Unified Treatment of Thermodynamic and Optical Variability in a
Document Sample
1 JULY 2003 JEFFERY AND AUSTIN 1621
Unified Treatment of Thermodynamic and Optical Variability in a Simple Model of
Unresolved Low Clouds
CHRISTOPHER A. JEFFERY AND PHILIP H. AUSTIN
Atmospheric Sciences Programme, University of British Columbia, Vancouver, British Columbia, Canada
(Manuscript received 16 May 2001, in final form 6 January 2003)
ABSTRACT
Comparative studies of global climate models have long shown a marked sensitivity to the parameterization
of cloud properties. Early attempts to quantify this sensitivity were hampered by diagnostic schemes that were
inherently biased toward the contemporary climate. Recently, prognostic cloud schemes based on an assumed
statistical distribution of subgrid variability replaced the older diagnostic schemes in some models. Although
the relationship between unresolved variability and mean cloud amount is known in principle, a corresponding
relationship between ice-free low cloud thermodynamic and optical properties is lacking. The authors present
a simple, analytically tractable statistical optical depth parameterization for boundary layer clouds that links
mean reflectivity and emissivity to the underlying distribution of unresolved fluctuations in model thermodynamic
variables. To characterize possible impacts of this parameterization on the radiative budget of a large-scale
model, they apply it to a zonally averaged climatology, illustrating the importance of a coupled treatment of
subgrid-scale condensation and optical variability. They derive analytic expressions for two response functions
that characterize two potential low cloud feedback scenarios in a warming climate.
1. Introduction qt q s , is a particularly acute problem (Manabe and
Wetherald 1967).
Statistical cloud schemes have a long history that The Sommeria–Deardorff–Mellor (SDM) statistical
dates back to the pioneering work of Sommeria and cloud scheme introduces a stochastic subgrid variable
Deardorff (1977) and Mellor (1977). Large-scale at- s that represents unresolved fluctuations in q s q t and
mospheric models typically contain temperature, pres- is assumed to be normally distributed.1 The variance of
sure, and total water (vapor liquid) fields that evolve s, 2, in more sophisticated schemes can be diagnosed
s
according to prescribed dynamical and thermodynami- from a turbulence model (Ricard and Royer 1993) or
cal equations. Traditionally these numerical models from neighboring cells (Levkov et al. 1998; Cusack et
would assign, for each field, a single average value to al. 1999) but, in practice, is often taken as a prescribed
an individual grid cell, thereby ignoring any variability fraction (Smith 1990) of q s . A key assumption in the
2
within the cell. The relative importance of this neglected SDM scheme is that each grid cell is assumed to contain
variability is, not surprisingly, scale dependent; for a complete ensemble of s from which the statistics of
large-scale climate models with grid spacings of 250 unresolved cloud are calculated, regardless of the size
km or greater the unresolved variability can be a sub- of the grid or the time step of the model. For example,
stantial fraction of the mean value (Barker et al. 1996). the mean liquid water to some power p, q lp , in the cloudy
Furthermore, the relative importance of subgrid vari- region of a cell is given by
ability is magnified manyfold by the presence of con-
densation, which is a small difference in two relatively qt qs
large scalar quantities: the saturation vapor density, q s , q lp Ad 1 p
a L (q t qs s) p Ps (s) ds
and the cell’s total vapor (or water) density, q t , prior to
condensation. Early climate modelers were well aware qt qs
that the use of ‘‘all-or-nothing’’ condensation schemes, Ad Ps (s) ds, (1)
whereby an individual grid cell is either completely
clear or completely cloudy depending on the difference
where Ps is the probability distribution function (e.g.,
Gaussian) of s; the cloud density, Ad , is the fraction of
Corresponding author address: Christopher A. Jeffery, Los Ala-
mos National Laboratory (NIS-2), P.O. Box 1663, Mail Stop D-436,
Los Alamos, NM 87545. 1
This notation differs from Mellor (1997), where s represents fluc-
E-mail: cjeffery@lanl.gov tuations in q t q s.
2003 American Meteorological Society
1622 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 60
grid cell occupied by cloud; and aL 1 is a parameter tion of sky covered by cloud when viewed from below.
that accounts for the subadiabatic liquid water profiles Thus, for example, an increase in q l (or ) caused by
typically observed in layer clouds. In what follows an increasing temperatures does not necessarily imply an
overbar is reserved to represent an average over the cloudy increase in R if A c decreases, producing an optically
fraction of a cell or column of cells, brackets · represent thicker cloud field with smaller cloud fraction. The cou-
a spatial average over the entire cell/column, and unre- pling of our statistical approach to surface temperature
solved variability in each cell is assumed to be centered allows us to investigate the combined ( A c / T, / T)
(i.e., have zero mean). response within an analytic framework.
Statistical cloud schemes, in their current form, pro- The first theoretical study, and one of the only studies
vide complete information about q l but only limited in- to date, that investigates the coupled ( A c , ) response
formation on cloud optical properties. This is because of a cloud layer to increasing temperature ( T) while
optical depth, , is a vertical integral of q lp (z) from cloud holding, alternatively, both A c and fixed is that by
base to cloud top and variability in q l has nonzero spatial Temkin et al. (1975). Temkin et al. (1975) compared
correlations produced by turbulence, that is, s(z1 )s(z 2 ) and contrasted the temperature sensitivity of A c at fixed
0. Thus while the SDM scheme does provide grid- , ( A c / T) , with the temperature sensitivity of at
column-averaged optical depth , it does not provide fixed A c , ( / T)Ac in a simplified atmosphere with one
or higher-order moments without further assumption. cloud layer and constant surface relative humidity (RH).
In section 2 below we link thermodynamic and optical They found ( A c / T) 0 and ( / T)Ac 0 using a
variability in the SDM scheme by first restricting s and nonstochastic model. These results indicate a negative
hence P s to be height independent in low clouds. At the low cloud feedback (LCF), where LCF is defined as the
same time, we consider a distribution of cloud-top change in the net downward shortwave flux at the top
height fluctuations (z top) that is distinct from a height- of the cloud layer produced by a positive temperature
independent P s . The resulting low-dimensional model change.
is analytically tractable and requires only the specifi- There have been numerous modeling and observa-
cation of the z-independent joint z top -s distribution func- tional studies that suggest a range of values for the sign
tion to completely determine P (Jeffery 2001). and magnitude of this cloud radiative response. For ex-
Our approach extends the results of Considine et al. ample, a negative low cloud ( ) feedback at high lati-
(1997), who showed that normally distributed cloud tudes is also implied by the positive liquid water sen-
thickness fluctuations can produce distributions of in- sitivity q l / T found by Somerville and Remer (1984)
in Russian aircraft measurements, assuming a negligible
tegrated cloud liquid water path (LWP) (or, equivalently
cloud thickness sensitivity. In contrast, Schneider et al.
