The effects of anomalous atmospheres on the accuracy of infrared

sea-surface temperature retrievals: Dry air layer intrusions over the

tropical ocean
M. Szczodrak , P.J. Minnett, R.H. Evans
Meteorology and Physical Oceanography, Rosenstiel School of Marine and Atmospheric Science, University of Miami, 4600 Rickenbacker Causeway, Miami, FL 33149, United States
a b s t r a c t a r t i c l e i n f o
Article history:
Received 12 October 2011
Received in revised form 10 September 2013
Accepted 13 September 2013
Available online 9 October 2013
Keywords:
Sea-surface temperature
Infrared radiometry
Atmospheric correction
Dry-layers
Anomalous atmospheric effects
The effects of layers inthe atmosphere withanomalously lowmoisture content on the accuracy of the sea-surface
temperature (SST) derived frommeasurements of infrared radiometers on earth observation satellites are quan-
tiedusing measurements taken fromresearchcruises and numerical simulations. Radiosonde data fromareas of
the oceans that are seasonally affected by intrusions of dry air masses of continental origin were used with the
Line-By-Line Radiative Transfer Model (LBLRTM) to simulate brightness temperatures measured in MODIS
bands 31 and 32 (at wavelengths of ~11 m and ~12 m). The radiosonde datasets contained proles with
and without a dry layer, representing the baseline (no dry layer aloft) and the anomalous conditions (dry layer
present). MODIS SST retrieval algorithmversions 5 and 6 were applied to the simulated brightness temperatures
to obtain the SST and the retrieval errors were examined. Whereas the average errors for the baseline no dry
layer conditions range between 0.29 and 0.72 K, in the case of very deep dry layers the errors can be N1 K. Sim-
ulations were also performed for atmospheric proles that were created from the measured proles by drying
layers at various altitudes. It was found that the retrieval errors 1) depend on the amount of water vapor in the
atmosphere, 2) change in a systematic way dependent on the presence and characteristics of a dry layer, 3) dry
layers in the lower troposphere make the SST retrieval errors more positive, and dry layers in the upper tropo-
sphere tend to introduce SST retrieval errors that are more negative.
2013 Elsevier Inc. All rights reserved.
1. Introduction
There are several sources of contributions to the uncertainties in sea-
surface temperature (SST) retrievals from infrared radiometers on satel-
lites, andone of the most important is the accuracy withwhichthe effects
of the intervening atmosphere are corrected. Other sources include the
uncertainties in the radiometric measurements of the spacecraft instru-
ment, and the effectiveness of the identicationof measurements that in-
clude radiances from clouds and aerosols. The presence of clouds in the
eld of view of an imaging infrared radiometer is a major hindrance to
the retrieval of SST and corrections cannot generally be made, which
leads to the exclusion of the cloud-contaminated measurements from
the SST retrieval process and can result in the rejection of ~8090% of
the infraredmeasurements (Kilpatrick, Podest, &Evans, 2001). For aero-
sols, there is some evidence that corrections can be made (Merchant,
Embury, Le Borgne, & Bellec, 2006; Nalli & Stowe, 2002), but as with
clouds, it is usually better to attempt to identify their presence and
discard the data fromthe SST retrieval process. The instrumental artifacts
comprise those that are essentially irreducible, such as the noise generat-
ed by the detectors and imperfections in the digitizers, and those for
which corrections can be derived given improved information about
the characteristics of the instrument, often gained from housekeeping
data and patterns in the errors in the derived variables (e.g. Xiong,
Barnes, Guenther, & Murphy, 2003). Thus, the contribution to the uncer-
tainties that is perhaps the most susceptible to improvement is that from
the atmospheric correction.
In the cases where the atmosphere is free of cloud and aerosols, the
radiative effects are due to the absorption and re-emission of infrared
photons by molecules in the atmosphere (see 3.4 of Kidder and Haar
(1995), or 4.5 of Martin (2004)). Since the atmosphere is nearly every-
where cooler than the sea surface, the consequence of the absorption
and re-emission is to make the temperatures measured by the satellite
radiometers, the brightness temperatures, cooler than the SST, and the
purpose of the atmospheric correction algorithm is to compensate for
this temperature decit (e.g. Zvody et al., 1995). Surface temperature
retrievals from satellite radiometers are made in the wavelength inter-
vals in the atmosphere's transmission spectrumwhere the transmissiv-
ity is high, the so-called atmospheric windows, so that most of the
signal originates at the surface and the temperature decit is relatively
small. In these spectral regions, such as in the wavelength range of
~1012.5 m, the main atmospheric constituent that interacts with
the infrared radiation is water vapor; and water vapor is very variable
in both space and time. Exploiting the wavelength dependence of the
Remote Sensing of Environment 140 (2014) 450465
Corresponding author. Tel.: +1 305 421 4996; fax: +1 305 421 4696.
E-mail address: gszczodrak@rsmas.miami.edu (M. Szczodrak).
0034-4257/$ see front matter 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.rse.2013.09.010
Contents lists available at ScienceDirect
Remote Sensing of Environment
j our nal homepage: www. el sevi er . com/ l ocat e/ r se
water vapor effect, an atmospheric correction algorithmcan be devised
based on the different brightness temperatures at different wavelengths
in the atmospheric window (McMillin, 1975; Minnett & Barton, 2010,
and references therein). Such atmospheric correction algorithms have
sets of coefcients that are generally derived empirically, through the
statistical analysis of matchups between the satellite brightness tem-
perature measurements and collocated, contemporaneous measure-
ments from surface thermometers or radiometers (e.g. Kilpatrick et al.,
2001). An alternative method to derive the coefcients uses radiative
transfer modeling to simulate the propagation of the spectrum of infra-
red emission fromthe sea-surface through a large number of atmospher-
ic temperature and humidity proles taken from radiosondes launch
from ships or coastal stations, or from the representation of the marine
atmosphere in the reanalysis elds of Numerical Weather Prediction
Models. The relative spectral response functions of the satellite radiome-
ter are used with the simulated spectra of infrared top-of-atmosphere
emission to derive a large set of simulated brightness temperatures
(Llewellyn-Jones, Minnett, Saunders, & Zvody, 1984; Zvody et al.,
1995). The simulated top-of-atmosphere brightness temperatures and
their associated SSTs are then subjected to the same statistical analysis
as the matchups to derive coefcients for the atmospheric correction
algorithm. In the empirical method there is an implicit assumption that
the distribution of atmospheric properties in the matchups is a good
representation of the range of cloud- and aerosol-free atmospheric con-
ditions to which the algorithmwill be applied. Similarly, in the modeling
approach, there is the assumption that the selection of atmospheric tem-
perature andhumidity proles usedto generate the simulatedbrightness
temperatures provides an appropriate representation of the atmospheric
conditions for the retrievals.
In applying the atmospheric correction algorithms to generate global
SST elds, it is to be expected that the uncertainties in the individual
retrievals will have a distribution, probably close to Gaussian, that re-
ects the deviations of the atmospheric conditions from the mean of
the distribution of those used to generate the coefcients used in the al-
gorithm. The bigger the deviation, the bigger the SST retrieval error
(Minnett, 1986). One feature that constitutes anomalous conditions,
and therefore can lead to larger SST errors, is the presence of an atmo-
spheric dry layer. Even if such layers exist in the selection of atmo-
spheres used in the coefcient derivation, they tend to be regionally or
seasonally constrained, and therefore will lead to uncertainties in the
SSTs that have a regional and seasonal character.
Barton (2011) discusses a dependence of the SST retrieval error for a
number of satellite SST sensors on the vertical distribution of water
vapor. Barton expresses this by the partition of water vapor between
lower and upper troposphere. In effect, dry layers are special (extreme)
cases of the vertical distribution of water vapor in the atmosphere so
one might suspect that under such conditions the SST error might be
large.
