Geology Reference
In-Depth Information
classification at a pixel level. Ice analysts in the CIS sub-
jectively evaluated classification results. The authors
claimed (correctly) that this approach was the first known
successful end‐to‐end process for operational SAR sea ice
image classification. It has the potential to improve the
operational SAR image analysis through an interactive
mode of operation.
At this point it is worth providing an example that high-
lights the value of combining optical and microwave
observations to enhance the retrieval of information
about sea ice types through visual analysis of the data.
Figure 10.5 shows an image of sea ice acquired by AVHRR
channel 1 (0.58-0.68
μ
m) over the North Water Polynya,
north of Baffin Bay in April 1998. The image is compared
against the same scene acquired by Radarsat‐1 after resa-
mpling to nearly the same resolution as of AVHRR. Note
the slight difference between the two orbits. The time dif-
ference between the two acquisitions is 35 min. The area
south of the ice arc in the Radarsat‐1 image appears to
have an elongated band of relatively bright signature with
different shades. This is likely to be a mix of thin ice and
OW. The same area in the AVHRR albedo image has a
uniformly dark signature (low albedo). If one relies on the
scene interpretation from the AVHRR image, it would be
difficult to conlude the presence of the ice in this area.
However, combining both sensors supports a conclusion
that the area has thin ice, mosty Nilas, and OW. Both
types have equally low albedo but differet backscatter.
This example shows how commingling data that are based
on different surface properties improves the information
retrieval. Note also the seawater band adjacent to the bor-
der of the ice arc, which envelops the thin ice, in the
Radarsat‐1 image. This is not visible in the AVHRR
image. Thicker ice types are unambigiuosly identified in
AVHRR image while the Radarsat‐1 image does not offer
the same clear distinction between ice and water in the
same area at least for nonexperienced interpreters. This is
another example that highlights the value of combining
the two data sources, but here the advantage of the optical
sensor in discriminating thick ice from water is undis-
puted. It should be pointed out, however, that Radarsat‐1
data provide more detailed information due to its much
finer resolution compared to AVHRR. The information is
retained even after the resampling of the data as shown in
this example.
Multichannel SAR data offer a better potential for ice
classification. The data can be presented in the form
of multifrequency, multipolarization (ultimately fully
polarimetric), or both. Multipolarization SAR data have
been available from a few space‐borne sensors such as
ESA's Sentinel, ENVISAT, CSA's Radarsat‐2, and JAXA's
ALOS PALSAR while fully polarimetric SAR have been
Ice arc
FY ice
Radarsat
AVHRR
Figure 10.5
Images of sea ice in the North Water Polynya north of Baffin Bay, Canadian Arctic. AVHRR channel
1 image (left) and Radarsat‐1 image (right) were acquired almost coincidentally on 20 April 1998. The contrast in
the AVHRR image between FYI on one hand and OW and YI on the other hand is clearly visible. The gray dots
correspond to the same geographic point in both images (Radarsat‐1 image © CSA).
