JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 110, D08206, doi:10.1029/2004JD005600, 2005
Multidecadal changes in near-global cloud cover and estimated cloud
cover radiative forcing
Joel R. Norris
Scripps Institution of Oceanography, La Jolla, California, USA
Received 11 November 2004; revised 28 January 2005; accepted 10 March 2005; published 30 April 2005.
[1] This study examines variability in zonal mean surface-observed upper-level
(combined midlevel and high-level) and low-level cloud cover over land during 1971–
1996 and over ocean during 1952–1997. These data were averaged from individual
synoptic reports in the Extended Edited Cloud Report Archive (EECRA). Although
substantial interdecadal variability is present in the time series, long-term decreases in
upper-level cloud cover occur over land and ocean at low and middle latitudes in both
hemispheres. Near-global upper-level cloud cover declined by 1.5%-sky-cover over land
between 1971 and 1996 and by 1.3%-sky-cover over ocean between 1952 and 1997.
Consistency between EECRA upper-level cloud cover anomalies and those from the
International Satellite Cloud Climatology Project (ISCCP) during 1984–1997 suggests the
surface-observed trends are real. The reduction in surface-observed upper-level cloud
cover between the 1980s and 1990s is also consistent with the decadal increase in all-sky
outgoing longwave radiation reported by the Earth Radiation Budget Satellite (ERBS).
Discrepancies occur between time series of EECRA and ISCCP low-level cloud cover due
to identified and probable artifacts in satellite and surface cloud data. Radiative effects of
surface-observed cloud cover anomalies, called ‘‘cloud cover radiative forcing (CCRF)
anomalies,’’ are estimated based on a linear relationship to climatological cloud radiative
forcing per unit cloud cover. Zonal mean estimated longwave CCRF has decreased
over most of the globe. Estimated shortwave CCRF has become slightly stronger over
northern midlatitude oceans and slightly weaker over northern midlatitude land areas. A
long-term decline in the magnitude of estimated shortwave CCRF occurs over low-latitude
land and ocean, but comparison with ERBS all-sky reflected shortwave radiation during
1985–1997 suggests this decrease may be underestimated.
Citation: Norris, J. R. (2005), Multidecadal changes in near-global cloud cover and estimated cloud cover radiative forcing,
J. Geophys. Res., 110, D08206, doi:10.1029/2004JD005600.
1. Introduction
[2] Clouds are major regulators of Earth’s radiation
budget. Typically, they reflect more solar or shortwave
(SW) radiation back to space than the unobscured surface,
thus decreasing the energy gained by the Earth. They also
usually emit less thermal infrared or longwave (LW) radiation to space than the unobscured surface, which decreases
energy loss by the Earth. Cloud radiative forcing (CRF) is
the difference between actual radiative flux and what it
would be were clouds absent. Negative CRF at the top of
the atmosphere (TOA) corresponds to a loss of energy by
the climate system due to cloud radiative effects. SWCRF is
most negative for clouds with large albedo under strong
insolation. LWCRF is most positive for non-transmissive
clouds high in the atmosphere since they are cold and
therefore emit less radiation than the unobscured surface.
Earth Radiation Budget Experiment (ERBE) satellite data
shows that SWCRF is larger than LWCRF in the global
average, thus indicating an overall cloud cooling effect in
Copyright 2005 by the American Geophysical Union.
0148-0227/05/2004JD005600$09.00
the present climate [Ramanathan et al., 1989]. Note that
SWCRF cooling primarily occurs at the surface and
LWCRF warming primarily occurs in the atmosphere.
[3] Despite the key role of clouds in the climate system,
we still have very limited understanding of the net cloud
response to climate change. At present it is not known
whether cloud cover, cloud reflectivity, and cloud height
will change in such a way as to mitigate or exacerbate
global warming [Moore et al., 2001]. In large part because
they do not correctly and consistently simulate clouds,
global climate models do not agree on the future magnitude
of global warming. Therefore it is essential to investigate
cloud and radiation variability observed over the past fifty
years, a time period of rapidly increasing anthropogenic
forcing on the climate system. Although relatively homogeneous satellite data sets have recently become long
enough to examine cloud and radiation variability over
1 – 2 decades [e.g., Wielicki et al., 2002a; Rossow and
Schiffer, 1999; Cess and Udelhofen, 2003], they only go
back to the mid-1980s. Synoptic reports of cloud cover
extend over a longer period of time, but many previous
studies of surface-observed cloud variations have not been
D08206
1 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
fully validated due to lack of independent cloud data [e.g.,
Henderson-Sellers, 1989; Kaiser, 2000; Norris, 1999; Sun
and Groisman, 2000; Sun et al., 2001]. It is now possible,
however, to begin addressing this shortcoming by comparing surface and satellite cloud data over a time period
approximately 15 years long [e.g., Sun, 2003].
[4] The present study uses surface synoptic cloud observations obtained from the Extended Edited Cloud Report
Archive (EECRA) [Hahn and Warren, 1999] to document
near-global variability in upper-level and low-level cloud
cover. Time series of midlatitude and low-latitude zonal mean
cloud cover are presented for land areas during 1971 – 1996
and ocean areas during 1952 –1997. These data are compared
to zonal mean time series of upper-level and low-level cloud
cover from the International Satellite Cloud Climatology
Project (ISCCP) [Rossow and Schiffer, 1999] during 1984 –
1996 to assess the consistency between the two independent
data sets. LW and SW radiation anomalies associated with
surface-observed cloud cover anomalies are furthermore
empirically estimated to partially quantify radiative impacts
of cloud variability, particularly in the pre-satellite era. These
are compared to all-sky LW and SW flux anomalies from the
Earth Radiation Budget Satellite (ERBS) [Wielicki et al.,
2002a] during 1985 – 1996. The documentation of multidecadal cloud cover trends and quantitative estimation of
their radiative impacts provided in this and subsequent
regional studies will improve our understanding of cloud
data quality and cloud feedbacks on the climate system.
2. Data and Methods
2.1. Extended Edited Cloud Report Archive (EECRA)
[5] The EECRA [Hahn and Warren, 1999] provides
individual surface synoptic cloud reports using a consistent
observing procedure globally over land during 1971 – 1996
and globally over ocean during 1952 – 1997. The land cloud
reports came from stations assigned official numbers by the
World Meteorological Organization (WMO). The ocean
cloud reports, primarily from Volunteer Observing Ships,
were originally archived in the Comprehensive OceanAtmosphere Data Set (COADS) [Woodruff et al., 1987].
[6] Synoptic code cloud parameters are N (sky cover by
all clouds), Nh (sky cover by the lowest cloud layer), CL,
CM, and CH (cloud types at low, middle, and high levels),
and ww (present weather) [WMO, 1987]. The standard
WMO code requires that N and Nh be reported in units of
eighths. A comparison of all-sky camera images with
standard visual observations reported by ships’ officers
found they agreed to within one-eighth of sky cover for
75% of the reports [Henderson-Sellers and McGuffie,
1988]. Henderson-Sellers and McGuffie report that some
of the disagreement arose from changes in cloudiness
between the time of the camera image and the time of the
visual report, and the only apparent bias was a tendency for
ships’ officers to slightly underreport the frequency of cirrus
and overestimate the amount of cirrus when it did occur.
EECRA observations are entirely independent from satellite
data and therefore represent an important additional
resource to studies of global cloud variability.
2.2. Earth Radiation Budget Satellite (ERBS)
[7] ERBS nonscanner data provides measured outgoing
LW radiation (OLR) and reflected SW radiation (RSW) at
D08206
10 10 grid resolution for the 1985 – 1999 time period.
OLR and RSW fluxes were corrected to account for
variations in satellite altitude using coefficients provided
by T. Wong (personal communication, 2004). Clear sky
radiation fluxes are not available, so the cloud impact on
TOA radiation cannot be directly observed. The ERBS orbit
precessed through 12 hours of local daytime and nighttime
sampling over 36 days, and averaging over monthly time
periods aliased the diurnal cycle of reflected SW radiation
into an apparent semiannual cycle in previous studies
[Wielicki et al., 2002a; Trenberth, 2002]. Averaging over
36-day intervals eliminates this problem [Wielicki et al.,
2002b]. 36-day ERBS nonscanner data always begin on
January 1, and the last five or six days of the year are
discarded. The sampling uncertainty of OLR data is less
than the sampling uncertainty of RSW data because
the former does not experience such a large diurnal cycle.
36-day values with less than twelve days contributing to the
mean were set to missing in the present study due to their
very large sampling uncertainty. Because ERBS had a lowinclination orbit, sampling is most frequent at low latitudes
and does not extend poleward of 60. Severe aliasing can
result poleward of 40 where 36 days are not sufficient
to sample the entire diurnal cycle. To prevent this, the
climatological seasonal cycle was removed and the available 36-day anomalies were then averaged to 72-day
anomalies. Another benefit of averaging to 72-day anomalies is reduction of general sampling uncertainty. ERBS data
are completely missing during the second half of 1993 and
during several shorter periods in 1998 and 1999. To avoid
aliasing problems, 72-day data were set to missing if data in
one of the contributing 36-day intervals were missing over a
large fraction or all of the globe.
[8] The ERBS scanner instrument provides information
on CRF that is not available from the ERBS nonscanner
instrument, but unfortunately only from 1985 to 1989
[Barkstrom et al., 1989]. Monthly all-sky fluxes and CRF
values at 2.5 2.5 grid resolution were averaged to 10 10 grid boxes with weighting according to area. These
were then converted to 72-day values by averaging with
weighting according to the number of days in each month
contributing to a 72-day period.
2.3. International Satellite Cloud Climatology
Project (ISCCP)
[9 ] The ISCCP [Rossow et al., 1996; Rossow and
Schiffer, 1999] provides cloud fraction, cloud top pressure,
and cloud optical thickness information retrieved from
geostationary and polar-orbiting weather satellites from July
1983 onwards. High clouds are defined as those with tops
above the 440 hPa level, midlevel clouds as those with tops
between 680 hPa and 440 hPa, and low-level clouds as
those with tops below the 680 hPa level. These height
categories are further divided into individual ‘‘cloud types’’
according to visible cloud optical thickness. This study
defines upper-level cloud fraction as the sum of ISCCP
high-level and midlevel cloud fraction. ISCCP low-level
cloud fraction takes into account only those low clouds that
are not obscured by higher clouds. Only ISCCP daytime
data are examined since ISCCP may have trouble correctly
detecting cirrus or low-level clouds using the IR channel
alone. Monthly ISCCP D2 data at 2.5 2.5 grid resolu-
2 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
tion were averaged to 72-day data at 10 10 grid
resolution with weighting according to area and the number
of days in each month contributing to a 72-day period.
