Geology Reference
In-Depth Information
generation of unique phase information associated with a
certain ice type. Second, it is not clear how the phase
information can be useful in resolving some of the snow
issues that hamper the sea ice classification from single‐
channel SAR data. Recall also that the phase information
is accessible only through the SLC SAR acquisition
mode. This mode is not available from the ScanSAR
mode, which is the most commonly used for operational
sea ice applications.
Multitemporal image acquisition is one way to increase
the dimensionality of SAR data from a single channel.
This approach has been used in applications where a
ground cover type can be identified based on its temporal
evolution of backscatter as in the case of vegetation
cover. It cannot be used to identify ice types because the
ice growth is not associated with an identifiable pattern
of parallel growth in its backscatter. Moreover, even if
there was an identifiable pattern, the SAR revisit rate
(frequency of satellite passes over the same area) may not
be fine enough to capture the transition between ice types.
The conclusion from this point is that multitemporal
SAR data are not useful as a tool for increasing the
dimensionality of the data in order to facilitate ice type
classification. Sequential SAR images of sea ice, however,
have been used successfully to identify the ice displace-
ment, which is a well‐established application to determine
ice motion field (section 10.7).
Multifrequency SAR observations have potential to
identify ice types. Among the first experimental SAR
multifrequency systems was NASA's Jet Propulsion
Laboratory AIRSAR operating in the C‐, L‐, and P‐band
frequency. The potential of this approach for sea ice
classification was presented in
Drinkwater et al
. [1991].
Another airborne multifrequency SAR system was devel-
oped in the late 1980s by Canada Centre for Remote
Sensing onboard the Convair 580 aeroplane [
Livingstone
,
1996]. It operated in the X‐ and C‐bands. These two sys-
tems were never intended for operational use, but studies
showed the potential of multifrequency data for discrimi-
nation between FY and MY ice types. A few studies have
been conducted lately to compare data from space‐borne
radar systems with different frequencies onboard differ-
ent satellites and integrate them to enhance information
retrieval.
Dierking and Busche
[2006] explored how the
C‐ and L‐band could complement each other and con-
cluded that their combination results in a more detailed
view of the sea ice cover state. It is not likely, however,
that a multifrequency space‐borne system will be devel-
oped in the near future.
Another approach to increase the dimensionality of
SAR data from a single channel is by using texture. A num-
ber of textural measures can be generated from a single‐
channel observation; many of them are uncorrelated. This
approach has been tested extensively since the early years
of SAR sea ice applications but resulted in limited success
[
Shokr
, 1991;
Soh and Tsatsoulis
, 1999;
Clausi and Yue
,
2004]. A key question regarding this approach is how to
capture the texture of different sea ice surfaces. Texture of
natural ice surface exists at different scales ranging from
small‐scale roughness to large‐scale blocks that form
ridges and rough ice surface. A texture technique employs
the texture operator within a window of a given size to be
convolved with the SAR image. The window captures
texture of scales that “resonate” with its size. Therefore, a
texture parameter calculated using a given window size
may fail to capture the texture that takes place at scale very
different from the window size. Moreover, many of the
suggested textural parameters are not independent and
thus offer only a limited number of new dimensions. More
on this subject is presented in section 10.1.2.
Perhaps the most viable approach to generate multi-
channel SAR data is through using multipolarization
channels. While multifrequency SAR data requires more
than one antenna, multipolarization requires one antenna
that transmits pulses in alternate polarization and receives
backscatter in the same alternating order from the same
or the orthogonal polarization. The multipolarization
SAR systems have been developed into dual‐polarization
or quad‐polarization (also called fully polarimetric) sys-
tems. A dual‐polarization SAR mode was included for
the first time in the ASAR system onboard the European
satellite ENVISAT (2002-2012). It recorded the data in
one of three alternating polarization combinations (HH
and HV), (VV and HV), or (HH and VV). Later, ALOS‐1
PALSAR (January 2006-April 2011) and Radarsat‐2
(December 2007-present) carried SAR with a dual‐polar-
ization mode operating with either one of two selections
(HH and HV) or (VV and VH). Fully polarimetric SAR
was one of ALOS‐1 PALSAR and Rardarsat‐2 modes. In
this mode the antenna transmits alternating horizontal
and vertical polarization pulses and records backscatter
from each pixel in the four basic polarization selections
HH, HV, VV, and VH. From these components several
independent parameters can be derived. The parameters
can also be viewed as an expanded set of “multichannel”
SAR data. The next section introduces some of those
parameters with elaboration on their relations to scatter-
ing mechanisms.
An example of a Radarsat‐2 multipolarization scene of
sea ice west of Ellesmere Island in the Arctic from the
three channels HH, VV, and HV is shown in Figure 7.33.
The images (not geographically projected) were con-
structed from fully polarimetric data acquired on 18 May
2010 using the fine beam mode. Integration of infor-
mation from the co‐ and cross‐polarization channels
provides better description of the scene. Smooth FY ice
(or land‐fast ice) has much lower cross‐polarization
backscattering than co‐polarization. Rough FY ice has
