4402 JOURNAL OF CLIMATE VOLUME 24

Atmospheric and Surface Contributions to Planetary Albedo

AARON DONOHOE AND DAVID S. BATTISTI

Department of Atmospheric Sciences, University of Washington, Seattle, Washington

(Manuscript received 28 July 2010, in final form 7 March 2011)

ABSTRACT

The planetary albedo is partitioned into a component due to atmospheric reflection and a component due to
surface reflection by using shortwave fluxes at the surface and top of the atmosphere in conjunction with
a simple radiation model. The vast majority of the observed global average planetary albedo (88%) is due to
atmospheric reflection. Surface reflection makes a relatively small contribution to planetary albedo because
the atmosphere attenuates the surface contribution to planetary albedo by a factor of approximately 3. The
global average planetary albedo in the ensemble average of phase 3 of the Coupled Model Intercomparison
Project (CMIP3) preindustrial simulations is also primarily (87%) due to atmospheric albedo. The intermodel
spread in planetary albedo is relatively large and is found to be predominantly a consequence of intermodel
differences in atmospheric albedo, with surface processes playing a much smaller role despite significant
intermodel differences in surface albedo. The CMIP3 models show a decrease in planetary albedo under
a doubling of carbon dioxide—also primarily due to changes in atmospheric reflection (which explains more
than 90% of the intermodel spread). All models show a decrease in planetary albedo due to the lowered
surface albedo associated with a contraction of the cryosphere in a warmer world, but this effect is small
compared to the spread in planetary albedo due to model differences in the change in clouds.

1. Introduction shortwave radiation (Hartmann 1994), which owes ap-

proximately one-third of its existence to meridional gra-
The ratio of reflected to incident shortwave radiation
dient in planetary albedo (Stone 1978). Climate states
at the top of the atmosphere (TOA), the earth’s plane-
with altered meridional structures of planetary albedo
tary albedo, is a function of climate state and exerts
exhibit varying magnitudes of meridional heat transport
a profound influence on the earth’s climate. As a refer-
(Enderton and Marshall 2009).
ence point, Budyko (1969) postulated that a change in
Finally, changes in planetary albedo associated with
global average planetary albedo of less than 0.02 units
both cloud and surface processes are among the most
could cause global glaciation of the climate system. The
powerful feedback processes in the response of the cli-
radiative forcing associated with a doubling of carbon
mate system to anthropogenic forcing (Bony et al. 2006),
dioxide above the preindustrial atmospheric concentra-
and variations in the planetary albedo feedback explain a
tion (3.7 W m22; Solomon et al. 2007) is approximately
large portion of the intermodel spread in climate sensi-
equivalent to a 0.01-unit change in global average plane-
tivity among global climate models (Bender et al. 2006).
tary albedo. The magnitude of temperature variability
Understanding the processes that control the earth’s
seen over the Holocene (Moberg et al. 2005) is consistent
planetary albedo in the current climate state is central to
with planetary albedo variations of 0.01 units or less
predicting how the planetary albedo will respond to fu-
(Cahalan et al. 1994; Bender 2011).
ture climate changes.
The meridional structure of planetary albedo is also
The planetary albedo is a function of the optical
inextricably linked to the strength of the atmospheric
properties of objects within the atmosphere (e.g., clouds,
and oceanic circulation. The large-scale circulation is
water vapor, and aerosols) and objects that constitute
primarily driven by the equator-to-pole gradient in net
the planet’s surface (e.g., ice, ocean, and trees) (Wielicki
et al. 1995; Hall 2004). It is convenient to lump the
Corresponding author address: Aaron Donohoe, University of atmosphere’s contribution to planetary albedo into
Washington, 408 ATG Building, Box 351640, Seattle, WA 98195. three bulk processes: 1) clouds and aerosols that di-
E-mail: aaron@atmos.washington.edu rectly reflect incident solar radiation back to space,

DOI: 10.1175/2011JCLI3946.1

Ó 2011 American Meteorological Society

15 AUGUST 2011 DONOHOE AND BATTISTI 4403

2) atmospheric opacity to downwelling shortwave radi- different climate states, ranging from the last glacial
ation that limits the amount of downwelling shortwave maximum to an ice-free aquaplanet (section 4). A
radiation reaching the surface, and 3) atmospheric summary and discussion of our results follows.
opacity to shortwave radiation upwelling from the sur-
face that limits the amount of shortwave radiation re-
2. Partitioning of planetary albedo in nature
flected by the surface that escapes to space (Qu and Hall
2005). Thus, the atmosphere influences the planetary In this section, we partition the observed planetary
albedo by way of direct reflection back to space (process albedo into atmospheric and surface components and as-
1) and by attenuating the effect of surface albedo on sess the contribution of each component to the hemi-
planetary albedo (processes 2 and 3). Both the atmo- spheric average planetary albedo.
spheric and surface contributions to planetary albedo
are functions of the climate state.
a. Datasets used
Several studies have examined the surface contribu-
tion to local planetary albedo in the polar regions. We use the TOA and surface shortwave radiation data
Gorodetskaya et al. (2006) used the satellite record of products from the Clouds and Earth’s Radiant Energy
planetary albedo in conjunction with observational da- System (CERES) experiment (Wielicki et al. 1996). We
tasets of snow and ice concentration to demonstrate that use Fasullo and Trenberth’s (2008a,b) long-term clima-
changing ice/snow concentrations from 0% to 100% tologies of the CERES TOA data that correct for missing
would result in local planetary albedo changes of order data and global average energy imbalances. For the sur-
0.20 with some regional dependence. This number is face shortwave fluxes, we use the CERES ‘‘AVG’’ fields,
significantly less than the change in surface albedo as- which are derived by assimilating the satellite observa-
sociated with the melting of the cryosphere (Hummel tions into a radiative transfer model to infer the surface
and Reck 1979), which suggests that the atmosphere fluxes (Rutan et al. 2001). All calculations are performed
strongly attenuates the impact of the surface albedo separately for each of the four CERES instruments (FM1
change on the planetary albedo. Qu and Hall (2005) and FM2 on Terra from 2000 to 2005 and FM3 and FM4
found that surface reflection accounts for less than 25% on Aqua from 2002 to 2005). We then average the results
of the climatological planetary albedo in the ice- and over the four instruments. Our calculations are performed
snow-covered regions of the planet and the remainder is on the annual average (solar weighted) data.
due to clouds. They also found that, although the year-to- The observed TOA fluxes from CERES differ from
year variability of planetary albedo in cryospheric regions previous satellite measurements of the Earth Radiation
is mainly due to changes in surface albedo, atmospheric Budget Experiment (ERBE; Barkstrom et al. 1989) by
processes attenuate the effect of the surface albedo as much as 8 W m22 in the zonal average (Trenberth
changes on the local planetary albedo by as much as 90% et al. 2009; Loeb et al. 2009). The instrumental absolute
(i.e., the change in planetary albedo is 10% of the change calibration error in the CERES TOA fluxes are as large
in surface albedo). as 4.2 W m22 (Loeb et al. 2009) and the total error in the
In this study, we quantify the relative contributions of global average shortwave flux is estimated to be
the surface and atmosphere to planetary albedo in the 2.6 W m22 (Bender et al. 2006). Similarly, the CERES
observations (section 2). We then perform the same surface shortwave fluxes have systematic errors ranging
analyses on the output of the preindustrial integrations from 3% to 5% (Kratz et al. 2010) and random errors in
from the climate models used in the International Panel the hourly retrievals of order tens of watts per square
on Climate Change (IPCC) Fourth Assessment Report meter (see Trenberth et al. 2009 for a discussion). In
(AR4) to determine whether the relative contributions appendix A, we propagate the errors in the observa-
of surface and atmospheric processes to the total plan- tional fluxes to obtain error estimates for the quantities
etary albedo observed in nature is captured in the model calculated in this study.
simulations. We also calculate the intermodel spread in
the hemispheric average planetary albedo and de- b. Partitioning planetary albedo into atmospheric
termine what portion of this spread is due to surface or and surface components
atmospheric processes (section 3). Finally, we analyze
1) METHODOLOGY
the changes in planetary albedo in the AR4 climate
models due to a doubling of atmospheric carbon dioxide We now describe a methodology for partitioning the
concentration and partition the changes in planetary planetary albedo into atmospheric and surface compo-
albedo between changes in atmospheric and surface nents, given the upwelling and downwelling solar fluxes
processes. We also examine planetary albedo in vastly at both the TOA and the surface. We will denote the
4404 JOURNAL OF CLIMATE VOLUME 24

