GEOPHYSICAL RESEARCH LETTERS, VOL. 39, L03701, doi:10.1029/2011GL050506, 2012
Global cloud height fluctuations measured by MISR on Terra
from 2000 to 2010
Roger Davies1 and Matthew Molloy1
Received 30 November 2011; revised 29 December 2011; accepted 31 December 2011; published 3 February 2012.
[1] Self-consistent stereo measurements by the Multiangle
Imaging SpectroRadiometer (MISR) on the Terra satellite
yield a decrease in global effective cloud height over the
decade from March 2000 to February 2010. The linear trend
is 44 22 m/decade and the interannual annual difference
is 31 11 m between the first and last years of the decade.
The annual mean height is measured with a sampling error of
8 m, which is less than the observed interannual fluctuation in
global cloud height for most years. A maximum departure
from the 10-year mean, of 80 8 m, is observed towards
the end of 2007. These height anomalies correlate well with
the changes in the Southern Oscillation Index, with the
effective height increasing over Indonesia and decreasing
over the Central Pacific during the La Niña phase of the
oscillation. After examining the net influence of Central
Pacific/Indonesia heights on the global mean anomaly,
we conclude that the integrated effects from outside
these regions dominate the global mean height anomalies,
confirming the existence of significant teleconnections.
Citation: Davies, R., and M. Molloy (2012), Global cloud height
fluctuations measured by MISR on Terra from 2000 to 2010,
Geophys. Res. Lett., 39, L03701, doi:10.1029/2011GL050506.
1. Introduction
[2] Changes in cloud properties in response to rising surface temperatures represent some of the strongest, yet least
understood, feedback processes in the climate system. The
radiative forcing due to increased CO2 over one decade
(≈0.28 W m 2) is equivalent in magnitude, for example, to
the shortwave forcing caused by a change of ≈0.3% in global
cloud fraction over the decade. In terms of its effect on
equilibrium surface temperature, the decadal change of CO2
is also equivalent in magnitude to the longwave forcing
caused by a decadal change of ≈19 m in effective cloud
height. (This number is obtained from a radiative-convective
equilibrium model [Davies and Radley, 2009] by perturbing
either the CO2 concentration or the current cloud height
distribution, with no other feedbacks.) By ‘effective cloud
height’, here we mean the expected altitude of the cloud
R∞
height distribution H = f(h)hdh, where f(h)dh is the prob0
ability of a detectable cloud top occurring in the range h to
h+dh. H also includes the probability of no cloud, when
h is the height of the surface. While the effective altitude from
which longwave radiation escapes to space is strongly
1
Department of Physics, University of Auckland, Auckland, New
Zealand.
Copyright 2012 by the American Geophysical Union.
0094-8276/12/2011GL050506
dependent on H, its complete determination would also require
accounting for the emission to space from gases at altitudes
above h. In the sense of radiative-convective equilibrium,
greater values of H imply higher surface temperatures, so that
in this paper the focus is therefore on changes in H over the last
decade, defined as H′ = H
〈H〉, where 〈〉 is the decadal
average. Note that because H is a functional that depends on
the probability of occurrence of cloud at altitude h, changes
in H will occur, for example, when the fraction of high cloud
changes, even though the total cloud fraction remains constant.
Further, H′ is affected more by changes in high cloud fraction
than by equivalent changes in low cloud fraction.
[3] Detection of decadal changes in cloud properties is
challenging. Ideally one seeks a time series long enough
to detect sustained change. While this may be possible for
historically well-measured variables such as surface temperature, it is much harder for global cloud properties. The
International Satellite Cloud Climatology Project [Rossow
et al., 1993], for example, suggests an uncertainty in satellite-derived mean cloud fraction of <10%. Norris and Slingo
[2009] also note the difficulty of sustaining a uniform time
series across multiple changes in satellite platforms. Here we
describe results from a new climate data record of cloud
properties, obtained using the MISR (Multiangle Imaging
SpectroRadiometer) instrument on the Terra satellite [Diner
et al., 1998]. This record is novel in the sense it provides a
geometric measure of H [Moroney et al., 2002]. Unlike more
traditional methods [Nieman et al., 1993], the geometric
technique does not require any assumptions about temperature profile in order to ascertain H. The climate record from
MISR has now passed its 10-year anniversary, allowing the
first decade to be assessed for potential change.
