulate the large warming-induced decrease in Southern Ocean cloud found in CCSM4. Two companion uncoupled SP-CAM4 simulations, one with a uniform 4 K sea-surface temperature increase and one with quadrupled CO2 but fixed SST, suggest that SP-CCSM4’s global-scale cloud changes are primarily mediated by the warming, rather than by rapid adjustments to increased CO2. SP-CAM4 show spatial patterns of cloud response qualitatively similar to the previous-generation superparameterized SP-CAM3, but with systematically more positive low cloud feedbacks over low-latitude land and ocean.

1. Introduction Recent rounds of the Coupled Model Intercomparison Project, CMIP3 and CMIP5, have emphasized the continuing uncertainty in simulating cloud feedbacks on climate change with coupled atmosphere-ocean general circulation models (CGCMs) [Soden and Held, 2006; Soden and Vecchi, 2011; Zelinka et al., 2013; Vial et al., 2013]. This uncertainty is in part due to the need for complex parameterizations of subgrid variability of clouds, precipitation and radiation due to cloud-forming eddies, including both boundary-layer turbulence and cumulus convection.

This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made.

BRETHERTON ET AL.

Superparameterized (SP) climate models [Grabowski, 2001; Khairoutdinov and Randall, 2001] explicitly simulate the largest and most organized circulations within deep cumulus cloud systems using a cloud-resolving model (CRM) to represent the typical atmospheric state within each grid column of the global model. For computational efficiency, the CRM grid is still too coarse (often 4 km horizontal grid spacing and around 30 vertical levels through the atmospheric column) to simulate realistic boundary-layer eddies. Nevertheless, SP has been quite beneficial in simulating aspects of deep cumulus convection that have been challenging to parameterize, for instance the diurnal cycle of deep convection over land and the Madden-Julian Oscillation [Khairoutdinov et al., 2005]. A better simulation of deep convection affects the entire tropical general circulation, summertime continental clouds and precipitation, and parts of midlatitude cyclones. Hence, SP has the potential to improve the simulation of all types of cloud systems, although it cannot be expected to realistically simulate boundary-layer clouds without a careful treatment of much smaller-scale turbulent eddies, e.g., through subgrid turbulence closure parameterizations. Even given its limitations for boundary-layer clouds, which are very important for global cloud feedback [Bony and Dufresne, 2005; Soden and Vecchi, 2011], SP provides an interesting comparison for cloud feedbacks simulated by conventionally parameterized atmospheric general circulation models (AGCMs) and CGCMs. However, until recently its computational expense limited the length and types of simulations used for this purpose. Wyant et al. [2006] compared the cloud distributions in a pair of 3.5 year simulations with a

C 2014. The Authors. V

1185

Journal of Advances in Modeling Earth Systems

10.1002/2014MS000355

superparameterized adaptation of the Community Atmosphere Model, Version 3 (SP-CAM3), one with an additional uniform 2 K warming of sea-surface temperature (SST). They noted an increase in subtropical low cloud cover in the warm-SST case, which Wyant et al. [2009] attributed to enhanced boundary-layer radiative cooling. Wyant et al. [2012] analyzed cloud and circulation response of a slightly different version of SPCAM3 to a CO2 quadrupling with no SST warming, using a pair of 2 year simulations. They found a shift of cloud cover and mean ascent from ocean to land regions and a reduction in marine boundary-layer cloudtop heights, with little net global change in cloud cover. In this paper, we analyze cloud feedbacks in a superparameterized adaptation of the Community Climate System Model Version 4 (SP-CCSM4), documented in detail by Stan and Xu [2014]. Stan and Xu [2014] and Arnold et al. [2014] have also recently discussed the climate change response of SP-CCSM4 to greenhouse gas perturbations, and Arnold et al. [2014] in particular noted a large increase in Arctic cloud cover in a 43CO2 climate, which they found was related in part to increased wintertime cumulus convection over an Arctic Ocean with much decreased sea ice. This paper will present a more comprehensive global view of SP-CCSM4 cloud feedbacks and compare them with the SP-CAM3 studies. We follow the simulation protocol adopted in CMIP5 [Taylor et al., 2012] in support of Round 2 of the Cloud Feedbacks Model Intercomparison Project (CFMIP2) [Bony et al., 2011]. This includes (1) differencing a 150 year simulation of the coupled response to an abrupt CO2 quadrupling with a coupled control simulation, (2) differencing two 35 year SP-CAM4 simulations to obtain the uncoupled response to a CO2 quadrupling with no SST warming, and (3) similarly obtaining the uncoupled response to a uniform 4 K SST warming. The latter simulations are long enough to greatly reduce the cloud response uncertainty due to internal variability compared to the studies of Wyant et al. [2006] and Wyant et al. [2012]. We broadly compare the SPCCSM4 cloud responses with the CMIP5 CGCMs to test whether the use of SP appears to have a large impact on cloud feedbacks. Andrews et al. [2012] show that the top of atmosphere global-mean radiative response of a fixed-SST AGCM simulation to CO2 quadrupling is well correlated across CMIP5 models with the effective radiative forcing derived as the y intercept of a ‘‘Gregory’’ plot of net radiative response versus global surface air temperature difference DT [Gregory and Webb, 2008]. Ringer et al. [2014] show that the temperature-mediated global cloud feedback of CMIP5 CGCMs is highly correlated across models to the cloud feedback inferred from a specified 4 K SST warming applied to their AGCMs. Gettelman et al. [2012] showed that the equilibrium climate sensitivity of slab-ocean versions of CAM4 and CAM5 were well predicted by the response of specified-SST simulations with the same SST and CO2 changes as the corresponding slab-ocean models (in this case the geographical pattern of SST change, not just the global-mean SST change, was specified based on the slab-ocean simulations). With these results in mind, we compare the SP-CAM4 AGCM simulations to the cloud and precipitation responses of the coupled simulation. In Section 2, we give some brief details of the configuration of the simulations. In Section 3, we discuss the global and zonal-mean cloud and precipitation response in the coupled model to the abrupt CO2 quadrupling. In Section 4 we analyze the resulting cloud feedbacks following Zelinka et al. [2012a], and relate our results to possible cloud feedback mechanisms. Section 5 discusses the coupling of the geographical response of clouds and cloud feedbacks to circulation, precipitation and land surface changes. Section 6 compares our findings to the SP-CAM3 cloud feedback studies, and Section 7 gives conclusions.

2. Simulations Performed and Terminology SP-CCSM4 is documented by Stan and Xu [2014]. For the simulations presented here, its atmospheric component, the SP-CAM4, is run at 2 by 2.5 degree resolution with a finite-volume dynamical core, and the ocean component has 1 degree resolution. The SAM CRM used in each grid column of this model is oriented east-west, has 32 horizontal grid points with a grid spacing of 4 km, and has 30 vertical levels. Table 1 lists the simulations performed. Nomenclature follows CMIP5 conventions, but for the coupled simulations there are minor deviations from the CMIP5 protocol. The 150 year SP-CCSM4 control and abrupt CO2 quadrupling (abrupt4xCO2) simulations are both initialized from restart files for CCSM4 for 1 January 2006 from an 1850–2010 historical-forcings run; (hence these runs use 2006 CO2 5 367 ppm rather than a

BRETHERTON ET AL.


1186


10.1002/2014MS000355

Table 1. Simulations Performed Case Control abrupt4xCO2 amip amip4CO2 amip4K a

Model

Descriptiona

Length

SP-CCSM4 SP-CCSM4 SP-CAM4 SP-CAM4 SP-CAM4

Coupled control abrupt 43CO2 Specified-SST control Quadrupled CO2, specified-SST Uniform 4 K SST increase

150 years 150 years 35 years 35 years 35 years

All specified-SST simulations have the same sea-ice extent.

preindustrial value as a baseline and have a short spin-up period in which they adjust to the superparameterized atmosphere). The 35 year fixed-SST control simulation, amip, is forced with the seasonal cycle of the 1949–2001 mean SST and sea ice data set described by Hurrell et al. [2008], with the same CO2 concentration as the coupled control simulation. We classify it as AMIP-like because it uses observed SSTs, but unlike a true AMIP simulation it does not include their interannual variability. The specified-SST perturbation simulations are amip4CO2 (CO2 quadrupling only) and amip4K (4 K SST increase only). Following the CMIP5 protocol, the same sea-ice data set is used in these simulations as in the fixed-SST control. Unless otherwise indicated, coupled results are means over years 101–150, and fixed-SST results are means over years 2–35. We use the following terminology for ‘‘experiments,’’ i.e., differences between simulation pairs: 435abrupt43CO22Control 43P05amip4CO2 2amipðCO2 quadrupling with fixed SSTÞ P45amip4K2amipð4 KSST increase without CO2 changeÞ 4x

43LC543P010:85P4ðspecified-SST linear combination matching DT Þ 4x 4xP0 Here DT 54:5 K is the 100–150 year 43 global mean surface air temperature increase, and DT 50:5 K P4 and DT 54:6 K are the 2–35 year 4xP0 and P4 global mean surface air temperature increases. The coefficient 0.85 is chosen such that 4x

DT 5DT

4xP0

10:85DT

P4

(1)

to match the 43 temperature and CO2 increases, such that 4xLC fixed-SST responses in different fields can be meaningfully compared with 43 coupled responses. Figure 1 shows maps of the temperature response for these four cases. The 43P0 (DCO2-only) warming is due both to small surface air temperature increases over the oceans (a reduced sea-air temperature difference) and larger greenhouse-induced surface air temperature increases over the continents [e.g., Wyant et al., 2012]. The 4xLC pattern has the same global mean DT as the coupled 43 case by construction. It also has a broadly similar spatial pattern of warming as 43, including stronger warming over the continents. However, 4xLC has more muted warming in the polar regions because it uses fixed sea ice. The most striking difference of the coupled 43 SST response from the uniform-SST idealization of 4xLC is the SST cooling response in 43 in the subpolar northeast Atlantic ocean, which is presumably driven by feedbacks with oceanic deep convection. The spatial patterns of the 43 SST increase can be expected to also manifest in its cloud and precipitation-related responses.