A well-known and sometimes extreme dry layer is the Saharan Air
Layer (SAL) that extends seasonally from the West African coast across
the tropical North Atlantic Ocean (Prospero, 1999; Prospero & Carlson,
1972), often, but not necessarily always, associatedwithmineral aerosols
(Zhang &Pennington, 2004). One consequence of the SAL is the suppres-
sion of the growth of Atlantic hurricanes (Dunion & Velden, 2004), and
the identication of the SAL for use in hurricane forecasting is facilitated
by the anomalous reduction in the satellite measurement of brightness
temperature differences between measurements at 10.7 and 12 m
(Dunion & Velden, 2004), the same pair of spectral measurements used
in the atmospheric correction for SST retrievals. Other occurrences of a
dry layer of continental origin extending over the tropical ocean are off
Northern Australia, and an example of this was seen in the Tropical
Warm Pool International Cloud Experiment (TWP-ICE) held in the
austral summer at the beginning of 2006 (May et al., 2008). An exten-
sive cyclonic feature centered over the Northern Territory caused the
winds to advect dry air at middle levels in the troposphere over the
Timor Sea.
Elsewhere, throughout the tropics and subtropics, dry layers result
as airmasses that have lost moisture through condensation events
descendover intervals of days to lower levels, achieving lowrelative hu-
midities, b20%, by adiabatic warming. These tend to formzonal bands in
the tropics of both hemispheres (Cau, Methven, & Hoskins, 2007).
This study is directed to improving our understanding of the effects
of dry layers on the uncertainties of SSTs derived frominfrared radiome-
ters on polar orbiting spacecraft, and is focused on events captured in
ship-based measurements in the tropical Atlantic Ocean (SAL events)
and in the Timor Sea (during the TWP-ICE campaign).
The algorithm used in retrieving SST values from top-of-atmosphere
radiance measurements of infrared radiometers, such as the Advanced
Very High Resolution Radiometer (Cracknell, 1997) and the MODerate-
resolution Imaging Spectroradiometer (Esaias et al., 1998) on earth-
observation satellites is based on statistical relationships between the
brightness temperatures and the corresponding SST. A commonly used
version is the Non-Linear SST algorithm (Walton, 1988) and was origi-
nally developed for AVHRR (May, Parmeter, Olszewski, & McKenzie,
1998). The MODIS version of the formula has the following form:
SST a
1
a
2
T
31
a
3
T
31
T
32
T
ref
a
4
T
31
T
32
sec 1 1
where T
31
and T
32
are brightness temperatures (BT) in MODIS channels
31 and 32 (wavelength bandwidths of 10.78 to 11.28 m and 11.77 to
12.27 m), T
ref
is a reference or rst guess temperature, which is
often taken as the Reynolds Optimally Interpolated SST (Reynolds
et al., 2007), is the satellite zenith angle, a
1
to a
4
are coefcients derived
from coincident measurements of subsurface temperatures from buoys,
and is a correction for the thermal skin effect so the retrieval is a skin
SST. The coefcients are dependent on the relative spectral responses
of the individual radiometers, through the conversion from measured
spectral radiance to the equivalent top-of-atmosphere (TOA) BT, and
on atmospheric variability, which results from the dependence of atmo-
spheric absorption and emission on variable constituents of the atmo-
sphere, especially water vapor. The second term on the right-hand side
of Eq. (1) represents the rst order effect relating the measured bright-
ness temperature to that of the emitting sea surface. The third termrep-
resents the correction for the absorption and emission of the intervening
atmosphere, primarily due to water vapor, and the fourth term accounts
for the increased length of a slant path the radiation propagates through
the atmosphere for off-nadir measurements. The skin effect correction
accounts for the difference between the subsurface SST, as measured by
the buoys, and the skin SST. The MODIS algorithms include the global
average of 0.17 K (Donlon et al., 2002; Minnett, Smith, & Ward,
2011) so that the MODIS retrievals are estimates of the skin SST.
In the current version of the MODIS SST retrieval algorithm(Version
5, referred to here as V5) the a coefcients are derived separately for
each month of the year from matchups between the MODIS BTs and
subsurface temperature measurements from drifting and moored
buoys. To avoid discontinuities at the start of each month a sliding tem-
poral tapered weighting function over ve months is used. Following
experience gained from the AVHRR Pathnder program (Kilpatrick
et al., 2001), the coefcients are derived separately for two populations
of the BT difference between MODIS channels 31 and 32 (BT
3132
),
which can be used a proxy for the precipitable water vapor (PWV).
One set of coefcients is derived for BT
3132
b 0.7 K and other for
BT
3132
N 0.7 K, corresponding to dry and moist atmospheres this
will be discussed further below. When the algorithm is applied to
MODIS measurements, the dry coefcients are applied where
BT
3132
b 0.5 K and moist for BT
3132
N 0.9 K; and a linear combi-
nation is used for the range between. The new Version 6 (V6) of the
algorithm that has been recently developed has coefcients depending
not only on the month of the year and BT
3132
, but also on six latitude
bands of the measurement to better capture the regional variability of
atmospheric conditions between the sea level and the satellite (Evans,
Minnett, & Podest, 2013). The latitude zones are 020, 2040 and
451 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
40 north and south of the Equator; this approach is similar to that used
for the ATSR (Along-Track Scanning Radiometer) atmospheric correc-
tion algorithm in which the zonal boundaries are at 25 and 50 north
and south (Zvody et al., 1995).
The global accuracy of the V5 retrievals is about 0.4 to 0.6 K when
expressed as a standard deviation about a mean that is close to zero for
comparisons with ship-based radiometric measurements of the skin
SST, or a mean close to the global average value of the thermal skin effect
when compared to sub-surface temperature measurement from drifting
buoys (Evans et al., 2013; Minnett, 2010). However, as withany statistical
approach, a question arises how accurate are these retrievals when they
are applied to conditions that depart fromthe average of the distribution
of those represented in derivation of the coefcients.
The objective of this study is to quantify the effects that the presence
and properties of dry layers in the atmosphere have on the SST retrieved
from infrared radiometers on spacecraft, with a focus on SST retrievals
using V5 and V6 of the MODIS algorithms.
2. Approach and data
2.1. Approach
To assess the dry layer effect we conducted numerical simulations of
MODIS TOA brightness temperatures and the retrieved SSTs for atmo-
spheres with dry layers at various levels, and compared the results
with simulations using atmospheres not affected by dry layers.
Incontext of this work we dene the accuracy of the MODIS retrieval,
SST, as the difference between the retrieved MODIS SST (SST
M
) and the
in situ SST (SST
i
),
SST SST
M
SST
i
: 2
The dry layer effect is the difference between the MODIS retrieval
accuracy for atmosphere with and without a dry layer.
The procedure employed is as follows:
(1) We used the Line by Line Radiative Transfer Model (LBLRTM;
Clough, Iacono, & Moncet, 1992; Clough & Iacono, 1995; Clough
et al., 2005) to compute the TOA radiance spectra in the infrared.
The LBLRTMrequires atmospheric proles of temperature and hu-
midity, as well as other atmospheric constituents, and the temper-
ature and emissivity of the underlying surface. As input to the
model computations we used:
a) Atmospheric proles measured by radiosondes during research
cruises in the northern tropical Atlantic (SAL region) in 2004
and 2006, and in the Eastern Indian Ocean (Timor Sea), north of
Darwin, Northern Territory, Australia, in 2006. The Timor Sea is
in the Tropical Warm Pool (TWP), and is another region where
extensive dry air layers in the atmosphere are observed. The
SAL radiosonde launches were generally timed so that the ra-
diosonde was in the air at the times of satellite overpasses;
those from the TWP were on a 3-hour schedule.
b) SSTs measured radiometrically from the ships by the Marine-
Atmosphere Emitted Radiance Interferometer (M-AERI, Minnett
et al., 2001) at the time of the radiosonde launches. When
M-AERI measurements were unavailable SST = T
ref
was assumed
for the SAL region and surface temperature of 28.9 C was
assumed for the TWP region. These temperatures represented
typical SST values in these regions during the cruises.
c) The atmospheric proles of six other atmospheric species
(CO
2
, O
3
, N
2
O, CO, CH
4
and O
2
) are those of the Standard Tropical
Atmosphere (STA).
d) The emissivity of the sea surface was taken from Filipiak (2008),
1
who gives refractive indices and emissivity of pure water and
seawater. The emissivities are tabulated as a function of
wave number (6003350 cm
1
), view angle (085), SST
(270310 K), and wind speed (025 ms
1
at 12.5 m height).
