Retrieving ice concentration from SMOS

Abstract

The Soil Moisture and Ocean Salinity (SMOS) Ice project explored the potential of retrieving sea-ice information from the SMOS satellite, a polar-orbiting L-band radiometer successfully launched in November 2009. Toward this end, radiance measurements were collected over the Northern Baltic during the Pol-Ice campaign. We test a simple ice-concentration retrieval algorithm on these data and compare the results with ARTIST Sea Ice (ASI) maps derived from the Advanced Microwave Scanning Radiometer on the Earth Observing System. All operational ice-concentration algorithms are based on the same principle which, for the campaign data, reduces to a linear scaling of the radiances because, effectively, only one channel was available. Because of biases introduced by the different footprint sizes of the two radiometers (airborne and satellite), the linear flight path, and pilot selection of preferred surface type, Pol-Ice and ASI concentrations were compared using three different levels of averaging. In the first case, the individual measurements from the airborne radiometer were compared with interpolated ASI values; in the second, they were averaged over the pixels in the ASI maps; and in the third, they were averaged by binning the ASI values in 1% intervals. The correlations were 0.59, 0.67, and 0.76, respectively. Because of the unique operating principle of SMOS, each ground point will be viewed at multiple effective angles within a short time span. It is proposed to exploit this extra information by interpolating to a single effective viewing angle.

Full text

IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 6, NO. 1, JANUARY 2007 1
Retrieving ice concentration from SMOS
Peter Mills, Georg Heygster, Member, IEEE,
Abstract—The SMOS-Ice project explored the potential of
retrieving sea ice information from the Soil Moisture and Ocean
Salinity satellite, a polar-orbiting, L-band radiometer successfully
launched in November 2009. Towards this end, radiance mea-
surements were collected over the Northern Baltic during the
Pol-Ice campaign. We test a simple ice concentration retrieval
algorithm on this data and compare the results to ARTIST
Sea Ice (ASI) maps derived from the Advanced Microwave
Scanning Radiometer on EOS (AMSR-E). All operational ice
concentration algorithms are based on the same principle which,
for the campaign data, reduces to a linear scaling of the
radiances, because effectively only one channel was available.
Because of biases introduced by the differing footprint sizes of
the two radiometers (airborne and satellite), the linear flight
path and pilot selection of preferred surface type, Pol-Ice and
ASI concentrations were compared using three different levels of
averaging. In the first case, the individual measurements from the
airborne radiometer were compared to interpolated ASI values,
in the second, they were averaged over the pixels in the ASI
maps and in the third, they were averaged by binning the ASI
values in one percent intervals. Correlations were 0.59, 0.67 and
0.76 respectively. Because of the unique operating principle of
SMOS, each ground point will be viewed at multiple effective
angles within a short time-span. It is proposed to exploit this
extra information by interpolating to a single effective viewing
angle.
Index Terms—
I. INT ROD UC TI ON
The Soil Moisture and Ocean Salinity (SMOS) instrument
is a new satellite microwave radiometer launched in November
2009. It measures in the L-band range at 1.4 GHz with the
capability of rendering all four Stokes parameters. To produce
a reasonably small footprint size at such a low frequency while
keeping antenna weight down, an array of detectors fold out
upon deployment to produce an effective aperture equivalent
to the span of the array—i.e. much larger than the individual
detectors.
While the chief focus of the instrument, as its name implies,
is the ocean surface and dry land, it will still return useful
information when it is pointing at the cryosphere—the sea ice
and glacial ice pack. Even more so than existing microwave
radiometers, SMOS will have the advantage of negligible
atmospheric contribution to the signal providing true all-
weather performance (since radiation at such low frequencies
has too little energy to interact strongly with most non-
magnetic and non-conducting materials, especially one as thin
as the atmosphere.) As preparation for the SMOS project,
the Pol-Ice field campaign was conducted in the Northern
Baltic in March 2007. This comprised airborne measurements
P. Mills and G. Heygster are both with the Institute of Environmental
Physics, University of Bremen, Otto-Hahn-Allee 1, 28359 Bremen, Germany,
e-mail: pmills@iup.physik.uni-bremen.de.
Manuscript received July 21, 2010
of sea ice brightness temperature using the EMIRAD L-band
radiometer [1] and helicopter measurements of ice thickness
using the E-M Bird ice thickness detection instrument [2].
Since brightness temperatures over open water tend to be
quite distinct from those over ice, the most obvious retrievable
quantity from SMOS data is ice concentration, defined as the
fraction of ice relative to the total area. Here we test a simple
concentration retrieval algorithm on Pol-Ice campaign data and
compare the results with satellite-based retrievals.
II. RETRIEVING ICE CONCENTRATION
Most ice concentration algorithms are predicated on the
dual obervation that: 1. different surface types have different,
strongly clustered, radiometric signatures and 2. the final
radiometric signature at the instrument head is a linear com-
bination of that of the surface types found in the footprint,
with weighting factors taking on the values of the relative
concentrations. If we form a vector-space from the measure-
ments in which the signatures of the different surface types
are assumed invariant and all but one are linearly independent,
then it becomes a straightforward matter to derive the relative
ice concentrations [2], [3].
We could express this mathematically as follows:
Tb=Tb0+
n
∑
i=1
(Tbi −Tb0)Ci(1)
where Tbis the vector of brightness temperatures at the
instrument head, Tbi are the brightness temperatures of the
ith surface type, or tie-point, Ciare the relative concentra-
tions and Tb0are the brightness temperatures of the nominal
background surface type, i.e., of open water. The NASA team
algorithm uses a slight variation on this principle: the radiance
measurements are transformed by taking the difference of
two channels and dividing by their sum, producing a slightly
nonlinear retrieval [4], [5]. The influence of ice temperature
is thus mitigated since, all other things being equal, brightness
temperature varies roughly linearly with temperature (see
Equation (2)) and since sea ice brightness temperatures at
different microwave channels are strongly correlated [3].
III. MOD EL LI NG I CE B RI GH TN ESS TEMPERATUR E
We will use a simple radiative transfer model to simulate
the brightness temperature of sea ice at 1.4 GHz. Sea ice is a
complex composite comprised mainly of ice crystals, included
brine pockets and air bubbles. Because of the small size of
the scatterers relative to the wavelength, volume scattering
at L-band can be neglected [6]–[9]. In the case of uniform
properties within the ice sheet and plane-parallel geometry, the
radiative transfer equation for discontinuous media reduces to

