Cirrus cloud top temperatures retrieved from radiances in
the National Polar-Orbiting Operational Environmental
Satellite System—Visible Infrared Imager Radiometer
Suite 8.55 and 12.0 ␮m bandpasses
Eric Wong, Keith D. Hutchison, S. C. Ou, and K. N. Liou
We describe what is believed to be a new approach developed for the National Polar-Orbiting Operational
Environmental Satellite System (NPOESS) to retrieve pixel-level, cirrus cloud top temperatures (CTTs)
from radiances observed in the 8.55 and 12.0 ␮m bandpasses. The methodology solves numerically a set
of nonlinear algebraic equations derived from the theory of radiative transfer based upon the correlation
between emissivities in these two bandpasses. This new approach has been demonstrated using NASA’s
Moderate Resolution Imaging Spectroradiometer (MODIS) as a proxy to Visible Infrared Imager Radiometer Suite (VIIRS) data. Many scenes have been analyzed covering a wide range of geophysical
conditions, including single-layered and multilayered cirrus cloud situations along with diverse backgrounds and seasons. For single-layer clouds, the new approach compares very favorably with the MODIS
5 km resolution cloud products; the mean CTT for both methods are very close, while the standard
deviation for the new approach is smaller. However, in multilayered cloud situations, the mean CTTs for
the new approach appear to be colder than the CTTs from MODIS cloud products, which are acknowledged to be too warm. Finally, partly because the new approach is applied at the pixel level, CTTs do not
increase toward cloud edges as is seen in the MODIS products. Based upon these initial results, the new
approach to retrieve improved VIIRS cloud top properties has been incorporated into the ground-based
data processing segment of the NPOESS system where prelaunch testing of all VIIRS algorithms
continues. © 2007 Optical Society of America
OCIS codes: 280.0280, 120.0280.
1. Introduction
The National Polar-Orbiting Operational Environmental Satellite System (NPOESS) satellites will
carry many sensors that collect data from the UV
through the microwave regions of the electromagnetic spectrum. The Visible Infrared Imager Radiometer Suite (VIIRS) is the NPOESS high-resolution
Earth-viewing sensor and has its heritage in three
When this research was performed E. Wong and K. D. Hutchison
were with Northrop Grumman Space Technology, Redondo Beach,
California, USA. K. D. Hutchison ([email protected]) is now
with the Center for Space Research, The University of Texas
at Austin, 3925 West Braker Lane, Suite 200, Austin, Texas 78759,
USA. S. C. Ou and K. N. Liou are with the Department of Atmospheric Sciences, University of California, Los Angeles, California,
USA.
Received 15 June 2006; revised 6 October 2006; accepted 9 October 2006; posted 9 November 2006 (Doc. ID 72016); published 20
February 2007.
0003-6935/07/081316-10$15.00/0
© 2007 Optical Society of America
1316
APPLIED OPTICS 兾 Vol. 46, No. 8 兾 10 March 2007
currently operating sensors, including the National
Oceanic and Atmospheric Administration’s Advanced Very High Resolution Radiometer, NASA’s
Moderate Resolution Imaging Spectroradiometer
(MODIS) onboard Terra and Aqua, and the Operational Linescan Sensor flown by the Defense Meteorological Satellite Program. Like MODIS, VIIRS
will provide highly calibrated data in all channels
collected using four focal plane assemblies. In addition, data collected by the VIIRS sensor will be
used to create nearly 30 environmental data products that will be delivered to a diverse user community in the Department of Commerce and
Department of Defense and at NASA. The products
that will be created include land and ocean surface
products, aerosol products, and a variety of cloud
products, including cloud masks, cloud phase, cloud
optical properties (COP), cloud top parameters (defined as pressure, temperature, and height), cloud
base heights, and cloud cover layers mapped to a
near-constant 6 km gridded field.
A previous publication1 described the baseline
NPOESS–VIIRS solar and thermal IR retrieval algorithms for COP and cloud top parameters. The IR
algorithm used a numerical approach to solve two
radiative transfer equations with three unknowns
based upon a parameterization of absorption coefficients in the M12 共3.70 ␮m兲 and M15 共10.76 ␮m兲
bands.2 (The CO2 slicing method was not used because the design of the VIIRS excludes all bandpasses in MODIS channels 33–36, which are needed
with this heritage algorithm.) This baseline IR algorithm also employed a statistical procedure to determine additional parameters needed to retrieve cloud
top parameters. These parameters were calculated,
based on a minimization method, on a pixel group of
approximately 100 ⫻ 100, which caused an unacceptable blockiness in the retrieved cloud top parameters,
which is shown in the results that follow. To remedy
this problem, Northrop Grumman Space Technology
(NGST) recommended a new approach that has been
developed at the University of California, Los Angeles (UCLA), and subsequently has undergone extensive testing at NGST. This new approach retrieves
cloud top parameters at the pixel level by retrieving
cloud top temperatures (CTTs) using the VIIRS M14
共8.55 ␮m兲 and M16 共12.0 ␮m兲 bands, while eliminating blockiness in the results, reducing processing
time requirement, and improving the accuracy of
CTTs when compared to MODIS results. The algorithm is discussed in this paper, but the complete
theoretical basis will be addressed in a separate one.
