Sea-surface temperature measured by the Moderate Resolution Imaging

Spectroradiometer (MODIS).
P.J. Minnett, R.H. Evans, E.J. Kearns and O.B. Brown.

Rosenstiel School of Marine & Atmospheric Science

University of Miami
4600 Rickenbacker Causeway, Miami, Florida, USA, 33149.

Abstract-The Moderate Resolution Imaging Spectroradiometer Here we introduce the SST measuring capabilities of the
(MODIS) on the NASA Earth Observing System satellites Terra and MODIS and present an initial determination of the accuracy
Aqua include infrared bands that are designed for the accurate with which these measurements are being made. Given an
measurement of sea-surface temperature (SST). In addition to the accurately-calibrated spacecraft infrared radiometer, the
conventional ‘split-window’ bands in the 10-12 µm atmospheric
uncertainties in the SST retrieval are determined largely by
window (similar in spectral properties to the corresponding channels
of the heritage instrument, the Advanced Very High Resolution inadequacies in the algorithm used for the correction of the
Radiometer (AVHRR)) the MODIS also has three bands in the 3.5 effects of the intervening atmosphere. The accuracy of the
to 4.2 µm interval for SST measurement. Here we present a MODIS SSTs is established by comparison with collocated,
validation of the MODIS SSTs, derived from the conventional independent measurements from ship-borne infrared spectro-
measurements at the 10-12 µm window, by comparison with radiometers and a large number of drifting buoys.
radiometrically-derived skin temperatures from ships, and with bulk
measurements from buoys. These show the MODIS SSTs are II. MODIS
comparable in accuracy to the AVHRR Pathfinder SST fields.
The MODIS is a 36-band sensor that spans the spectral
range from 0.4 to 14.4µm, i.e. through the visible into the
I. INTRODUCTION thermal infrared. These bands were selected to satisfy the
needs of a large user community in many fields of
The temperature of the ocean surface is an important
environmental science. The oceanographic applications of
variable for the study of the earth’s climate system, for
MODIS data include measurements of ocean color and SST
weather forecasting, and for oceanographic research. Patterns
[1]. The bands used for SST measurement are in the
of sea-surface temperature (SST) reveal the complexity of the
atmospheric ‘windows’ at wavelengths of 3.5 to 4.2 µm and
ocean surface currents, and large scale SST anomalies in
critical oceanic areas presage climate perturbations, such as 10 to 12 µm. There are two spectral bands (numbered 31 and
El-Niño events. The SST is a controlling variable in the 32) in the longer wavelength window, which correspond
coupling of the ocean and atmosphere in terms of the fluxes closely to the corresponding AVHRR channels (AVHRR
of heat, moisture, momentum and greenhouse gasses, such as channels 4 and 5), but there are three in the mid-infrared
CO2. Satellite remote sensing has provided a mechanism for window (bands 20, 22 and 23) where AVHRR has just one
deriving global SST fields by measuring the infrared and (AVHRR channel 3). These bands have noise equivalent
microwave thermal emission from the sea-surface. Infrared temperature differences (NE∆Ts) of ~0.05K, and a spatial
radiometry provides high spatial resolution (~1 km2), but resolution of ~1km2 at nadir. The swath width is 2330 km,
suffers from the fact that clouds obscure view of the sea- which results in small regions near the equator not being
surface, while microwave radiometry provides data largely sampled on adjacent orbits; these are covered on the next day.
independent of cloud effects, but at a much coarser spatial In-flight calibration of the infrared bands is achieved using a
resolution (~625 km2). Infrared radiometers, specifically measurements of cold space and of a well-characterized,
successive models of the Advanced Very High Resolution internal black-body reference target, the temperature of which
Radiometer (AVHRR) flown on the NOAA series of polar is carefully monitored. In addition to the multiple spectral
orbiting satellites, have provided SST measurements for over bands in the mid-infrared window, the technological
two decades. With the launch of the Earth Observing System innovations of MODIS include a bank of 10 detectors for
satellite Terra in December 1999, a new generation of each spectral band, a large two-sided ‘paddle-wheel’ scene
imaging radiometer, the MODIS (Moderate Resolution mirror to scan the swath on the earth’s surface, and 12-bit
Imaging Spectroradiometer) was introduced which combines digitization for all data. The on-board data recording
a more extensive suite of spectral measurements with capabilities allow all of the data from all detectors to be
improvements in instrument technology, and with which the stored and telemetered to ground receiving stations, offering
time-series of infrared measurements of SST can be extended full 1km2 data over the entire oceans. Some of the
into the future. radiometric resolution in Bands 31 and 32 offered by the 12-

