FEATURE

SAR TOMOGRAPHY: AN ADVANCED TOOL

FOR 4D SPACEBORNE RADAR SCANNING
WITH APPLICATION TO IMAGING AND
MONITORING OF CITIES AND SINGLE BUILDINGS
G. Fornaro1, Senior Member IEEE, A. Pauciullo1, D. Reale1 Member IEEE,
X. Zhu2,3 Member IEEE, and R. Bamler2,3 Fellow IEEE
1
Institute for the Electromagnetic Sensing of the Environment (IREA), National Research Council
(CNR) 80124 Napoli, Italy (e-mail: [email protected], [email protected], [email protected]).
2
Remote Sensing Technology Institute (IMF), German Aerospace Center (DLR),
82234 Oberpfaffenhofen ([email protected], [email protected]).
3
Technische Universität München, Lehrstuhl für Methodik der Fernerkundung,
80333 Munich, Germany.

1. Introduction elevation direction by exploiting multiple passes of the radar

Synthetic Aperture Radar (SAR) has become a routine geo- at slightly different orbit positions to establish a virtual array
information source complementing optical imagery. The of antennas, as depicted in Figure 1. The synthetic aperture in
microwaves used by SAR can penetrate canopy, soil, snow elevation allows reducing the width of the elevation antenna
and ice allowing for estimation of volumetric properties, e.g. beam providing a fine beam “radar scanner” from the space
biomass. SAR also can employ polarization for deriving able to generate high resolution 3D images, hence the addi-
structural parameters of objects [1] [2]. Another strong and tional name of 3D Imaging.
unique selling point of SAR is that surface displacements and The inherent (Rayleigh) elevation resolution ts of the to-
object deformations can be measured to mm-level accuracy mographic arrangement is related to the spread Tb of this ar-
by exploiting multitemporal acquisitions. In particular, the ray [3]–[6]:
recent technological advancement in SAR systems has pro-
mr
vided very high resolution (VHR) X-Band sensors, such as ts = (1)
2Tb
the TerraSAR-X/TanDEM-X mission and the Cosmo-Skymed
constellation, characterized by spatial resolutions of the order and can reach values of the order of a few meters.
of 1 meter. With respect to the former generation of medium By stacking all the multiview coherent images and by
resolution (a ten, or a few tens of meters) C-Band SAR sys- performing the tomographic processing, s-profiles can be
tems, the increase of the resolution of such sensors allows
much higher details of ground structures to be captured.
On the downside we have to cope with an imaging geom- b
etry that does not support an easy interpretation of SAR im-
ages, especially in complex scenarios. A single SAR image is Δb
the projection of the 3D scene into the azimuth (x) – range (r)
coordinates. Whenever only surface scattering is present and
the local slopes of the surfaces are smaller than the local inci-
dence angle this mapping is injective: in this case the image Elevation
can be easily converted to any other map projection. However, Aperture
for volumetric scatterers (e.g. forests) and steep – or even ver-
tical – surfaces, like in urban areas, SAR imaging becomes Δs s
non-injective and real 3D imaging is required. z 3-D Reflectivity
SAR Tomography (TomoSAR) is a technique that allows Distribution
x y γ (x, r, s)
resolving scatterer densities in the third native radar co-ordi- =0
urface s
nate “elevation (s)” (also referred to as slant–height, orthogo- Reference S r
nal to the azimuth-range plane). It extends the synthetic aper-
ture principle – as used in the azimuth direction – also to the Figure 1. TomoSAR geometry