This paper focuses on this new algorithmic approach to retrieving the CTT of cirrus clouds with
VIIRS data. Emphasis is placed on the analysis of
single-layered cloud patterns but includes multilayered clouds, which represent the greatest challenge
for the MODIS algorithms used to generate the cloud
products that are available over the Earth Observing
System (EOS) Data Gateway (EDG). Multilayered
cirrus clouds are herein defined as ice clouds over
lower-level water clouds within a single analysis region. This analysis region for VIIRS is the individual
pixel, i.e., nominally 1 km for MODIS and 750 m for
VIIRS. On the other hand, the MODIS algorithms
used to retrieve CTT consider a 5 ⫻ 5 pixel group
as the analysis area. Thus in the following sections
the performance requirements for VIIRS cloud top
parameters are highlighted, along with the VIIRS
algorithms used to retrieve CTTs, which are then
converted to cloud top pressure and height fields using a priori knowledge of atmospheric profiles. In
nighttime situations, CTT is also retrieved along with
COP in cirrus cloud atmospheres. Case studies are
then presented to show the expected accuracy of the
retrieved CTT.
2. Retrieval of Cirrus Cloud Top Temperatures in the
NPOESS Era
Stringent requirements have been established for the
retrieval of CTT with the VIIRS sensor as shown in
Table 1, which is taken from the NPOESS System
Specification, the document that defines acceptable
Table 1. Performance Requirements for the VIIRS Cloud Top
Temperature as Shown in the NPOESS System Specificationa
Paragraph
40.4.9-1
40.4.9-12
40.4.9-2
40.4.9-3
40.4.9-4
40.4.9-5a
40.4.9-5b
40.4.9-5c
40.4.9-6
40.4.9-7
40.4.9-8
40.4.9-9
40.4.9-10
40.4.9-13
40.4.9-14a
40.4.9-14b
40.4.9-15
40.4.9-16
Attribute of the NPOESS Cloud
Top Temperature Product
a. Horizontal cell size
1. Edge of swath
2. Nadir
b. Horizontal reporting
interval
c. Horizontal coverage
d. Measurement range
e. Measurement accuracy
1. Cloud layer optical
thickness ⬎1, water cloud,
day
2. Cloud layer optical
thickness ⬎1, water cloud,
night
3. Cloud layer optical
thickness ⬎1, ice cloud
4. Cloud layer optical
thickness ⬍1 ice cloud
f. Measurement precision
g. Long-term stability
h. Mapping uncertainty,
3 Sigma
i. Maximum local average
revisit time
j. Latency
k. Nadir measurement
uncertainty
1. Water
2. Ice
l. Degraded daytime
measurement condition:
Sun glint ⬍ 36°
m. Excluded measurement
condition: aerosol optical
thickness ⬎ 1.0
Required
Performance
6 km
6 km
HCS
Global
180 to 310 K
2K
3K
3K
6K
1.5 K
1K
1.5 km
3.9 h
3K
5K
Night
performance
a
A dictionary of terms can be found at http://npoess.noaa.gov/
library_NPOESS.html.
performance for the total NPOESS system and each
subsystem. These requirements vary according to
cloud top phase and cloud optical depth with more
stringent requirements levied for cloud optical depths
that exceed unity. For example, water CTT must be
retrieved with an accuracy of 2 K during daytime and
3 K during nighttime conditions. On the other hand,
the accuracy for the retrieval of ice clouds is 6 K if the
cloud optical depth is less than or equal to unity but
3 K otherwise.
The VIIRS cloud mask algorithm, shown at the
beginning of the cloud processing chain in Fig. 1,
provides a pixel-level cloud confidence along with
cloud top phase classification.3 Next, cloud phase is
determined for all pixels classified as confidently
or probably cloudy with an approach described by
Pavolonis and Heidinger.4 The VIIRS cloud phase
algorithm produces seven possible classes, including
water, opaque cirrus, cirrus (single layer), overlap
(cirrus over water clouds), mixed, partly cloudy, and
10 March 2007 兾 Vol. 46, No. 8 兾 APPLIED OPTICS
1317
Fig. 1. (Color online) Architecture of the VIIRS cloud algorithms: COT, cloud optical thickness; CPS, cloud effective particle size; CTH,
cloud top height; CTT, cloud top temperature; SDR, VIIRS sensor data records; VCM, VIIRS cloud mask. Products listed in the boxes filled
with the lightest shade of gray are final products that will be accessible by the user community. Products in black boxes represent inputs
to the algorithms that are shown in boxes in the middle shade of gray.
clear. Partly cloudy is assigned to pixels classified as
probably clear by the VIIRS cloud mask algorithm.
