near-surface air temperature retrieval derived from amsu-a and sea surface temperature
TRANSCRIPT
Near-Surface Air Temperature Retrieval Derived from AMSU-A andSea Surface Temperature Observations
DARREN L. JACKSON
Cooperative Institute for Research in Environmental Sciences, and NOAA/Earth System
Research Laboratory, Boulder, Colorado
GARY A. WICK
NOAA/Earth System Research Laboratory, Boulder, Colorado
(Manuscript received 29 October 2009, in final form 3 March 2010)
ABSTRACT
A 10-m air temperature (Ta) retrieval using Advanced Microwave Sounding Unit A (AMSU-A) and
satellite-derived sea surface temperature (Ts) observations is presented. The multivariable linear regression
retrieval uses AMSU-A brightness temperatures from the 52.8- and 53.6-GHz channels and satellite-derived
daily sea surface temperatures to determine Ta. A regression error of 0.838C using 841 matched satellite and
ship observations demonstrates a high-quality fit of the satellite observations with in situ Ta. Validation of the
retrieval using independent International Comprehensive Ocean–Atmosphere Dataset (ICOADS) ship and
buoy observations results in a bias of 20.218C and root-mean-square (RMS) differences of 1.558C. A com-
parison with previous satellite-based Ta retrievals indicates less bias and significantly smaller RMS differ-
ences for the new retrieval. Regional biases inherent to previous retrievals are reduced in several oceanic
regions using the new Ta retrieval. Satellite-derived Ts–Ta data were found to agree well with ICOADS buoy
data and were significantly improved from previous retrievals.
1. Introduction
High-resolution and accurate estimates of turbulent
heat fluxes are considered essential for improving fore-
casts of numerical weather prediction models and global
climate models. Retrieving turbulent heat fluxes is chal-
lenging because of inaccuracies from parameterization
using bulk formulas and difficulties in acquiring accurate
input parameters for the bulk formulation. Accurate
meteorological inputs to the bulk formulas have been
difficult to acquire from satellite remote sensing methods
because of limitations in direct retrieval of near-surface
parameters, so these inputs have typically been acquired
from ship and buoy observations. However, unlike ship
and buoy observations, satellite observations can provide
global coverage on a daily time scale, thus providing the
potential for daily global heat flux products. Accurate
global observations of air temperature (Ta) are consid-
ered essential for more accurate retrieval of sensible heat
flux at the ocean surface (Curry et al. 2004). Therefore,
a new satellite method to retrieve Ta is presented in this
paper.
High-quality Ta observations remain difficult to re-
trieve using satellites because of difficulties in resolving
the temperature layer near the ocean surface using exist-
ing satellite instrumentation. Therefore, indirect satellite
methods have generally been utilized to derive Ta ob-
servations. Liu (1988) used near-surface humidity re-
trievals to estimate Ta by assuming 80% relative humidity
over the ocean surface. Jourdan and Gautier (1995) used
satellite-derived precipitable water (PW) to estimate Ta
by deriving a relationship between monthly mean PW and
Comprehensive Ocean–Atmosphere Dataset (COADS)
Ta data. Konda et al. (1996) retrieved Ta by using satellite-
derived estimates of sea surface temperature (Ts), wind
speed, and specific humidity. Clayson and Curry (1996)
used Ts observations and a parameterization of Ta–Ts
based on a satellite-derived cloud classification tech-
nique. Jones et al. (1999) used a neural network approach
for retrieving Ta using monthly mean satellite-derived
Corresponding author address: Darren L. Jackson, Cooperative
Institute for Research in Environmental Sciences, University of
Colorado/NOAA/Earth System Research Laboratory, 325
Broadway R/PSD2, Boulder, CO 80305–3337.
E-mail: [email protected]
OCTOBER 2010 N O T E S A N D C O R R E S P O N D E N C E 1769
DOI: 10.1175/2010JTECHA1414.1
� 2010 American Meteorological Society
PW and Ts data to train the retrieval. Jackson et al.