IEEE Geoscience and Remote Sensing Society Newsletter • December 2012 9

retrieved for every x–r pixel. These profiles can be continuous g = Rc + f (3)
in the case of forest biomass imaging [3] or may consist only
of a few discrete scatterers, typically corresponding to scatter- where g is the vector of measurements according to eq. (2), c
ers located on the ground, facade and roof, in the case of urban is the elevation profile, R is the irregular Fourier matrix com-
mapping. This article is devoted to the latter application. posed of the so-called steering vectors and f is noise.
Since the elevation antenna array is in the far-field of the The spectral estimation problem of s-profile reconstruction
imaged objects, the complex signal received at any of the ra- can be framed as the inversion of this linear system. The inver-
dar positions bn is a sample of the Fourier transform of the sion must be carefully implemented because: (i) the Fourier
reflectivity profile in elevation c (s): samples are irregularly spaced at pn, (ii) their number N may
be small, (iii) the SNR may be low for the majority of the pix-
gn = # c (s) exp (- j2rp n s) ds, n = 1 , f, N (2) els, (iv) the data may contain non-Gaussian phase noise due
Ts to uncompensated atmospheric delay and unmodeled motion
where p n = -2b n / (mr) is the spatial (elevation) frequency. and (v) the orbit tube of modern SAR satellites is tight leading
Therefore, retrieval of the s-profile is framed as a spectral to a small Tb and, hence, to a low elevation resolution. Many
estimation problem and reliable scatterers showing a good different MDI algorithms have been proposed in the recent
degree of coherence can be identified by looking for the literature to cope with these problems.
peaks in the focused reflectivity function. The simplest algorithm is based on the Beam-Forming
Since the multipass dataset is acquired at different time in- (BF) that is the matched filter, i.e., ct = R H g . It computes
stants, sometimes over a period of years, possible motion and the amount of backscattered energy at different elevations by
deformation of objects must be additionally considered in the digitally steering (through the column vectors of R) the beam
process of estimation of the s-profiles, either as useful infor- of the multibaseline array. Because of the irregular acquisition
mation (subsidence, tectonics, landslides, etc.) or simply as distribution BF reconstruction exhibits poor performances in
nuisance parameters. Space/velocity (4D) imaging techniques, terms of large sidelobes and also it does not allow exceeding
also known as Differential SAR Tomography (D-TomoSAR), the Rayleigh resolution of eq. (1) [5].
extends the 3D imaging and can be applied to measure also the To overcome such limitations, advanced inversion ap-
deformation parameters (velocity spectrum) of any temporal proaches have been proposed in the literature. The need to
coherent scatterer in the focused 3D space [7][8][6]. If motion achieve super-resolution involves a discretization that exceeds
is considered, eq. is in fact extended to a 2D or even higher the Rayleigh limit thus making the problem in (3) underde-
dimensional Fourier transform, depending on how many mo- termined. A class of super-resolution TomoSAR algorithms is
tion modes are accounted for (e.g. linear, periodic, thermal, based on regularized inversion and tries to find the solution
etc.) [7]–[10]. In this case the technique is referred to as Multi- among the infinitely many solutions of the underdetermined
Dimensional (MD) SAR imaging. Finally, the possibility of system model by minimizing:
screening the reflectivity function in elevation allows discrim-
inating the presence of multiple peaks, even exhibiting differ- t
c = arg min " < g - Rc <22 + b < Cc < qp , (4)
c
ent velocities, and hence solve the interference and increase
the density of monitored scatterers [8][11][6]. A rather simple and easily to be implemented solution is
MD imaging (MDI) is in effect a technique that extends the based on the use of the Singular Value Decomposition (SVD)
Persistent Scatterer Interferometry (PSI) [12]–[15] approach. regularized linear inversion [5][6]. In this case, Tikhonov
PSI assumes the presence of a dominant scattering mecha- (Wiener) filtering choices of the singular values allow achiev-
nism in each pixel and therefore cannot resolve the layover ing the solution to the problem in (4) with b = SNR , C = I
problem. Moreover, theoretical and experimental results on and p = 2 and q = 2 [6]. Another solution strictly related to
both simulated and real data have shown [16] that the use of the above approach is based on Truncated SVD: in this case
an imaging approach (i.e. SAR Tomography), which exploits b = 0 and a hard limitation of the singular values is used [5]
the phase as well as the amplitude information, performs bet- to control the ill-conditioning nature of the inversion. Super-
ter even in the detection of dominant persistent scatterers and resolution can be achieved by reducing the scene support with
in the estimation of their localization and deformation param- respect to the theoretical limit given by Ts = mr/ (2bu ) , where
eters with respect to classical PSI which uses only the phase bu is the average baseline separation, and controlling the inver-
information. sion by choosing a suitable number of singular values during
the inversion process, provided that the noise level is suffi-
2. Tomographic SAR Inversion Algorithms ciently low. SVD achieves typically also better sidelobe sup-
In the following for sake of simplicity we refer to the 3D pression than plain BF [5][6].
reconstruction case. Discretizing the elevation profile in eq. For many acquisition configurations SVD super-resolution
leads to this standard linear system equation: is not sufficient. The orbits of TerraSAR-X, e.g., are controlled