Extensive testing of the cloud phase algorithm supports the following generalizations: (1) The algorithm
accurately differentiates between regions of overlap,
opaque cirrus, and single-layered cirrus clouds, and
(2) it performs very well during daytime conditions
but tends to underspecify cloud overlap during nighttime conditions. The ability to identify regions of
cloud overlap provides highly valuable information
that can be used to better understand the results
obtained by CTT generated by the VIIRS and MODIS
algorithms.
With the specification of cloud phase, the determination of CTT and COP follows one of two paths, i.e.,
one for ice clouds and another for water clouds. In the
case of water clouds and daytime conditions, cloud
effective particle size and cloud optical thickness are
used to determine CTT for all situations including
semitransparent water clouds.5 For nighttime conditions, the CTT of water clouds are retrieved using a
similar IR algorithm as in COP except that it employs
the radiances from 3.7 and 10.76 ␮m channels.1,2 In
addition, clouds classified as mixed phase are treated
as water clouds. However, the treatment of cirrus
CTT with VIIRS data will follow one of two approaches, depending upon whether the data are collected in daytime or nighttime conditions.
A. Theoretical Formulation of a New Infrared Approach to
Retrieve Cirrus Cloud Top Temperatures
The new approach to retrieve cirrus CTT, the mean
effective ice crystal size, and optical depth from the
upwelling radiance of VIIRS cloud retrieval channels follows the principles of the dual-IR-channel
technique.6 – 8 The VIIRS 3.7, 8.55, 10.76, and 12.0
1318
APPLIED OPTICS 兾 Vol. 46, No. 8 兾 10 March 2007
␮m radiances have been selected for use. A major
advantage of using these four channels for cirrus retrievals is that the radiances of these window bands
are less affected by the presence of water vapor than
are the other bands. The retrieval program is based
on the numerical solution of three groups of equations, which are derived from radiative and microphysical parameterizations. Similar principles of
retrieval have been applied to atmospheric infrared
sounder data in the IR region.9
From the theory of radiative transfer, we may express the upwelling radiance at top of the atmosphere
(TOA) for the 3.7, 8.55, 10.76, and 12.0 ␮m (VIIRS
M12, M14, M15, and M16, respectively) channels
over a cirrus cloudy atmosphere in terms of the cirrus
CTT, Tc, and emissivities, ␧i, as follows:
Ri ⫽ 共1 ⫺ ␧i兲Rai ⫹ ␧iBi共Tc兲 i ⫽ 12, 14, 15, 16, (1)
where Rai denotes the upwelling radiance reaching
the cloud base for the two spectral bands and Bi共Tc兲
are the respective Planck functions at Tc. The first
term on the right-hand side of Eq. (1) represents the
contribution of the transmitted radiance from below
the cloud. The second term denotes the emission contribution from the cloud itself. The emission by water
vapor above the cirrus cloud has been neglected. The
effects of cloud reflectivity, which are generally less
than 3% of the incident radiance based on exact
radiative transfer calculations, have also been neglected.
To solve for the cloud top temperature implied in
B16共Tc兲 from Eq. (1) we relate B14共Tc兲 with B16共Tc兲,
correlate ␧14 and ␧16, and statistically determine the
mean clear radiance Rai for the M14 and M16 bands
using clear pixels in the scene. To relate the emissivities for the M14 and M16 channels we follow the
approach proposed by Liou et al.6; with that, we parameterize cirrus emissivities for the two bands in
terms of the visible optical depth, ␶, as follows:
␧i ⫽ 1 ⫺ exp共⫺ki␶兲 i ⫽ 14, 16.
(2)
The exponential term represents the effective
transmissivity. The parameters ki represent the effective extinction coefficients for the two channels
accounting for the effects of multiple scattering. Their
values are obtained from an adding– doubling radiative transfer model that includes multiple scattering
effects.7,10 All ki are smaller than 1 because the effect
of multiple scattering is smaller with IR than with
the visible. Thus the products ki␶ may be considered
as the effective optical depth that would yield the
same emissivity values for the pure absorption conditions at these wavelengths. By eliminating ␶ from
Eq. (2) for M14 and M16, we obtain
共1 ⫺ ␧14兲1兾k14 ⫽ 共1 ⫺ ␧16兲1兾k16.
(3)
Equation (3) correlates ␧14–␧16 directly. A further
combination of Eq. (1) for the M14 and M16 bands
and Eq. (3) leads to the following:
关共R16 ⫺ B16共Tc兲兲兾共Ra16 ⫺ B16共Tc兲兲兴 ⫺ 关共R14 ⫺ B14共Tc兲兲兾
(4)
共Ra14 ⫺ B14共Tc兲兲兴k16兾k14 ⫽ 0.