2.4. Derivation of EECRA Upper-Level and
Low-Level Cloud Cover
[10] Upper-level cloud cover (NU), defined in this study
as the coverage by midlevel and high-level clouds, was
inferred by assuming random overlap with obscuring lowerlevel clouds, i.e., NU = (N Nh)/(1 Nh). In general it is
not possible to calculate midlevel cloud cover and highlevel cloud cover separately since only total cover (N) and
lowest-level cloud cover (Nh) are reported. Random overlap
assumes that upper-level cloudiness covers the same relative
fraction of sky where it is obscured by lower clouds as
where it is not obscured. Studies of clouds in several
different meteorological regimes have found that random
overlap is the best assumption for non-contiguous cloud
layers, as well as for widely separated vertical levels of a
single thick cloud layer [e.g., Tian and Curry, 1989; Hogan
and Illingworth, 2000; Mace and Benson-Troth, 2002].
Both of these conditions commonly characterize the twolayer cloud division used in this investigation. Mazin et al.
[1993] found that cirrus cloud cover estimated from surface
synoptic reports using random overlap differed from cirrus
cloud cover measured in aircraft soundings by less than
10% over the former U.S.S.R region. In the case that the
synoptic code parameter CL reported no low-level cloud
types (CL = 0), upper-level cloud cover was set to the value
of N. Upper-level cloud cover was set to 100% due to
identification of nimbostratus in the case of sky obscuration
(N = 9) or overcast low-level cloudiness (Nh = 8 and CL =
1 – 9) with non-drizzle precipitation (ww = 60– 75, 77, 79–
99). Since shallow clouds can nevertheless drizzle, the
presence of drizzle precipitation did not lead to the identification of upper-level nimbostratus unless the overcast lowlevel clouds were cumulonimbus or bad-weather stratus
(CL = 3, 7, 9). No determination of upper-level cloud cover
could be made in other cases of overcast low-level cloudiness or sky-obscuration, and this study assumes that
average upper-level cloud cover is the same for when it
cannot be seen as when it can be seen. Comparison of
EECRA and ISCCP upper-level cloud cover climatologies
indicates the above assumption is less valid for marine
subtropical stratocumulus regions where upper clouds preferentially occur when passing synoptic disturbances break
up overcast stratocumulus. Given the difficulty of distinguishing the specific meteorological environment of each
overcast cloud report, however, it was deemed simplest to
assume no preference for or against upper-level cloudiness,
unless non-drizzle precipitation was present and nimbostratus could thus be diagnosed.
[11] Surface-observed low-level cloud cover cannot be
directly compared to satellite-observed low-level cloud
cover because higher clouds often block the satellite’s view
of low-level clouds. EECRA low-level cloud cover values
were therefore adjusted to represent the ‘‘satellite view’’ by
removing the portion of low-level cloud cover overlapped
by higher clouds. This was accomplished by subtracting
upper-level cloud cover from total cloud cover, i.e., NL =
Nh (1 N)/(1 Nh), where NL is ‘‘satellite view’’ lowlevel cloud cover.
D08206
[12] A code change introduced in 1982 instructed observers to report Nh, CL, CM, and CH as ‘‘no information’’ if the
sky were completely clear (N = 0). This revised coding
method could create a bias in the calculation of average
upper-level and low-level cloud cover over ocean since
some ships routinely fail to report Nh, and thus after 1982 it
is not possible to distinguish routinely non-reporting ships
from those ships not reporting simply because N = 0.
Including routinely non-reporting ships in the calculation
of average upper-level cloud cover would result in a clearsky bias since such ships contribute only when upper-level
cloud cover is zero (clear sky). A similar bias could result
from ships that routinely do not report Nh instead of merely
when the sky is obscured by fog or precipitation (N = 9).
Application of the averaging method described by Norris
[1999, Appendix] avoids ‘‘clear-sky’’ and ‘‘obscured-sky’’
biases. Potential clear-sky and obscured-sky biases are
nevertheless negligible over most of the ocean since clear
sky and obscured sky are rare over most of the ocean.
Potential biases are also small over land since very few
stations fail to report Nh, and the averaging methods are
identical when Nh is always reported.
[13] Since human observers have difficulty identifying
cloudiness under conditions of poor illumination (little or no
moonlight), only a fraction of nighttime observations are
reliable [Hahn et al., 1995]. To avoid biases resulting from
non-uniform sampling of the diurnal cycle, 36-day average
cloud cover values were separately calculated for each
observing time (00, 06, 12, 18 UTC over ocean and 00,
03, 06, 09, 12, 15, 18, 21 UTC over land). Daytime-only
cloud cover values were averaged solely from observing
times when the sun was above the horizon. As shown in
Figure A1, changes in the diurnal cycle of cloud cover are
small compared to variability in the diurnal mean.
[14] Anomaly values were generated for individual land
station time series by subtracting the long-term 36-day
means at each observing time, and anomalies for all stations
within a 5 5 grid box were averaged together. Longterm means for each station were separately averaged and
then added to the grid box anomaly. This procedure avoids
spurious variability that could result if incomplete time
series from stations with differing climatological cloud
cover were averaged together. Individual ocean observations were separately averaged from land values into 5 5 grid boxes, and 5 5 land and ocean values were
separately averaged to the 10 10 ERBS grid. Spurious
variability can result from non-uniform sampling within a
10 10 grid box; this is especially a problem for ocean
data because ships often travel along narrow trade routes
and may pass through a climatologically less cloudy portion
of the grid box at one time and a climatologically cloudier
portion at another time. To limit spatial, temporal, and
diurnal sampling biases, 36-day 5 5 6-hourly anomalies
were averaged to 72-day 10 10 diurnal mean anomalies
with weighting by the number of reports contributing to each
36-day 5 5 6-hourly anomaly. 36-day 5 5 6-hourly
long-term mean values were averaged to 72-day 10 10
diurnal mean values with equal weighting in time and
weighting by area in space. This procedure enables those
5 5 regions and hours of the day with better sampling to
contribute more to the 10 10 diurnal mean anomaly
without introducing spurious variability due to a climato-
3 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
D08206
Figure 1. The 72-day anomalies in EECRA upper and total cloud cover (thin red), ISCCP upper and
total cloud cover (thick blue), ERBS nonscanner OLR and RSW (thick black), and ERBS scanner
LWCRF and SWCRF (thin green) averaged over land and ocean grid boxes between (a) 30S – 30N,
(b) 30– 60N, and (c) 30– 60S with weighting according to grid box area. Note that the signs of OLR
and SWCRF time series have been reversed for better comparability. Only daytime hours contributed to
cloud time series (VIS/IR detection for ISCCP). Anomalies are referenced from the 1985 – 1989 mean.
logical diurnal cycle or spatial gradient in cloud cover. After
combining 72-day 10 10 anomalies and long-term
means, land and ocean values were averaged with weighting
according to land/ocean fraction in the 10 10 grid box.
3. Observed Cloud and Radiation Variations
[15] Figure 1 displays zonal mean time series of upperlevel and total cloud cover reported by the EECRA together
with satellite data for the 1984 – 1999 time period. A
reduction in surface-observed upper-level cloud cover
occurs between 1984 and 1996 at low and middle latitudes
in both hemispheres. This near-global decline in upper-level
cloud cover is also evident in the ISCCP data (Figure 1).
Interannual anomalies in EECRA and ISCCP upper-level
cloud cover have greatest agreement at northern middle
latitudes where surface sampling density is highest and least
agreement at southern middle latitudes where the scarcity of
ship observations introduces substantial noise into the
EECRA time series. The discrepancy between EECRA
and ISCCP during 1991 –1993 is due to volcanic aerosols
from the Mount Pinatubo eruption that caused ISCCP
to misidentify some high-level optically thin clouds as
low-level clouds, thus creating an apparent decrease in
upper-level cloud cover during that interval [Luo et al.,
2002]. The decadal decrease in EECRA and ISCCP upperlevel cloudiness is consistent with the decadal increase in
OLR reported by the ERBS nonscanner (plotted with
reversed sign in Figure 1 for better comparability to the
cloud time series). Wielicki et al. [2002a] previously documented a decadal increase in ERBS tropical OLR, attributed
to reduced cloudiness by Chen et al. [2002], and Figures 1b
and 1c indicate that the upward trend in OLR occurred at
higher latitudes as well. Additional corroboration of the
decadal decrease in tropical upper-level cloud cover is the
decline in uppermost opaque cloud height measured by
the Stratosphere Aerosol and Gas Experiment (SAGE) II
between 1985 and 1998 [Wang et al., 2002].
[16] Less agreement occurs between the cloud data sets
and ERBS for total cloud cover and RSW. ERBS reports
a strong decline in RSW at low latitudes (Figure 1a) but
only slight downward trends in RSW at middle latitudes
4 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
Table 1. Linear Correlations Between 10 10 Cloud and
Radiation Anomalies During 1985 – 1989
Correlation Parameters
Low Latitude
(30S – 30N)
NH Midlatitude
(30 – 60N)
Nonscanner OLR-scanner OLR
Scanner OLR-LWCRF
LWCRF-ISCCP upper cover
LWCRF-EECRA upper cover
Nonscanner RSW-scanner RSW
Scanner RSW-SWCRF
SWCRF-ISCCP total cover
SWCRF-EECRA total cover
SWCRF-ISCCP cloud albedo
+0.87
0.97
+0.79
+0.51
+0.82
0.99
0.71
0.56
0.45
+0.81
0.68
+0.52
+0.43
+0.58
0.82
0.41
0.49
0.39
(Figures 1b and 1c). At low latitudes the decrease in ISCCP
total cloud cover is in the same direction as the RSW trend,
as noted by Cess and Udelhofen [2003], but the increase in
EECRA total cover is not. Conversely, at middle latitudes
the near-zero trend in EECRA total cloud cover is similar to
the near-zero RSW trend, but the larger decline in ISCCP
total cover is not. Note that EECRA cloud variations are
not expected to match the large positive anomaly in RSW
and negative anomaly in OLR during 1991 –1993 caused
by Pinatubo aerosols. Differences between EECRA and
ISCCP total cloud cover will be discussed in more detail
later on.