2 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 6, NO. 1, JANUARY 2007
the following, closed-form equation, which will henceforth be
referred to as the three-layer (air, ice and water) model:
Tb=
(Ria −1) {[Rwiτ2+ (1 −Rw i)τ−1]Tice +
(Rwi −1)τ Tw+ (Ria −1)Rwi τ2Tsky }
(RiaRwi τ2
−1) ,
(2)
where Tbis the modelled brightness temperature, Ria and
Rwi are the reflection coefficients at the ice-air and water-ice
interfaces, calculated via the Fresnel equations (see (5) and
(6), below), Tice is the temperature of the ice, Twis the water
temperature, Tsky is the downwelling brightness temperature
from the sky and τis the transmission coefficient:
τ= exp (−
4πνhImnice
ccos θt)(3)
where νis the frequency, his ice thickness, nice is the complex
refractive index of the ice, cis the speed of light and θtis
the angle of the radiation as it is transmitted through the ice,
calculated from Snell’s law:
n1sin θ1=n2sin θ2(4)
where n1is the refractive index of the first medium, n2is
the refractive index of the second medium, θ1is the angle of
the ray in the first medium (relative to a normal drawn with
respect to the interface) and θ2is the angle of the ray in the
second medium. Once we have the two angles, the Fresnel
equations follow:
Rv=