A similar equation for the M12 and M15 bands, which
is to be used for nighttime cirrus cloud retrievals can
also be obtained as follows:
关共R15 ⫺ B15共Tc兲兲兾共Ra15 ⫺ B15共Tc兲兲兴 ⫺ 关共R12 ⫺ B12共Tc兲兲兾
(5)
共Ra12 ⫺ B12共Tc兲兲兴k15兾k12 ⫽ 0.
B.
Daytime Algorithm for Cirrus Cloud Top Temperature
During daytime conditions, defined as the solar zenith angle ⬍75°, the VIIRS approach to the retrieval
of COP and CTT relies upon two algorithms: a solar
and an IR algorithm. The solar algorithm closely follows the heritage MODIS algorithms.11,12 One exception is the exclusive use of lookup tables (LUTs) with
the VIIRS approach, while the MODIS approach
applies asymptotic theory once clouds reach a sufficiently large optical thickness, e.g., approximately
6. In addition, recent changes have been made in
the MODIS ice crystal habit (VIIRS LUTs assumed
randomly distributed hexagonal ice crystal), which
may eventually cause differences between VIIRS
and MODIS results for cloud optical thickness and
effective particle sizes; however, these changes are
recent and do not affect MOD06 Collection 4 results
in the CTT used in the case studies in this research.
An overview is first provided of the MODIS algorithms used to retrieve the CTT since the results of
the VIIRS algorithms are compared against MODIS
results in the absence of a ground truth database.
There are also two approaches used in the MODIS
algorithms to retrieve the cloud top (temperature
and pressure) parameters that are reported in the
MODIS cloud products. These approaches are briefly
highlighted here, but more information can be obtained from Menzel et al.13 The primary MODIS algorithm for the retrieval of cloud top parameters
relies upon the CO2 slicing method while the alternative approach is based on the 11 ␮m brightness
temperature, i.e., TB11. The retrieval of cloud top parameters with these two MODIS algorithms begins
with the use of global National Center for Environmental Prediction, 1° ⫻ 1° latitude兾longitude, and six
hourly analysis fields that are vertically interpolated
to 101 pressure levels of temperature and water vapor mixing ratio. Next, transmittance profiles are
computed from the atmospheric profiles for each
MODIS band used to retrieve cloud top pressure, i.e.,
MODIS bands 31 and 33–36. No horizontal interpolation is used except for surface temperature and pressure. Next, average radiances in these MODIS bands
are calculated over a 5 ⫻ 5 pixel grouping; then clearsky radiances are determined along with the CO2
slicing computations. A window channel value is obtained, i.e., TB11, then the CO2 slicing solution. If a
CO2 slicing solution is not available, the window
channel result will be reported as the cloud top pressure, which will be one of the 101 pressure levels,
rounded to the nearest 5 hPa increment. The reported cloud top temperature is simply the temperature associated with the pressure level chosen for the
cloud top parameter solution. The CO2 slicing method
is used as long as the cloud signal in the MODIS
13 ␮m band remains sufficiently strong; i.e., cloud
top heights are above approximately 3 km or lower
than approximately 700 hPa.14 The TB11 method assumes a cloud emissivity of unity.13
In the new VIIRS approach, cloud effective particle
diameter 共De兲 and optical thickness are obtained from
LUTs that are based upon MODIS observations in
the VIIRS M5 共0.672 ␮m兲, M8 共1.240 ␮m兲, and M10
共1.610 ␮m兲 bandpasses. Next, the ratio between absorption coefficients between the VIIRS M16 共12.0 ␮m兲
and M14 共8.55 ␮m兲 bandpasses, i.e., the k ratio, is
calculated based upon the retrieved effective particle
diameter of the ice crystals, shown as
k16兾k14 ⫽ 1.596 ⫺ 0.004*De.
(6)
This correlation was obtained through radiative
transfer simulations of various ice cloud models that
specifically include multiple-scattering effects. With
knowledge of the k ratio, cloud top temperature is
solved directly using Eq. (4).
C.
Nighttime Algorithm for Cirrus Cloud Top Temperature
For nighttime conditions, the retrieval of CTT with
the MODIS algorithms is essentially the same as for
daytime except that the ice cloud effective particle
10 March 2007 兾 Vol. 46, No. 8 兾 APPLIED OPTICS
1319
diameter must be determined iteratively. However,
MODIS produces COP products, i.e., optical thickness and particle size only for daytime conditions, not
for nighttime. Since VIIRS requires cloud optical
properties to be retrieved during both daytime and
nighttime conditions, it was necessary to develop a
new formalism to retrieve COP during nighttime conditions. The approach retrieves COP along with CTT
iteratively. Initially, we set the k ratio to 1.1 and
retrieve the cirrus CTT with Eq. (4) using the M16
共12.0 ␮m兲 and M14 共8.55 ␮m兲 bandpasses. Then using the CTT in combination with the radiative equation for M15 共10.76 ␮m兲 we calculate the cloud optical
thickness as in Eqs. (1) and (2). Subsequently, the
effective particle diameter is determined by using the
CTT, the combined radiative equations as in Eq. (5),
and the parameterization equation for the k ratio (in
terms of De) as in Eq. (7). Alternatively, the effective
particle diameter can also be calculated based on a
different parameterization equation as shown in Eq.