atmospheric contribution to the planetary albedo as

aP,ATMOS and the surface contribution to planetary al-
bedo as aP,SURF in the remainder of this paper.
At each grid point we build a single-layer model of
solar radiation that accounts for three shortwave pro-
cesses: atmospheric reflection, atmospheric absorption,
and surface reflection. We assume that each of these
processes is isotropic: a certain percentage of the inci-
dent radiation is absorbed per pass through the atmo-
sphere and a different percentage of the incident radiation
is reflected per pass through the atmosphere. For example,
of the total downwelling solar radiation incident at the
TOA S, a fraction R is reflected by the atmosphere,
a fraction A is absorbed by the atmosphere, and the re-
mainder is transmitted to the surface. Of the transmitted FIG. 1. Schematic representing the first two reflections in the
radiation, a fraction a (the surface albedo) is reflected at single-layer solar radiation model. Moving from left to right, the
the surface back toward the atmosphere. Of this reflected arrows represent the radiative fluxes associated with the incident
radiation, a portion R is reflected back to the surface by solar, first reflection, and second reflection. The variables A, R, and
a are the atmospheric absorption fraction during a single pass
the atmosphere, a portion A is absorbed within the atmo-
through the atmosphere, the fraction of cloud reflection, and the
sphere, and the remainder is transmitted to space (Fig. 1). surface albedo, respectively. The solid arrows at the TOA repre-
These processes are repeated for an infinite number of sent the radiative fluxes that we associated with cloud reflection
reflections. Hence, the annual mean upwelling solar flux and the dashed lines represent the radiative fluxes that we associ-
at each grid point at the TOA is ated with the surface reflection.

F[TOA 5 S[R 1 a(1 2 R 2 A)2 1 a2 R(1 2 R 2 A)2 1 a3 R2 (1 2 R 2 A)2


]

(1 2 R 2 A)2
5 SR 1 Sa(1 2 R 2 A)2 [1 1 (aR) 1 (aR)2


] 5 SR 1 Sa , (1)
1 2 aR

where F[TOA is the upwelling solar flux at the TOA and although it is not clear to us whether a solution to the
the convergence of the infinite series to the final ex- generalized system of equations must exist. Nonethe-
pression on the right-hand side is ensured because both less, solutions to Eqs. (1)–(3) exist at all grid points for
R and a are less than one (Qu and Hall 2005). Similar the datasets and GCM output discussed in this paper.
convergent infinite series can be obtained for the downw- Furthermore, all solutions (A and R values at each grid
elling and upwelling solar fluxes at the surface: point) discussed here are unique.
Solving these equations results in spatial maps of R
(1 2 R 2 A) (Fig. 2d) and A (not shown). Dividing Eq. (1) by S and
FYSURF 5 S , and (2)
1 2 aR separating the two terms allows us to partition the plan-
etary albedo into aP,ATMOS and aP,SURF components:
(1 2 R 2 A)
F[SURF 5 aS 5 aFYSURF . (3)
1 2 aR

TABLE 1. Variables used in this study.

Therefore, given datasets of shortwave radiative
fluxes on the left-hand side of Eqs. (1)–(3) and S, these Symbol Meaning
equations represent a system of three equations in terms a Surface albedo
of three unknown variables: A, R, and a (see Table 1 for aP Planetary albedo 5 TOA albedo
A Percentage of absorption during each
descriptions of all variables). In practice, the ratio of
pass through the atmosphere
upwelling to downwelling radiation at the surface [Eqs. R Percentage of reflection during each
(3) and (2)] defines a such that the system can be re- pass through the atmosphere
duced to two equations [Eqs. (1) and (2)] and two un- aP,ATMOS Atmospheric contribution to planetary albedo
knowns (A and R). One can show that all possible aP,SURF Surface contribution to planetary albedo
x Atmospheric attenuation of surface albedo
solutions to our equations have 0 # R # 1 and 0 # A # 1,
15 AUGUST 2011 DONOHOE AND BATTISTI 4405

FIG. 2. (a) Surface and (c) planetary albedo, and (b) surface and (d) atmospheric contributions to (c). All quantities
are expressed as a percentage, where 1% corresponds to 0.01 units of albedo.

aP,ATMOS 5 R, and (4) and aP,SURF are reported in appendix B. The qualitative
conclusions found in this study are independent of the
(12R2A)2 assumptions made in the simplified radiative transfer
aP,SURF 5 a . (5) model.
1 2 aR

Equation (4) is due to direct reflection by the atmo- 2) RESULTS

sphere [the first term on the right-hand side of Eq. (1)].
All of the solar radiation that is reflected by the surface The maps of surface and planetary albedo exhibit
and eventually passes through the TOA [the second term large values in the polar regions, with larger spatial
on the right-hand side of Eq. (1)] is attributed to aP,SURF. differences in the meridional direction than in the zonal
By definition, the surface and atmospheric contributions direction (Fig. 2); the predominant spatial structure in
to planetary albedo sum to the planetary albedo: both maps is an equator-to-pole gradient. Significant
meridional gradients in surface albedo are constrained
aP 5 aP,ATMOS 1 aP,SURF . (6) to be at the transition to the cryospheric regions (around
708 in each hemisphere), whereas the meridional gradi-
Maps of aP,SURF and aP,ATMOS are shown in Figs. 2b,d. ents in planetary albedo are spread more evenly across
We calculate aP,ATMOS and aP,SURF using annual av- the storm-track regions (from 308 to 608).
erage (solar weighted) data. We have also performed the The percentage of solar radiation absorbed during a
partitioning on the climatological monthly mean data and single pass through the atmosphere (not shown) features
then averaged the monthly values of aP,ATMOS and aP,SURF a predominant equator-to-pole gradient with tropical
to obtain the annual average climatology. The annual and values of order 25% and high-latitude values of order
zonal average aP,ATMOS calculated from the monthly 15% with still smaller values occurring over the highest
data agree with that calculated directly from the annual topography. The global pattern of atmospheric solar
average data to within 1% of aP,ATMOS at all latitudes. absorption is virtually identical to the pattern of verti-
We note that Taylor et al. (2007, hereafter T07) used cally integrated specific humidity [from National Cen-
a similar simplified radiative transfer model to partition ters for Environmental Prediction (NCEP) reanalysis]
planetary albedo into surface and atmospheric compo- with a spatial correlation coefficient of 0.92; this is
nents. In contrast to our formulation, T07 assumed ab- expected because the atmospheric absorption of solar
sorption only occurs on the first downward pass through radiation is predominantly due to water vapor and ozone
the atmosphere and occurs above the level of atmo- (Chou and Lee 1996) while clouds and black carbon each
spheric reflection. The impact of the differences in contribute less than 10% of the atmospheric absorption
model formulation on the derived quantities aP,ATMOS (Kim and Ramanathan 2008).
4406 JOURNAL OF CLIMATE VOLUME 24

