You are on page 1of 17

This article has been accepted for inclusion in a future issue of this journal.

Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 1

Tomographic Imaging of Fjord Ice Using a Very


High Resolution Ground-Based SAR System
Temesgen Gebrie Yitayew, Student Member, IEEE, Laurent Ferro-Famil, Member, IEEE,
Torbjørn Eltoft, Member, IEEE, and Stefano Tebaldini, Member, IEEE

Abstract— This paper presents new experimental results of due to its low sensitivity to weather conditions, to its
3-D imaging using tomographic techniques over a snow covered high-resolution imaging capabilities as well as to its
sea ice medium, sensed with an X-band radar system. The sensitivity to volumetric scattering effects. This last feature
available data are from a ground-based synthetic aperture radar
data collection campaign carried out over Kattfjord, Tromsø, of SAR is linked to the penetrating capability of microwave
Norway. Direct imaging of the vertical structures of the radar frequencies deep into semitransparent volumetric media.
reflectivity of the snow and sea ice layers is achieved by focusing A few examples of studies addressing the interaction of radar
the signal from a 2-D synthetic array in the 3-D space. The waves with snow and sea ice, and the exploitation of SAR
effect of a change in propagation velocity of the wave inside data for sea ice remote sensing applications may be found
the considered medium is investigated in the focusing process,
and the tomograms are effectively corrected for this effect. in [3]–[5]. One main application of SAR data for Cryosphere
The distribution of the scattering contributions in the vertical characterization concerns the retrieval of different physical
direction reveals a strong response from the sea ice cover. and electrical properties of snow and sea ice. The refractive
Tomograms at two different polarizations are investigated and index of snow and sea ice is one such parameter which has
compared. The results and the interpretations are also supported a major role in determining the interaction of radar waves
by the simulated data from the same system.
with the medium. The estimation of this quantity for sea ice
Index Terms— Ground-based SAR (GB-SAR), SAR polarime- is a challenge due to the fact that sea ice is a very complex
try, SAR tomography, sea ice, synthetic aperture radar (SAR). medium containing water, salt, brine, and air bubbles.
Examples of previous studies concerning the estimation of
I. I NTRODUCTION the refractive indices of snow and sea are found in [6]–[8].
Volumetric information of snow and sea ice, such as snow
T HE characterization of sea ice and its snow cover plays a
pivotal role in understanding and monitoring changes in
the global climate and ecosystem. The physical and electrical
depth, ice thickness, and snow and ice layering, is important
inputs to ecological process modeling, and understanding
properties of the ice and its snow cover control the amount of and monitoring changes in the ecosystem [9], [10]. In [11],
solar radiation reflected to the atmosphere, absorbed within we demonstrated the use of a very high resolution (VHR)
snow and ice, and transmitted into the ocean beneath the ground-based SAR (GB-SAR) measurement system in a tomo-
ice [1]. The precise analysis of the local distribution of the graphic configuration for the 3-D characterization of snow
scatterers in snow and sea ice requires the use of imaging covered fjord ice. A fundamental advantage of this approach
techniques describing this complex volumetric medium in a compared with the classical 2-D SAR imaging is its capability
3-D space. Vertical sections of such 3-D images can be used of direct imaging of the vertical structure of the reflectiv-
to reveal the vertical layering structures formed as a result of ity of the scene. This is due to the fact that the classical
changes in the amount of impurities (brine and air) trapped SAR system is inherently a 2-D imaging system, where the
in sea ice, an important information that can be linked to the 3-D scene scattering properties are projected onto the
growth rate of the ice [2]. 2-D azimuth-range plane.
Snow and sea ice characteristics have been studied using SAR tomography is the extension of the conventional
electromagnetic (EM) remote sensing techniques with passive 2-D SAR imaging to three dimensions, and is achieved by
or active sensors at frequencies ranging from metric wave- the formation of an additional synthetic aperture in elevation.
lengths to the visible domain. Synthetic aperture radar (SAR) The coherent combination of images acquired from several
is among the most widely used measurement techniques, parallel flight tracks using multibaseline SAR interferometric
techniques allows direct imaging of the volume, and therefore,
Manuscript received July 24, 2015; revised August 19, 2016; accepted it is a promising technique for the study of volumetric proper-
September 11, 2016. ties and improved estimation of geophysical parameters. It was
T. G. Yitayew and T. Eltoft are with the University of Tromsø-The Arctic
University of Norway, 9019 Tromsø, Norway. first demonstrated in 2000 using L-band multibaseline airborne
L. Ferro-Famil is with the University of Rennes 1, 35000 Rennes, France. SAR data over forested area [12]. It has been recently applied
S. Tebaldini is with the Politecnico di Milano, 20133 Milan, Italy. for forest vertical reflectivity retrieval [13], [14] and the
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. 3-D characterization of buildings in urban areas [15], [16].
Digital Object Identifier 10.1109/TGRS.2016.2613900 Very recently, the technique has been applied for the
0196-2892 © 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

2 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

3-D imaging of snowpack [17] and sea ice [11], [18].


A similar technique called tomographic profiling has been used
in [19] for snow pack application. However, the system does
not employ a 2-D synthetic array.
In this paper, we present the experimental results from the
3-D imaging of snow covered sea ice using a GB-SAR system
and investigate the vertical distribution of the radar reflectivity
in the volume. One of the main objectives is to illustrate
the important potential of coherent 3-D imaging in providing
information such as the number and location of the main
contributions that cannot be accessed through the 2-D imaging.
A related objective is to provide a quantitative description
of the scattering contributions made by the snow and ice
layers, and show how such an analysis can be used to explain
the information contained in the 2-D SAR images acquired
with other airborne or space-borne SAR sensors operating
with the same frequency as the one used here. In addition,
we aim to analyze the polarization aspect of the response
from the considered medium and compare the measurements
at two different polarizations. The data sets considered here
are acquired by a GB-SAR system, developed at the IETR,
University of Rennes 1, over Kattfjord, Tromsø, Norway. Fig. 1. Setup of the GB-SAR system.
It consists of multiple X-band measurements at VV and HV
polarizations.
The remainder of this paper is organized as follows.
In Section II, a brief description of the GB-SAR system is
provided. This section also introduces the test site where mea-
surements were collected, and the 3-D SAR focusing approach
used in this paper together with some of the processed results.
Section III covers the investigation of the effects of change
in propagation velocity of the wave inside the considered
3-D complex medium. Section IV is devoted to the charac-
terization of the scene in terms of the scattering mechanisms
based on tomographic measurements. Finally, the concluding
remarks are made in Section V.

II. E XPERIMENTAL S ETUP AND DATA P ROCESSING


A. Ground-Based Tomographic SAR System
The system used for this paper is a GB-SAR system, named
Pocket SAR (PoSAR), developed at the IETR, University of
Rennes 1. The setup of the system is shown in Fig. 1. A vector
network analyzer, which accurately controls the transmission
and reception of the radar signals, and a set of four horn
antennas are contained in a metallic box, which in turn is
mounted on a long linear rail, whose effective aperture length Fig. 2. Aperture synthesis from a single pass. An unevenly spaced antenna
is 3 m. A stepped motor is used to displace the metallic array, shifted in both azimuth and elevation (top and middle), produces a
box along the rail parallel to the observed scene to acquire six-element uniformly spaced equivalent antenna array (bottom).
multiple images exploiting the SAR technique. The rail is
fixed on two vertical supports (poles). The PoSAR system
can operate at different frequency bands, including C-, X-, The antenna array can be manually shifted to increase the
and Ku-bands, and over different polarization channels. The vertical aperture in the repeat-pass mode. Fig. 4(a) shows a
four horn antennas are displaced in the azimuth and elevation simplified geometrical configuration of the acquisition system
direction. They are also unevenly spaced for the purpose of for three passes, which is equivalent to 18 acquisition posi-
achieving a uniformly spaced equivalent antenna array [20]. tions in the vertical direction. The revisit time between two
Each antenna can be operated either as transmitter or receiver. consecutive passes may vary from 10 to 20 min.
As a result, six equivalent monostatic phase centers with a The GB-SAR system employs a stepped frequency con-
vertical offset can be achieved in a single pass (see Fig. 2). tinuous wave (SFCW) form. The system parameters are
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 3