[17] Comparison of ERBS nonscanner all-sky flux and
ERBS scanner CRF time series in Figure 1 indicates that
clouds were the primary source of zonal mean variability in
OLR and RSW during 1985 – 1989 [e.g., Hartmann et al.,
1992] (note that the sign of the SWCRF time series has been
reversed for better comparability to the other time series). In
fact, the agreement between RSW and SWCRF is probably
greater than that seen in Figure 1 because the much larger
field of view of the nonscanner (1000 km) made scene
identification difficult, leading to less accurate predictions
of albedo dependence on solar zenith angle (B. A. Wielicki,
personal communication, 2004). The weaker amplitude of
LWCRF anomalies relative to OLR anomalies, however,
suggests that other atmospheric properties, such as water
vapor and temperature, have substantial effects on OLR
variability. Changes in temperature can also potentially
cause changes in LWCRF even if cloud properties remain
the same. Nevertheless, the similarity of LWCRF and OLR
time series implies that these other constituents largely covary with clouds. Table 1 lists linear correlations between
10 10 anomalies of various cloud and radiation parameters for low latitudes and for northern middle latitudes
during 1985 – 1989 (southern middle latitudes are excluded
due to the lower density of surface observations). At low
latitudes, variations in LWCRF explain more than 90% of
the variance in OLR and variations in ISCCP upper-level
cloud cover explain more than 60% of the variance in
LWCRF. Variations in ISCCP total cloud cover explain
50% of the variance in SWCRF, which is substantially
more than the 20% explained by variations in cloud albedo.
Sampling uncertainty due to the smaller numbers of surface
observations bring down the magnitude of correlations
between CRF and EECRA cloud cover relative to correlations between CRF and ISCCP cloud cover. Possible
reasons for the weaker correlations at middle latitudes are
greater variability in surface properties and difficulty in
D08206
obtaining accurate retrievals from satellites orbiting at low
inclination.
[18] The decadal changes in OLR and RSW reported by
ERBS are not reproduced by current GCM simulations of
historical climate variability, nor do their time series and
spatial pattern resemble the pattern of ENSO [Allan and
Slingo, 2002]. Although the validity of the ERBS nonscanner decadal record has been questioned [Trenberth,
2002], it appears consistent with independent scanner data
[Wielicki et al., 2002a]. Furthermore, surprisingly similar
OLR and RSW variations are seen in independent radiation
flux calculations based on application of sophisticated
radiative transfer models to ISCCP cloud properties and
other atmospheric constituents [Hatzianastassiou et al.,
2004; Hatzidimitriou et al., 2004; Zhang et al., 2004].
The study of Hatzidimitriou et al. [2004] found that a
decreasing trend in high cloud cover reported by ISCCP
was the primary cause for the increasing trend in OLR
between 1984 and 2000.
[19] Figure 2 displays low- and midlatitude zonal mean
time series of EECRA upper-level, total, low-level, and
cumulus cloud cover separately averaged for ocean during
1952– 1997 and for land during 1971 – 1996. Ocean-only
zonal means were created using ocean-only values from all
grid boxes with at least 50% ocean, and an identical
procedure was used to create land-only zonal means.
Weighting was according to 10 10 grid box area,
irrespective of land/ocean fraction because the ERBS data
does not provide separate values for the contributions from
land and ocean portions of a grid box. As seen in Figures 2a
and 2b, upper-level cloud cover has decreased over low and
northern midlatitude land areas since 1971. Fitting a linear
trend to the time series, the change from 1971 to 1996 is
about 1.5%-sky-cover (Table 2). While it is possible that
increasing anthropogenic haze may have made it more
difficult for surface observers to detect optically thin high
clouds, the consistency between EECRA and ISCCP upperlevel cloud cover during the period of overlap suggests that
the zonal mean downward trends are largely real. Substantial interdecadal variability occurs in upper-level cloud
cover over low- and especially midlatitude oceans, and
the 1990s had less upper-level cloud cover than the 1950s
(Figures 2c, 2d, 2e, and 2f). Although not a good statistical
model for the ocean time series, fitting a linear trend
indicates upper-level cloud cover decreased by 1.3%-skycover between 1952 and 1997 over the near-global ocean
(Table 2). The reduction in upper-level cloud cover is
consistent with the satellite study of Bates and Jackson
[2001], which found a decline in zonal mean upper tropospheric humidity at subtropical and middle latitudes
between 1979 and 1998, although it is possible that the
satellite data might be inhomogeneous. It appears that that
the decrease in upper-level cloud cover reported by previous
studies not only occurs over a larger range of latitudes but
also over a longer period of time.
[20] As previously noted, inconsistent trends occur in
time series of zonal mean ISCCP and EECRA total cloud
cover. These clearly stem from discrepancies in low-level
cloud cover since upper-level cloud cover time series
correspond well, especially where surface sampling is
dense, such as the middle latitudes of the Northern Hemisphere. Examination of regional time series (not shown)
5 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
Figure 2. The 72-day anomalies of upper-level, total, low-level, and cumulus cloud cover averaged
over land between (a) 30S – 30N and (b) 30 –60N and over ocean between (c) 30S – 30N, (d) 30 –
60N, (e) 30– 60S, and (f) 60S – 60N for EECRA (thin red) and ISCCP (thick blue). Only daytime
hours contributed to the time series (VIS/IR detection for ISCCP). EECRA low-level and cumulus cloud
cover are only that not overlapped by higher clouds. Note that the ISCCP ‘‘cumulus’’ type does not
necessarily correspond to visually identified cumulus clouds, but rather is defined as low-level cloudiness
with the least optical thickness. Anomalies are referenced from the 1985– 1989 mean, and 1-2-1
smoothing was applied.
6 of 17
D08206
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
Table 2. Changes in Zonal Mean Upper-Level Cloud Cover From
1971 to 1996 Over Land and From 1952 to 1997 Over Ocean With
95% Confidence Intervalsa
Latitude Zone
NH middle (30 – 60N)
Low (30S – 30N)
SH middle (30 – 60S)
Near-global (60S – 60N)
Land
(%-Sky-Cover)
1.4
1.6
1.7
1.5
±
±
±
±
1.2
1.2
3.7
0.9
Ocean
(%-Sky-Cover)
1.0
1.1
1.9
1.3
±
±
±
±
1.5
1.0
1.8
1.1
a
Changes were calculated by fitting least squares linear trends to the time
series and multiplying by 25.8 years (land) or 45.8 years (ocean). The
effective sample size was assumed to be one-third the nominal sample size.
indicates that surface-satellite disagreement is not geographically uniform and that time series of ISCCP and
surface-observed total and low cloud cover have good
correspondence in certain areas [e.g., Sun, 2003; Sun and
Bradley, 2004]. The decreasing trend in ISCCP low-level
cloud cover at low latitudes primarily results from less
coverage by optically thin low-level clouds (defined as
‘‘cumulus’’ by ISCCP, but not necessarily corresponding
to cumulus identified visually) (Figures 2a and 2c). The
post-1985 increasing trend in EECRA low-level cloud cover
over the ocean primarily results from more coverage by
cumulus clouds (CL = 1, 2, 4) (Figures 2c and 2d). This
increase in low-level cloud cover occurs in the original data
[Norris, 1999] and is not merely a consequence of the
adjustment to ‘‘satellite view.’’ It is interesting to note that
SAGE II reports that the uppermost opaque cloud layer in
the Tropics occurred more frequently at levels below 4 km
during 1995 – 1998 than during 1985 – 1990 [Wang et al.,
2002, Figures 1a and 2]. While some or most of the
apparent low cloud increase may merely be the result of a
decrease in high cloud cover that has allowed the satellite to
see lower in the atmosphere, the downward trend in SAGE
II uppermost opaque cloud frequency above 12.5 km is
actually less than the upward trend in uppermost opaque
cloud frequency below 12.5 km.
[21] The discrepancy between ISCCP and EECRA may
in part result from the differing methods of observation, an
issue particularly germane to cumulus clouds since surface
observers include the sides of cumulus clouds as part of
sky-dome cover [Henderson-Sellers and McGuffie, 1990].
Consequently, variations in cumulus sky-cover seen from
the surface are less related to variations in retrieved satellite
cloud fraction than is the case for layer clouds. This is
consistent with the results of Sun [2003], who found that
ISCCP cloud type anomalies over the United States were
highly correlated with surface-observed anomalies for stratus, stratocumulus, nimbostratus, and cirrus categories but
not the cumulus category. Additionally, Meerkötter et al.
[2004] compared total cloud cover based on NOAA/
AVHRR retrievals with that from surface synoptic reports
over Europe and found that the greatest disagreement
occurred during summer, when the satellite data reported
15% less cloud cover than the surface data. Summer is the
season when cumulus clouds are most prevalent, and it is
likely that many cumulus clouds smaller than the spatial
scale of the satellite pixel are not bright enough to surpass
the detection threshold. Analysis of Landsat images with
30 m pixel size indicates that many trade cumuli will not be
D08206
apparent in the 4– 7 km pixel size used by ISCCP [Wielicki
and Parker, 1992]. The overestimation of cumulus cloud
fraction by surface observers due to inclusion of cloud sides
and the failure by satellites to detect small cumulus underscore the difficulty of measuring cumulus horizontal cloud
fraction.