The map of aP,ATMOS (Fig. 2d) shows several antici- TABLE 2. Global average planetary albedo, its decomposition
pated features: 1) high values associated with the in- into atmospheric and surface components (as described in the
text), and global average surface albedo for the observations and
tertropical convergence zone (ITCZ) in the equatorial
the CMIP3 multimodel average and spread (2s). The observa-
Pacific, 2) low values in the drier subtropics, 3) high tional uncertainty on total planetary albedo is taken from Bender
values in the NH Atlantic and Pacific storm tracks and in et al. (2006). The observational uncertainty on aP,ATMOS and
the nearly zonally symmetric SH storm track associated aP,SURF is assessed by the method described in appendix A. All
with synoptic storms, 4) high values in the Arctic and entries are percentages.
West Antarctica that tend to be cloudy during the local aP aP,ATMOS aP,SURF a
summer (Curry et al. 1996), and 5) high values over
Obs 29.8 26.2 3.6 12.3
Southeast Asia associated with extensive cold season Observational uncertainty (2s) 0.8 1.8 0.9 —
stratocumulus (Hahn and Warren 2003). Model average 30.4 26.4 4.0 13.2
As is true for the planetary albedo and aP,ATMOS, Model spread (2s) 1.6 2.3 1.3 3.1
aP,SURF is greater in the polar regions than in the tropics.
More important, however, is that aP,SURF is greatly at-
tenuated by the atmosphere (cf. Figs. 2a and 2b)—so much more of the climatological planetary albedo
much so that almost everywhere the planetary albedo ($75%) than the surface.’’ The hemispheric average
is largely determined by the atmospheric and not the planetary albedo is very similar in both hemispheres
surface contribution (cf. Figs. 2b and 2d). Our simple (0.299 in the NH versus 0.298 in the SH), although the
radiative model predicts that aP,SURF will be attenuated partitioning between aP,ATMOS and aP,SURF differs
by square of the atmospheric transmissivity: (1 2 R 2 A)2 slightly between the hemispheres (aP,ATMOS contributes
[Eq. (5) and Fig. 1]. In this framework, we can see that 86% of the hemispheric average planetary albedo in the
while the atmospheric opacity primarily acts to damp NH as compared to 90% in the SH).
the impact of the surface albedo on planetary albedo, The zonal average planetary albedo, partitioned into
the atmosphere also imparts some spatial structure to aP,ATMOS and aP,SURF, is shown in Fig. 3a. At all lati-
aP,SURF by virtue of the spatial structure in atmospheric tudes, aP,ATMOS is greater than aP,SURF (higher values
opacity. For example, the surface albedo over Antarc- of aP,SURF are over the cryospheric and NH landmass
tica is fairly uniform, but aP,SURF is much larger over regions, owing to the high surface albedo in those re-
East Antarctica because the atmosphere over East Ant- gions). The largest values of aP,ATMOS are in the storm-
arctica is less opaque (both less cloud and less absorption track regions and the Arctic where clouds are prevalent,
by water vapor) than that over West Antarctica. and the smallest values are in the dry subtropics; both
Atmospheric processes play a much greater role than the meridional structure of cloud fraction and solar ze-
surface processes in determining the amplitude and nith angle (Minnett 1999) contribute to the meridional
structure of planetary albedo, both regionally and in structure of aP,ATMOS.
the global average. The (solar weighted) global average As can be seen in Fig. 3b (or by comparing Figs. 2a and
planetary albedo of 0.298 is partitioned into a global 2b), aP,SURF is much smaller than a, because 1) the
average aP,ATMOS of 0.262 (88% of the total) and downwelling solar radiation at the TOA is attenuated
a global average aP,SURF of 0.036 (12%; Table 2). These by atmospheric reflection and absorption as it passes
results are in qualitative agreement with Qu and Hall’s downward through the atmosphere and 2) the solar
(2005) conclusion that ‘‘the atmosphere accounts for radiation reflected by the surface is attenuated by

FIG. 3. (a) Zonal mean planetary albedo partitioned between atmospheric and surface components. (b) Zonal mean surface albedo and
surface component of planetary albedo. (c) Atmospheric attenuation of surface albedo.
15 AUGUST 2011 DONOHOE AND BATTISTI 4407

atmospheric absorption and reflection as it passes up- b. Results

ward from the surface to the TOA. Even over the polar
regions, where the zonal mean surface albedo exceeds Averaged over all the CMIP3 models, the global av-
70%, aP,SURF never exceeds 30%. We define the at- erage planetary albedo is 0.304 (Table 2). This value is
mospheric attenuation of surface albedo x as within the range of observational estimates based on the
corrected ERBE data (0.313) and CERES data (0.298)
aP,SURF (1 2 R 2 A)2 (see Fasullo and Trenberth 2008a,b and Loeb et al. 2009
x51 2 51 2 . (7)
a 1 2 Ra for a thorough discussion of the different satellite da-
The atmospheric attenuation of the surface albedo (Fig. tasets and their uncertainties). The intermodel spread
3c) ranges from a maximum of 80%–85% in the storm (two standard deviations or 2s) in planetary albedo is
tracks to a minimum of 60% in the dry subtropics. The 0.016, corresponding to a TOA net shortwave differ-
meridional variations in atmospheric attenuation of ence of approximately 5.5 W m22 across the models.
surface albedo are small compared to the global mean Averaged across the models, aP,ATMOS contributes
atmospheric attenuation despite the large meridional 87% of the global average planetary albedo, which
variations in cloud properties; this is partly because solar agrees with the observed partitioning of global average
radiation is attenuated by clear-sky processes (i.e., solar planetary albedo between atmospheric and surface
atmospheric absorption), which contribute as much as processes.
50% of the atmospheric attenuation in the global mean As is true of the observations, the two hemispheres
(Bony et al. 2006). contribute nearly equally to the global average planetary
Over the observational period analyzed in this study, albedo in the models (spread on the ordinate of Fig. 4).
atmospheric processes are found to be the dominant Also in agreement with the observations, aP,SURF con-
(88%) contributor to global average planetary albedo tributes slightly more to the hemispheric average plan-
while surface processes make a much smaller contribu- etary albedo in the NH than in the SH (the offset on
tion to the global average planetary albedo. abscissa of Fig. 4b between the NH and SH data).
The intermodel spread in hemispheric average
aP,ATMOS (2s 5 0.022) is approximately twice as large as
3. Partitioning of planetary albedo in the models
that of aP,SURF (2s 5 0.012). Overall, the intermodel
used in IPCC AR4
spread of hemispheric average planetary albedo is well
In this section, we look at the intermodel spread in correlated with the intermodel spread in hemispheric
planetary albedo in the preindustrial (PI) simulations of average aP,ATMOS (Fig. 4a; R2 5 0.53 and 0.71 in the NH
the climate system performed using the climate models and SH); it is not correlated with the hemispheric av-
that were used in the IPCC AR4. We then employ the erage aP,SURF (Fig. 4b). The hemispheric average
same methodology used in section 2 to partition plane- planetary albedo is poorly correlated with the hemi-
tary albedo into its atmospheric and surface compo- spheric average aP,SURF because the spread in hemi-
nents. We find, not surprisingly, that the intermodel spheric average aP,SURF is small compared to the spread
spread in planetary albedo is due to the differences in in aP,ATMOS. Furthermore, the intermodel spreads in
atmospheric processes and only weakly related to dif- aP,ATMOS and aP,SURF are negatively correlated as ex-
ferences in surface albedo. pected from Eq. (8) and discussed later in this section.
The large intermodel spread in global and hemi-
a. Model runs analyzed
spheric average planetary albedo is primarily due to
We use data from the World Climate Research Pro- atmospheric processes as opposed to surface albedo for
gramme’s (WCRP) Coupled Model Intercomparison three reasons: 1) the climatological planetary albedo is
Project phase 3 (CMIP3) multimodel dataset: an archive primarily (87%) determined by atmospheric reflection,
of a suite of standardized coupled simulations from 25 2) the surface albedo contribution to planetary albedo is
global climate models that were included in the Inter- strongly attenuated by the atmospheric opacity, and 3)
national Panel on Climate Change’s Fourth Assessment a portion of the intermodel spread in surface albedo is
Report (https://esgcet.llnl.gov:8443/index.jsp). The set of a consequence of the intermodel spread in atmospheric
model simulations is commonly referred to as the WCRP’s opacity. As a consequence, the hemispheric average
CMIP3 multimodel dataset (Meehl et al. 2007). We use surface albedo is very weakly correlated with hemi-
output from the PI simulations, which are forced with spheric average planetary albedo (R2 5 0.03 in the NH
temporally invariant external forcing (CO2 is set to and 0.12 in the SH; Fig. 5) despite the relatively large
280 ppm). We use 15 coupled models that provided the intermodel spread in hemispheric average surface al-
output fields required for our analysis (Table 3). bedo (2s 5 0.022 in the NH and 0.036 in the SH).
4408 JOURNAL OF CLIMATE VOLUME 24

TABLE 3. Models used in this study and their resolution. The horizontal resolution refers to the latitudinal and longitudinal grid spacing or
the spectral truncation. The vertical resolution is the number of vertical levels.