(2006) introduced a multi-instrument approach that used
Advanced Microwave Sounding Unit A (AMSU-A)
and Special Sensor Microwave Imager (SSM/I) micro-
wave brightness temperatures directly to determine Ta.
Roberts et al. (2010) recently developed a neural net-
work approach that uses SSM/I radiances and Ts ob-
servations to retrieve several surface parameters including
Ta. Only Jackson et al. (2006) and Roberts et al. (2010)
use satellite brightness temperatures directly to retrieve
Ta, and these studies provide methods that can increase
the spatial and temporal resolution of the satellite Ta
products and reduce the retrieval error that has plagued
earlier studies.
Jackson et al. (2006) determined that a multisensor sat-
ellite approach could achieve more accurate satellite-
derived Ta observations by combining AMSU-A and SSM/I
satellite observations. This resulted in root-mean-square
(RMS) differences with independent validation data of
1.968C, which was a 1.08C less than previously published
algorithms. The AMSU-A 52.8-GHz channel was found
to significantly improve the Ta retrieval because this
temperature-sensing channel has a weighting function
that peaks in the lower troposphere near the 900-mb
level. However, a significant cold bias for Ta , 158C and
an RMS difference near 28C still did not meet the re-
quirement of a Ta accuracy of 0.968C (Jackson et al.
2006) needed to achieve a sensible heat flux accuracy of
10 W m22. The goal of this study is to develop a satellite
Ta retrieval that further reduces these Ta retrieval
errors.
2. Retrieval and validation
Recognizing the generally close relationship between
Ts and Ta resulted in consideration of using Ts as a
predictor for the Ta regression retrieval. Therefore, the
satellite-derived Ts data described in Reynolds et al.
(2007) were used as input to the regression model to de-
termine the regression error impact on the Ta retrieval.
The version of Reynolds Ts data used in this study only
used satellite data from the Advanced Very High Reso-
lution Radiometer (AVHRR) satellite. AMSU-A data
were taken from the National Oceanic and Atmospheric
Administration’s (NOAA) NOAA-15 satellite in this
study to develop the retrieval; however, other satellites
flying AMSU-A instruments could be used to retrieve
Ta. The input training data were taken from Jackson
et al. (2009), which used high-quality research vessel
observations primarily from NOAA’s Earth System Re-
search Laboratory (ESRL). Collocation requirements
between the ship and satellite observations were 3 h and
50 km. Ship Ta observations were adjusted to 10-m
height using the Coupled Ocean–Atmospheric Response
Experiment (COARE) version 3 bulk flux model (Fairall
et al. 2003). The final regression equation used to predict
Ta at 10 m is
Ta 5�67.8532 1 0.801 546Ts 1 0.443 879Tb52.8
�0.170 060Tb53.6
, (1)
where Ts is the sea surface temperature in degrees Cel-
sius, and Tb52.8 and Tb53.6 are AMSU-A brightness
temperatures in kelvins at 52.8 and 53.6 GHz, re-
spectively. The 53.6-GHz channel was also included
because it has some sensitivity to lower tropospheric
temperature even though its weighting function peaks in
the middle troposphere, and it potentially removes tem-
perature variability detected above the lower tropo-
sphere. The regression RMS difference was 0.838C
using 841 satellite matches to ship observations. Jackson
et al. (2006) included SSM/I observations in their Ta
algorithm (hereafter called AMMI), but these data did
not improve the regression when using both AMSU-A
and Ts data as input to the regression. A benefit of not
including SSM/I is the potential for increased temporal
and spatial Ta coverage because some of the AMSU-A
instruments on other satellites do not match well with
SSM/I observations since they do not have overpasses
with SSM/I that meet the requirement for 3-h time co-
incidence. Not including Ts with Tb52.8 and Tb53.6 re-
sulted in a regression RMS difference of 1.938C, thus
demonstrating a greater than 18C improvement in re-
gression error from including Ts.