[ 22 ] Another contributor to this discrepancy is the
presence of identified and unidentified observational artifacts in ISCCP. Campbell [2004] found that a substantial
portion of the downward trend in ISCCP total cloud amount
could be explained by a systematic dependence of cloud
retrievals on view angle. The ISCCP algorithm detects more
cloud cover at high viewing angles than low viewing angles
because the slant path through a cloud is greater at high
viewing angles, thus making optically thin clouds appear
thicker. An increase in the number of geostationary satellites
over time has produced a tendency towards lower viewing
angles at many locations, thus generating an apparent
decline in cloud cover. Further artifacts might also exist
[Campbell, 2004], as suggested by the presence of coherent
changes throughout geostationary satellite fields of view
[Norris, 2000a]. The question then arises: how can the
existence of spurious trends in ISCCP be reconciled with
the close agreement between fluxes based on ISCCP and
fluxes reported by ERBS [Hatzianastassiou et al., 2004;
Hatzidimitriou et al., 2004; Zhang et al., 2004]? Although a
comprehensive and quantitative assessment is beyond the
scope of the present study, the likely explanation is that the
apparently spurious cloud cover variability principally
occurs for optically thin clouds and especially optically thin
low-level clouds (Figures 2a and 2c). In other words, the
largest problems occur for those ISCCP clouds that have the
smallest impact on radiation flux. The decline in average
viewing angle over time has evidently acted to decrease the
average slant path through optically thin clouds such that
marginal pixels identified as cloudy early in the record were
seemingly later identified as clear. A decreasing trend in
low-level cloud cover as well as an increasing trend in the
average optical thickness of the remaining cloudy pixels
(not shown) consequently result, and these two effects
compensate in the flux calculation. The presence of such
an artifact certainly does not exclude the possibility of a
real decline in low-level and total cloud cover, but it will
be difficult to distinguish between real and spurious variability in low-level and total cloud cover until the ISCCP
data are reprocessed to take changing viewing angle into
account.
[23] The suspicious character of the strong increase in
EECRA low-level cloud cover over the ocean (Figures 2c
and 2d) suggests that an artifact may be present in the
surface observations even though no potential causes have
been identified [Norris, 1999]. Time series of ISCCP lowlevel cloud cover and cloud optical thickness regrettably
cannot be used as a benchmark for evaluating surfaceobserved low-level cloud cover for reasons described in
the preceding paragraph. The quality of cloud reports from
Volunteer Observing Ships (VOS) can be evaluated in the
pre-satellite era by comparing them to cloud reports from
nearby Ocean Weather Stations (OWS), which presumably
had observers with better training. Figure A2 shows, in
most cases, that co-located OWS and VOS time series
exhibit similar large decadal variations in upper-level and
7 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
low-level cloud cover. Moreover, midlatitude cloud trends
have been found to be consistent with trends in physically
related parameters [e.g., Norris and Leovy, 1994; Norris,
2000b]. Unfortunately, no OWS were located at low latitudes, so it is not possible to verify the upward trend in
tropical oceanic low-level cloud cover. The increase in
tropical oceanic low-level cloud cover is questionable
because it would act to increase RSW, something not
evident in the ERBS record (Figure 1a). The amount of
potential discrepancy, however, depends on the relative
values of and possible changes in the optical thickness of
low-level clouds, upper-level clouds, and aerosols. Any
possible artifact in surface observations apparently equally
affects total and low-level cloud cover reports since EECRA
and ISCCP upper-level cloud time series substantially agree
even though upper-level cloud cover is partially derived
from low-level cloud cover through the random overlap
equation. The decrease in surface-observed upper-level
cloud cover is not merely due to an increase in low-level
cloud cover; instead it comes from a reported increase in
low-level cloud cover that is larger than the increase in total
cloud cover. Identification of specific artifacts and spurious
variability in surface-observed tropical oceanic low-level
cloud cover is beyond the scope of this study, particularly
because a reliable independent measure of low-level cloudiness is currently not available.
[24] Although zonal mean time series of land cloud cover
presented in Figures 2a and 2b cannot be directly compared
to those from other studies that examined smaller regions
and different time periods, the results are nonetheless
broadly consistent. The previously documented large
increases in cloud cover over several land areas [e.g., Angell,
1990; Henderson-Sellers, 1989; Jones and HendersonSellers, 1992; Karl and Steurer, 1990; Sun and Groisman,
2000; Sun et al., 2001] occur before 1971 and consequently
are not inconsistent with the slightly decreasing total cloud
cover trend displayed in Figures 2a and 2b. In fact, there has
been a large decline in total cloud cover over China since the
mid-1970s [Kaiser, 2000]. The increasing trends in total
cloud cover and low-level cloud cover over the ocean shown
in Figures 2c, 2d, 2e, and 2f are consistent with trends noted
in earlier studies based on similar data [Warren et al., 1988;
Norris, 1999]. The decreasing trend in upper-level cloud
cover, however, contradicts the increasing trends in surfaceobserved midlevel and high-level cloud cover previously
computed by Warren et al. [1988]. Recent calculations on
updated COADS data now show decreasing midlevel cloud
trends for the same time period, in agreement with this
study, and the reason for the previous results has not yet
been identified (S. G. Warren, personal communication,
2004).
4. Estimated Cloud Cover Radiative Forcing
and ERBS Radiation Variations
4.1. Estimation of Cloud Cover Radiative
Forcing Anomalies
[25] The good agreement of EECRA and ISCCP upperlevel cloud cover time series with the ERBS OLR time
series (Figure 1) combined with the demonstrated importance of cloud cover variability to OLR variability (Table 1)
suggests that surface synoptic cloud observations can be
D08206
used to estimate in the pre-satellite era the component of
OLR variability due to upper-level cloud cover variability.
Cloud cover anomalies also make a substantial contribution
to RSW anomalies (Table 1), thus motivating a similar
estimation of that parameter. These estimated radiation
anomalies will be called ‘‘cloud cover radiative forcing
(CCRF) anomalies’’ since they are similar to but not the
same as regular CRF anomalies, which include effects of
anomalies in cloud albedo, emissivity, and other properties
in addition to cloud cover. CCRF anomalies were estimated
from EECRA cloud cover anomalies by multiplying cloud
cover anomalies by LWCRF per unit cloud cover or
SWCRF per unit cloud cover. Since radiative properties of
clouds aggregated over a 72-day period and 10 10 grid
are much more likely to be consistent for a particular season
and geographical location than those for instantaneous local
observations, CCRF anomalies were calculated from 72-day
10 10 cloud cover anomalies rather than individual
synoptic cloud reports.
[26] The following procedures for estimating radiation
flux from surface cloud observations assume a linear
relationship between radiation anomalies and cloud cover
anomalies that is identical to the ratio of climatological
radiation flux to climatological cloud cover. Variations in
cloud albedo, cloud emissivity, and cloud top temperature
are not considered, except for climatological geographical
differences and the seasonal cycle. Previous studies have
documented long-term changes in frequency of different
cloud types over land regions and oceans [e.g., Bajuk and
Leovy, 1998; Sun and Groisman, 2000; Sun et al., 2001; Sun
and Groisman, 2004], implying corresponding changes to
cloud albedo and emissivity, but the present study did not
attempt to estimate the radiative effects of changes in cloud
type since there is not a one-to-one correspondence between
surface-observed cloud type and cloud radiative properties
[Hahn et al., 2001]. Effects of varying greenhouse gases,
aerosols, and other environmental properties are also not
explicitly included, but effects of certain atmospheric constituents, such as water vapor, may be implicitly included if
they happen to co-vary with cloud cover and LWCRF.
[27] To determine the value of LWCRF per unit cloud
cover at each grid box, 1985 – 1989 mean monthly 10 10 values of LWCRF obtained from the ERBE scanner
instrument were divided by 1985 – 1989 mean monthly
10 10 values of upper-level cloud cover. Both parameters used diurnal mean values. The monthly ratios were
then converted to 72-day values by averaging with weighting according to the number of days in each month
contributing to a 72-day period. The conversion from
monthly to 72-day data introduces negligible error because
the seasonal cycle of LWCRF per unit cloud cover is small.
LW CCRF anomalies were estimated for the entire EECRA
time period by multiplying upper-level cloud anomalies by
1985– 1989 mean LWCRF per unit cloud cover at each grid
box. High-level clouds affect OLR more than midlevel
clouds, but insufficient information is available in the
synoptic code to determine sky cover by high-level clouds
separately from midlevel clouds. Low-level cloud anomalies were not considered since they are warm and emit
almost the same amount of OLR as the unobscured surface.
[28] Values of SWCRF per unit cloud cover were separately calculated for upper-level and low-level cloud cover
8 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
because these may have different optical properties and
temporal variability. The individual contributions of upperlevel and low-level clouds cannot be determined from
ERBE SWCRF, so monthly values of ISCCP visible cloud
optical thickness were instead converted to cloud albedo
using ISCCP lookup tables (identical to Figure 3.13 of
Rossow et al. [1996]). This procedure was done for each
ISCCP cloud type, and the cloud albedo values were
averaged together with weighting by the cloud fraction of
each type to produce monthly upper-level cloud albedo and
low-level cloud albedo. Upper-level and low-level cloud
albedo values were averaged from monthly to 72-day and
from 2.5 2.5 to 10 10 in the same manner as
described previously. Weighting by insolation was applied
so that average albedo would be physically consistent with
average RSW. As was the case for LWCRF per unit cloud
cover, converting from monthly to 72-day data causes little
error. Long-term mean values were obtained by averaging
over 1984 –1996.
[29] ERBE broadband albedo values are less than ISCCP
narrowband visible albedo values due to solar absorption at
near IR wavelengths. This was taken into account by scaling
ISCCP upper-level and low-level cloud albedo values by the
ratio of ERBE total cloud albedo divided by ISCCP total
cloud albedo. ERBE total cloud albedo was calculated from
ERBE all-sky and clear-sky albedo using ISCCP total cloud
cover. Another issue is that ISCCP cloud albedo values are
given for clouds over a black surface with no atmosphere,
and RSW is not directly proportional to cloud albedo if the
underlying surface and atmosphere have non-zero albedo.
This effect was addressed by multiplying cloud albedo
values by the factor (1 as)2/(1 as ac), where ac is
cloud albedo over a black surface and as is surface albedo
obtained from ERBE. This admittedly crude adjustment,
derived in Appendix A, accounts for multiple reflections
between the cloud and the surface while assuming no cloud
absorption and an isotropic angular distribution of radiation
through a perfectly transmissive atmosphere. The other
uncertainties in radiative properties of surface-observed
clouds are so large that applying a more sophisticated
radiative transfer scheme would not be worth the effort
required. Grid boxes over the Sahara Desert and Arabian
Peninsula (10 – 30N, 20W – 60E) were excluded from
subsequent zonal averages because it is very difficult to
accurately retrieve SWCRF and estimate SW CCRF when
cloud fraction is small and the surface is bright.