2. The MISR Standard Product for Cloud Height
[4] Measurements of spectral shortwave radiance at
0.67 mm are made at ≈275 m horizontal resolution at nadir
and two near-nadir viewing angles, continuously over the
daylight half of Terra’s sun-synchronous orbit. Most scenes
provide reflectivity patterns that are rapidly compared using a
fast stereo matcher [Muller et al., 2002]. The 14-bit dynamic
range of the MISR detectors ensures that the detection of
contrast patterns is independent of scene illumination or
radiometric calibration changes. The parallax between images from different directions then yields a first-order measure
of the altitude of the reflectivity pattern above the reference
surface (the WGS84 ellipsoid). For precision work involving
regional case studies, the height can also be corrected for the
effects of any along-track wind component, using stereo
matches from more oblique camera pairs [Davies et al.,
2007], or from reanalysis data. After examination, this correction had negligible effect on the interannual variability of
mean heights over large regions, so we use the uncorrected
L03701
1 of 6
L03701
DAVIES AND MOLLOY: MISR CLOUD HEIGHTS
L03701
Figure 1. Deseasonalized anomalies of global effective cloud-top height from the 10-year mean. Solid line: 12-month running mean of 10-day anomalies. Dotted line: linear regression. Gray error bars indicate the sampling error (8 m) in the
annual average.
height fields to determine H. Here we use heights processed
to a horizontal resolution of 2.2 km, referred to as the
Reflecting Layer Reference Altitudes (RLRA). There are
times when the stereo matcher fails to retrieve any height,
usually due to a lack of spatial contrast in the reflectivity.
This typically occurs over cloud-free oceanic scenes. When
this happens, a radiometric cloud mask is referenced and
scenes that are high confidence clear over ocean have their
RLRA assigned to the ocean surface. All other non-retrievals
are flagged as such and are omitted from subsequent analysis.
3. Analysis Approach
[5] We analyzed the entire dataset of RLRA values available from the Level 2 processing for the time period March
2000 to February 2010. These were summarized initially at
the ‘block’ level for each orbit. A ‘block’ represents ≈140 km
in the along-track direction, with an across-track width of
≈380 km. About 140 blocks are sunlit at any given time of the
year. Over 106 RLRA’s are typically found per orbit. Terra
completes 14.56 orbits a day, but there are occasional data
outages due to spacecraft and instrument operations, with an
average 22% loss of processed data. The only month to suffer
a major loss was October 2007, which still had 152 useful
orbits. Overall, over 109 RLRAs went into the analysis.
[6] The height analysis is routine, with data first being
averaged by time-of-year for specific regions over the entire
record. Discrete time-of-year intervals are chosen as the first
and second 10-day intervals for each month, with the third
interval for each month being from the 21st to the end. Height
‘anomalies’ are then defined as the difference between the
average for a given interval and the 10-year average for the
same region and time of year. When regions are integrated
to obtain zonal or global values, the local anomalies are
weighted by area, provided they are sampled more than a
minimum threshold, set here to be 100 samples.
4. Results
[7] The solid line in Figure 1 shows the global H′ as a
function of time, after smoothing using a 12-month running
mean (a 6-month mean at the ends). The 1s sampling error
for anomalies that are averaged over 12 months is 8 m,
indicated by error bars. Overall, the global H changed over
the decade by 44 22 m when a linear trend is fitted (see
auxiliary material), due in part to higher values at the start of
the record and the negative anomaly in late 2007.1 If only the
first and last years are differenced, the change is 31 11 m.
[8] The heights were also examined regionally, averaged
over 20° lat-lon areas. Figure 2 shows the rms interannual
fluctuation in the regional H. The maximum fluctuation is
over 700 m, and is located in the central Pacific (0°N, 180°E).
A secondary maximum, with an interannual fluctuation of
400 m is located in the vicinity of Indonesia (0°N, 122°E).
For use later, we enclose both regions in a so-called Pacific
Box (100°E–130°W, 30°S–30°N). This box encloses just
over 1/6 of the Earth’s surface area.
[9] Using NCEP/NCAR reanalysis data for the same time
period, we calculated the correlation between regional H′
1
Auxiliary materials are available in the HTML. doi:10.1029/
2011GL050506.
2 of 6
L03701
DAVIES AND MOLLOY: MISR CLOUD HEIGHTS
L03701
Figure 2. Regional standard deviations in annual H′ for the 2000–2010 decade. The outlined region of maximum standard
deviations defines the Pacific Box.