TABLE I
GB-SAR S YSTEM PARAMETERS

summarized in Table I. The setup considered in this paper


operates at X-band with the center frequency fc of 10 GHz
and the bandwidth B f of 4 GHz. The slant range resolution δr
that corresponds to this bandwidth is 3.75 cm [see Fig. 4(b)].
Referring to Fig. 4, azimuth and elevation resolutions are
determined by the synthetic aperture length L x and L z ,
respectively. For L x = 3 m, the azimuth resolution δx at
θ = 45° is 1.73 cm. The vertical synthetic aperture length
is dependent on the number of parallel acquisition tracks in
the vertical direction. For example, for 18 acquisitions that
correspond to a 34-cm vertical baseline [see Fig. 4(a)], the
vertical resolution δz at θ = 45° is 14 cm. Therefore, the
instrument is a VHR system. Table I summarizes the different
components of the resolution of the system. Typical scene
dimensions that can be imaged using the system are 6 and 8 m Fig. 3. Ice cores and the depth of snow and ice covers at the test site.
in the azimuth and ground range directions, respectively. Even (a) Sketch of the depth of snow and ice at the test site. (b) Photos of ice
cores showing irregularly shaped, randomly oriented air bubbles.
though the illuminated scene is limited to a few square meters,
the VHR capability of the instrument allows to reveal fine
details of the considered 3-D medium.
was about 28 cm [see Fig. 3(a)]. The seawater at the fjord
was fresh, and therefore, the ice formed can be seen under the
B. Data Collection
framework of low salinity sea ice. This is mainly attributed to
Using the system described earlier, a GB-SAR data the fresh water coming from the surrounding mountains during
collection campaign was carried out by a team from the early periods of the winter. The ice contains the significant
the IETR, University of Rennes 1, in collaboration with the amounts of air bubbles of size in the order of few millimeters
members of the EO lab at the University of Tromsø, in 2013 in (ranging from 0.5 to 7 mm) (see Fig. 3b). The snow cover on
Tromsø, Norway. Two sets of tomographic measurements are top of the ice was dry.
considered in this paper. The first constitutes 36 measurements
acquired at VV polarization. The second set contains a pair of
15 measurements acquired at VV and HV polarizations. Since C. Tomographic Data Processing
the set containing 36 VV measurements provides a better A simplified geometry of the tomographic imaging system
resolution than the 15 VV measurements of the second set, is shown in Fig. 4(a). It consists of M vertically shifted
it will be used for most of the investigations of the scene in acquisition positions (from now on, we call them M parallel
this paper. The 15 dual polarization measurements that were tracks), which form a second aperture in the elevation direc-
taken at the same time will only be used to investigate and tion. Fig. 4(b) shows how the construction of this second
compare the response of the scene at the two polarizations. aperture makes the direct imaging of the vertical structures
The sea ice formed at Kattfjord is seasonal ice that can have of the reflectivity of a stratified volumetric medium possible.
a life of up to three to four months depending on the season. In contrast, the classical 2-D SAR system has limitations in
Here, we would like to note that throughout this paper, the term imaging such environments. This is due to the fact that, as it is
“sea ice” and “fjord ice” are used interchangeably. On the day shown in the same figure, the 3-D scene scattering properties
which the data were collected, the weather was cold and dry, are projected onto the 2-D azimuth-slant range plane.
with a light, dry, and intermittent snowfall. The temperature It is demonstrated elsewhere that 3-D SAR focusing can
reading, as it is obtained from the Norwegian Meteorological be formulated as a spectral estimation problem, where both
Institute, was −8 °C for the minimum and −2 °C for the nonparametric approaches such as beamforming and Capon,
maximum. Snow depth on top of the ice at the location of the and parametric spectral estimation techniques such as MUSIC
test site was about 24 cm, and the depth of the sea ice as it can be applied [21], [22]. All the spectral estimation tech-
was measured by drilling a hole into the ice using a hand drill niques for tomographic focusing share one common step that
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

4 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

transmitted wave is a narrowband signal. The ground-based


tomographic SAR system considered here uses a wideband
wave form, and operates from a few meters of altitude.
This latter feature of the system might cause a nonnegligible
variation of incidence angles for vertical displacements of the
sensor. In contrast, for space-borne and airborne tomographic
SAR systems, this variation of incidence angle is so small that
it is safe to assume that the radar is always looking at the same
resolution cell despite making small vertical displacements.
Therefore, for the acquisition system considered here, other
accurate tomographic SAR focusing techniques should be
considered.
In this paper, the 3-D signal focusing was achieved by
jointly processing multiple SAR acquisitions through time-
domain backprojection (TDBP) [23], [24]. The signal received
by the considered SFCW radar system can be formulated
as follows. The 2-D synthetic array shown in Fig. 4(a) is
formed from N acquisition positions in azimuth and M par-
allel tracks in the elevation direction. Let the antenna posi-
tion of each transmitted pulse be denoted by am,n , where
m = 1, . . . , M, and n = 1, . . . , N. In the SFCW, the
selected frequency bandwidth of the system is sampled at
frequencies f k , where k = 1, . . . , K . Hence, for each antenna
position, the received signal, sr ( am,n , fk ), in the frequency
domain can be modeled as the sum of the responses due to
each individual scatterer in the 3-D scene

4π f k
sr (
am,n , f k ) = ρ(r )e− j c ||an,m −r || d r (1)
V
where V ∈ R3
represents the 3-D space in the considered
volume, r is the scatterer position in the volume,
ρ(r ) represents frequency- and aspect angle-dependent
complex reflectivity of the medium, and || am,n − r||
represents the sensor-scatterer distance.
The 3-D matched filter response of a target at a given
focusing position r0 (x 0 , y0 , z 0 ) in a 3-D focusing grid can be
written as

K
4π f k
||
am,n −r0 ||
A(r0 ) = sr (
am,n , fk )e j c . (2)
M N k=1
In the TDBP focusing using (2), inverse Fourier transfor-
mation is applied on the frequency domain measurements of
the target’s spectral profile sr (
am,n , f k ). Then, for each 3-D
location on the focusing grid, the contributions from each of
the parallel tracks are summed together to reconstruct the full
Fig. 4. Geometry of the tomographic SAR measurement system. x, y, and z 3-D scene information. The 3-D focused signal can then be
are azimuth, ground range, and elevation directions, respectively, whereas written as
r and n are slant range and cross-range directions, respectively. (a) Simpli-
fied geometry of the tomographic SAR acquisition system. (b) 2-D versus 
M

3-D resolution cell. A(r0 ) = si (r0 )e j φi (r0 ) (3)


i=1
where si (r0 ) is the i th focused signal from each of the
a 2-D azimuth-range SAR focusing should first be applied M parallel tracks at each 3-D location r0 of the focusing grid,
separately to the acquired signal from each of the available and φi (r0 ) is the phase compensation term used in the TDBP
parallel tracks by synchronizing them to a common elevation algorithm.
position. Then, the spectral estimation techniques will be This focusing technique is very accurate as it is based on
applied to the stack of the 2-D azimuth-range focused images computing the correct radar-scatterer distances. However, it is
to reconstruct the scene information in the third dimension. also important to note that vacuum propagation velocity c is
However, the formulation is based on the assumption that the assumed in the focusing process.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 5

obtained by averaging 37 azimuth looks. Here, we would


like to note that I (r0 ) is computed from the radiometrically
corrected data, and hence, it represents the backscattering
coefficients of the medium. Details regarding the correction
procedures applied to the data can be found in Section III-B.
Referring to the 2-D SAR images of Fig. 5, as a conse-
quence of projecting all the contributions of the 3-D volume
on a 2-D focusing plane (inherent in classical 2-D SAR
imaging), it is difficult to associate the information to any of
the characteristics of the scene. However, from the tomograms,
it can be clearly seen that the reconstructed volumetric medium
is characterized by a complex multilayered vertical structure.
By looking at the intensity tomograms of Fig. 5, and compar-
ing them with the ground truth information of Fig. 3(a), it is
difficult to associate those structures, having strong backscatter
power, to either the layers or to the interfaces between the
layers. This is due to the fact that the structures are displaced
from their actual positions. Referring to Fig. 3(a), the snow
plus ice medium under investigation is about 52 cm thick.
However, from the tomograms of Fig. 5, the reconstructed
medium appears dilated at near range and contracted at far
range. As it is shown in Section III, this geometrical distortion
is due to wave refraction phenomena and the associated change
in propagation velocity of the wave in the observed medium.

III. I NVESTIGATION OF WAVE P ROPAGATION I NSIDE


A D ENSE M EDIUM OF S NOW AND I CE
As it is observed from the tomograms [see Fig. 5(a)],
all the structures in the tomograms appear tilted with a positive
slope from near range to far range, and this tilting effect is
more pronounced at the bottom layers. As a result of this
distortion, the different layers and interfaces are displaced
from their actual positions. This phenomenon is due to vari-
ations of propagation velocity of the transmitted wave and
the associated change in its propagation direction in the ana-
lyzed stratified medium as it is described by the well-known
Snell’s law [25]. In the TDBP focusing using (2), vacuum
propagation velocity c was assumed despite the fact that the
radar wave is propagating in a dense medium (snow and ice).
Fig. 5. Two-dimensional SAR images and intensity tomograms
based on a different number of acquisitions. (a) Two-dimensional In this section, this variation of wave propagation velocity
SAR image and intensity tomogram from 36 acquisitions at VV. will be investigated in detail. For a simple demonstration,
(b) Two-dimensional SAR image and intensity tomogram from homogeneous and perfectly horizontal snow and ice layers
15 acquisitions at HV.
are considered. Let the propagation velocity (and refractive
index) of the snow and ice layers be v s (n s ) and v ice (n ice ),
D. Tomograms respectively. The air–snow and snow–ice interfaces are consid-
In this section, the processed data sets from the tomographic ered, and the principle of wave propagation across interfaces
acquisition system described in Section II-A will be intro- of media with differing permittivity is applied. The following
duced. Throughout this paper, the 3-D information of the scene derivation is an extension of the derivation given in [26] for
is presented in the form of tomograms (2-D sectional images) one interface to two interfaces.
by vertically sectioning the 3-D focused image of the medium. Fig. 6 shows the path followed by a wave transmitted
Fig. 5(a) shows a 2-D SAR image (top) from one of the from an antenna elevation position H under two conditions.
36 acquisitions, focused on an azimuth-ground range plane, The first one is when the wave is assumed to continue
and a multilooked intensity tomogram (bottom) displayed propagating inside the snow and ice with vacuum propagation
as a height-ground range section, from 36 acquisitions at velocity c (shown in black), and the second one is when
VV polarization. Fig. 5(b) shows a similar presentation from the wave experiences a change in wave propagation velocity
a different data set, which consists of 15 acquisitions at and direction due to a change in the medium from air-to-
HV polarization. The 3-D intensity I (r0 ) is computed from (3) snow and snow-to-ice (shown in red). Inside the considered
as I (r0 ) = |A(r0 )|2 . All the tomograms in this paper are medium, the apparent positions of the scatterers are designated
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