10 IEEE Geoscience and Remote Sensing Society Newsletter • December 2012

so accurately that Tb is typically in the order of 250–350 m ers dramatically compared to PSI. As a consequence, precise
leading to an elevation resolution of 30–50 m, which is unac- monitoring of even single buildings and in general of infra-
ceptably large compared to the range and azimuth resolutions structures as well as cultural heritage is possible. In the fol-
of 1–3 m. The classical super-resolving spectral estimators that lowing, experiments on X-Band data acquired by both the
are used for TomoSAR are adaptive beam-forming (CAPON) TerraSAR-X and Cosmo-Skymed sensors are discussed.
[17] (non-linear, non-parametric) or MUSIC and ESPRIT (both Processing is carried out with implementation of MDI at
non-linear, parametric) [18][19]. The latter parametric methods IREA-CNR and DLR-IMF.
do not retrieve continuous s-profiles but rather estimate the po-
sitions of individual scatterers. They need the number of scat- 3.1. MDI Systems of IREA-CNR and DLR-IMF
terers as prior information. These methods are computationally
fast. However, they require the estimation of covariance ma- 3.1.1. MDI system of IREA-CNR
trices which is usually done by multilooking and reduces the The MDI approach developed in the last few years at IREA-
azimuth and range resolution. In the case of two or more scat- CNR exploits a simple tomographic processing based on
terers these estimators are not efficient, i.e. they do not reach beam-forming to estimate the backscattering distribution in
the Cramér-Rao Lower Bound. They are also not energy con- the elevation/velocity domain (4D MDI) and to identify scat-
serving and the strength of the estimated spectral lines are not terers (up to two), possibly interfering within the same image
straightforwardly related to the reflectivity of the scatterer. pixel [25]. The algorithm separates the problem in two steps:
The optimum parametric method – under Gaussian noise the first step is devoted to estimate the scatterers parameters,
assumption – is the non-linear least-squares estimator. How- i.e. the position in space and deformation, whereas the second
ever, it would require a combinatorial search of scatterer posi- step concerns the selection of the number of scatterers in a
tions. Here the theory of Compressive Sensing (CS) comes detection framework, i.e., paying attention to achieve high
into play. detection performance for a given false alarm rate. Particularly,
CS is able to reconstruct sparse signals from their irregu- the selection stage exploits a detection scheme which is based
larly sampled Fourier transform in a quasi-parametric way. on the sequential use of a detector based on the Generalized
Indeed the elevation profiles of urban objects usually contain Likelihood Ratio Test for single scatterers in [13][16] .
only a few scatterers, e.g. one at the ground and one on the The MDI technique developed in IREA has demonstrated,
façade. Since the elevation resolution ts from eq. (1) is often for the first time, the capability of this processing approach
much worse than the range resolution, the elevation extent of to resolve multiple scatterers interfering in the same pixel
these scatterers is much smaller than ts , rendering these scat- by using C-Band data in [26]: further significant results on
terers discrete. These are the very prerequisites for using the C-Band data were reported in [8] and [11]. Nevertheless, the
theory of sparse signal reconstruction and CS. By referring to first demonstration of the capability of MDI to resolve dis-
(4), CS allows finding the solution by selecting b = b K , C = I tributed layover occurring over vertical structures as buildings
and p = q = 1, where K is the number of sparse targets [20]. have been shown in [32]: in this case the MDI beam-forming
The first CS TomoSAR simulations were presented in [21] and tomographic processing was applied to a dataset of TerraSAR-
the SR capability of CS for TomoSAR reconstruction and its X data acquired over the area of Las Vegas. Particularly rel-
robustness on elevation estimation against phase noise have evant were in this case the results achieved over the Mirage
been proven in [22]. To overcome the drawbacks of a simple hotel which are not shown here for brevity.
CS estimator, the “Scale-down by L1 norm Minimization, In order to face the higher sensitivity of X-Band radiation
Model selection, and Estimation Reconstruction” (SL1M- to small changes of targets, such as those caused by thermal
MER) algorithm has been proposed in [23][24], a spectral esti- dilation, MDI processing has been extended to measure even
mation algorithm based on CS, with an additional model order this component in the 5D domain (space/velocity + thermal
selection and final maximum likelihood parameter estimation. dilation). In fact, even if already observed in C-Band (ERS
As a last remark it is important to point out that not only and Envisat) data [28][29], this effect is much more evident
estimation of motion parameters (e.g., velocity, topography, in X-Band data due to the increase of sensitivity associated
etc.) associated with scatterers interfering in the same pixel with the shorter wavelength [30]–[32]. To account also for the
can be achieved, but also separation of time series is possible thermal dilation component, the deformation model, typically
by adoption of proper tomographic based filtering techniques. made of the linear temporal term, is extended with a new one,
Results of experiments with real data have confirmed this pe- measured in mm/K, which is linearly related to the tempera-
culiarity of MDI data processing [11]. ture of the area at the acquisition instants [10].

