IEEE Geoscience and Remote Sensing Magazine - June 2020 - 7

3D SAR TECHNIQUES
The 3D mapping of cities is of great interest for many applications, such as environmental risk management, urban
planning [1], and the compilation of 3D city models and
high-level geographic information system data [2]. Thanks
to the meter resolution of recent spaceborne SAR sensors,
interferometric (InSAR) and TomoSAR techniques can now
address this issue [3]. Moreover, the ability of SAR sensors
to monitor subcentimeter deformation displacements in
urban areas from space [4], both over extended areas and
at the single-infrastructure level, has generated increasing
interest from the scientific community. The past decade has
seen the development of increasingly advanced processing
techniques, allowing the achievement of more ambitious
objectives and results with increased accuracy.
In recent years, many techniques for ground-heightprofile reconstruction using InSAR data have been developed. Cross-track InSAR systems exploit multiple acquisitions over the same ground scene taken along different
sensor trajectories separated in the direction orthogonal
to both the line of sight and the track (spatial baselines)
[5]. The first applications of InSAR techniques, which date
back approximately 30 years, were based on the use of only
one phase interferogram and required a phase unwrapping
operation [6]-[8]. These techniques suffered from the problem of solution ambiguities, which could be overcome by
imposing a regularity constraint on the height profile of
the scene, that is, by assuming the absence of strong height
discontinuities and/or high slopes [9]. Because this assumption is not met in many cases, such as in urban and mountainous areas, many methods exploiting multibaseline SAR
phase interferograms have been developed to solve the
solution ambiguity and improve height reconstruction accuracy [10]-[13].
When multibaseline images are taken with a temporal
separation (temporal baselines), differential interferometric techniques (DInSAR) [4] can be applied to reveal Earth
displacement maps with subcentimeter accuracy. This has
been widely used for the monitoring of large areas affected
by slow and long-term movements, for instance, due to
landslides, subsidence, and water or oil extraction.
Several DInSAR techniques have been developed in the
past few decades [14]. Those based on persistent scatterer
interferometry (PSI), which are essentially based on phaseonly models, have proven to be very effective and have been
extensively applied [14], [15]. PS-InSAR is a standard tool for
monitoring slow deformations, including ground [4], [16],
single buildings [17], [18], and infrastructures [19], [20].
InSAR and DInSAR techniques are based on the assumption that backscattering predominantly occurs as a single
dominant scatterer within a given range/azimuth-resolution cell, which is characterized by a high signal-to-noise
ratio (SNR). Therefore, these techniques cannot provide reliable results in the presence of volume scattering or in areas with a strong layover phenomenon. In the last decade,
many advances have been made in the development of new
JUNE 2020

IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE

interferometric methods that allow the management of distributed scatterers (DSs) [14], but they can be exploited only
if the DSs form homogeneous groups of pixels that are large
enough to allow a statistical analysis [21]. These models do
not adapt well to urban applications, where the scene varies
on a small spatial scale and several coherent scatterers often
interfere in the same range/azimuth-resolution cell.
More recently, TomoSAR techniques have extended interferometric methods to multidimensional imaging [22].
By exploiting multibaseline acquisitions, TomoSAR allows the ground-reflectivity profile to be projected in 3D
Euclidean space [23]. When the data are acquired on different dates (multitemporal data), it is possible to obtain
the image in the 4D (3D space plus deformation velocity)
or 5D (3D space plus deformation velocity plus thermal
dilation) spaces [24], [25]. The 3D imaging (focusing) provides a reconstruction of the scene's scattering structure
along the vertical direction, which is not possible with InSAR techniques.
This is a very important feature that has been widely exploited in applications involving volume-distributed scattering along the vertical direction, such as in the case of
forested areas [26], where the incident wave can penetrate
the vegetation layer when lower frequency sensors (i.e.,
P-band or L-band SAR systems), are used. In this case, the
backscattered signal is the superposition of different contributions from the vegetation layer and ground, which must
be separated to reconstruct the ground profile and forest
vertical structure.
In contrast, in the case of urban areas, the incident wave
does not penetrate the scene, and surface and multireflection scattering mechanisms (for instance, double bounce)
are usually dominant [27], [28]. Also, due to the presence
of buildings with different heights in this case, multiple
targets responding in the same range/azimuth-resolution
cell can be present, so scatterer unmixing is required to
perform 3D imaging [29].
The applications of TomoSAR to urban areas are numerous [30]-[34]. These techniques use both the amplitude and phase of a stack of interferometric images to focus
the 3D-reflectivity profile and separate multiple scatterers.
This processing amounts to the synthesis of a synthetic
aperture along the direction orthogonal to the range and
azimuth through the coherent superposition of a stack of
multibaseline complex-valued interferometric images. It
consists of the inversion of the model linking the multibaseline images to the 3D reflectivity profile of the scene,
which essentially involves a Fourier reconstruction starting from measurements that are not uniformly spaced [23],
[35]. Several works have considered the use of polarimetric
information in TomoSAR 3D reconstructions, both for vegetation and urban applications, due to the improved ability to separate different scattering mechanisms thanks to
polarization [36]-[43].
Several practical problems must be addressed to obtain accurate tomographic reconstructions, including the
7
IEEE Geoscience and Remote Sensing Magazine - June 2020

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - June 2020