Validation of the Ts and AMSU-A retrieval for Ta
(hereafter referred to as AMSST) was carried out using
independent Ta observations from the International
COADS (ICOADS) International Maritime Meteoro-
logical Archive (IMMA) version 2.4 dataset (Worley
et al. 2005). The ICOADS data are primarily a combi-
nation of observations from voluntary observing ships
(VOS) and buoys. These data have been used in previous
studies for satellite validation of 10-m specific humidity
(Qa) (i.e., Bentamy et al. 2003; Jackson et al. 2009) and
heat flux (DaSilva et al. 1994; Zeng et al. 2009) data.
ICOADS air temperature observations are taken at
various heights from the ocean surface with ICOADS
ship observations generally measured at levels exceed-
ing 10 m and buoy observations typically measured at
3 m. This height difference could result in a bias be-
tween buoy and ship observations and a bias between
ICOADS and satellite 10-m Ta observations. However,
difficulties in adjusting the ICOADS observations to
10 m are due to a lack of sensor height information for
many of the observations. This issue is discussed in greater
1770 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 27
detail in Jackson et al. (2009). For the current study, 645
coincident nighttime ICOADS buoy and ship air tem-
perature observations for 1999 were found to have an
absolute difference of 0.298C. This difference can be
considered the uncertainty in the Ta uncertainty due to
measurement height differences of the ICOADS data.
The retrieval validation and product comparison were
conducted with datasets using SSM/I observations, so the
sampling of the AMSST Ta data for this validation was
restricted to locations where AMSU and SSM/I obser-
vations were coincident within these grid cells. A land
mask was applied to all satellite observations since the
highly variable and large land surface emissivity is not
accounted for in this retrieval. Furthermore, rain-rate
observations exceeding 0 mm h21 and cloud liquid wa-
ter (CLW) observations exceeding 0.1 mm retrieved from
the SSM/I observations were excluded in the comparison
since rainfall and CLW can contaminate SSM/I-derived
products used to derive Ta retrieval in methods used to
compare with the AMSST Ta retrieval.
Validation of the Ta retrieval used the same ICOADS
observations and technique described in Jackson et al.
(2009). Both satellite and ICOADS observations were
averaged to 0.58 spatial and 3-h temporal resolution
maps. Matches between satellite and in situ observations
were determined from these grid maps. Figure 1 gives
the validation results for both the AMSST and the
AMMI Ta retrieval using matched satellite/ICOADS
data for 49 167 observations in 1999. The AMSST Ta
shows a bias of 20.218C and RMS difference of 1.558C,
which are both significantly smaller than the AMMI Ta
retrieval. AMSST Ta is slightly lower than ICOADS for
most ICOADS Ta bins but generally warmer than the
AMMI Ta retrieval. AMSST Ta is warmer than ICOADS
for Ta , 08C, whereas the AMMI Ta is less biased but has
large RMS differences. At this low Ta range, greater
scatter for both satellite brightness temperatures and Ts
observations occur in the training dataset, and satellite
brightness temperatures show generally less sensitivity
to Ta fluctuations. This problem has also been shown
with Ta values derived from Roberts et al. (2010). AMSST
and AMMI retrievals are both colder than ICOADS for
Ta exceeding 308C. A small number of observations in the
training dataset for this Ta range may contribute to this
validation bias. Bias and RMS differences for other re-
trievals such as Jackson et al. (2006), Liu (1988), Jourdan
and Gautier (1995), and Konda et al. (1996) are provided
in Table 1. These retrievals have biases exceeding 18C
and RMS differences exceeding 48C when using all of the
matched data from 1999.
Local solar heating of daytime ICOADS Ta sensors
(Kent et al. 1993) may explain part of the bias between
ICOADS and AMSST Ta. Table 1 shows that when the
validation results are separated by day and night matches
for AMSST, the bias for day-only observations is larger
than the night-only observations and is consistent with
FIG. 1. Scatter diagram comparing AMSST and AMMI Ta re-
trievals with ICOADS Ta for 1999. Error bars represent one stan-
dard deviation. Bins are offset between AMSST and AMMI Ta data.
TABLE 1. Ta validation statistics using ICOADS for 1999. Bias is defined as satellite Ta 2 ICOADS Ta.