For the emissivity calculations, a wind speed of 2 ms
1
was
assumed for the entire dataset; the wind-speed dependence of
the apparent emissivity at these wavelengths in the infrared is
small (Hanan & Minnett, 2005).
e) Detector-specic relative spectral response functions (rsrs) were
obtained from NASA.
2
MODIS channel 31 and 32 rsrs, from the
instrument on Aqua, were used with the LBLRTM simulated
TOA radiance spectra to calculate the MODIS channel 31 and 32
radiances. Channel 31 and 32 radiances were then converted to
brightness temperatures T
31
and T
32
using inverse Planck func-
tion look-up-tables for the central wavelength of each channel.
We choose to conduct the calculations with the rsrs of detector
No 1, as the differences in the rsrs for the entire detector set are
small. Furthermore, this study focuses on differences in the simu-
lated BTs andthe derived SSTs introduced by dry layers, so uncer-
tainties in the measured rsrs become second order effects.
(2) Two values of the MODIS SST
M
were calculated for each set of
brightness temperatures T
31
and T
32
according to Eq. (1), SST
5
and SST
6
corresponding to versions 5 and 6 of the a coefcients
used with the SST algorithm (Eq. (1)). For this paper the
LBLRTM simulations and the SST retrievals were performed for
measurements at nadir. Since the skin SST is used in the simula-
tions, in Eq. (1) is set to zero.
(3) Version 5 and 6 MODIS SST simulated accuracies, SST
5
and
SST
6
, were calculated according to Eq. (2) where the in situ
SSTs were those used in the LBLRTM simulation.
2.2. LBLRTM input data
The Aerosol and OceanScience Expedition (AEROSE) was a multidis-
ciplinary oceanographic eld campaign conducted in the tropical North
Atlantic Ocean in March of 2004. The main goals of the expedition were
1) to obtain measurements of the microphysical properties and evolu-
tion of the Saharan Air Layer (SAL) and dust aerosol as they traverse
the Atlantic Ocean, and 2) to assess the impact on Saharan dust on the
atmospheric radiation and satellite remote sensing in visible and infra-
red wavelengths (e.g., Nalli et al., 2006, 2005). The 2004 SAL region
cruise took place on board of the NOAA ship Ronald H. Brown (RHB).
RHB set out from Bridgetown, Barbados, on 29 February and traveled
eastward toward Africa. Near the African coast, the ship turned north
toward the Canary Islands and, after a port call in Las Palmas de Gran
Canaria, the ship headed to San Juan, Puerto Rico, arriving on 26
March 2004. The cruise track is shown in Fig. 1. During the cruise radio-
sondes were launched fromthe RHB approximately every 3 h (e.g., Nalli
et al., 2005). The M-AERI was installed on board the ship where mea-
surements of the sea surface ahead of the bow-wave can be made, and
delivered radiometric measurements of skin SST approximately every
15 min.
The 2006 SAL region cruise (AEROSE-II) was conducted in conjunc-
tion with the African Monsoon Multidisciplinary Analyses (AMMA;
Redelsperger et al., 2006), an international project aimed at improving
the ability to predict and monitor weather and climate of West Africa
and Tropical Western Atlantic. The cruise took place on board of RHB
which headed east and south from San Juan, Puerto Rico, on 24 May
and after a port call in Recife, Brazil, returned to Charleston, SC on 15
July, 2006. Fig. 2a and b shows the tracks of legs 1 and 2 of the cruise.
The TWP cruise was conducted on the Australian research vessel
Southern Surveyor (SS) between 19 January and 13 February, 2006. This
cruise was part of the Tropical Warm Pool International Cloud Experi-
ment (TWP-ICE; May et al., 2008), an intense measurement campaign
1
Available at http://datashare.is.ed.ac.uk/
2
See ftp://mcst.hbsss-sigma.com/pub/permanent/MCST/FM1_RSR_LUT_07-10-01/
452 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
conducted as a collaborative effort between the US DOE ARM
(Atmospheric Radiation Measurement; Stokes & Schwartz, 1994)
project, the Australian Bureau of Meteorology, NASA, the European
Commission DGRTD-1.2 and several United States, Australian, Canadian
and European Universities. TWP-ICE was undertaken over a four week
period corresponding to the wet phase of the Australia monsoon with
a goal of describing the properties of tropical cirrus and the convection
that leads to their formation. Radiosondes were launched from the SS
every 3 h, and M-AERI skin SST measurements were made from the
ship every 15 min (Fig. 3).
The three measurement campaigns will be referred to hereafter as
SAL04, SAL06 and TWP06. In total, there are 359 radiosonde proles
from all three cruises for which the MODIS SST retrieval simulations
were performed. By visual inspection, the measured proles of the
SAL06 and TWP06 datasets were divided into four classes:
1 Proles that approximate the STA and show no dry layer
2 Proles with a distinct dry layer at any level
3 Proles with either weak or fragmented dry layer or proles that
cannot be clearly classied as class 1 or 2
4 Cloudy proles.
We do not consider the retrievals simulated for cloudy conditions to
be reliable and exclude cloudy proles from further analysis.
Fig. 4 shows examples of proles of classes 1, 2 and 3 from the
TWP06 dataset. The green and black lines are the measured temperature
(T), and dewpoint temperature (T
d
), and the blue and red lines are the T
and T
d
of the STA.
One might expect that the statistical MODIS SST retrieval coefcients
are in some way representative of typical atmospheric conditions for
the times of year when the MODIS measurements were taken, and in
the case of V6 coefcients also for the location of the measurement or
at least the latitude zone. Since the majority of the measurements we
are concerned with here take place in the tropics, one might expect
that the best agreement between the retrieved SST
5,6
(meaning the
derived SSTs using V5 and V6 algorithms) and the input SST
i
would
be obtained for proles that are close to the STA prole, presuming
this approximates well the mean atmospheric conditions used in the
derivation of the atmospheric correction algorithmcoefcients. The de-
parture of a measured prole from the STA prole could be seen as a
predictor of the SST
5,6
SST
i
difference. The STA prole also provides
guidance for allocating each radiosonde prole into the 4 categories
given above. Table 1 lists number of proles of each class for each of
the two research cruises.
The SAL04 dataset differs from the other two as it contains proles
with exceptionally deep dry layers. Therefore, for the SAL04 data we
modied the denition of prole classes as follows:
1 as before: proles that followthe STAprole and showno dry layer
2 as before: proles with distinct dry layer at any level
2 proles with a very deep dry layer
3 as before: proles with either weak or fragmented dry layer or
proles that cannot be clearly classied as class 1 or 2.
Additionally, there were four cloudy proles in the SAL04 data, but
these were not used in the analysis presented here.
An example of SAL04 Class 2 prole is shown in Fig. 5. The number
of proles of each class in the SAL04 dataset is shown in Table 2.
3. Results
3.1. Naturally occurring dry layers
TOA radiance spectra were calculated with the LBLRTMfor all the at-
mospheric proles of temperature andhumidity describedinSection2.2
and MODIS SST retrievals were simulated according to the procedure
outlined in Section 2.1. The average SST accuracies of retrievals with
versions 5 and 6 of the atmospheric correction algorithms were calcu-
lated for each class of proles in all datasets. The meanSST uncertainties
and standard deviation (in brackets) of retrievals for versions 5 and 6 of
the MODIS atmospheric correction algorithms are listed in Table 3 for
SAL06 and TWP06, and in Table 4 for the SAL04 dataset.
Also shown in Tables 3 and 4 are the differences between the
class-average measured PWV, calculated by integrating the water vapor
-70 -60 -50 -40 -30 -20 -10
0
10
20
30
15 20 25 30
Sea Surface Skin Temperature
o
C
Fig. 1. Track of Ronald H. Brown during the SAL04 cruise. Colors indicate the skin sea-surface temperature. The asterisks mark the positions of radiosonde launches.
453 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
prole measured by each radiosonde, and the STA total water vapor
(PWV
STA
=4.08 cm). This difference (PWV) can be viewed as a mea-
sure of the departure of the measured atmospheric proles from the
standard conditions. In theory, we would expect the retrievals to be
most accurate for the baseline (Class 1) cases as these represent the
state of the atmosphere closest to standard. This is indeed the case for
the TWP06 and SAL04 retrievals but not for the SAL06.
