validation of cloud property retrievals from mtsat-1r imagery using modis observations
Post on 09-Apr-2017
212 Views
Preview:
TRANSCRIPT
Validation of cloud property retrievals from MTSAT-1R imagery usingMODIS observations
YONG-SANG CHOI†‡ and CHANG-HOI HO‡*
†Department of Earth, Atmospheric and Planetary Sciences, Massachusetts Institute of
Technology, Cambridge, Massachusetts, USA
‡School of Earth and Environmental Sciences, Seoul National University, Seoul, Korea
(Received 21 September 2007; in final form 17 February 2008)
A cloud property retrieval algorithm optimized for five channels (centred at 0.6,
3.7, 6.7, 10.8, and 12.0 mm) has been explored for application to onboard meteor-
ological radiometers on geostationary satellites; however, its validity remains to be
established. Here, we present validation results for the cloud properties retrieved
by the developed algorithm from the full-disk imagery of the Multi-functional
Transport Satellite (MTSAT-1R) for August 2006. The considered cloud proper-
ties include cloud phase (CP), cloud optical thickness (COT), effective radius (ER)
and cloud top pressure (CTP). Their one-month averages, daily variations, and
respective collocated values are compared with the Moderate Resolution Imaging
Spectroradiometer cloud data. Our validation results show that an additional
6.7 mm brightness temperature test in CP retrieval identifies water and ice phases
that may be overlooked in the 10.8- and 12.0-mm bands. Our method to extract
cloud-reflected radiances at the 0.6- and 3.7-mm bands contributes to the accuracy
of the COT for values between 5 and 60, and the ER for values less than 40 mm.
Estimating high-cloud top pressure from the radiance ratio in the 6.7- and 10.8-mm
bands remarkably reduces (by up to 70%) large uncertainties in the CTP, which
may be found in the presence of high thin cirrus clouds.
1. Introduction
Atmospheric parameters retrieved from satellite algorithms contain inevitable uncer-
tainties that may originate from both modelling and physical sources. Reflection or
emittance function look-up libraries generated from radiative transfer (RT) models
for the retrieval algorithm have model-induced biases; further, physical instrument
errors and various atmospheric effects cause significant uncertainties (King et al.
1997). The uncertainties are even amplificatory, particularly in temporally and spa-
tially heterogeneous fields of measurement, such as cloud properties that include the
cloud phase (CP), cloud optical thickness (COT), effective radius (ER), cloud toppressure (CTP), etc. Considering that clouds play a crucial role in the Earth’s hydro-
logical cycle due to their large and varying areal coverage as well as their complex
microphysical properties, the validation of retrieved cloud properties is an essential
task for improving our understanding of the Earth’s weather/climate and guarantee-
ing a reliable product.
*Corresponding author: Email: hoch@cpl.snu.ac.kr
International Journal of Remote SensingISSN 0143-1161 print/ISSN 1366-5901 online # 2009 Taylor & Francis
http://www.tandf.co.uk/journalsDOI: 10.1080/01431160902791887
International Journal of Remote Sensing
Vol. 30, No. 22, 20 November 2009, 5935–5958
Validation of the cloud property retrievals has been performed on a case-by-case
basis by using in situ measurements onboard aircraft and/or by intercomparison
with other collocated data from satellite instruments (e.g. Nakajima and Nakajima
1995, King et al. 1997, Baum et al. 2000). Ground-based measurements of cloud
radiative and microphysical properties using lidars and radars are useful for valida-tion (e.g. Mace et al. 2005). However, these methods for the identification of the
cloud top and optical thickness have significant limitations, such as their insensi-
tivity to small-sized particles at the cloud top and/or the attenuation of the light
beam by low thick clouds (Clothiaux et al. 2000); accordingly, such ground mea-
surements of cloud properties contain additional uncertainty. The Moderate
Resolution Imaging Spectroradiometer (MODIS) instrument, a reliable source of
satellite data, measures various cloud properties via 36 visible and infrared (IR)
spectral bands (King et al. 1997). The prominent capacity for retrieval by MODISdue to the advent of the informative bands has made the MODIS cloud properties
useful references for the validation of the cloud properties retrieved using other
satellites.
To create a weather service with a higher temporal resolution using the
Communication, Ocean and Meteorological Satellite (COMS) imager scheduled to
be launched in 2009 in Korea, a new cloud analysis algorithm (CLA) was explored on
the basis of five channels (centred at 0.6, 3.7, 6.7, 10.8 and 12.0 mm) from weather
imagers in geosynchronous earth orbit (Choi et al. 2007). The algorithm introducesthe following retrieval methods: (i) CP based on spectral absorptivity in the 6.7m m
band in addition to the 10.8- and 12.0-mm bands; (ii) COT and ER using a combina-
tion of cloud-reflected radiances in the 0.6- and 3.7-mm bands through thermal
correction; and (iii) CTP using the brightness temperature in the 10.8-mm band for
optically thick clouds (COT . ,10), and the radiance ratio in the 6.7- and 10.8-mm
bands for optically thin high clouds (CTP � 400 hPa). These three methods were
developed based on RT theory.
The retrieval methods have been previously validated by comparing MODIS clouddata and the collocated cloud properties from CLA by using the MODIS level-2
calibrated radiances (Choi et al. 2007); however, these validation results may indicate
only the uncertainty induced by RT modelling. Because the MODIS radiances have
spectral information within finer bandwidths than those of most five-channel meteor-
ological imagers, the results may not indicate the substantial physical uncertainties
that would emerge from the coarser bandwidths of these imagers. The Japanese
Advanced Meteorological Imager (JAMI) onboard the Multi-functional Transport
Satellite (MTSAT-1R) is also a five-channel weather imager and has been in operationsince 28 June 2005 (Puschell et al. 2006). JAMI has almost the same response func-
tions as the COMS imager, and its orbit and field-of-view (FOV) are similar to those
of the COMS satellite. Therefore, the imagery of JAMI can be used as an input for the
CLA, which produces the cloud properties to be validated.
The present paper attempts to validate the pre-performed CLA algorithm by using
hourly MTSAT-1R imagery via comparison with the MODIS cloud data. A detailed
description of the data and methods is provided in §2. We begin with large-scale and
time-series analyses of cloud products derived for August 2006 (§3). The duration ofone month is determined by considering the limited computational space; however,
hemispheric FOV for the duration considered retains all kinds of sufficient condi-
tions, including surface and cloud types, atmospheric gas profiles, geometry, etc.
Since one of the objectives of the COMS imager is severe weather forecast, we
5936 Y.-S. Choi and C.-H. Ho
subsequently examine the performance of the algorithm during the lifespans of
tropical cyclones (Saomi and Bopha in this study) on the pixel level. The scenes of
tropical cyclones varying dynamically from their generation to decay are overcast by
various types of clouds; this allows investigation of the capacity of the algorithm to
diagnose the stage of a tropical cyclone based on the changes in cloud properties.Finally, a summary and discussion are provided in §4.
2. Data and methods
2.1 MTSAT-1R radiance and MODIS cloud products
We used the hemispheric full-disk images of hourly JAMI/MTSAT-1R calibrated
radiance (or converted brightness temperature) for August 2006 as inputs for the CLA
developed by Choi et al. (2007). The spatial resolution of the JAMI radiances and the
corresponding geometrical angles used in this study are 4 km · 4 km and 0�, 20�, 40�,60� and 80�. The five JAMI onboard spectral channels are centred at 0.725 (hereafter
VIS), 10.8 (IR1), 12.0 (IR2), 6.75 (IR3), and 3.75 mm (IR4). The full-disk image covers
East Asia, West Pacific, Australia, and a part of the Antarctic (80.5� S-80.5� N, 60.4�
E-139.4� W), which may be similar to the COMS imagery, although its final location
has not yet been determined (figure 1).
Figure 1. JAMI/MTSAT-1R radiance imagery for the five spectral channels centred at 0.725(VIS) (a), 10.8 (IR1) (b), 12.0 (IR2) (c), 6.75 (IR3) (d), and 3.75 mm (IR4) (e) for 0333 UTC 7August 2006. Except for the VIS channel, the brighter colour corresponds to a relatively lowvalue in W m-2 sr-1 mm-1. The full-disk imagery covers East Asia, West Pacific, Australia, and apart of the Antarctic region (80.5� S-80.5� N, 60.4� E-139.4� W).
Validation of MTSAT-1R cloud property 5937
To compare with the cloud products from the JAMI imagery on the pixel level, we used
the MODIS cloud product (MOD06, collection 5) data that contain various retrieval
results of cloud properties such as the CP and CTP (COT and ER) at a 5 km · 5 km
(1 km · 1 km) nadir resolution (Platnick et al. 2003). Details of the improvements in the
collection 5 data over those of previous collections can be found elsewhere (Baum et al.