in their approximation, optical thickness) that qualita-
(1978) suggested that warming leads to increased con-
tively match Landsat satellite cloud observations for a
vection and vertical q t transport and a resulting atmo-
range of cloud fractions. Our approach also builds upon
sphere that is unable to increase q t sufficiently to main-
the recent work of Wood and Taylor (2001), who linked
tain constant RH. A relative drying of the lower at-
s and P but did not consider z top. Below we derive mosphere with warming implies that global cloud feed-
general forms for LWP(s, z top) and (s, z top) in a layer back may not be negative. Support for a positive cloud
with horizontally fluctuating cloud top and cloud base. feedback is provided by the global climate model
We also adopt a radiation parameterization that incor- (GCM) study of Hansen et al. (1984), who found that
porates P into the calculation of longwave and short- clouds contribute a feedback of 1.5 C—nearly 1 C
wave fluxes. due to a reduction in low clouds—resulting in a net
Our analytic expressions for allow us to combine climate sensitivity double that found in an earlier study
fluctuations in s and z top into a single subgrid variable with fixed clouds (Manabe and Stouffer 1979). By 1990,
s*, and we examine the radiative response of the sta- all 19 of the GCMs compared by Cess et al. (1990)
tistical cloud scheme to changes in the variance of s* predicted a decrease in globally averaged A c with in-
and hence the optical thickness distribution P . We creasing temperature, although the sign and magnitude
choose a form for P s* suitable for large-scale models of the net cloud feedback varies considerably from mod-
and similar to the triangle distribution of Smith (1990), el to model. More recently, Tselioudis et al. (1993) an-
adopting his choice for the temperature dependence of alyzed global satellite observations of low cloud and
unresolved variability, 2 (T). With this modeled
* qs
2
s found a generally negative optical depth sensitivity, /
coupling of P to the surface temperature through q s , T, (positive cloud feedback) which increases from the
we use the parameterization to investigate the change midlatitudes to the Tropics. A positive low/midlatitude
in net mean cloud reflectivity for a specified temperature low cloud ( ) feedback is also implied by the temper-
change, given an idealized climatology. ature dependence of satellite observations of liquid wa-
Understanding the response of cloud-layer reflectivity ter path (Greenwald et al. 1995).
to increasing temperature is complicated by the fact that In this article we expand on the approach of Temkin
the total reflectivity ( R A c R ) of a cloud layer is a et al. (1975) and calculate analytic response functions
nonlinear function of and cloud fraction A c : the frac- for our statistical treatment of low cloud optical vari-
1 JULY 2003 JEFFERY AND AUSTIN 1623
ability. We find ( A c / T) 0 in contradistinction with {A} H : {A 0} H 0, {A 0} H A is a Heaviside
Temkin et al. (1975). This result links the observational bracket and we have used z bot w (q 0
1
q t s) from
evidence of a largely negative sensitivity (Tselioudis assumption 2. The key feature of Eqs. (2) and (3) is
et al. 1993; Greenwald et al. 1995) with GCM simu- that fluctuations in z bot are defined by inverting q l (z bot ,
lations that predict a negative A c response (Cess et al. s) 0, whereas the unresolved cloud-top height fluc-
1990) as we discuss in section 4. tuations z top are absorbed into the new subgrid variability
The article is organized as follows. In section 2 we s*. Thus our distribution Pz top can, in principle, be com-
derive our statistical model and compare the predictions bined with P s to give P s* . This is advantageous because
of the model with satellite data taken from Barker et al. unresolved variability of q t , q s , and z top is contained in
(1996). The behavior of our scheme is analyzed in sec- the single parameter s* and P s* is z independent. In
tion 3 using an idealized zonally averaged climatology analogy with the SDM scheme we assume that s* is
and in section 4 we present A c– –T response functions. centered and P s* is known. The rhs of Eq. (2) should
Section 5 contains a summary. not be confused with q lp 1(z top , x)a L 1 w 1/(p 1) from
Eq. (1) since the statistics of s* generally differ from
the statistics of s.
2. Model description
Formulation of the shortwave and longwave optical
Our model of boundary layer cloud optical variability depths follows from Eq. (2) given the appropriate func-
is based on two assumptions: 1) horizontal subgrid var- tional relation # func(q l , . . .) dz. At this point, it
iability in the boundary layer of large-scale models ex- is convenient to introduce the cloud thickness h(x)
ceeds vertical variability; and 2) cloud liquid water in- z top (x) z bot (x) so that Eq. (2) is simply
creases linearly with height above cloud base, that is, z top(x)
(a L )p p 1
w
q s (z) q0 w z where w 0. Assumption 1 is ac- q lp (z, x) dz h (x). (4)
curate for large-scale temperature and moisture fluctu- z bot(x)
p 1
ations because the horizontal length of a grid cell in a The longwave optical depth, because it depends primarily
climate model is much greater than the boundary layer on the LWP, is strictly given by the ‘‘h 2 model’’:
height. However, it does not hold near cloud top where
the vertical dependence of s at the cloud boundary is (x) h 2 (x)
complex. We overcome this deficiency by introducing aL
a distribution of unresolved cloud-top height fluctua- {q z q0 s*(x)} H .
2
(5)
2 w t w top
tions, P z top, that is distinct from a z-independent P s , al-
though s and z top may be correlated. Our second as- In contrast, a number of different formulations exist
sumption is well supported both numerically and ex- for the shortwave optical depth. For example, writing
perimentally in the literature. in terms of LWP and a constant effective radius (reff )
Given assumptions 1 and 2 we are now in a position recovers the h 2 model. In layer clouds a better approach
to calculate the optical statistics of low clouds. For clar- due to Pontikis (1993) is to assume proportionality of
ity and brevity, we first introduce the notation. The var- the effective and volume-averaged radius, reff q1/3,
l
iable dependence (x) labels unresolved horizontal var- giving p 2/3 and a 5/3 dependence of on h [Pontikis
iability, whereas (z) indicates a vertical dependence, 1993, (Eq. 5)]. Thus we have the ‘‘h 5/3 model’’ for short-
which, by assumption 1 is nonstochastic; that is, subgrid wave optical depth:
vertical fluctuations are assumed negligible. Further-
more, (x) represents unresolved variability in a single 3a L 3
2/
(x) {q z q0 s*(x)} H 3 .
5/
(6)
5 w t w top
cell whereas (z) is a continuous dependence that may
extend through a column of cells.