Horizontal Vertical
Abbreviation Full name (Country) resolution resolution
BCCR-BCM2.0 Bjerknes Centre for Climate Research Bergen Climate Model T63 L31
version 2.0 (Norway)
CCCma-CGCM3.1 Canadian Centre for Climate Modelling and Analysis Coupled T47 L31
General Circulation Model, version 3.1 (Canada)
CNRM-CM3 Centre National de Recherches Météorologiques Coupled Global T63 L45
Climate Model, version 3 (France)
CSIRO-MK3.0 Commonwealth Scientific and Industrial Research Organisation Mark T63 L18
version 3.0 (Australia)
GFDL-CM2.0 National Oceanic and Atmospheric Administration (NOAA)/Geophysical 2.08 3 2.58 L24
Fluid Dynamics Laboratory GFDL Climate Model version 2.0 (USA)
GISS-ER National Aeronautics and Space Administration (NASA)/Goddard 48 3 58 L20
Institute for Space Studies Model E-R (USA)
IAP-FGOALS Institute of Atmospheric Physics Flexible Global Ocean– T42 L26
Atmosphere–Land System Model,
National Key Laboratory of Numerical Modeling for Atmospheric
Sciences and Geophysical Fluid Dynamics (LASG) (China)
MPI-ECHAM5 Max Planck Institute for Meteorology/ECHAM5 (Germany) T63 L31
INM-CM3.0 Institute of Numerical Mathematics Coupled Model, version 3.0 (Russia) 48 3 58 L21
IPSL CM4.0 L’Institut Pierre-Simon Laplace Coupled Model, version 4 (France) 2.58 3 3.758 L19
Miroc3.2(hires) Model for Interdisciplinary Research on Climate 3.2, high-resolution T106 L56
version (Japan)
MRI-CGCM2.3.2a Meteorological Research Institute Coupled General Circulation Model, T42 L30
version 2.3.2a (Japan)
NCAR-CCSM3.0 NCAR Community Climate System Model, version 3 (USA) T85 L26
UKMO HadCM3 Third climate configuration of the Met Office Unified Model (UK) 2.58 3 3.88 L19
MIUB-ECHOg Meteorological Institute of the University of Bonn, ECHO-G T30 L19
Model (Germany)

The spatial map of the intermodel average planetary observational uncertainty in planetary albedo over
albedo is very similar to that observed except in Saharan those regions (Loeb et al. 2009). The intermodel spread
Africa, the Arabian Peninsula, and the subtropical in planetary albedo is greatest in the midlatitude storm-
stratocumulus regions (not shown); in these regions, the track regions over the North Atlantic, North Pacific, and
models are biased low (the difference exceeds 2s) Southern Ocean and in the cryosphere regions of both
compared to the observations. This bias exceeds the hemispheres (not shown).

FIG. 4. (a) Hemispheric average aP vs hemispheric average aP,ATMOS in the CMIP3 PI simulations (crosses),
observations (solid squares), and altered climate states. Red symbols and lines are for the SH while blue denotes the
NH. The dashed lines are linear best fits to the PI simulations and the solid line is the 1:1 line with zero intercept. (b)
As in (a), except plotted against hemispheric average aP,SURF.
15 AUGUST 2011 DONOHOE AND BATTISTI 4409

reflectance would have a reduced aP,SURF because the

reflective surface is less visible from the TOA. This re-
sult is interesting because it suggests that a portion of the
intermodel spread in aP,SURF is due to atmospheric
variability across models as opposed to the intermodel
spread of surface albedo.

4. Planetary albedo in simulations of altered

climate states
In this section, we analyze the change in planetary
albedo and its partitioning between surface and atmo-
spheric processes due to a doubling of carbon dioxide
from the preindustrial values and in climate states with
severely altered surface albedo distributions.
FIG. 5. Hemispheric average aP vs hemispheric average a in the
PI CMIP3 simulation (crosses), observations (solid squares), and a. Carbon dioxide doubling experiments
altered climate states. Red symbols and lines are for the SH while
blue denotes the NH. The axes have a 1:1 ratio. 1) MODEL RUNS USED
Here we analyze the output from the ‘‘1pctto2x’’
The zonal and annual average surface albedo, plane- simulations in the CMIP3 archive (Meehl et al. 2007).
tary albedo, aP,ATMOS, and aP,SURF for each member of The initial conditions for each model come from the
the CMIP3 PI ensemble are coplotted (dashed black equilibrated PI simulations. Atmospheric CO2 is increased
lines) in Figs. 6a–d alongside the observations (light blue at 1% yr21 until CO2 has doubled relative to the PI con-
lines). (Results from some additional model experi- centration (70 yr). The simulations are then run forward
ments are also shown in Fig. 6 and will be discussed in for an additional 150 yr with carbon dioxide fixed at twice
section 4.) In agreement with the observations, the the PI concentration. We average the model output over
aP,ATMOS in the CMIP3 ensemble is larger than aP,SURF the last 20 yr of these simulations (years 201–220) and
at all latitudes with the exception of the high Arctic and compare the climatological fields to their counterparts in
Antarctic (cf. Figs. 6c and 6d) . The meridional structure that model’s PI simulation. Theses runs will be referred to
of surface albedo is consistent with the observations. as the 2 3 CO2 and PI runs, respectively.
However, there are substantial differences between
2) RESULTS
models in annual average surface albedo, especially in
the cryosphere regions as discussed in Roesch (2006). Averaged across all models, the doubling of CO2 re-
The zonal average planetary albedo (Fig. 6b) exhibits sults in a reduction in the hemispheric average planetary
substantial intermodel spread, exceeding 0.08 units (2s) albedo of 20.007 6 0.009 in the NH and 20.004 6 0.009
in the high latitudes (Fig. 7), which is attributed to dif- in the SH; the range quoted is 2 standard deviations. The
ferences in aP,ATMOS in our analysis. Only in the high intermodel spread in the change of hemispheric average
polar regions does aP,SURF contribute significantly to the planetary albedo is strongly correlated with the change
intermodel spread in planetary albedo (Fig. 7). This in aP,ATMOS in both hemispheres (Fig. 8a) with an R2
result suggests that model differences in the meridional value of 0.91 in the NH and 0.96 in the SH (both sig-
structure of planetary albedo are primarily determined nificant at the 99% confidence interval). In contrast, the
by atmospheric (cloud) processes and only weakly by hemispheric average aP,SURF change is not significantly
intermodel differences in surface processes. correlated with the change in planetary albedo in either
With the exception of the SH ice margin, the inter- hemisphere (Fig. 8b). This is primarily because the inter-
model standard deviation of aP,ATMOS exceeds the total model spread in hemispheric average aP,ATMOS change
intermodel standard deviation of planetary albedo (Fig. is approximately 3 times as large of that in aP,SURF.
7); this result is a consequence of the negative correla- Although the change in hemispheric average aP,SURF is
tion between aP,ATMOS and aP,SURF (not shown) that negative in all models (abscissa of Fig. 8b) because of
was also noted above. The negative correlation is ex- decreasing ice and snow cover, the intermodel spread in
pected based on our solar radiation model [i.e., Eq. (5)]; the aP,ATMOS overwhelms the change in surface contri-
if two models had identical surface albedo in the Arctic bution resulting in a total planetary albedo feedback
but different cloud reflectance (and hence atmospheric that is ambiguous in sign among the ensemble members.
transmissivity) the model with the larger atmospheric This result is in qualitative agreement with Kato et al.’s
4410 JOURNAL OF CLIMATE VOLUME 24

FIG. 6. Zonal annual mean (a) a, (b) aP, (c) aP,SURF, and (d) aP,ATMOS in the PI simulations from the CMIP3
models (dashed black lines). Also shown are the observations (solid light blue line) and model simulations of altered
climate states (other colored lines).