[30] The adjusted cloud albedo values were multiplied by
the mean seasonal cycle of insolation and the mean seasonal
cycles of ISCCP upper-level and low-level cloud fraction at
each grid box to create values of SWCRF due to upper-level
clouds and SWCRF due to low-level clouds. Interannual
variability in insolation was not taken into account because
the effect is very small. The use of daytime means for all
parameters ignores the strong diurnal variation in insolation,
but 6-hourly sampling by surface observations is insufficient to resolve the diurnal cycle. Figure A1 indicates that
long-term variations in the diurnal cycle are weaker than
variations in the diurnal mean. Upper-level and low-level
SWCRF values were divided by 1984 – 1996 mean values of
EECRA upper-level and low-level cloud cover. SW CCRF
anomalies due to upper-level and low-level clouds were
estimated for the entire EECRA period by multiplying cloud
D08206
cover anomalies by SWCRF per unit cloud cover at each
grid box.
4.2. Comparison of Estimated Cloud Cover Radiative
Forcing and ERBS Radiation Flux
[31] The first step to evaluating the quality of estimated
CCRF is comparison to similar observations. Since CRF
measurements are not available during the full 1985 – 1997
time period, let alone CCRF measurements, ERBS all-sky
fluxes are used as a proxy for the component of radiation
variability due to cloud cover variability. Aside from
sampling uncertainty, complete agreement cannot be
expected between estimated CCRF and ERBS fluxes since
changes in non-cloud meteorological properties will have
affected all-sky flux to an unknown extent. Lack of distinction between cloud-related and other contributions to the
ERBS nonscanner fluxes adds some ambiguity to the
following comparisons, but the demonstrated dominance
of clouds in the radiation budget (Table 1) provides strong
evidence that variations in ERBS nonscanner flux are
mainly caused by cloud variations. Note that a positive
cloud cover anomaly produces a positive LW CCRF anomaly, a negative OLR anomaly, a negative SW CCRF
anomaly, and a positive RSW anomaly. For convenience,
CCRF anomalies are plotted with reversed sign so that they
go in the same direction as OLR and RSW anomalies.
[32] Figure 3 displays linear correlation coefficients between ERBS and estimated CCRF anomaly time series for
each grid box. The mid-1991 through mid-1993 time period
was excluded from the correlation calculation due to the
large and non-cloud related radiative signal of Mount
Pinatubo volcanic aerosol in the ERBS data. The substantial
autocorrelation in many of the time series was taken into
account following an approach loosely based upon Zwiers
and von Storch [1995]. Examination of linearly detrended
72-day anomaly time series of upper-level cloud cover
indicates that the effective sample size is at least one third
of the nominal sample size for 96% of the grid boxes. Using
this ratio, correlation values greater than 0.43 are deemed
statistically significant, but this threshold is actually a lower
limit since many areas of the globe have decorrelation times
shorter than seven months. The presence of strong and
significant correlations over a large fraction of the globe
confirms that cloud cover anomalies are a dominant factor
in creating both OLR and RSW anomalies. Slightly higher
correlations exist between estimated CCRF and ERBE CRF
during 1985– 1989 (not shown). The scarcity of ship reports
over the eastern South Pacific and midlatitude Southern
Ocean contribute to the weaker correlations in those
regions, and undersampling of ERBS measurements contributes to weaker correlations at middle latitudes in both
hemispheres. SW correlations are weaker than LW correlations because the ERBS SW data have greater uncertainties
than the LW data. Net upward correlations (not shown) are
much weaker than LW and SW correlations because net
upward radiation is generally a small difference between the
opposing larger LW and SW anomalies, resulting in relatively greater sampling uncertainty.
[33] Figure 4 shows 1985 – 1996 zonal mean time series
of reversed-sign LW, SW, and net CCRF estimated from
EECRA cloud observations and OLR, RSW, and net upward radiation anomalies reported by ERBS. The data were
9 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
D08206
Figure 3. Linear correlation between 72-day CCRF anomalies estimated from EECRA cloud
observations and all-sky flux anomalies reported by ERBS for (a) LW CCRF and OLR and (b) SW
CCRF and RSW. Signs of CCRF anomalies were reversed so that good correlations would be positive.
Correlations were calculated for the 1985– 1996 time period excluding mid-1991 through 1993.
Assuming that the effective sample size is one-third the nominal sample size, correlation coefficients
greater than 0.43 are significant at the 95% level (one-sided).
averaged over land and ocean grid boxes between 30S–
30N and 30 – 60N excluding the Sahara Desert and
Arabian Peninsula (southern middle latitudes are not shown
due to the low density of surface observations). Although
very few 10 10 72-day values were missing, identical
spatial coverage was ensured by averaging only where both
satellite and surface data existed. Table 3 lists linear
correlation coefficients between zonal mean time series of
ERBS flux and estimated CCRF anomalies. The fact that
correlations generally increase with the size of the averaging
region suggests that undersampling by ERBS and surface
observers is the primary reason why the 10 10 grid box
correlations are not higher than they are. Aside from
volcanic episodes, a linear relationship to upper-level cloud
cover explains more than half of the variance in zonal mean
ERBS OLR on multi-year time scales. Attribution of the
OLR increase to an upper-level cloud cover decrease is not
necessarily inconsistent with the decrease in tropical uppermost opaque cloud height reported by SAGE II [Wang et
al., 2002] because a lowering of cloud top height could cooccur with a decrease in surface-observed cloud cover. The
estimated SW CCRF time series resembles the ERBS RSW
time series at northern middle latitudes but not at low
latitudes. This suggests that the discrepancy is related to
the unexplained increase in surface-observed tropical oceanic cumulus cloud cover. Estimated SW CCRF is much
Figure 4. The 72-day anomalies in LW, SW, and net CCRF estimated from EECRA cloud observations
(thin red) and all-sky OLR, RSW, and net upward radiation reported by ERBS (thick black) averaged
over land and ocean regions between (a) 30S – 30N and (b) 30 –60N. Note that the signs of CCRF time
series have been reversed for better comparability. The Sahara Desert and Arabian Peninsula (10 – 30N,
20W – 60E) were excluded from the averages. The mid-1991 through 1993 time period is excluded
since it is greatly affected by Mount Pinatubo volcanic aerosols that have little relation to cloud cover.
Anomalies are referenced from the 1985– 1989 mean.
10 of 17
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
D08206
Table 3. Linear Correlations Between Zonal Mean Estimated
CCRF and ERBS All-Sky Flux Anomaliesa
Latitude Zone
NH middle (30 – 60N)
Low (30S – 30N)
SH middle (30 – 60S)
Near-global (60S – 60N)
Land/Oceanb
LW
SW
Net
land +
ocean
land
land +
ocean
land
land +
ocean
land
land +
ocean
land
0.80
0.77
0.70
0.77
0.79
0.63
0.27
0.29
0.65
0.77
0.79
0.55
0.42
0.39
0.52
0.59
0.68
0.58
0.21
0.25
0.57
0.54
0.63
0.51
0.40
0.45
0.23
0.40
0.29
0.06
0.18
0.23
0.33
0.01
0.04
0.22
ocean
ocean
ocean
ocean
a
Calculated over 1985 – 1996 (land only and land + ocean) or 1985 –
1997 (ocean only), excluding the mid-1991 through 1993 period of Mount
Pinatubo volcanic aerosol influence. Signs of CCRF anomalies were
reversed so that good correlations would be positive. Assuming that the
effective sample size is one-third the nominal sample size, correlation
coefficients greater than 0.43 are significant at the 95% level (one-sided).
The Sahara Desert and Arabian Peninsula (10 – 30N, 20W – 60E) were
excluded from the averages.
b
Ocean-only and land-only zonal means were averaged from 10 10
grid boxes with at least 50% ocean or land area.
closer to ERBS RSW if the tropical Atlantic and eastern
tropical Pacific Oceans are excluded from the zonal average
(Figure 5a). The reason for the large difference in these
particular regions of the ocean (Figure 5b) is not known.
[34] Although the results presented in Figures 4b and 5a
provide substantial support for the co-occurrence of decadal
changes in cloud cover and radiation flux over much of
the globe, the prospect of an artifact in the EECRA raises
the possibility that the agreement between ERBS and the
surface cloud observations may only be coincidental. The
regional discrepancy between ERBS RSW and estimated
SW CCRF, however, is not sufficient to determine the
specific magnitude of any spurious variability in surfaceobserved low-level cloud cover because many other factors
besides cloud cover can affect all-sky flux. For example,
one simplification of the CCRF estimation method is the
D08206
assumption that clouds have unchanging albedo. If the
increase in low-level cloud cover only came from clouds
with relatively low albedo, then the method would overestimate their impact. Similarly, a decrease in zonal mean
cloud albedo could compensate for the increase in zonal
mean low-level cloud cover, or a larger upper-level cloud
albedo could give more radiative weight to the decrease in
upper-level cloud cover. Another possibility is that part of
the decline in ERBS RSW is not cloud related but instead
comes from an increase in absorbing aerosol. An additional
complicating factor is that there may not be a linear
relationship between changes in TOA radiation flux and
changes in surface-observed sky dome cover by cumulus
clouds, which can be as tall as they are wide. Until reliable
independent information on variability in low-level cloud
cover and optical thickness becomes available, it will not be
possible to quantify the magnitude of spurious variability
occurring in surface cloud observations over the tropical
Atlantic and eastern tropical Pacific and perhaps elsewhere.