Note that inherent in Eq. (6) is the approximation that
the cloud droplet concentration, N, is independent of
a. Linking Ps and Pz top to P s*, an approximation that is likely accurate if the ma-
jority of low clouds in a grid cell are non- or weakly
Consider the integral of q lp(z, x) from cloud base
precipitating.
z bot (x) to cloud top z top (x) z top z top(x):
The constant of proportionality in Eq. (6) goes as N 1/3
z top(x) (Pontikis 1993; see also appendix C). In what follows we
q lp (z, x) dz ignore the effect of unresolved fluctuations in N on the
z bot(x) statistics of . This approximation is unlikely to be valid
p near large sources of N, for example, major industrial
aL w
1
{q t z
w top q0 p
s*(x)} H 1 , (2) cities, but is justifiable elsewhere because the 5/3 moment
p 1 of ql acts to magnify fluctuations while the 1/3 moment
where of N acts to damp fluctuations. This behavior is illustrated
with the following example. Consider the dependence of
s*(x) s(x) z (x),
w top (3) (Y 0 Y ) on mean value Y 0 and the variance of Y ,
1624 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 60
2
Y , where Y is normally distributed. Writing (Y 0 Y ) from LESs of a cloud-topped boundary layer. Thus we
Y0 Y we find ( 1/3, 0.15) and ( 5/3, find that the variance of s*
0.5) for the range Y Y 0 2 Y . Hence, increasing 2
ql at fixed ql,0 acts to increase (ql,0 ql ) 5/3 as expected q0
but a similar increase in N decreases (N 0 N )1/3 be-
2
s* T { z
w top Ri 1 } 2 . (8)
T
cause of the damping effect of the 1/3 exponent. This result
suggests that the normalized variance of N would have to As mentioned previously s (z) is usually assumed to be
be 3 to 4 times larger than the normalized variance of ql proportional to q s (z), that is, T constant. Assuming
for unresolved droplet number fluctuations to have a com- that ztop and Ri are T independent as well permits a
parable effect on the statistics of . particularly simple form for the temperature dependence
The moments of can be calculated from P s* in anal- of P s* .
ogy with P s in Eq. (1) while A c is given by Recent observational studies (Norris 1998a,b; Bajuk
and Leovy 1998; Chen et al. 2000) have indicated the
q t q s(z top)
importance of cloud type in the analysis of cloud prop-
Ac Ps* (s*) ds*. (7) erties; in principle the ratio s* / s could be parameter-
ized as a function of cloud type diagnosed from various
Intuitively, we might expect the maximum cloud overlap stability and potential energy considerations. On the oth-
assumption, A c max(A d ) A d (z top ), to hold for a er hand, in consideration of the poor boundary layer
model of cloud variability that ignores vertical varia- vertical resolution of typical GCMs and in the absence
tions in s. However, comparing Eqs. (1) and (7) we find of knowledge of s* / s , we follow current GCM pa-
that it does not hold generally since A c is calculated rameterizations and assume s* s q s in section
from P s* and not P s . The failure of the maximum cloud 3. Note that z top is coupled to q s through w in Eq. (3).
overlap assumption is due to the independence of z top Since s* is z independent by definition in our scheme
and s in our approach. Note that the usual cell averaged we will use q s (z) at the surface [i.e., q 0 (T)] to evaluate
quantities, For example, q l (z), are independent of z top the temperature dependence of s* in section 3.
and should be calculated with P s .
c. Radiation
b. Approximations for Ps and Ps* Currently, most GCMs lack the methodology to in-
clude unresolved variability in the calculation of cloud
Knowledge of both P s and P z top, and hence P s* , is
reflectivity (R) or emissivity ( ). This deficiency may
limited. Cloud ensemble (Xu and Randall 1996), large have important implications for the prediction of global
eddy simulations (LES; Cuijpers and Bechtold 1995), cloud feedback discussed above. Through the use of a
and observational studies (Larson et al. 2001) provide statistical cloud scheme [e.g., SDM, Eq. (1)], many
support for a Gaussian P s . Comparatively less is known modern GCMs couple changes in cloud properties [e.g.,
about P z top. Ground-based (Boers et al. 1988; Albrecht ( Ac , )] in a changing climate to the distribution of
et al. 1990) and space-based (Strawbridge and Hoff the subgrid variability s. But they also use the plane-
1996; Loeb et al. 1998) retrievals of z top suggest a stan- parallel homogeneous (PPH) assumption Rpph R( ),
dard deviation of 50–100 m for marine stratus over which decouples the optical properties R and from the
typical GCM length ( 100 km) and time ( 2 h) scales. underlying thermodynamic cloud variability. The con-
A comprehensive analysis of the shape of Pz top is cur- vexity of the functions relating R and to ensures that
rently lacking. the optical bias incurred from the PPH assumption is
Observational (Klein and Hartmann 1993; Oreopou- positive, that is, R and are overestimated. To reduce
los and Davies 1993; Norris and Leovy 1994; Klein et this bias many current GCMs use an effective optical
al. 1995; Bony et al. 1997) studies of the marine bound- depth eff where 0.7 to calculate Rpph (Cahalan
ary layer over relatively long timescales suggest that s et al. 1994). Although may be tuned in a particular
is largely a function of boundary layer temperature, T, GCM to reproduce the measured radiative stream, this
while z top is largely controlled by the jump in potential approach is ad hoc in nature and becomes increasingly
temperature ( ) at the top of the boundary layer. De- inaccurate as the climate departs from its present state.
fining an interfacial Richardson number (Deardorff Below we present a unified treatment of the thermo-
1981) dynamic and optical variability of boundary layer clouds
Ri (g/T )z top /w*
2 based on the SDM scheme.
The utility of Eqs. (5) and (6) is not in the calculation
where w * is Deardorff (1974)’s convective velocity of directly but, rather, in providing a methodology to
scale, Moeng et al. (1999) estimate the standard devi- include subgrid variability in the reflectivity and emis-
ation of z top as sivity. We do this by calculating P ( ) from P s* using
the equations above and a change of variable; then by
z top /z top 0.6 Ri 1
definition
1 JULY 2003 JEFFERY AND AUSTIN 1625
X X( )P ( ) d , (9)
where X R or and P is the distribution of in the
cloudy part of the column. This approach is discussed
in Considine et al. (1997) and Pincus and Klein (2000),
albeit not in the context of a generalized framework of
unresolved variability. Unfortunately, the analytic ex-
pressions for R and are sufficiently unwidely to prevent
an analytic evaluation of Eq. (9).
Another approach, pioneered by Barker (1996), is to
assume an analytically friendly form for P that is both
sufficiently general to approximate P over a wide range
of conditions and that allows a closed-form expression
for X . Barker et al. (1996) analyzed satellite data of
marine low clouds and found that a generalized dis-
tribution, P ( ), closely approximates the observed dis-
tribution and allows Eq. (9) to be integrated analytically.
The last step in our treatment of unresolved optical
variability is to relate P to Eqs. (5) or (6) and P s* . The
2
shape of P is controlled by the parameter / 2,
which is a measure of the width of the distribution rel- FIG. 1. Plot of A c vs for the h 2 and h 5/3 models calculated using
Eqs. (5)–(7), and P s from appendix B. Landsat data in the range
ative to its mean. Barker (1996), who introduced P ( , 6.5 from Table 2 *of Barker et al. (1996) is also shown for com-
), did not relate , and in particular 2 , to the unre- parison.
solved thermodynamic variability s. Using the frame-
work we have presented thus far, we calculate and
2
(and thus ) using Eqs. (5) or (6), and P s* substituted therefore introduce a modified triangle distribution (ap-
for P s in Eq. (1). Analytic expressions for R ( , ) and pendix B) that is similar to Smith’s scheme (or a Gauss-
( , ) used in our analysis in section 3 are given in ian) at large A c but better reproduces Gaussian behavior
Barker (1996) and Barker and Wielicki (1997), respec- at small A c . In particular, our distribution gives P s*
tively. 0 for a wider range of s*, | s* | (35/3)1/2 s* , than
Smith (1990)’s triangle, | s* | (6) 1/2
s . An expression
*
for the arbitrary moment is given in appendix B.