AMSST Jackson et al. (2006) (AMMI) Liu (1988) Jourdan and Gautier (1995) Konda et al. (1996)
Obs. No. Bias RMS Bias RMS Bias RMS Bias RMS Bias RMS
Day 1
night
All 49167 20.21 1.55 21.33 2.64 21.60 4.46 21.28 4.12 22.52 6.40
Ship 34677 20.25 1.67 21.48 2.81 21.86 4.82 21.52 4.44 22.92 6.87
Buoy 14400 20.08 1.24 20.96 2.15 20.95 3.44 20.69 3.18 21.53 5.06
Day
only
All 31620 20.40 1.57 21.21 2.58 21.24 4.36 20.98 4.07 21.95 6.03
Ship 21029 20.49 1.75 21.36 2.80 21.51 4.80 21.22 4.48 22.36 6.18
Buoy 10496 20.21 1.11 20.88 2.06 20.68 3.25 20.47 3.06 21.09 4.59
Night
only
All 22885 0.07 1.42 21.37 2.55 21.82 4.29 21.46 3.91 22.90 6.44
Ship 13910 0.12 1.57 21.65 2.83 22.39 4.83 21.97 4.38 23.77 7.25
Buoy 9124 0.01 1.15 20.91 2.03 20.92 3.25 20.66 3.02 21.52 4.92
OCTOBER 2010 N O T E S A N D C O R R E S P O N D E N C E 1771
possible solar heating of the sensors. Interestingly, the
other retrievals show less bias for day-only than the night-
only matches with ICOADS. Even though the average
ICOADS Ta does increase 0.888C from night-only to day-
only observations, all satellite retrievals except for the
AMSST retrieval respond with a greater increase in Ta
from night to day. Diurnal changes in SST are not con-
sidered a factor since the average daytime increase in
ICOADS Ts was only 0.098C. The bias for buoy-only
observations for all of the retrievals gives less bias than
ship-only observations. Nighttime observations from the
buoys are likely to be the most accurate observations, and
the AMSST buoy night-only data give the lowest bias
of 0.018C. Generally, the AMSST Ta compares most
favorably with ICOADS by having the lowest bias
and RMS differences for all matched conditions with
ICOADS.
Jackson et al. (2009) demonstrated that Qa bias can be
affected by large-scale subsidence-induced temperature
inversions in the North Pacific during the summer months;
therefore, regional bias of the AMSST Ta was examined
and compared to previous retrievals. The geographical
distribution of Ta bias is given in Fig. 2 for AMSST and
AMMI Ta. Bias is computed using ICOADS Ta data
over a 1-yr period during 1999 and binned onto a 0.58
spatial resolution grid map. Bias with the AMMI Ta
retrieval in Fig. 2a shows satellite retrieval under-
estimation of Ta in several regions such as the Medi-
terranean Sea, the east coast of Australia, the Caribbean
Sea, and the northeastern Atlantic Ocean. This pattern
of bias is also true for Ta retrievals of Liu (1988), Jourdan
and Gautier (1995), and Konda et al. (1996) (not shown).
Figure 2b shows that the AMSST Ta retrieval gives
generally less bias in these regions. Other regions where
AMMI overestimates Ta such as the western coastlines of
North and South America also show less bias for AMSST.
The mean bias for AMSST Ta averaged for all map lo-
cations is 20.218C and for AMMI Ta is 21.338C. This
shows significant reduction in bias using the AMSST Ta
retrieval. Despite these improvements, there are some
regions where AMSST has significant positive bias such
as regions in the Atlantic along the east coast of North
FIG. 2. Ta bias maps for 1999 of (a) AMMI 2 ICOADS Ta and (b) AMSST 2 ICOADS Ta.
1772 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 27
America and in the northern Pacific near Japan. The
warm Gulf Stream and the Kuroshio likely cause over-
estimation of Ta, particularly when cold air outbreaks
occur over these currents. Overall, Figs. 2a and 2b show
that the regional bias relative to ICOADS has been sig-
nificantly reduced by using Ts in the AMSST Ta retrieval.