n2cos θ1−n1cos θ2
n2cos θ1+n1cos θ2



2
(5)
Rh=



n1cos θ1−n2cos θ2
n1cos θ1+n2cos θ2



2
(6)
where Rvand Rhare the reflection coefficients at vertical and
horizontal polarization, respectively.
Model results summarizing the approximate behaviour of
the signal as the temperature, salinity and thickness are varied
can be seen in Figure 1. The figure shows brightness tem-
peratures computed for ice thicknesses ranging from 0 to 2
meters, correcting for the fact that a standard three layer model
does not converge to the open water case [10]. The salinity is
modelled as an exponential function of ice thickness [11]–[14]
for parent waters with two different salinities: the world oceans
and for the Baltic sea. Salinity and temperature determine the
complex permittivity through the mixture models of Vant et
al. [7]. The results suggest that it may be possible to retrieve
ice thickness simultaneously with ice concentration only at
temperatures close to melting.
For a more complete description of the model, see [2].
IV. DATA
The Pol-Ice campaign was conducted in March 2007 in
the Northern Baltic and comprised fully polarimetric measure-
ments from an aircraft-mounted, L-band radiometer as well as
ice-thickness measurements [2] which will not be used in this
study. It is the only reliable source of L-band measurements
over sea ice to date. The radiometer used was the EMIRAD
which has an angular field-of-view (full-width, half-maximum)
Fig. 1. Model curves for ice at different temperatures in the Baltic (S=5)
and in the ocean (S=35). Thickness is varied between 0 and 2 m and marked
at intervals by solid dots. Salinity varies as an exponential function of ice
thickness. X-axis is the polarization difference or second Stokes component,
Q=Tbv −Tbh.
Fig. 2. Map of all Pol-Ice EMIRAD measurements. Flights are labelled by
date, name and number.

MILLS AND HEYGSTER: ICE CONCENTRATION FROM SMOS 3
Fig. 3. Measured L-band brightness temperatures from the Pol-Ice field
campaign compared to ice concentrations interpolated from ASI ice maps.
of 13.16 degrees [1]. Since the aircraft was flying between 500
m and 600 m, this translates to a footprint size on the order of
250 m. Figure 2 shows most of the radiometer flights labelled
by date and time, flight number and flight name.
Ice concentrations for comparison with those derived from
the field data are derived from the ASI (ARTIST Sea Ice)
algorithm which uses the 89 GHz channel of the Advanced
Microwave Scanning Radiometer on EOS (AMSR-E) to derive
ice concentrations [15]. These are averaged once a day to
maps rectangularly-gridded on a polar-stereographic projection
centered at 70 degrees latitutde. The 89 GHz channel is
used because it is the highest frequency, therefore it has
the best resolution with the final averaging resulting in a
mean grid-cell size of 6.25 by 6.25 km. The full record of
SSM/I and AMSR-E derived sea ice maps is available online:
http://www.iup.uni-bremen.de/iuppage/satellite index.html
V. RE SU LTS
It was found that the vertical polarization of the aft-looking
radiometer was mal-functioning. Since the third and fourth
Stokes components are generated by correlating the hand v
components, this makes these two channels suspect also. Thus,
there is only one channel to work with and the concentration
retrieval in (1) reduces to a linear rescaling of a single
brightness temperature. Figure 3 compares ice concentration
interpolated from ASI ice maps with EMIRAD brightness
temperatures from the Pol-Ice campaign. The correlation, at
0.59, is relatively low, however the radically different footprint
sizes—over 3 km for AMSR-E, while for EMIRAD it will
be less than 300m—make direct comparison difficult. In par-
ticular, the small footprint of the EMIRAD instrument means
that pure signals will be more common, thus ice concentration
becomes more of an on-off value—either there is ice or there
is water. This can be clearly seen in Figure 3 and allows us
to easily pick out the two tie-points.
The ice concentrations algorithm will use tie-points of 80 K
and 200 K for open water and ice, respectively. Values lower
Fig. 4. Ice concentration retrieved from Pol-Ice EMIRAD measurements
compared with ASI ice maps. Retrieved concentrations are first averaged over
each pixel of the map.
Fig. 5. Ice concentration retrieved from Pol-Ice EMIRAD measurements
compared with ASI ice maps. Retrieved concentrations are binned in one
percent invervals based on the ASI values and averaged.
than 80K and higher than 200K are set to 0% and 100% ice
concentration, respectively.
The justification for these two tie points is further reinforced
by the model results in Figure 1. The model results suggest that
sea ice brightness temperatures converge to roughly the same
value independent of either the temperature or salinity of the
parent water. The values are also close to those chosen for the
retrieval, however the retrieval uses more moderate values for
the tie points because of the variability in the signatures of pure
ice types around the mean. We want to push concentrations