d. Comparison with Landsat data A comparison of A c versus is shown in Fig. 1 for
Our scheme provides a one-to-one relationship be- the h 2 and h 5/3 models calculated using Eqs. (5)–(7), and
tween and A c for a given P s* that can be tested against our new triangle distribution (appendix B). Also shown
the Landsat satellite data compiled by Barker et al. is a subset, ∈ (0, 6.5), of the Landsat data tabulated
(1996). To make such a comparison we need to specify in Table 2 of Barker et al. (1996). Agreement between
the distribution P s* . Considine et al. (1997) assumed a the data and both theoretical models is good despite
normally distributed h—equivalent to a normally dis- uncertainties in the retrieval of A c from the satellite
tributed s* in our formulation—and found that the A c scenes (Barker et al. 1996). It should be noted that ∈
dependence of the LWP distribution predicted by the h 2 (6.5, 12) has a significant impact on the calculation of
model is consistent with Landsat data. Using the (Gauss- R and and is in close agreement with the model, while
ian) Considine model, Wood and Taylor (2001) derived the selected data shown in Fig. 1 emphasize the region
an approximate relationship between LWP and LWP for ∈ (0.5, 3). However, it is encouraging that the as-
large A c and verified this relationship using First ISCCP ymptotic behavior near 0.5 predicted by our model
(International Satellite Cloud Climatology) Regional is consistent with the data.
Experiment data. In appendix A we present exact an-
alytic results for LWP and LWP 2 , and hence LWP , that 3. Low cloud radiative feedback using a simple
are valid for all A c . climatology
A potential disadvantage in assuming a normally dis-
tributed s* is that closed-form expressions for nonin- As a demonstration of the behavior of our statistical
teger moments, for example, the h 5/3 model, are not cloud scheme, we now consider the coupling of unre-
available. The triangle distribution first used by Smith solved variability and cloud feedback in the model of
(1990) is a computationally efficient surrogate for the section 2 using a zonally averaged climatology. We con-
Gaussian distribution that is analytically tractable, but sider only the response of low clouds and we specify a
it does not accurately mimic a Gaussian at small A c . We fixed (2 C) surface temperature perturbation. Since the
1626 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 60
prognosed cloud changes do not feed back into the tem-
perature perturbation our model experiments are an
open-loop study. While our model neglects the merid-
ional structure of cloud amount caused by atmospheric
dynamics, for example, the storm tracks, and meridional
variations in surface properties or s (Rotstayn 1997),
we incorporate what we consider to be the major lati-
tudinal dependencies: solar zenith angle, saturation va-
por density, and surface albedo. We further assume that
droplet number concentration is T independent.
In this section and in the spirit of Temkin et al. (1975),
we consider three different responses of a zonally av-
eraged climatology to a fixed global increase in tem-
perature. In our ‘‘observationally constrained ’’ re-
sponse (CTobs ) model mean optical depth decreases with
increasing T according to a parameterization (appendix
D) of the satellite observations of Tselioudis et al.
(1993). By specifying the sensitivity and assuming
q s (z 0, T) q 0 (T) we then predict A c using
s
*
Eqs. (6) and (7). Following Temkin et al. (1975) we
also consider (i) a ‘‘constrained ’’ response (CT) model
with constant , and (ii) a ‘‘constrained A c’’ response FIG. 2. Comparison of R and with PPH values Rpph R( ) and
(CA) model where A c remains constant and we deter- pph ( ). The convexity of R and ensures that the PPH approx-
imation overestimates R and . Shortwave [Eq. (6)], longwave
mine from our unified scheme. In these calculations [Eq. (5)], and A c [Eq. (7)] are all calculated using P s from appendix
we do not determine q t and z top independently; for ex- B. Expressions for R and are from Barker (1996) *and Barker and
ample, the negative optical depth sensitivity observed Wielicki (1997), respectively. Reflectivities are diurnally averaged at
by Tselioudis et al. (1993) could result from a decrease equinox. See appendix C for parameter values.
in z top despite increasing specific humidity (Tselioudis
et al. 1998). Here our CTobs , CT, and CA model exper-
iments represent three possible scenarios of the climate’s mation used by GCMs. The well-documented plane-
response to increasing temperatures that may have very parallel albedo (reflectivity) bias (Cahalan et al. 1994)
complex dynamical origins and spatial structure. We do of Rpph , roughly 0.06 in our model, is visible in the lower
not make any claims that either CTobs , CT, or CA is a half of the figure, but it is overshadowed by the much
‘‘most probable’’ low cloud response. Rather, we hope larger bias of pph that averages near 0.3. Early studies
to use these experiments to gain some insight into the of feedback (Temkin et al. 1975; Somerville and Re-
relationship between low cloud radiative feedback and mer 1984) assumed that longwave optical properties of
the coupled ( A c , ) response. clouds are saturated ( 1 pph ) and as a result,
First consider our base-state climatology. Our zonally changes in only affect the cloud’s shortwave proper-
averaged model extends over latitudes 60 S to ties. Although, as shown in Fig. 2, is not saturated at
60 N where our ‘‘grid cells’’ encompass one latitudinal global scales in our model, assumptions concerning the
band. We assume that our new triangle distribution (ap- behavior of are not significant for low cloud radiative
pendix B) defines the shape of the distribution P s* that forcing calculations since the longwave forcing is very
represents meridional fluctuations in ‘‘subgrid’’ vari- small. Note that the PPH biases calculated using zonally
ability, and that this form is independent of and T. averaged values of and A c are larger than the biases
Our base-state climatology is constructed from the A c ( ) associated with a typical, partially cloudy GCM grid
measurements of Warren et al. (1988) [See Ramaswamy cell for which the cloud fraction tends to exceed that
and Chen (1993) and Kogan et al. (1997) for a similar of our zonal climatology. Figure 2 reiterates that a cou-
approach.] Mean optical depth ( ) ∈ (3.5,6.2) is es- pled treatment of thermodynamic and optical variability
timated from the satellite measurements presented in can substantially impact the predicted values of low
Hatzianastassiou and Vardavas (1999). Parameter values cloud R and in a GCM cloud parameterization.
are given in appendix C. By specifying A c , , T( ), and We now turn our attention to the modeled response of
q s (T), we solve for the two unknowns s* and q t low cloud properties to warming. Consider a T 2C
w z top . We use the more accurate h model for (h 2 globally uniform warming where the sensitivity (T
5/3
model for longwave ). T) is prescribed to be (mostly) negative in the CTobs
The mean reflectivity and emissivity of our base-state model, (T T) (T) constrains the CT model and
climatology, averaged over the diurnal cycle at equinox, Ac(T) Ac(T T) constrains the CA model. The low
is shown in Fig. 2 along with the corresponding values cloud feedback predicted by the CTobs , CT, and CA models
predicted by the plane-parallel homogeneous approxi- is shown in Fig. 3. As before we define LCF as the change
1 JULY 2003 JEFFERY AND AUSTIN 1627
ative contribution of the A c response has been shaded
in Fig. 3. The shading reveals that the total CTobs feed-
back is dominated by the negative A c response, even in
the Tropics where the change in is largest (Tselioudis
et al. 1993). This behavior is in qualitative agreement
with the recent 2 CO 2 GCM experiments of Tselioudis
et al. (1998, their Fig. 14), which show a relatively small
feedback of 0.2 C compared to the 1.5 C A c feed-
back reported with an older version of the same GCM
(Hansen et al. 1984). The dominance of A c over feed-
back is also in agreement with the regional observational
studies of Oreopoulos and Davies (1993) (Tropics) and
Bony et al. (1997) (subtropics), which imply that the
negative A c response may make a larger contribution to
shortwave low cloud feedback than the negative re-
sponse. Several other observational studies (Klein and
Hartmann 1993; Norris and Leovy 1994; Klein et al.