6 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

this interface. Here again, θ0 , θ1 , and θ2 are related to each


other by Snell’s law.
Similar to the case of the snow layer, a relationship between
the actual z ice and the apparent z ice elevation positions of a
scatterer in the ice layer can be established by the following:
1) replacing (τ/2) in (7) with ((τ0 )/2) − ((z ice  )/

(c(cos θ0 ))), which is obtained from (4);


2) replacing ((τds )/2) in (7) with ((τ0 )/2) + ((ds )/
(v s cos θ1 )), which is obtained from (5) after replacing z s
with −ds .
After the above replacements, a relationship between
 can be obtained from (7), and can be written as
z ice and z ice

z ice cos θ2 n s cos θ2
Fig. 6. Apparent target height resulting from a change in propagation velocity z ice = + ds − ds (8)
of the transmitted wave inside the snow and ice. B is the apparent position n ice cos θ0 n ice cos θ1
of target A when c is used for focusing.
where Snell’s law is used again in replacing the velocity ratios
with refractive index ratios.
Investigating (6) and (8) reveals that the targets at near
by (yk , z k ), and the corresponding actual positions are desig-
range appear at a larger depth whereas those at far range
nated by (yk , z k ), where the subscript “k” stands for either the
appear at a smaller depth than their actual depth if vacuum
snow layer “s” or the ice layer “ice.”
propagation velocity is erroneously used for focusing. This
From Fig. 6, ds and dice are positive constants, which
is also shown in Fig. 7. The top panel of the figure shows
represent the thickness of the snow and ice layers, respec-
a plot of the actual and corresponding apparent position of
tively. For a certain pair of incidence angle and delay (θ0 , τ )
the interfaces. From the ground truth information of Fig. 3(a),
(see Fig. 6), the apparent location of a target inside the medium
the actual positions of the interfaces are known. The air–snow
(kth layer) assuming propagation in air is given by
interface is located at z = 0 cm, the snow–ice interface is
cτ c located at z = −24 cm, and the ice–seawater interface is
z k = H − cos θ0 = − (τ − τ0 ) cos θ0
 2  2 located at z = −52 cm. These interfaces are represented
yk = H − z k tan θ0 (4) by horizontal lines and are plotted in red. Using (5) and (6)
where τ0 is the delay corresponding to a target at z = 0, and for the snow–ice interface, and (7) and (8) for the ice–water
θ0 = arctan(y/H ) at z = 0. Accounting for wave propagation interface, their corresponding apparent positions (y  , z  ) are
inside the snow layer (k = s), the actual location of a scatterer computed and plotted in black. The plots clearly show the
inside this layer (ys , z s ) will be given by deformation undergone by horizontal interfaces in a medium
of propagation velocity v, when vacuum propagation velocity c
−v s
zs = (τ − τ0 ) cos θ1 is used for focusing. These apparent interface position plots
2 are superimposed on the intensity tomogram (see the bottom
ys = H tan θ0 − z s tan θ1 (5)
panel of Fig. 7). It is interesting to note from the figure that
where θ0 and θ1 are related to each other by Snell’s law. the apparent positions of the interfaces which are derived from
After replacing (τ/2) in (5) with ((τ0 )/2) − ((z s )/(c(cos θ0 ))) the model approximately follow the shape of the structures
which is obtained from (4), the following relationship between with strong backscatter power in the tomograms. Therefore,
the actual (z s ) and the apparent (z s ) elevation positions of a the dilation at near range and contraction at far range of the
scatterer in the snow layer can be obtained as tomograms is indeed due to the change in propagation velocity
of the wave inside the snow and ice layers. This distortion of
v s cos θ1 n 0 cos θ1 z  cos θ1
z s = z s = z s = s (6) the tomograms can also be easily simulated (see Section IV-A).
c cos θ0 n s cos θ0 n s cos θ0 Summing up, using the horizontal and homogeneous snow and
where Snell’s law is used in replacing the velocity ratios with ice layers hypothesis stated earlier, we have shown that the
refractive index ratios. In this paper, n 0 , the refractive index distortion in the tomograms can be explained to be due to
of air, is assumed to be unity. changes in wave propagation velocity in the analyzed medium.
Similarly, accounting wave propagation inside the ice layer Therefore, for a proper interpretation of the tomograms, the
(k = ice), the actual location of a scatterer inside the ice layer effect of propagation velocity has to be taken into account for
(yice , z ice ) will be given by in the focusing process.
−v ice
z ice = (τ − τds ) cos θ2 − ds
2 A. Estimation of Refractive Indices of Snow and Fjord Ice
yice = H tan θ0 + ds tan θ1 − (ds + z ice ) tan θ2 (7)
As it is stated elsewhere [27], [28], there are two sets of
where ds is the thickness of the snow layer, θ2 is the angle parameters that influence the strength of the radar backscatter
of refraction at the interface between the snow and ice layers, signal from natural media. The first set is the sensor para-
and τds is the two way propagation delay from the antenna to meters, such as frequency, polarization, and incidence angle,
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 7

constant  of the medium. In [29], a waveguide apparatus was


used to measure the dielectric constant of low salinity sea
ice from the Baltic sea at 10 GHz. It is reported that about
28 samples with different ranges of salinity and temperature
values are used for the experiment. The reported refractive
index values vary from 1.73 to 2.12. Another experimental
study in [30] was conducted to investigate the dielectric prop-
erties of both fresh ice and sea ice at microwave frequencies.
Depending on temperature, salinity, and density of the ice,
the reported values of the refractive indices of sea ice show
a wide range of variation (from 1.58 to 1.95). Therefore, our
estimated value, n ice = 1.7 ± 20%, for fjord ice, which is
expected to be low saline, is within the range of the reported
values in the works just mentioned.
Depending on the wetness, density, age, and grain size
Fig. 7. Actual (red) and apparent (black) locations of the interfaces. of the snow, the refractive index values of snow at
microwave frequencies reported in the literature show a big
scatter [8], [31]–[33]. However, for dry snow, the refractive
and the second set is the physical properties of the observed index at microwave frequencies is mainly dependent on the
medium. The two most important physical properties are the snow density [34], [35]. Typical refractive index values of
surface roughness and the dielectric constant of the medium dry snow reported in these two previous studies vary between
under consideration. Much of the early works on sea ice 1.1 and 1.4, where the higher values correspond to aged
type mapping using SAR data have been relied on these two seasonal snow. Depending on the density, the refractive index
important parameters [3], [4]. The dielectric constant is a key values of dry snow at 9.375 GHz are reported to vary
parameter of wave scattering in dense media as it affects between 1.18 and 1.73 [33]. A data set acquired by the
scattering from volumes as well as surfaces. This section deals same tomographic SAR system as used here has been used
with the estimation of the dielectric constants of the snow in [36] for snowpack permittivity retrieval. The reported snow
and ice layers from the deformation of the tomograms of the refractive index values vary between 1.1 and 1.7. Therefore,
Kattfjord data. our estimated value of 1.4 ± 17% for the refractive index of
From the model derived in Section III and the discussion dry and seasonal snow corresponds well to the values reported
therein, (6) and (8) can be used to estimate the refractive in the aforementioned previous works.
indices of the medium. Approximate readings of the apparent Another parameter that signifies the knowledge of the
interface positions (ys , z s ) and (yice
 , z  ) can be obtained from
ice dielectric constant of snow is snow density. The real part of
the distorted tomograms. Then, the refractive index of the the dielectric constant of dry snow is a sole function of its
snow can be estimated by looking for a value of n s in (6) density. There are a number of models that can be used to
that makes the slop of (ys , z s ) matches with the slop of the relate the dielectric constant of dry snow with its density [32],
snow–ice interface in the tomogram. Referring to Fig. 7, the [34], [37]. Therefore, once the refractive index of the snow is
idea is to find a black line for the corresponding red line, estimated from the tomograms, the density of the snow layer
which matches the interfaces in the tomogram. In the figure, may also be estimated.
the approximate fit for the snow–ice interface is found for a
value of n s = 1.4±17%. Similarly, (8) can be used to estimate B. Correction of Tomograms for Propagation Delay
the refractive index of the ice. An approximate fit for the We have seen from the preceding investigation that the
ice–seawater interface was found for a value of n ice = tomograms are distorted as a result of the radar waves having
1.7 ± 20%. The errors reported here are associated with the a different propagation velocity inside the considered dense
uncertainties of locating the apparent interfaces in the distorted medium than in vacuum. Once the refractive indices of the
tomograms. Here, it is important to emphasize that the esti- snow and ice are estimated, this distortion can be corrected.
mation technique just presented is based on the assumption The correction can be done in two ways. The first one is by
that the snow and ice layers are homogeneous and can be refocusing the data after replacing the distance || am,n − r0 ||
represented by their respective bulk dielectric constant values. in (2) of the TDBP focusing with the electrical distance com-
In other words, local inhomogeneities are ignored, and as puted using the estimated refractive indices [36], [38]. The sec-
a result, the estimated refractive index values can only be ond method is a postfocusing approach where once the data are
considered as approximate. focused using the assumption of vacuum propagation velocity,
Estimation of refractive index and/or dielectric constant of the tomograms are corrected by interpolating them from the
sea ice and snow has been studied by different authors. In this apparent to the true positions. In this paper, for simplicity rea-
text, the values of refractive indices of snow and sea ice are sons, the latter approach is used. For each scatterer’s apparent
either directly reported as appeared in the respective literature position (y  , z  ) in the distorted tomogram, (4), (6), and (8)
or obtained from the dielectric constant as the two are related are used to compute the corresponding actual posi-
by n = (())(1/2) , where () is the real part of the dielectric tions (y, z) by making use of the estimated refractive indices.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