2005, King et al. 2006, Yang et al. 2007). In this study, we selected the MOD06 granules in
the north-western Pacific (10�–30� N, 113�–149� E) during 5–11 August 2006.
The recent CP product in MOD06 is derived in two ways: the bispectral IR test,
and the shortwave IR test. The bispectral IR technique using the 8.5- and 11-mm
bands is a simplified version of the trispectral IR technique (Strabala et al. 1994)
using the 8.5-, 11- and 12-mm bands (Menzel et al. 2006). The shortwave IR test uses
visible, shortwave IR and thermal bands, and is a part of the cloud optical property
module (King et al. 2006). Since the shortwave IR test uses more information, ityields more reliable estimates. However, this is only available during the daytime
and the capability cannot be applied to MTSAT-1R. Thus, we used the MODIS CP
data derived from the bispectral IR test. This CP product includes the following
categories: water, ice, mixed, and uncertain phases.
The COT and ER in MOD06 are the total-column values determined by the
combination of visible channels and a near-IR channel (0.6, 0.8, 1.2, and 2.1 mm).
The minimum detectable COT in MOD06 is about 0.1 (Choi et al. 2005), and the
maximum is set to 100. For ER, the valid retrieval range is within 2 and 30 mm (5 and90 mm) for liquid (ice) clouds. Both COT and ER are retrieved to the second decimal
place. To compare with the MODIS data, the retrieval extent of the COT and ER in
the CLA was set to be the same.
On the other hand, the CTP in MOD06 is retrieved by a CO2 slicing method (also
called radiance ratioing method) using CO2 absorption bands within 13.2–14.4 mm
(Menzel et al. 1983, 2006). The CTP has an interval larger than 10 hPa between 95 hPa
and 1040 hPa, which is finer than the obtainable CLA products (i.e., 50 hPa within
100 hPa and 1000 hPa).The MODIS gridded level-3 daily atmospheric product (MOD08, collection 5) is also
collected for the analysis period. MOD08 contains the 1� · 1� gridded values calculated
from the MOD06 cloud properties. We used MOD08 to calculate the averages for the
analysis period and the time-series of the cloud properties for the given grids.
2.2 Cloudy scene identification
We identified ‘cloudy’ pixels from the images by a bispectral test using the VIS and
IR1 channels before using the CLA to retrieve the cloud properties. The bispectral test
for cloud detection is based on the International Satellite Cloud Climatology Project
(ISCCP) algorithm (Rossow and Garder 1993a), but it is simplified as in the followingrelations:
Clear :ðTclrIR1 � TIR1Þ � �IR and ðLVIS � Lclr
VISÞ � �VIS (1a)
Cloudy :ðT clrIR1 � TIR1Þ>�IR or ðLVIS � Lclr
VISÞ � �VIS; (1b)
where TIR1, TclrIR1, LVIS, and Lclr
VIS are, respectively, the IR1 total-sky brightness
temperature, the IR1 clear-sky brightness temperature, the VIS total-sky radiance,
and the VIS clear-sky radiance. Note that LVIS is a scaled radiance in percentage, as
5938 Y.-S. Choi and C.-H. Ho
used in the ISCCP algorithm. The values of dIR and dVIS are 12.0 K and 6.0% for land
(3.5 K and 3.0% for ocean), respectively. The validity of the cloud detection must be
judged primarily by the accuracy of the clear-sky radiances (see Rossow and Garder
1993b). In this work, T clrIR1 (Lclr
VIS) is obtained as the maximum (minimum) observed
value for each UTC for the 31 days of August 2006. dVIS is the same as that of theISCCP, but dIR is higher than the values given by Rossow and Garder (1993a) (by 6 K
for land and 1 K for ocean) due to higher IR clear-sky values. At night, only the IR
conditions in equation (1) are used.
The cloud fraction in the FOV obtained by this method is 57.3% on average for
August 2006, which is comparable with other measurements of global cloud fractions:
62.7%, 61.2%, 61.4% and 51.8% in the ISCCP C2 dataset (1984–1988), gridded surface
weather station reports (SOBS) (1971–1981), METEOR (1976–1988), and Nimbus-7
(1980–1984) images, respectively (Rossow et al. 1993). It is noted that the cloud fractionidentified using the five channels of MTSAT-1R is much less than that from MODIS
(77.6% for the same period and domain of this study), since the MODIS cloud
discrimination, which includes thin cirrus clouds, is much more robust due to the use
of about 18 different bands and a smaller FOV. The underlying uncertainties in the
cloud mask may exist in the derived MTSAT-1R cloud properties; however, the thin
clouds overlooked by our simple thresholds will have little impact on COT and ER.
2.3 Algorithm description and adjustment
Since MTSAT-1R imagery is used as the input for the CLA, all the look-up libraries
and thresholds in the CLA should be set for the MTSAT-1R spectral response
functions and images (see Puschell et al. 2006). The libraries and thresholds were
recalculated in the same manner as described in Choi et al. (2007) by using the discrete
ordinates radiative transfer (DISORT) models (Stamnes et al. 1988): Streamer
(Key and Schweiger 1998) and Santa Barbara DISORT Atmospheric Radiative
Transfer (SBDART) (Ricchiazzi et al. 1998). In brief, the model limitations to beaddressed are the following. The parameterization of the scattering of nonspherical
ice crystals is imperfect in the two DISORT models. The only gases considered are
water vapour, oxygen, carbon dioxide and ozone. No atmospheric refraction or
spherical geometry is included; therefore, the results for solar zenith angles over
about 70� are subject to errors in Streamer. Further, Streamer exhibits a number of
problems in simulating radiance for near-IR bands (3.4–4.0 mm), since both the
reflected and emitted energy values are factored in (Key 2002). These problems may
affect the CLA retrievals in various ways.The CP algorithm consists of brightness temperature threshold tests using the IR1,
IR2 and IR3 channels. Choi et al. (2007) noted that the IR3 band is a useful alter-
native when the 8.7-mm band is absent; this should be validated in the present study.
Our RT modelling for the MTSAT-1R spectral response function reveals that some
clouds with a mixed phase can have a brightness temperature in the IR3 band (TIR3)
between 228 K and 239K. Thus, the IR3 threshold for ice phase discrimination is
changed to the more rigorous values of 228 K and 239 K from the original 234 K and
250 K in order to prevent false identification of an ice phase from the MTSAT-1Rradiance (compare with table 2 in Choi et al. 2007). The changed criteria and test logic
for determining a cloud phase are shown in table 1.
The COT and ER values are derived simultaneously by using the cloud-reflected
radiance components in the VIS and IR4 channels – the sun reflection method. The
Validation of MTSAT-1R cloud property 5939
desired cloud-reflected components can be decoupled from the undesired ones such as
the ground-reflected components, and the cloud and ground thermal radiances can be
resolved by the rapid method using only one look-up library, which should also be
validated in this study. The COT and ER look-up library comprises the VIS and IR4
radiances, various geometric angles (0�, 20�, 40�, 60� and 80�), and ground albedo
(Ag = 0 and 0.5). Here, we used Ag = 0.5 instead of unity suggested by Choi et al.(2007) because most values of Ag in nature are lower than 0.5 in the MTSAT
FOV. Theoretically, the ground-reflected radiance changes almost linearly in propor-
tion to Ag under the assumption of very small multiple reflections between the ground
and the upper layer (Choi et al. 2007). Once the angular variables and Ag are known
for a given pixel, the simulated thermal-free radiance (i.e. the sum of the cloud- and
ground-reflected radiances) for Ag = 0 is subtracted from that for Ag = 0.5 in the look-
up table and multiplied by a given Ag. In this study, the Ag database used for
individual pixels is the space-dependent (in 2.5�) one-month average from theEuropean Centre for Medium-Range Weather Forecasts, while MODIS uses a
quite sophisticated Ag database with higher temporal (16 day) and spatial (2 km)
resolutions (Moody et al. 2005, 2007).
To increase computational efficiency, the COT and ER libraries were calculated by
using a single wavelength of the peak of the spectral bandwidth, i.e. 0.725 and 3.75
mm. King et al. (1997) described that possible model uncertainties can be induced by
using four factors – a single wavelength, spectrally averaged radiance, a shift in the
central wavelengths, and false optical constants. Among these four factors, the errorin the COT and ER retrievals caused by using a single wavelength is known to be
relatively small if the spectral centre is chosen accurately. While not shown in the
figure, the spectrally averaged radiances weighted with several wavelengths for the
IR4 (or VIS) band have an almost exact linear relationship with the radiance of a
single wavelength. However, this linear relationship is not well represented for solar
and satellite zenith angles greater than or equal to 60�. Therefore, we retrieved the
COT and ER only in the limited daytime FOV with solar and satellite zenith angles
less than 60� (see the second and third panels in figure 2).The CTP is retrieved by a combination of two methods: the IR-window channel
(simply IR1) estimate and the radiance ratioing method. The IR1 estimate is a typical
method in which the TIR1 value is compared with the vertical temperature profile in
the area of interest. This method is valid only for thick clouds (COT� 10) (Choi et al.