[35] Despite the shortcomings described in the previous
paragraph, the generally good agreement between estimated
CCRF and ERBS flux in many regions provides confidence
that the historical influence of cloud cover variability
on radiation flux variability can be estimated with some
accuracy over much of the globe. Figure 6 displays zonal
mean time series of reversed-sign estimated LW, SW, and
net CCRF anomalies and ERBS all-sky flux anomalies
separately averaged for ocean during 1952 – 1997 and for
land during 1971– 1996 as in Figure 2. Missing data were
assigned a value of zero and averaged with the rest of the
10 10 values, thus creating a bias in the time series
towards climatology in the absence of data. Actual biases
are quite small since very few data are missing. The Sahara
Desert, Arabian Peninsula, tropical Atlantic Ocean, and
eastern tropical Pacific Ocean were excluded from the
low-latitude average because the estimation of SW CCRF
fails in those regions (Figure 5). An ENSO signal is
also apparent in the estimated low-latitude time series
(Figures 6a and 6c), which exhibit negative LW CCRF
and positive SW CCRF anomalies during warm phases.
Figure 5. As in Figure 4, but averaged over low-latitude (a) Region A and (b) Region B. Region A
comprises 30S–30N except the tropical Atlantic (10 – 30N, 20– 80W and 30S – 10N, 40W –10E)
and eastern tropical Pacific(30S – 10N 80– 130W). Region B comprises the excluded regions.
11 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
D08206
Figure 6. The 72-day anomalies in LW, SW, and net CCRF estimated from EECRA cloud observations
(thin red) and OLR, RSW, and net upward radiation reported by ERBS (thick black) averaged only over
land areas between (a) 30S – 30N and (b) 30– 60N and over ocean areas in (c) low-latitude Region A,
(d) 30 –60N, (e) 30– 60S, and (f) 60S – 60N excluding Region B. Region A and B are defined in
Figure 5. Stars indicate individual values, and 1-2-1 smoothing was applied.
[36] The consistency between ERBS OLR and estimated
LW CCRF suggests the decreasing trend in LW CCRF
evident over low- and northern midlatitude land regions
since 1971 is real (Figures 6a and 6b). Although estimated
SW CCRF does not exhibit as large an anomaly as does
ERBS RSW after 1994 in Figure 6a, the otherwise good
agreement between the time series provides support for
the reality of decreasing trends in cloud cover over tropical
12 of 17
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
D08206
D08206
Table 4. Changes From 1952 to 1997 in Ocean-Only Zonal Mean Estimated CCRF With 95% Confidence
Intervalsa
Latitude Zoneb
Fraction of
Global Area
NH middle (30 – 60N)
Low (30S – 30N)
SH middle (30 – 60S)
Near-global (60S – 60N)
0.09
0.26
0.18
0.53
LW CCRF,
W m2
1.4
3.0
1.7
2.3
±
±
±
±
1.5
1.6
1.9
1.3
SWR CCRF,
W m2
1.2 ± 2.0
+0.5 ± 1.3
0.4 ± 1.8
0.1 ± 1.0
Net CCRF,
W m2
2.7
(2.4
(2.1
(2.4
±
±
±
±
2.5
2.7)
2.0)
1.6)
a
Changes were calculated by fitting least squares linear trends to the time series and multiplying by 45.8 years. Confidence
intervals were calculated from uncertainties in estimated CCRF anomalies combined with the uncertainty in trend fitting.
Uncertainties in estimated CCRF anomalies were determined by comparison to ERBS all-sky flux anomalies. The effective
sample size was assumed to be one-third the nominal sample size. Parentheses indicate trends from time series exhibiting large
systematic differences with ERBS.
b
Excluding the tropical Atlantic (10 – 30N, 20 – 80W and 30S – 10N, 40W – 10E) and eastern tropical Pacific (30S –
10N, 80 – 130W).
land regions previously noted by Hahn et al. [1994].
Estimated zonal mean SW CCRF has slightly weakened
over northern midlatitude land regions between 1971 and
1996 (Figure 6b), but the SW CCRF time series cannot be
used to test the hypothesis that surface solar radiation has
decreased between 1961 and 1990 (i.e., ‘‘global dimming’’)
[Liepert, 2002] because such a reduction, if real, comes
from a change in optical thickness rather than cloud cover.
[37] Marine synoptic cloud reports are available back to
1952, and Figure 6c shows zonal mean time series for the
low-latitude ocean excluding the tropical Atlantic and
eastern tropical Pacific. Multi-year variations in estimated
CCRF and ERBS flux are qualitatively similar, suggesting
that long-term downward trends in tropical oceanic upperlevel cloud cover (Figure 2c) and LW CCRF are real. The
decadal change in LW CCRF is larger than the change in
OLR and the decadal change in SW CCRF is smaller than
the change in RSW, but it is not known whether these
differences stem from an artifact in the observations or are
merely due to contributions of factors besides cloud cover to
ERBS all-sky flux variability. The trend in net CCRF is
sensitive to errors in the estimation of LW and SW CCRF
and is probably not trustworthy. Multi-year variations in
estimated CCRF and ERBS flux averaged over northern
midlatitude oceans exhibit excellent agreement (Figure 6d),
even for decadal changes in net upward flux and net CCRF.
Despite the presence of substantial interdecadal variability,
it appears that upper-level cloud cover and LW CCRF have
generally declined since 1952. Increasing low-level cloud
cover (Figure 2d) has apparently compensated decreasing
upper-level cloud cover to produce a slight overall strengthening (more negative) of SW CCRF since 1952. Although
sampling is poor over the midlatitude Southern Ocean, it
appears that 1952 – 1997 CCRF trends are similar to those
over northern oceans (Figure 6e). For the ocean as a whole
(excluding high latitudes, the tropical Atlantic, and the
eastern tropical Pacific), LW CCRF has become less positive, SW CCRF has exhibited little change, and the trend in
net CCRF is probably not reliable (Figure 6f).
[38] Linear trends are a convenient way to summarize
changes in estimated CCRF between 1952 and 1997,
although by doing so it is not suggested that they are
statistically good models for the time series or that future
changes will continue in the same direction. Table 4 lists
46-year trend values along with 95% confidence intervals
for the oceanic zonal mean time series displayed in Figure 6.
The trends were obtained by the least squares method since
it offers a straightforward procedure for determining confidence intervals [von Storch and Zwiers, 2001], but the more
robust technique of calculating the trend by the median
of pairwise slopes [e.g., Lanzante, 1996] generates very
similar results. The confidence intervals are based on the
independent combination of the uncertainty of trend fitting
and the uncertainty of the estimated CCRF values. Calculation of the standard error by treating estimated CCRF as
the independent parameter and ERBS all-sky flux as the
predicted parameter while assuming a 1:1 slope between
them provided a measure of the uncertainty of an arbitrary
estimated CCRF value. The mid-1991 through 1993
Pinatubo interval was excluded from the calculation. Undersampling and errors in cloud cover and ERBS data, inaccuracies in the estimation method, and radiative effects of
all other cloud and environmental properties except volcanic aerosol contributed to the standard error. All calculations
assumed that the effective sample size was one-third of the
nominal sample size. Decreasing trends in LW CCRF are
statistically different from zero for the tropical ocean and
the near-global mean but not for northern midlatitude
oceans due to substantial nonlinear interdecadal variability
that lowers the level of significance. Increasingly negative
SW CCRF over northern midlatitude oceans combines
decreasingly positive LW CCRF to produce a statistically
significant negative net CCRF trend (corresponding to more
cooling by clouds on the climate system). Significant
negative net CCRF trends occuring at other latitudes
are probably untrustworthy due to the large systematic
differences their time series exhibit with respect to ERBS.
Although cloud cover variability apparently makes the
largest contribution to all-sky flux variability on interannual
and decadal time scales, changes in greenhouse gas concentrations, aerosol characteristics, temperature, humidity,
surface albedo, and other climate parameters are influential
on longer time scales. For this reason, the estimated CCRF
time series presented in this study do not represent changes
in all-sky flux on multidecadal time scales.
5. Summary
[39] This study averaged individual surface synoptic
cloud reports into 72-day 10 10 values of low-level
cloud cover and upper-level cloud cover during 1971 – 1996
over land and 1952 – 1997 over ocean. Upper-level cloud
cover was calculated from total and low-level cloud cover
observations assuming random overlap, and low-level cloud
13 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
cover was adjusted to correspond to that not overlapped by
higher clouds for comparability with satellite data. The data
show that zonal mean upper-level cloud cover at low and
middle latitudes decreased by 1.5%-sky-cover between
1971 and 1996 over land and by about 1%-sky-cover
between 1951 and 1997 over ocean. The good agreement
between surface-observed and ISCCP upper-level cloud
cover variability during the period of overlap suggests that
the multidecadal downward trends in surface-observed
upper-level cloud cover are real. Substantial disagreement
occurs between zonal mean time series of surface-observed
and ISCCP low-level and total cloud cover, due in part to
apparently spurious trends generated by secular changes in
average satellite view angle. Artifacts may also exist in time
series of surface-observed low-level cloud cover, but it is
currently impossible to assess the specific magnitude of
spurious variability since independent and reliable information on variability in low-level cloud cover and cloud
optical thickness is not yet available from ISCCP.
[ 40 ] The generally close correspondence between
surface and satellite observations of upper-level cloud cover
suggests synoptic reports provide a useful record of cloud
cover variability and associated radiative effects in the presatellite era. Anomalies in LW CCRF were estimated from
anomalies in cloud cover by multiplying by the ratio of
long-term mean LWCRF divided by long-term mean cloud
cover at each grid point. Multiyear and decadal zonal mean
variations in estimated LW CCRF are generally consistent
with ERBS all-sky OLR during the period of overlap. Not
taking into account any changes in other cloud and atmospheric properties, the downward trend in upper-level cloud
cover has led to a reduction of zonal mean estimated LW
CCRF over land and ocean at low and middle latitudes.
Between 1952 and 1997, estimated LW CCRF averaged over
most of the global ocean has decreased by about 2 W m2.
[41] Anomalies in SW CCRF were estimated in a similar
manner, and multiyear variations in estimated SW and net
CCRF are consistent with ERBS all-sky RSW and net flux
over northern midlatitude land and ocean regions. Between
1952 and 1997 estimated SW CCRF averaged over
midlatitude oceans has become more negative by about
1 W m2, and estimated SW CCRF averaged over northern
midlatitude land areas has become slightly less negative.