1995) also provide support for a negative A c response.
4. A c– –T response functions
FIG. 3. Plot showing LCF predicted by the CTobs , CT, and CA Observational studies of cloud fraction sensitivity ( Ac /
models for a uniform 2 C increase in global mean temperature. Since T) and optical depth sensitivity ( / T) are often used to
low clouds cool the earth by reflecting solar radiation, less low cloud
(CTobs and CT) enhances warming (positive cloud feedback) while
provide insight into cloud feedback. As pointed out by
more low cloud (CA) buffers the warming (negative cloud feedback). Arking (1991), the information provided by these studies
The shaded region indicates the dominant contribution of the A c re- is limited because it is not known which parameters are
sponse to the overall CTobs feedback. See appendix D for calculation held fixed and which are allowed to vary. In this section
details. we present analytic response functions, ( Ac / T) and
( / T)Ac, for our subgrid-scale cloud parameterization that
in the net (positive downward) radiative flux at the top of do not suffer from this deficiency. Our use of the termi-
the boundary layer. The calculations employ the diurnally nology ‘‘response function’’ is an analogy to response
averaged equinox R and predicted by our unified ap- functions in the theory of thermodynamics, for example,
proach and the approximation s* (T T) 1 q 0 (T specific heat and adiabatic compressibility of an ideal gas.
T) where 1 s (T)/q 0 (T). The figure illustrates that We compare our results with the earlier study by Temkin
*
the CTobs and CT feedbacks are positive and considerably et al. (1975), discussed in section 1, and assess the impact
larger in magnitude than the negative LCF ( feedback) of coarse vertical resolution on a discrete numerical eval-
of the CA model. Moreover, in the tropical regime 20 uation of these functions.
20 the CTobs cloud feedback—a mixture of neg- Combining Eqs. (6) and (7), s* 1 qs , w 2 qs
ative Ac and response—is as much as 3.5 times as large and Smith’s triangle distribution for subgrid variability
as the CT cloud feedback. The CTobs and CT cloud feed- (Smith 1990), we derive
backs are also generally larger than the 2 CO 2 forcing
of 4 W m 2. However, it is important to emphasize that lnA c 4 1 qs 4 L
q (10)
LCF is not a top-of-the-atmosphere feedback. In particular, T ,
5 s T 5 R T2
modulation of the longwave stream through changes in
high cloud properties could enhance or buffer the net cloud ln 2 L
feedback. , (11)
T 3 R T2
Further analysis (not shown) reveals that LCF is rel- Ac,
atively insensitive to the treatment of unresolved optical valid for Ac 0.5, where { 1, 2 }, L is the latent
variability; LCF computed using plane-parallel homo- heat of vaporization, R is the gas constant for water vapor,
geneous optical properties overestimates the CTobs cloud and recall that qs must be evaluated at some fixed height
feedback by 15% and the CA feedback by 35% com- (e.g., at the surface) since s* and w are z independent
pared to the predicted values shown in Fig. 3. Thus, by definition. For Ac 0.5, Eq. (11) remains valid but
clearly it is the constrained response of our modeled for Eq. (10) ( lnAc / T) , decays monotonically to zero
climate, that is, CTobs vs. CT vs. CA, and not the pa- as Ac → 1. The disappearance of the Ac response as Ac
rameterization of optical variability that determines LCF → 1 reflects the increasing independence of Ac to small
to first order. This finding is consistent with the GCM changes in s* in the limit of vanishing unresolved var-
sensitivity study of Rotstayn (1999), among others. iability. Overall the more general result ( lnAc / T) ,
To further explore the CTobs cloud feedback, the rel- 0 and ( ln / T)Ac, 0 is valid for all Ac.
1628 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 60
We can interpret Eqs. (10) and (11) as representing ticular GCMs tend to underpredict persistent marine
two potential low cloud shortwave feedback scenarios stratocumulus cloud sheets in eastern ocean subsidence
in a warming climate demarcated by / T 0 and A c / regions (Browning 1994; Bushell and Martin 1999).
T 0, respectively. Let F 0 be the net (positive Typically GCMs have only four to six model levels in
downward) shortwave radiative flux reflected by the the boundary layer (BL) and the vertical resolution of
(unforced) low clouds and T 0 be the thermal forc- these levels usually decreases with height. As a result,
ing. Consider the small approximation F A c . Then the top model level in the BL will dominate the discrete
Eq. (10) implies LCF (4/5)F T/T*, a positive cloud integration of q lp [Eq. (2)] and hence . In this low
feedback, while for Eq. (11), LCF (2/3)F T/T*, a resolution limit the h 5/3 model for shortwave optical
negative cloud feedback, where T* R T 2 /L . depth [Eq. (6)] becomes
Although we make no claims regarding the likelihood
of the two scenarios described by Eqs. (10) and (11), (x) {q t z
w top q0 s*(x)} 2/3 z,
the difference in sign of Eqs. (10) and (11) leads to a where z is the thickness of the model level centered
nontrivial asymmetry between the (A c , ) response and at z top . Computing the low vertical resolution response
LCF. Using Eq. (10) we find that ( / T) 0 is a functions we find that the response [Eq. (11)] remains
sufficient condition for a positive LCF while Eq. (11) unchanged while the A c response becomes
implies that ( A c / T) 0 is a sufficient condition for
a negative LCF. These relations follow from the positive lnA c L
coupling between A c and , that is, ( / A c ) T, 0. On 2 , (13)
T , 1, z
R T2
the other hand, ( / T) 0 and ( A c / T) 0 do not
uniquely specify the sign of the LCF. Thus our statistical independent of w . A comparison of Eqs. (10) and (13)
approach links the observational evidence of a largely reveals that RH-based implementations of statistical
negative sensitivity (Tselioudis et al. 1993; Greenwald cloud schemes in low vertical resolution GCMs tend to
et al. 1995; Bony et al. 1997) with GCM simulations overestimate the unresolved low cloud A c response by
(Hansen et al. 1984; Wetherald and Manabe 1986; Col- a factor of 2.5 for A c 0.5, compared to the same
man and McAvaney 1997; Yao and Del Genio 1999) statistical cloud scheme run at higher vertical resolution.
that predict a negative A c sensitivity and a positive LCF.
It is important to note that although these GCMs do not 5. Summary
explicitly use a statistical cloud scheme, their RH-based
grid-cell parameterizations for A d (and hence A c ) are Understanding the complex interaction of clouds, ra-
formally analogous to Eq. (1) for A d with 2 s q s (T).