(2006) conclusion that the recent decline in Arctic sea state, the surface albedo contribution to planetary al-
ice has a negligible signature on the Arctic radiative bedo could change by virtue of an atmospheric opacity
budget at the TOA because of atmospheric attenuation change. Nonetheless, the changes in aP,SURF in the
and uncertainty in cloud feedbacks. 2 3 CO2 runs are dominated by the first term on the rhs
The change in hemispheric average aP,SURF in the of Eq. (8) because the fractional changes in a are much
2 3 CO2 runs is well correlated with the change is hemi- larger than the fractional changes in either R or A (not
spheric average surface albedo (Fig. 9a, R2 5 0.77 in the shown). Furthermore, because both R and a are less than
NH and 0.87 in the SH). This result can be understood by unity, Ra/(1 2 Ra)  1, Eq. (8) can be approximated by
differentiating Eq. (5) with respect to a, R, and A:
(1 2 R 2 A)2
2  DaP,SURF ’ Da [ (1 2 x)Da. (9)
(1 2 R 2 A) Ra 1 2 Ra
D(aP,SURF ) 5 11 Da
(1 2 Ra) 1 2 Ra
Equation (9) neglects how changes in atmospheric
2a(1 2 R 2 A) a(1 2 R 2 A) opacity and changes in multiple reflections impact
2 DA 2
1 2 Ra 1 2 Ra aP,SURF. It predicts that the changes in planetary albedo
 
12R2A due to surface reflections are equal to the change in
3 22a DR.
1 2 Ra surface albedo modified by the atmospheric attenuation
(8) of surface albedo in the mean climate. If all climate
models had the same atmospheric attenuation of surface
Equation (8) highlights that changes in aP,SURF are albedo [and Eq. (9) held perfectly], then we would ex-
functions of both surface and atmospheric state; even if pect the change in aP,SURF in the 2 3 CO2 to be pro-
surface albedo remains constant in an altered climate portional to Da with a slope equal to (1 2 x). The strong
15 AUGUST 2011 DONOHOE AND BATTISTI 4411

have replicated Qu and Hall’s result using their method

on the datasets considered in this study suggesting the
different quantitative conclusions are a consequence
of methodological differences. The sensitivity of the
atmospheric attenuation of surface albedo to the as-
sumptions made in our simplified radiative transfer
model (Fig. 1) is further discussed in appendix B.
Figure 9b shows that there is no correlation between
the change in planetary albedo and the change in surface
albedo, averaged over a hemisphere. This is a conse-
quence of the atmospheric opacity attenuating the con-
tribution of surface albedo changes to F[TOA and the
large intermodel spread in changes in atmospheric
(cloud) reflectivity. This result suggests that changes in
surface albedo associated with poleward retreat of the
FIG. 7. The intermodel standard deviation of zonal average aP cryosphere relative to its location in the present day cli-
(solid line), aP,ATMOS (dashed line), and aP,SURF (dashed–dotted
mate are a poor predictor of changes in planetary albedo
line) for the CMIP3 PI simulations.
because the shortwave cloud feedback uncertainties exert
a more profound influence on the F[TOA than the direct
correlation in Fig. 9a indicates that the models have
radiative forcing of the surface albedo. These results
a similar atmospheric attenuation of surface albedo and
agree with Trenberth and Fasullo (2009) who found that
that intermodel spread in the surface albedo feedback is
decreases in planetary albedo in the CMIP3 ensemble
primarily driven by model differences in the surface al-
simulations for the A1B and A2 CO2 emission scenarios
bedo response to warming. This result is in qualitative
are primarily a consequence of reduced cloud cover.
agreement with Qu and Hall’s (2005) conclusion that
the surface albedo feedback spread in climate models is
b. Simulations with specified surface albedo
due to spread in the simulated surface albedo changes
distributions
and not in the intermodel spread of how surface albedo
changes contribute to F[TOA. The slope of the best-fit In this section, we explore the relationship between
lines in Fig. 9a are 0.34 in the NH and 0.30 in the SH, surface albedo and planetary albedo in models of radi-
suggesting that the change in planetary albedo associ- cally different climates that feature severely altered
ated with a change in surface albedo is approximately spatial patterns of surface albedo.
one-third of the magnitude of the change in surface al-
1) MODEL RUNS USED
bedo. This result is at odds with Qu and Hall’s (2005)
finding that the change in planetary albedo per unit We consider three greatly different climate states: the
change in surface albedo is approximately one-half. We Last Glacial Maximum (LGM), the equilibrated climate

FIG. 8. (a) Change in hemispheric average planetary albedo in the 2 3 CO2 runs (relative to the PI simulations) vs
change in hemispheric average aP,ATMOS. The lines are the linear best fits in each hemisphere. (b) As in (a), but vs
hemispheric average aP,SURF. All quantities are expressed as a percentage, where 1% corresponds to 0.01 units of albedo.
4412 JOURNAL OF CLIMATE VOLUME 24

FIG. 9. (a) Change in hemispheric average aP,SURF in the 2 3 CO2 runs (relative to the PI simulations) vs change in
hemispheric average a. The lines are the linear best fits in each hemisphere. (b) Change in hemispheric average aP in
the 2 3 CO2 runs (relative to the PI simulations) vs change in hemispheric average a.

under 4 times the PI CO2 concentration (QUAD), and an the QUAD climate has a reduced high-latitude surface
aquaplanet (AQUA) with no sea ice. The LGM simula- albedo because of a poleward shift in the ice edge, while
tion is performed with the National Center for Atmo- the LGM run has greatly enhanced surface albedo in the
spheric Research’s (NCAR) Community Atmosphere NH midlatitudes mainly because of the presence of large
Model version 3 (CAM3) at a resolution of T42; it is continental ice sheets.
forced with glacial orbital parameters and greenhouse
gas concentrations, specified land ice albedo and topog-
2) RESULTS
raphy and with sea ice concentrations and sea surface
temperatures prescribed from the output of a fully cou- While the changes in sea ice extent and in the distri-
pled simulation of the LGM [see Li and Battisti (2008) for bution of land ice in the altered climate states greatly
a description]. The QUAD simulation is also performed affect the surface albedo, the changes in planetary al-
with CAM3 at a resolution of T42 with the greenhouse bedo are much smaller than the changes in surface al-
gas concentrations fixed at 1000 ppm and with sea surface bedo (Figs. 6, 10). For example, in the LGM simulations
temperatures, sea ice concentrations, and snow extent the surface albedo in the midlatitude NH increased by 0.25
prescribed as the output of an equilibrated coupled sim- (from ;0.15 in the PI simulation to ;0.40 in the LGM
ulation. The LGM and QUAD runs are compared to experiment), but the planetary albedo only increased by
a T42 CAM3 PI industrial run (the asterisks in Figs. 4 approximately 0.10 relative to the PI simulations be-
and 5) as opposed to the T63 coupled CCSM3 PI run that cause of a 0.06 increase in aP,SURF and a 0.04 increase in
is included in the CMIP3 ensemble. The AQUA simu- aP,ATMOS (Fig. 10; the latter is due to more persistent
lation is performed with Geophysical Fluid Dynamics clouds over the continental ice sheets).
Laboratory Atmosphere Model version 2 (GFDL AM2) In the global average, the planetary albedo increases
coupled to a 50-m-depth slab ocean and sea ice formation from 0.300 in the PI to 0.320 in the LGM and is parti-
is prohibited even if the temperature drops below the tioned into a 0.012 increase in aP,SURF and a 0.008 in-
freezing point [this run is identical to Kang et al. (2008), crease in aP,ATMOS (Table 4). The global average surface
except with a 50-m slab ocean]. All simulations are run albedo increase of 0.040 exceeds the aP,SURF change by
for 50 yr and we average the output over the last 20 yr of more than a factor of 3, again suggesting that the impact
a single ensemble member of each simulation. The of surface albedo changes on the TOA are attenuated by
AQUA simulation is compared to the fully coupled PI a factor of approximately 3 by atmospheric opacity.
simulation included in the CMIP3 multimodel ensemble. These results are in close agreement with T07, who found
The prescribed spatial distribution of surface albedo that the surface albedo changes over the ice sheets alone
varies widely in these runs (Figs. 6a, 10). The AQUA accounted for a 0.008 and 0.007 change in global average
climate has almost no meridional gradient in surface al- planetary albedo in two different simulations of the
bedo. (The small gradient in surface albedo is due to the LGM. Our global average aP,SURF changes also include
angle of incidence dependence of ocean albedo.) Relative changes in surface albedo away from the ice sheets as-
to the control PI climate simulated by the same model, sociated with a reduction in global sea level; we estimate
15 AUGUST 2011 DONOHOE AND BATTISTI 4413

FIG. 10. (a) Zonal average a in the GFDL simulations of the PI (black) and AQUA (green). (b) Zonal average
aP,ATMOS, and aP,SURF in the GFDL simulations of the PI (black) and AQUA (green). (c),(d) As in (a) and (b), but
for the NCAR simulations of the PI (black), QUAD (red), and LGM (blue).