(8) in which the mean ice water content, 具IWC典, mean
effective particle size 具De典, mean cloud thickness, ⌬z,
and cloud optical thickness, ␶, are used. This completes one iteration loop, and the CTT is then tested
for convergence. If the convergence criterion is not
met, the k ratio is recalculated as in Eq. (6) using the
previously calculated De. From the cases studied we
found that the iteration typically completes in just a
few cycles. The new approach calculates parameters
such as the k ratio on a pixel level instead of block
level used in the previous baseline algorithm.1,2 Specifically, the approach can be described as follows:
(1) Assume k16兾k14 ⫽ 1.1 and solve for pixel-level
CTT 共Tc兲 using Eq. (4).
(2) Solve Eq. (1) for M15 emissivity based upon
the Tc from step 1.
(3) Solve Eq. (2) for M15 for cloud optical depth (␶),
where k15 ⫽ 0.52.
(4) Solve Eq. (5) for k15兾k12.
(5) Solve the following equation for De,k:
k15兾k12 ⫽
2
兺 bnDe,k⫺n,
n⫽0
(7)
where bn are parameterized coefficients (VIIRS Algorithm Theoretical Basis Document for Cloud Optical
Properties, 2006)
(6) Solve the following equation for De,m:
De,m ⫽ c兵␶兾关⌬z共␣ ⫹ ␤兾De兲 具 IWC典兴其1兾3 具 De 典 .
(8)
(7) Convergence criteria for cloud top temperature
are set in view of VIIRS system specifications, e.g., at
0.5 K for cloud optical thickness less than unity. If
these convergence criteria are met, we have a converged solution for CTT; otherwise, we repeat the
cycle by updating the k ratio using the newly calculated De in Eq. (6).
1320
APPLIED OPTICS 兾 Vol. 46, No. 8 兾 10 March 2007
3. Results
The analyses of MODIS granules are presented to
demonstrate the expected performance of the new
VIIRS CTT algorithms and to compare the results
with MODIS Collection 4 products available over the
EDG. The case studies selected for analysis contain a
variety of cirrus clouds ranging from optically thin
cirrus to opaque (optically thick) and overlap situations. Data are shown for daytime and nighttime
conditions.
The first case study, shown in a true color composite in Fig. 2(a), is a daytime MODIS granule
(MOD2001.032.1750) that shows an extensive region
of single-layer, cirrus clouds over the southwest
United States extending from the Gulf of California
into the southern Rockies. A false-color composite
shown in Fig. 2(b) is displayed with the 1.38 ␮m band
in the red gun of the display device, the 1.6 ␮m band
in the green gun, and the 11.0 ␮m band in the blue
gun. In this color composite, cirrus clouds appear as a
combination of the red and blue guns (purplish hue)
while water clouds appear more turquoise (green and
blue) in color. VIIRS and MODIS cloud phase analyses, shown in Figs. 2(c) and 2(d), respectively, show
general agreement that cirrus clouds are present over
much of the region. The VIIRS cloud phase analysis
shows this cirrus to be predominantly single layered,
i.e., neither overlap nor opaque, as indicated by cloud
phase class 6 displayed as red. MODIS shows only
cirrus cloud (class 4 as turquoise) as the phase with
some regions of uncertainly (class 6 shown as dark
brown).
Figures 2(e) and 2(f) show individual results for the
VIIRS and MODIS CTT, respectively, that were retrieved for the scene, while Figs. 2(g) and 2(h) directly
compare the results from these two sets of analyses.
For these comparisons, the pixel-level VIIRS CTTs
have been aggregated to the 5 ⫻ 5 MODIS analysis
size. Two features are obvious in the results shown in
Figs. 2(e) and 2(f). First, both results appear to be
very similar with CTT ranging between approximately 220 and 245 K. Second, the MODIS analysis
has several regions of much higher temperatures
in the left middle to upper part of the region. Similar temperatures are noticed in the MODIS analysis
at the edge of clouds found across the lower section of
the region.