4 IEEE GEOSCIENCE AND REMOTE SENSING LETTERS, VOL. 6, NO. 1, JANUARY 2007
Fig. 6. Viewing angle dependence of sea-ice emissivity simulated using
three-layer RT model for three different scenarios of ice thickness, complex
permittivity and concentration.
close to the pure types over the border to a pure type if the
radiance signature is within this variability.
To address the issue of different footprint sizes, we average
all the EMIRAD measurements within each pixel of the ice
map and compare ice concentrations pixel-by-pixel. The ASI
ice maps are averaged daily to bins that are regularly-gridded,
along a polar-stereographic projection. This averaging does
not significantly reduce the resolution of the instrument and
is done mainly to take advantage of the many overlapping
swaths at high latitudes. EMIRAD measurements from Pol-
Ice are similarly averaged by first converting measurement
locations to the projection coordinates and then searching for
the matching ASI grid point. Results are shown in Figure 4.
Even though we have matched the instrument resolution to
the lower of the two, results may still be biased because the
pilot may be deliberately searching out either ice-covered or
open water surface and because the linear flight path traces
out only a narrow line of measurements across the broad
satellite footprint. To address this, we further average the
results by collecting them in one percent bins based on ASI
ice concentration. This generates the graph seen in Figure 5
which shows a much improved correlation.
VI. DISCUSSION
The unique operating principle of the SMOS instrument
means that each point on the ground will be viewed from
multiple angles by overlapping measurements —each mea-
surement from the instrument will consist of a single large
footprint comprising multiple pixels of varying size and ef-
fective viewing angle [16]. With a high sampling rate, these
footprints will overlap. Can extra information be gained by
having measurements from different viewing angles? Results
from the model described in Section III suggest that the best
use of these extra measurements is to interpolate them to a
single effective viewing angle.
Figure 6 shows modelled sea ice emissivity as a function
of viewing angle for three scenarios of varying ice concentra-
tion, thickness and complex permittivity. These scenarios are
designed to be degenerate at an angle of 25 degrees. Although
Fig. 7. Viewing angle dependence of sea-ice emissivity simulated using
three-layer RT model and of fitted, re-scaled Fresnel equations. For RT model,
complex permittivity is ϵ= 3.5 + 0.05iand ice thickness is h=0.5m. Fresnel
equations are for a real refractive index of n= 1.91 and have been rescaled
using e′
p=aep+bwhere epis emissivity at polarization pand a= 0.901
and b= 0.050 are constants.
we would expect them to diverge at other viewing angles,
they are almost identical, certainly to within the instrument
noise. It was found that the viewing angle dependence of ice
emissivity models of the type described in Equation (2) can be
well approximated by Fresnel equations that have been linearly
re-scaled:
e′
p=aep+b(7)
where e′
pis the fitted emissivity at polarization p(hor v) and
ep= 1 −Rpis the emissivity as calculated from the Fresnel
equations (5) and (6). Figure 7 demonstrates the procedure
which works as well if brightness temperature is modelled
with more than one ice layer.
VII. CONCLUSION
Ice concentrations derived from Pol-Ice campaign data were
compared with ASI satellite- based retrievals. Correlations
were between 0.59 and 0.76 depending on the level of av-
eraging empoyed. Several influences make the averaging pro-
cedures necessary: the two instruments (satellite and airborne)
have very different footprint sizes as well as different temporal
resolutions; the airborne radiometer had a linear path, during
which the pilot may have been searching out specific surface
types. Another possible error source: the ASI algorithm is
designed for the high salinity of the open ocean rather than
the brackish waters of the Baltic Sea.
This comparison exercise was conducted as part of the
SMOSIce project, which aims to prepare retrieval algorithms
for data collected over sea ice by the Soil Moisture and Ocean
Salinity (SMOS) satellite which was successfully launched in
November 2009. The SMOS instrument will sample the same
stretch of ground at multiple effective viewing angles within
a short time span. Ice emissivity models suggest that this data
will provide little extra information. An effective use of it,
however, would be to interpolate to a single effective viewing
angle, thus helping to stabilize the measurements.