this effect causes our global average aP,SURF changes to simulation in which the subtropics are relatively cloud
be approximately 20% greater than that reported by T07. free during all seasons.
In the AQUA simulation, the midlatitude and high- It is worth discussing how the partitioning of plan-
latitude total planetary albedo and aP,SURF decreases etary albedo in the altered climate state experiments
(relative to the PI simulations) by approximately one-third
of the surface albedo change in the same regions. In the TABLE 4. Change in hemispheric average planetary albedo and
global average, however, the decrease in planetary al- the partitioning of the changes into atmospheric (aP,ATMOS) and
surface (aP,SURF) contributions in the altered climate state simu-
bedo (relative to the GFDL PI simulation) is due to lations. Each difference is taken relative to the PI simulation in the
nearly equal decreases in aP,SURF and aP,ATMOS (Figs. same model used to simulate the altered climate state (and at the
4a,b and Table 4). Further inspection (Fig. 10b) shows same resolution). The last two rows show the CMIP3 intermodel
that the decrease in hemispheric average aP,ATMOS in spread (2s) of the same variables.
the AQUA simulation is almost entirely due to reduced Hemispheric average aP (%)
atmospheric reflection in the subtropics. This surprising
Model Hemisphere aP aP,ATMOS aP,SURF
result is due to differences in the seasonal migration of
the ITCZ between the AQUA and PI runs; the ITCZ QUAD–PI NH 20.2 10.4 20.6
SH 20.3 10.1 20.4
remains on the equator during all seasons in the AQUA
LGM–PI NH 12.2 10.6 11.6
simulation, while the ITCZ migrates into the summer SH 11.9 11.1 10.8
hemisphere in the PI simulation. This causes the sub- AQUA–PI NH 24.2 21.8 22.4
tropics to be cloudier during the sunny season in each SH 23 21.5 21.5
hemisphere in the PI climate resulting in a larger annual CMIP 2s NH 1.8 2.7 1.4
SH 1.8 2.1 1.1
average planetary albedo compared to the AQUA
4414 JOURNAL OF CLIMATE VOLUME 24

would change if our partitioning methodology was with the spread in surface albedo despite considerable
systematically underestimating aP,SURF relative to the intermodel differences in surface albedo.
‘‘true’’ surface contribution to planetary albedo; in this Surface albedo decreases with global warming (2 3 CO2
case, aP,ATMOS would be biased high relative to its runs) in all models but the magnitude of the change varies
‘‘true’’ value [by Eq. (6)] especially over the cryospheric significantly. However, the intermodel spread in the change
regions. If our aP,SURF values were biased low, we would in surface albedo is poorly correlated with the change in
expect to see large changes in aP,ATMOS collocated with planetary albedo. The planetary albedo changes asso-
the regions of large changes in surface albedo (and with ciated with global warming were found to be primarily
the same sign as the surface albedo change). This does due to changes in cloud albedo (93% of the inter-
not occur. In the AQUA simulation, the changes in model spread). This result is unsurprising given that
aP,ATMOS are very small (except in the subtropics where cloud albedo plays the dominant role in setting the
the changes in surface albedo are also small, Fig. 10). In planetary albedo in the unperturbed climate and that
the QUAD simulation the changes in aP,ATMOS and the surface albedo’s impact on the TOA radiative
surface albedo have opposite signs in the cryosphere budget is strongly attenuated (approximately three-
region; aP,ATMOS increases in the high latitudes where fold) by the atmosphere.
the surface albedo has decreased relative to the PI and Experiments with prescribed and severely altered
LGM simulations with the same model (Fig. 10). Thus, surface albedo distributions associated with the Last
the partitioning of planetary albedo in the altered cli- Glacial Maximum and an aquaplanet demonstrated that
mate states suggests that the aP,SURF values calculated changes in aP,SURF were approximately one-third the
by our methodology are not biased low. magnitude of the surface albedo changes because of
atmospheric attenuation. Furthermore, the changes in
planetary albedo in these experiments were due to
5. Summary and discussion
nearly equal magnitude contributions from changes in
We have partitioned the earth’s planetary albedo into cloud properties and the prescribed changes in surface
a component due to the reflection of incoming radiation albedo. This suggests that cloud feedbacks in greatly
by objects in the atmosphere aP,ATMOS and a compo- altered climate states are likely to play as important of
nent due to reflection at the surface aP,SURF. In the a role in setting planetary albedo as even zeroth-order
global average, the vast majority (88%) of the observed changes in surface albedo (Eastman and Warren 2010).
planetary albedo is due to aP,ATMOS. We demonstrate The most surprising result from this study is that the
that the primary conclusions reached in this manuscript differences in planetary albedo in climates that have
are robust to uncertainties in the observational datasets severely altered surface albedo are relatively modest
(appendix A) and the assumptions made in our simpli- compared to the spread in planetary albedo in the PI
fied radiative transfer model (appendix B). The CMIP3 simulations of the CMIP3 models. For example, the
PI ensemble intermodel average planetary albedo is also difference in hemispheric average planetary albedo be-
primarily due to aP,ATMOS (87%). The intermodel spread tween the simulations of the PI world and the LGM is
in global average planetary albedo is large, corresponding less than the spread in the PI simulations using the
to radiative differences at the top of the atmosphere (2s 5 CMIP3 models. Similarly, the CMIP3 ensemble average
5.5 W m22) that exceed the radiative forcing of doubling change in planetary albedo due to global warming in the
carbon dioxide. Intermodel differences in aP,ATMOS ex- 2 3 CO2 simulations is less than 30% of the intermodel
plain the majority (.60%) of the intermodel spread in spread in simulations of the present day climate. These
hemispheric average planetary albedo. results follow because planetary albedo is largely de-
Surface albedo makes a small contribution to the termined by atmospheric processes (cloud reflection)
global average planetary albedo because atmospheric and there are large intermodel differences in cloud dis-
processes attenuate the contribution of surface albedo tributions in simulations of the PI climate. In contrast,
to planetary albedo by a factor of approximately 3. The planetary albedo is largely insensitive to surface pro-
atmosphere attenuates aP,SURF relative to the surface cesses. Thus, changes in surface albedo make a small
albedo because the atmosphere is somewhat opaque to contribution to planetary albedo changes relative to the
solar radiation and thus attenuates the amount of large uncertainties in cloud distribution changes.
shortwave radiation incident on the TOA that reaches Our analysis has focused on global and hemispheric
the surface and the amount of shortwave radiation re- average planetary albedo. We note that surface albedo,
flected at the surface that is transmitted back to the aP,SURF, and aP,ATMOS changes associated with global
TOA. As a consequence, the CMIP3 PI ensemble in- warming and paleoclimate states have a spatially non-
termodel spread in planetary albedo is poorly correlated uniform structure, often peaking at the high latitudes.
15 AUGUST 2011 DONOHOE AND BATTISTI 4415

The meridional structure of planetary albedo is in-

extricably linked to the equator-to-pole temperature
gradient and the total heat transport in the climate sys-
tem (Stone 1978) and these issues are currently under
further investigation.