8 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Fig. 9. Multilooked tomograms based on 36 acquisitions. (a) Intensity


tomogram. (b) CR tomogram.

IV. C HARACTERIZATION OF S CATTERING M ECHANISMS


This section is devoted to the analysis of the collected
tomographic data sets. From the result in Section III-B [see
Fig. 8(c)], one can make a general statement that the portion
of the tomogram at or around the interfaces is character-
ized by a relatively strong response than the portion of the
tomogram farther away from the interfaces. To highlight the
dominant scattering mechanism in the different sections of the
Fig. 8. Correction of VV tomograms for propagation delay inside the tomogram, a second quantity called coherence ratio (CR) is
considered medium. (a) Before antenna pattern and range spread loss
correction. (b) After antenna pattern and range spread loss correction.
defined in (9). The CR depicts, for each 3-D bin, the coherent
(c) After propagation delay correction. to incoherent focusing amplitude ratio. The numerator of
CR is the modulus of the matched filter response given
Then, the tomograms are corrected by interpolating the data by (3), whereas the denominator represents the incoherent
from the apparent to the true positions. However, before the sum of the amplitudes sampled over the M multibaseline
interpolation process is applied to correct for propagation acquisitions
 M 
 j φi (r0 ) 
i=1 si (r0 )e |A(r0 )|
delay, radiometric correction is applied to the data. This
includes an antenna pattern correction, range spread loss CR(r0 ) = M = M . (9)
correction and radiometric calibration of the data in reference i=1 |si (r0 )| i=1 |si (r0 )|
to a TerraSAR-X image acquired over the fjord on the same From the nature of this definition, the quantity CR will have
date. The antenna pattern correction is a postfocusing imple- the tendency to enhance the amount of correlated information
mentation applied after computing the angles by which each contained within the M echoes at location r0 . Therefore,
apparent position in the tomogram is observed by the system, CR values will be higher for point-like scatterers, for which
i.e., θ0 of Fig. 6. All the intensity tomograms in this paper the phase variation over the M images exactly follows
are obtained after the aforementioned radiometric correction the model used to perform 3-D focusing. In other words,
procedures are applied to the data. Tomograms before and if φi (r0 ) ≈ φk (r0 ) for k = 1, 2, . . . , M, which is a char-
after antenna pattern correction are shown in Fig. 8(a) and (b), acteristics of isolated scatterers, CR values will be very
respectively. Finally, the correction for the distortion due close to unity. If the medium is relatively well localized, but
to propagation delay is applied and the result is shown its phase variation does not exactly follow the model used
in Fig. 8(c). As it can be seen from Fig. 8(c), the upper for 3-D focusing, the CR values will still be relatively high but
surface of the snow is at zero elevation position, and the snow– not close to unity. In other words, for a relatively well localized
ice and ice–water interfaces are at around −24 and −52 cm, media, similar to the one investigated here, higher CR values
respectively, which agrees with the ground truth measurements at a given location indicate the presence of a dominant contri-
of Fig. 3(a). It can also be observed that, after the correction, bution at that point compared with other points, which are at
all the interfaces are approximately horizontal. the same range inside a given resolution cell. This is clearly
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 9

depicted in the CR tomogram of Fig. 9(b). Both the corrected


intensity and CR tomograms obtained from 36 acquisition at
VV polarization are shown in Fig. 9. Therefore, in analyzing
tomograms, the CR can be used to enhance the dominant
contributions within the limits of the tomographic resolution
cell. Further discussions on the CR response are given at the
end of this section.
Referring to the intensity tomogram of Fig. 9(a), it can
be clearly seen that the regions around the interfaces are
characterized by a stronger intensity response than the regions
farther away from the interfaces. The backscatter from a
snow covered sea ice medium is generally considered as a
sum of several contributing backscattering sources. These are
surface/interface backscattering from air–snow, snow–ice and
ice–water interfaces, and volume scattering contributions from
inhomogeneities within the snow and the ice layers [27]. The
examples of inhomogeneities in sea ice include brine pockets,
air bubbles, and drainage structures. The relative strength of
the backscatter power at the interfaces is dependent on the
roughness of the surfaces with respect to the wavelength used
and the level of dielectric mismatch between the two media
forming the interface. On the other hand, the relative strength
of the backscatter power from volume inhomogeneities is
a function of the dielectric contrast between volume inho-
mogeneities and the background, size (with respect to the
wavelength), shape, orientation, and correlations in positions
of the inhomogeneities within the volume. Fig. 10. Dry snow layer from top to bottom. It is also homogenous across
the whole snow layer except at the top where there is about 5 mm snow
The relatively strong air–snow interface backscattering, crust. (a) Thin layer of snow crust (about 5 mm) at the top of the snow layer.
compared with the snow volume response, can be interpreted (b) Dry and homogeneous (except at the very top) snow layer.
to be related to the condition of the snow at the test site.
During data collection, it was observed that the snow layer
was dry from top to bottom [see Fig. 10(b)]. Moreover, it was relationship between radar backscatter cross section of slightly
observed that the snow layer is homogeneous (filled with dry rough natural surfaces and incidence angle for monostatic
powder of snow grains) except at the top. As it can be seen radar measurements [35].
from Fig. 10(a), there is a thin snow crust of about 5 mm thick, Apart from the thin crust and the small-scale roughness
formed at the top surface of the snow layer. This is a relatively structures at the top, the rest of the snow layer appears dry
hard, upper layer of compacted snow. Such structures can be and homogeneous. At X-band, volume scattering from dry
formed as a result of melting and refreezing of the top layer snow is very low [17], [41], [42]. Therefore, the backscattering
of the powder in response to variations in temperature and contribution from the rest of the snow volume is expected to
wind conditions [39]. The sources of temperature variation be very low. The space (volume) between the first and the
at the air–snow interface are thermodynamic process in the second interfaces is characterized by a relatively low response
surrounding atmosphere and the snow and ice layers [40]. [see Fig. 9(a)]. Therefore, the backscatter contribution from the
Therefore, scattering from this thin, relatively hard crust will snow cover is mostly due to the crust at the top and the small-
contribute to the observed strong response at and around the scale roughness from the upper surface of the snow cover. This
air–snow interface. is clearly depicted in the intensity tomogram of Fig. 9(a) and
Another snow property that could possibly contribute to the intensity plot of Fig. 17(b) by the relatively strong response
the relatively strong response from the air–snow interface is around the zero elevation position.
the scattering due to the small-scale roughness of the top From the intensity tomogram of Fig. 9(a), it is noticed
surface of the snow layer. Fig. 11 shows the top surface of that the ice layer has a stronger response at and around the
the snow layer. Even though the roughness is not analyzed snow–ice interface, compared with the response from the
quantitatively, a closer look at the surface in Fig. 11(b) reveals section of the ice layer farther away from this interface. This is
that it is far from being plain and perfectly smooth. The more pronounced for ground range values greater than 2.5 m.
small scale structures can easily be noticed from the photo. The strong backscatter response at and around the snow–ice
Therefore, these structures could cause surface scattering from interface is mainly attributed to the dielectric contrast between
the top of the snow layer. One indication for this, as it can snow and sea ice, the roughness of the top surface of the ice,
be clearly seen from the intensity tomogram of Fig. 9(a), and the air bubbles close to the top surface of the ice layer.
is the relatively low response of the air–snow interface at Fig. 12 shows a photograph of the top surface of the ice layer.
large incidence angles. This is due to the well-known inverse As it can be seen from the figure, the ice surface contains
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