Our expectation is that the dry layers adversely affect the accuracies
of the MODIS SST retrievals. Since atmospheres with a dry layer will
likely have a lower PWV than others, one would expect the accuracy
of the retrieval to dependon the PWV or on the departure fromthe stan-
dard conditions (PWV). For a given dataset, PWV should also be
directly related to the class of atmospheric prole. This is indeed the
case. One can see in Tables 3 and 4 that in TWP06 and SAL06 datasets
the dry layer (Class 2) atmospheres are the driest followed by Class
3, and nally Class 1. In the SAL04 case, Class 2, the class with a very
deep dry layer, is the driest. Fig. 6ac shows how the average accuracy
of the V5 and V6 retrievals for each class of proles depends on the
class-average PWV in each dataset.
In considering the characteristics of the atmospheric proles, and
their inuence on the SST retrieval accuracies, it is useful to examine
both the class of prole as well as the PWV as each conveys information
that, while linked, is separately relevant. Also the geographic locations
play a role. There is a progression fromvery moist to very dry conditions
as we move from the TWP06 dataset, through SAL06 to SAL04. For the
TWP06, the average Class 1 prole is 1.16 cm moister than the STA
while for the SAL04 case the average Class 1 prole is only 0.11 cm
drier. In this sense the datasets span a wide range of conditions. The
best accuracy of both V5 and V6 retrievals is achieved for Class 1 of
TWP06 proles which are the moistest proles in all the datasets. In
Fig. 2. Track of Ronald H. Brown during Leg 1 (top) and Leg 2 (bottom) of SAL06. Colors indicate the skin sea-surface temperature.
454 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
terms of PWVSAL06 Class 1 is closer to Class 3 of TWP06 thanto Class 1
of the TWP06 dataset. The accuracy of the SST retrievals for Class 1 of
SAL06 is also closer to the accuracy of Class 3 of TWP06. For TWP06
the SST
5,6
(meaning SST
5
SST
i
and SST
6
SST
i
) for Class 3 are
increased with respect to Class 1 by factors close to 2 but for SAL06 the
increase in SST
5,6
going from Classes 1 to 3 is much less pronounced.
In terms of PWV, Class 1 of SAL04 falls between Classes 1 and 3 of
SAL06 and the similarity of these cases is also reected in the close
values of SST
5,6
. Class 2 of SAL04 is signicantly drier than Class 2 of
either TWP06 or SAL06 and this is also reected in larger SST
5,6
for
Class 2 of SAL04 than the other datasets. As expected, the largest SST
5,6
were obtained for Class 2 of the very dry SAL04: SST
5
is 75% and
SST
6
is 84% higher than average values for proles with no dry layer.
Looking at Fig. 6ac together one also notices that as the proles be-
come drier going from 1.16 cm to about 0.5 cm above STA, the SST
5,6
increase, and then between 0.5 cm and about 1.0 cm below STA
the SST
5,6
almost level off only to increase again for proles drier
than 1.0 cm.
It is also noteworthy that V5 retrievals are more accurate for all cases
except Class 2 in TWP06, implying that even in the base state, without a
dry layer, the atmospheres in the study areas are somewhat drier than
the zonal averaged conditions in the tropics.
3.2. Dry layer experiments
To gain a better understanding of the impact of dry layers on the ac-
curacy of MODIS SST retrievals we conducted numerical experiments
where we simulated MODIS SST retrievals for atmospheric conditions
that were created from the measured radiosonde proles by drying
layers at various altitudes. This allowed us to control the strength and
altitude of the dry layer and to determine their impact on the retrieval
accuracy. All available proles are subjected to this modication.
The drying of a prole was accomplished by reducing the water
vapor volume mixing ratio to 10% of the original value in a selected
layer. Four layers were considered: 250 to 500 hPa, 450 to 700 hPa,
550 to 800 hPa, and 750 to 900 hPa. As an example Fig. 7 shows
Class 1 prole from Fig. 4 that has been dried between 450 and
700 hPa.
As before, we examine proles in the classes previously dened in
Section 3. The class denition applies to the radiosonde prole as
originally measured.
Since most of the atmospheric moisture resides in the boundary
layer andlower troposphere, drying layers closer to the surface typically
leads to greater reduction in PWV than drying a layer higher in the at-
mosphere. Thus, examining changes in SST
5,6
as a function of the alti-
tude of the dry layer includes a component determinedby the variations
in PWV. In Fig. 8a, b we show, for the three classes of the original pro-
les of the SAL06 dataset, howthe SST
5,6
change in response to drying
of the proles at different levels in the atmosphere. In Fig. 8 and the
following gures the drying is expressed in terms of PWV. The geo-
graphical area of the SAL06 dataset is large. The SAL evolves as it advects
to the west and this can lead to regional variations in the errors in the
SST retrievals (e.g. Vzquez-Cuervo, Armstrong, & Harris, 2004). To in-
vestigate regional variations, we divided all radiosondes into three geo-
graphical regions: north-east (NE), north-west (NW), and South (S).
The zonal division runs along the Equator, and the meridional division
is along 30W. In Fig. 8 we present the results for all proles together,
and NE and NW proles separately. We do not present the southern
proles separately since the number of proles in this section is small
(7). Fig. 9 is as Fig. 8, but for the TWP06 dataset.
129 129.50 130 130.50 131
-13
-12.50
-12
-11.50
-11
26 28 30 32
Sea Surface Skin Temperature
o
C
Fig. 3. Track of Southern Surveyor during TWP06. Colors indicate the skin sea-surface temperature.
455 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
For both datasets the relationship between SST
5,6
and PWV is very
similar for all classes of proles. The right-most data point in all panels
of Figs. 8 and 9 represents the original unaltered prole. As discussed
before, the change to the PWV is the smallest when the drying takes
place at the highest level (250 to 500 hPa). So, going from right to left
in the panels of Figs. 8 and 9 corresponds to drying lower and lower
layers of the atmosphere. As can be seen in the plots, as we move the
dry layer downward in the atmosphere, SST
5,6
initially decrease then
almost level off and then increase signicantly for the dry layer at the
lowest level (750 to 900 hPa).
Version 6 retrievals perform better (i.e. give lower values of mean
SST) than version 5 only for Class 2 of the TWP06 dataset. For SAL06,
except Class 2 V6 retrievals, there is no signicant difference in the re-
trieval accuracy for NE and NW subset even though Class 2 NE proles
are on average drier than their NW counterparts. Thus, for the SAL04
dataset that also stretches over a large geographical area we show
only class averages over the entire region. The SST
5,6
and PWV rela-
tionship for SAL04 is shown in Fig. 10. The character of this relationship
is again very similar to what we see in Figs. 8 and 9 but the values of
SST
5,6
are greater. Since the SAL04 proles are in general drier than
in the other datasets the range of PWV is smaller. This is especially
true for Class 2 of SAL04, i.e. the deep dry layer proles (the percentage
change of a small PWV is small). Figs. 8 to 10, showing the effects of dry
layers on the MODIS SST retrievals, indicate that unaltered Class 2 pro-
les (rightmost data point in the top right panels) are similar to class
1 proles where a dry layer has been introduced. A comparison of
these original dry layer cases with the cases of Class 1 proles with
dry layers introduced at various levels (shown the top left panel) can
serve as a consistency check; i.e. one expects the SST of the original
dry layer prole to be close to the SST of the synthetic prole with
the dry layer at the level representative of the typical level of the dry
layer in the group of observed proles. That is in fact the case for all
three datasets. For the TWP and the SAL06 data the SST for the dry
layer is in the range of SST for Class 1 proles that were dried out at
the level of approximately 750 hPa which corresponds closely to the
median of the observed base levels of the dry layer in TWP (725 hPa)
and SAL06 (775 hPa). Similarly, for the SAL04 dataset the observed
Class 2 SST corresponds to Class 1 dried in a layer with a base at ap-
proximately 850 mb which is close to the observed dry layer base
level of Class 2 proles (median 875 hPa). For Class 2 the median ob-
served dry layer base level is at 962 hPa and the SST corresponds clos-
est to a synthetic prole with the dry layer based at 900 hPa.