3. Discussion
Differences between Ts and Ta drive the sensible heat
flux exchange between the ocean and atmosphere, so the
relationship of Ta with Ts is examined for the satellite
and ICOADS data. One important question is whether
the use of Ts in the AMSST Ta retrieval diminishes the
utility for using these Ta observations for sensible heat
flux retrievals. A Ta retrieval was developed only using
a linear fit to the Ts data, hereafter referred to as SST Ta,
to help examine the spatial and temporal affects of
combining AMSU-A and Ts observations. The question
is whether the temporal and spatial structure of the
AMSST Ta field is dominated by Ts. Figure 3 shows the
Ta difference between AMSST and SST Ta retrievals for
January and July of 1999. The largest differences occur
over the North Pacific and Atlantic Oceans. Cold air
outbreaks from Asia result in colder air overlying a rel-
atively warm ocean. Negative Ta differences around
Japan, over the Mediterranean Sea, and over the Gulf
Stream east of North America indicate that retrievals
using the AMSU-A satellite observations result in colder
Ta observations. Conversely, July Ta differences show
values to be warmer than what would be predicted using
a retrieval based solely on Ts. The structure of this Ta
difference is also evident over Southern Hemisphere and
shows seasonal variability. Therefore, the AMSST Ta
retrieval does show structure that differs from Ts.
The relationship of the Ts–Ta difference as a function
of Ts is compared between satellite and ICOADS ob-
servations. Figure 4 gives a binned scatter diagram show-
ing the Ts–Ta relationship for both satellite (Reynolds
Ts 2 AMSST Ta) and ICOADS Ts–Ta as a function of Ts
FIG. 3. Maps of AMSST Ta 2 SST Ta for 1999 for (a) January and (b) July.
OCTOBER 2010 N O T E S A N D C O R R E S P O N D E N C E 1773
for night-only matched observations. The matched satel-
lite and ICOADS observations are filtered to only include
data where jReynolds Ts 2 ICOADS Tsj , 0.258C so to
isolate the differences between the satellite and ICOADS
data caused by differences in the Ta data. The Ts–Ta
relationship is shown to have the same mean difference
for most values of Ts, and Ts is approximately 1.308C
warmer than Ta on average for both products. While the
mean Ts–Ta relationship for satellite and ICOADS is
very close, the variability in the Ts–Ta differences is
smaller for the satellite data. Standard deviations for the
ICOADS data are 1.768C whereas the satellite data have
standard deviations of less than half that value. The
satellite retrieval is less likely to capture large Ts–Ta
deviations for several reasons. Since the AMSST Ta
retrieval uses Ts as a predictor for the retrieval, the re-
trieval will have a tendency to converge toward Ts. The
satellite observations also measure radiance for broad
layers of the atmosphere, so shallow layers of cold or
warm air are difficult to detect. Also the footprint of the
satellite data is much larger than the point measurement
of the ship, so the satellite measurements smooth out
temperature gradients that could cause significant de-
partures in Ts–Ta.
Figure 5 gives histograms of the Ts–Ta relationship
for the ICOADS data and for the satellite products
(Reynolds Ts and AMSST Ta) for all available ship and
buoy matches during 1999. The matches were filtered
the same as data in Fig. 4 so that only Ts differences
between ICOADS and satellite that are less than 0.258C
are included. This restriction more significantly reduced
the number of ship observations by 79% compared to
the 51% reduction in buoy observations. The satellite
histograms are shifted right of the ICOADS histograms
and show a more narrow distribution. The best agree-
ment between satellite and ICOADS occurs for the
buoy histograms where the peak values occur at the
same Ts–Ta bin. Most of the large positive and negative
ICOADS Ts–Ta values that contribute to the broader
ICOADS distribution originate from the ship observa-
tions. Potentially larger errors in the ship-only Ts and Ta
data contribute to part of this wider distribution since
VOS observations are generally considered to have larger
errors than buoy observations. Differences in the geo-
graphical distribution in sampling of the ship-only and
buoy-only observations may also contribute these dif-
ferences. Buoy observations are concentrated in the
tropical Pacific and coastal regions of North America
and Europe, whereas the ship data are concentrated
along shipping routes in the North Pacific and Atlantic
Oceans.