0-7803-7536-X/$17.00 (C) 2002 IEEE 1177

bit digitizer has been lost by expanding the dynamic range of IV. VALIDATION
these bands to permit wild-fire monitoring (a situation largely
remedied in the MODIS to be flown on Aqua). Additional An initial determination of the accuracy of the MODIS
data quality problems have resulted from the use of multiple SSTs has been made by comparison with collocated,
detectors and a two-sided scene mirror used in such a way independent measurements of SST made by Fourier-
that the angle of incidence changes across the scan. This was Transform infrared interferometric spectroradiometers, M-
exacerbated in the infrared at wavelengths longer than ~8µm AERIs (Marine – Atmospheric Emitted Radiance Interfero-
by the use of a multi-layer coating on the mirror which causes meters [7]) mounted on ships, and by conventional in situ
a marked change in the mirror reflectivity across the scan. thermometers mounted on drifting buoys.
Corrections for these effects have been devised successfully
and implemented. A. M-AERI skin temperatures.
Here we focus on the SSTs derived from measurements of
Bands 31 and 32 as these correspond closely to the data from The M-AERI is a robust, accurate, self-calibrating, sea-
AVHRR. going Fourier-transform infrared spectroradiometer that is
deployed on marine platforms to measure the emission
III. ATMOSPHERIC CORRECTION ALGORITHM spectra from the sea surface and marine atmosphere. The
instrument works continuously under computer control with a
Building on the experience of deriving SST from AVHRR, high rate of data return. Spectral measurements are made in
the algorithm used to correct for the atmosphere is the same the range of ~3 to ~18 µm wavelength, and are calibrated
as that used for the Pathfinder SST product [2] which is based using two internal, NIST-traceable blackbody cavities. The
on the Non-Linear SST algorithm [3]. This takes the form environmental variables derived from the spectra include the
surface skin temperature of the ocean, surface emissivity,
SST = a + bT31 + c(T31-T32)Tsfc+ d(T31-T32)(sec(θ)-1) (1) near-surface air temperature and profiles of temperature and
humidity through the lower troposphere. The absolute
where T31 and T32 are the brightness temperatures measured uncertainty of the derived skin SSTs is <0.1K.
in Bands 31 and 32, θ is the satellite zenith angle measured at Comparisons have been made between the MODIS SSTs
the sea surface, and Tsfc is a ‘first guess’ SST value, taken derived with empirical coefficients and M-AERI data from
from the optimally interpolated SST weekly National Centers four cruises, in the Mediterranean Sea (R/V Urania, April
for Environmental Prediction (NCEP) SST analysis of global 2000), the Pacific Ocean (USCGC Polar Sea, March – April
buoy, ship, and AVHRR SST data [4]. The numerical values 2001; and NOAA S Ronald H. Brown, March – April 2001),
of the coefficients a, b, c and d were derived in two ways: and the Eastern Caribbean (Explorer of the Seas[8], April
empirically by regression against SST values derived from 2001). The mean error (MODIS – M-AERI) is 0.20K and the
AVHRR and theoretically by numerical simulations of the scatter has a standard deviation of 0.26K (n=242). These
infrared radiative transfer through the atmosphere. results are shown in Figure 1.
The empirical coefficient set was derived by identifying
areas of the ocean that were cloud-free in both AVHRR and
MODIS images on the same day (i.e. separated by about three
hours), and performing a regression analysis between the
MODIS brightness temperatures and the AVHRR SSTs.
Independent validation of the AVHRR SSTs shows these to
have small mean error (~0.1K) and small scatter (~0.3K) [5].
This approach leads to MODIS SST fields that should be
directly compatible with the extant AVHRR Pathfinder SSTs.
The theoretically-derived coefficients were obtained by
regression analysis of simulated MODIS brightness
temperatures predicted by a line-by-line radiative transfer
model [6]. The range of atmospheric variability is defined in
these simulations by a large set of marine atmospheres
derived from the data assimilation model of the European
Centre for Medium-range Weather Forecasting (ECMWF).
This approach should result in MODIS SSTs that are based Fig. 1. MODIS SSTs errors determined by comparison with
on our understanding of the physics of the measurement. M-AERI skin SSTs. The colors indicate different cruises:
blue – R/V Urania; red – USCGC Polar Sea; pink - NOAA S
Ronald H. Brown; green – Explorer of the Seas.

1178
B. Buoy bulk temperatures Captains, officers and crews of the R/V Urania, USCGC
The comparison between MODIS SSTs, derived with both Polar Sea, NOAA S Ronald H. Brown and the Explorer of the
the empirical and theoretical coefficients, and the bulk SSTs Seas. Funding from NASA is acknowledged.
from drifting buoys is shown in Figure 2. The statistics of the

REFERENCES

[1] Esaias, W.E., M.R. Abbott, I. Barton, O.B. Brown, J.W.