3. Global and Zonal-Mean Cloud and Precipitation Responses in SP-CCSM4 Simulation Figure 2 shows 150 year time series of selected global annual means from the SP-CCSM4 control and abrupt CO2 quadrupling simulations. Figure 2a shows surface air temperature, and Figure 2b shows net top-ofatmosphere radiation R. Since both runs are identically initialized from a historical run of the conventionally parameterized CCSM4, each undergoes a 2 year adjustment phase, during which the control run undergoes

BRETHERTON ET AL.


1187


10.1002/2014MS000355

ΔT [K]

Latitude [deg N]

Latitude [deg N]

−10

90 50 30

−5

0

5

4x: global mean = 4.46 K

a)

90 50 30

0

0

−30 −50 −90

−30 −50 −90

90 50 30

c)

4xP0: global mean = 0.49 K 90 50 30

0 −30 −50 −90 0

10

b)

4xLC: global mean = 4.46 K

d)

P4: global mean = 4.64 K

0

60

120 180 240 Longitude [deg E]

300

−30 −50 −90 360 0

60


300

360

Figure 1. Surface air temperature increase relative to control simulations in 100–150 year mean of coupled (43) and in 2–35 year mean of specified SST experiments (4xP0: fixed SST, 4xCO2 only; P4: 4 K SST increase only). P4LC is a linear combination of the 4xP0 and P4 responses with the same global surface air temperature and CO2 increase as 43.

Figure 2. Global mean time series from SP-CCSM4 Control and abrupt4xCO2 simulations, whose difference gives the coupled ‘‘43’’ response to abrupt quadrupling of CO2. (a) Surface air temperature T , (b) net TOA incoming radiation R, (c) fraction of surface area covered by sea-ice, and (d) fractional total cloud cover.

BRETHERTON ET AL.


1188


a temperature drop and an increase in R. This adjustment should only have a minor effect on their difference (the 43 response). Most of this difference develops in the first 5 years, although it is still increasing after 150 years. There is also a rapid adjustment of sea-ice cover (Figure 2c) and total cloud cover (Figure 2d) within the first year. The sea-ice is reduced more than 60% in abrupt43CO2 after 100 years.

10 (LW+SW) full sky LW clear sky SW clear sky LW CRE SW CRE

−2 ΔR [W m ]

8 6

10.1002/2014MS000355

4 2 0 1

2

3 ΔT [K]

4

4xLC

4xP0

Figure 3 shows a Gregory plot of the 43 difference of annual global mean TOA radiation components. All components show reasonably linear behavior with DT . A linear Figure 3. Gregory plot of 43 global-annual mean TOA radiative flux changes. Symbols show corresponding results from specified-SST experifit of the TOA net incoming radiation differments 4xP0 (magenta) and 4xLC (green). ence DR versus DT implies an equilibrium D T at quadrupled CO2 of 5.6 K, i.e., an equilibrium climate sensitivity to CO2 doubling of 2.8 K, and a feedback parameter k50:80 K W– 1 m2. The positive slope dSWclr =d T reflects positive clear-sky snow/ice albedo feedback. The slopes dLWCRE=d T and dSWCRE=dT are both close to zero, but these should not be interpreted as cloud feedbacks because of water vapor and surface albedo masking effects [Soden et al., 2004]. In section 4, we will show the global shortwave and longwave cloud feedbacks are both slightly positive, 0.3 and 0.2 W m– 2 K– 1, respectively. The y intercept of the SWCRE fit line is slightly positive, consistent with the slight reduction of total cloud cover during the initial rapid adjustment to the CO2 increase. The y intercept of the LWCRE fit line is mainly due to CO2 masking. −2 0

5

6

The larger symbols in Figure 3 show corresponding results from the specified-SST experiments 4xP0 (DCO2only, magenta outline) and 4xLC (additional uniform SST increase, green outline). For the cloud radiative effects and clear-sky long wave radiation, these are consistent with the 43 coupled experiment. The clear-sky shortwave radiation and net radiation are less sensitive to DT than in 43 because the sea-ice extent is fixed. 3.1. Zonal-Mean Cloud and Precipitation Responses Figure 4 shows selected zonal mean 43 differences averaged over years 100–150 (the period shown by the red bar in Figure 2a); plots (b) and (c) instead show the Control and abrupt4xCO2 precipitation and total cloud cover, so that their differences can be compared with the control climatology. Figure 4a shows that the surface temperature difference is nearly uniformly 4 K equatorward of 40 latitude in both hemisphere, with more than two-fold polar amplification in both hemispheres after 100 years. In this panel, the corresponding temperature difference profiles from the specified-SST cases introduced in Section 2. are also plotted for reference. The P4 case shows only weak polar amplification due to its fixed sea-ice extent and uniform SST increase. The 4xP0 case shows the largest response at Northern Hemisphere high latitudes with a large fractional land cover, where CO2 can radiatively warm the surface efficiently. Returning to the coupled case, precipitation is enhanced in the equatorial belt and the high latitudes (Figure 4b), and total cloud cover shifts slightly poleward in the midlatitudes of both hemispheres (Figure 4c). These behaviors are both qualitatively and quantitatively within the range of CMIP5 CGCMs [Collins et al., 2013]. Figures 4d–4f shows the high (< 400 hPa), midlevel (400–700 hPa) and low (> 700 hPa) cloud responses. As with precipitation, there is a large increase in high cloud in high latitudes, with a slight decrease in the subtropics and a slight increase near the equator. Low cloud also increases in polar ocean regions at the expense of the midlatitudes, likely in connection with enhanced surface-driven convection over open water, as suggested by Arnold et al. [2014]. Midlevel cloud decreases at almost all latitudes, except right over the polar caps. 3.2. Latitude-Binned Gregory Plots of Cloud Types Figure 5 shows Gregory plots of the responses of high, middle and low cloud, as well as ice water path (IWP) and liquid water path (LWP) to the abrupt CO2 increase. Tropical (60 N/S) annual mean responses are plotted for each simulated year. As in Figure 3, the magenta and green symbols show corresponding 4xP0 and 4xLC specified-SST results. All of the plots show an approximately linear behavior of cloud properties with surface temperature difference DT with a y intercept that is close to zero. Following Gregory and Webb [2008], this suggests that these large-scale measures of SP-CCSM4’s cloud response are mainly temperature-mediated; rapid adjustment plays only a minor role. Except for low cloud and LWP, the 4xP0 and 4xLC results are also approximately on the same fit lines, supporting this interpretation and suggesting that regional SST differences are not crucial to these responses. The consistency between the specified-SST and coupled cloud responses on these very large space scales is consistent with the findings of Andrews et al. [2012] and Ringer et al. [2014]. A Gregory plot of precipitation (Figure 5d), on the other hand, shows negative intercepts in the tropics and midlatitudes, corresponding to the expected decrease in global precipitation that accompanies a abrupt CO2 increase [e.g., Bala et al., 2010]. It is interesting that this is not accompanied by a larger corresponding rapid adjustment in cloud properties. Figures 5a–5d corroborate the zonal-mean plots of cloud and precipitation response in Figure 4, showing increases in polar high cloud and precipitation, decreases in midlevel cloud in the tropics and especially midlatitudes, and low cloud increases in polar regions and decreases in the tropics. The ice water path (Figure 5e) increases in polar regions in association with the high cloud increase, while the liquid water path (Figure 5f) increases in polar regions in association with the low cloud increase. Shifts in the freezing level must also contribute to the LWP/IWP partitioning, but do not dominate the impacts of cloud fraction changes. The 4xLC LWP in Figure 5f is far above the coupled result for the midlatitude and polar regions; we do not have a good explanation for this.

BRETHERTON ET AL.


1190

Journal of Advances in Modeling Earth Systems b)

0.04

0.02 0 −0.02

0.5

0.02 0 −0.02

d)

15

4xLC

−0.06

0.02 0 −0.02 −0.04

4xP0

4xP0

−0.06

0.04

−0.04 4xLC

−0.04

0.06

−0.06

e)

40

4xLC

0.04

Tropical Mid−latitude High−latitude

4xP0

0.06

c)

ΔCLDLOW

0.06

ΔCLDMED

ΔCLDHGH

a)

10.1002/2014MS000355

f)

ΔLWP [g m−2]

5

−2

ΔIWP [g m ]

0

0 −5

20

0

2 ΔT [K]

4

2 ΔT [K]

4

4xLC

−40 0

4xP0

−15 0

4xLC

−10

4xP0

−0.5 0

4xLC

−20 4xP0

ΔPREC [mm d−1]

10

2 ΔT [K]

4

Figure 5. Gregory plot of annual mean tropical (30 S–30 N), midlatitude (30–60 S and N) and high-latitude (60–90 S and N) changes in selected cloud-related quantities. These regions are indicated on Figure 4a to emphasize their different areal coverage.