A histogram comparison of Ts–Ta for buoy-only ob-
servations for the current and previously published re-
trievals is shown in Fig. 6. The previous retrievals all
show a broad distribution with significantly more large
negative and positive values than are shown from the
FIG. 4. A comparison of 1999 night-only data between Reynolds
Ts–AMSST Ta and ICOADS Ts 2 ICOADS Ta data plotted as
function of Ts using 28C bins. Squares and dotted curve represent
mean and one standard deviation for satellite Ts–Ta, and triangle
and dotted curves represent mean and one standard deviation for
ICOADS Ts–Ta. Satellite and ship matches were filtered such that
jReynolds Ts 2 ICOADS Tsj , 0.258C.
FIG. 5. Ts–Ta histogram data for 1999 using satellite data
(Reynolds Ts 2 AMSST Ta) and ICOADS data for all matched
observations, ship-only observations, and buoy-only observations.
Histogram bins have a width of 0.58C. Matches were filtered as in
Fig. 4.
1774 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 27
ICOADS buoy data. The AMMI distribution does have
a more narrow distribution relative to the other satellite
products and a reduced number of large negative values,
but the AMMI distribution peaks at much warmer Ts–Ta
value than the buoy distribution. The AMSST distri-
bution has a similar width to the buoys and the peak
agrees with the buoys, but this distribution does show
fewer negative Ts–Ta observations and warmer Ts–Ta
values than the buoys and is slightly shifted to right of
the ICOADS buoy distribution. It should be noted that
Roberts et al. (2010) (not shown here) has also shown
a Ts–Ta histogram that compares well with in situ data.
They also used SST as a predictor; however, they uti-
lized a neural network approach that uses SSM/I rather
than AMSU-A satellite observations. Therefore, even
though Ts–Ta using AMSST has a more narrow distri-
bution relative to other satellite products, it does agree
better with Ts–Ta buoy observations.
4. Conclusions
These results indicate that combining AMSU-A ob-
servations with Ts daily mean observations provides a
useful method for the retrieval of 10-m air temperature
observations. Regression RMS differences of 0.838C are
significantly less than in previously published methods.
Validation with ICOADS data indicates an overall bias
of 20.218C and RMS difference of 1.558C, which was
shown to be significantly better than previous satellite
Ta retrievals. Regional biases were found to be smaller
for the AMSST retrieval than previous methods for
several oceanic regions. Examination of the Ts–Ta re-
lationship, which is important for computation of sen-
sible heat flux using the bulk formula, indicates that the
mean satellite Ts–Ta observations agree well with
ICOADS Ts–Ta observations for most observations.
Reduced variability for the AMSST Ta retrieval does
hinder detection of the largest Ts–Ta differences, likely
because the Ta algorithm uses Ts as a predictor in its
retrieval. The Ts–Ta histogram distributions for satellite
and ICOADS are similar with satellite Ts–Ta slightly
warmer than ICOADS, and the ICOADS distributions
were generally broader particularly for the ship obser-
vations. However, the Ts–Ta relationship using the
AMSST Ta retrieval, when compared with previous
retrievals, did show a substantial improvement when
compared with the Ts–Ta distribution from buoy ob-
servations in ICOADS. The AMSST Ta retrieval pro-
vides a relatively simple global method for observing
near-surface temperature and provides a potentially use-
ful dataset for computing global fields of sensible heat flux
over the oceans of the earth.
Acknowledgments. This project supported in part by
funding from NASA Grant NNM04AA67I in collabo-
ration with Dr. Franklin R. Robertson at the Marshall
Space Flight Center. The ICOADS data are from the
Research Data Archive (RDA), which is maintained by
the Computational and Information Systems Labora-
tory (CISL) at the National Center for Atmospheric
Research (NCAR). NCAR is sponsored by the National
Science Foundation (NSF). The original data are available
from the RDA (http://dss.ucar.edu) in dataset ds540.1.