MILLS AND HEYGSTER: ICE CONCENTRATION FROM SMOS 5
ACKNOWLEDGMENT
The authors would like to thank the following people
involved in the SMOS-Ice project for valuable discussions:
Catherine Bouzinac, Mark Drinkwater. Matthias Drusch, Ste-
fan Hendricks, Lars Kaleschke, Christian Maetzler and Ras-
mus Tonboe. This work was supported by ESA contract
21130/08/NL/EL, L-Band Radiometry for Sea Ice Applica-
tions.
REF ER EN CE S
[1] N. Skou, S. S. Sobjaerg, and J. Balling, “EMIRAD-2 and its use in
the CoSMOS Campaigns,” Electromagnetic Systems Section, Danish
National Space Center, Technical University of Denmark, Tech. Rep.
ESTEC Contract No. 18924/05/NL/FF, 2007.
[2] G. Heygster, S. Hendricks, L. Kaleschke, N. Maass, P. Mills, D. Stam-
mer, R. T. Tonboe, and C. Haas, “L-Band Radiometry for Sea-Ice
Applications,” Institute of Environmental Physics, University of Bremen,
Tech. Rep. final report for ESA/ESTEC Contract N. 21130/08/NL/EL,
2009.
[3] D. A. Rothrock, D. R. Thomas, and A. S. Thorndike, “Principal
Component Analysis of Satellite Passive Microwave Data Over Sea Ice,”
Journal of Geophysical Research, vol. 93, no. C3, pp. 2321–2332, March
1988.
[4] J. C. Comiso, D. J. Cavalieri, C. L. Parkinson, and P. Gloersen, “Passive
microwave algorithms for sea ice concentration: A comparison of two
techniques,” Remote Sensing of the Environment, vol. 60, pp. 357–384,
1997.
[5] T. Markus and D. J. Cavalieri, “An Enhancement of the NASA Team Sea
Ice Algorithm,” IEEE Transactions on Geoscience and Remote Sensing,
vol. 38, no. 3, pp. 1387–1398, May 2000.
[6] D. G. Barber, A. K. Fung, T. C. Grenfell, S. V. Nghiem, R. G.
Onstott, V. I. Lytle, D. K. Perovich, and A. J. Gow, “The role of snow
on microwave emission and sacttering over first-year sea ice,” IEEE
Transactions on Geoscience and Remote Sensing, vol. 36, no. 5, pp.
1750–1763, September 1998.
[7] M. R. Vant, R. O. Ramseier, and V. Makios, “The complex-dielectric
constant of sea ice at frequencies in the range 0.1-40 GHz,” Journal of
Applied Physics, vol. 49, no. 3, pp. 1264–1280, March 1978.
[8] A. Stogryn, “A study of microwave brightness temperatures of snow
from the point of view of strong fluctuation theory,” IEEE Transactions
on Geoscience and Remote Sensing, vol. GE-24, no. 2, pp. 220–231,
March 1986.
[9] K.-P. Johnsen, “Radiometrische Messungen im arktischen Ozean–
Vergleich von Theorie und Experiment,” Ph.D. dissertation, Universitaet
Bremen, Otto-Hahn Allee 1, Bremen, Germany, 1998.
[10] J. D. Menashi, K. M. St. Germain, C. T. Swift, J. C. Comiso, and A. W.
Lohanick, “Low-frequency passive-microwave observations of sea ice in
the weddell sea,” Journal of Geophysical Research, vol. 98, no. C12,
pp. 22 569–22 577, December 1993.
[11] W. F. Weeks and S. F. Ackley, The Geophysics of Sea Ice, ser. NATO
ASI Series B: Physics. Plenum Press, 1985, vol. 146, ch. 2.
[12] G. Cox and W. Weeks, “Numerical simulations of the profile properties
of undeformed first-year sea ice during the growth season,” Journal of