Figures 2(g) and 2(h) illustrate direct comparisons
between the CTT retrieved with the VIIRS and the
MODIS algorithms. These comparisons contain only
those pixels identified as cirrus clouds by both the
MODIS and VIIRS cloud phase analyses. No water
clouds, mixed phase, or overlap clouds are considered
in these results. A review of Fig. 2(g) shows good
agreement for many analysis regions. However, the
MODIS CTT analyses of CTT that fall in the
250–280 K range are clearly not realistic for cirrus
clouds and show the impact of edge effects. These
unrealistically high CTT may be due to the averaging
of radiances over the 5 ⫻ 5 pixel grouping prior to
retrieving cloud top parameters with the MODIS al-
Fig. 2. (Color online) Results from analysis of single-layered cirrus in daytime data found in MODIS granule 2002.032.1750. (a) True-color
composite of MODIS data; (b) false-color composite showing cloud phase; (c) VIIRS cloud phase analysis; (d) MODIS cloud phase from EDG;
(e), (f) comparison of VIIRS versus MODIS cloud top temperatures; and (g), (h) distributions of VIIRS and MODIS cloud top temperatures
retrieved from their respective algorithms. (i) Blockiness associated with the optimization method used in the original VIIRS algorithms
that are now eliminated by the new VIIRS CTT algorithms.
10 March 2007 兾 Vol. 46, No. 8 兾 APPLIED OPTICS
1321
Fig. 2. (Continued).
gorithm. These effects are subsequently referred to as
edge effects since they appear to be associated with
smaller-scale cloud features. The pixel-level VIIRS
analyses show no such effects. This impact of edge
effects is further seen by examination of the CTT
distributions in Fig. 2(h). While the mean CTT of both
algorithms are very similar, i.e., 232 K for MODIS
versus 231 K for VIIRS, the standard deviation for
the VIIRS algorithm is 7.0 K, while the MODIS algorithm has a standard deviation of 14.6 K. Figure 2(i)
shows the results obtained with the original algorithms delivered to NGST that employed the minimization method, on a pixel group of approximately
100 ⫻ 100. A comparison between the results shown
in Fig. 2(e) and replicated in Fig. 2(i) clearly shows
blockiness in the retrieval that was found to be unacceptable. In addition, results from the original algorithms produced CTT that were substantially
warmer than those found in the results shown for
MODIS and the new VIIRS CTT algorithm.
For the next case study, regions of multilayered
and single-layered cirrus clouds are examined in the
scene shown in Fig. 3. Panel descriptions for this
scene are identical to those shown in Fig. 2 for
MODIS granule 2002.001.0340, which contains data
collected over northwest China. This scene was originally selected to evaluate performance of the VIIRS
cloud mask, since it contains highly diverse backgrounds, including snow, desert, and sparsely and
more densely vegetated land all in one region. Comparisons between the cloud phase analyses in Fig.
3(c) for VIIRS and Fig. 3(d) for MODIS with the color
composite in Fig. 3(b) reveal that the MODIS cloud
mask misclassifies a significant number of pixels in
this scene by calling cloud-free pixels as confidently
cloudy. However, the upper-left quadrant of this figure represents the area of interest for assessing the
performance of the VIIRS and MODIS CTT algorithms. In this region, both algorithms accurately
detect the cirrus clouds, although VIIRS classifies
correctly many of the pixels and overlap, i.e., cloud
phase class 7 shown as dark brown, while other pixels
are classified as cloud phase 6 (single-layered ice
cloud) and shown as red. MODIS calls all the pixels
ice (cloud phase 4 as turquoise) while some are clas1322
APPLIED OPTICS 兾 Vol. 46, No. 8 兾 10 March 2007
sified as uncertain (cloud phase 6 as dark brown).
Next, comparisons are made for all pixels classified as
cloud phase 6 or 7 by the VIIRS algorithm and cloud
phase 4 by the MODIS algorithm. These comparisons
are similar to those presented in Fig. 2, which examined only pixels classified as ice by both algorithms.
Figures 3(e)–3(h) show comparisons between the
VIIRS and MODIS CTT retrievals for those pixels
classified as cirrus clouds by both the VIIRS and
MODIS cloud phase algorithms. Again, the VIIRS
CTTs were aggregated to 5 ⫻ 5 regions for the comparisons shown in these panels. First, it is evident by
comparisons shown in Fig. 3(h) that the MODIS CTT,
contained in Fig. 3(f) are warmer than those produced by the VIIRS algorithms and shown in Fig.
3(e). The distribution of VIIRS CTT is unimodal with
the majority of retrievals lying in the 210–240 K
range and a maximum population near 230 K. On the
other hand, the MODIS distribution is bimodal with
a maximum number of retrievals near approximately
240 K. The mean CTT results for the ice cloud population are 228 K for VIIIRS with a standard deviation of 8 K, while the MODIS results have a mean of
232.5 K and a standard deviation of 12 K.
The bimodal distribution of CTT in the MODIS
results may be attributable to recognized errors inherent in the CO2 slicing algorithm that occur in
multilayered or cloud overlap situations. It is known
that in situations where cirrus and lower-level water
clouds are present within the same analysis area, the
CO2 slicing algorithm may retrieve cloud top pressures with errors up to 100 hPa (Para 3.1.3.a.3,
Menzel et al.).12 Assuming a standard atmospheric
lapse rate, errors of this magnitude could translate to
errors in CTT that are too warm by as much as approximately 15 K. Therefore the bimodal distribution
of MODIS CTT, shown in Fig. 3(h), could result from
the presence of single-layered clouds with a maximum distribution near 220 K and the multilayered
clouds (cirrus over water clouds) with a maximum
distribution near 240 K.