4. Diagnosis of Cloud Feedbacks in SP-CCSM We use the method of Zelinka et al. [2012a] to diagnose cloud feedbacks in SP-CCSM4. This method uses International Satellite Cloud Climatology Program (ISCCP) simulator diagnostics in combination with a radiative kernel to diagnose the TOA longwave and shortwave radiative feedback of cloud changes partitioned by location, cloud-top height and optical depth. Dr. Mark Zelinka kindly provided software to perform this analysis. We do not break the feedback into fast-adjustment and temperature-mediated components, since for SP-CCMS4 we have already shown the latter component dominates the global-scale cloud response. Also, as the ISCCP simulator is not run during polar night, the annual average of longwave cloud feedback at these latitudes is computed only from months when the sun is above the horizon. [Zelinka et al., 2012a, Appendix B, section d] suggest that the errors induced by this approximation are not significant. Figure 6 shows the geographical patterns of the net, longwave and shortwave cloud feedbacks. There is a positive global longwave cloud feedback of 0.30 W m–2 K–1 and a positive shortwave cloud feedback of 0.19 W m–2 K–1. These values are quite similar to the CMIP5 multimodel mean cloud feedbacks (without rapid adjustments separately broken out) in Zelinka et al. [2013, Figure 3]. For reference, Figure 7 shows the easier-to-compute longwave and shortwave cloud radiative effect (here normalized by DT ), as well as the change in 500 hPa vertical pressure velocity x500, which we use as a measure of circulation changes. As expected, the CRE changes and cloud feedbacks are highly spatially correlated for both longwave and shortwave, with a spatially variable offset of their difference that reflects cloud masking effects. Globally, the longwave and shortwave feedbacks are 0.65 and 0.16 W m– 2 K– 1 larger than the respective global CREs. The longwave feedback is spatially anticorrelated with the x500 change, since high-topped clouds are favored in regions of mean ascent. It is rather insensitive to land-ocean boundaries. The shortwave feedback, on the other hand, is strongly positive over all low-latitude land masses except Australia (where there is very little low cloud to start with); we will relate this to low cloud reductions over land in SP-CCSM4 in a

BRETHERTON ET AL.


1191

Journal of Advances in Modeling Earth Systems −1

K

5

0

0

−30 −50 −90 90 50 30

−5 −2

b)

4x: global mean = 0.30 W m

−1

K

5

0

0

−30 −50 −90 90 50 30

−5 c)

4x: global mean = 0.19 W m−2 K−1 5

0 −30 −50 −90 0

0

−5 60


Net Cld Fdbk [W m−2 K−1]


300

360

LW Cld Fdbk [W m−2 K−1]

a)

SW Cld Fdbk [W m−2 K−1]

Latitude [deg N]

Latitude [deg N]

Latitude [deg N]

−2

90 50 30

Figure 6. Maps of SP-CCSM4 net, longwave and shortwave cloud feedback deduced from 43 changes in ISCCP simulator cloud fraction.

10.1002/2014MS000355

warmer climate. It also tends to spatially anticorrelate with the longwave feedback due to changes in middle and hightopped cloud cover. Figure 8 shows the global, landonly, and ocean-only annual average 43 cloud cover changes binned by cloud top pressure and optical depth. The large global increase of high-top cloud in the uppermost cloud top pressure bin is due partly to the rise of the tropopause in a warmer climate [Hansen et al., 1984; Hartmann and Larson, 2002], and partly to the previously noted increase in high cloud at high latitudes. Consistent with our earlier findings, there is a decrease in midtop cloud of all optical depths. Low cloud cover of all optical depths also decreases over land, but low cloud of intermediate optical depths increases over the oceans and in the global mean. For midtop and high-top cloud, the 43 cloud changes are fairly similar between land and ocean, though as previously noted they do depend on latitude.

The global, land-only, and ocean-only longwave, shortwave and net cloud feedbacks due to these cloud changes are also given in Figure 8. Since the main difference in ocean versus land cloud changes is the large decrease in low cloud over land, the longwave cloud feedback is similar over ocean and land, while the shortwave cloud feedback is strongly positive over land (0.79 W m– 2 K– 1) but weakly negative over ocean. Overall, SP-CCSM4’s net cloud feedback is 0.99 W m– 2 K– 1 over land and 0.28 W m– 2 K– 1 over ocean. Thus, the 30% of Earth’s surface that is land contributes 60% of SP-CCSM4’s positive global net cloud feedback. We also partitioned the global net feedback into contributions from altitude, cloud fraction and optical depth changes, following Zelinka et al. [2012b]; of these the altitude increase of high clouds explains 0.30 W m– 2 K– 1 of the global net cloud feedback, all in the longwave, while the cloud fraction and optical depth changes and a residual term all make smaller positive contributions to the feedback. 4.1. Cloud Feedbacks in SP-CCSM4 Versus CCMS4 and Other Climate Models The moist physical parameterizations in most conventionally parameterized AGCMs are arguably more similar to each other than they are to superparameterization. Cloud feedbacks and equilibrium climate sensitivity are strongly tied to the atmospheric representation in a climate model, as can be seen by the strong correlation between cloud feedbacks inferred from fully coupled climate models and those inferred from running an AGCM with idealized SST perturbations. From this perspective, it is appropriate to compare SPCCSM4 to the whole class of conventionally parameterized GCMs, not just CCSM4. Figure 6 can be compared with CMIP3 multimodel mean cloud feedbacks deduced using a different technique by Soden and Vecchi [2011, Figure 1]. There is a reassuring similarity in the broad geographical pattern of feedbacks, e.g., the positive longwave feedbacks in the wetter parts of the tropics, as well as in high

BRETHERTON ET AL.


1192

4x: global mean = −0.00 Pa s−1 0.04

0

−0.04

−1

K

0

−5 −2

c) 4x: global mean = 0.03 W m

−1

K

5

0

0

−30 −50 −90 0

−5 60


300

360

K−1]

−30 −50 −90

ΔLW CRE/ΔT [W m

−2

5

−1

4x: global mean = −0.35 W m

K ]

−2

b)

Δω

−0.02

0

90 50 30

500

0

−30 −50 −90 90 50 30

−1

0.02

[Pa s ]

a)

−2

90 50 30

ΔSW CRE/ΔT [W m

Latitude [deg N]

Latitude [deg N]

Latitude [deg N]


Figure 7. Maps of SP-CCSM4 43 responses of x500, longwave and shortwave cloud radiative effect.

10.1002/2014MS000355

latitudes, as well as the large positive shortwave feedbacks over the low-latitude continents outside Australia, though there are many regional differences. However, examination of Soden and Vecchi [2011, Figure 1] suggests that the SP-CCSM4 net and shortwave cloud feedbacks averaged over ocean areas are somewhat smaller over ocean compared to the CMIP3 composite, and that in the CMIP3 composites there is no clear difference between the land-mean versus ocean-mean cloud feedbacks. An exception to the argument that motivated a broad comparison of cloud feedbacks in SP-CCSM4 with CMIP models is the possible role of the land model, which is the same for superparameterized and conventionally parameterized versions of the same climate model. That is, could SP-CCSM4’s strong positive shortwave cloud feedback over land be due in part to the land model? If so, the conventional CCSM4 might also show a similar response.

Hence, we also compare cloud feedbacks in SP-CCSM4 versus the conventionally parameterized CCSM4. Zelinka et al. [2013] calculated cloud feedbacks for fully coupled versions of CCSM4 and other CMIP5 models

Pressure

a) ΔGlobal cloud, avg = −0.7% Net/LW/SW Cloud Feedback −2 −1 = 0.49/0.30/0.19 W m K

b) ΔLand cloud, avg = −2.2% Net/LW/SW Cloud Feedback −2 −1 = 0.99/0.20/0.79 W m K

c) ΔOcean cloud, avg = −0.1% Net/LW/SW Cloud Feedback −2 −1 = 0.28/0.34/−0.05 W m K

180

180

180

310

310

310

440

440

440

560

560

560

680

680

680

800

800

800

0 0.3 1.3 3.6 9.4 23 60 380 Optical Depth

−0.01

0 0.3 1.3 3.6 9.4 23 60 380 Optical Depth

0 0.3 1.3 3.6 9.4 23 60 380 Optical Depth

−0.005 0 0.005 4x: ΔCloud Fraction from ISCCP Simulator

0.01

Figure 8. Global, land-only, and ocean-only annual average 43 cloud cover changes binned into ISCCP cloud top pressure and optical depth categories. Plot titles include the integrated cloud cover change and cloud feedbacks.

BRETHERTON ET AL.