3. Application Examples 3.1.2. MDI system of DLR-IMF: Tomo-GENESIS

The advanced processing via MDI of VHR X-Band data The workhorse of interferometric processing at DLR is the
allows nowadays increasing the density of monitored scatter- GENESIS system [33]. It has been the basis for the

IEEE Geoscience and Remote Sensing Society Newsletter • December 2012 11

 makes super-resolving TomoSAR algorithms particularly im-
portant for urban mapping.
Super-resolution factor: κ50% = ρs/ρ50%

a1/a2 = 1 The SL1MMER algorithm is demonstrated to be an ef-

a1/a2 = 1.5 #(.)dΔϕ ficient estimator and achieves super-resolution factors of
a1/a2 = 2
1.5~25 at the interesting parameter range for TomoSAR (see
101 a1/a2 = 2.5 SR
Figure 2), i.e. N = 10~100 and SNR = 0~10 dB [22][23].
a1/a2 = 3 factors:
The results shown in Figure 3 are approximately applicable
1.5–25 to nonlinear least-squares estimation as well, and hence, al-
though it is derived experimentally, they can be considered
as a fundamental bound for SR of spectral estimators. In [24]
the super-resolution capability of the SL1MMER algorithm is
100 demonstrated using TerraSAR-X data.
10 15 20 25 30 For an input data stack, the Tomo-GENESIS system re-
N*SNR [dB] trieves the following information: number of scatterers inside
a pixel, amplitude and phase, topography and motion parame-
Figure 2. Fundamental bound of super-resolution (SR): SR factor of ters (e.g. linear deformation velocity and amplitude of thermal
the SL1MMER algorithm as a function of N SNR under different dilation induced seasonal motion) of each detected scatterer.
amplitude ratios a1/a2 of two close scatterers [24]. Compared to other existing MDI processing systems, it has
the following new features: the time warp method for multi-
component nonlinear motion estimation [9]; the CS based
developments of DLR’s operational SRTM and TanDEM-X SL1MMER algorithm and super-resolution capability [22]–
processors. An extended version of it (PSI-GENESIS) han- [24]; fusion of PSI and TomoSAR processing for operational
dles PSI processing of medium resolution data, wide-swath purpose [35]; RANSAC based point cloud fusion algorithm
mosaics as well as very high resolution spotlight data. During [36][35]. Currently, the system is extending for object recon-
the last years several new algorithms for TomoSAR process- struction from the unstructured TomoSAR point clouds [37].
ing have been developed at DLR extending the system to what
we introduce as “Tomo-GENESIS” [34]. 3.2. Application Examples
The layover phenomenon in a SAR image of an urban area
is mainly caused by the following two scenarios: (i) buildings 3.2.1. Space radar scanning with SAR tomography
with different heights in layover with the ground or (ii) taller To show the capabilities of MDI to achieve “synthetic” radar
building in layover with the ground and the roof of a lower scanning for imaging ground structures we present the fol-
building. Both scenarios suggest that double scatterer pairs lowing results relevant to the San Paolo Stadium in the city of
with smaller elevation distances will be more frequent than Naples, Italy. A dataset of 28 images acquired by the Cosmo-
those with larger distances. Therefore, SR is crucial for VHR Skymed constellation on descending passes in the standard
tomographic SAR reconstruction in urban environment. This stripmap mode (~3 m spatial resolution) from February 2010
to February 2011 were processed with the 5D MDI algorithm.
Figure 3 shows the three-dimensional reconstruction of the
stadium visualized on a Google Earth map: the colors are set
according to the estimated height. Note that even in this mod-
erate resolution mode a large density of targets compared to
PSI is achieved, especially on the roof [30]–[32].

3.2.2. Compressed Sensing SAR Tomography for

super resolution spaceborne radar scanners
Figure 4 presents a 3D view of the single buildings visual-
ized in GoogleEarth reconstructed by SL1MMER using a
stack of 25 TerraSAR-X images [24]. The test building is the
Bellagio hotel at downtown Las Vegas. Compared to PSI,
TomoSAR offers in general a tremendous improvement in
Figure 3. 3D view of the San Paolo Stadium, Naples, Italy, recon- detailed reconstruction and monitoring of urban areas.
structed by the 5D MDI with COSMO/SKYMED data provided by Experiments using TerraSAR-X high resolution spotlight
the Italian Space Agency. Colormap is set according to the estimated data stacks show the scatterer density to be in the order of
height [30]–[32]. 600,000~1,000,000/km2 compared to a PS density in the