2007). In the radiance ratioing method, the observed radiance ratio defined by
LIR1 � LclrIR1
� ��� LIR3 � Lclr
IR3
� �where the subscripts and superscripts have similar
meanings to those used in equation (1) is compared with the simulated radiance
ratio. Since the radiance ratio is a function of the CTP regardless of the opticaldepth, the method compensates for the limitation of the IR1 estimate for thin clouds
(COT , 10). The radiance ratioing method using the IR1 and IR3 bands must be
Table 1. Criteria (required tests) for determining cloud phase.
Ice Mixed Water
TIR1 , 238 K or For no ice For no ice/mixedTIR1-TIR2 � 4.5 K or 238 K � TIR1 , 268 K and TIR1 � 285 K orTIR3 , 228 K 228 K � TIR3 , 239 K TIR3 � 239 K
5940 Y.-S. Choi and C.-H. Ho
100 1000hPa
IceMixedWater
0 100
Saomai
Bopha
0 64µm
(i ) Base products (ii ) Final products
100 1000hPa
(b)
Clo
ud o
ptical th
ickne
ss
IceMixedWater
(a)
Clo
ud p
hase
(d)
Clo
ud
to
p p
ressure
0 100
Saomai
Bopha
Saomai
Bopha
0 64µm
(c)
Eff
ective
ra
diu
s
Figure 2. Cloud properties derived by the CLA from the JAMI level-1b calibrated radiancesshown in figure 1. Cloud phase (a), cloud optical thickness (b), effective radius (c), and cloud toppressure (d) are shown. Base products (i) are the results of conventional methods or withoutcorrection methods, and final products (ii) from improved methods or with the correctionmethods developed in the present study.
Validation of MTSAT-1R cloud property 5941
validated in this study. The CTP library that includes the relation between the
radiance ratio and the CTP is calculated for the standard profiles of the tropics,
mid-latitude summer, and mid-latitude winter (Ellingson et al. 1991) and applied to
the ranges of latitudes 30� S-30� N, 30–60� N, and 30–60� S, respectively.
2.4 Validation methods
Since we wish to validate the methods, two types of cloud products are retrieved,
namely, base products and final products, in a stepwise manner from the previous or
conventional algorithm and the present algorithm, respectively. In detail, the base CP
is retrieved using the IR1 and IR2 channels only, and the final CP using the IR1, IR2
and IR3 channels in order to determine the role of the IR3 test. Similarly, the base
COT and ER are retrieved without using the decoupling method, and the final COTand ER are retrieved using the decoupling method. The CTP is retrieved simply by
IR1 estimate (base CTP), and corrected by the radiance ratioing method (final CTP).
The terms of the products used in this analysis are summarized in table 2.
We compared the base, final and MODIS products by the following four-fold
procedure: (i) scene analysis, (ii) large-scale climatology comparison, (iii) time-series
comparison and (iv) pixel-by-pixel comparison. Scene analysis was the first step to
validate the cloud products with the JAMI imagery. Then, we analysed the large-scale
climatology of the cloud products for August 2006 in comparison with the MOD08data. The climatology was calculated for many restricted conditions such as August
2006, daytime (solar zenith angle , 80�), night-time, liquid clouds, ice clouds, north-
ern and southern hemispheres, and polar (latitude 60–90�), tropical (equator-30�),and mid-latitude regions (30–60�). These conditions are necessary because the average
values of the parameters do not reveal the underlying factors that cause biases.
Next, the daily variation in the cloud products was examined in 1� · 1� regions of
interest for August 2006. The regions were selected considering various surface condi-
tions including land, ocean, desert and snow/ice, as well as different latitudes includingtropics, mid-latitudes and polar regions; the selected nine regions were Seoul, the north-
east China plain, the Gobi Desert, the Tibetan Plateau, the South China Sea, the
Philippine Sea, the East Pacific, the Bering Sea and the Antarctic region.
Table 2. Definitions of terms used in this analysis.
Term Unit Definition
Base CP unitless Cloud phase is determined by TIR1 and TIR1-TIR2 tests intable 1.
Final CP unitless Cloud phase is determined by TIR1, TIR1-TIR2, and TIR3
tests in table 1.Base COT/ER unitless/mm Cloud optical thickness and effective radius are roughly
retrieved by using measured VIS and IR4 radiances thatremain to include both thermal and reflected components.
Final COT/ER unitless/mm Cloud optical thickness and effective radius are retrieved bythe sun reflection method that uses the decoupledradiances, i.e. cloud-reflected components.
Base CTP hPa Cloud top pressure is retrieved by the IR1 window estimate.Final CTP hPa Cloud top pressure is retrieved by both the IR1 window
estimate and the radiance ratioing method using IR1 andIR3 radiances.
5942 Y.-S. Choi and C.-H. Ho
Finally, the retrieved cloud properties were also analysed on the pixel-level scale to
comprehend their error ranges. We compared the MOD06 collection 5 cloudy pixels
and the derived cloud properties from the JAMI images. In this analysis, we selected
the north-western Pacific (10–30� N, 113–149� E) during 5–11 August 2006. Various
types of clouds can be detected since many tropical cyclones passing the area arecharacterized by well-organized deep convection with intense cyclonic winds and
spiral rain bands (c.f. Kim et al. 2006). In order to avoid the possible mismatch due
to the temporal and spatial differences between the bi-daily MODIS and hourly JAMI
images, we allowed a 50-km distance within 30 min, considering the wind trajectories.
Under this condition, we obtained about 2 160 000 cloudy pixels for the analysis (half
for COT/ER). Note that the cloud properties were produced for each given cloudy
pixel at a 4-km resolution by the CLA, while the MOD06 contains the CTP and CP
(COT and ER) at a 5-km (1-km) resolution.
3. Results
3.1 Scene analysis
Figure 1 shows an example of JAMI radiance imagery for 0333 UTC 7 August 2006.Clouds are clearly shown along the intertropical convergence zone (ITCZ) as are
tropical cyclones (Saomi and Bopha centred at 18� N, 139� E, and 23� N, 129� E,
respectively). In the VIS image, optically thicker clouds appear brighter due to solar
light reflection, and they are well distinguished from darker surfaces (e.g., the ocean)
(figure 1(a)). In the IR images, the brighter colour corresponds to relatively lower
values (figures 1(b)–1(e)). Thus, higher clouds appear brighter due to low emitted IR
radiance from the cloud top. It is noted that only high clouds (,,400 hPa) appear
brighter in the IR3 channel due to the strong water vapour absorption in the low- andmid-troposphere, while low clouds are distinguishable in the atmospheric window
channels, i.e. IR1 and IR2. The IR4 radiance is generally higher for smaller-sized
cloud particles and liquid clouds than for larger particles and ice clouds if the other
conditions are identical.
Allowing for the spectral characteristics of the five images, we may infer three major
features in the example: (i) the presence of very high and optically thick clouds in the
tropical West Pacific, including the known tropical cyclones Saomai and Bopha
(characterized by high VIS, low IR1, and low IR2 radiances); (ii) the presence ofhigh thin clouds in the East Pacific (characterized by low VIS, low IR1, and low IR2
radiances); and (iii) the presence of widely spread low thin clouds and coexistent high
thick clouds over the south-western seas of Australia (characterized by the widespread
region of low VIS, high IR1, and high IR2 radiances, and the branch-shaped area of
high VIS and low IR3 radiances running through the widespread cloud fields). Later,
we will relate the inferred features in points (i) to (iii) with the cloud properties
retrieved by the CLA.
Figure 2 shows the cloud properties retrieved by the CLA for the same period as infigure 1. The CP (a), COT (b), ER (c), and CTP (d) are retrieved in a stepwise fashion
from the base products (i) to the final products (ii). Overall, noticeable differences
exist between the base and final products. The final CP shows more water clouds and
fewer uncertain clouds (white) over the globe than the base CP (the first panel). The
final COT (ER) values in the centres of tropical cyclones and ITCZ are larger than the
base COT (ER) (the second and third panels). More clouds have low CTP (i.e. higher
top) in the final product in the East Pacific than in the base product (the fourth panel).
Validation of MTSAT-1R cloud property 5943
These differences make the three abovementioned features inferred from the radiance
scenes in figure 1 more noticeable in the final products than in the base products. Very
high and thick clouds are found more clearly in the tropical West Pacific in the final
products than in the base products. Although the optical properties of some clouds
are not retrieved due to high geometric angles, the final CTP and COT more obviouslyshow the high thin clouds in the East Pacific and the branch-shaped high thick clouds
over the south-western seas of Australia than the base products.