1193


10.1002/2014MS000355

Latitude [deg N]

Net Cld Fdbk [W m−2 K−1]

using the same approach that we used for SP-CCSM4, but showed only 5 multimodel mean results. Bitz et al. [2012] calculated CCSM4 cloud feedbacks on the equilibrium climate 0 0 change produced by CO2 doubling −30 over a slab ocean, based on applying −50 −5 masking corrections derived using a −90 0 60 120 180 240 300 360 kernel method to the maps of longLongitude [deg E] wave and shortwave cloud radiative forcing change; Zelinka et al. [2012a] Figure 9. Map of net cloud feedback for the conventionally parameterized CCSM4 based on Bitz et al. [2012, Figure 8]. showed that this method gives similar cloud feedbacks to theirs. Figure 9 shows a map of the CCSM4 net cloud feedback deduced by Bitz et al. [2012]. Their Figure 8 shows the longwave and shortwave contributions to this feedback. CCSM4: global mean = 0.48 W m−2 K−1

90 50 30

Figure 9 shows qualitatively similar geographical patterns of net cloud feedback over land to SP-CCSM4 (Figure 6c), suggesting that interaction with the land model may indeed be regionally important. However, CCSM4’s shortwave and net cloud feedbacks averaged over land are 0.35 and 0.52 W m– 2, approximately half as large as in SP-CCSM4. Over ocean, these feedbacks are 0.22 and 0.46 W m– 2, both larger than for SPCCSM4. Thus, CCSM4 does not show the large enhancement of net cloud feedback over land compared to over ocean seen in SP-CCSM4. CCSM4’s spatial patterns of cloud feedback over the ocean also have some similarities to SP-CCSM4, but also notable differences. In particular, SP-CCSM4 does not reproduce the prominent band of positive cloud feedback over the Southern Ocean found in CCSM4. CCSM4’s global-mean long wave, shortwave and net cloud feedbacks are 0.23, 0.25 and 0.48 W m22 K21, fairly similar to SP-CCSM4. Bitz et al. [2012] also find that CCSM4 has a climate sensitivity of 3.2 K, which is slightly larger than for SP-CCSM4, but that this estimate is somewhat dependent on the methodology used to compute it, as well as the horizontal resolution. It remains an interesting question how sensitive a superparameterized climate model is to the details of its host model, assuming that the embedded CRMs are not changed. We will revisit this issue in Sec. 6., where SP results using the CAM4 AGCM are compared with previous studies that used SP-CAM3. 4.2. Relation of Global Cloud Feedbacks to Cumulus Mixing Efficiency Metrics Sherwood et al. [2014] introduced two metrics, S and D, of lower to midtropospheric cumulus mixing efficiency in the current climate that empirically are correlated with climate sensitivity and hence global cloud feedbacks in the CMIP3/CMIP5 models. These metrics can be construed as measures of how well GCMsimulated shallow cumulus convection (reaching no higher than the freezing level) can keep up with deep cumulus convection in the regions in which deep convection is most prevalent. They might also be relevant to how aggressively shallow cumulus convection vertically mixes in the subsiding branch of the Hadley circulation, which supports most off the boundary layer cloud and is responsible for much of the intermodel spread in cloud feedbacks. SP-CCSM4 does not use a cumulus parameterization, relying instead on an explicit, albeit under-resolved simulation of cumulus convection. Hence, it is of interest (a) how well SPCCSM4 agrees with the observed value of these metrics, and (b) whether its climate sensitivity lies on the CMIP-based regression lines between climate sensitivity and each of S and D. Following the methods section of Sherwood et al. [2014], at each month and tropical (30 N–30 S) ocean grid point, we computed s 5 0:5ðdR2dT=9Þ, where d denotes a difference between monthly mean values at 700 hPa and 850 hPa, R is relative humidity, and T is temperature in K. This is a measure of lowtropospheric temperature and moisture stratification, which Sherwood et al. [2014] argue is regulated by convective mixing efficiency. For each month, we select the tropical ocean grid points in regions of 500 hPa mean ascent 2x500 > 0, spatially average s over the quartile of these grid points with strongest mean ascent (which mainly lie in the west Pacific warm pool). We then average the resulting quantity across months to get a statistic S, which we find is 0.40 for SP-CCSM4. Based on Sherwood et al. [2014, Figure 5a], this is within the range 0.35–0.5 suggested by radiosonde observations and reanalyses. SP-CCSM4’s climate sensitivity of 2.8 K is also within the fairly broad range (2–4.5 K) of climate sensitivities of GCMs whose S is

BRETHERTON ET AL.


1194


10.1002/2014MS000355

within the range 0.35–0.45. For this metric, SP-CCSM4 is consistent with observations and behaves similarly to conventional GCMs. The second metric D is a ratio of shallow versus deep overturning in regions of mean low-level ascent over the ITCZ sectors (160 W–30 E) of the tropical oceans, where ‘‘bottom-heavy’’ mean vertical motion profiles are common. In the discussion below, ‘‘800 hPa’’ will be used as a proxy for an average of 850 and 700 hPa data, and ‘‘500 hPa’’ as a proxy for an average of 400, 500, and 600 hPa data, to simplify the description of Sherwood et al. [2014]’s ratio. For each month, those grid columns within this region with 800 hPa mean ascent and net lateral outflow of mass from the lower troposphere (500–800 hPa) are selected. The shallow overturning is the area integral of this lateral mass flux. Then, those grid points within the ITCZ region with 500 hPa monthly mean ascent are selected, and the deep overturning is defined as the area integral of this upward mass flux through the 500 hPa level. The monthly ratio of shallow to deep overturning is averaged across the annual cycle to obtain D. For SP-CCSM4, following this prescription we obtain D 5 0.32, which is at the upper end of the CMIP5 model range and somewhat less than two reanalysis-based estimates of 0.38 and 0.45. We note that most of the contribution to D comes from ‘‘marginal’’ precipitation regions outside the core of the ITCZ, and is sensitive to the extent of these regions. We also computed D using the specified-SST control run of SP-CAM4, and found a significantly smaller value D 5 0.21 that is below the median of CMIP5 models shown in Sherwood et al. [2014, Figure 5b]. To the extent that D is an indicator of simulated convective mixing, it is not obvious why it should be substantially different in the coupled and specified-SST simulations, raising concerns about the robustness of D for this purpose. The climate sensitivity of SP-CCSM4 is on the low end of the CMIP5 models with S 1 D values similar to the 0.72 of SP-CCSM4, but it would be comparable to CMIP5 models with if we used D 5 0.21 to calculate S1D50:61 instead.

4.3. Subtropical Cloud Feedbacks Versus Predictions With CGILS Column Tests The CGILS intercomparison [Zhang et al., 2013; Blossey et al., 2013] aimed to understand mechanisms of subtropical low cloud feedback using single column experiments. Realistic steady ‘‘control’’ (CTL) large-scale advective forcings and boundary conditions corresponding to mean summertime conditions in three northeast Pacific locations, S6, S11 and S12, were applied to obtain simulated equilibrium cloud-topped boundary layer for those locations, which in reality support shallow cumulus cloud regimes, cumulus rising into stratocumulus, and well-mixed stratocumulus, respectively. Then the column model was brought into equilibrium with a perturbed forcing (P2S), including a 2 K SST and column temperature increase and a 10% decrease in subsidence, to mimic effects of climate change. The hope was that differences (DP2S) of low cloud between these two column experiments would approximately correspond to those simulated by the full GCMs. This did not prove to be the case for conventional climate models [Zhang et al., 2013], because the equilibrium responses of single column models do not respond smoothly to small perturbations in the forcings. In fact, the single-column model responses were dominated by grid-locking effects in which the cloud response of a single column model sensitively depended on whether or not the capping inversion shifted up or down a grid level due to the forcing perturbation. In the superparameterized SP-CAM, each grid column of the AGCM is simulated by an entire CRM, so one might imagine the mean state of the corresponding atmospheric column model would have more freedom to evolve smoothly in response to a small forcing perturbation. Thus, we used a 2D version of the SAM CRM with a grid, time step, and physical parameterizations configured as closely as possible to that in SP-CAM4, to perform the CGILS tests. We call this configuration SP-SAM. Table 2 summarizes our findings. For brevity, SWCRE is the sole metric used for cloud response; following Blossey et al. [2013] this is calculated for the CGILS cases from simulated 8–10 day means. We find that both the mean state and perturbation response of SP-SAM are significantly different from those of the fullresolution three-dimensional version of SAM presented by Blossey et al. [2013]. We also compare the SPSAM CGILS results with the summertime (June–July–August or JJA) climatology of the SP-CAM4 amip and amip4K runs at the same three locations. This makes a better comparison than the coupled results, for two reasons. First, the specified-SST control run enforces the same observed SST climatology as used in the CGILS boundary conditions, Second, the P4 experiment, like the CGILS P2S perturbation, only includes a warming perturbation and not a CO2 perturbation, while allowing mean subsidence to respond to the SST

BRETHERTON ET AL.