10 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Fig. 12. Roughness of the top surface of the ice layer due to small patches
of snow sticked to the surface.

the roughness of the top surface of the ice and the bubbles in
the region. However, quantifying the individual contributions
of these components needs further investigation. The relative
weakening of the response of the ice layer as a function of
depth of the ice can be interpreted to be associated with
the volume extinction phenomenon caused by the increased
involvement of the air bubbles deep in the ice layer in multiple
scattering processes. Finally, the ice–water interface response
at the bottom is associated with the big dielectric contrast
between ice and seawater, and the facets observed at the
bottom face of the ice layer (see Fig. 19).
It is stated in many sea ice studies in the literature that
first year sea ice contains significant amount of brine pockets
but fewer air bubbles [27], [41], [43]. As a result, in such
types of sea ice, it is suggested that the brine pockets are the
main inhomogeneities that have significant contribution to the
backscatter signal. On the contrary, multiyear sea ice is char-
acterized by its significant amount of air bubbles, fewer brine
pockets, and high level of surface deformation. Therefore, in
such types of sea ice, it is generally agreed that the air bubbles
Fig. 11. Photographs of the top surface of the snow layer. (a) Top surface are the main sources of volume backscatter. The physical
of the snow layer. (b) Closer look of the top surface of the snow layer. description of both first and multiyear sea ice is discussed
in detail in [2] and [27]. In our scene, it is pointed out in
Section II-B that the ice at Kattfjord is seasonal ice having
small patches of connected snow grains that are sticked to the only a few months of life time, with low salinity characteristics
ice surface, and these structures will act as roughness elements. and containing significant amounts of air bubbles (see Fig. 3b).
Such structures can be formed as a result of melting and The contribution of these bubbles is depicted by the rela-
refreezing of the top surface of the ice layer [27]. Moreover, tively strong response of the ice medium in the intensity
the ice cores from Fig. 3b reveal that the ice layer contains tomogram.
enormous amount of air bubbles embedded inside the ice layer. Another observation that can be made from Fig. 9(a) is
These bubbles will contribute to the stronger backscatter signal the difference in the level of response among the different
around the interface by creating more dielectric discontinuity interfaces. As it can be seen from the figure, compared with the
in the region. Therefore, the strong response at and around the air–snow and ice–seawater interface backscattering, the snow–
snow–ice interface can be interpreted as a combined effect of ice interface backscattering is significantly stronger. This is
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 11

Fig. 13. Local incidence angles at the three interfaces.

mainly due to the difference in the scale of roughness among


the three interfaces and the way radar backscatter behaves with
respect to a change in wavelength and local incidence angle.
Due to the refraction phenomenon discussed in Section III, the
range of local incidence angles varies at the interfaces. This
variation depends on the rate at which the refractive indices
of adjacent layers changes, and it is governed by Snell’s law.
Fig. 13 shows a plot of local incidence angles versus ground
range at the three interfaces. As it can be observed from
the figure, the range of local incidence angles decreases as a
function of depth of the interfaces. In general, this would mean Fig. 14. Variation of CR across the tomogram is linked to the characteristics
of the CR response in enhancing dominant contributions within the limits of
that, for a relatively similar roughness scale of the interfaces, the tomographic resolution cells.
the response from the deeper interfaces at a given ground range
position would be stronger than the corresponding response
from the shallower ones [36]. In the context of our obser- The difference in the level of responses from the interfaces
vations, this will have an enhancing effect on the response of is also reflected in the CR tomogram of Fig. 9(b), which
the snow–ice and ice–water interfaces, compared the air–snow agrees with the intensity tomogram. However, there are some
interface. notable differences between Fig. 9(a) and (b). For example,
One more phenomenon, which causes enhancement of the the air–snow interface has a higher CR response at near range
responses from the deeper interfaces compared with the shal- than at far range, whereas the ice–water interface has a higher
lower ones, is the shift in wavenumber due to the presence CR response at far range than at near range. In contrast, the
of the upper layers. In [44]–[46], it is demonstrated that the CR response from the middle interface is fairly constant for
response from the ground is higher when there is a snow cover longer ground range extent. This is linked to the nature of
on top of the ground than when it is snow free. This is because CR as discussed at the beginning of this section. The CR
of the shift in wavenumber experienced by the wave while it enhances the dominant contributions within the limits of the
travels through the snow cover. Due to the relatively higher tomographic resolution cells. This is illustrated by a simple
refractive index of the snow compared with that of air, the sketch in Fig. 14. Sample points “A” to “D” are chosen
wavelength of the propagating wave will be shortened. This at different points within cells “1,” “2,” and “3.” From the
shortening of the wavelength experienced by the wave, while it intensity tomogram, sample point “C” is more dominant
travels through the snow layer will make the deeper interfaces than points “B” and “D” inside cell “2.” As a result, the
appear rougher for the radar compared with the shallower ones. CR enhanced point “C” than points “B” and “D.” Similarly,
The same argument is valid for the ice layer versus the ice– Points “A” and “E” are the dominant sample points inside
water interface. This will have an enhancement effect on the cell “1” and cell “3,” respectively. As a result, the CR has an
response of the deeper interfaces than the shallower ones. In enhancement effect at these two points than any other point in
summary, the effect of the two enhancing phenomena just dis- their respective cells. In summary, a relatively high CR value
cussed will be more pronounced at the already rougher snow– at a given point implies that, at that point, there is a strong
ice interface. This is clearly depicted in the intensity tomogram response corresponding to a relatively well localized medium
of Fig. 9(a). whose response is significantly higher than the response
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

12 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Referring to Fig. 15 and assuming homogeneous and per-


fectly horizontal layers, y can be computed as
Hβ z 1 n 0 β
y=  + (11)
1 − β2 n0
1− n1 β2
z 2 n 0 β z 3 n 0 β
+ +
n0 n0
1− n2 β2 1− n3 β2

where β = sin θ0 , n i and z i for i = 1, 2, 3 are the refractive


index and the thickness of the i th layer, respectively.
The total electrical distance de between the sensor posi-
tion H and a point at P can be computed as
de = n 0 d0 + n 1 d1 + n 2 d2 + n 3 d3 . (12)
Fig. 15. Radar-scatterer distance computation by taking into account the By making use of (11), de in (12) can be written as
variations in propagation velocity of the transmitted wave inside a multilayered
dense medium. Many of the symbols have already been defined in connection H z 1 n 0
to Fig. 6. de = n 0  + n1 (13)
1 − β2 n0
1− n1 β2
z 2 n 0 z 3 n 0
of any other point located at the same range. Therefore, +n 2 + n3 .
n0 n0
the use of the CR information for interpreting tomograms 1− n2 β2 1− n3 β2
should take into account this effect. This variation in CR
response as a function of ground range is also simulated Equation (13) is used to compute radar-scatterer distances
in Section IV-A. during the simulation process. Then, the simulated raw data
are focused by using geometric distances computed based on
the assumption of vacuum propagation velocity throughout the
whole medium.
A. Simulating Some of the Effects Observed in the In raw SAR data simulation [47], the complex reflectivity
Tomograms map ρ of the medium in (10) can be computed from
In this section, we present simulation experiments to val- EM-scattering models. For rough surfaces, models, such as
idate some of the phenomena observed in the tomograms. small perturbation model (SPM), integral equation model, and
Special attention will be given to the dilation at near range Kirchhoff scattering models, can be used [48]. For volumetric
and the contraction at far range of the tomograms, the angular media, models derived based on the first Born approximation,
scattering pattern at the interfaces, and the variation of CR distorted Born approximation, strong fluctuation theory, and
as a function of ground range. For a simple illustration, a radiative transfer approaches are commonly used [48], [49].
hypothetical model that pictures the considered multilayered In modeling scattering from rough surfaces, roughness
medium as a volume containing ideal discrete scatterers and parameters, such as the surface height standard deviation (σ )
interfaces with a small-scale roughness is considered. In and correlation length (L), are used to characterize the
simulating the raw tomographic SAR data, the discrete version randomness of the surface [48]. Therefore, knowledge
of (1) is used. It can be written as of these parameters is vital. In modeling the scattering
processes from a volumetric medium, depending on whether a
 4π f k continuous or discrete random volume model is used, different
sr (
am,n , fk ) = ρ(r )e− j c d
(10) input model parameters are required. The discrete model
V simulates the scattering volume as containing inhomogeneities
(discrete scatterers) embedded in a homogeneous background.
where d is the radar-scatterer distance, and all the remaining Therefore, the input parameters in such cases include the
parameters are defined in Section II-C. As it is discussed in permittivities of both the inhomogeneities and the background,
Section III in detail, the distortion of the tomograms is a the fractional volumes of the constituents, the shapes, the
consequence of focusing the signal without taking into account sizes, the orientation distributions of the scatterers, and the
the fact that the wave is propagating in a dense medium with layer thickness. In cases where there are more than one type of
a varying velocity. This effect can be simulated by using inhomogeneity in the volume, a mixture formula has to be used
electrical distances de instead of the geometric distances d to determine the equivalent mixture permittivity. In the case
in (10). For each simulated scatterer position r, de can be of a continuous model, the variance of the dielectric constant
computed by using the estimated refractive indices of the snow function is used to measure the strength of the dielectric
and ice layers. Fig. 15 shows how de can be computed for fluctuation across the volume. Typical input parameters for
three layers. the continuous model include variances, correlation lengths,
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 13

