Geology Reference
In-Depth Information
Edge detection techniques are commonly used to
delineate pixels with high gradient of gray tone values in
a satellite image. Ridges are defined by their high gradi-
ent, although the reverse is not true. This means that the
gradient must be complemented by the gradient must be
complemented by at least one more criterion. This can
be the spatial arrangement of the high gradient pixels.
Ridge pixels are typically organized along a linear or
quasi‐linear shape. If the high‐gradient pixels are spread
over an area, they probably designate rubble or brash ice.
With this approach the task is reduced to searching for
an optimal edge detection technique that marks continu-
ous edge pixels with minimum noise pixels (noise lead
definition of false edges). Isolated pixels of high gray
tone or gradient of gray tone represent noise and should
therefore be removed. This can be achieved using the fast
Fourier transform (FFT) technique accompanied with a
high pass filter. Since the criterion that has to be employed
in the edge detection or the threshold approach is the
contrast between pixel values, calibrated backscatter is
not needed.
One of the first studies that addressed the identification
of ridges in SAR images was conducted by
Vesecky et al.
[1990] using the L‐band SAR onboard Seasat. It uses what
has become a traditional approach of image processing to
identify and characterize ridges in radar imagery. It com-
bines a threshold to identify bright pixels in the image and
a line detection operator using a set of masks that are
convolved with a 5 × 5 pixel window. This produces linear
segments that can potentially form a ridge. The rest of the
work is to examine the connectivity of those segments to
decide if they structurally make a ridge.
Zhou and Li
[2000]
combined the backscatter threshold with a supervised
classification to identify ridges and brash ice in SAR
images. As these two objects have equally high backscatter,
they can be distinguished using an object delineation
method to demarcate the ice floe boundaries. The high
backscattering pixels located within an ice floe can then be
uniquely identified as ridge pixels, while those located at
the edges or between floes can be labeled as brash ice.
The above discussions imply that the image analysis
approach would be more successful if the surface defor-
mation features are readily visible in the original SAR
images. It is, therefore, essential to select the mode and the
SAR parameters (frequency, polarization, and incidence
angle) that enhance the visibility of the ridges in particu-
lar. Discussions of the effect of each parameter are pre-
sented in the following, but it should be noted that a ridge
becomes more visibly identifiable in a SAR image if it is
oriented parallel (or nearly parallel) to the satellite ground
track. Ridges that have no definite orientation or more
rubble‐like spatial configuration are difficult to detect.
As a rule of thumb, higher incidence angles (i.e., shal-
low angles) are preferred for ridge detection in SAR
images. At these angles the backscatter caused by surface
roughness is reduced while the backscatter from the ridges
is enhanced because most of the scattering is directed
back to the antenna.
Melling
[1998] concluded that the
ERS‐1 C‐band operating at a small incidence angle
(around 25°) is only of marginal use for ridge detection,
whereas an airborne X‐band SAR with a larger incidence
angle (between 40° and 70°) enables unambiguous dis-
crimination of ridges.
As for radar frequency, the longer wavelength L‐band
(23 cm) delineates ridges and brash ice better than the
C‐band (5.4 cm).
Dierking and Busche
[2006] compared
coincident images from the L‐band JERS‐1 and the C‐
band ERS‐1 and concluded that although the C‐band is
regarded as a reasonable choice for all‐season ice
monitoring capabilities, the L‐band is superior for the
specific task of mapping the surface deformation features.
Figure 9.2 shows a comparison between JERS‐1 and
ERS‐1 SAR images. It demonstrates the better capability
of JERS‐1 to delineate different forms of ice deformation:
ridges, shear zone, and brash ice. Note the better contrast
of brash ice between the small ice floes in zone C in the
JERS‐1 image. This makes L‐band imagery better suited
for delineating brash ice fields than C‐band imagery. In
comparing the two images in Figure 9.2 it should be noted
that the JERS-1 image was acquired at larger incidence
angle. This adds another advantage to being acquired
with the longer L-band wavelength. JERS‐1 SAR polari-
zation is HH while ERS‐1 is VV but that should not make
a difference as they both affect the ridges in almost the
same manner. Two important observations in Figure 9.2
are worth mentioning, though not relevant to the surface
deformation,. The first is that the backscatter from MY ice
floes (e.g., the floe marked B) is lower in the JERS‐1
images. This is because the surface becomes less rough
with respect to the longer wavelength of the L‐band.
Moreover, the L‐band wavelength (~23 cm) is much bigger
than the air bubble responsible for volume scattering in the
subsurface layer. Therefore, both surface and volume scat-
tering are reduced. The second observation is the much less
backscatter from the thin ice (zone A) in the L‐band than
the C‐band. These two observations have an impact on ice
type discrimination as explained in section 10.1.2.
A similar study to compare the L‐band and C‐band
SAR data is presented in
Arkett et al
. [2008]. They com-
pared the L‐band HH ALOS‐PALSAR images of differ-
ent sea ice scenes in the Arctic and the east coast of
Canada against near coincident Radarasat‐1 (C‐band)
ScanSAR wide images. Figure 9.3 shows a Radarsat‐1
image acquired on 13 January 2008 and a PALSAR wide‐
beam image acquired almost 5 h later. The Radarsat
image analysis from CIS identifies this area as a mixture
of thick FY ice and MY ice. Although MY ice is visible
in the PALSAR image, its tone is very similar to that of