12 IEEE Geoscience and Remote Sensing Society Newsletter • December 2012

order of 40,000~100,000 PS/km2 [34][38]. In particular,
together with its SR power, SL1MMER provides ultimate
information one can retrieve from the data stack. Figure 5
presents the number of scatterers map obtained by SVD-
Wiener inversion (left) and SL1MMER (right) over the test
area where blue indicates zero scatterers inside the azimuth-
range pixel, green stands for one and red for two. Non-
parametric estimators can only detect two scatterers with an
elevation distance larger than approximately the Rayleigh
elevation resolution unit ts. Therefore, it is not surprising
that the double scatterers detected by the linear estimator are
mainly located on the upper part of the building façade. The
result of SL1MMER shows a much denser red color which Figure 4. 3D view of the single building visualized in GoogleEarth
indicates a larger amount of detected double scatterers. For reconstructed by SL1MMER using a stack of 25 TerraSAR-X images
an urban area like this typically about 30~40% of the scat- (the color represents height) [24].
terers detected by SL1MMER are double scatterers com-
pared to less 10~20% detected by linear estimators.
orbit and another stack of 79 SAR images from descending
3.2.3. TomoSAR point cloud fusion orbit are processed. For both stacks, the acquisition time span
Due to the side-looking geometry of SAR, a single stack of is about four years. Figure 6 present the 3D positions of the
SAR images only provides information on one side of a build- fused point clouds. From Figure 6 Figure 6: small structures
ing. To serve the function of urban structure monitoring, like the Victory Column, i.e. the statue at the center of the park
fusion of the TomoSAR results of multiple stacks from differ- can be easily identified. For this test-site, about 40 million
ent view angles can provide us with a shadow-free point cloud scatterers are detected from the two data stacks.
with high degree of coverage over the entire urban area.
At the test site of Berlin, we have a luxury data archive 3.2.4. Multicomponent motion estimation
with a large number of TerraSAR-X high resolution spotlight D-TomoSAR was originally proposed in [7] for estimating
images including a stack of 94 SAR images from ascending linear motion of multiple scatterers inside a pixel. Motion,

(a) (b)

Figure 5. Double scatterer density for SVD-Wiener (a) and SL1MMER (b) reconstruction (Blue: Null scatterers per pixel; Green: Single;
Red: Double) [24]. The superresolution capability of SL1MMER leads to more detected double scatterers, in particular at the lower parts
of the facades.

IEEE Geoscience and Remote Sensing Society Newsletter • December 2012 13

Figure 6. Fusion of two point clouds generated from TerraSAR-X data stacks of ascending and descending orbit. The color represents
height [35].

however, is often nonlinear (periodic, accelerating, stepwise, and casino (see Figure 4). Figure 8(a) shows in the native ra-
etc.). Conventional D-TomoSAR has been extended to esti- dar geometry (range and azimuth are the horizontal and verti-
mate multicomponent nonlinear motion in [9] by proposing cal directions, respectively), the topography retrieved by the
the generalized ”time warp” method. It rewrites the MDI Beam-Forming imaging approach by using only single
D-TomoSAR system model to an M+1-dimensional standard scatterers. Figure 8(b) and 8(c) show the estimated topogra-
spectral estimation problem, where M indicates the user phy and thermal dilation with single and double scatterers by
defined motion model order and, hence, enables the motion using the 5D Beam-Forming imaging: note that thermal dila-
estimation for all possible complex motion models. tion mostly affects the outer parts of building and exhibits dif-
Figure 7 shows an example of multicomponent motion esti- ferent behavior on the same building according to the different
mation. Since July 2009, the selected area over Las Vegas (see projection of thermal dilation along the radar line-of-sight.
Figure 7a) is undergoing a pronounced subsidence centered
at the convention center. Together with the thermal dilation
induced seasonal motion of the metallic building structure, the 4. Conclusion and Further Developments
selected area is characterized by a two-component nonlinear With reference to the current status of VHR tomographic
motion. Here, we choose the motion basis function as a sine SAR inversion presented in this article, the following conclu-
function with a period of one year for seasonal motion and sions can be drawn:
linear function for linear subsidence. And hence the motion • VHR tomographic SAR inversion is able to reconstruct
parameters to be estimated are amplitude of seasonal motion the shape and motion of individual buildings and entire
and linear deformation velocity. The final estimation results city areas.
of the generalized time-warp method are presented in Figure • Super-resolution is crucial and possible for VHR tomo-
7 including elevation estimates in meter (b), amplitude of sea- graphic SAR inversion for urban infrastructure.
sonal motion in millimeter (c) and the LOS linear deformation • TomoSAR reconstruction from multiple tracks enables
velocity in millimeter/year (d). us to reconstruct the complete structure of individual
If we know the temperature at the time of data acquisitions, buildings and to generate 3D point clouds of the illumi-
we can also accordingly choose the temperature as the basis nated area with a point density comparable to LiDAR
function to model thermal dilation induced deformation [10] • The motion or deformation of buildings is often nonlin-
[30][31]. In this case, the thermal coefficient is estimated that ear (periodic, accelerating, stepwise, etc.). This is par-
represents the strength of undergoing thermal dilation induced ticularly true with VHR SAR data. Multicomponent
deformation. An example is presented in Figure 8. The site nonlinear motion of multiple scatterers can be separated
under investigation is relevant to the area of the Bellagio hotel and further estimated by tomographic reconstruction.

