Geology Reference
In-Depth Information
images.
Burns
[1992] used a dynamic threshold technique
applied to six combinations of the five spectral channels
of the AVHRR.
Nolin et al.
[2002] explored a different approach using
data from the multispectral Multiangle Image
SpectroRadiometer (MISR) onboard NASA's Terra sat-
ellite. The spectral information is provided by 4 channels
(3 in the VIS and 1 in the NIR). The angular information
is provided by 9 cameras operating at different angles. One
camera is pointed straight down (nadir direction), 4 are
pointed at different forward angles and the other 4 four
at the same backward angles. This makes a total of 36
reflectance measurements.
Nolin et al.
[2002] explored the
potential uses of the multiangular observations for map-
ping sea ice types based on surface roughness properties.
The premise is that the scale of surface roughness deter-
mines the pattern of surface scattering. For example, the
ice surface is a forward scatterer at relatively fine rough-
ness scales but becomes a backward scatterer when sur-
face microtopography dominated. Snow sastrugi, defined
as wind‐eroded snow with curvlinear protrusions deter-
mined by the wind direction, affects the angular observa-
tion of the ice surface from optical channels. The study
classified the TOA reflectance data obtained from the 9
angular MISR channels using the standard unsupervised
classification technique ISODATA [
Tou and Gonzalez
,
1974]. It demonstrated that sea ice surface has character-
istic angular signatures that can be used in much the same
way as spectral signatures for ice classification. It used the
MISR angular data to distinguish between MY and FY
ice types. The discrimination, however, was not as clear as
it could be from using SAR data. In summer, when sur-
face flood renders SAR data useless for ice classification
the multiangular data from MSR provide key ice‐mapping
information. The optimal number and type of MISR
channels needed to achieve the ice classification has yet to
be researched.
ratio. This is achieved using the familiar linear equation
that combines emission from OW and ice (
T
b
):
TTCT C
b
1
/
(10.2)
,
ice
b
ice
b
,
ow
ice
where
T
b
,ow
is the tie point of the OW and
C
ice
is the ice
concentration.
Hughes
[2009] combined the seven channels of SSM/I
to explore their potential use in an unsupervised classifi-
cation scheme designed for operational use in the
Norwegian Ice Service. The author incorporated data
from the first weeks of March, June, September, and
December in the seven‐dimensional parameter space
and examined their distribution. The
K
‐means clustering
technique [
Hartigan
, 1975] was used to partition the data
into four clusters such that the within‐cluster sum of
squares is minimized. This can be achieved by selecting
initial centers of the clusters and assigning each observa-
tion to its closest cluster. After all points are assigned,
cluster centers are updated and the process is repeated
until there are no further tangible changes in cluster
centers. Clusters are then labeled to ice types according
to some ancillary information. Three ice types were used
in addition to OW; young ice, FY ice, and MY ice.
Weekly maps were produced for three selected ice sea-
sons. Generally speaking, the ice type maps generated
from passive microwave data are too coarse to be of
practical value for operational marine users. Yet they
provide synoptic coverage and can be used as a baseline
for detailed maps of local coverage.
For the active microwave data, SAR images have been
the most promising source for ice type classification. Their
fine resolution (a few tens of meters to 100 m) satisfies
the tactical scale requirements of marine navigation.
Classification at coarser resolution (around 1 km or
greater) is also useful for weather and climate modeling.
SAR data can be used to examine the subpixel informa-
tion in coarser resolution data where pixels usually feature
heterogeneous composition of ice types and OW (e.g.,
from passive microwave sensors). The interest in ice
classification from SAR started after the astounding ice
information obtained from the first space‐borne SAR
onboard Seasat. A working group, called Radar Age/Type
Algorithm Group (RAGTAG), was formed and held its
first workshop in Seattle in July 1988. The mandate was to
review and devise methods to retrieve age‐based ice types
from SAR in anticipation for data from the “future” SAR
systems onboard the European ERS and the Canadian
Radarsat satellites. This effort stimulated several investi-
gators to experiment with SAR data that were available at
that time from Seasat and a few airborne systems. Most of
the methods were geared toward using available image
processing techniques, including texture analysis. That
10.1.2. Ice Classification from Microwave Data
While most of the ice type classification algorithms
from microwave data are based on SAR images, a few
algorithms utilize passive microwave data. For example,
SAF‐OSI, incorporates an ice type product derived from
SSM/I data. The algorithm (described in
Andersen
[2000])
identifies two ice types: FY ice and MY ice. It is based on
the gradient ratio GR
19
V
37
V
defined by equation (8.11).
This parameter is positive for OW, negative for MY ice,
and takes on values close to zero for FY ice (Figure 8.19).
The algorithm accounts for the fact that a mixture of MY
ice and OW will produce GR
19
V
37
V
similar to that of FY
ice. This is done by considering only the part of bright-
ness temperature produced by the ice fraction in a hetero-
geneous footprint (
T
b
,ice
) when calculating the gradient