Estimated SW CCRF is less consistent with ERBS all-sky
RSW at low latitudes, particularly over the Sahara Desert,
the Arabian Peninsula, the tropical Atlantic, and the eastern
tropical Pacific. Aside from the problem areas mentioned
above, low-latitude land and ocean regions nonetheless
exhibit a weakening of estimated SW CCRF from the
1980s to the mid-1990s that is consistent with the decrease
in all-sky RSW reported by ERBS. Large systematic differences between estimated net CCRF and ERBE net all-sky
flux preclude meaningful estimates of long-term trends in
net CCRF over low-latitude land and ocean.
[42] The consistency of multi-year variations in surfaceobserved upper-level cloud cover, ISCCP upper-level cloud
cover, estimated LW CCRF, and ERBS OLR provides
substantial support for a large decadal change in tropical
mean cloud cover and radiation flux between the 1980s and
the mid-1990s. Moreover, it appears that a similar decadal
change in upper-level cloud and LW radiation occurred at
middle latitudes as well, and the decrease in upper-level
D08206
Figure A1. Upper-level cloud cover time series for 02–
04, 08 –10, 14– 16, and 20– 22 LT with (a) seasonal means
removed, and (b) diurnal means removed. The cloud data
were averaged from ocean regions between 20– 70E, 110–
160E, 200– 250E, 290– 340E, and 20S–20N. Anomalies are referenced from the 1985 – 1989 mean, and 1-2-1
smoothing was applied.
cloud cover has occurred over a much longer time period.
Balanced against this agreement between three independent
data sets is the existence of identified and probable artifacts
in surface and satellite data that makes reconciliation of
variability in surface-observed and ISCCP low-level cloud
cover and ERBS SW radiation more difficult. Further work
is required to understand the sensitivity of the reported
decadal change to potential artifacts and to remove identified artifacts from the data sets. Additional research is
needed to understand factors contributing to cloud cover
changes over the past fifty years and whether recent trends
will continue in the same direction in the future.
Appendix A
A1. Changes in Diurnal Cycle
[43] It is important to examine the possibility of decadal
changes in the diurnal cycle before analyzing the radiative
14 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
D08206
Figure A2. Daytime-only seasonal anomalies of (a) upper-level cloud cover and (b) low-level cloud
cover for five OWS (thick black) and adjacent 10 10 grid boxes (thin red). OWS locations and grid
box averaging regions are OWS B: 56.5N, 309E and 50– 60N, 300– 320E; OWS C: 52.75N,
324.5E and 50– 60N, 320 –330E; OWS N: 30N, 220E and 20– 40N, 210 –230E; OWS P: 50N,
215E and 40– 60N, 210 – 220E; OWS V: 34N, 164E and 30– 40N, 160 – 170E. Low-level cloud
cover is only that not overlapped by higher clouds. Anomalies are referenced from the 1956 – 1971 mean,
and 1-2-1 smoothing was applied.
effects of decadal changes in diurnal mean cloud cover.
This is especially true for SW reflection by clouds since
there is a strong diurnal cycle in insolation. The large
diurnal amplitude of high cloud cover measured by ISCCP
over much of the tropical ocean [Cairns, 1995; Bergman
and Salby, 1996] motivates a search for possible long-term
variations in the diurnal cycle in surface-observed upperlevel cloud cover. This was carried out by converting
oceanic cloud observations reported at 00, 06, 12, 18 UTC
to local time (LT) and then averaging them within 40 or
50 longitude bands to obtain upper-level cloud cover
averaged within 3-hour intervals for each band. Data from
alternating longitude bands were then averaged together to
provide tropical ocean upper-level cloud cover centered on
00, 06, 12, 18 LT or 03, 09, 15, 21 LT. Figure A1a shows
seasonal upper-level cloud anomalies averaged over the
longitude bands 20– 70E, 110– 160E, 200– 250E, and
290 –340E for 02– 04, 08– 10, 14– 16, and 20– 22 LT.
The time series exhibit similar variations, particularly the
daytime time series, which have many more good cloud
reports than do the nighttime time series. Time series with
the diurnal mean removed, plotted in Figure A1b, have
less variance than those in Figure A1a. This indicates that
changes in the diurnal cycle are weaker than changes in
the diurnal mean. Relative to the diurnal mean, upperlevel cloud cover values for 08– 10 and 14– 16 LT have
slightly increased over time. Upper-level cloud cover for
11 – 13 LT (averaged over 20W – 20E, 70– 110E, 160 –
200E, 250– 290E) also exhibits a slight increase (not
shown). The fact that similar trends occur at all daytime
hours suggests that the daytime mean adequately represents the influence of upper-level cloud variability on SW
reflection. Changes in the diurnal cycle of upper-level
cloud cover are smaller than changes in the diurnal mean
for tropical land and midlatitude land and ocean (not
shown). The same is true for the diurnal cycle of lowlevel cloud cover (not shown).
A2. Comparisons of Ocean Weather Stations and
Volunteer Observing Ships
[44] Zonal average time series of surface-observed cloudiness over the ocean exhibit large decadal variations
(Figures 2c, 2d, 2e, and 2f). ERBS and ISCCP data
corroborate the variations in upper-level cloud cover during
the latter part of the record, but unfortunately few resources
are available for assessing the accuracy of the earlier
EECRA data. One valuable data set is a collection of
synoptic cloud reports from OWS, which made regular
weather observations at fixed locations for several decades
[Norris, 1998; Norris and Klein, 2000]. These data were
excluded from the previous averaging of VOS cloud reports
to gridded values and thus provide an independent measure
of cloud variability. Seasonal values of upper-level and lowlevel cloud cover were calculated for each OWS in the same
manner as for the VOS observations, excepting spatial
averaging. Figure A2 compares time series of upper-level
and low-level cloud cover reported by five OWS with time
series from adjacent 10 10 VOS grid boxes. Values
from more than one 10 10 grid box were averaged
together if the OWS location was near or at the border
between grid boxes. Although OWS time series exhibit
greater decadal variability, they are generally similar to the
gridded VOS time series. Close agreement cannot be
expected since OWS reported cloudiness at a single point
that may not always be representative of the area average.
These results provide confidence in the realism of surfaceobserved cloud variability over northern midlatitude ocean.
[45] No OWS were located at low latitudes, so the
validity of the increase in cumulus cloud cover reported
by VOS cannot be evaluated. In most cases decadal vari-
15 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
ability in island low-level cloud cover time series does not
resemble that for the ocean grid box containing the island.
However, climatological cloud cover reported by island
stations is typically different from that reported by ships
in the nearby open ocean, indicating that a comparison
between the two is probably not meaningful.
A3. Derivation of the Cloud Albedo Adjustment to
Account for Multiple Reflections
[46] The treatment of multiple reflections between the
cloud and the underlying surface was simplified by assuming no cloud absorption and an isotropic angular distribution
of radiation through a perfectly transmissive atmosphere.
For the case of 100% cloud cover,
aall ¼ ac þ ð1 ac Þas ð1 ac Þ þ ð1 ac Þas ac as ð1 ac Þ
þ ð1 ac Þas ac as ac as ð1 ac Þ þ . . .
where aall is the all-sky albedo, ac is cloud albedo over a
black surface, and as is surface albedo. The first term on the
right side of the equation accounts for the initial reflection
by the cloud; the second term accounts for transmission
through the cloud, reflection by the surface, and transmission through the cloud to space; the third term accounts for
transmission through the cloud, reflection by the surface,
reflection back to the surface by the cloud, a second
reflection by the surface, and transmission through the cloud
to space; etc. Summing the infinite series, one obtains,
aall ¼ ac þ as ð1 ac Þ2 =ð1 ac as Þ:
Effective cloud albedo, ac0 , is aall as, or,
a0c ¼ ac ð1 as Þ2 =ð1 ac as Þ:
[47] Acknowledgments. An NSF CAREER award, ATM02-38527,
and a NASA GWEC grant, NAG5-11731, supported this work. T. Wong
graciously provided 36-day ERBS data very similar to those used by
Wielicki et al. [2002b]. ERBE CRF data were obtained from the
NASA Langley Research Center Atmospheric Sciences Data Center
(http://eosweb.larc.nasa.gov). ISCCP data were obtained from the NASA
Goddard Institute for Space Studies (http://isccp.giss.nasa.gov). The author
thanks S. G. Warren and C. J. Hahn for creating the EECRA, T. Wong, B. A.
Wielicki, B. J. Soden, W. B. Rossow, and an anonymous reviewer for
helpful comments, B. D. Norris for editing the manuscript, and countless
observers for making reliable cloud observations over the years.
References
Allan, R. P., and A. Slingo (2002), Can current climate model forcings
explain the spatial and temporal signatures of decadal OLR variations?,
Geophys. Res. Lett., 29(7), 1141, doi:10.1029/2001GL014620.
Angell, J. K. (1990), Variation in United States cloudiness and sunshine
duration between 1950 and the drought year of 1988, J. Clim., 3, 296 –
308.
Bajuk, L. J., and C. B. Leovy (1998), Are there real interdecadal variations
in marine low clouds?, J. Clim., 11, 2910 – 2921.
Barkstrom, B., et al. (1989), Earth Radiation Budget Experiment (ERBE)
archival and April 1985 results, Bull. Am. Meteorol. Soc., 70, 1254 –
1262.
Bates, J. J., and D. L. Jackson (2001), Trends in upper-tropospheric humidity, Geophys. Res. Lett., 28, 1695 – 1698.
Bergman, J. W., and M. L. Salby (1996), Diurnal variation of cloud cover
an their relationship to climatological conditions, J. Clim., 9, 2802 – 2820.
Cairns, B. (1995), Diurnal variations of cloud from ISCCP data, Atmos.
Res., 37, 133 – 146.
Campbell, G. G. (2004), View angle dependence of cloudiness and the
trend in ISCCP cloudiness, paper presented at 13th Conference on
D08206
Satellite Meteorology and Oceanography, Norfolk, Va., 20 – 23 Sept.
(Available at http://ams.confex.com/ams/pdfpapers/79041.pdf)
Cess, R. D., and P. M. Udelhofen (2003), Climate change during 1985 –
1999: Cloud interactions determined from satellite measurements, Geophys. Res. Lett., 30(1), 1019, doi:10.1029/2002GL016128.
Chen, J., B. E. Carlson, and A. D. Del Genio (2002), Evidence for strengthening of the tropical general circulation in the 1990s, Science, 295, 838 –
841.
