8 Satellite Microwave Radar Observations of Antarctic Sea Ice
8.4.2.2
Multichannel and Multisensor Data Fusion
Further preliminary work to classify sea ice in ERS scatterometer data is being undertaken by Early (unpublished) using isodata clustering algorithms on three-dimensional ERS-1 scatterometer image data sets. Three dimensions are constructed from sixday A and B images and an isotropy image, where the primary A image is a standard
EScat enhanced product comprising a weekly mean 40 0 incidence angle normalized
backscatter coefficient image, the B image is the gradient in scattering coefficient in the
20-60 0 incidence angle range, and the isotropy image characterizes the azimuthal variance in scattering coefficient and combines information acquired using different
beams of the scatterometer instrument. As illustrated by the section above, the main
problem in one-dimensional ice classification using EScat data is the separation of
bright returns from pancake ice, multiyear ice and rough first-year and brash ice. While
the former and latter ice types are typically found seasonally in a specific area, it is difficult to unambiguously distinguish perennial ice near the ice margin. The primary
advantage of this three-dimensional method is that pancakes and brash in the MIZ can
be distinguished from multiyear ice in high concentrations. Similarly, rough ice with
isotropic scattering characteristics can be distinguished from bright patches of pancake growth, where wave effects create a large degree of anisotropy.
Multidimensional algorithms can further address the problems of overlap between
the backscatter coefficient distributions by using multispectral approaches which
merge microwave radar and radiometer data. Beaven and Gogineni (this volume), for
instance, merge SAR and passive microwave data in an attempt to improve sea ice classification. Future studies will attempt to fuse ERS, NSCAT, and SSM/I data to improve
ice classifications, especially in situations where atmospheric water vapor limits the
accuracy of passive microwave retrievals.
8.5
Antarctic Sea Ice Dynamics
One of the primary scientific goals of employing high-resolution satellite radar is to generate SAR ice kinematics data products with which to develop a spatial and temporal
picture of Antarctic ice drift and opening and closing in response to the momentum
transfer (Drinkwater 1995b). The benefit of radar is that it can image the ice night and
day, such that a continuous database of images may be acquired for ice motion tracking
processing. Until now, ice-motion studies utilized satellite-tracked buoy data as a means
of characterizing the drift and dynamics primarily in the Weddell Sea. As such,
Argos/GPS buoys are also the only source of detailed independent information on ice
kinematics. Their drawback, on the other hand, is that they do not provide information
on the spatial characteristics of drift unless they are carefully deployed in a network.
8.5.1
Length Scales of Antarctic Sea-Ice Drift
Vihma et al. (1996) and Kottmeier and Sellmann (1996) have investigated the spatial
correlation function of drift in the Weddell Sea using various spatial arrangements of
8.4.2.2
Multichannel and Multisensor Data Fusion
Further preliminary work to classify sea ice in ERS scatterometer data is being undertaken by Early (unpublished) using isodata clustering algorithms on three-dimensional ERS-1 scatterometer image data sets. Three dimensions are constructed from sixday A and B images and an isotropy image, where the primary A image is a standard
EScat enhanced product comprising a weekly mean 40 0 incidence angle normalized
backscatter coefficient image, the B image is the gradient in scattering coefficient in the
20-60 0 incidence angle range, and the isotropy image characterizes the azimuthal variance in scattering coefficient and combines information acquired using different
beams of the scatterometer instrument. As illustrated by the section above, the main
problem in one-dimensional ice classification using EScat data is the separation of
bright returns from pancake ice, multiyear ice and rough first-year and brash ice. While
the former and latter ice types are typically found seasonally in a specific area, it is difficult to unambiguously distinguish perennial ice near the ice margin. The primary
advantage of this three-dimensional method is that pancakes and brash in the MIZ can
be distinguished from multiyear ice in high concentrations. Similarly, rough ice with
isotropic scattering characteristics can be distinguished from bright patches of pancake growth, where wave effects create a large degree of anisotropy.
Multidimensional algorithms can further address the problems of overlap between
the backscatter coefficient distributions by using multispectral approaches which
merge microwave radar and radiometer data. Beaven and Gogineni (this volume), for
instance, merge SAR and passive microwave data in an attempt to improve sea ice classification. Future studies will attempt to fuse ERS, NSCAT, and SSM/I data to improve
ice classifications, especially in situations where atmospheric water vapor limits the
accuracy of passive microwave retrievals.
8.5
Antarctic Sea Ice Dynamics
One of the primary scientific goals of employing high-resolution satellite radar is to generate SAR ice kinematics data products with which to develop a spatial and temporal
picture of Antarctic ice drift and opening and closing in response to the momentum
transfer (Drinkwater 1995b). The benefit of radar is that it can image the ice night and
day, such that a continuous database of images may be acquired for ice motion tracking
processing. Until now, ice-motion studies utilized satellite-tracked buoy data as a means
of characterizing the drift and dynamics primarily in the Weddell Sea. As such,
Argos/GPS buoys are also the only source of detailed independent information on ice
kinematics. Their drawback, on the other hand, is that they do not provide information
on the spatial characteristics of drift unless they are carefully deployed in a network.
8.5.1
Length Scales of Antarctic Sea-Ice Drift
Vihma et al. (1996) and Kottmeier and Sellmann (1996) have investigated the spatial
correlation function of drift in the Weddell Sea using various spatial arrangements of