3.3. MODIS SST algorithm
To better understand the results of Sections 3.1 and 3.2 we examine
more closely the MODIS SST algorithm. For a satellite zenith angle of 0
Fig. 4. TWP06 dataset, examples of prole classes: Class 1 (top-left), Class 2 (top-right)and Class 3 (bottom-left). The green and black lines are the measured T and T
d
; the blue and red
dashed lines are the T and T
d
of the Standard Tropical Atmosphere (STA). PWV and PWVSTA give precipitable water vapor measured by radiosonde and that of the Standard Tropical
Atmosphere.
Table 1
Number of radiosonde proles of each class for the SAL06 and TWP06 cruises.
Cruise Class 1 Class 2 Class 3 Total
SAL06 13 15 50 74
TWP06 37 18 49 104
456 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
the only varying terms in Eq. (1) are the second and third. If the actual
SST remains unchanged, how these terms change in relation to each
other as the atmospheric proles change determines whether the
retrieved SST
5,6
increase or decrease, and hence howthe retrieval uncer-
tainty is inuenced.
Our drying experiment can be traced in the plots shown in Figs. 11
to 13, which show the change in term 3 of Eq. (1) between the dried
prole and the original prole plotted against the change in term 2 for
each class of proles of the three datasets: SAL06 in Fig. 11, TWP06 in
Fig. 12 and SAL04 in Fig. 13. Figs. 11a, 12a and 13a show the plots for
V5 retrievals and Figs.11b, 12b, and 13b for V6. Each symbol corre-
sponds to a single prole that has been modied by drying a certain
layer in the atmosphere. Blue dots indicate a dry layer between 250
and 500 hPa, red stars a dry layer between 450 and 700 hPa, green cir-
cles a dry layer between 550 and 800 hPa, and the purple squares a dry
layer between 700 and 900 hPa. The black y = x line indicates no-
change. If the symbols fall on this line the changes in the contributions
to the atmospheric correction of terms 2 and 3 cancel and there is no
change in the retrieved SST
5,6
, i.e. the presence of a dry layer does not in-
troduce additional uncertainties in the SST retrieval. If the symbols fall
above the no-change line the increase in term 2 is greater than the
drop in term 3 and the retrieved SST
5,6
is greater than the SST
5,6
re-
trieved for the original prole. The opposite is true for the points
below the no-change line.
Figs. 11 to 13 reveal that the response of the MODIS retrieved SST
5,6
(whether it increases or decreases) as dry layers are introduced to the
intervening atmosphere depends largely on the altitude of the dry
layer. The changes are smallest in Class 2 proles as these originally
contained a dry layer and the additional drying of the atmosphere has
a more muted effect.
As seen in these gures, if dry layers occupy the lower atmosphere
(700900 hPa), the MODIS algorithm tends to retrieve a higher SST
compared to the standard no dry layer conditions (purple squares). If
a dry layer is located higher in the atmosphere (250500 hPa, and
450700 hPa) the retrieved SST
5,6
are lower than for standard condi-
tions (blue dots and red stars). This behavior is typical for all classes of
proles, all datasets and both versions of the algorithm. The case of
dry layer at 550800 hPa shows most variability between datasets, pro-
le class, and algorithm version. Another dependence revealed by the
plots is that typically the greater the absolute change in terms 2 and 3
the greater the change in SST
5,6
.
4. Discussion
The focus of this research is the introduction of errors in MODIS SST
retrievals over specic areas of the ocean with tropical and subtropical,
typically very moist, atmospheres by the presence of dry layers inthe at-
mosphere. We have focused on the mean errors (error being dened in
Eq. (2)) caused by the presence of dry layers, and not the scatter in the
errors, as the undetected increases in bias errors can lead to signicant
adverse consequences in many applications of satellite-derived SSTs,
especially in climate predictions (e.g. Ashfaq, Skinner, & Diffenbaugh,
2011) and in the depiction of tropical conditions in climate models
(e.g. Li & Xie, 2012).
For the atmospheric state most resembling the STA, SST retrievals
are overestimated by 0.29 to 0.49 K for the V5 algorithm, and by 0.36
to 0.72 K for V6. The best results were obtained for the moistest
atmospheres.
This overestimate is not surprising for V5 retrievals as the retrieval
coefcients were obtained for optimal global performance and might
not represent very accurately the specic location, season and condition
of the datasets considered here. One would expect the V6 to perform
better as the coefcients are latitude dependent but this is not the
case for any of the datasets considered here. The best V6 results were
obtained for the TWP06. This could be related to the very small geo-
graphical extent of the TWP06 data compared to the SAL04 and 06
datasets, and that the atmospheric conditions appear to coincide well
with those of the tropical zone.
The departure of atmospheric conditions from the STA caused by
intrusions of dry air layers typically leads to increases in SST
5,6
. The
effects of the dry layer in the retrieved MODIS SST can be summarized
as follows:
1) SST
5,6
depends on the amount of water vapor in the atmosphere
(PWV)
Fig. 5. Example of SAL04 Class 2 prole (very deep dry layer). The green and black lines
are the measured T and T
d
; the blue and red dashed lines are the T and T
d
of the Standard
Tropical Atmosphere (STA). PWVandPWVSTAgive precipitable water vapor measuredby
radiosonde and that of the Standard Tropical Atmosphere.
Table 2
Number of radiosonde proles of each class for the SAL04 cruises.
Cruise Class 1 Class 2 Class 2 Class 3 Total
SAL04 17 45 44 42 148
Table 3
Mean SST uncertainties and standard deviation (in brackets) of retrievals for versions 5
and 6 of the MODIS atmospheric correction algorithms. SAL06 and TWP06 cruises.
Cruise SAL06 TWP06
SST
5
[K]
SST
6
[K]
PWV
[cm]
SST
5
[K]
SST
6
[K]
PWV
[cm]
Class 1 0.44 (0.23) 0.69(0.24) 0.35 0.29 (0.29) 0.36 (0.24) 1.16
Class 2 0.35 (0.21) 0.65 (0.22) 0.95 0.66 (0.22) 0.51 (0.17) 0.73
Class 3 0.45 (0.33) 0.84 (0.40) 0.35 0.57 (0.34) 0.63 (0.28) 0.57
Table 4
Mean SST uncertainties and standard deviation (in brackets) of retrievals for versions 5
and 6 of the MODIS atmospheric correction algorithms. SAL04 cruise.
Cruise SAL04
SST
5
[K] SST
6
[K] PWV [cm]
Class 1 0.49 (0.18) 0.72 (0.17) 0.11
Class 2 0.67 (0.22) 0.95 (0.22) 1.6
Class 2 0.86 (0.12) 1.19 (0.23) 2.74
Class 3 0.55 (0.19) 0.90 (0.18) 0.95
457 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
2) The simulated MODIS SST
5,6
change in a systematic way if the inter-
vening atmosphere is modied to include dry layers
3) Dry layers in the lower troposphere lead to more positive SST
5,6
4) Dry layers in the upper troposphere lead to more negative SST
5,6
.
Thus, the presence and properties of a dry layer in the atmosphere
do inuence the uncertainty in the MODIS SST retrievals.
Our results are in broad agreement with those of the recent study
of Barton (2011) who investigated another SST algorithm (slightly dif-
ferent than MODIS) and concludes that knowledge of the vertical distri-
bution of water vapor can positively inuence the SST retrieval accuracy
for the two (11 and 12 m) channel SST retrieval algorithm. For exam-
ple, Barton (2011) shows that an atmospheric prole with
PWV b 4.5 cm and with a dry layer above 750 hPa leads to SST of ap-
proximately 0.4 K which is within the range of SST reported in
Tables 3 and 4. We conclude that the retrieval coefcients which are ef-
fective for standard conditions are not necessarily appropriate for atmo-
spheres with vertically extensive or strong dry layers, i.e. they are not
robust to the presence of dry layers. This endorses the results of an earlier
numerical study of the effects of anomalously dry atmospheric conditions
over the North Atlantic on TOA BTs (Minnett, 1986).
Since the occurrence of a dry layer is typically accompanied by
cloudless skies the satellite retrievals of SST under these conditions
are likely to be agged as high quality even though the SSTs are poten-
tially large. Thus, the dry layer effect might be behind some of the out-
liers, both positive and negative, of the SST
5,6
distributions.