3.2 Large-scale climatology comparison
In this section, the retrieved cloud properties are compared quantitatively with the
MODIS retrievals for August 2006. Table 3 shows the one-month climatology of the
final, base, and MODIS CP. Overall, there are large differences between the derived CPand the MODIS CP. Water and ice clouds are noticeably more prevalent in the final CP
than the base CP, and their proportion is closer to that in the MODIS CP. This indicates
that the additional water and ice clouds in the final CP are identified by adding the IR3
test to the conventional IR1 and IR2 tests. Without the IR3 test, a large number of
cloudy pixels remain undecided. This improvement is actually seen regardless of the
local time and region, as shown in table 3. Meanwhile, mixed clouds are relatively
smaller in the final CP than in the base CP. This reduction in mixed clouds should be
due to the additional IR3 test. Despite these improvements in the final CP, a largeamount – to the extent in the MODIS CP (roughly 15%) – of clouds remain unknown.
The characteristics of the CP in the polar regions where the averaged temperature is
less than -30�C at all levels (-40�C over 700 hPa) are noteworthy. Our final CP has a
large portion of ice and mixed phases in this region (54.8 + 36.3 = 91.1%) and a
negligible portion of water phase (0.7%). Interestingly, the MODIS bispectral IR
algorithm provides a much larger number of water clouds (28.4%) and large unknown
clouds (17.4%). Some middle-level clouds with temperatures greater than about
-20�C can contain supercooled water droplets that coexist with ice particles(Liou 2002). Recent reports from field/aircraft-based experiments in the Arctic also
indicate that cloud particles can exist in the liquid phase at temperatures down to
-30�C (Shupe et al. 2006, Verlinde et al. 2007). However, this remains to be clarified in
further analysis with more observational evidence.
Table 3. Comparison of cloud phases from the JAMI/MTSAT-1R and the MODIS algorithm inAugust 2006. The numbers indicate the percentage of cloud phase (water/ice/mixed/uncertain)
over the total cloud fraction. All results are calculated in the FOV of JAMI.
MTSAT CP
Domain Final Base MODIS CP
GlobalAugust 39.4/24.6/20.5/15.5 20.9/19.1/29.2/30.8 50.3/29.5/5.6/14.7Day 34.2/28.2/18.2/19.4 20.9/21.6/26.0/31.4 53.2/26.7/5.1/15.0Night 34.5/29.2/21.1/15.2 16.0/20.4/32.9/30.8 46.1/33.4/6.4/14.2
Northern hemisphere 30.2/29.1/22.3/18.3 21.8/23.7/29.4/25.1 48.1/32.4/4.4/15.1Southern hemisphere 50.6/19.0/18.2/12.2 19.9/13.5/29.0/37.6 52.5/26.5/6.7/14.2Polar 0.7/54.8/36.3/8.2 0.2/31.5/59.8/8.5 28.4/40.3/14.1/17.4Mid-latitude 28.9/19.6/29.8/21.7 8.8/12.9/40.4/37.9 52.7/23.8/6.8/16.7Tropical 46.1/25.7/15.3/12.9 27.7/21.6/22.5/28.2 55.1/30.6/2.0/12.3
5944 Y.-S. Choi and C.-H. Ho
In the mid-latitudes and tropics, a similar contrast is also revealed (i.e. large ice and
mixed clouds in the derived CP versus large water clouds in MODIS). Moreover,
mixed clouds hold a larger portion in the derived CP than in the MODIS CP, and
clouds of this type were possibly flagged as being in the water phase in MODIS. The
high proportion of mixed phase in the derived CP as compared with that inthe MODIS CP is perhaps because the key channel (e.g. 8.5 mm), which is used in
the MODIS algorithm to identify a mixed phase, is absent in our algorithm. Since it is
known that the MODIS CP algorithm fails to identify high or cirrus clouds as
belonging to the ice phase (Menzel et al. 2006), the comparison in this study recon-
firms that contemporary CP algorithms from passive IR radiometers are still
imperfect.
Despite large differences between the final CP and the MODIS CP, the differences
are comparable to some extent under reciprocal conditions. The night-time CP hasmore clouds of the ice phase than the daytime CP in both the final and the MODIS
data, indicating the presence of a diurnal variation in CP. Further, the northern
hemispheric CP has more of the ice phase than the southern hemispheric CP in both
products. This is because the ITCZ is located in the northern hemisphere in August
(see figure 2). However, these differences cannot be seen in the base CP.
The relative frequency of the one-month climatology of the COT is examined for all
the clouds for various conditions (figure 3). The bin size is set to a value based on the
criteria of the COT for the ISCCP cloud classification (Rossow and Schiffer 1999).The dotted, solid and thick solid lines indicate the base, final and MODIS products,
respectively. The derived COT is generally underestimated in comparison with the
MODIS COT. The highest frequency is seen in bins 4–9 and 9–25 in the final and
MODIS COT, respectively. The frequency of ‘COT , 4’ is greater in the derived data
than in the MODIS COT data, while the frequency of ‘COT = 9–25’ is less in the
derived data than in the MODIS COT data in all cases. This systematic bias may
result from the modelling calculation, and it should be minimized by tuning the COT
library.The bias can be also induced by the different reflectivity between liquid and ice
clouds. In figure 3, the bias is much larger for liquid clouds than for ice clouds. Water
droplets reflect solar insolation more effectively than ice crystals; consequently, the
reflected radiance from the liquid clouds is greater than that from the ice clouds for
the same COT. Accordingly, the CLA can underestimate the COT if ice clouds are
falsely identified as liquid clouds. This indicates the important role of the CP classi-
fication result in the COT retrieval.
A similar analysis result is shown for the ER in figure 4. Our derived ER is slightlyless than the MODIS ER. However, the bias between them is relatively small in
comparison with that of the COT (compare figures 3 and 4). In comparison with
the liquid and ice phases, the frequency of ER in both the datasets shows consistently
smaller liquid particles with ER between 2 mm and 30 mm and larger ice crystals with
ER between 5 mm and 64 mm. The highest frequency is shown in the bins 0–20 mm and
20–30 mm for liquid and ice clouds, respectively. On the other hand, no noticeable
difference is found between the ER values of the northern and southern hemispheres.
While the bias is relatively large in the tropics, we note that in the mid-latitudes, therelative frequencies in the final and MODIS data show fairly good agreement. Since
ER retrieval is highly sensitive to the IR4 reflected radiance, the results show that our
method may be particularly effective in the mid-latitudes to decouple the thermal and
surface-reflected components from the observed IR4 radiance (Choi et al. 2007).
Validation of MTSAT-1R cloud property 5945
With regard to the comparison between the final and the base products, the one-
month climatology of the COT and ER agrees with those of the MODIS products to
some extent due to the decoupling method employed in the algorithm (compare the
dotted and solid lines). By using the decoupling method, the frequency of thin clouds
(COT , 4) decreases and that of thick clouds (COT � 4) increases, except for iceclouds (figure 3). Similarly, the frequency of clouds of small particles (ER , 10)
decreases and that of clouds of large particles (ER � 10) increases (figure 4). Since
clouds with optical depths between 1 and 9 are particularly significant in the tropical
energy balance (Choi and Ho 2006), the change in the corresponding cloud amounts
by the decoupling method can give rise to a change in the estimation of the cloud
radiative effect in the tropics. Clearly, the decoupling method contributes to improv-
ing not only the COT and ER retrievals but also our understanding of cloud radiative
effects.
0~1 1~2 2~3 3~4 4~9 9~25 25~60 60~
(a) August
Cloud optical thickness
Fre
qu
en
cy
(%)
0
10
20
30
40
Simple
Decoupled
MODIS
(b) Liquid
Cloud optical thickness
0~1 1~2 2~3 3~4 4~9 9~25 25~60 60~
Fre
qu
en
cy
(%)
0
10
20
30
40
(c) Ice
Cloud optical thickness
0~1 1~2 2~3 3~4 4~9 9~25 25~60 60~
Fre
qu
en
cy
(%)
0
10
20
30
40
(d) NH
Cloud optical thickness
0~1 1~2 2~3 3~4 4~9 9~25 25~60 60~
Fre
qu
en
cy
(%)
0
10
20
30
40
(e) SH
Cloud optical thickness
0~1 1~2 2~3 3~4 4~9 9~25 25~60 60~
Fre
qu
en
cy
(%)
0
5
10
15
20
25
30
35
( f ) Mid-latitude
Cloud optical thickness
0~1 1~2 2~3 3~4 4~9 9~25 25~60 60~
Fre
qu
en
cy
(%)
0
10
20
30
40
( g ) Tropical
Cloud optical thickness
0~1 1~2 2~3 3~4 4~9 9~25 25~60 60~
Fre
qu
en
cy
(%)
0
10
20
30
40
Figure 3. Relative frequency (in %) of cloud optical thickness without using the decouplingmethod (i.e. base products), using the decoupling method (i.e. final products), and MODIS datato the total clouds for the corresponding conditions. SH and NH stand for the northern andsouthern hemispheres, respectively.