2
diation, and climate is a formidable challenge; the sign
It is interesting to compare our negative A c response and magnitude of the global cloud feedback remains a
function, Eq. (10), with the result ( A c / T) 0 derived question of concern and debate. In this study we focus
by Temkin et al. (1975) using a nonstochastic model. on one facet of the cloud–climate interaction problem,
In the Temkin et al. (1975) model, the increase in avail- namely, the relationship between the thermodynamic
able liquid water with increasing temperature (recall RH cloud properties A c and and the optical properties R
is fixed) is placed in a formerly clear column that there- and within the context of a statistical cloud scheme.
by increases A c . In contrast, in our statistical approach We restrict our attention to low clouds where the vertical
the cloud thickness decreases in the face of increasing profile of cloud liquid water is linear and where hori-
w resulting in a negative A c sensitivity. zontal variability dominates. Assuming a known distri-
We extend Eq. (11) to another useful form through bution of unresolved variability that includes cloud-top
the approximation ( ln / T)Ac, 2( lnR / T)Ac, valid height fluctuations, we derive a self-consistent and com-
for O(5), giving putationally efficient set of equations for A c and the
moments of , thereby incorporating subgrid optical
ln R 1 L fluctuations into the statistical cloud schemes first in-
. (12)
T 3 R T2 troduced in the 1970s (Sommeria and Deardorff 1977;
Ac,
Mellor 1977). This unified treatment of thermodynamic
Since LCF is relatively insensitive to changes in we and optical variability is particularly well suited for use
can combine Eqs. (10) and (12): in a GCM that incorporates a subgrid-scale turbulence
|(LCF) , | scheme (Ricard and Royer 1993).
2.4, When cloud-top height fluctuations and temperature/
|(LCF) A c , |
moisture fluctuations are treated as a single random var-
which illustrates that, in general, Ac feedback dominates iable, then our model of longwave optical depth (liquid
the feedback in this model. The approximate 1:2.4 LCF water path) reduces to the Considine et al. (1997) model
ratio is illustrated by the CT and CA models in Fig. 3. if this new random variable is normally distributed. This
We can also use our response functions to quantify approach, however, is not always valid. For example, a
the effect of low model vertical resolution on LCF. The minimum large-scale lifting condensation level—break-
representation of low clouds in GCMs is poor; in par- ing the reflection symmetry of cloud-base and cloud-
1 JULY 2003 JEFFERY AND AUSTIN 1629
top height fluctuations—requires that cloud-base and using the h 2 model for LWP [see Eq. (5)], z top 0,
cloud-top height fluctuations be treated distinctly (Jef- Gaussian P s , and assuming small s /q c where q c qt
fery and Davis 2002). Recent improvements in the re- w z top q 0 . Wood and Taylor (2001) state Eq. (A1)
trieval of cloud physical properties using multiple re- is accurate to better than 5% for s /q c 1/2. Below
mote sensors (Clothiaux et al. 2000; Wang and Sassen we present analytic expressions for LWP and LWP 2 ,
2001) should provide more information on the joint sta- and hence LWP , that are valid for all s /q c .
tistics of cloud-base and cloud-top height fluctuations Using Eq. (5) and Gaussian P s we find
that could, in principle, be incorporated into our treat- *
ment of low-cloud optical depth. aL qc s* 2 2
Our unified approach can also be used to probe the LWP q2 2
Ac 1 e q c / 2 s*
2 w c s*
2
sensitivity of parameterized cloud fraction and optical
depth to changes in temperature. The coupled ( A c , aL
2
) global response of clouds to increasing temperature LWP 2 q4 6q c
2 2
3 4
4 w c
2 s* s*
is analogous to the response of an open thermodynamic
system. Although the particular thermodynamic trajec-
tory that the system follows may be very sensitive to Ac 1 2 2
(q c
3
s* 5q c 3
s* )e q c / 2 s*
, (A2)
external forcing and boundary conditions, much can be 2
learned by computing response functions where one of
the thermodynamic coordinates is fixed along the tra- where A c erfc( q c / 2 s )/2. Expanding (A2) to
*
jectory. This approach was first considered by Temkin fourth order in small s /q c gives
*
et al. (1975), who found ( A c ) 0 and ( )Ac 0 1/ 2 2 1/ 2
using a nonstochastic model of a simplified atmosphere 2a L aL s*
with one cloud layer and constant surface RH. LWP s* LWP ,
w 4 w
Using our statistical treatment of cloud optical vari-
ability, we derive analytic response functions in (Ac, , T) from which (A1) follows approximately. Using Eq. (A2)
space that demonstrate the overall dominance of the cloud we find that (A1) is accurate to better than 7% for s* /
fraction feedback in the model. In contradistinction to Te- qc 1/2.
mkin et al. (1975), we find ( Ac ) 0. In particular, we A potential disadvantage of Eq. (A2) is that corre-
show that the global observational evidence of a largely sponding closed-form expressions for the h 5/3 model are
negative optical depth sensitivity presented by Tselioudis not available; the modified triangle distribution intro-
et al. (1993) produces in the model a much stronger neg- duced in appendix B has tractable noninteger moments
ative cloud fraction response and therefore a net positive and exhibits Gaussian behavior in close agreement with
low cloud feedback. Also we find that low model vertical (A2).
resolution can cause a significant overestimation of the
unresolved low cloud Ac response by a factor of around APPENDIX B
2.5. The accuracy of these results rests upon the crucial
assumption that low-cloud Ac may be parameterized as a
function of only relative humidity, an assumption that is Modified Triangle Distribution
typically made in large-scale models. Improvement in our
Our modified triangle distribution is
understanding of the factors that control Ac at large scales
is therefore a necessary next step towards the refinement 3
in the formulation of the (Ac, ) response functions intro- 3 5|s| |s|
P(s) 1 1 ,
duced in this work. 2 0 3 0 0
Acknowledgments. We are grateful to Nicole Jeffery 0 s 0 , (B1)
for a careful reading of the manuscript. We thank three
anonymous reviewers for very thorough and construc- where 2 0 (35/3) 2. Using Eqs. (1) and (B1), cloud
s
tive comments. This work was supported through fund- fraction A c A d (z top ) is
ing of the Modeling of Clouds and Climate Proposal by
the Canadian Foundation for Climate and Atmospheric 0 QN 1
Sciences, the Meteorological Service of Canada, and (1 Q N ) 4 (1 Q N )/2 1 QN 0
the Natural Sciences and Engineering Research Council. Ac
1 (1 Q N ) (1 Q N )/2 0 QN 1
4
APPENDIX A 1 1 QN ,
Gaussian Relations for the h 2 Model where Q N q c / 0 and q c qt w z top q 0 . The -
th moment of the cloud liquid water used in the cal-
Recently Wood and Taylor (2001) derived
1/ 2 culation of via Eqs. (5) or (6) follows in a similar
LWP (2a L w 1 )1/ 2 LWP s (A1) manner:
1630 JOURNAL OF THE ATMOSPHERIC SCIENCES VOLUME 60
0 QN 1 REFERENCES
A c F1
1
1 QN 0
(q c s) Albrecht, B. A., C. W. Fairall, D. W. Thomson, and A. B. White,
A c (F1 F2 ) 0 QN 1
1
1990: Surface-based remote sensing of the observed and the
adiabatic liquid water content of stratocumulus clouds. Geophys.