The accuracy of the MODIS SST could be improved if
1) a special set of retrieval coefcients were developed to represent the
dry layer conditions
2) a way to identify dry layers (ideally using the satellite measurements
themselves) were to be developed so dry layer coefcients could
be applied whenever a dry layer were identied in the atmosphere.
In fact, point 2) might have to be addressed rst so a database of
MODIS SST measurements coincident with in situ SST measurements
under dry layer atmospheres can be assembled.
In Fig. 14 we show the difference of the LBLRTM calculated bright-
ness temperatures in MODIS channels 31 and 32 (BT
3132
) versus the
precipitable water vapor of the atmosphere derived from radiosonde
measurements for all available radiosonde proles of the three datasets
TWP06, SAL04 and SAL06. There is an approximately linear dependence
between BT
3132
and the PWV, at least in range of PWV between 1 and
8 cm. Thus the brightness temperature difference between MODIS chan-
nels 31 and 32 canpossibly be used as a measure of the PWVin the atmo-
sphere, as similarly demonstrated by Dunion and Velden (2004) using
GOES data, and by Dalu (1986) and Eck and Holben (1994) for AVHRR.
Low values of BT
3132
should be indicative of dry atmosphere and
high values will represent moist conditions. This relationship supports
the concept of using BT
3132
to divide the PWV range for the generation
of coefcients for the atmospheric correction algorithm. Typical tropical
atmospheres are moist (PWV N 4 cm) andsuchconditions correspondto
large channel 31and32brightness temperature differences. Anobserved
BT
3132
lower than a threshold value for a given region over the tropical
ocean could indicate the presence of a dry layer.
In case of the SAL region the dry air outbreaks often carry large load-
ings of Saharan dust (Zhang & Pennington, 2004). Thus the existence of
the dry layers in this area can often be conrmed with measurements
from the MODIS channels in the visible part of the spectrum sensitive
to dust aerosols (Remer et al., 2005), but this option is available only
during the sunlit part of each orbit. During the night there is evidence
that the presence of dust can produce a signature in the infrared
bands (Merchant et al., 2006). This results, at least in part, from the
spectral wings of the reststrahlen feature of silica (SiO
2
), a main compo-
nent of the Saharan aerosol layer (Moreno et al., 2006). The reststrahlen
results from a resonance between the frequency of the infrared radia-
tion and the vibrations between the silicon and oxygen atoms in
the silica crystal lattice that is centered at a wavelength near 9 m
(Salisbury, Wald, & D'Aria, 1994) andcauses an increase in the spectral
reectivity at and around the resonant wavelength. In the 1012 m
wavelength range where SST is measured from space, the reststrahlen
effects are greater at shorter wavelengths; this is contrary to the effect
Fig. 6. MODIS SST average class retrieval accuracy vs average class PWVdecit withrespect
to Standard Tropical Atmosphere for (a) SAL06, (b) TWP06 and (c) SAL04 data sets.
Fig. 7. Example of prole modied by drying a layer (black line). PWVorg and PWVdry
give precipitable water vapor of the original prole and the prole modied by the intro-
duction of the dry layer correspondingly.
458 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
of water vapor. By reducing the BT
3132
the silica aerosols reinforce the
radiative effects of a dry layer. There are additional geometrical scatter-
ing effects as the single scattering albedo, specic extinction coefcient
and asymmetry factor of Saharan mineral dust all decrease with
wavelength for N 10 m (Haywood et al., 2005) so the mineral dust
layer can result in an increase in the BT
32
relative to BT
31
(Highwood,
Haywood, Silverstone, Newman, &Taylor, 2003). The situation is further
complicated as radiative effects of the mineral dust in the infrared
Fig. 8. SAL06 dataset: average change in (a) MODIS V5 and (b) MODIS V6 SST relative to true SST in response to introduction of dry layer to the original proles. Rightmost data point
represents the original prole. Decreasing PWV corresponds to lower altitude of the dry layer. Sum of all proles (blue) as well as NE (red) and NW (green) region proles
plotted separately.
Fig. 9. TWP06 dataset: average change in (a) MODIS V5 and (b) MODIS V6 SST relative to true SST in response to introduction of dry layer to the original proles. Rightmost data point
represents the original prole. Decreasing PWV corresponds to lower altitude of the dry layer.
459 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
transmission window are dependent on the height of the dust
(Merchant et al., 2006), as are the effects of the dry layer. If the existence
of a dry layer is conrmed, the SST retrieval algorithm should use the
dry layer coefcient set instead of the standard set, as mentioned
above. However, the combined presence of mineral dust and a dry
layer leads to signals in the infrared that may be difcult to disentangle,
as the dust signature also inuences the TOA BTs in the 1012.5 m at-
mospheric window: thus the presence of a combineddry layer withmin-
eral dust could lead to signals in the infrared that are indistinguishable
froma deepdry layer inthe absence of dust, or a dust layer inthe absence
of a dry layer. This reinforces the need to use independent information,
such as the reected solar radiation at shorter wavelengths, to help iden-
tify and characterize the dust during the day, but these estimates are also
subject to uncertainty (Redemann et al., 2006) that can introduce addi-
tional complications to the SST retrieval. The retrieval of accurate SSTs
in the presence of dry layers, with and without the attendant mineral
dust, is clearly a topic requiring additional research.
With further improved prole retrievals fromatmospheric sounding
instruments on satellites, and more accurate atmospheric radiative
transfer modeling, the prospect of deriving SSTs as part of the prole re-
trieval process or by using the derived proles to solve the forward
problem(Merchant &LeBorgne, 2004) becomes increasingly attractive;
this would potentially avoid the consequences of anomalous atmo-
spheric conditions on atmospheric correction algorithms basedon a sta-
tistical approach. Another avenue worthexploring is the use of radiative
transfer modeling with atmospheric proles derived from Numerical
Weather Prediction (NWP) models (Le Borgne, Roquet, & Merchant,
2011), as, for example, ECMWF (European Center for Medium-range
Weather Forecasts) provide a good representation of the SAL in their
modeled elds (Szczodrak, Minnett, Nalli, & Feltz, 2007).
5. Summary and conclusions
Using infrared radiative transfer modeling through the atmosphere,
with properties determined by ship-launched radiosonde proles, we
have investigated the effects of atmospheric dry layers on the accuracy
of retrievals of skin SST using a split-window atmospheric correction
algorithm. The data are taken from research cruises where dry layers
were encountered: in the Tropical Atlantic Oceanwhere the dry Saharan
Air Layer occurs, with and without associated mineral dust aerosols, and
froma cruise in the Timor Sea off northwest Australia. To permit a more
thorough analysis of the effects of dry layers, the natural dry layers
sampled by the radiosondes were augmented by introducing articial
dry layers in the proles. The results from naturally occurring and
articial dry layers were comparable.
The clear-sky atmospheric correction algorithms developed for
MODIS on Aqua were used with the simulated brightness temperatures
to assess the impacts of the presence of dry layers. The dry layers intro-
duce bias errors in the retrieved SSTs that are signicant in comparison
to the near-zero bias errors found in analyses of large oceanareas and in
comparison with the accuracies required for climate studies (~0.1 K;
Ohring, Wielicki, Spencer, Emery, & Datla, 2005). The presence of such
Fig. 10. SAL04 dataset: average change in (a) MODIS V5 and (b) MODIS V6 SST relative to true SST in response to introduction of dry layer to the original proles. Rightmost data point
represents the original prole. Decreasing PWV corresponds to lower altitude of the dry layer.
460 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
biases has potentially severe consequences on the use of satellite SST
elds in climate research (e.g. Ashfaq et al., 2011; Li & Xie, 2012) as
well as other applications.
While it is feasible to identify some cases where Saharan dust aero-
sols compromise the SST retrievals and reduce the SST retrieval error by
using an atmospheric correction algorithm that is more robust to the
presence of such aerosols (e.g. Merchant et al., 2006), correcting for
the broader problem of dry layers, with or without aerosols, poses a
more difcult problem. The variability in the thickness, height and
humidity of the dry layers results in ambiguous signals in the BT
3132
(Fig. 14) and therefore prevents a selection of an appropriate set of coef-
cients (or algorithm formulation) based on the MODIS measurements
alone. Ancillary information about the presence and characteristics of
dry layers would be needed to provide a robust retrieval of SST from
multi-spectral infrared imaging radiometers in such conditions. Alterna-
tives to measuring SST in the presence of dry layers include dual-view
infrared radiometry and microwave radiometry as these have been
shown to be less inuenced by aerosols (Corlett et al., 2006; Reynolds,
Gentemann, & Corlett, 2010) and are expected to be more robust to
anomalous water vapor distributions. However, dual-view radiometers
have a narrow swath (~500 km) and microwave radiometers have a
poor spatial resolution (~50 km).
