Geology Reference
In-Depth Information
estimates by the CIS analysts). Values of
W
i
were selected
to be proportional to the number of pixels available from
the homogeneous polygons of the relevant ice type and
inversely proportional to the standard deviation of their
backscatter. The selected values were 0.6 for NI, GWI, FY
thick, and MY ice types, and 0.4 for Nilas, GI, SY ice and
OW (see Table 11.1 for definitions and thickness ranges of
these ice types). If no homogeneous polygons were availa-
ble for a certain type
i
, then
W
i
was set to 0. The selection
of
W
i
values, however, has a minor impact on the results
because the majority of the polygons are usually heteroge-
neous, hence the second term in equation (8.3) carries
much more weight than the first.
After solving equations (8.7) for
b
i
es
, the two residue
terms defined in equations (8.4) and (8.5) can be deter-
mined. The first (
Sample results of this technique are presented in
Table 8.2. The results were obtained from an analysis of
Radarsat images of ice types that exist during the freeze‐
up period in the Canadian eastern Arctic. The table
includes the backscatter as a function of the radar inci-
dence angle. Backscatter data calculated from homoge-
neous IAPs are also included. The significant difference
of backscatter of seawater between homogeneous and
heterogeneous polygons in the case of the steep incidence
angle (20°-30°) can be attributed to the difference between
water surface roughness in the open ocean and within the
pack ice (i.e., when the water is confined between ice
floes). In the latter case, the wind does not induce as much
roughness to the water surface and that leads to a lower
backscatter as seen in Table 8.2. For an ice type that can
hardly appear in homogeneous IAP such as the gray‐
white ice (GWI), the backscatter estimate from using this
method would converge into a more accurate value as the
number of heterogeneous polygons that include this type
increases. This can be seen in the table as the difference
between the two estimates (from using homogeneous and
heterogeneous polygons) increases.
One of the challenges in discriminating sea ice from
OW in SAR images is the wide range of the OW signa-
ture that overlaps heavily with the sea ice signatures.
Similar to the ice signatures, the OW signature is influ-
enced by the SAR incidence angle. Additionally, it is
influenced by the wind‐driven surface roughness. In fact,
the high dielectric constant of seawater leads to signifi-
cant surface scattering, which is highly sensitive to sur-
face roughness (Section 7.7.2). This sensitivity has caused
errors in ice‐water classification and ice concentration
estimates from SAR data (especially from using a single‐
channel SAR). The error is more pronounced in the case
of data from wide swath SAR such as the Radarsat
ScanSAR‐wide mode. This is the primary data source for
the operational ice monitoring. It features a 500 km
swath width with an incidence angle range from 20° in
the near‐range to 49° in the far range. A useful review of
the effects of the wind speed and SAR incidence angle
on radar backscatter from ocean surface is presented in
Robinson
[2004].
in es
represents the degree of con-
formity of the estimated backscatter of ice type
i
with the
value obtained from its homogeneous training areas (if they
exists). This is called residue of type 1 (RT1). It can be
regarded as a quantitative measure of the difficulty at which
backscatter can be used as a sole criterion in identifying
the relevant ice type in a Radarsat image. That is because
b
i
in
is strictly based on backscatter, whereas
b
i
es
indirectly
incorporates all ancillary data used in the Radarsat image
analysis (in addition to backscatter). A small value of RT1
means there is a better possibility of using backscatter to
identify the given ice type and vice versa. The second resi-
due term (
bb
i
)
i
ex ob
represents the degree of consistency
of the analysis of polygon
j
(in terms of ice types and
concentrations) with respect to the analyses of all pre-
vious polygons. This will be called residue of type 2
(RT2). A low value means higher consistency and vice
versa. Anomalous polygons, defined based on a selected
threshold on RT2, can then be identified. Consequently,
they can be either revised by the CIS analysts or removed
from the set equations (8.1). The record of RT2 can be
considered a quantitative measure of the robustness of
the visual analysis of Radarsat images at CIS. This is
particularly important as results from CIS Radarsat
image analysis are increasingly used in algorithm devel-
opment, sea ice flux studies and data assimilation.
The aforementioned mathematical scheme was applied
to three sets of IAPs, each set was associated with a spe-
cific range of radar incidence angle: 20°-30°, 30°-40°,
and 40°-50° from the Radarsat ScanSAR mode. The
input is a table where each line includes the ice types and
concentrations from a single polygon
j
as well as the aver-
age backscatter
B
j
ob
from the polygon. This represents a
single equation in the set equations (8.1). As more ana-
lyzed polygons are introduced, the number of equations
increases and an updated solution is obtained. As the
number of occurrences of an ice type
i
in the analyzed
polygons increases, the number of estimates of
b
i
es
also
increases and eventually stabilizes around a final value.
BB
j
)
j
8.1.2. Effect of Surface Wind Speed over
Ocean on Backscatter
Wind speed over ocean triggers surface waves with a
wide range of amplitudes and wavelengths. This affects
the backscatter because the amplitude and wavelength
of the ocean wave, along with the incidence angle of the
radar beam, determine the mechanism of the interaction
of the incident radar beam with the surface (e.g., forward,
random, or Bragg scattering as described in section 7.7.2).
As a rule of thumb, steep incidence angles (around nadir