Fig. 16. Comparison of simulated and real tomograms from fjord ice. (a) Simulated with n 1 = 1.4 and n 2 = 1.7, X-band, VV, 18 acquisitions. (b) Tomogram
from real data (Kattfjord), X-band, VV, 18 acquisitions.

effective permittivity of the equivalent medium, and thickness The following setup of the GB-SAR system is considered.
of layers [48], [49]. Therefore, modeling scattering from The height of the reference antenna is 2.4 m, the azimuth
natural media requires precise quantitative description of the aperture length is 1 m, and three passes were used to simulate
physical state of the medium. However, our measurement 18 acquisitions. Even though 36 acquisitions are used in
lacks such detailed quantitative description of the medium. the analysis of the real data above, only 18 acquisitions are
As a result, the actual radar backscatter from the considered considered here. This is to reduce the amount of time required
snow and fjord ice scene cannot be accurately modeled, to generate the raw data. The dimension of the simulated
and that is not our objective here. The objective is to try to volume is 1.2 m in azimuth, 4.5 m in range, and 0.52 m in the
simulate the above-mentioned effects without giving too much height direction. The air–snow interface is placed at z = 0 m,
attention to the magnitude of the modeled backscattering snow–ice interface at z = −0.24 m, and ice–sea interface at
coefficients. What is presented here is an illustration. z = −0.52 m, which matches with the depth of the different
To illustrate the angular radiation pattern observed in layers in the considered fjord ice scene [see Fig. 3(a)].
the tomograms, the SPM is used to simulate the responses After simulating the raw multibaseline SAR data
from the interfaces. All the three interfaces, i.e., air–snow, using (10), the TDBP focusing approach that was discussed
snow–ice, and ice–seawater, are assumed to have a small- in Section II-C is used to focus the signal. Here, it is important
scale roughness. The values of surface roughness parameters, to note that vacuum propagation velocity is used for focusing.
which are used as inputs to the SPM, are chosen based on Tomograms from the simulated tomographic SAR data are
the validity conditions stated in [28]. The three conditions are shown in Fig. 16(a). The values in the intensity tomogram
kσ < 0.3, k L < 3, and σ/L < 0.3, where k = 2π/λ. From are scaled values, and cannot be considered as modeled
the second criterion, the maximum value of the correlation backscattering coefficients. However, the dilation at near
length for the frequency bandwidth used in this paper is range and the contraction at far range are clearly apparent.
about 1.79 cm. This value is used for the surface correlation The shape and the position of these interfaces also agree
lengths of the three interfaces. Then, within the limits of the with the shape and position of the corresponding interfaces
validity conditions, arbitrary σ values are used. Here, it is in the real tomogram of Fig. 16(b). The uncorrected intensity
important to note that, in practice, the σ and L values should tomogram from 18 acquisitions is shown in Fig. 16(b)
be taken from the actual measurements of the roughness of for comparison. This clearly demonstrates that in focusing
the three interfaces of the considered medium. The space tomographic SAR measurements from semitransparent media
between the interfaces (volume) is represented by idealized such as snow and ice, failing to account for the change in
scatterers having some complex reflectivity values. According propagation velocity of the wave would cause distortion of
to the speckle model [50], the real and imaginary components the tomograms and this in turn would affect the interpretation
of these complex coefficients can be modeled as Gaussian of the results.
random variables. In the simulation, we have used scaled The angular radiation pattern at the interfaces, i.e., the
complex coefficients according to this model. Then, during decrease in the backscatter signal as a function of incidence
the generation of the raw SAR data, these complex reflectivity angle, can also be noted in the simulated tomogram. This
values are inserted in place of ρ in (10). is more apparent in the first and last interfaces than the
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

14 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

Fig. 17. Intensity plots derived from the tomogram of Fig. 9(a). (a) Sum of intensity values in the snow and ice layers. (b) Intensity plots as a function of
depth for three representative incidence angles.

middle one. A possible explanation for this would be the fact Quantitatively, the ice cover accounts for about 84% of the
that there will be more mixing of the signal coming from the total backscattering contribution. The remaining 16% corre-
surrounding volume with the middle interface than the top sponds to the snow cover.
and bottom ones. Finally, the strong CR response at the top The contribution from the multilayered complex medium
left and bottom right corner of the CR tomogram that was can also be quantified by plotting the height profile at a given
discussed in Section IV is clearly visible in the simulated incidence angle θ0 . Fig. 17(b) shows the vertical profile of
CR tomogram too. As it is discussed in that section, this the snow plus ice complex medium for three representative
is inherently linked to the shape of the 3-D resolution cell. incidence angles. Plots such as the ones shown in Fig. 17
These are the three effects that can be easily illustrated with- can provide complementary information for 2-D SAR images
out the backscattering coefficients being accurately modeled. acquired using a different sensor but operating with the same
Had enough quantitative measurements regarding the physical frequency. The ideal example here is TerraSAR-X. Consider a
state of the medium been made, a similar procedure would single pixel from a 2-D SAR image of the Kattfjord scene
have been used to compare measured and simulated backscat- acquired using TerraSAR-X. The backscattering signal of
tering coefficients. This will be further investigated in the such a single pixel is composed of contributions from the
future when data supported by such ground truth information snow volume, the ice volume, and all the interfaces. In a
are available. stripmap mode, TerraSAR-X operates with incidence angle
values between 20° and 45° [51]. The representative incidence
B. Quantitative Characterization angles chosen in Fig. 17(b) fall in this interval. The plots show
So far, the level of the contributions from each of the how the vertical profile of a single pixel from TerraSAR-X
layers and interfaces has only been described qualitatively. acquired at a given incidence angle could look like. As it is
In this section, quantitative description is presented to get observed from the tomograms, the plots show a general trend
more specific information regarding the contribution of the of a higher response at/around the interfaces compared with
different components. The ultimate objective is to show how the region farther away from the interfaces. This is depicted
such information can be used to explain measurements made by the peaks of the plots at around 0, −0.24 and −0.52 m,
by other SAR sensors, such as TerraSAR-X. Quantitative which corresponds to the location of the air–snow, snow–ice,
characterization of the contributions from the different parts and ice–seawater interfaces, respectively. Moreover, it can
of the medium can be done in many different ways. For be clearly seen from the plots that, at a given incidence
example, the total contribution from each of the snow and angle, the major contribution is made by the ice cover
ice covers (surface plus volume contributions) can be plotted (represented by the ice volume and the top surface of the
as a function of incidence angle or ground range. Fig. 17(a) ice layer). However, it is important to note that for a proper
shows such a plot, which shows the sum of all contributions comparison, absolute radiometric calibration of the systems is
from the snow and the ice layers as a function of incidence vital.
angle. The intensity values are plotted from the corrected
tomogram of Fig. 9(a). As it can be clearly seen from the C. Comparison Between VV and HV Tomograms
figure, the sum of all signal contributions from the ice layer is A comparison of the vertical structure of the reflectivity
superior compared with the one coming from the snow layer. of the medium at VV and HV polarizations using another
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 15

Fig. 18. Comparison of VV and HV tomographic data sets. Multilooked corrected tomograms at VV (left) and HV (right) based on 15 acquisitions.
(a) VV tomogram. (b) HV tomogram. (c) Corresponding intensity plots for VV tomogram. (d) Corresponding intensity plots for HV tomogram.

not surprising as cross-polarized backscatter from relatively


smooth surfaces is usually much lower than the correspond-
ing copolarized backscatter [41], [52]. Moreover, the low
HV response from snow compared with the ice indicates that
the snow grains are relatively spherical in shape, whereas the
air bubbles have irregular, randomly oriented shapes, as it was
observed during data collection [see Fig. 3(b)]. Comparing the
intensity tomograms of Fig. 18(a) and (b), it can be noted that
the HV response from the considered medium is lower than
that of the VV response. Moreover, the relatively strong ice–
seawater interface response of HV compared with other parts
of the medium may be associated with the oriented facets
observed during data collection at the bottom surface of the
ice layer. This can also be clearly seen from the bottom surface
of the ice cores (see Fig. 19). Another interesting observation
that can be made from the HV tomogram is the uniform spread
of power in the sea ice layer. This is associated with the
depolarizing effect of the randomly distributed air bubbles,
and it is an indication that they have a random orientation.
Fig. 19. Facets at the bottom surface of the ice layer.
V. C ONCLUSION
set of measurements is shown in Fig. 18. The left column In this paper, the vertical structure of the radar reflectivity of
[Fig. 18(a) and (c)] is obtained by processing 15 acquisitions a snow covered sea ice medium is investigated. The test site
at VV, and the right column [Fig. 18(b) and (d)] corresponds under consideration was Kattfjord, Tromsø, Norway. A 3-D
to the same number of acquisitions at HV polarization. The GB-SAR system was used to acquire multiple SAR images
combined 30 measurements were taken at the same time. from slightly different elevation positions. The interfaces
The three interfaces can easily be identified from the between the different layers of the considered medium were
VV tomogram. However, the HV tomogram is characterized easily identified from the tomograms, which were displayed as
by very weak response from the snow layer. This result is a height-ground range sections. It is found that the strongest
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