Hahn, C. J., and S. G. Warren (1999), Extended edited synoptic cloud
reports from ships and land stations over the globe, 1952 – 1996,
NDP026C, 71 pp., Carbon Dioxide Inf. and Anal. Cent., Oak Ridge Natl.
Lab., Oak Ridge, Tenn. (Available at http://cdiac.esd.ornl.gov/epubs/ndp/
ndp026c/ndp026c.html)
Hahn, C. J., S. G. Warren, and J. London (1994), Edited synoptic cloud
reports from ships and land stations over the globe, 1982 – 1991,
NDP026B, 45 pp., Carbon Dioxide Inf. and Anal. Cent., Oak Ridge Natl.
Lab., Oak Ridge, Tenn.
Hahn, C. J., S. G. Warren, and J. London (1995), The effect of moonlight
on observation of cloud cover at night, and application to cloud climatology, J. Clim., 8, 1429 – 1446.
Hahn, C. J., W. B. Rossow, and S. G. Warren (2001), ISCCP cloud properties associated with standard cloud types identified in individual surface
observations, J. Clim., 14, 11 – 28.
Hartmann, D. L., M. E. Ockert-Bell, and M. L. Michelsen (1992), The
effect of cloud type on Earth’s energy balance: Global analysis, J. Clim.,
5, 1281 – 1304.
Hatzianastassiou, N., A. Fotiadi, C. Matsoukas, K. G. Pavlakis, E. Drakakis,
D. Hatzidimitriou, and I. Vardavas (2004), Long-term global distribution
of Earth’s shortwave radiation budget at the top of the atmosphere, Atmos.
Chem. Phys., 4, 1217 – 1235.
Hatzidimitriou, D., I. Vardavas, K. G. Pavlakis, N. Hatzianastassiou,
C. Matsoukas, and E. Drakakis (2004), On the decadal increase in the
tropical mean outgoing longwave radiation for the period 1984 – 2000,
Atmos. Chem. Phys., 4, 1419 – 1425.
Henderson-Sellers, A. (1989), North American total cloud amount variations this century, Palaeogeography Palaeoclimatol. Palaeoecol., 75,
175 – 194.
Henderson-Sellers, A., and K. McGuffie (1988), A quantitative evaluation
of surface-based observations of oceanic cloudiness, Atmos. Res., 21,
241 – 260.
Henderson-Sellers, A., and K. McGuffie (1990), Are cloud amounts estimated from satellite sensor and conventional surface-based observations
related?, Int. J. Remote Sens., 11, 543 – 550.
Hogan, R. J., and A. J. Illingworth (2000), Deriving cloud overlap statistics
from radar, Q. J. R. Meteorol. Soc., 126, 2903 – 2909.
Jones, P. A., and A. Henderson-Sellers (1992), Historical records of cloudiness and sunshine in Australia, J. Clim., 5, 260 – 267.
Kaiser, D. P. (2000), Decreasing cloudiness over China: An updated analysis examining additional variables, Geophys. Res. Lett., 27, 2193 –
2196.
Karl, T. R., and P. M. Steurer (1990), Increased cloudiness in the United
States during the first half of the twentieth century: Fact or fiction?,
Geophys. Res. Lett., 17, 1925 – 1928.
Lanzante, J. R. (1996), Resistant, robust and non-parametric techniques
for the analysis of climate data: theory and examples including applications to historical radiosonde station data, Int. J. Climatol., 16, 1197 –
1226.
Liepert, B. G. (2002), Observed reductions of surface solar radiation at sites
in the United States and worldwide from 1961 to 1990, Geophys. Res.
Lett., 29(10), 1421, doi:10.1029/2002GL014910.
Luo, Z., W. B. Rossow, T. Inoue, and C. J. Stubenrauch (2002), Did the
eruption of Mt. Pinatubo volcano affect cirrus properties?, J. Clim., 15,
2806 – 2820.
Mace, G. G., and S. Benson-Troth (2002), Cloud-layer overlap characteristics derived from long-term cloud radar data, J. Clim., 15, 2505 – 2515.
Mazin, I. P., S. N. Burkovskaya, and E. T. Ivanova (1993), On the climatology of upper-layer clouds, J. Clim., 6, 1812 – 1821.
Meerkötter, R., C. König, P. Bissolli, G. Gesell, and H. Mannstein (2004),
A 14-year European Cloud Climatology from NOAA/AVHRR data in
comparison to surface observations, Geophys. Res. Lett., 31, L15103,
doi:10.1029/2004GL020098.
Moore, B., W. L. Gates, L. J. Mata, A. Underdal, R. J. Stouffer, B. Bolin,
and A. Ramirez Rojas (2001), Advancing our understanding, in Climate
Change 2001: The Scientific Basis, edited by J. T. Houghton et al.,
pp. 769 – 786, Cambridge Univ. Press, New York.
Norris, J. R. (1998), Low cloud type over the ocean from surface observations. Part I: Relationship to surface meteorology and the vertical distribution of temperature and moisture, J. Clim., 11, 369 – 382.
Norris, J. R. (1999), On trends and possible artifacts in global ocean cloud
cover between 1952 and 1995, J. Clim., 12, 1864 – 1870.
16 of 17
D08206
NORRIS: GLOBAL CLOUD AND RADIATIVE CHANGES
Norris, J. R. (2000a), What can cloud observations tell us about climate
variability, Space Sci. Rev., 94, 375 – 380.
Norris, J. R. (2000b), Interannual and interdecadal variability in the storm
track, cloudiness, and sea surface temperature over the summertime
North Pacific, J. Clim., 13, 422 – 430.
Norris, J. R., and S. A. Klein (2000), Low cloud type over the ocean from
surface observations. Part III: Relationship to vertical motion and the
regional surface synoptic environment, J. Clim., 13, 245 – 256.
Norris, J. R., and C. B. Leovy (1994), Interannual variability in stratiform
cloudiness and sea surface temperature, J. Clim., 7, 1915 – 1925.
Ramanathan, V., R. D. Cess, E. F. Harrison, P. Minnis, B. R. Barkstrom,
E. Ahmad, and D. Hartmann (1989), Cloud-radiative forcing and climate:
results from the Earth Radiation Budget Experiment, Science, 243, 57 –
63.
Rossow, W. B., and R. A. Schiffer (1999), Advances in understanding
clouds from ISCCP, Bull. Am. Meteorol. Soc., 80, 2261 – 2287.
Rossow, W. B., A. W. Walker, D. E. Beuschel, and M. D. Roiter (1996),
International Satellite Cloud Climatology Project (ISCCP) documentation
of new cloud data sets, WMO/TD 737, 115 pp., World Meteorol. Org.,
Geneva. (Available at http://isccp.giss.nasa.gov/docs/documents.html)
Sun, B. (2003), Cloudiness over the contiguous United States: Contemporary changes observed using ground-based and ISCCP D2 data, Geophys.
Res. Lett., 30(2), 1053, doi:10.1029/2002GL015887.
Sun, B., and R. S. Bradley (2004), Reply to comment by N. D. Marsh and
H. Svensmark on ‘‘Solar influences on cosmic rays and cloud formation:
A reassessment,’’ J. Geophys. Res., 109, D14206, doi:10.1029/
2003JD004479.
Sun, B., and P. Y. Groisman (2000), Cloudiness variations over the former
Soviet Union, Int. J. Climatol., 20, 1097 – 1111.
Sun, B., and P. Y. Groisman (2004), Variations in low cloud cover over the
United States during the second half of the 20th century, J. Clim., 17,
1883 – 1888.
Sun, B., P. Y. Groisman, and I. I. Mokhov (2001), Recent changes in cloudtype frequency and inferred increases in convection over the United
States and the Former USSR, J. Clim., 14, 1864 – 1880.
Tian, L., and J. A. Curry (1989), Cloud overlap statistics, J. Geophys. Res.,
94, 9925 – 9935.
D08206
Trenberth, K. E. (2002), Changes in tropical clouds and radiation, Science,
296, 2095a.
von Storch, H., and F. W. Zwiers (2001) Statistical Analysis in Climate
Research, 484 pp., Cambridge Univ. Press, New York.
Wang, P.-H., P. Minnis, B. A. Wielicki, T. Wong, and L. B. Vann (2002),
Satellite observations of long-term changes in tropical cloud and outgoing longwave radiation from 1985 to 1998, Geophys. Res. Lett.,
29(10), 1397, doi:10.1029/2001GL014264.
Warren, S. G., C. J. Hahn, J. London, R. M. Chervin, and R. L. Jenne
(1988), Global distribution of total cloud cover and cloud type amounts
over the ocean, NCAR/TN-317+STR, 42 pp. + 170 maps, Natl. Cent. for
Atmos. Res., Boulder, Colo.
Wielicki, B. A., and L. Parker (1992), On the determination of cloud cover
from satellite sensors: The effect of sensor spatial resolution, J. Geophys.
Res., 97, 12,799 – 12,823.
Wielicki, B. A., et al. (2002a), Evidence for large decadal variability in the
tropical mean radiative energy budget, Science, 295, 841 – 844.
Wielicki, B. A., et al. (2002b), Response, Science, 296, 2095a.
Woodruff, S. D., R. J. Slutz, R. L. Jenne, and P. M. Steurer (1987), A
comprehensive ocean-atmosphere data set, Bull. Am. Meteorol. Soc.,
68, 1239 – 1250.
World Meteorological Organization (WMO) (1987), International cloud
atlas, vol. 1, WMO Publ. 407, 212 pp., Geneva.
Zhang, , Y. W. B. Rossow, A. A. Lacis, V. Oinas, and M. I. Mishchenko
(2004), Calculation of radiative fluxes from the surface to top of atmosphere based on ISCCP and other global data sets: Refinements of the
radiative transfer model and the input data, J. Geophys. Res., 109,
D19105, doi:10.1029/2003JD004457.
Zwiers, F. W., and H. von Storch (1995), Taking serial correlation into
account in tests of the mean, J. Clim., 8, 336 – 351.
J. R. Norris, Scripps Institution of Oceanography, University of
California, San Diego, 9500 Gilman Drive, DEPT 0224, La Jolla, CA
92093-0224, USA. (jnorris@ucsd.edu)
17 of 17