Figure 3. Global distributions of effective height relationships: (top) correlation of H′ with anomalies in surface pressure;
(middle) correlation of H′ with surface temperature anomalies; (bottom) correlation of H′ with Southern Oscillation Index.
The color scale indicates the correlation coefficient.
3 of 6
L03701
DAVIES AND MOLLOY: MISR CLOUD HEIGHTS
L03701
Figure 4. Ten-year time series of H′ for the Indonesian region (dashed) and the Central Pacific region (dotted) compared
with the Southern Oscillation Index (solid). The Central Pacific height anomalies have been inverted.
with fluctuations in other key variables, as shown in Figure 3.
Figure 3 (top) shows that for much of the globe H′ has,
unsurprisingly, a strong negative correlation with anomalies
in surface pressure. The correlation with surface temperature
anomalies is more varied, as shown in Figure 3 (middle).
Over much of the ocean, and high northern latitudes, the
correlation is strongly positive. This relationship is reversed
over tropical land, likely associated with monsoon activity.
Figure 5. As for Figure 1, but also showing the contribution to the global anomaly from the Pacific Box (dotted) separately
from the rest of the world (dashed).
4 of 6
L03701
DAVIES AND MOLLOY: MISR CLOUD HEIGHTS
Of particular interest is the correlation in H′ with the Southern
Oscillation Index (SOI, obtained from the SOI archive,
Australian Government Bureau of Meteorology), shown in
Figure 3 (bottom). There are two distinct regions of strong
correlation in the Pacific. One is strongly negative, and centered in the Central Pacific (0°N, 180°E) with connecting
branches to the southeast and northeast across to the Americas. The other is strongly positive, and centered over Indonesia (0°N, 120°E) with a connecting branch to the southeast
that appears coincident with the South Pacific Convergence
Zone.
[10] To further explore the connection with the SOI, we
plot the time series of H′ from the two dominant regions in
Figure 4. In addition to the SOI time series, Figure 4 shows H′
for a (20°lat, 20°lon) region centered at (0°N, 120°E), labeled
Indonesia (dashed), and H′ for a similar region centered at
(0°N, 180°E), labeled Central Pacific (dotted). A 12-month
running mean has been applied to each time series. Because
the correlations with the SOI are reversed for the two regions,
the Central Pacific anomalies have been inverted on the plot
to show the relationship more clearly. There is a remarkably
tight relationship between H′ and the SOI. During the
La Niña phase (positive SOI), the effective height decreases
over the Central Pacific and increases over Indonesia.
[11] Given the tight relationship between SOI and H′, we
broke the global time series of H′ into two components, one
arising inside the Pacific Box, and the other from outside. As
shown in Figure 5, the dip in global H′ at the end of 2007 is
dominated by the ≈5/6 of the planet that is outside the box,
with the bimodal nature of H′ within the box largely cancelling out. The teleconnections (suggested by the coherent
correlation patterns of Figure 3 (bottom)) to heights outside
the box thus appear to significantly affect the anomalies in
global effective height.
5. Summary
[12] MISR has been generating a new climate data record
of globally distributed effective cloud heights, H, as a function of time since the start of consistent data collection in
March 2000. H is obtained from stereo retrievals based on
shortwave reflectivity patterns. The main limitations of the
technique are: a sampling time at only 10:30 am local time;
the omission of thin clouds (cirrus with optical depth less
than ≈0.3); and the omission of very homogeneous cloud
(some anvil cirrus). The main advantages are: the retrieval of
H that is independent of atmospheric temperature profile;
consistent measurement from year to year at the same local
time with no concerns about calibration drift; and the abundance of data at high spatial resolution yielding a very low
sampling error. Because H is due to the topmost non-thin
cloud, changes in effective cloud height (due to changes in
the probability of occurrence of cloud as a function of altitude) over time directly affect the emission to space of
longwave radiation, independent of changes in atmospheric
temperature. This climate data record can be used in several
ways, notably establishing correlations with changes in other
local variables to examine potential feedback mechanisms,
and monitoring the long-term record for hints of secular
trends that act to amplify or dampen the effects of rising
surface temperatures.
