The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
180
3. INPUT DATA
ENVISAT and ALOS SAR imagery were tested using DInSAR
for monitoring the ground deformation due to mining at the
study area shown in Table 1. The spatial and temporal
separations between the images of the interferometric pair are
indicated as perpendicular baseline (Bperp) and temporal
baseline (Btemp).
Sensor
Master
(dd/mm/yyyy
)
Slave
(dd/mm/yyyy
)
Bperp
(m)
Btemp
(day)
ALOS/
PALSAR
14/6/2007*
30/7/2007*
515
46
30/7/2007*
14/9/2007*
143
46
15/12/2007**
30/01/2008**
351
46
ENVISAT
/ASAR
3/1/2008
7/2/2008
361
35
Tabled. Interferometric pairs tested using DInSAR for
monitoring mine subsidence at the study area. Note: * and **
denote the ALOS data acquired in single and dual polarization
mode, respectively.
4. LIMITATIONS OF DINSAR MEASUREMENT
Radar interferometry is the technique to measure geodetic
information, e.g. elevation or height change, by exploiting the
phase information of at least two images. The precision of the
result depends on not only the wavelength of the microwave
signals emitted by the SAR antenna, but also the geometry of
interferometric pair. Shorter wavelengths are more sensitive to
the height change of terrain than the longer wavelengths.
However, the shorter wavelengths can be more influenced by
vegetations. Geometry of interferometric pair determines the
sensitivity of interferometric phases with respect to topography.
After all, the precision of radar interferometry depends on how
accurate the process is to convert the interferometric phases to
height or height changes. The limitations of DInSAR
measurement are discussed in the following subsections.
4.1 Decorrelation
The interferometric quality of an identical imaging pixel in two
SAR images can be assessed by measuring the correlation
between the pixels. The quantity of correlation, or the so-called
coherence, can be considered as a direct measure for the
similarity of the dielectric properties of the same imaging pixels
between two SAR acquisitions. Decorrelation degrades the
coherence between two SAR images and results phase noise.
Decorrelation is site-specific. It depends upon local climate
conditions, vegetation cover, complexity of the terrain
(moderate v.s. hilly or mountainous), land use, etc. As a rule of
thumb, decorrelation is more severe for the area having heavily
vegetated cover, complex terrain or constantly changing climate
conditions. Decorrelation can be classified into three main
categories: spatial, temporal and volume decorrelation (Zebker
and Villasenor, 1992).
Spatial decorrelation is caused by the physical separation, or so-
called perpendicular baseline, between the locations of the two
SAR antennas. The maximum allowable baseline before the
pixels get totally decorrelated is often referred to as the critical
baseline (Zebker et al., 1994). The perpendicular baseline of the
interferometric pair for DInSAR processing is therefore
preferred to be as small as possible.
Temporal decorrelation is caused by the variation of the
dielectric properties of ground objects over time between two
repeat-pass acquisitions. In repeat-pass interferometry the sub
resolution properties of the imaged scatterers (sub-scatterers)
may change between surveys, e.g. by meteorological
differences at the time of image acquisitions, the movement of
leaves and branches, water surface, or vegetation growth.
Therefore, urban regions normally illustrate much higher
coherence than vegetated/agricultural areas over a period of
time.
Volume decorrelation is also about the phase stability of the
imaging cell (pixel) over time, especially over vegetated
regions. The dielectric properties and alignments of tree leaves
and branches in the canopy may result in the variation of the
backscattered radar signals. They are also affected easily by the
local weather conditions, such as rains and winds. Therefore, it
is desirable to have a finer SAR imaging resolution in order to
minimise the variation of the random scattering phases.
The histograms of the coherence value of 4 selected
interferometric pairs as given in Table 1 are shown in Figure 1.
The mean coherence of the ALOS pair, 15/12/2007 -
30/01/2008, illustrated higher value than other two ALOS pairs
acquired in June, July and September in 2007.
It is interesting to note that even the perpendicular baseline of
the ALOS pair, 30/7/2007 - 14/9/2007, is 143m its coherence
over the selected scene is not as good as expected. After
comparing this coherence map to a high resolution optical
image of the identical area, the areas showing low coherence
value (<0.3) are actually farmlands. It suggests the local
agriculture pattern may cause additional noise in the scene.
Based on the coherence value shown in Figure 1, the
decorrelation due to agriculture is more severe in summer (June
- September) than winter (December - February). In addition,
the ALOS data acquired on 15/12/2007 and 30/01/2008 are in
single polarization mode, which has twice better imaging
resolution than the images acquired in dual polarization mode.
Therefore, the interferometric pairs with better imaging
resolution will be less subject to volume decorrelation, hence
high interferometric coherence value.
The ENVISAT pair covers a similar period as the 3 rd ALOS
pair. However, it has less mean coherence of 0.379. The
decorrelation may be caused by the spatial decorrelation as the
perpendicular baseline of this ENVISAT pair is about 361m.
4.2 High phase gradient
As mentioned earlier, the interferometric phase variation caused
by land deformation is linearly depending on the height change
of terrain. Therefore, a large vertical ground movement over a
small spatial extent causes the rapid changes of phases in the
differential interferogram. When the phase gradient gets too
large, it becomes ambiguous and it is impossible to recover the
height change accurately. The problem can be solved or
reduced by having interferometric pairs with higher imaging
resolution, signals with longer wavelength and/or a shorter site
revisit cycle. This is demonstrated in Figure 2 by comparing the
DInSAR results derived using ENVISAT (C-band, wavelength
= 5.6cm) and ALOS (L-band, wavelength = 23.5cm) data.