1195


10.1002/2014MS000355

Table 2. CGILS Control and P2 SWCRE Responses (W m22) for SP-SAM Versus High-Resolution 3-D SAM, SP-CAM4, and Satellite Observations S6 Modela SP-SAM Hi-res 3-D SAMa SP-CAM4 (JJA)b Observed (JJA) c

S11

S12

CTL

DP2S

CTL

DP2S

CTL

DP2S

2100 230 293 225

8 3 21.4

2210 2145 2107 283

114 0 1.7

2219 2133 2154 284

27 21 5.2

a

From Blossey et al. [2013]. From sampling of the SP-CAM4 P4 experiment at the CGILS locations. DP2S perturbation is estimated as half the P4 perturbation. c Based on CERES_EBAF_Ed2.8 [Loeb et al., 2009] climatology from 2000 to 2013 (current climate only). b

warming. Because the P4 SST perturbation is twice as strong as that used for CGILS, in Table 2 we halve the P4 cloud-radiative response for comparison with the CGILS DP2S response. We find that SP-SAM has much more SWCRE in the control case, and much more positive cloud feedbacks, than the full SP-CAM4 at all three locations. These results bear further study. Inspection of the time series and vertical structure of the control and perturbed cloud responses (not shown) suggests that grid locking remains an important issue for SP-SAM, like conventional single-column models, because the vertical grid spacing is coarse and the strong simulated inversions remain flat across the CRM grid. For instance, the very large SP-SAM warming response in the S11 case is due to the inversion moving up a layer and causing partial cloud dissipation, while in the S12 case, the inversion also moves up a layer and this induces decoupling of the boundary layer, also contributing to stratocumulus thinning. Adding a carefully formulated transient forcing component could reduce grid-locking artifacts [Brient and Bony, 2013]. For now, we conclude that (a) as currently configured, the CGILS cases are not a good guide to subtropical cloud feedback for SP-CCSM4, and (b) that a much higher vertical and horizontal grid resolution (or at least a more sophisticated sub grid cloud and turbulence parameterization) in the embedded CRMs of SP-CCSM4 could substantially change its cloud feedbacks. Lastly, we note that the thermodynamic and radiative mechanisms of positive marine subtropical low cloud feedback deduced from the CGILS LES results [Bretherton et al., 2013], respectively due to overall warming of the lower atmosphere and to the increased longwave emissivity of the overlying free troposphere, are not apparent in the SP-CCSM4 results. Instead, SP-CCSM4 shows very little marine subtropical low cloud change in the face of increased greenhouse gasses and a warmer atmospheric column. A key issue for future study is to resolve whether the weak subtropical low cloud changes in SP-CCSM4 are due to inadequate CRM resolution or to other confounding consequences of greenhouse warming.

5. Changes in the Geographical Distribution of Cloud and Precipitation As seen above, the regional response of clouds to a changing climate is tightly coupled to circulation changes [Bony et al., 2004], as well as differential changes between land and ocean. These relationships affect zonal means, but the full regional response, which is relevant to climate impacts, is best visualized with selected geographical maps, presented in this section. 5.1. Midlevel Cloud Response Figure 10a shows that in the control run, midlevel cloud is much more prevalent at latitudes poleward of 45 in both hemispheres. Figure 10b shows the coupled response of midlevel cloud to the abrupt CO2 increase. As suggested by the zonal-mean behavior, midlevel cloud decreases nearly everywhere except the polar caps and the central equatorial Pacific, with typical relative reductions of 10–20%. The spatial modulations in midlevel cloud change correlate well with 2x500 (Figure 7a). There is an offset, reflected in regions such as the North Pacific storm track with increased upward motion but midlevel cloud decreases, suggesting that the ratio of midlevel cloud to upward mass flux in storm systems may tend to decrease in SP-CCSM4 in a warmer climate. Zelinka et al. [2013] showed that in contrast to SP-CCSM4, the CMIP5 multimodel mean exhibits a rapid CO2-forced reduction of midlevel cloud, but only a weak overall further temperature-mediated decrease.

BRETHERTON ET AL.


1196

a)

Control: global mean = 0.19 1 0.75

0

0.5

−30 −50 −90 90 50 30

0.25 0 4x: global mean = −0.02

b)

0.2 0.1

0 −30 −50 −90 0

CLDMED

90 50 30

0

CLDMED

Latitude [deg N]

Latitude [deg N]


−0.1 60


300

360

−0.2

10.1002/2014MS000355

5.2. Low Cloud Changes Figure 11a shows the spatial distribution of low cloud in the control simulation. The persistent low cloud over the midlatitude oceans is well captured, but the subtropical stratocumulus regions less well so [Stan and Xu, 2014]. Figure 11b shows the 43 low cloud response. There is a low cloud decrease over the midlatitude oceans and there are increases in the central subtropical oceans, corresponding to a westward shift of the subtropical stratocumulus regions that could be interpreted as a downstream delay of the subtropical stratocumulus to cumulus transitions.

The most striking feature is the low cloud decrease over most land areas. Figure 11c shows the 43 near-surface relative humidity response. There is a strong spatial correlation between near-surface relative humidity and low cloud cover changes over both land and ocean. As noted by Joshi et al. [2008] and others, the nearsurface relative humidity decreases sharply over land as the climate warms because the land surface becomes more moisture-limited. This inhibits low cloud development since it requires a deeper layer of boundary layer turbulent mixing before condensation will occur. Figure 10. (a) Midlevel cloud in Control, and (b) 43 changes in midlevel cloud.

5.3. Coupled Versus Specified-SST Cloud-Radiative and Low Cloud Responses In this and the following subsection, we will compare the spatial structure of the coupled (43) response of selected quantities to that due to the specified-SST perturbations, 4xP0 (DCO2-only) and P4 (DSST-only), and their temperature-matched linear combination 4xLC. We have already learned from the Gregory plots (Figure 5) that for averages of most cloud quantities over the entire tropical, midlatitude, or polar latitude belts, the DCO2-only response is fairly small, and the 4xLC combination matches the coupled run fairly well. On these nearly global scales, then, the SP-CCSM4 coupled cloud response is dominated by temperaturemediated changes which are not sensitive to the subtle spatial gradients in SST increase (Figure 1a). We now test whether the same is true for cloud-related responses at smaller regional scales. Since tropical circulations are sensitive to SST gradients, and SST does not increase in the 43 run exactly uniformly, like it does in the 4xLC analogue, some regional differences between the 43 and 4xLC patterns of cloud-related response are to be expected. In addition, the climatologies of the coupled and specified-SST control simulations are somewhat different. For instance, the coupled control has a much stronger double-ITCZ rainfall and circulation bias than the specified-SST control [Stan and Xu, 2014]. But are these large enough to make the 43 versus 4xLC comparison meaningless? We start by considering the cloud radiative response, which we define as the difference in TOA radiative flux (defined positive downward) between a pair of simulations due only to the changes in the threedimensional cloud distribution, computed from the ISCCP-binned cloud histograms following Zelinka et al. [2012a]. For the 43 (coupled) case, this is the cloud feedback multiplied by the temperature change (recall that in this paper, we are including fast cloud adjustment within the cloud feedback). For all simulation pairs, it differs from the CRE response due to cloud masking effects, and gives a truer measure of the importance of cloud changes alone to the TOA radiation budget. Figure 12 compares the spatial structure of the 43 (coupled) net (longwave plus shortwave) cloud radiative response with the DCO2-only, P4 and combined (4xLC) responses. The DCO2-only response is fairly weak, consistent with the Gregory plots, but with positive net radiative response over many land areas leading to a positive global-mean DCO2-only response of 0.8 W m– 2. The temperature-mediated cloud radiative response is stronger, with generally positive values at lower latitudes and slightly negative values at higher

BRETHERTON ET AL.


1197

Control: global mean = 0.35

a)

1 0.75

0

0.5

−30 −50 −90

CLDLOW

90 50 30

0.25

0.2 0.1

0

0

−30 −50 −90

CLDLOW

90 50 30

0

4x: global mean = −0.00

b)

−0.1 −0.2 4x: global mean = −0.97

c)

90 50 30

20 10

0

0

−30 −50 −90 0

−10 60


300

360

Surface RH, %

Latitude [deg N]

Latitude [deg N]

Latitude [deg N]


−20

Figure 11. (a) Low cloud in Control, and 43 changes in (b) low cloud and (c) nearsurface relative-humidity.

10.1002/2014MS000355

latitudes, especially over oceans. The 4xLC net cloud radiative response has a fairly similar zonal-mean structure to the coupled 43 case, and both are positive over nearly all land regions. 43 and 4xLC also both tend to show negative cloud radiative response in the subtropical stratocumulus to cumulus transition regions. The negative equatorial cloud radiative response in the coupled simulation is not apparent in the 4xLC analogue. This may result from more upward motion (Sec. 5..4) and optically thick cloud due to a slight equatorial Pacific enhancement of the SST warming in 43 that is robustly seen in CMIP5 models and has been attributed to weakening of the Walker circulation [Chadwick et al., 2013]. Figure 13 shows the analogous plots for low cloud response. All panels look strikingly similar to the net cloud radiative response, except with reversed sign. This echoes the well-known correlation between net cloud radiative effect and low cloud amount in the

Net Cloud Radiative Response [W m−2]

Latitude [deg N]

Latitude [deg N]

−20

90 50 30

−10 a)

0

10

−2


90 50 30

0

0

−30 −50 −90

−30 −50 −90

90 50 30

c)

4xP0: global mean = 0.82 W m−2 90 50 30

0 −30 −50 −90 0

20 −2

b)

4xLC: global mean = 2.23 W m

d)

P4: global mean = 1.65 W m−2

0

60


300

−30 −50 −90 360 0

60


300

360

Figure 12. Net cloud radiative response: coupled (43) versus specified SST (4xP0, P4, 4xLC) experiments.