[13] When H′ is expressed as deseasonalized 10-day
(Figure 1) or monthly (Figure 5) departures from the 10-year
L03701
mean, the time series shows departures that exceed the
expected sampling uncertainty of the global average. The
main ‘event’ of the decade was a maximum anomaly of
80 m that began in late 2007, coincident with a moderately
strong La Niña. H′ also shows (Figure 3) distinct regional
patterns of correlation with coincident anomalies in surface
pressure and surface temperature, and especially with the
Southern Oscillation Index. The correlations with SOI are
of particular interest, showing distinct regional patterns of
coherent correlation over large areas. The strongest correlations are centered on two regions: positive correlations at
0°N, 120°E (Indonesia); and negative correlations at 0°N,
180°E (Central Pacific). The relationship between SOI
and effective height for the Indonesian and Central Pacific
regions is consistently strong when plotted using a 12-month
running mean (Figure 4).
[14] The coherent nature of these correlations suggests that
the measurement technique is yielding a record that should
prove interesting to other researchers. Because these correlation patterns are large-scale, they should be a useful diagnostic test for a well-constructed dynamic climate model that
can relate changes in dynamic circulation to changes in the
presence and altitude of clouds.
[15] Finally, we note that the climate data record of H
anomalies may ultimately indicate a measure of long-term
cloud feedback that may be quite separate from the correlations discussed above. Ten years is unfortunately too short
a span for any definitive conclusion, as the linear trend in
global cloud height of 44 22 m over the last decade is
partly influenced by the La Niña event, and may prove
ephemeral. The difference between the first and last year
of the decade, not directly affected by the La Niña event, is
31 11 m. If sustained, such a decrease would indicate a
significant measure of negative cloud feedback to global
warming, as lower cloud heights reduce the effective altitude
of emission of radiation to space with a corresponding cooling effect on equilibrium surface temperature. Given the
precision of the MISR measurements, we look forward to the
extension of this climate data record with great interest.
[16] Acknowledgments. We thank the MISR group, especially D. J.
Diner, C. M. Moroney and M. J. Garay, for helpful discussions and data
acquisition. This research was supported by subcontract 1281858 between
the California Institute of Technology/Jet Propulsion Laboratory and Auckland UniServices Limited.
[17] The Editor thanks the two anonymous reviewers for their assistance
in evaluating this paper.
References
Davies, R., and C. Radley (2009), Radiative-convective equilibrium
revisited: The greenhouse effect of clouds, in Current Problems in
Atmospheric Radiation (IRS 2008), edited by T. Nakajima and M. A.
Yamasoe, AIP Conf. Proc., 1100, 667–670.
Davies, R., Á. Horváth, C. Moroney, B. Zhang, and Y. Zhu (2007), Cloudmotion vectors from MISR using sub-pixel enhancements, Remote Sens.
Environ., 107, 194–199, doi:10.1016/j.rse.2006.09.023.
Diner, D. J., et al. (1998), Multi-angle Imaging SpectroRadiometer (MISR)
instrument description and experiment overview, IEEE Trans. Geosci.
Remote Sens., 36, 1072–1087, doi:10.1109/36.700992.
Moroney, C., R. Davies, and J.-P. Muller (2002), Operational retrieval of
cloud-top heights using MISR data, IEEE Trans. Geosci. Remote Sens.,
40, 1532–1540, doi:10.1109/TGRS.2002.801150.
Muller, J.-P., A. Mandanayake, C. Moroney, R. Davies, D. J. Diner, and
S. Paradise (2002), MISR stereoscopic image matchers: Techniques and
results, IEEE Trans. Geosci. Remote Sens., 40, 1547–1559, doi:10.1109/
TGRS.2002.801160.
Nieman, S. J., J. Schmetz, and W. P. Menzel (1993), A comparison of
several techniques to assign heights to cloud tracers, J. Appl. Meteorol.,
5 of 6
L03701
DAVIES AND MOLLOY: MISR CLOUD HEIGHTS
32, 1559–1568, doi:10.1175/1520-0450(1993)032<1559:ACOSTT>2.0.
CO;2.
Norris, J. R., and A. Slingo (2009), Trends in observed cloudiness
and Earth’s radiation budget what do we not know and what do we
need to know?, in Clouds in the Perturbed Climate System, edited by
J. Heintzenberg and R. J. Charlson, pp. 17–36, MIT Press, Cambridge,
Mass.
L03701
Rossow, W. B., A. W. Walker, and L. C. Garder (1993), Comparison of
ISCCP and other cloud amounts, J. Clim., 6, 2394–2418, doi:10.1175/
1520-0442(1993)006<2394:COIAOC>2.0.CO;2.
R. Davies and M. Molloy, Department of Physics, University of Auckland,
Private Bag 92019, Auckland 1142, New Zealand. ([email protected])
6 of 6