14 IEEE Geoscience and Remote Sensing Society Newsletter • December 2012

 100

80

60

40

20

–20

–40

(a) (b)

15
0

10
–5

5
–10

0 –15

–5 –20

–10 –25

–15 –30

(c) (d)

Figure 7. TomoSAR estimates of the selected area in Las Vegas: TerraSAR-X intensity map (a); Elevation estimates (b; unit: m) amplitude of
seasonal motion (c; unit: mm) and linear deformation velocity (d; unit: mm/y) [9].

A few topics for further study are outlined which mainly acquire data pairs simultaneously and repeatedly in
concern 1) tomographic SAR reconstruction from mixed sin- time. The TanDEM-X data pairs are free of motion,
gle-pass/multipass data stacks 2) object reconstruction from atmosphere and temporal decorrelation and, hence,
TomoSAR point clouds. possess much higher data quality. The fusion of
• Tomographic SAR reconstruction from mixed single- TerraSAR-X and TanDEM-X data, i.e. adding a cou-
pass/repeat-pass data stacks: ple of TanDEM-X acquisition pairs to the TerraSAR-X
So far, the data used for spaceborne VHR tomograph- data stacks, can be used to improve the result of
ic SAR inversion are repeat-pass data stacks. With tomographic SAR inversion on the one hand, and to
TanDEM-X, for the first time there is a real multi- explore the limits of tomographic reconstruction on
antenna array system in space. It enables us to the other hand [39].

IEEE Geoscience and Remote Sensing Society Newsletter • December 2012 15

 –50 +120 –50 +120 –0.5 +0.5
meters meters mm/K
(a) (b) (c)

Figure 8. Thermal dilation analysis with TerraSAR-X data. (a) estimated topography with the 4D imaging (single scatterers); (b) esti-
mated topography with the 5D imaging (single and double); (c) estimated thermal coefficients with the 5D imaging (single and double)
[10][30][31].

References
[1] J.C. Curlander and R.McDonough, “Synthetic Aperture
1,050
Radar – System and Signal Processing”, New York, John
1,000
Wiley & Sons, Inc , 1991.
950 [2] G. Franceschetti and R. Lanari, “Synthetic Aperture
900 Radar Processing”, CRC Press, Boca Raton FL, 1999.
3
.998 [3] A. Reigber and A. Moreira, “First demonstration of air-
983
3.9 982
3.9 981
borne SAR tomography using multibaseline L-band data”,
3.9 81
× 106 3.99
6.6465 IEEE Trans. Geosci. Remote Sens., vol. 38, no. 5, pp. 2142–
81 6.646
3.99 98 6.6455 2152, Sep. 2000.
3.9
79 6.645 × 105 [4] F. Gini, F. Lombardini, and M. Montanari, “Layover solu-
3.99 7 9 6.6445
3 .99 9 tion in multibaseline SAR interferometry”, IEEE Trans. Aerosp.
E 97 6.644
3.9 8
97 6.6435 N Electron. Syst., vol. 38, no. 4, pp. 1344–1356, Oct. 2002.
3.9
[5] G. Fornaro, F. Serafino, and F. Soldovieri, “Three-
Figure 9. From TomoSAR point clouds to objects: reconstructed dimensional focusing with multipass SAR data”, IEEE Trans.
building façade of the test building Bellagio hotel in Las Vegas [37]. Geosci. Remote Sens., vol. 41, no. 3, pp. 507–517, Mar. 2003.
[6] X. Zhu and R. Bamler, “Very High Resolution Spaceborne
SAR Tomography in Urban Environment,” IEEE Trans.
• Object reconstruction from TomoSAR point clouds: Geosci. Remote Sens., vol. 48, no. 12, pp. 4296–4308, Dec.
These tomographic point clouds can be potentially 2010.
used for building façade reconstruction in urban envi- [7] F. Lombardini, “Differential tomography: A new frame-
ronment from space with some special considerations work for SAR interferometry,” IEEE Trans. Geosci. Remote
such as side-looking geometry, anisotropic estimation Sens., vol. 43, no. 1, pp. 37–44, Jan. 2005.
accuracy and decorrelation. Yet in order to provide a [8] G. Fornaro, D. Reale, and F. Serafino, “Four-dimensional
high quality spatio-temporal 4D city model, object SAR imaging for height estimation and monitoring of single
reconstruction from these TomoSAR point clouds is and double scatterers”, IEEE Trans. Geosci. Remote Sens.,
emergent. A 3D view of the reconstructed façades over vol. 47, no. 1, pp. 224–237, Jan. 2009.
a test building Bellagio hotel in Las Vegas (see Fig- [9] X. Zhu, and R. Bamler, “Let’s Do the Time Warp:
ure 4) using point clouds from multiple viewing angles, Multicomponent Nonlinear Motion Estimation in Differential
i.e. both ascending and descending orbits, is exempli- SAR Tomography,” IEEE Geosci. Remote Sens. Lett., vol. 8,
fied in Figure 9. no. 4, pp. 735–739, July 2011.