Acknowledgments. We acknowledge the modeling

groups, the Program for Climate Model Diagnosis and
Intercomparison (PCMDI) and the WCRP’s Working
Group on Coupled Modelling (WGCM) for their roles
in making available the WCRP CMIP3 multimodel
dataset. Support of this dataset is provided by the Office
of Science, U.S. Department of Energy. Dargan Frierson
ran the seasonal aquaplanet simulation utilized in this
work and provided feedback on the discussion in this
text. Camille Li ran and provided the model output for FIG. A1. Zonal average aP, aP,ATMOS, aP,SURF, and calculated
the LGM and QUAD CO2 runs. Stephen Warren’s ad- from the observations and their uncertainties (2s) estimated from
vice on the observational datasets used and comments the Monte Carlo simulations described in the text. The shaded
were greatly appreciated. Tra Dinh and Tom Ackerman areas are the uncertainty if the observational errors are random
and uncorrelated at each grid point. The dashed lines are the un-
provided a radiative transfer model that aided in veri-
certainty if the errors are systematic and perfectly correlated at all
fying our methodology and were also instrumental aids locations in each zonal band.
in understanding the radiative components of this work.
This work was inspired by conversations with Gerard Roe,
Kyle Armour, Ed Wrigglesworth Blanchard IV, Brian that our partitioning algorithm is sufficiently linear over
Rose, John Marshall, Valerio Lucarini, Rowan Sutton, the range of the uncertainty in the observed radiative
Jonathan Gregory, and Patrick Taylor and benefitted from fluxes. The standard deviation of the aP,ATMOS over the
the critiques of three anonymous reviewers. 100 realizations is approximately 5% of the average
aP,ATMOS over most of the domain, while the standard
APPENDIX A deviation of aP,SURF is approximately 10% of the average
aP,SURF. The standard deviation of both aP,ATMOS and
aP,SURF is significantly larger (reaching 25% of the av-
Uncertainties in Observational Estimates of
erage value) over the Greenland Ice Sheet and East
aP,ATMOS and aP,SURF
Antarctica where the surface albedo is large and the
The satellite-derived TOA and surface shortwave higher-order reflections contribute to the partitioning
fluxes used in this study are both uncertain. Here we algorithm (not shown).
analyze how the uncertainty in these fields propagates to We propagate the uncertainty in aP,ATMOS and
uncertainties in aP,SURF, aP,ATMOS, and aP. aP,SURF onto the zonal averages (Fig. A1) with error
The climatological CERES TOA shortwave fluxes bars representing two standard deviations. If we assume
have a total uncertainty of order 6 W m22 or about 2% that the uncertainties are random, and uncorrelated at
of the incident radiation (Loeb et al. 2009); the error in adjacent grid points, the uncertainties in the zonal av-
the CERES surface shortwave fluxes is approximately erage aP,ATMOS and aP,SURF (shaded areas) are less than
5% (Kratz et al. 2010). We propagate these errors to 0.008 at all locations. In this case the standard deviation
obtain error estimates in aP,ATMOS and aP,SURF using a of the zonal average is approximately equal the standard
Monte Carlo simulation. At each grid point, we add a deviation at a given grid point divided by the square root
random error to the observed flux; the error is assumed to of the number of zonal grid points (zonal averaging re-
be Gaussian with a standard deviation of 2% of the ob- duces the uncertainty by more than a factor of 10). On
served flux for the TOA fluxes and 5% of the observed the other hand, if the errors are systematic and have the
flux for the surface fluxes. We then take 100 random re- same sign at all zonal locations, then the uncertainty in
alizations of the radiative fluxes and calculate aP,ATMOS the zonal mean is equal to the zonal average uncertainty
and aP,SURF using the methodology outlined in section 2. (dashed lines in Fig. A1 give 2s of this quantity). The
The aP,ATMOS and aP,SURF maps averaged over the true zonal average uncertainty is most likely somewhere
100 realizations are statistically indistinguishable from in between these two estimates; observational errors at
the maps calculated using the original data, suggesting one grid point are expected to be correlated with those
4416 JOURNAL OF CLIMATE VOLUME 24

at their adjacent grid points with a spatial decorrelation are not depleted in a single pass through the atmosphere.
scale that is regional (i.e., smaller than the zonal cir- In this section, we explore the sensitivity of our results to
cumference assumed if the errors are systematic). these differing assumptions.
If the observational errors are systematic, the two- We apply the methodology of T07 to the observa-
standard-deviation range of the global average tional and model datasets. We further illuminate the
aP,ATMOS is 0.244–0.280 (82%–94% of the total plane- effects of the differing assumptions 2 and 3 by per-
tary albedo); the two-standard-deviation range of global forming the planetary albedo partitioning with a model
average aP,SURF is 0.028–0.045 (9%–15% of the total that assumes absorption occurs during the first pass
planetary albedo). through the atmosphere only, but that the absorption
and reflection occur at the same level of the atmosphere.
APPENDIX B These three different models for partitioning planetary
albedo are summarized in Table B1.
The partitioning of planetary albedo into aP,ATMOS
Sensitivity of aP,ATMOS and aP,SURF Partitioning
and aP,SURF using each of these models is shown in
Methodology
Fig. B1 for both the observations and the ensemble av-
The partitioning of planetary albedo into surface and erage of the CMIP3 PI simulations. The model used in
atmospheric components presented in this paper is this study and the T07 model are in very close agreement
contingent on the assumptions underlying the simple in the partitioning of the observed planetary albedo at all
single-layer shortwave radiative transfer model (Fig. 1 latitudes. The T07 model finds lower values of aP,ATMOS
and section 2b). Specifically, we assume that 1) absorp- (and consequentially, higher values of aP,SURF) than the
tion and reflection of radiation is isotropic, 2) atmo- model used in this study at all latitudes. The average
spheric absorption occurs during each pass through the difference in aP,ATMOS estimated using the two models
atmosphere, and 3) absorption and reflection of short- is approximately 3% of the average aP,ATMOS and does
wave radiation occurs at the same vertical level of the not exceed 10% at any latitude. Our model finds that
atmosphere. T07 partitioned the planetary albedo 88% of the global average planetary albedo is due to
feedback between atmospheric and surface processes aP,ATMOS, whereas the T07 model finds a value of 85%
using a similar conceptual model except that they as- (Table B1). The single absorption model finds smaller
sumed that the absorption of shortwave radiation in the values of aP,ATMOS (and larger values of aP,SURF) than
atmosphere occurred only during the first downward the other two models; on average, the single absorption
pass through the atmosphere (differs from our assump- aP,ATMOS values are 4% smaller than the Taylor model
tion 2) and that the absorption occurs above the level of and 9% smaller than the values found in this study.
atmospheric reflection (differs from our assumption 3; The three methodologies yield similar results for the
see their Fig. 1). Specifically, T07 assume that down- partitioning of planetary albedo between aP,ATMOS and
welling shortwave radiation in the atmospheric absorp- aP,SURF in the CMIP3 PI ensemble average. The
tion bands is entirely depleted prior to arriving at the smallest aP,ATMOS values are obtained using the single
cloud-top level. (These assumptions allow the parti- absorption model and the model used in this study finds
tioning to be performed analytically.) In contrast, we the highest aP,ATMOS (Fig. B1).
assume that shortwave radiation reflected from the Although all three models give similar results for the
surface also experiences some absorption on its upward partitioning of aP, the three models disagree on the
path toward the TOA because calculations using a ra- magnitude of the atmospheric attenuation of surface
diative transfer model (Kato et al. 1999) show that the albedo K. Our model finds that the atmosphere atten-
atmospheric absorption of solar radiation increases as uates the surface albedo’s contribution to the planetary
the surface albedo is increased from 0 to 1, indicating albedo by 69% in the global average, whereas the Taylor
that solar radiation in the atmospheric absorption bands and single absorption models find values of 61% and

TABLE. B1. Summary of the different simplified radiative transfer models used to calculate aP,ATMOS and aP,SURF.

Number of passes with Vertical location Global average Global average

Model name atmospheric absorption of absorption aP,ATMOS aP,SURF
This study Infinite (all passes) Collocated with reflection 0.262 (88%) 0.036 (12%)
T07 1 Above reflection 0.253 (85%) 0.045 (15%)
Single absorption 1 Collocated with reflection 0.246 (83%) 0.052 (17%)
15 AUGUST 2011 DONOHOE AND BATTISTI 4417

FIG. B1. (a) Planetary albedo partitioning in the observations using the model described in this study, the single
absorption model, and the model of T07. (b) Surface albedo and surface contribution to planetary albedo using the
three methods of planetary albedo partitioning. (c),(d) As in (a) and (b), but for the CMIP3 intermodel average.

57%. The differences in atmospheric attenuation be- (neglecting the effect of multiple reflections) and is
tween the three models are largest in the dry subtropics constrained by the input radiative data used to partition
where atmospheric reflection is relatively small and the planetary albedo. In our model, the atmospheric
thus absorption contributes to a larger fraction of the transmissivity is 1 2 R 2 A, whereas in Taylor’s model1
atmospheric opacity. Our model determines signifi- it is (1 2 RT)(1 2 AT), where the subscripts have been
cantly larger values of attenuation in this region relative added to denote that the R and A values determined by
to the T07 and single absorption models because ab- the two models are not equivalent. Expanding out the
sorption increases the atmospheric opacity to upwelling opacity in T07’s model, and knowing that values for A
shortwave radiation in our model and thus a smaller and AT are similar (although A is always greater than
fraction of the radiation reflected by the surface escapes or equal to AT), one can show that RT . R. This is be-
to space. cause the radiation reaching the cloud top has been
Although an evaluation of which model most accu- depleted by absorption in the T07 model (and not in our
rately captures the radiative behavior of the atmosphere model). Hence, clouds in the T07 model must reflect a
is beyond the scope of the present work, we will briefly larger fraction of the radiation incident at cloud top to
comment on the source of differences in the results satisfy the other radiative constraints imposed by the input
obtained using the three models, and why the T07 model data. The opacity to upwelling radiation in our model
yields results that are in closer agreement with our re-
sults than those using the single absorption model. The
ratio of downwelling radiation at the surface and at the 1
T07 call this quantity m(1 2 g), where m is (1 2 AT) and g is
TOA is roughly equal to the atmospheric transmissivity reflection.
4418 JOURNAL OF CLIMATE VOLUME 24