The final case study examines the retrieval of ice
clouds in a multilayered situation in nighttime
MODIS imagery. The scene selected is from MODIS
granule 2002.001.0435 and is located off the coast of
Chile. The results are presented in Figs. 4(a)– 4(h).
Figure 4(a) contains a color composite of the nighttime MODIS imagery constructed by assigning the
3.7 ␮m band to the red gun of the color display, the
11.0 ␮m band to green, and the 12.0 ␮m band to blue.
Water clouds have a pinkish hue and thin cirrus
clouds a bluish hue in the composite. Optically thick
cirrus appears more white than blue. Figure 4(b)
shows a manually generated cloud mask constructed
to assess performance of the VIIRS and MODIS cloud
mask algorithms. The cloud phase analyses shown in
Figs. 4(c) and 4(d) are similar to those shown in Figs.
2 and 3. It is seen from the cloud phase analysis that
the MODIS Collection 4 cloud mask has classified
many more pixels as water clouds than does the VIIRS
Fig. 3. (Color online) Results from analysis of single-layered and overlap cirrus in daytime data found in MODIS granule 2002.001.0340.
(a) True-color composite of MODIS data; (b) false-color composite showing cloud phase; (c) VIIRS cloud phase analysis; (e), (f) comparison
of VIIRS versus MODIS cloud top temperatures; (g), (h) distributions of VIIRS and MODIS cloud top temperatures retrieved from their
respective algorithms.
10 March 2007 兾 Vol. 46, No. 8 兾 APPLIED OPTICS
1323
Fig. 4. (Color online) Results from analysis of single-layered and overlap cirrus in nighttime data found in MODIS granule 2002.001.0435.
(a) False-color composite of MODIS data; (b) manually generated cloud mask used to assess performance of VIIRS and MODIS cloud
masks; (c) VIIRS cloud phase analysis; (d) MODIS cloud phase from EDG; (e), (f) comparison of VIIRS versus MODIS cloud top
temperatures; (g), (h) distributions of VIIRS and MODIS cloud top temperatures retrieved from their respective algorithms.
cloud mask, while the VIIRS cloud mask appears to be
in good agreement with the manually generated cloud
mask.
1324
APPLIED OPTICS 兾 Vol. 46, No. 8 兾 10 March 2007
The results from the cirrus CTT algorithms are
shown in Figs. 4(e)– 4(h) in the manner previously
shown in Figs. 2 and 3 for other MODIS granules. In
this multilayered cloud situation, which is evident
from the color composite shown in Fig. 4(a), the mean
CTT for the VIIRS algorithm is 234 K, while the
mean for the MODIS is 239 K. The standard deviations are 5.6 and 7.5 K, respectively. Therefore, comparisons between the VIIRS and MODIS algorithms
with nighttime data are similar to the results obtained during daytime conditions.
4. Conclusions
A new approach has been developed for the retrieval
of cirrus CTT during the NPOESS era. The new approach relies upon the radiative transfer of ice clouds
measured in the VIIRS M14 共8.55 ␮m兲 and M16
共12.0 ␮m兲 bands that directly yields the CTT. During
nightime conditions, the CTT the and COP are retrieved using an iterative method. During daytime
conditions, however, the COP are retrieved separately using the reflected solar radiation bands, thus
providing the needed effective particle diameter input for direct calculation of the CTT. The previous
baseline IR algorithms relied on the 3.7 ␮m band for
daytime conditions, which contains a large solar component that must be removed. The new approach
does not use the 3.7 ␮m band and therefore is expected to be much more accurate.
The VIIRS algorithms have been used to analyze
numerous MODIS granules, and results have been
compared against those generated by the MODIS
cloud top parameter algorithms. The results from this
initial testing have shown that VIIRS and MODIS
algorithms compare very favorably for single-layered
ice clouds where the MODIS approach is known to
perform well. The differences between the means of
the MODIS and the VIIRS CTT distributions are of the
order of 1 K, although the standard deviation for the
VIIRS algorithms is much tighter than that observed
in the MODIS data. One reason for the larger standard
deviations in the MODIS results appears to be associated with edge effects in the MODIS analyses, where
the CTT for smaller-scale ice cloud fields are unrealistically warmer and in obvious error.