16 IEEE Geoscience and Remote Sensing Society Newsletter • December 2012

[10] D. Reale, G. Fornaro, and A. Pauciullo, Extension of 4D [26] G. Fornaro, F. Serafino, “Imaging of Single and Double
SAR Imaging to the Monitoring of Thermally Dilating Scatterers in Urban Areas via SAR Tomography”, IEEE
Scatterers, IEEE Trans. Geosci. Remote Sens., accepted for Trans. Geosci. Remote Sens, 44(12), pp. 3497–3505, 2006.
publication. [27] D. Perissin and F. Rocca “High-Accuracy Urban DEM
[11] G. Fornaro, F. Serafino, and D. Reale, “4D SAR imaging: Using Permanent Scatterers”, IEEE Trans. Geosci. Remote
The case study of Rome,” IEEE Geosci. Remote Sens. Lett., Sens., vol. 44, no. 11, pp. 3338–3347, Nov. 2006.
vol. 7, no. 2, pp. 236–240, Apr. 2010. [28] D. Perissin and A. Ferretti, “Urban-Target Recognition
[12] A. Ferretti, C. Prati, and F. Rocca, “Nonlinear Subsidence by Means of Repeated Spaceborne SAR Images”, IEEE
Rate Estimation Using Permanent Scatterers in Differential Trans. Geosci. Remote Sens., vol. 45, no. 12, pp. 4043–4058,
SAR Interferometry”, IEEE Trans. Geosci. Remote Sens., vol Dec. 2007.
38, pp. 2202–2212, 2000. [29] O. Monserrat, M. Crosetto, M. Cuevas, and B. Crippa,
[13] A. Ferretti, C. Prati, and F. Rocca, “Permanent Scatterers “The Thermal Expansion Component of Persistent Scatterer
in SAR Interferometry”, IEEE Trans. Geosci. Remote Sens., Interferometry Observations,” IEEE Geosci. Remote Sens.
vol. 39, no. 1, pp. 8–20, Jan. 2001. Letters, vol. 8, no. 5, pp. 864–868, Sept. 2011.
[14] N. Adam, R. Bamler, M. Eineder, B. Kampes, Parametric [30] G. Fornaro, D. Reale, and S. Verde, “Potential of SAR for
estimation and model selection based on amplitude-only data Monitoring Transportation Infrastructures: an Analysis with
in ps-interferometry. In: Proceedings of FRINGE, 2005. the Multi-Dimensional Imaging Technique,” J. Geophys.
[15] B. Kampes, Radar interferometry: persistent scatterer Eng. 9 S1, 2012, doi:10.1088/1742-2132/9/4/S1.
technique. Springer, 2006. [31] G. Fornaro, D. Reale, and S. Verde, “Bridge Thermal
[16] A. De Maio, G. Fornaro, A. Pauciullo, “Detection of Dilation Monitoring With Millimeter Sensitivity via
Single Scatterers in Multi-Dimensional SAR Imaging”, IEEE Multidimensional SAR Imaging,” IEEE Geosci. Remote Sens.
Trans. Geosci. Remote Sens., vol. 47, no. 7, July 2009. Lett., early access article, doi: 10.1109/LGRS.2012.2218214.
[17] F. Lombardini, F. Gini, P. Matteucci, “Application of [32] D. Reale, G. Fornaro, A. Pauciullo, X.Zhu, and R. Bamler,
array processing techniques to multibaseline InSAR for lay- “Tomographic Imaging and Monitoring of Buildings With
over solution”, Proc. of the 2001 IEEE Radar Conf., pp. Very High Resolution SAR Data,” IEEE Geosci. Remote Sens.
210–215, 2001. Letters, vol. 8, no.4, pp. 661–665, July 2011.