5946 Y.-S. Choi and C.-H. Ho
The one-month climatology of the CTP is also compared with the MODIS CTP for
the analysis period (figure 5). It is noted that our CTP and the MODIS CTP are
produced in the units of 50 and 25 hPa, respectively. In the analysis of the MODIS
CTP (figure 5(a)), the CTP frequency is highest at the 300–400, 700–800 and 900–1000
hPa levels in August. This distribution varies between day and night, northern and
southern hemispheres, and tropical and polar regions. The base CTP from the IR1
estimate reveals a similar shape of CTP distribution to that of MODIS. However, no
high frequencies at the high- and mid-levels are found, and low clouds (CTP = 900–1000hPa) are more frequent in the IR1-estimated CTP (figure 5(b)). When the CTP is
corrected by the radiance ratioing method, the frequency of high clouds (300–400 hPa)
increases by 4% and that of low clouds (900–1000 hPa) decreases by 7% (figure 5(c)).
Consequently, the values from the final CTP at both levels become fairly close to those of
MODIS (23.5% versus 20.7% at the low level, 12.2% versus 12.2% at the high level).
Interestingly, the regional characteristics of the CTP distribution are present in both the
final CTP and the MODIS CTP data; in the tropics, very high (200–300 hPa) and very
(a) AugustSimple
Decoupled
MODIS
(b) Liquid (c) Ice
(d) NH (e) SH (f) Mid-latitude
(g) Tropical
(a) August
Effective radius (µm)
2~5 5~1010~20
20~3030~40 60~
50~6040~50
Fre
qu
ency (
%)
0
10
20
30
40
50
60
Effective radius (µm)
2~5 5~1010~20
20~3030~40 60~
50~6040~50
Fre
qu
ency (
%)
0
10
20
30
40
50
60
Effective radius (µm)
2~5 5~1010~20
20~3030~40 60~
50~6040~50
Fre
qu
ency (
%)
0
10
20
30
40
50
60
Effective radius (µm)
2~5 5~1010~20
20~3030~40 60~
50~6040~50
Fre
qu
ency (
%)
0
10
20
30
40
50
60
Effective radius (µm)
2~5 5~1010~20
20~3030~40 60~
50~6040~50
Fre
qu
ency (
%)
0
10
20
30
40
50
60
Effective radius (µm)
2~5 5~1010~20
20~3030~40 60~
50~6040~50
Fre
qu
ency (
%)
0
10
20
30
40
50
60
Effective radius (µm)
2~5 5~1010~20
20~3030~40 60~
50~6040~50
Fre
qu
ency (
%)
0
10
20
30
40
50
60Simple
Decoupled
MODIS
(b) Liquid (c) Ice
(d) NH (e) SH ( f ) Mid-latitude
(g) Tropical
Figure 4. Same as figure 3 but for cloud effective radius (in mm).
Validation of MTSAT-1R cloud property 5947
low clouds (� 1000 hPa) are the most common, while middle clouds (600–800 hPa) are
the most common in the polar regions. Therefore, the radiance ratioing method using the
IR1 and IR3 channels is found to be useful in identifying high clouds above 400 hPa that
are falsely flagged as low clouds using the IR1 estimate alone. It should be noted that
some middle clouds identified by the MODIS CO2 slicing method cannot be discrimi-nated by our method, since the radiance ratioing method is valid only above 400 hPa.
In order to quantitatively estimate the uncertainty, we calculated the difference
between the derived MTSAT CTP and the MODIS CTP (figure 6). The derived CTP
includes the CTP from the IR1 estimate (i.e. base CTP), and its corrected CTP
IR1 only
12 km
60 km
100 km
220 km
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
MTSAT − MODIS distribution (%)
–20 –10 0 10 20
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
IR1 only
12 km
60 km
100 km
220 km
( f ) Polar (h ) Tropical(g ) Mid-latitude
(a) August
(e) SH
(b) Day (c) Night (d) NH
Figure 6. The difference between MTSAT-1R (both base and final) CTP and MODIS CTPvalues (in %) shown in figure 5. The radiance ratios are calculated by using clear-sky radiancesobtained within various spatial resolutions: 12 (3 · 3), 60 (15 · 15), 100 (25 · 25), and 220 km(55 · 55 pixels).
Distribution (%)
0 5 10 15 20 25 30 35
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
Distribution (%)
0 5 10 15 20 25 30 35
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
August
Day
Night
NH
SH
Polar
Tropical
Midlatitude
Distribution (%)
0 5 10 15 20 25 30 35
Clo
ud
to
p p
ressu
re (
hP
a)
0
200
400
600
800
1000
(a) MODIS CTP (b) Base CTP (c) final CTP
Figure 5. Relative frequency distribution (in %) of MODIS CTP (a), base CTP retrieved bythe IR1 estimate only (b), and final CTP corrected by the radiance ratioing method (c) for thetotal clouds. The results are shown for August 2006, daytime, night-time, northern and south-ern hemispheres, and the polar, tropical and mid-latitude regions.
5948 Y.-S. Choi and C.-H. Ho
(i.e. final CTP) using the radiance ratio in various spatial domains. For the polar
region, the radiance ratioing method cannot be applied (Choi et al. 2007). In the
MODIS CTP algorithm, the radiance ratio using CO2 channels is calculated by using
the clear-sky radiance obtained from 5 · 5 km, i.e. 5 · 5 MODIS pixels. As the
spatial resolution of a pixel in JAMI is 4 km in the IR bands, the minimal domain-to-domain clear-sky radiance for a given cloudy pixel should be 12 · 12 km (i.e. 3 · 3
pixels). Taking a larger domain would cause larger uncertainties in the CTP due to
false clear-sky radiances; however, more clear-sky values would be obtained. Thus,
the uncertainty of the final CTP can be more than that of the base CTP when the size
of the domain is large (e.g. 220 km). Figure 6 shows clearly that the final CTP using a
12-km domain has the frequencies nearest to those of the MODIS CTP. The absolute
value of the CTP from the 12-km domain minus the corresponding MODIS CTP
distribution is highest at the 100–200 hPa and 700–800 hPa levels (3% and 7%,respectively). Therefore, the accuracy of the obtained clear-sky radiance is one of
the major factors that determine the CTP uncertainties.
3.3 Time-series comparison
We have investigated the averaged values of cloud properties for various conditions
during our analysis period. Although the analysis of the averages is important for
validating the retrieved products, it cannot be guaranteed that the results representactual coherency with the MODIS products, which are considered as true reference
values in this study. Therefore, we examine the time series of cloud properties in the
nine selected regions. Since the MTSAT data are hourly at a 4 km · 4 km resolution,
the cloud properties were averaged for a 1� grid for each hour in order to compare it
with MOD08. MODIS/Terra passes at about 10:30 am local time over all the
regions. Therefore, the time series of the hourly MTSAT data is expected not to
be the same as that of MODIS, only retaining similar daily variations with hourly
fluctuation.As expected, the time series of the ratio of ice clouds to the total clouds in the base
CP (i) and the final CP (ii) includes large hourly fluctuations, indicating a rapid
change in the cloud properties in nature (figure 7). The figure also indicates that the
effect of adding the IR3 test to the bispectral CP test is dependent on region, on
comparing (i) and (ii). The time series of the ratios in most regions shows that
the derived and the MODIS data are nearly in phase. The ratios become higher
when the IR3 test is incorporated; therefore, the ratios in the final CP are closer to
the MODIS daily values. However, in the Antarctic region, the ratios in the final CPare relatively high in comparison with the MODIS values (11–20 and 25–31 August).
Considering that our analysis period corresponds to wintertime in this region, this
high ratio of ice clouds may be additional evidence showing the effect of the IR3 test to
identify ice clouds in the cold air of the polar region. As we will show the time series of
the CTP in the same regions later, the frequent occurrence of high clouds above 500
hPa in this region also strongly supports the high ratio of ice clouds.
The COT and ER are retrieved for the FOVs with solar and satellite zenith angles
less than 60�. Thus, the time series of the base (i) and final products (ii) are obtained inonly five of the nine regions at daytime (figures 8 and 9); better agreement with the
MODIS data is found for the final data in most regions than for the base data
(compare (i) and (ii) in both figures). This improvement in the COT and ER retrievals
via the decoupling method is seen regardless of the region. However, large hourly
Validation of MTSAT-1R cloud property 5949
variations in the COT and ER retrievals exist. This is perhaps related to the influence
of the change in the solar zenith angle on the retrieval; large bias must occur around
sunrise or sunset.
