IEEE Geoscience and Remote Sensing Magazine - March 2013 - 32

z (m)

A further possibility for tomographic
imaging consists in acquiring
772
data by means of circular trajectories
770
Canopy
instead of straight lines, the so-called
768
circular SAR (CSAR) [159]-[165]. In
766
CSAR, the sensor follows a 360° cir764
cular flight around a spotted region,
-385
having 3-D imaging capabilities and
762
Trunk
allowing the maximum attainable
-515
760
Canopy
-380
resolution of a fraction of the wave-510
758
length in the ground plane. Never-385
-505
-380
x (m)
-510 -515 y (m) -375
y (m)
-375
theless, both potentials are linked to
-505
x (m)
the backscattering stability of the targets for different illumination angles,
FIGURE 24. Fully polarimetric-holographic tomogram of a single tree in Pauli basis and
hence limiting the performance in
focused with beamforming (BF). A total of 21 circular passes were processed into a volume
real-world scenarios, where the tarand added coherently [166].
gets tend to be anisotropic. In this
sense, CSAR can also allow the study
of the anisotropic properties of targets.
sampling call for more advanced processing schemes in
The acquisition of several circular trajectories over
order to increase the resolution beyond (47) and reduce
the same area further extends the potentials of CSAR,
the sidelobe level. The most commonly used solutions
where several options are possible [162], [165], [166]. On
are based on MUSIC, CAPON, or, more-recently, comthe one hand, the circle can either be divided in small
pressive sensing (CS). MUSIC and CAPON are non-linear
angular regions and a conventional tomographic promethods, while MUSIC is also parametric, i.e., it requires a
cessing be performed, so that the final combination of
priori information concernall sub-angle tomograms results in a holographic represening the number of scatterers
tation of the scene. A second option is to project each
to be detected. Also, both
SAR tomography
circle completely into a volume to finally add all of them
require the estimation of the
exploits a synthetic
coherently and achieve the maximum possible resolucovariance matrix, implyaperture in elevation
tion. Fig. 24 shows an example of the latter approach
ing a resolution loss due to
with data of a multi-circular campaign performed with
the averaging operation. On
to retrieve the
DLR's F-SAR sensor at L-band, where a total of 21 circular
the
other
hand,
CS
works
vertical distribution
passes were acquired.
at
full
resolution
and
can
of scatterers.
reconstruct non-uniformly
VII. Future Developments
sampled sparse signals, the
The last decades have witnessed a tremendous increase
latter meaning that the elevaof Earth observation applications that take advantage of
tion profile to be estimated must be discrete. If the signal of
the unique properties of high-resolution SAR images.
interest is indeed sparse, the CS theory guarantees the posTo study dynamic processes on the Earth surface, more
sibility to obtain it at a rate significantly below the Nyquist
and more users ask for time series or stacks of coherent
one [150].
radar images acquired in repetition intervals that are as
The use of CS in the frame of urban monitoring in
short as possible (Fig. 25, see also [167], [83]). The curcombination with PS has been a topic of research in the
rent generation of SAR instruments is, however, limited
recent years. Indeed, differential SAR tomography [20],
in their capability to acquire radar images with both
[151]-[153] allows the discrimination of multiple scathigh-resolution and wide-swath coverage. This immediterers in layover, e.g., ground and building facades, and
ately impacts the acquisition frequency if large contiguat the same time the retrieval of their respective deforous areas shall be mapped systematically with a single
mation velocities. These approaches have been mainly
satellite. The resolution versus swath width restriction is
exploited using current high-resolution spaceborne senfundamental and closely connected to the intricacies of
sors [106], [154]-[157], becoming a powerful tool for
the SAR data acquisition process: SAR imaging exploits
urban monitoring, as already mentioned in Section IV.
the Doppler shift arising from the sensor movement relHowever, in the frame of forest monitoring, CS as such
ative to the ground to improve the azimuth resolution
does not apply so well, as the elevation profile is indeed
well beyond the diffraction limit of the radar antenna.
not sparse in the Fourier basis. One possible solution is
To achieve a high azimuth resolution, a broad Doppler
to make use of a wavelet basis in order to obtain a sparse
spectrum has to be acquired. A broad Doppler spectrum
representation of the vertical structures, hence allowing
means, in turn, that the system has to be operated with
the use of CS also for the imaging of forested areas [158].
32

ieee Geoscience and remote sensing magazine

march 2013

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - March 2013