[18] F. Gini, F. Lombardini, “Multilook APES for multibase- [33] N. Adam, B. Kampes, M. Eineder, J. Worawattanamateekul,
line SAR interferometry,” IEEE Trans. on Signal Processing, M. Kircher, The development of a scientific permanent scat-
50 (7): 1800–1803, 2002. terer system, ISPRS Hannover Workshop, Hannover, 2003.
[19] M. Nannini, R. Scheiber, A. Moreira, Estimation of the [34] X. Zhu, Very High Resolution Tomographic SAR
minimum number of tracks for SAR tomography. IEEE Inversion for Urban Infrastructure Monitoring – A Sparse
Trans. Geosci. Remote Sens. 47 (2): 531–543, 2009. and Nonlinear Tour, Deutsche Geodätische Kommission,
[20] A. Budillon, A. Evangelista, A.; and G. Schirinzi, Reihe C, Nr. 666, Verlag der Bayerischen Akademie der
“Three-Dimensional SAR Focusing From Multipass Signals Wissenschaften, ISBN 978-3-7696-5078-5, 2011.
Using Compressive Sampling”, IEEE Trans. Geosci. Remote [35] Y. Wang, X. Zhu, Y. Shi., R. Bamler, “Operational
Sens., 49(1), 488–499, 2011. TomoSAR Processing Using TerraSAR-X High Resolution
[21] A. Budillon, A. Evangelista, G. Schirinzi, “SAR Spotlight Stacks from Multiple View Angles”, Proc. IEEE
Tomography from sparse samples,” Proc. IEEE 2010 IGARSS 2012 IGARSS Conf., Munich, Germany, 2012.
Conf., Cape Town, Africa, 2010. [36] S. Gernhardt, X. Cong, M. Eineder, S. Hinz, & R.
[22] X. Zhu, R. Bamler, Tomographic SAR Inversion by L1 Bamler, Geometrical Fusion of Multitrack PS Point Clouds
Norm Regularization – The Compressive Sensing Approach, IEEE Geoscience and Remote Sensing Letters, Vol. 9 (1), pp.
IEEE Trans. Geosci. Remote Sens., 48(10), pp. 3839–3846, 2010. 38–42, 2012.
[23] X. Zhu, R. Bamler, “Super-Resolution Power and [37] X. Zhu, M. Shahzad, R. Bamler, From TomoSAR Point
Robustness of Compressive Sensing for Spectral Estimation Clouds to Objects: Façade Reconstruction, Proc. TyWRRS
with Application to Spaceborne Tomographic SAR,” IEEE 2012, Naples, Italy, 2012.
Trans. Geosci. Remote Sens., 50(1), pp. 247–258, 2012. [38] S. Gernhardt, N. Adam, M. Eineder, and R. Bamler,
[24] X. Zhu, R. Bamler, “Demonstration of Super-resolution “Potential of Very High Resolution SAR for Persistent
for Tomographic SAR Imaging in Urban Environment,” IEEE Scatterer Interferometry in Urban Areas,” Ann. GIS, vol. 16,
Trans. Geosci. Remote Sens., 50(8), pp. 3150–3157, 2012. no. 2, pp. 103–111, Jun. 2010.
[25] A. Pauciullo, D. Reale, A. De Maio, G. Fornaro, [39] X. Zhu, R. Bamler, “Sparse Tomographic SAR Reconstruction
“Detection of Double Scatterers in SAR Tomography,” IEEE from Mixed TerraSAR-X/TanDEM-X Data Stacks,” Proc. IEEE
Trans. Geosci. Remote Sens., vol. 50, no. 9, Sept. 2012. 2012 IGARSS Conf, Munich, Germany, 2012.

IEEE Geoscience and Remote Sensing Society Newsletter • December 2012 17