In more complex cirrus cloud situations, where
cirrus clouds lie over lower-level water clouds, the
VIIRS CTTs show no degradation from the results
observed with single-layered cirrus clouds. This is
considered a major achievement since all cloud retrieval approaches assume the presence of a single
cloud in the analysis region. In fact, in these multilayered situations, the MODIS algorithmic approach
is acknowledged to produce errors of up to 100 hPa in
cloud top pressures of the cirrus clouds, which could
result in 15 K errors in the CTT. In these situations,
the mean CTTs of the VIIRS algorithms are colder
than those of MODIS, and the standard deviations
remain smaller. These characteristics in the comparisons between VIIRS and MODIS CTTs are similar in
both daytime and nighttime data.
References
1. S. C. Ou, Y. Takano, K. N. Liou, G. J. Higgins, A. George, and
R. Slonaker, “Remote sensing of cirrus cloud optical thickness
and effective particle size for the National Polar-Orbiting Operational Environmental Satellite System Visible兾Infrared
Imager Radiometer Suite: sensitivity to instrument noise and
uncertainties in environmental parameters,” Appl. Opt. 42,
7202–7204 (2003).
2. S. C. Ou, K. N. Liou, and Y. Takano, VIIRS Cloud Optical
Properties Algorithm Theoretical Basis Document (Raytheon
Information Technology and Scientific Services, 2003), available online at http://www.ipo.noaa.gov/library_NPOESS.html.
3. K. D. Hutchison, J. Roskovensky, J. Jackson, A. Heidinger,
M. J. Pavolonis, R. Frey, and T. Kopp, “Automated cloud detection and typing of data collected by the Visible Infrared
Imager Radiometer Suite (VIIRS),” Int. J. Remote Sens. 20,
4681– 4706 (2005).
4. M. J. Pavolonis and A. K. Heidinger, “Daytime cloud overlap
detection from AVHRR and VIIRS,” J. Appl. Meteorol. 43,
762–778 (2004).
5. H.-L. Huang, Cloud Top Parameters, Algorithm Theoretical
Basis Document, Version 5.0 (Raytheon Information Technology and Scientific Services, 2002), available online at http://
www.ipo.noaa.gov/library_NPOESS.html.
6. K. N. Liou, S. C. Ou, Y. Takano, F. P. J. Valero, and T. P.
Ackerman, “Remote sounding of tropical cirrus cloud temperature and optical depth using 6.5 and 10.5 ␮m radiometers
during STEP,” J. Appl. Meteorol. 29, 716 –726 (1990).
7. S. C. Ou, K. N. Liou, W. M. Gooch, and Y. Takano, “Remote
sensing of cirrus cloud parameters using advanced very-highresolution radiometer 3.7 and 10.9 ␮m channels,” Appl. Opt.
32, 2171–2180 (1993).
8. S. C. Ou, K. N. Liou, Y. Takano, N. X. Rao, Q. Fu, A. J.
Heymsfield, L. M. Miloshevich, B. Baum, and S. A. Kinne,
“Remote sounding of cirrus cloud optical depths and ice crystal
sizes from AVHRR data: verification using FIRE II IFO composite measurements,” J. Atmos. Sci. 52, 4143– 4158 (1995).
9. M. Chahine, H. Aumann, M. Goldberg, L. McMillin, P. Rosenkranz, D. Staelin, L. Strow, J. Susskind, and M. Gunson, AIRS
Team Retrieval for Core Products and Geophysical Parameters,
Algorithm Theoretical Basis Document, Vol. 2.2 (NASA-JPL,
2001).
10. Y. Takano and K. N. Liou, “Solar radiative transfer in cirrus
clouds. Part I and II,” J. Atmos. Sci. 46, 3–36 (1989).
11. T. Nakajima and M. D. King, “Determination of the optical
thickness and effective particle radius of clouds from reflected
solar-radiation measurements. 1. Theory,” J. Atmos. Sci. 47,
1878 –1893 (1990).
12. M. D. King, S.-C. Tsay, S. E. Platnick, M. Wang, and K. N.
Liou, Cloud Retrieval Algorithms for MODIS: Optical Thickness, Effective Particle Radius, and Thermodynamic Phase,
MODIS Algorithm Theoretical Basis Document, No. ATBDMOD-05 (NASA-GSFC, 1997).
13. W. P. Menzel, B. A. Baum, K. I. Strabala, and R. A. Frey, Cloud
Top Properties and Cloud Phase Algorithm Theoretical Basis
Document, MODIS ATBD Reference Number: ATBD-MOD-04
(NASA-GSFC, 2002).
14. S. Platnick, M. D. King, S. A. Ackerman, W. P. Menzel, B. A.
Baum, J. C. Riedi, and R. A. Frey, “The MODIS cloud products:
algorithms and examples from Terra,” IEEE Trans. Geosci.
Remote Sens. 41, 459 – 473 (2003).
10 March 2007 兾 Vol. 46, No. 8 兾 APPLIED OPTICS
1325