The time series of the derived CTP generally follow that of the MODIS values in all
the analysis regions (figure 10). The effect of applying the ratioing method may be to
change the overestimated base CTP values (i) into more realistic, lower final CTP
values (ii). In comparison with the MODIS values, the final CTP values may be
(i )
(ii )
(i )
(ii )
(i )
(ii )
1 4 7 10 13 16 19 22 25 28 31
Ice /
clo
ud
s (
%)
020
40
60
80
100
Day of month
Day of month
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
1 4 7 10 13 16 19 22 25 28 31Ic
e /
clo
ud
s (
%)
020
40
60
80
10
0
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Ice /
clo
uds (
%)
020
40
60
80
100
Day of monthDay of month
Day of monthDay of month
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
10
0
MTSAT MODIS
(a) Seoul (37°–38°N, 126°–127°E) (b) Northeast China Plain (32°–33°N, 115°–116°E) (c) Gobi desert (40°–41°N, 104°–105°E)
(i )
(ii )
MTSAT MODIS (i)
(ii)
(i )
(ii )
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
Day of month1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
1 4 7 10 13 16 19 22 25 28 31
Ice /
clo
ud
s (
%)
020
40
60
80
100
Day of month1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
MTSATMODIS
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
Day of month
Day of month Day of month Day of month
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
MTSATMODIS
(d) Tibetan Plateau (35°–36°N, 90°–91°E) (e) The South China Sea (10°–11°N, 115°–116°E) ( f ) The Philippine Sea (10°–11°N, 135°–136°E)
o o o o
(i )
(ii )
o o o o
(i )
(ii )
o o o o(g) The East Pacific (0 −1 N, 150 −151 W) (h) The Bering Sea (59 −60 N, 179 −180 W) (i) The Antarctic (70 −71 S, 159 −160 E)
(i )
(ii )
1 4 7 10 13 16 19 22 25 28 31
Ice /
clo
ud
s (
%)
020
40
60
80
100
Day of month
1 4 7 10 13 16 19 22 25 28 31
Ice /
clo
uds (
%)
020
40
60
80
100
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
louds (
%)
020
40
60
80
100
Day of month
1 4 7 10 13 16 19 22 25 28 31
Ice
/ c
lou
ds (
%)
020
40
60
80
100
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Ice /
clo
ud
s (
%)
020
40
60
80
100
MTSAT MODIS
Day of month
Day of month Day of month Day of month
1 4 7 10 13 16 19 22 25 28 31
Ice /
clo
uds (
%)
020
40
60
80
100
Figure 7. Time series of the ratio of ice clouds to the total clouds at nine selected sites; base CPfrom IR1 and IR2 (i), and final CP from IR1, IR2, and IR3 (ii).
5950 Y.-S. Choi and C.-H. Ho
1 4 7 10 13 16 19 22 25 28 31
Optical th
ickn
ess
020
40
60
80
Day of month1 4 7 10 13 16 19 22 25 28 31
Optical th
ickne
ss
020
40
60
80
(i)
(ii)
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Op
tical th
ickness
020
40
60
80
Day of month1 4 7 10 13 16 19 22 25 28 31
Optica
l th
ickn
ess
020
40
60
80
(i)
(ii)
MTSAT MODIS
Day of month
Day of monthDay of month
(i )
(ii )
MTSAT MODIS (i )
(ii )
MTSAT MODIS
(a) Seoul (37°–38°N, 126°–127°E) (b) Northeast China Plain (32°–33°N, 115°–116°E)
1 4 7 10 13 16 19 22 25 28 31
Optical th
ickne
ss
020
40
60
80
Day of month1 4 7 10 13 16 19 22 25 28 31
Optical th
ickne
ss
020
40
60
80
(i)
(ii)
MTSAT MODIS
Day of month
(i )
(ii )
MTSAT MODIS
(c) Gobi desert (40°–41°N, 104°–105°E)°
1 4 7 10 13 16 19 22 25 28 31
Op
tica
l th
ickn
ess
02
040
60
80
Day of month
1 4 7 10 13 16 19 22 25 28 31
Op
tica
l th
ickn
ess
020
40
60
80
(i)
(ii)
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Optical th
ickness
020
40
60
80
Optical th
ickness
020
40
60
80
Day of month
Day of month
Day of month
1 4 7 10 13 16 19 22 25 28 31
(i)
(ii)
MTSAT MODIS
(d) The South China Sea (10°–11°N, 115°–116°E)
(e) The Philippine Sea (10°–11°N, 135°–136°E)
(i )
(ii )
MTSAT MODIS
(i )
(ii )
MTSAT MODIS
–11 –116
Figure 8. Same as figure 7 but for base COT using the VIS and IR4 radiances (i), and finalCOT corrected using the decoupling method in order to have a reflected component fromclouds only in the radiances (ii).
Validation of MTSAT-1R cloud property 5951
1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
adiu
s (
µ m)
01
02
03
040
50
60
Day of month1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
adiu
s (
µ m)
01
02
03
040
50
60
(i
(ii )
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
adiu
s (µ
m)
01
02
03
04
05
060
Day of month1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
adiu
s (
µ m)
010
20
30
40
50
60
(i ))
(ii )
MTSAT MODIS
Day of month Day of month
°–38°N, 126°–127°E) (b) Northeast China Plain (32°–33°N, 115°–116°E)
1 4 7 10 13 16 19 22 25 28 31
01
02
03
04
05
060
1 4 7 10 13 16 19 22 25 28 31
(a) Seoul (37
1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
adiu
s (
µ m)
010
20
30
40
50
60
Day of month
Day of month
1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
adiu
s (
µ m)
010
20
30
40
50
60
(i )
(ii )
MTSAT MODIS
(c) Gobi desert (40°–41°N, 104°–105°E)
1 4 7 10 13 16 19 22 25 28 31
c
1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
adiu
s (
µ m)
010
20
30
40
50
60
Day of month
1 4 7 10 13 16 19 22 25 28 31
Eff
ective r
ad
ius (
µ m)
010
20
30
40
50
60
(i )
(ii )
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Eff
ective
ra
diu
s (
µ m)
010
20
30
40
50
60
Day of month
Day of month
Day of month
1 4 7 10 13 16 19 22 25 28 31
Eff
ective
ra
diu
s (
µ m)
010
20
30
40
50
60
(i )
(ii )
MTSAT MODIS
(d ) The South China Sea (10°–11°N, 115°–116°E)
(e ) The Philippine Sea (10°–11°N, 135°–136°E)
Figure 9. Same as figure 8 but for base ER (i) and final ER (ii).
5952 Y.-S. Choi and C.-H. Ho
underestimated to some extent in Seoul, the north-east China plain, the East Pacificand the Bering Sea. There are several reasons for this underestimation: (i) middle clouds
are not commonly identified; (ii) the radiance ratio library for the mid-latitudes is
applied to the 30–60� latitudes, while the atmospheric conditions over the Bering Sea
(located between 59� and 60� N) may be far from the model profile; and (iii) in the mid-
latitudes, the radiance ratio is not distinctly sensitive to the CTP (refer to figure 7 of
Choi et al. 2007). Despite these several sources of uncertainties, the final CTP agrees
fairly well with the MODIS values in the other regions, including the Gobi desert,
Tibetan plateau, South China Sea, Philippine Sea and Antarctic region.
1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
10
04
00
700
1000
Day of month1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
100
400
700
1000
(i )
(ii )
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
10
04
00
700
1000
Day of month1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
100
400
700
1000
(i )
(ii )
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Heig
ht
(hP
a)
100
400
700
1000
Day of month
Day of month Day of month Day of month
1 4 7 10 13 16 19 22 25 28 31
Heig
ht
(hP
a)
10
04
00
700
1000
(i )
(ii )
MTSAT MODIS
(a) Seoul (37°–38°N, 126°–127°E) (b) Northeast China Plain (32°–33°N, 115°–116°E) (c) Gobi desert (40°–41°N, 104°–105°E)
700
1000
1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
100
400
700
1000
Day of month1 4 7 10 13 16 19 22 25 28 31
He
igh
t (h
Pa
)100
400
700
1000
(i )
(ii )
MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Heig
ht
(hP
a)
100
400
700
1000
Day of month1 4 7 10 13 16 19 22 25 28 31
Heig
ht
(hP
a)
100
400
700
1000
(i )
(ii )
MTSATMODIS
1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
100
400
700
1000
Day of month
Day of month
1 4 7 10 13 16 19 22 25 28 31
Heig
ht (h
Pa
)100
400
700
1000
(i )
(ii )
MTSATMODIS
(d ) Tibetan Plateau (35°–36°N, 90°–91°E) (e) The South China Sea (10°–11°N, 115°–116°E) ( f ) The Philippine Sea (10°–11°N, 135°–136°E)
(g) The East Pacific (0o−1oN, 150o−151oW)
1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
100
400
700
100
0
Day of month1 4 7 10 13 16 19 22 25 28 31
Heig
ht
(hP
a)
100
40
070
01000
(i )
(ii )
MTSAT MODIS
(h) The Bering Sea (59o−60oN, 179o−180oW)
1 4 7 10 13 16 19 22 25 28 31
He
ight
(hP
a)
100
400
700
100
0
Day of month
Day of monthDay of month Day of month
Day of monthDay of month Day of month
1 4 7 10 13 16 19 22 25 28 31
Heig
ht
(hP
a)
100
40
070
01000
(i )
(ii )
MTSAT MODIS
(i) The Antarctic (70o−71oS, 159o−160oE)
1 4 7 10 13 16 19 22 25 28 31
He
igh
t (h
Pa)
100
400
70
01
000
(i ) MTSAT MODIS
1 4 7 10 13 16 19 22 25 28 31
Heig
ht
(hP
a)
100
400
700
1000
(ii )
1 −
He
igh
t (h
Pa)
Figure 10. Same as figure 7 but for base CTP (i) and final CTP (ii).