Geophysical Research, vol. 93, no. C10, pp. 12499–12 460, October
1988.
[13] W. B. Tucker, D. K. Perovich, A. J. Gow, W. F. Weeks, and M. R.
Drinkwater, “Physical Properties of Sea Ice Relevant to Remote Sens-
ing,” in Microwave Remote Sensing of Sea Ice, ser. Geophysical Mono-
graphs. American Geophysical Union, 1992, no. 68, ch. 2, pp. 9–28.
[14] J. K. Ehn, B. J. Hwang, R. Galley, and D. G. Barber, “Investigations
of newly formed sea ice in the Cape Bathurst polynya: 1. Structural,
physical and optical properties,” Journal of Geophysical Research, vol.
112, no. C05002, 2007, doi=10.1029/2006JC003702.
[15] G. Spreen, L. Kaleschke, and G. Heygster, “Sea ice remote sensing using
AMSR-E 89 GHz channels,” Journal of Geophysical Research, vol. 113,
no. C02S03, 2008, doi=10.1029/2005JC003384.
[16] Y. H. Kerr, “Soil Moisture and Ocean Salinity: Radiometry applied to
soil Moisture and Salinity Measurements,” CESBIO, Toulouse, Tech.
Rep., November 1998, the Living Planet Programme Earth Explorer
Missions.
Peter Mills was born on December 22, 1973 in
Montreal, Canada. He received a Bachelor’s Degree
in Physics from the University of Waterloo in On-
tario, Canada and a Master’s Degree in Environmen-
tal Physics from the University of Bremen, Germany.
Peter has worked most of his adult life in the fields
of climate and remote-sensing research, including
some of the most hotly discussed areas such El
Ninio/Southern Oscillation and ozone detection. He
has a passion for all forms of rational inquiry,
particularly those that cross disciplines. Examples
include chaos theory, artificial life, complexity theory and cognitive science.
He has authored two papers on chaos theory.
Georg Heygster Georg Heygster (M’00) received
the Diploma degree in solid-state physics and the
Ph.D. degree in digital image processing from the
University of Goettingen, Goettingen, Germany, in
1976 and 1979, respectively.
He served as a Consultant with the Computer
Center of the University of Bremen, Bremen, Ger-
many, from 1979 to 1988. Since then, after working
for a year on the imaging mechanisms of scanning
acoustic microscopes, he has been Head of the
Geophysical Analysis of Satellite Images group at
the Institute of Environmental Physics, University of Bremen. His research
activities include passive and active microwave remote sensing, particularly of
both surface and atmospheric parameters in the high latitudes, various aspects
of the hydrological cycle, long-term trends, and retrieval techniques. He was
or still is Principal Investigator or Coinvestigator of many research projects
funded by the European Union, the European Space Agency, the German
Research Council, and the Japan Aerospace Exploration Agency. These
projects include the development of sensor soft- and hardware, conducting
campaigns, the final data analysis from multi- and single-sensor data to
geophysical parameters, and the interpretation and application of these results
in many areas such as meteorology, climatology, and oceanography.