BRETHERTON ET AL.


1198


10.1002/2014MS000355

ΔCLDLOW

Latitude [deg N]

Latitude [deg N]

−0.2

90 50 30

−0.1

0

0.1

4x: global mean = −0.00

a)

90 50 30

0

0

−30 −50 −90

−30 −50 −90

90 50 30

c)

4xP0: global mean = −0.01 90 50 30

0 −30 −50 −90 0

0.2

b)

4xLC: global mean = −0.01

d)

P4: global mean = 0.00

0

60


300

−30 −50 −90 360 0

60


300

360

Figure 13. Low cloud change: coupled versus specified SST experiments.

current climate [Klein and Hartmann, 1993]. Thus, low cloud changes explain almost all the cloud-radiative changes in the previous figure; in particular the low cloud decrease over land in both 43 and 4xLC explains the positive net cloud radiative response over land in these simulations. Both the DCO2-driven and DSSTdriven low cloud responses contribute to these low cloud decreases over land. Using a subset of CMIP5 models, Zelinka et al. [2013] found low cloud reduction over land that in the multimodel mean is weaker than SP-CCSM4 and develops mainly in the initial DCO2-induced rapid adjustment. Since low cloud reduction over land is important to the positive low cloud feedbacks in SP-CCSM4, and since shallow convective cloud over land is very difficult to parameterize well in conventional GCMs, this issue bears further study, perhaps using a superparameterized model with a higher-resolution CRM that can better simulate shallow convection over land [Lenderink et al., 2004]. 5.4. Coupled Versus Specified-SST Responses in x500 and Precipitation Figure 14 compares the 43 x500 response to the specified-SST cases. The CO2-only response is weak, though with systematic ascent ðx500 < 0Þ over land and subsidence over ocean [Wyant et al., 2012]. The P4 response is stronger in most regions, and dominates the 4xLC response. The spatial pattern of x500 response in 4xLC is fairly similar to the 43 (coupled) case. An important common feature between the two cases is a reduction of the tropical overturning circulation (a positive x500 response in the rainy regions and a negative x500 response in the subsidence regions) [Chadwick et al., 2013]. For the 43 case, there is also more ascent along the equatorial Pacific, also like most CMIP models, for reasons noted in the previous subsection. The 4xLC case also exhibits this response, but to a lesser extent. The 43 and specified-SST precipitation responses, shown in Figure 15, have spatial structure anticorrelated with x500. This is particularly evident in the 4xP0 (DCO2-only) response, for which their pattern correlation is 20.80. In the other cases, the atmospheric warming introduces a thermodynamic component into the precipitation response [Bony et al., 2013], due to Clausius-Clapeyron-driven increase of the precipitation produced by a given upward mass flux [Held and Soden, 2006]. Nevertheless, the 43 (coupled) pattern correlation of x500 and precipitation is 20.75. We interpret 43 precipitation increases in the equatorial Pacific as being mainly dynamically driven [Bony et al., 2013].

BRETHERTON ET AL.


1199

Journal of Advances in Modeling Earth Systems Δω

500

Latitude [deg N]

Latitude [deg N]

−0.04

90 50 30

−0.02

[Pa s−1] 0

0.02

4x

a)

90 50 30

0

0

−30 −50 −90

−30 −50 −90

90 50 30

4xP0

c)

10.1002/2014MS000355

90 50 30

0

b)

4xLC

d)

P4

0.04

0

−30 −50 −90 0

60


−30 −50 −90 360 0

300

60


300

360

Figure 14. 500 hPa omega change: coupled versus specified SST experiments.

ΔPREC [mm d

Latitude [deg N]

Latitude [deg N]

−5

90 50 30

−2.5 a)

2.5

4x: global mean = 0.15 mm d −1 90 50 30 0

−30 −50 −90

−30 −50 −90 c)

4xP0: global mean = −0.15 mm d −1 90 50 30

0 −30 −50 −90 0

]

0

0

90 50 30

−1

5

b)

4xLC: global mean = 0.18 mm d −1

d)

P4: global mean = 0.39 mm d −1

0

60


300

−30 −50 −90 360 0

60


300

360

Figure 15. Precipitation change: coupled versus specified SST experiments.

BRETHERTON ET AL.


1200

ΔCloud Radiative Effect [W m−2]


10.1002/2014MS000355

P4: SP−CAM4

P2: SP−CAM3 5

10

0

0

−5

−10

ΔNet CRE ΔLW CRE ΔSW CRE

−10 −90 −50 −30 0 30 50 90 Latitude [deg N]

−20 −90 −50 −30 0 30 50 90 Latitude [deg N]

Figure 16. Zonal mean CRE change for (a) P2 climate perturbation using 3.5 year SP-CAM3 simulations, and (b) P4 SP-CAM4 climate perturbation, with vertical scale doubled to allow fair comparison with a P2 simulation. The gray shading on both plots bounds the 62r uncertainty on the mean net CRE due to interannual variability estimated from the 34 years of the SP-CAM4 P4 experiment.

The 4xP0 and P4 precipitation responses in Figures 15b and 15d can be approximately compared with the CMIP5 multimodel mean responses of Bony et al. [2013, Figures 2d and 2e], keeping in mind that the latter were derived from Gregory regressions rather than fixed-SST simulations. As with other fields, the DCO2only response is weaker in SP-CAM4 than the CMIP5 multimodel mean, although it has a broadly similar geographical structure. SP-CAM4 is not unique in this respect. Bony et al. [2013] note that the intermodel variability between CMIP5 GCMs in the DCO2-only response accounts for a much larger fraction of the intermodel variability in 43 precipitation changes that does the temperature-mediated response.

6. Comparison With SP-CAM3 Cloud Responses SP-CAM4 is not dramatically different from SP-CAM3. The CRM is nearly unchanged between the two models. SP-CAM4 uses a finite-volume dynamical core versus a semi-Lagrangian tracer advection scheme in SPCAM3. There are also differences in the land model, but one might anticipate that the cloud response of the two models to a climate perturbation would be similar, at least over the oceans. Comparison of our results with those published for SP-CAM3 suggests otherwise. As discussed in the introduction, Wyant et al. [2006] examined the cloud response of SP-CAM3 to a uniform 2 K SST increase (P2), using 2 years of output after spinup. They found an increase in marine low cloud cover and negative shortwave and net CRE changes at most latitudes in the warmer simulation. Our SP-CAM4 results suggest low cloud decreases in some latitude belts, especially the tropics, both for the coupled (43) case and for the P4 case analogous to Wyant et al.’s [2006] study. Similarly, Wyant et al. [2012] noted slight low cloud increases over land in their SP-CAM3 fixed-SST CO2-quadrupling study, using 3.5 years of control and perturbed simulation, while our 4xP0 experiment shows cloud decreases over land. In this section, we test whether these differences are meaningful or whether they can be explained purely by interannual variability of the cloud response. The SP-CAM3 papers attempted to address this issue using the interannual differences between their simulated years, but with only 2–3 years of model output this is not very reliable. Our 35 year long SP-CAM4 specified-SST experiments are long enough to accurately estimate a year-to-year standard deviation in any cloud or radiation statistic, which we further assume is not substantially different in SP-CAM3. 6.1. Response to a Uniform SST Increase Figure 16 compares the shortwave, longwave, and net CRE from the SP-CAM3 P2 climate perturbation (adapted from Wyant et al. [2006, Figure 1]) and the SP-CAM4 P4 climate perturbation. The latter is plotted with a doubled vertical scale for a fair comparison. In both plots, the gray shading is a 62r uncertainty range of net CRE over the averaging period. Here the estimated r5N21=2 ri , where ri is the corresponding interannual standard deviation based on years 2–35 of the SP-CAM4 P4 experiment, and N is the number of simulated years (3 for SP-CAM3, 34 for SP-CAM4). The uncertainty range is much larger for SP-CAM3 because fewer years are represented in the average.

BRETHERTON ET AL.