Validation of MTSAT-1R cloud property 5953
(a) CTP difference (b) Error in retrieved CTP
(c) COT difference (d) Error in retrieved COT
(e) ( f)ecnereffidRE Error in retrieved ER
FinalBase
r
FinalBase
µ
Figure 11. Relative frequency of MTSAT minus MODIS CTP (a), COT (c), and ER (e) for themaximum values. Errors in the retrieved CTP (b), COT (d), and ER (f) (in %) with respect to thecorresponding parameters. The solid and dotted lines indicate values from the final (corrected)and base (uncorrected) products, respectively.
5954 Y.-S. Choi and C.-H. Ho
Over the South China and Philippine seas, strong convection systems appear to
occur between 7 and 13 August, as high ratio of ice clouds, high COT, and low CTP
are consistently revealed (figures 7, 8, and 10). It should be noted that the convective
clouds over the South China Sea are generally more developed vertically than those
over the Philippine Sea (Park et al. 2007). This may be indicated by consistentresults: higher ratio of ice clouds, higher COT, and lower CTP over the South
China Sea.
3.4 Pixel-by-pixel comparison
The results of the pixel-by-pixel comparison of cloud properties are presented in this
subsection. Figure 11 shows the relative frequency of the differences between the
derived MTSAT and the MODIS retrievals to the maximum values, and the errors inthe retrievals with respect to the corresponding MODIS values. The error is expressed
in terms of the ratio (in percentage) of the difference between the MTSAT and
MODIS retrievals to the MODIS retrievals.
Most (90%) of the final CTP and MODIS CTP values are in agreement with each
other (figure 11). A large difference in CTP over 300 hPa also occurs for about 0.1%
of the total pixels. This uncertainty may result from a number of factors such as the
discrepancies in the retrieval resolutions, radiance ratios, constructed clear-sky
radiances and the numerical atmospheric profiles employed. A comparison of thebase and final CTP values clearly reveals that our correction method (i.e. the
radiance ratioing method for high-cloud top pressure) is effective particularly
when the MTSAT CTP is higher than the MODIS CTP (figure 11(a)). Large errors
for low clouds (CTP � 700 hPa) are dramatically decreased by this method (figure
11(b)). The error in the final CTP is found to be less than roughly 10%, and it is
lowest for high clouds between 200 and 300 hPa. These results confirm the effect of
the radiance ratioing method using the IR1 and IR3 bands in the present CTP
algorithm.The final COT and the MODIS COT values are mostly identical to within a
difference of � 5. Only a few pixels (,2% of the total pixels) are in disagreement
with the MODIS COT (figure 11(c)). The disagreement and the error in the final COT
values are much lower than those for the base COT values, especially for clouds with
COT , 60 (figures 11(c) and 11(d)). The errors in both the base and the final COT
values are very small for very (optically) thick clouds with COT � 60. This is some-
what contradictory when considering that the optical thickness of thick clouds may
have much more uncertainty due to physical reasons such as the high sensitivity of theCOT value to VIS radiance (Choi et al. 2007). Our further analysis shows that the
occurrence of very thick clouds is not sufficient in nature. In addition, the collocation
problems are immanent in any case. These reasons would lead to false errors, and
often to low errors by chance.
Unlike the COT, the final ER values still have large uncertainty when compared to
the MODIS values. The disagreement between the final and MODIS ER values
occurs for a considerable portion of the pixels, and it is likely to have originated
from the relatively large ER values (. 40 mm) (figures 11(e) and 11(d)). This uncer-tainty probably results from the insensitivity of IR4 radiance to large-sized cloud
particles (Choi et al. 2007). Therefore, reliable ER retrieval from MTSAT-1R images
can be guaranteed only up to a value of approximately 40 mm. We note that the ER
from MODIS provides estimates for liquid (ice) clouds only up to 30 (90 mm).
Validation of MTSAT-1R cloud property 5955
4. Summary and discussion
We have presented validation results of the CLA (Choi et al. 2007) by comparing the
base, final and MODIS cloud properties. The base and final cloud properties are
defined as the products from the hourly MTSAT five-channel imagery by the previous
and present algorithm, respectively (table 2). For the validation, we carried out scene
analysis, large-scale climatology comparison, time-series comparison, and pixel-by-
pixel comparison. Based on our analysis, the final CP, COT, ER and CTP values are
consistently closer to the MODIS reference data than the base products. This
indicates that the present developed schemes for the CLA are, in effect, of meritdespite the limitation of the channel number. The major results are summarized as
follows.
The final CP has considerably more water and ice clouds (or fewer mixed and
unknown clouds) than the base CP, which indicates the role of the additional IR3 test.
Compared with the MODIS CP, however, the proportion of water (mixed) clouds
across the globe is still very low (high); it is lower (higher) by about 11% (15%). This
uncertainty between the water and mixed phases remains as the limitation of the
present IR CP algorithm due to the absence of the 8.5-mm channel in the MODISalgorithm. It is noted that the proportion of ice clouds is comparable with that in the
MODIS CP data, but is very high (by about 15%) only in the polar region.
The final COT and ER values have less bias against the MODIS values than the
base products, particularly in the mid-latitudes, which indicates the effect of the
radiance decoupling method to extract cloud-reflected radiances in the VIS and IR4
channels. When the decoupling method is incorporated, the bias is dramatically
reduced for 5 � COT , 60 and for ER , 40 mm. The large uncertainty in the COT
and ER retrievals due to high solar zenith angles and very large COT (.. 60) and ERvalues (.. 40 mm) remains to be resolved.
Reliable CTP values can be retrieved by correcting the high-cloud top pressure
(, 400 hPa) via the radiance ratio obtained from the clear-sky radiance, which is
defined as the maximal radiance in the domain of 12 km · 12 km. The accuracy of
CTP retrievals largely depends on the accuracy of the obtained clear-sky radiance.
Bias against the MODIS CTP still exists in the mid-troposphere between the 700 and
800 hPa levels.
The cloud properties retrieved from the validated CLA in this study can be used inmany fields of study such as cloud microphysics, severe weather forecast, and climate
research. The CLA is intended for use in the COMS meteorological data processing
system in Korea. The present study details the physical uncertainties and limitations
underlying the algorithm, based on intercomparison with the MODIS cloud data. In
fact, by using the imagery from onboard five-channel passive radiometers on present
and past geostationary satellites, we believe that this CLA can also be operated via an
off-line system to construct a global map of cloud properties with a high temporal and
spatial resolution. The accumulated and reconstructed cloud properties would beuseful in investigating numerous natural phenomena of interest, including general
circulation, cloud-radiative feedback, global/regional climate changes, etc.
Acknowledgements
This study was funded by the Korea Meteorological Administration Research and
Development Program under Grant CATER 2006–4204. The MODIS data were
obtained from LAADS at the Goddard Space Flight Center.
5956 Y.-S. Choi and C.-H. Ho
References
BAUM, B.A., SOULEN, P.F., STRABALA, K.I., KING, M.D., ACKERMAN, S.A., MENZEL, W.P. and
YANG, P., 2000, Remote sensing of cloud properties using MODIS airborne simulator
imagery during SUCCESS, 2, Cloud thermodynamic phase. Journal of Geophysical
Research, 105, pp. 11781–11792.
BAUM, B.A., YANG, P., HEYMSFIELD, A.J., PLATNICK, S., KING, M.D., HU, Y.X. and BEDKA, S.T.,
2005, Bulk scattering properties for the remote sensing of ice clouds. Part II:
Narrowband models. Journal of Applied Meteorology, 44, pp. 1896–1911.