A c (F1 F2 F3 ) 1 QN ,
1
Res. Lett., 17, 89–92.
Arking, A., 1991: The radiative effects of clouds and their impact
where on climate. Bull. Amer. Meteor. Soc., 72, 795–813.
Austin, P., Y. Wang, R. Pincus, and V. Kujala, 1995: Precipitation in
0 (1 QN ) 4
3 5Q N 4 20Q N stratocumulus clouds: Observational and modeling results. J.
F1 Atmos. Sci., 52, 2329–2352.
2 1 2 Bajuk, L. J., and C. B. Leovy, 1998: Seasonal and interannual var-
6 30Q N 12 20Q N iations in stratiform and convective clouds over the tropical Pa-
cific and Indian Oceans from ship observations. J. Climate, 11,
3 4 2922–2941.
Barker, H. W., 1996: A parameterization for computing grid-averaged
5(1 QN ) solar fluxes for inhomogeneous marine boundary layer clouds.
Part I: Methodology and homogeneous biases. J. Atmos. Sci.,
5 53, 2289–2303.
——, and B. A. Wielicki, 1997: Parameterizing grid-averaged long-
0QN 1
QN 3Q N
3
QN 9Q N
3
9Q N
3
wave fluxes for inhomogeneous marine boundary layer clouds.
F2 J. Atmos. Sci., 54, 2785–2798.
16 1 2 3 ——, ——, and L. Parker, 1996: A parameterization for computing
grid-averaged solar fluxes for inhomogeneous marine boundary
3Q N
3
layer clouds. Part II: Validation using satellite data. J. Atmos.
4 Sci., 53, 2304–2316.
Boers, R., J. D. Spinhirne, and W. D. Hart, 1988: Lidar observations
of the fine-scale variability of marine stratocumulus clouds. J.
0 ( 1 QN ) 4
3 5Q N 4 20Q N Appl. Meteor., 27, 797–810.
F3 Bony, S., K.-M. Lau, and Y. C. Sud, 1997: Sea surface temperature
2 1 2
and large-scale circulation influences on tropical greenhouse ef-
6 30Q N 12 20Q N fect and cloud radiative forcing. J. Climate, 10, 2055–2077.
Brenguier, J.-L., H. Pawlowska, L. Schu ¨ller, R. Preusker, J. Fischer,
3 4 and Y. Fouquart, 2000: Radiative properties of boundary layer
clouds: Droplet effective radius versus number concentration. J.
5(1 QN ) Atmos. Sci., 57, 803–821.
. Browning, K. A., 1994: Survey of perceived priority issues in the
5 parameterizations of cloud-related processes in GCMs. Quart.
J. Roy. Meteor. Soc., 120, 483–487.
Bushell, A. C., and G. M. Martin, 1999: The impact of vertical res-
APPENDIX C olution upon GCM simulations of marine stratocumulus. Climate
Dyn., 15, 293–318.
Parameter Values Cahalan, R. F., W. Ridgway, W. J. Wiscombe, T. L. Bell, and J. B.
Snider, 1994: The albedo of fractal stratocumulus clouds. J. At-
Parameter values are q 0 (T) (1.826 10 9 g m 3 ) mos. Sci., 51, 2434–2455.
exp{ R /(L T)}, R 461.5 J K 1 kg 1 , L 2.5 Cess, R. D., G. L. Potter, and J. P. Blanchet, 1990: Intercomparison
and interpretation of climate feedback processes in 19 atmo-
10 6 J kg 1 , w (4 10 3 K m 1 ) {L /(R T 2 )}q 0 (T), spheric general circulation models. J. Geophys. Res., 95 (D10),
aL 0.75, and the longwave absorption coefficient is 16 601–16 615.
0.15 g 1 m 2 . The constant of proportionality in Eq. (6) Chen, T., W. B. Rossow, and Y. Zhang, 2000: Radiative effects of
is 2(k )1/3 (4/3 w ) 2/3 N 1/3 (Pontikis 1993) with param- cloud-type variations. J. Climate, 13, 264–286.
eter values k 1, w 1 g cm 3 , and droplet number Clothiaux, E. E., T. P. Ackerman, G. G. Mace, K. P. Moran, R. T.
Marchand, M. A. Miller, and B. E. Martner, 2000: Objective
density N 200 10 6 m 3 . The parameter k relates determination of cloud heights and radar reflectivities using a
the effective and volume averaged radii. Parameter val- combination of active remote sensors at the ARM CART sites.
ue a L 0.75 is consistent with a range of observations J. Appl. Meteor., 39, 645–665.
(e.g., Austin et al. 1995; Brenguier et al. 2000). Colman, R. A., and B. J. McAvaney, 1997: A study of general cir-
culation model climate feedbacks determined from perturbed sea
surface temperature experiments. J. Geophys. Res., 102 (D16),
APPENDIX D 19 383–19 402.
Considine, G., J. A. Curry, and B. Wielicki, 1997: Modeling cloud
LCF Parameter Values fraction and horizontal variability in marine boundary layer
clouds. J. Geophys. Res., 102 (D12), 13 517–13 525.
Shortwave cloud forcing: no absorption, surface al- Cuijpers, J. W. M., and P. Bechtold, 1995: A simple parameterization
bedo is from Robock (1980), ln / T 0.01 0.14 of cloud water related variables for use in boundary layer models.
exp( 0.00175 2 ) K 1 is parameterized from Tselioudis J. Atmos. Sci., 52, 2486–2490.
Cusack, S., J. M. Edwards, and R. Kershaw, 1999: Estimating the
et al. (1993), and the solar constant is 1365 W m 2 . subgrid variance of saturation, and its parameterization for use
Longwave cloud forcing: water vapor forcing is ignored in a GCM cloud scheme. Quart. J. Roy. Meteor. Soc., 125, 3057–
and cloud-top temperature is T 6.5 C. 3076.
1 JULY 2003 JEFFERY AND AUSTIN 1631
Deardorff, J. W., 1974: Three-dimensional numerical study of tur- and cloud cover on sea surface temperature for two tropical
bulence in an entraining mixed layer. Bound.-Layer Meteor., 7, marine stratocumulus regions. J. Climate, 6, 2434–2447.
199–226. Pincus, R., and S. A. Klein, 2000: Unresolved spatial variability and
——, 1981: On the distribution of mean radiative cooling at the top microphysical process rates in large-scale models. J. Geophys.
of a stratocumulus-capped mixed layer. Quart. J. Roy. Meteor. Res., 105 (D22), 27 059–27 065.
Soc., 107, 191–202. Pontikis, C., 1993: Parameterization of the cloud optical thickness:
Greenwald, T. J., G. L. Stephens, S. A. Christopher, and T. H. Vonder Influence of clear air entrainment. Geophys. Res. Lett., 20, 2655–
Haar, 1995: Observations of the global characteristics and re- 2658.
gional radiative effects of marine cloud liquid water. J. Climate, Ramaswamy, V., and C.-T. Chen, 1993: An investigation of the global
8, 2928–2946. solar radiative forcing due to changes in cloud liquid water path.