is 1 2 R 2 A, whereas in the T07 model it is 1 2 RT. The for a shortwave radiative transfer model. J. Quant. Spectrosc.
exclusion of absorption in atmospheric opacity to up- Radiat. Transfer, 62, 109–121.
——, N. G. Loeb, P. Minnis, J. A. Francis, T. P. Charlock, D. A.
welling radiation in the T07 model is partially balanced by
Rutan, E. E. Clothiaux, and S. Sun-Mack, 2006: Seasonal and
the fact that RT . R. As a consequence, our model and interannual variations of top-of-atmosphere irradiance and
the T07 model produce results that are in close agreement. cloud cover over polar regions derived from CERES data set.
Geophys. Res. Lett., 33, L19804, doi:10.1029/2006GL026685.
REFERENCES Kim, D., and V. Ramanathan, 2008: Solar radiation and radiative
forcing due to aerosols. J. Geophys. Res., 113, D02203,
Barkstrom, B., E. Harrison, G. Smith, R. Green, J. Kibler, and doi:10.1029/2007JD008434.
R. Cess, and The ERBE Science Team, 1989: Earth Radiation Kratz, D. P., S. K. Gupta, A. C. Wilber, and V. E. Sothcott, 2010:
Budget (ERBE) archival and April 1985 results. Bull. Amer. Validation of the CERES edition 2B surface-only flux algo-
Meteor. Soc., 70, 1254–1262. rithms. J. Appl. Meteor. Climatol., 49, 164–180.
Bender, F. A.-M., 2011: Planetary albedo in strongly forced cli- Li, C., and D. S. Battisti, 2008: Reduced Atlantic storminess during
mate, as simulated by the CMIP3 models. Theor. Appl. Cli- Last Glacial Maximum: Evidence from a coupled climate
matol., doi:10.1007/s00704-011-0411-2, in press. model. J. Climate, 21, 3561–3579.
——, H. Rodhe, R. J. Charlson, A. M. L. Ekman, and N. Loeb, Loeb, N. G., B. A. Wielicki, D. R. Doelling, G. L. Smith, D. F.
2006: 22 views of the global albedo—Comparison between 20 Keyes, S. Kato, N. Manalo-Smith, and T. Wong, 2009: To-
GCMs and two satellites. Tellus, 58A, 320–330. wards optimal closure of the earth’s top-of-atmosphere radi-
Bony, S., and Coauthors, 2006: How well do we understand climate ation budget. J. Climate, 22, 748–766.
change feedback processes? J. Climate, 19, 3445–3482. Meehl, G. A., C. Covey, T. Delworth, M. Latif, B. McAvaney,
Budyko, M. I., 1969: The effect of solar radiation variations on the J. F. B. Mitchell, R. J. Stouffer, and K. E. Taylor, 2007: The
climate of the Earth. Tellus, 21, 611–619. WCRP CMIP3 multi-model dataset: A new era in climate
Cahalan, R. F., W. Ridgway, W. J. Wiscombe, T. L. Bell, and J. B. change research. Bull. Amer. Meteor. Soc., 88, 1383–1394.
Snider, 1994: The albedo of fractal stratocumulus clouds. Minnett, P. J., 1999: The influence of solar zenith angle and cloud
J. Atmos. Sci., 51, 2434–2460. type on cloud radiative forcing at the surface in the Arctic.
Chou, M. D., and K. T. Lee, 1996: Parameterizations for the ab- J. Climate, 12, 147–158.
sorption of solar radiation by water vapor and ozone. J. Atmos. Moberg, A., D. M. Sonechkin, K. Holmgren, N. M. Datsenko, and
Sci., 53, 1203–1208. W. Karlén, 2005: Highly variable Northern Hemisphere tem-
Curry, J. A., W. B. Rossow, D. Randall, and J. L. Scramm, 1996: peratures reconstructed from low- and high-resolution proxy
Overview of Arctic cloud radiation characteristics. J. Climate, data. Nature, 433, 613–617, doi:10.1038/nature03265.
9, 1731–1762. Qu, X., and A. Hall, 2005: Surface contribution to planetary albedo
Eastman, R., and S. G. Warren, 2010: Interannual variations of variability in the cryosphere regions. J. Climate, 18, 5239–5252.
Arctic cloud types in relation to sea ice. J. Climate, 23, 4216– Roesch, A., 2006: Evaluation of surface albedo and snow cover in
4232. AR4 coupled climate models. J. Geophys. Res., 111, D15111,
Enderton, D., and J. Marshall, 2009: Controls on the total dy- doi:10.1029/2005JD006473.
namical heat transport of the atmosphere and oceans. J. At- Rutan, D. A., F. G. Rose, N. M. Smith, T. P. Charlock, 2001:
mos. Sci., 66, 1593–1611. Validation data set for CERES surface and atmospheric radia-
Fasullo, J. T., and K. E. Trenberth, 2008a: The annual cycle of the tion budget (SARB). WCRP/GEWEX Newsletter, Vol. 11,
energy budget. Part I: Global mean and land–ocean ex- International GEWEX Project Office, Silver Spring, MD, 11–12.
changes. J. Climate, 21, 2297–2312. Solomon, S., D. Qin, M. Manning, M. Marquis, K. Averyt, M. M. B.
——, and ——, 2008b: The annual cycle of the energy budget. Part Tignor, H. L. Miller Jr., and Z. Chen, Eds., 2007: Climate
II: Meridional structures and poleward transports. J. Climate, Change 2007: The Physical Science Basis. Cambridge Uni-
21, 2313–2325. versity Press, 996 pp.
Gorodetskaya, I. V., M. A. Cane, L.-B. Tremblay, and A. Kaplan, Stone, P. H., 1978: Constraints on dynamical transports of energy
2006: The effects of sea ice and land snow concentrations on on a spherical planet. Dyn. Atmos. Oceans, 2, 123–139.
planetary albedo from the Earth Radiation Budget Experi- Taylor, K. E., M. Crucifix, P. Braconnot, C. D. Hewitt, C. Doutriaux,
ment. Atmos.–Ocean, 44, 195–205. A. J. Broccoli, J. F. B. Mitchell, and M. J. Webb, 2007: Estimating
Hahn, C. J., and S. G. Warren, 2003: Cloud climatology for land shortwave radiative forcing and response in climate models.
stations worldwide, 1971–1996. Numerical data package NDP- J. Climate, 20, 2530–2543.
026D, Carbon Dioxide Information Analysis Center, 35 pp. Trenberth, K. E., and J. T. Fasullo, 2009: Global warming due to
Hall, A., 2004: The role of surface albedo feedback in climate. increasing absorbed solar radiation. Geophys. Res. Lett., 36,
J. Climate, 17, 1550–1568. L07706, doi:10.1029/2009GL037527.
Hartmann, D. L., 1994: Global Physical Climatology. Academic ——, ——, and J. Kiehl, 2009: Earth’s global energy budget. Bull.
Press, 411 pp. Amer. Meteor. Soc., 90, 311–324.
Hummel, J. R., and R. A. Reck, 1979: A global surface albedo Wielicki, B. A., R. D. Cess, M. D. King, D. A. Randall, and E. F.
model. J. Appl. Meteor., 18, 239–253. Harrison, 1995: Mission to planet Earth—Role of clouds and
Kang, S. M., I. M. Held, D. M. W. Frierson, and M. Zhao, 2008: The radiation in climate. Bull. Amer. Meteor. Soc., 76, 2125–2153.
response of the ITCZ to extratropical thermal forcing: Idealized ——, B. R. Barkstrom, E. F. Harrison, R. B. Lee, G. Louis Smith,
slab-ocean experiments with a GCM. J. Climate, 21, 3521–3532. and J. E. Cooper, 1996: Clouds and the Earth’s Radiant En-
Kato, S., T. P. Ackerman, J. H. Mather, and E. Clothiaux, 1999: ergy System (CERES): An Earth Observing System experi-
The k-distribution method and correlated-k approximation ment. Bull. Amer. Meteor. Soc., 77, 853–868.