CHOI, Y.-S. and HO., C.-H., 2006, Radiative effect of cirrus with different optical properties
over the tropics in MODIS and CERES observations. Geophysical Research Letters, 33,
L21811, doi:10.1029/2006GL027403.
CHOI, Y.-S., HO, C.-H. and SUI, C.-H., 2005, Different optical properties of high cloud in GMS
and MODIS observations. Geophysical Research Letters, 32, L23823, doi:10.1029/
2005GL024616.
CHOI, Y.-S., HO, C.-H., AHN, M.-H. and KIM, Y.-M., 2007, An exploratory study of cloud
remote sensing capabilities of the Communication, Ocean and Meteorological Satellite
(COMS) Imagery. International Journal of Remote Sensing, 28, pp. 4715–4732.
CLOTHIAUX, E.E., ACKERMAN, T.P., MACE, G.G., MORAN, K.P., MARCHAND, R.T., MILLER,
M.A. and MARTNER, B.E., 2000, Objective determination of cloud heights and radar
reflectivities using a combination of active remote sensors at the ARM CART Sites.
Journal of Applied Meteorology, 33, pp. 645–665.
ELLINGSON, R.G., ELLIS, J. and FELS, S., 1991, The intercomparison of radiation codes used in
climate models: long wave results. Journal of Geophysical Research, 96, pp. 8929–8953.
KEY, J.R., 2002, Streamer Version 3.0 User’s Guide (Madison, WI: NOAA/NESDIS). Available
online at: stratus.ssec.wisc.edu (accessed 23 July 2009).
KEY, J.R. and SCHWEIGER, A., 1998, Tools for atmospheric radiative transfer: Streamer and
FluxNet. Computers and Geosciences, 24, pp. 443–451.
KIM, J.-H., HO, C.-H., LEE, M.-H., JEONG, J.-H. and CHEN, D., 2006, Large increase in heavy
rainfall associated with tropical cyclone landfalls in Korea after the late 1970s.
Geophysical Research Letters, 33, L18706, doi:10.1029/2006GL027430.
KING, M.D., TSAY, S.C., PLANTNICK, S.E., WANG, M. and LIOU, K.N., 1997, Cloud retrieval
algorithms for MODIS: optical thickness, effective particle radius, and thermodynamic
phase. MODIS Algorithm Theoretical Basis Document, NASA.
KING, M.D., PLATNICK, HUBANKS, P.A., ARNOLD, G.T., MOODY, E.G., WIND, G. and WIND, B.,
2006, Collection 005 Change Summary for the MODIS Cloud Optical Property
(06_OD) Algorithm. Available online at: http://modis-atmos.gsfc.nasa.gov/
C005_Changes/C005_CloudOpticalProperties_ver311.pdf (accessed 23 July 2009).
LIOU, K.N., 2002, An Introduction to Atmospheric Radiation, 2nd ed. (San Diego: Academic Press).
MACE, G.G., ZHANG, Y., PLATNICK, S., KING, M.D., MINNIS, P. and YANG, P., 2005, Evaluation
of cirrus cloud properties derived from MODIS data using cloud properties derived
from ground-based observations collected at the ARM SGP site. Journal of Applied
Meteorology, 44, pp. 221–240.
MENZEL, W.P., SMITH, W.L. and STEWART, T.R., 1983, Improved cloud motion wind vector and
altitude assignment using VAS. Journal of Climate and Applied Meteorology, 22,
pp. 377–384.
MENZEL, W.P., FREY, R.A., BAUM, B.A. and ZHANG, H., 2006, Cloud top properties and cloud
phase algorithm theoretical basis document. MODIS Algorithm Theoretical Basis
Document, NASA.
MOODY, E.G., KING, M.D., PLATNICK, S., SCHAAF, C.B. and GAO, F., 2005, Spatially complete
global spectral surface albedos: Value-added datasets derived from Terra MODIS land
products. IEEE Transactions on Geoscience and Remote Sensing, 43, pp. 144–158.
MOODY, E.G., KING, M.D., SCHAAF, C.B., HALL, D.K. and PLATNICK, S., 2007, Northern
Hemisphere five-year average (2000–2004) spectral albedos of surfaces in the presence
Validation of MTSAT-1R cloud property 5957
of snow: Statistics computed from Terra MODIS land products. Remote Sensing of
Environment, 111, pp. 337–345.
NAKAJIMA, T.Y. and NAKAJIMA, T., 1995, Wide-area determination of cloud microphysical
properties from NOAA AVHRR measurements for FIRE and ASTEX regions.
Journal of the Atmospheric Sciences, 52, pp. 4043–4059.
PARK, M.-S., CHOI, Y.-S., HO, C.-H., SUI, C.-H., PARK, S.-K. and AHN, M.-H., 2007,
Regional cloud characteristics over the tropical northwestern Pacific as revealed
by TRMM PR and TMI. Journal of Geophysical Research, 112, D05209,
doi: 10.1029/2006JD007437.
PLATNICK, S., KING, M.D., ACKERMANN, S.A., MENZEL, W.P., BAUM, B.A., RIEDI, J.C. and
FREY, R.A., 2003, The MODIS cloud products: Algorithms and examples from Terra.
IEEE Transactions on Geoscience and Remote Sensing, 41, pp. 456–473.
PUSCHELL, J.J., OSGOOD, R., AUCHTER, J., HURT, W.T., HITOMI, M., SASAKI, M., TAHARA, Y.,
TADROS, A., FALLER, K., MCLAREN, M., SHEFFIELD, J., GAISER, J., KAMEL, A. and
GUNSHOR, M., 2006, In-flight performance of the Japanese Advanced Meteorological
Imager, SPIE Proceedings, 6296, 62960N.
RICCHIAZZI, P., YANG, S., GAUTIER, C. and SOWLE, D., 1998, SBDART: A research and teaching
software tool for plane-parallel radiative transfer in the earth’s atmosphere. Bulletin of
the American Meteorological Society, 79, pp. 2101–2114.
ROSSOW, W.B. and GARDER, L.C., 1993a, Cloud detection using satellite measurements of
infrared and visible radiances for ISCCP. Journal of Climate, 6, pp. 2341–2369.
ROSSOW, W.B. and GARDER, L.C., 1993b, Validation of ISCCP cloud detections. Journal of
Climate, 6, pp. 2370–2393.
ROSSOW, W.B. and SCHIFFER, R.A., 1999, Advances in understanding clouds from ISCCP.
Bulletin of the American Meteorological Society, 80, pp. 2261–2287.
ROSSOW, W.B., WALKER, A.W. and GARDER, L.C., 1993, Comparison of ISCCP and other cloud
amounts. Journal of Climate, 6, pp. 2394–2418.
SHUPE, M.D., MATROSOV, S.Y. and UTTAL, T., 2006, Arctic mixed-phase cloud properties
derived from surface-based sensors at SHEBA. Journal of the Atmospheric Sciences,
63, pp. 697–711.
STAMNES, K., TSAY, S.C., WISCOMBE, W. and JAYAWEERA, K., 1988, Numerically stable algo-
rithm for discrete-ordinate-method radiative transfer in multiple scattering and emit-
ting layered media. Applied Optics, 27, pp. 2502–2509.
STRABALA, K.I., ACKERMAN, S.A. and MENZEL, W.P., 1994, Cloud properties inferred from
8–12-mm data. Journal of Applied Meteorology, 33, pp. 212–229.
VERLINDE, J., HARRINGTON, J.Y., MCFARQUHAR, G.M., YANNUZZI, V.T., AVRAMOV, A.,
GREENBERG, S., JOHNSON, N., ZHANG, G., POELLOT, M.R., MATHER, J.H., TURNER, D.D.,
ELORANTA, E.W., ZAK, B.D., PRENNI, A.J., DANIEL, J.S., KOK, G.L., TOBIN, D.C., HOLZ, R.,
SASSEN, K., SPANGENBERG, D., MINNIS, P., TOOMAN, T.P., IVEY, M.D., RICHARDSON, S.J.,
BAHRMANN, C.P., SHUPE, M., DEMOTT, P.J., HEYMSFIELD, A.J., and SCHOFIELD, R., 2007,
The mixed-phase arctic cloud experiment. Bulletin of the American Meteorological
Society, 88, pp. 205–221.
YANG, P., ZHANG, L., HONG, G., NASIRI, S.L., BAUM, B.A., HUANG, H.L., KING, M.D. and
PLATNICK, S., 2007, Differences between collection 4 and 5 MODIS ice cloud optical/
microphysical products and their impact on radiative forcing simulations. IEEE
Transanctions on Geoscience and Remote Sensing, 45, pp.2886–2899.
5958 Y.-S. Choi and C.-H. Ho
Copyright of International Journal of Remote Sensing is the property of Taylor & Francis Ltd and its content
may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express
written permission. However, users may print, download, or email articles for individual use.
top related