1201


10.1002/2014MS000355

The longwave CRE is similar between the models. Throughout most of the tropics, the shortwave 0.05 and net CRE are negative in SP0 0 CAM3, while they are near zero or positive in SP-CAM4. These differen−30 −0.05 −50 ces in net CRE somewhat exceed −90 −0.1 the 62r uncertainty range, and 0 60 120 180 240 300 360 Longitude [deg E] probably reflect real effects from the changes in model version. As a Figure 17. Map of annual-mean SP-CAM3 P2 low cloud change. For fair comparison result, the P2-deduced climate senwith SP-CAM4 P4 results shown in Figure 13d, the range of the color scale has been sitivity parameter of SP-CAM3 is halved compared to that figure. 0.41 K W– 1 m2 [Wyant et al., 2006], while we compute the P4-deduced climate sensitivity parameter of SP-CAM4 to be 0.58 K W– 1 m2, reflecting the systematically more positive cloud feedbacks in SP-CAM4 compared to SP-CAM3. SP−CAM3: global mean = 0.01

0.1

Low Cloud Fraction

Latitude [deg N]

90 50 30

Figure 17 shows the regional structure of the low cloud changes in the SPCAM3 P2 simulations, which are the most important determinants of its net cloud feedback. The color scale is designed for a fair comparison with the SP-CAM4 P4 results in Figure 13d, in which the climate perturbation is twice as large. In SP-CAM3, the spatial pattern of low cloud change is qualitatively similar to SP-CAM4, but low cloud increases more over the low-latitude oceans and decreases less over low-latitude land. 6.2. Response to a CO2 Quadrupling With Fixed SST The spatial patterns of longwave and shortwave cloud adjustment for SP-CAM3 in response to CO2 quadrupling with fixed SST [Wyant et al., 2012, Figure 2] are qualitatively similar to the SP-CAM4 4xP0 results in most locations (not shown). However, there are quantitative differences in mean tropical (30S-30N) cloud and radiation changes, given in Table 3. Even the tropical cloud climatologies of the two SP-CAM control runs are different. SP-CAM4 has larger fractional cover than SP-CAM3 in all vertical layers, low, medium and high. Table 3 includes 62r uncertainty ranges for the time-mean 4xP0 cloud and radiation responses for both models, calculated as in the previous section, but now with N 5 2 for SP-CAM3. The midlevel and high cloud changes from both SP-CAM4 and SP-CAM3 are positive over land and negative over ocean, and their 62r uncertainty ranges nearly overlap. For low cloud over land, SP-CAM4 shows decreases over land, while SP-CAM3 suggests highly uncertain increases. The change in net CRE is correspondingly less negative over land in SP-CAM4 than in SP-CAM3 (to translate net CRE change into a tropical-mean cloud radiative response, add a CO2-masking correction of 1.1 W m– 2 for either model [Wyant et al., 2012]). 6.3. Why Did Low Cloud Response Change From SP-CAM3 to SP-CAM4? Overall, SP-CAM4 appears to have stronger temperature-mediated low cloud reduction, somewhat more positive cloud feedbacks, and a stronger reduction in low cloud over land than SP-CAM3. It goes beyond the scope of this paper to analyze exactly what caused these changes. The change in land model may contribute but cannot explain why the low cloud changes are systematically more negative in SP-CAM4 than in SP-CAM3 over the ocean as well as over land.SP-CAM4/SP-CCSM4 simulations of other moist phenomena such as the MJO and double-ITCZ biases [Stan and Xu, 2014] also differ from their SP-CAM3/SP-CCSM3

Table 3. SP-CAM3 Versus SP-CAM4 Tropical-Mean (30S–30N) Cloud Cover and Net Cloud Radiative Effect, and Their Sensitivities Over Land and Ocean Regions to CO2 Quadrupling With Fixed SST Model SP-CAM3

SP-CAM4

a

a

Case

Low cld (%)

Med cld (%)

High cld (%)

Net CRE (W m22)

CTL Dland Docn amip Dland Docn

27 0.6 6 0.8 20.1 6 0.3 31 20.9 6 0.2 20.3 6 0.1

7 0.6 6 0.5 20.3 6 0.2 11 1.2 6 0.1 20.6 6 0.04

17 2.0 6 1.0 20.4 6 0.4 28 1.9 6 0.2 20.8 6 0.1

229 21.06 1.3b 20.9 6 0.5 238 0.0 6 0.4 20.8 6 0.1

From Wyant et al. [2012, Table 1]. Estimation of 62r uncertainty ranges for both models is described in text.

b

BRETHERTON ET AL.


1202


10.1002/2014MS000355

counterparts [Stan et al., 2010]. The reasons underlying these differences are under investigation using simulations with intermediate model configurations.

7. Conclusions We have analyzed the cloud response of the super-parameterized coupled climate model SP-CCSM4 to climate perturbations following CFMIP protocols, using an abrupt 43CO2 simulation and related specified-SST perturbation experiments. We use a combination of Gregory regression plots and ISCCP-simulator cloud feedback analysis [Zelinka et al., 2012a]. In most respects, SP-CCSM4’s cloud adjustment and feedbacks lie within the typical range of conventionally parameterized CMIP5 GCMs. Since SP-CCSM4 does not employ a cumulus parameterization, and cumulus parameterization details can substantially affect global cloud feedbacks [Gettelman et al., 2012; Zhao, 2014], this is reassuring (with the caveat that the embedded cloud-resolving models in SP-CCSM4 do not resolve shallow cumuli or boundary-layer turbulence that dominate cloud feedback uncertainty).

Acknowledgments We acknowledge support from the NSF Science and Technology Center for Multi-Scale Modeling of Atmospheric Processes (CMMAP), led by David Randall and managed under the leadership of by Colorado State University under cooperative agreement ATM-0425247. The computations underlying this research used resources of the National Energy Research Scientific Computing Center, which is supported by the Office of Science of the U.S. Department of Energy under Contract DE-AC0205CH11231, and the Extreme Science and Engineering Discovery Environment (XSEDE), which is supported by National Science Foundation grant number OCI1053575. The model output data used to produce the results of this publication can be obtained upon request to C. Stan. The CERES data were obtained from the NASA Langley Research Center CERES ordering tool at http://ceres.larc.nasa.gov/. We would like to thank Mark Zelinka for the code to calculate cloud feedbacks from ISCCP simulator histograms, Cecilia Bitz and Karen Shell for providing the model output used to analyze and plot CCSM4 cloud feedbacks, Matt Wyant for assisting with the analysis of the SP-CAM3 simulations, Marat Khairoutdinov for his sustained contributions to the development of SAM and SP-CAM, and two anonymous reviewers for comments that helped improve the paper.

BRETHERTON ET AL.

Compared to most CMIP5 GCMs, SP-CCSM4’s simulated clouds respond weakly to CO2 changes alone. This is corroborated by fixed-SST simulations with quadrupled CO2. As global surface temperature increases, SPCCSM4 simulates strong reductions in midlevel cloud worldwide and strong increases in Arctic high and low cloud, as noted by Arnold et al. [2014]. Both of these are common in CMIP5 models. SP-CCSM4 also simulates strong reductions in boundary layer cloud over land, more so than most CMIP5 models, including the conventionally parameterized CCSM4, which shares the same land model and has geographically similar patterns of low cloud change over land. These reductions are highly spatially correlated to reduced nearsurface relative humidity driven by reduced moisture availability of land in a warmer climate, and are the main driver of SP-CCSM4’s weakly positive global shortwave cloud feedbacks. SP-CCSM4 also shows the upward shift of high cloud seen in CMIP5 models and the associated positive longwave cloud feedback [Zelinka et al., 2012b]. Overall, SP-CCSM has a moderate positive total cloud feedback of 0.5 W m– 2 K– 1, consistent with its equilibrium climate sensitivity of 2.8 K. As with the CMIP5 models [Ringer et al., 2014], the cloud response of SP-CCSM4 at global and near-global scales is also captured in atmosphere-only simulations with a specified uniform 4 K SST increase. These simulations even reproduce some regional features of the coupled cloud response, such as the low cloud reduction over the continents and slight marine low cloud increases downstream of the subtropical stratocumulus to cumulus transition regions. We compared these atmosphere-only simulations with much shorter published simulations with the previous-generation superparameterized SP-CAM3, finding that the new model has broadly similar patterns of warming-induced cloud change but significantly more positive low cloud feedback in low latitudes, for reasons that appear to have more to do with the host climate model than the embedded CRMs, but are otherwise poorly understood. As superparameterized models are developed with more skillful subgrid turbulence parameterizations or better CRM resolutions, it will be important to see how this affects both the overall magnitude and spatial pattern of their cloud feedbacks. Based on the current study, even just the first 20 years of a coupled abrupt 43CO2 simulation pair or 10 years of specified-SST simulation pairs will give an excellent guide to the largescale cloud response of an SP-CCSM version to various combination of CO2 and surface warming perturbations.

References Andrews, T., J. M. Gregory, M. J. Webb, and K. E. Taylor (2012), Forcing, feedbacks and climate sensitivity in CMIP5 coupled atmosphereocean climate models, Geophys. Res. Lett., 39, L09712, doi:10.1029/2012GL051607. Arnold, N. P., M. Branson, M. A. Burt, D. S. Abbot, Z. Kuang, D. A. Randall, and E. Tziperman (2014), The effects of explicit atmospheric convection at high CO2, Proc. Natl. Acad. Sci. U. S. A., 111, 10,943–10,948, doi:10.1073/pnas.1407175111. Bala, G., K. Caldeira, and R. Nemani (2010), Fast versus slow response in climate change: Implications for the global hydrological cycle, Clim. Dyn., 35, 423–434, doi:10.1007/s00382-009-0583-y. Bitz, C. M., K. M. Shell, P. G. Gent, D. A. Bailey, G. Danabasoglu, K. C. Armour, M. M. Holland, and J. T. Kiehl (2012), Climate sensitivity of the Community Climate System Model, Version 4, J. Clim., 25, 3053–3070. Blossey, P. N., C. S. Bretherton, M. Zhang, A. Cheng, S. Endo, T. Heus, Y. Liu, A. Lock, S. R. de Roode, and K. M. Xu (2013), Marine low clouds to an idealized climate change: The CGILS LES intercomparison, J. Adv. Model. Earth Syst., 5, 234–258, doi:10.1002/jame.20025. Bony, S., and J. Dufresne (2005), Marine boundary layer clouds at the heart of tropical cloud feedback uncertainties in climate models, Geophys. Res. Lett., 32, L20806, doi:10.1029/2005GL023851.