REFERENCES
Bentamy, A., K. B. Katsaros, A. M. Mestas-Nunoz, W. M. Drennan,
E. B. Forde, and H. Roquet, 2003: Satellite estimates of wind
speed and latent heat flux over the global oceans. J. Climate, 16,637–656.
Clayson, C. A., and J. A. Curry, 1996: Determination of surface
turbulent fluxes for the Tropical Ocean–Global Atmospheric
Coupled Ocean–Atmosphere Response Experiment: Com-
parison of satellite retrievals and in situ measurements.
J. Geophys. Res., 101, 28 515–28 528.
Curry, J. A., and Coauthors, 2004: SEAFLUX. Bull. Amer. Meteor.
Soc., 85, 409–424.
DaSilva, A. M., C. C. Young, and S. Levitus, 1994: Algorithms and
Procedures. Vol. 1, Atlas of Surface Marine Data 1994, NOAA
Atlas NESDIS 6, 83 pp.
Fairall, C. W., E. F. Bradley, J. E. Hare, A. A. Grachev, and
J. B. Edson, 2003: Bulk parameterization of air–sea fluxes:
Updates and verification for the COARE algorithm. J. Cli-
mate, 16, 571–591.
Jackson, D. L., G. A. Wick, and J. J. Bates, 2006: Near-surface re-
trieval of air temperature and specific humidity using multisensor
FIG. 6. Ts–Ta histogram data for 1999 comparing satellite data
(Reynolds Ts 2 satellite Ta) of previous retrievals and AMSST
with ICOADS Ts–Ta observations. The matches were filtered as in
Fig. 4 and only buoy observations are included for these data.
OCTOBER 2010 N O T E S A N D C O R R E S P O N D E N C E 1775
microwave satellite observations. J. Geophys. Res., 111, D10306,
doi:10.1029/2005JD006431.
——, ——, and F. R. Robinson, 2009: Improved multisensor ap-
proach to satellite-retrieved near-surface specific humidity
observations. J. Geophys. Res., 114, D16303, doi:10.1029/
2008JD011341.
Jones, C., P. Peterson, and C. Gautier, 1999: A new method for deriving
ocean surface specific humidity and air temperature: An artificial
neural network approach. J. Appl. Meteor., 38, 1229–1245.
Jourdan, D., and C. Gautier, 1995: Comparison between global latent
heat flux computed from multisensor (SSM/I and AVHRR) and
from in situ data. J. Atmos. Oceanic Technol., 12, 46–72.
Kent, E. C., R. J. Tiddy, and P. K. Taylor, 1993: Correction of
marine air temperature observations for solar radiation ef-
fects. J. Atmos. Oceanic Technol., 10, 900–906.
Konda, M., N. Imasato, and A. Shibata, 1996: A new method to
determine near-sea surface air temperature by using satellite
data. J. Geophys. Res., 101, 14 349–14 360.
Liu, W. T., 1988: Moisture and latent heat flux variabilities in the
tropical Pacific derived from satellite data. J. Geophys. Res.,
93, 6749–6760.
Reynolds, R. W., T. M. Smith, C. Liu, D. B. Chelton, K. S. Casey,
and M. G. Schlax, 2007: Daily high-resolution-blended
analyses for sea surface temperature. J. Climate, 20, 5473–
5496.
Roberts, B., C. A. Clayson, F. R. Robertson, and D. L. Jackson,
2010: Predicting near-surface characteristics from SSM/I using
neural networks with a first guess approach. J. Geophys. Res.,
in press.
Worley, S. J., S. D. Woodruff, R. W. Reynolds, S. J. Lubker, and
N. Lott, 2005: ICOADS Release 2.1 data and products. Int.
J. Climatol., 25, 823–842.
Zeng, L., P. Shi, W. T. Liu, and D. Wang, 2009: Evaluation of
a satellite-derived latent heat flux product in the South China
Sea: A comparison with moored buoy data and various
products. Atmos. Res., 94, 91–105.
1776 J O U R N A L O F A T M O S P H E R I C A N D O C E A N I C T E C H N O L O G Y VOLUME 27