The limited information in the measurements of the BTs measured
in two spectral intervals renders any statistical atmospheric correction
algorithm susceptible to error in anomalous conditions. An alternative
approach to the SST retrieval problem is an Optimum Estimation tech-
nique based on an accurate representation of the atmospheric state
and simulation of the measured BTs using radiative transfer modeling,
Fig. 11. SAL06 dataset: Difference between the value of term3 of the (a) MODIS V5 and (b) MODIS V6 SST retrieval algorithmfor a prole modied by drying a layer and the value of this
term for the original prole vs. the corresponding difference in values of term 2 of the algorithm.
461 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
as used in this study. Such an approach has been shown to be feasible
when applied to infrared measurements of SEVIRI (Spinning Enhanced
Visible and Infrared Imager) on the EUMETSAT Meteosat geosynchro-
nous satellite (Merchant, LeBorgne, Marsouin, &Roquet, 2008). Adapting
such an approach to MODIS and other well-calibrated infrared imagers is
a topic worthy of further research.
Although this study has been directed towards the infrared SST
retrievals of MODIS on Aqua, the results and conclusions are applicable
to SST retrievals from other scanning infrared radiometers on polar-
orbiting satellites, such as the AVHRR on the current NOAA-n and
Met-Op series and the Visible Infrared Imaging Radiometer Suite
(VIIRS) on the Suomi-NPP satellite, and on Joint Polar Satellite System
(JPSS) satellites inthe future, andradiometers ongeostationary satellites.
The numerical results presented here will of course change in value
because each sensor has its own specic spectral response functions,
but the qualitative results are expected to hold across a wide range of
sensors.
Acknowledgments
This study was supported through the NASA Physical Oceanography
Program (NNG04HZ33C and NNX07AE83G). The radiosonde data from
the TWP-ICE cruise of the RV Southern Surveyor were provided by the
Australian Bureau of Meteorology (Alex Williams), and from the NOAA
S Ronald H. Brown, by NOAA/NESDIS (Nick Nalli). The at-sea support of
the captains, ofcers and crews of the ships is greatly appreciated. The
ECMWF ERA-Interim data used in this study were obtained from the
ECMWF data server. The valuable suggestions of three anonymous
reviewers have helped to clarify the presentation.
Fig. 12. TWP06 dataset: Difference between the value of term3 of the (a) MODIS V5 and (b) MODIS V6 SST retrieval algorithmfor a prole modied by drying a layer and the value of this
term for the original prole vs. the corresponding difference in values of term 2 of the algorithm.
462 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
Fig. 13. SAL04 dataset: Difference between the value of term3 of the (a) MODIS V5 and (b) MODIS V6 SST retrieval algorithmfor a prole modied by drying a layer and the value of this
term for the original prole vs. the corresponding difference in values of term 2 of the algorithm.
463 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
References
Ashfaq, M., Skinner, C. B., & Diffenbaugh, N. S. (2011). Inuence of SST biases on future
climate change projections. Climate Dynamics, 36, 13031319.
Barton, I. J. (2011). Improving satellite-derived sea surface temperature accuracies using
water vapor prole data. Journal of Atmospheric and Oceanic Technology, 28, 8593.
Cau, P., Methven, J., & Hoskins, B. (2007). Origins of dry air in the tropics and subtropics.
Journal of Climate, 20, 27452759.
Clough, S. A., & Iacono, M. J. (1995). Line-by-line calculations of atmospheric uxes and
cooling rates II: Application to carbon dioxide, ozone, methane, nitrous oxide, and
the halocarbons. Journal of Geophysical Research, 100, 16,519516,535.
Clough, S. A., Iacono, M. J., & Moncet, J. -L. (1992). Line-by-line calculation of atmospheric
uxes and cooling rates: Application to water vapor. Journal of Geophysical Research,
97, 1576115785.
Clough, S. A., Shephard, M. W., Mlawer, E. J., Delamere, J. S., Iacono, M. J., Cady-Pereira, K.,
et al. (2005). Atmospheric radiative transfer modeling: A summary of the AER codes.
Journal of Quantitative Spectroscopy and Radiative Transfer, 91, 233244.
Corlett, G. K., Barton, I. J., Donlon, C. J., Edwards, M. C., Good, S. A., Horrocks, L. A., et al.
(2006). The accuracy of SST retrievals from AATSR: An initial assessment through
geophysical validation against in situ radiometers, buoys and other SST data sets.
Advances in Space Research, 37, 764769.
Cracknell, A. P. (1997). The advanced very high resolution radiometer. Taylor and Francis
(534 pp.).
Dalu, G. (1986). Satellite remote sensing of atmospheric water vapour. International
Journal of Remote Sensing, 7, 10891097.
Donlon, C. J., Minnett, P. J., Gentemann, C., Nightingale, T. J., Barton, I. J., Ward, B., et al.
(2002). Toward improved validation of satellite sea surface skin temperature measure-
ments for climate research. Journal of Climate, 15, 353369.
Dunion, J. P., &Velden, C. S. (2004). The impact of the Saharan air layer on Atlantic tropical
cyclone activity. Bulletin of the American Meteorological Society, 85, 353365.
Eck, T. F., & Holben, B. N. (1994). AVHRR split window temperature differences and total
precipitable water over land surfaces. International Journal of Remote Sensing, 15,
567582.
Esaias, W. E., Abbott, M. R., Barton, I., Brown, O. B., Campbell, J. W., Carder, K. L., et al.
(1998). An overview of MODIS capabilities for ocean science observations. IEEE
Transactions on Geoscience and Remote Sensing, 36, 12501265.
Evans, R. H., Minnett, P. J., & Podest, G. (2013). Status of MODIS AQUA and TERRA SST;
Transition from V5 to V6. Presented at MODIS Science Team Meeting (Available
from http://modis.gsfc.nasa.gov/sci_team/meetings/201304/plenary.php)
Filipiak, M. J. (2008). Refractive indices (5003500 cm
1
) and emissivity (6003350 cm
1
)
of pure water and seawater [dataset]. http://hdl.handle.net/10283/17
Hanan, J. A., &Minnett, P. J. (2005). Infrared-emissivity measurements of a wind-roughened
sea surface. Applied Optics, 44, 398411.
Haywood, J. M., Allan, R. P., Culverwell, I., Slingo, T., Milton, S., Edwards, J., et al. (2005).
Can desert dust explain the outgoing longwave radiation anomaly over the Sahara
during July 2003? Journal of Geophysical Research, 110, D05105.
Highwood, E. J., Haywood, J. M., Silverstone, M.D., Newman, S. M., & Taylor, J. P. (2003).
Radiative properties and direct effect of Saharan dust measured by the C-130 aircraft
duringSaharanDust Experiment (SHADE): 2. Terrestrial spectrum. Journal of Geophysical
Research, 108, 8578.
Kidder, S. Q., & Haar, T. H. V. (1995). Satellite meteorology: An introduction. Academic Press
(466 pp.).
Kilpatrick, K. A., Podest, G. P., & Evans, R. H. (2001). Overview of the NOAA/NASA path-
nder algorithm for sea surface temperature and associated matchup database.
Journal of Geophysical Research, 106, 91799198.
Le Borgne, P., Roquet, H., & Merchant, C. J. (2011). Estimation of sea surface temperature
from the spinning enhanced visible and infrared imager, improved using numerical
weather prediction. Remote Sensing of Environment, 115, 5565.
Li, G., &Xie, S. -P. (2012). Origins of tropical-wide SST biases inCMIPmulti-model ensembles.
Geophysical Research Letters, 39, L22703.