16 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING

backscatter comes from the sea ice layer. Such information [10] B. T. Saenz and K. R. Arrigo, “Simulation of a sea ice ecosystem using
is vital for EM-scattering modelers as it reveals information a hybrid model for slush layer desalination,” J. Geophys. Res., Oceans,
vol. 117, no. C05007, May 2012, doi: 10.1029/2011JC007544.
about the scattering power distribution as a function of depth. [11] T. G. Yitayew, L. Ferro-Famil, and T. Eltoft, “3-D imaging of sea
The effect of a change in propagation velocity of the wave ice using ground-based tomographic SAR data and comparison of
inside the considered medium is investigated in the focusing the measurements with TerraSAR-X data,” in Proc. IEEE Int. Geosci.
Remote Sens. Symp. (IGARSS), Jul. 2014, pp. 1329–1332.
process, and the tomograms are effectively corrected for this [12] A. Reigber and A. Moreira, “First demonstration of airborne SAR
effect. This propagation velocity difference is identified as tomography using multibaseline L-band data,” IEEE Trans. Geosci.
the main source of distortion of the 3-D information, and Remote Sens., vol. 38, no. 5, pp. 2142–2152, Sep. 2000.
[13] Y. Huang, L. Ferro-Famil, and M. Neumann, “Tropical forest struc-
this might hinder the correct interpretation of the tomograms. ture estimation using polarimetric SAR tomography at P-band,” in
Therefore, this phenomenon should be considered in focusing Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), Jul. 2012,
tomographic SAR data and interpreting tomograms of semi- pp. 7593–7596.
[14] S. Tebaldini and F. Rocca, “Multibaseline polarimetric SAR tomography
transparent volumetric media, such as snow and sea ice. Our of a boreal forest at P- and L-bands,” IEEE Trans. Geosci. Remote Sens.,
results from comparing VV and HV measurements reveal that, vol. 50, no. 1, pp. 232–246, Jan. 2012.
even if the response at HV is weaker than the response at VV, [15] Y. Huang and L. Ferro-Famil, “3-D characterization of buildings in a
dense urban environment using L-band Pol-InSAR data with irregular
most of the HV responses are associated with the bottom part baselines,” in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS),
of the medium, mainly the ice and the ice–seawater interface. vol. 3. Jul. 2009, pp. III-29–III-32.
Using a simple technique, we have estimated the refractive [16] D. Reale, G. Fornaro, A. Pauciullo, X. Zhu, and R. Bamler, “Tomo-
graphic imaging and monitoring of buildings with very high resolution
indices of snow and sea ice from the 3-D information of a SAR data,” IEEE Geosci. Remote Sens. Lett., vol. 8, no. 4, pp. 661–665,
single polarization measurement, and the results are compared Jul. 2011.
with values from the previous experimental studies. Moreover, [17] S. Tebaldini and L. Ferro-Famil, “High resolution three-dimensional
imaging of a snowpack from ground-based SAR data acquired at
by plotting the scattering contribution of the different parts of X and Ku band,” in Proc. IGARSS, Jul. 2013, pp. 77–80.
the considered medium as a function of depth and incidence [18] T. G. Yitayew, L. Ferro-Famil, and T. Eltoft, “High resolution three-
angle, we have demonstrated the idea that such an information dimensional imaging of sea ice using ground-based tomographic SAR
data,” in Proc. 10th Eur. Conf. Synth. Aperture Radar (EUSAR),
can be useful to compare and validate measurements acquired Jun. 2014, pp. 1–4.
by other 2-D SAR sensors operating at the same sensor [19] K. Morrison and J. Bennett, “Tomographic profiling—A technique
parameters. Finally, we believe that the results presented in for multi-incidence-angle retrieval of the vertical SAR backscattering
profiles of biogeophysical targets,” IEEE Trans. Geosci. Remote Sens.,
this paper highlight the immense potential of the 3-D imaging vol. 52, no. 2, pp. 1350–1355, Feb. 2014.
and SAR tomography for simplifying EM scattering models as [20] L. Ferro-Famil, D. Cristallini, D. Pastina, and P. Lombardo, “Improving
it reveals the dominant sources of information and the prepon- SAR tomography performance using efficient sensor configurations,”
in Proc. IEEE Int. Geosci. Remote Sens. Symp. (IGARSS), Jul. 2011,
derant scattering mechanisms in a semitransparent medium, pp. 4225–4228.
such as snow and sea ice. [21] F. Gini, F. Lombardini, and M. Montanari, “Layover solution in multi-
baseline SAR interferometry,” IEEE Trans. Aerosp. Electron. Syst.,
ACKNOWLEDGMENT vol. 38, no. 4, pp. 1344–1356, Oct. 2002.
[22] Y. Huang, “Tomographic processing of polarimetric and interferometric
The authors would like to thank C. Leconte and F. Boutet, SAR data for urban and forestry remote sensing,” Ph.D. dissertation,
who were part of the team during the data collection process. IETR, Univ. Rennes 1, Rennes, France, 2011.
[23] L. M. H. Ulander, H. Hellsten, and G. Stenstrom, “Synthetic-aperture
R EFERENCES radar processing using fast factorized back-projection,” IEEE Trans.
Aerosp. Electron. Syst., vol. 39, no. 3, pp. 760–776, Jul. 2003.
[1] M. Nicolaus, C. Petrich, S. R. Hudson, and M. A. Granskog, “Variability [24] L. A. Gorham and L. J. Moore, “SAR image formation toolbox for
of light transmission through arctic land-fast sea ice during spring,” MATLAB,” Proc. SPIE, vol. 7699, p. 769906, Apr. 2010.
Cryosphere, vol. 7, pp. 977–986, Jun. 2013. [25] C. A. Balanis, Advanced Engineering Electromagnetics, vol. 20.
[2] W. F. Weeks and S. F. Ackley, “The growth, structure, and properties of New York, NY, USA: Wiley, 1989.
sea ice,” in The Geophysics of Sea Ice, N. Untersteiner, Ed. New York, [26] L. Ferro-Famil, S. Tebaldini, M. Davy, and F. Boute, “3D SAR imag-
NY, USA: Springer, 1986, pp. 9–64. ing of the snowpack at X- and Ku-band: Results from the AlpSAR
[3] R. G. Onstott, “SAR and scatterometer signatures of sea ice,” in campaign,” in Proc. 10th Eur. Conf. Synth. Aperture Radar (EUSAR),
Microwave Remote Sensing of Sea Ice, F. D. Carsey, Ed. Washington, Jun. 2014, pp. 1–4.
DC, USA: American Geophysical Union, 1992, pp. 73–104. [27] F. D. Carsey, Ed., Microwave Remote Sensing of Sea Ice. Washington,
[4] R. Kwok, E. Rignot, B. Holt, and R. Onstott, “Identification of sea ice DC, USA: AGU, 1992.
types in spaceborne synthetic aperture radar data,” J. Geophys. Res., [28] A. K. Fung and K. S. Chen, Microwave Scattering and Emission Models
Oceans, vol. 97, no. C2, pp. 2391–2402, 1992. for Users. Norwood, MA, USA: Artech House, 2010.
[5] K. Nakamura, H. Wakabayashi, K. Naoki, F. Nishio, T. Moriyama, and [29] M. T. Hallikainen, M. V. O. Toikka, and J. M. Hyyppa, “Microwave
S. Uratsuka, “Observation of sea-ice thickness in the sea of Okhotsk dielectric properties of low-salinity sea ice,” in Proc. Remote Sens., Mov-
by using dual-frequency and fully polarimetric airborne SAR (pi-SAR) ing Toward 21st Century, Int. Geosci. Remote Sens. Symp. (IGARSS),
data,” IEEE Trans. Geosci. Remote Sens., vol. 43, no. 11, pp. 2460–2469, vol. 1. Sep. 1988, pp. 419–420.
Nov. 2005. [30] M. R. Vant, R. B. Gray, R. O. Ramseier, and V. Makios, “Dielectric
[6] S. Hong, “Detection of small-scale roughness and refractive index of properties of fresh and sea ice at 10 and 35 GHz,” J. Appl. Phys., vol. 45,
sea ice in passive satellite microwave remote sensing,” Remote Sens. no. 11, pp. 4712–4717, 1974.
Environ., vol. 114, no. 5, pp. 1136–1140, May 2010. [31] T. H. Achammer and A. Denoth, “Snow dielectric properties: From DC
[7] G. A. Maykut and B. Light, “Refractive-index measurements in freez- to microwave X-band,” Ann. Glaciol., vol. 19, no. 1, pp. 92–96, 1994.
ing sea-ice and sodium chloride brines,” Appl. Opt., vol. 34, no. 6, [32] M. T. Hallikainen, F. Ulaby, and M. Abdelrazik, “Dielectric properties
pp. 950–961, 1995. of snow in the 3 to 37 GHz range,” IEEE Trans. Antennas Propag.,
[8] M. N. O. Sadiku, “Refractive index of snow at microwave frequencies,” vol. 34, no. 11, pp. 1329–1340, Nov. 1986.
Appl. Opt., vol. 24, no. 4, pp. 572–575, 1985. [33] W. A. Cumming, “The dielectric properties of ice and snow at
[9] H. Cattle, J. Crossley, and D. J. Drewry, “Modelling arctic climate 3.2 centimeters,” J. Appl. Phys., vol. 23, no. 7, pp. 768–773, 1952.
change,” Philos. Trans. R. Soc. London A, Math. Phys. Sci., vol. 352, [34] C. Mátzler, “Microwave permittivity of dry snow,” IEEE Trans. Geosci.
no. 1699, pp. 201–213, 1995. Remote Sens., vol. 34, no. 2, pp. 573–581, Mar. 1996.
This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