Campbell, K.L. Carder, D.K. Clark, R.H. Evans, F.E.
Hoge, H.R. Gordon, W.M. Balch, R.Letelier, and P.J.
Minnett., "An Overview of MODIS Capabilities for
Ocean Science Observations.," IEEE Transactions on
Geoscience and Remote Sensing, vol. 36, p. 1250-1265,
1998.
[2] Kilpatrick, K.A., G.P. Podesta, and R.H. Evans,
"Overview of the NOAA/NASA Pathfinder algorithm for
Sea Surface Temperature and associated Matchup
Database.," J. Geophys Research, vol. 106, p. 9179-9198,
2001.
[3] Walton, C.C., W.G. Pichel, J.F. Sapper, and D.A. May,
"The development and operational application of
Fig 2. MODIS SSTs errors determined by comparison with nonlinear algorithms for the measurement of sea surface
buoy bulk SSTs. temperatures with the NOAA polar-orbiting
environmental satellites," Journal of Geophysical
errors (as mean ± standard deviation) are: Research, vol. 103, p. 27,999-28,012, 1998.
a) empirical coefficients: -0.070 ± 0.433K [4] Reynolds, R.W. and T.M. Smith, "Improved global sea
b) theoretical coefficients: -0.239 ± 0.534K surface temperature analysis using optimum
There are 174 matchups in this analysis. interpolation," J. Climate, vol. 7, p. 929-948, 1994.
[5] Kearns, E.J., J.A. Hanafin, R.H. Evans, P.J. Minnett, and
V. DISCUSSION O.B. Brown, "An independent assessment of Pathfinder
AVHRR sea surface temperature accuracy using the
The initial validation of Terra MODIS SSTs reported here
Marine-Atmosphere Emitted Radiance Interferometer (M-
is only for the temperatures derived from Bands 31 and 32;
AERI)." Bulletin of the American Meteorological
validation of the mid-infrared SSTs will follow. The initial
Society., vol. 81(7), p. 1525-1536, 2000.
results show that MODIS SSTs are comparable to the best
[6] Závody, A.M., C.T. Mutlow, and D.T. Llewellyn-Jones,
SSTs produced by AVHRR [5], and the improved accuracy
"A radiative transfer model for sea-surface temperature
determined by M-AERI validation, compared to buoy
retrieval for the Along Track Scanning Radiometer," J.
validation is also in line with the findings of Kearns et al. [5]
Geophys. Res., vol. 100(C1), p. 937-952, 1995.
and others [9]. The poorer performance of the theoretical
[7] Minnett, P.J., R.O. Knuteson, F.A. Best, B.J. Osborne,
coefficients is somewhat surprising, and probably indicates
J.A. Hanafin, and O.B. Brown, "The Marine-Atmospheric
unresolved sources of error that are self-compensating in the
Emitted Radiance Interferometer (M-AERI), a high-
empirical coefficient set. These results are initial and
accuracy, sea-going infrared spectroradiometer," Journal
represent the start of a continuing effort to validate MODIS
of Atmospheric and Oceanic Technology., vol. 18(6), p.
SSTs in a wide range of conditions, and the estimates of the
994-1013, 2001.
uncertainties will likely change as more data are analyzed.
[8] Prager, E., P.J. Minnett, B. Albrecht, H. Maring, P.
The second MODIS is scheduled for launch on Aqua in
Ortner, D. Wilson, S. Cummings, D. Palmer, and E.
April 2002, and a comparable validation exercise will be
Williams, "Explorer of the Seas: A Revolution in Marine
undertaken. It is anticipated that the SST accuracies will
Research and Education," EOS, Trans. Am. Geophys.
improve, particularly as the dynamic range of Bands 31 and
Union. In preparation, 2002.
32 have been reduced, thereby increasing the radiometric
[9] Donlon, C.J., P.J. Minnett, C. Gentemann, T.J.
resolution.
Nightingale, I.J. Barton, B. Ward, and J. Murray,
"Towards improved validation of satellite sea surface skin
ACKNOWLEDGEMENT
temperature measurements for climate research.," J.
Thanks are extended to colleagues at RSMAS: V. Banzon,
Climate, vol. 15, p. 353-369, 2002.
W. Barringer, J. Brown, J. Hanafin, E. Key, K. Kilpatrick, A.
Kumar, K. Maillet, G. Szczodrak, S. Walsh, and to the

1179