1203


10.1002/2014MS000355

Bony, S., J.-L. Dufresne, H. Le Treut, J.-J. Morcrette, C. Senior (2004), On dynamic and thermodynamic components of cloud changes, Clim. Dyn., 22, 71–86. Bony S., M. Webb, C. Bretherton, S. Klein, P. Siebesma, G. Tselioudis, and M. Zhang (2011), CFMIP: Towards a better evaluation and understanding of clouds and cloud feedbacks in CMIP5 models, CLIVAR Exch., 56, 20–22. [Available at http://www.clivar.org/publications/ exchanges/688Exchanges_56.pdf.] Bony, S., G. Bellon, D. Klocke, S. Sherwood, S. Fermepin, and S. Denvil (2013), Robust direct effect of carbon dioxide on tropical circulation and regional precipitation, Nat. Geosci., 6, 447–451, doi:10.1038/ngeo1799. Bretherton, C. S., P. N. Blossey and C. R. Jones (2013), Mechanisms of marine low cloud sensitivity to idealized climate perturbations: A single-LES exploration extending the CGILS cases, J. Adv. Model. Earth Syst., 5, 316–337, doi:10.1002/jame.20019. Brient, F., and S. Bony (2013), Interpretation of the positive low-cloud feedback predicted by a climate model under global warming, Clim. Dyn., 40, 2415–2431, doi:10.1007/s00382-011-1279-7. Collins, M., et al. (2013), Long-term climate change: Projections, commitments and irreversibility, in Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, chap. 12, pp. 1029–1136, Cambridge Univ. Press. Chadwick, R., I. Boutle, and G. Martin (2013), Spatial patterns of precipitation change in CMIP5: Why the rich do not get richer in the tropics, J. Clim., 26, 3803–3822. Gettelman, A., J. E. Kay, and K. M. Shell (2012), The evolution of climate sensitivity and climate feedbacks in the Community Atmosphere Model, J. Clim., 25, 1453–1469. Grabowski, W. (2001), Coupling cloud processes with the large-scale dynamics using the cloud-resolving convection parameterization (CRCP), J. Atmos. Sci., 58, 978–997. Gregory, J., and M. Webb (2008), Tropospheric adjustment induces a cloud component in CO2 forcing, J. Clim., 21, 58–71. Hansen, J., A. Lacis, D. Rind, G. Russell, P. Stone, I. Fung, R. Ruedy, and J. Lerner (1984), Climate sensitivity: Analysis of feedback mechanisms, in Climate Processes and Climate Sensitivity, Geophys. Monogr. 29, edited by J. E. Hansen and T. Takahashi, pp. 130–163, AGU, Washington, D. C. Hartmann, D. L., and K. Larson (2002), An important constraint on tropical cloud-climate feedback, Geophys. Res. Lett., 29(20), 1951, doi: 10.1029/2002GL015835. Held, I. M., and B. J. Soden (2006), Robust responses of the hydrological cycle to global warming, J. Clim., 19, 5686–5699. Hurrell, J. W., J. J. Hack, D. Shea, J. M. Caron, and J. Rosinski (2008), A new sea surface temperature and sea ice boundary dataset for the Community Atmosphere Model, J. Clim., 21, 5145–5153, doi:10.1175/2008JCLI2292.1. Joshi, M. M., J. M. Gregory, M. J. Webb, D. M. H. Sexton, T. C. Johns (2008), Mechanisms for the land/sea warming contrast exhibited by simulations of climate change, Clim. Dyn., 30, 455–465, doi:10.1007/s00382-007-0306-1. Khairoutdinov, M. F., and D. A. Randall (2001), A cloud-resolving model as a cloud parameterization in the NCAR Community Climate System Model: Preliminary results, Geophys. Res. Lett., 28, 3617–3620. Khairoutdinov, M. F., C. DeMott, and D. A. Randall (2005), Simulation of the atmospheric general circulation using a cloud-resolving model as a super-parameterization of physical processes, J. Atmos. Sci., 62, 2136–2154. Klein, S. A., and D. L. Hartmann (1993), The seasonal cycle of low stratiform clouds, J. Clim., 6, 1587–1606. Lenderink, G., et al. (2004), The diurnal cycle of shallow cumulus clouds over land: A single-column model intercomparison study, Q. J. R. Meteorol. Soc., 130, 3339–3364. Loeb, N. G., B. A. Wielicki, D. R. Doelling, G. L. Smith, D. F. Keyes, S. Kato, N. Manalo-Smith, and T. Wong (2009), Toward optimal closure of the Earth’s top-of-atmosphere radiation budget, J. Clim., 22, 748–766, doi:10.1175/2008JCLI2637.1. Ringer, M. A., T. Andrews, and M. J. Webb (2014), Global-mean radiative feedbacks and forcing in atmosphere-only and coupled atmosphere-ocean climate change experiments, Geophys. Res. Lett., 41, 4035–4042, doi:10.1002/2014GL060347. Sherwood, S. C., S. Bony and J.-L. Dufresne (2014), Spread in climate model sensitivity traced to atmospheric convective mixing, Nature, 505, 3742. Soden, B. J., and I. M. Held (2006), An assessment of climate feedbacks in coupled ocean-atmosphere models, J. Clim., 19, 3354–3360. Soden, B. J., and G. A. Vecchi (2011), The vertical distribution of cloud feedback in coupled ocean-atmosphere models, Geophys. Res. Lett., 38, L12704, doi:10.1029/2011GL047632. Soden, B. J., A. J. Broccoli, and R. S. Hemler (2004), On the use of cloud forcing to estimate cloud feedback, J. Clim., 17, 3661–3665. Stan, C., and L. Xu (2014), Climate simulations and projections with the super-parameterized CCSM4, Environ. Model. Software, 60, 1234– 1252, doi:10.1016/j.envsoft.2014.06.013. Stan, C., M. Khairoutdinov, C. A. DeMott, V. Krishnamuthy, D. M. Straus, D. A. Randall, J. L. Kinter III, and J. Shukla (2010), An oceanatmosphere climate simulation with an embedded cloud resolving model, Geophys. Res. Lett., 27, L01702, doi:10.1029/2009GL040822. Taylor, K. E., R. J. Stouffer, and G. A. Meehl (2012), An overview of CMIP5 and the experiment design, Bull. Am. Meteorol. Soc., 93, 485–498. Vial, J., J.-L. DuFresne, and S. Bony (2013), On the interpretation of inter-model spread in CMIP5 climate sensitivity estimates, Clim. Dyn., 41, 3339–3362, doi:10.1007/s00382-013-1725-9. Wyant, M. C., M. Khairoutdinov, and C. S. Bretherton (2006), Climate sensitivity and cloud response of a GCM with a superparameterization, Geophys. Res. Lett., 33, L06714, doi:10.1029/2005GL025464. Wyant, M. C., C. S. Bretherton, and P. N. Blossey (2009), Subtropical low cloud response to a warmer climate in a superparameterized climate model. Part I: Regime sorting and physical mechanisms, J. Adv. Model. Earth Syst., 1, 7, doi:10.3894/JAMES.2009.1.7. Wyant, M. C., C. S. Bretherton, P. N. Blossey, and M. Khairoutdinov (2012), Fast cloud adjustment to increasing CO2 in a superparameterized climate model, J. Adv. Model. Earth Syst., 4, M05001, doi:10.1029/2011MS000092. Zelinka, M. D., S. A. Klein, and D. L. Hartmann (2012a), Computing and partitioning cloud feedbacks using cloud property histograms. Part I: Cloud radiative kernels, J. Clim., 25, 3715–3735, doi:10.1175/JCLI-D-11-00248.1. Zelinka, M. D., S. A. Klein, and D. L. Hartmann (2012b), Computing and partitioning cloud feedbacks using cloud property histograms. Part II: Attribution to changes in cloud amount, altitude, and optical depth, J. Clim., 25, 3736–3754, doi:10.1175/JCLI-D-11-00249.1. Zelinka, M. D., S. A. Klein, K. E. Taylor, T. Andrews, M. J. Webb, J. M. Gregory, and P. M. Forster (2013), Contributions of different cloud types to feedbacks and rapid adjustment in CMIP5, J. Clim., 26, 5007–5027, doi:10.1175/JCLI-D-12-00555.1. Zhang, M., et al. (2013), CGILS: Results from the first phase of an international project to understand the physical mechanisms of low cloud feedbacks in general circulation models, J. Adv. Model. Earth Syst., 5, 826–842, doi:10.1002/2013MS000246. Zhao, M. (2014), An investigation of the connections among convection,clouds, and climate sensitivity in a global climate model. J. Clim., 27, 1845–1861.

BRETHERTON ET AL.


1204