Hansen, J., A. Lacis, D. Rind, and G. Russell, 1984: Climate sen- J. Geophys. Res., 98 (D9), 16 703–16 712.
sitivity: Analysis of feedback mechanisms. Climate Processes Ricard, J. L., and J. F. Royer, 1993: A statistical cloud scheme for
and Climate Sensitivity, Geophys. Monogr., No. 29, Amer. Geo- use in an AGCM. Ann. Geophys., 11, 1095–1115.
phys. Union, 130–163. Robock, A., 1980: The seasonal cycle of snow cover, sea ice and
Hatzianastassiou, N., and I. Vardavas, 1999: Shortwave radiation bud- surface albedo. Mon. Wea. Rev., 108, 267–285.
get of the northern hemisphere using International Satellite Rotstayn, L. D., 1997: A physically based scheme for the treatment
Cloud Climatology Project and NCEP/NCAR climatological of stratiform clouds and precipitation in large-scale models. I:
data. J. Geophys. Res., 104 (D20), 24 401–24 421. Description and evaluation of the microphysical processes.
Jeffery, C. A., 2001: Statistical models of cloud-turbulence interac- Quart. J. Roy. Meteor. Soc., 123, 1227–1282.
tions. Ph.D. thesis, University of British Columbia, Vancouver, ——, 1999: Climate sensitivity of the CSIRO GCM: Effects of cloud
Canada, 122 pp. modeling assumptions. J. Climate, 12, 334–356.
——, and A. B. Davis, 2002: Signature of cloud-base-height skew- Schneider, S. H., W. M. Washington, and R. M. Chervin, 1978: Cloud-
ness in ARM microwave water radiometer data: Implications for iness as a climatic feedback mechanism: Effects on cloud
cloud radiative parameterizations in GCMs. Proc. SPIE, 4815, amounts of prescribed global and regional surface temperature
9–19. changes in the NCAR GCM. J. Atmos. Sci., 35, 2207–2221.
Klein, S. A., and D. L. Hartmann, 1993: The seasonal cycle of low Smith, R. N. B., 1990: A scheme for predicting layer clouds and their
stratiform clouds. J. Climate, 6, 1587–1606. water content in a general circulation model. Quart. J. Roy.
——, ——, and J. R. Norris, 1995: On the relationships among low- Meteor. Soc., 116, 435–460.
cloud structure, sea surface temperature, and atmospheric cir- Somerville, R. C. J., and L. A. Remer, 1984: Cloud optical thickness
culation in the summertime northeast Pacific. J. Climate, 8, feedbacks in the CO 2 climate problem. J. Geophys. Res., 89 (D6),
1140–1155. 9668–9672.
Kogan, Z. N., Y. L. Kogan, and D. L. Lilly, 1997: Cloud factor and Sommeria, G., and J. W. Deardorff, 1977: Subgrid-scale condensation
seasonality of the indirect effect of anthropogenic sulfate aero- in models of nonprecipitating clouds. J. Atmos. Sci., 34, 344–
sols. J. Geophys. Res., 102 (D22), 25 927–25 939. 355.
Larson, V. E., R. Wood, P. R. Field, J.-C. Golaz, T. H. Vonder Haar, Strawbridge, K. B., and R. M. Hoff, 1996: LITE validation experiment
and W. R. Cotton, 2001: Small-scale and mesoscale variability along California’s coast: Preliminary results. Geophys. Res. Lett.,
of scalars in cloudy boundary layers: One-dimensional proba- 23, 73–76.
bility density functions. J. Atmos. Sci., 58, 1978–1994. Temkin, R. L., B. C. Weare, and F. M. Snell, 1975: Feedback coupling
Levkov, L., B. Rockel, H. Schiller, and L. Kornblueh, 1998: 3-D of absorbed solar radiation by three model atmospheres with
simulation of clouds with subgrid fluctuations of temperature clouds. J. Atmos. Sci., 32, 873–880.
and humidity. Atmos. Res., 47–48, 327–341. Tselioudis, G., A. A. Lacis, D. Rind, and W. B. Rossow, 1993: Po-
´
Loeb, N. G., T. Varnai, and D. M. Winker, 1998: Influence of subpixel- tential effects of cloud optical thickness on climate warming.
scale cloud-top structure on reflectances from overcast stratiform Nature, 366, 670–672.
cloud layers. J. Atmos. Sci., 55, 2960–2973. ——, A. D. Del Genio, W. Kovari, and M.-S. Yao, 1998: Temperature
Manabe, S., and R. T. Wetherald, 1967: Thermal equilibrium of the dependence of low cloud optical thickness in the GISS GCM:
atmosphere with a given distribution of relative humidity. J. Contributing mechanisms and climate implications. J. Climate,
Atmos. Sci., 24, 241–259. 11, 3268–3281.
——, and R. J. Stouffer, 1979: A CO 2-climate sensitivity study with Wang, Z., and K. Sassen, 2001: Cloud type and macrophysical prop-
a mathematical model of the global climate. Nature, 282, 491– erty retrieval using multiple remote sensors. J. Appl. Meteor.,
493. 40, 1665–1682.
Mellor, G. L., 1977: The Gaussian cloud model relations. J. Atmos. Warren, S. G., C. J. Hahn, J. London, R. M. Chervin, and R. L. Jenne,
Sci., 34, 356–358; Corrigendum, 34, 1483. 1988: Global distribution of total cloud cover and cloud type
Moeng, C. H., P. P. Sullivan, and B. Stevens, 1999: Including radiative amounts over the ocean. Tech. Note TN-317 STR, NCAR,
effects in an entrainment rate formula for buoyancy-driven PBLs. Boulder, CO, 305 pp.
J. Atmos. Sci., 56, 1031–1049. Wetherald, R. T., and S. Manabe, 1986: An investigation of cloud
Norris, J. R., 1998a: Low cloud type over the ocean from surface cover change in response to thermal forcing. Climatic Change,
observations. Part I: Relationship to surface meteorology and 8, 5–23.
the vertical distribution of temperature and moisture. J. Climate, Wood, R., and J. P. Taylor, 2001: Liquid water path variability in
11, 369–382. unbroken marine stratocumulus cloud. Quart. J. Roy. Meteor.
——, 1998b: Low cloud type over the ocean from surface obser- Soc., 127, 2635–2662.
vations. Part II: Geographical and seasonal variations. J. Climate, Xu, K.-M., and D. A. Randall, 1996: Evaluation of statistically based
11, 383–403. cloudiness parameterizations used in climate models. J. Atmos.
——, and C. B. Leovy, 1994: Interannual variability in stratiform Sci., 53, 3103–3119.
cloudiness and sea surface temperature. J. Climate, 7, 1915– Yao, M.-S., and A. D. Del Genio, 1999: Effects of cloud parame-
1925. terization on the simulation of climate changes in the GISS
Oreopoulos, L., and R. Davies, 1993: Statistical dependence of albedo GCM. J. Climate, 12, 761–779.