YITAYEW et al.: TOMOGRAPHIC IMAGING OF FJORD ICE USING A VHR GB-SAR SYSTEM 17

[35] F. Ulaby and D. Long, Microwave Radar and Radiometric Remote Laurent Ferro-Famil (S’99–A’00–M’04) received
Sensing. Ann Arbor, MI, USA: Univ. of Michigan Press, 2014. the M.S. degree in electronic systems and computer
[36] B. Rekioua, M. Davy, L. Ferro-Famil, and S. Tebaldini, “Snow- engineering and the Ph.D. degree from the Univer-
pack permittivity profile retrieval from tomographic SAR data,” sity of Nantes, Nantes, France, in 1996 and 2000,
Comp. Rendus Phys., 2016. [Online]. Available: http://dx.doi.org/ respectively.
10.1016/j.crhy.2015.12.016 In 2001, he became an Associate Professor
[37] E. Kärkäs, T. Martma, and E. Sonninen, “Physical properties and stratig- with the University of Rennes 1, Rennes, France.
raphy of surface snow in western Dronning Maud Land, Antarctica,” Since 2011, he has been a Full Professor with the
Polar Res., vol. 24, nos. 1–2, pp. 55–67, Jul. 2005. University of Rennes 1, where he is currently the
[38] J. Fortuny-Guasch, “A novel 3-D subsurface radar imaging technique,” Head of the Remote Sensing Department, Institute
IEEE Trans. Geosci. Remote Sens., vol. 40, no. 2, pp. 443–452, of Electronics and Telecommunications of Rennes.
Feb. 2002. His current research interests include analog electronics, digital communi-
[39] K. W. Birkeland, “Terminology and predominant processes associated cations, microwave theory, signal processing, and polarimetric SAR remote
with the formation of weak layers of near-surface faceted crystals in the sensing, polarimetric SAR signal statistical processing, radar polarimetry
mountain snowpack,” Arctic Alpine Res., vol. 30, no. 2, pp. 193–199, theory, and natural media remote sensing using multibaseline PolInSAR data,
1998. with application to classification, electromagnetic scattering modeling and
[40] A. Langlois and D. G. Barber, “Passive microwave remote sensing of physical parameter retrieval, time-frequency analysis, and 3-D reconstruction
seasonal snow-covered sea ice,” Prog. Phys. Geogr., vol. 31, no. 6, of environments using SAR tomography.
pp. 539–573, Dec. 2007.
[41] Y.-S. Kim, R. Moore, and R. Onstott, “Theoretical and experimental
study of radar backscatter from sea ice,” Univ. of Kansas Remote
Sensing Lab., Tech. Rep. CRINC/RSL-TR-331-37, 1984.
[42] M. R. Drinkwater and G. Crocker, “Modelling changes in scattering
properties of the dielectric and young snow-covered sea ice at GHz
requencies,” J. Glaciol., vol. 34, no. 118, pp. 274–282, 1988.
[43] D. P. Winebrenner, L. Tsang, B. Wen, and R. West, “Sea-ice character-
ization measurements needed for testing of microwave remote sensing Torbjørn Eltoft (M’92) received the M.S. and
models,” IEEE J. Ocean. Eng., vol. 14, no. 2, pp. 149–158, Apr. 1989. Ph.D. degrees from the University of Tromsø-The
[44] J. Shi and J. Dozier, “Estimation of snow water equivalence using Arctic University of Norway, Tromsø, Norway,
SIR-C/X-SAR. I. Inferring snow density and subsurface properties,” in 1981 and 1984, respectively.
IEEE Trans. Geosci. Remote Sens., vol. 38, no. 6, pp. 2465–2474, In 1988, he joined the Faculty of Science and
Nov. 2000. Technology, University of Tromsø-The Arctic Uni-
[45] A. Martini, J.-P. Dedieu, L. Ferro-Famil, and É. Pottier, “Dry snow versity of Norway, where he was the Head of
discrimination in a high mountainous environment (French Alps) using the Department of Physics and Technology from
multi-frequency and multi-temporal polarimetric SAR data,” Télédétec- 2013 to 2015. He is currently leading CIRFA, a new
tion, Contemp. Pub. Int., vol. 6, no. 1, pp. 45–55, 2006. Center for Research-based Innovation, University of
[46] X. V. Phan et al., “1D-Var multilayer assimilation of X-band SAR Tromsø-The Arctic University of Norway. He is also
data into a detailed snowpack model,” Cryosphere, vol. 8, no. 5, a Professor of electrical engineering and an Adjunct Professor with the
pp. 1975–1987, 1975. Northern Research Institute (Norut), Tromsø. He has supervised 15 Ph.D. and
[47] S. Khwaja, “Fast raw data generation of realistic environments for a SAR numerous master’s students. He has a significant publication record in the area
system simulator,” Ph.D. dissertation, Dept. Eng. Sci., Univ. Rennes 1, of research signal processing and remote sensing. His current research interests
Rennes, France, 2008. include multidimensional signal and image analysis with application in radar
[48] F. T. Ulaby, F. Kouyate, A. K. Fung, and A. J. Sieber, “A backscatter remote sensing, statistical models, neural networks, and machine learning.
model for a randomly perturbed periodic surface,” IEEE Trans. Geosci. Dr. Eltoft was a recipient of the 2000 Outstanding Paper Award in Neural
Remote Sens., vol. GE-20, no. 4, pp. 518–528, Oct. 1982. Networks awarded by the IEEE Neural Networks Council and the Honourable
[49] S. Van Nghiem, “Electromagnetic wave models for polarimetric remote Mention for the 2003 Pattern Recognition Journal Best Paper Award.
sensing of geophysical media,” Ph.D. dissertation, Dept. Elect. Eng.
Comput. Sci., Massachusetts Inst. Technol., Cambridge, MA, USA,
1991.
[50] J.-S. Lee and E. Pottier, Polarimetric Radar Imaging: From Basics to
Applications. Boca Raton, FL, USA: CRC Press, 2009.
[51] M. Eineder, T. Fritz, J. Mittermayer, A. Roth, E. Boerner, and H. Breit,
“TerraSAR-X ground segment, basic product specification document,”
German Aerospace Center, Tech. Rep. TX-GS-DD-3302, 2008.
[52] F. T. Ulaby, R. K. Moore, and A. K. Fung, Microwave Remote Sensing: Stefano Tebaldini (M’06) received the M.S. degree
Active and Passive: Radar Remote Sensing and Surface Scattering and in telecommunication engineering and the Ph.D.
Emission Theory, vol. 2. Reading, MA, USA: Addison-Wesley, 1982, degree from the Politecnico di Milano, Milan, Italy,
pp. 457–1064. in 2005 and 2009, respectively.
Since 2005, he has been with the Digital
Signal Processing Research Group, Politecnico di
Milano, where he has been a Permanent Researcher
since 2011. He is with the Politecnico di Milano,
where he teaches courses on signal theory and
remote sensing. He has been involved as a Key
Temesgen Gebrie Yitayew (S’12) received the
Scientist in several studies by the European Space
master’s degree in physics from the, University of
Agency (ESA) concerning the tomographic phase of Biomass. He is one of the
Tromsø-The Arctic University of Norway, Tromsø,
inventors of a new technology patented by Tele-Rilevamento Europa (T.R.E.)
Norway, in 2012, where he is currently pursuing the
for the exploitation of multiple interferograms in the presence of distributed
Ph.D. degree.
scattering. He also collaborates with the SAPHIR Team, University of Rennes
His research interests include remote sensing of
1, Rennes, France. His current research interests include earth observation with
sea ice, synthetic aperture radar (SAR) polarimetry,
synthetic aperture radar and radar design and processing.
SAR interferometry, and SAR tomography applied
Dr. Tebaldini is a member of the SAOCOM-CS Expert Group at ESA.
for 3-D imaging and parameter retrieval of volumet-
He co-authored the Best Poster Paper Award and the Second Best Student
ric media.
Paper Award at the European Conference on Synthetic Aperture Radar 2012.

You might also like