Llewellyn-Jones, D. T., Minnett, P. J., Saunders, R. W., & Zvody, A.M. (1984). Satellite
multichannel infrared measurements of sea-surface temperature of the N.E. Atlantic
Ocean using AVHRR/2. Quarterly Journal of the Royal Meteorological Society, 110,
613631.
Martin, S. (2004). An introduction to ocean remote sensing. : Cambridge University Press
(426 + xxvii pp.).
May, P. T., Mather, J. H., Vaughan, G., Jakob, C., McFarquhar, G. M., Bower, K. N., et al. (2008).
The tropical warm pool international cloud experiment. Bulletin of the American
Meteorological Society, 89, 629645.
May, D. A., Parmeter, M. M., Olszewski, D. S., & McKenzie, B.D. (1998). Operational pro-
cessing of satellite sea surface temperature retrievals at the Naval Oceanographic
Ofce. Bulletin of the American Meteorological Society, 79, 397407.
McMillin, L. (1975). Estimation of sea-surface temperatures fromtwo infrared windowmea-
surements with different absorption. Journal of Geophysical Research, 80, 51135117.
Merchant, C. J., Embury, O., Le Borgne, P., & Bellec, B. (2006). Saharan dust in nighttime
thermal imagery: Detection and reduction of related biases in retrieved sea surface
temperature. Remote Sensing of Environment, 104, 1530.
Merchant, C. J., & LeBorgne, P. (2004). Retrieval of sea surface temperature from space,
based on modeling of infrared radiative transfer: Capabilities and limitations.
Journal of Atmospheric and Oceanic Technology, 21, 17341746.
Merchant, C. J., LeBorgne, P., Marsouin, A., & Roquet, H. (2008). Optimal estimation of sea
surface temperature from split-window observations. Remote Sensing of Environment,
112, 24692484.
Minnett, P. J. (1986). A numerical study of the effects of anomalous North Atlantic atmo-
spheric conditions on the infrared measurement of sea-surface temperature from
space. Journal of Geophysical Research, 91, 85098521.
Minnett, P. J. (2010). The validation of sea surface temperature retrievals fromspaceborne
infrared radiometers. In V. Barale, J. F. R. Gower, &L. Alberotanza (Eds.), Oceanography
from space, revisited (pp. 273295). Springer Science+Business Media B.V.
Minnett, P. J., & Barton, I. J. (2010). Remote sensing of the earth's surface temperature. In
Z. M. Zhang, B. K. Tsai, & G. Machin (Eds.), Radiometric temperature measurements and
applications (pp. 333391). Academic Press/Elsevier.
Minnett, P. J., Knuteson, R. O., Best, F. A., Osborne, B. J., Hanan, J. A., & Brown, O. B. (2001).
The Marine-Atmospheric EmittedRadiance Interferometer (M-AERI), a high-accuracy,
sea-going infrared spectroradiometer. Journal of Atmospheric and Oceanic Technology,
18, 9941013.
Minnett, P. J., Smith, M., & Ward, B. (2011). Measurements of the oceanic thermal skin
effect. Deep Sea Research Part II: Topical Studies in Oceanography, 58, 861868.
Moreno, T., Querol, X., Castillo, S., Alastuey, A., Cuevas, E., Herrmann, L., et al. (2006).
Geochemical variations in aeolian mineral particles from the SaharaSahel Dust
Corridor. Chemosphere, 65, 261270.
Nalli, N. R., Clemente-Coln, P., Minnett, P. J., Szczodrak, M., Jessup, A., Branch, R., et al.
(2006). Ship-based measurements for infrared sensor validation during AEROSE
2004. Journal of Geophysical Research, 111, D09S04.
Nalli, N. R., Clemente-Coln, P., Morris, V., Joseph, E., Szczodrak, M., Minnett, P. J., et al.
(2005). Prole observations of the Saharan air layer during AEROSE 2004. Geophysical
Research Letters, 32.
Nalli, N. R., & Stowe, L. L. (2002). Aerosol correction for remotely sensed sea surface tem-
peratures from the National Oceanic and Atmospheric Administration advanced very
high resolution radiometer. Journal of Geophysical Research, 107.
Ohring, G., Wielicki, B., Spencer, R., Emery, B., & Datla, R. (2005). Satellite instrument
calibration for measuring global climate change: Report of a workshop. Bulletin of
the American Meteorological Society, 86, 13031313.
Prospero, J. M. (1999). Long-range transport of mineral dust in the global atmo-
sphere: Impact of African dust on the environment of the southeastern United
States. Proceedings of the National Academy of Sciences of the United States of America,
96, 33963403.
Prospero, J. M., & Carlson, T. N. (1972). Vertical and areal distribution of Saharan dust over
the western Equatorial North Atlantic Ocean. Journal of Geophysical Research, 77,
52555265.
Redelsperger, J. -L., Thorncroft, C. D., Diedhiou, A., Lebel, T., Parker, D. J., &Polcher, J. (2006).
African monsoon multidisciplinary analysis: An international research project and
eld campaign. Bulletin of the American Meteorological Society, 87, 17391746.
Redemann, J., Zhang, Q., Schmid, B., Russell, P. B., Livingston, J. M., Jonsson, H., et al.
(2006). Assessment of MODIS-derived visible and near-IR aerosol optical properties
and their spatial variability in the presence of mineral dust. Geophysical Research
Letters, 33, L18814.
Remer, L. A., Kaufman, Y. J., Tanr, D., Mattoo, S., Chu, D. A., Martins, J. V., et al. (2005). The
MODIS aerosol algorithm, products, andvalidation. Journal of the Atmospheric Sciences,
62, 947973.
Reynolds, R. W., Gentemann, C. L., & Corlett, G. K. (2010). Evaluation of AATSR and TMI
satellite SST data. Journal of Climate, 23, 152165.
Reynolds, R. W., Smith, T. M., Liu, C., Chelton, D. B., Casey, K. S., &Schlax, M. G. (2007). Daily
high-resolution-blended analyses for sea surface temperature. Journal of Climate, 20,
54735496.
Salisbury, J. W., Wald, A., & D'Aria, D.M. (1994). Thermal-infrared remote sensing and
Kirchhoff's law: 1. Laboratory measurements. Journal of Geophysical Research: Solid
Earth, 99, 1189711911.
0 1 2 3 4 5 6 7 8
0
0.5
1
1.5
2
2.5
3
PWV
sonde
[ cm ]
B
T
3
1

B
T
3
2

[

K

]
SAL 2004
SAL 2006
TWP 2006
Class 1
Class 2
Class 3
Class 4
Class 2
Fig. 14. LBLRTMbrightness temperature difference betweenMODIS channels 31and 32 vs.
precipitable water vapor for all measured radiosonde proles of the three dataset: TWP06,
SAL04 and SAL06.
464 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465
Stokes, G. M., & Schwartz, S. E. (1994). The Atmospheric Radiation Measurement (ARM)
Program: Programmatic background and design of the cloud and radiation test bed.
Bulletin of the American Meteorological Society, 75, 12011221.
Szczodrak, M., Minnett, P. J., Nalli, N. R., &Feltz, W. F. (2007). Proling the lower troposphere
over the ocean with infrared hyperspectral measurements of the Marine-Atmosphere
Emitted Radiance Interferometer. Journal of Oceanic and Atmospheric Technology, 24,
390402.
Vzquez-Cuervo, J., Armstrong, E. M., &Harris, A. (2004). The effect of aerosols and clouds on
the retrieval of infrared sea surface temperatures. Journal of Climate, 17, 39213933.
Walton, C. C. (1988). Nonlinear multichannel algorithms for estimating sea surface tem-
perature with AVHRR satellite data. Journal of Applied Meteorology, 27, 115124.
Xiong, X., Barnes, W. L., Guenther, B., & Murphy, R. E. (2003). Lessons learned from
MODIS. Advances in Space Research, 32, 21072112.
Zvody, A.M., Mutlow, C. T., & Llewellyn-Jones, D. T. (1995). A radiative transfer model for
sea-surface temperature retrieval for the along track scanning radiometer. Journal of
Geophysical Research, 100, 937952.
Zhang, C., &Pennington, J. (2004). African dry air outbreaks. Journal of Geophysical Research,
109, D20108.
465 M. Szczodrak et al. / Remote Sensing of Environment 140 (2014) 450465