130
D.P. WINEBRENNER, D.G. LONG, B. HOLT
interpret the observations in a quantitative physical sense, and we must implement the
interpretative operations on machines that can keep pace with the torrent of raw data
that satellites provide.
The fine spatial resolution provided by synthetic aperture radar (SAR) can aid quantitative, physical interpretation of observations in cases where, as in the case of sea ice,
the geophysical medium of interest varies greatly on horizontal scales of tens of meters
to hundreds of kilometers. Spatial resolution allows one to focus attention on sea ice of
a particular type (e.g., leads, thick first-year ice, multiyear ice, ridges, etc.) for which a
theoretical understanding of signatures may be available. The use of fine spatial resolution avoids mixing signatures of various ice types into an average which cannot, in
general, be uniquely unmixed, or which can be unmixed only by a laborious analysis
using ancillary information.
There are, however, prices for to be paid the advantages of SAR. One of these has been
a limitation in swath width and temporal coverage; various design and information
throughput constraints limit the width of the SAR image strip and the interval between
revisits of a given geophysical target. Thus what is gained in interpretability may be lost
in lack of context over the largest relevant spatial scales or shortest time scales. There
is, in addition, a danger of drowning in data while trying to extract a relatively small
volume of information. This is the background against which recent work to utilize SAR
to observe seasonal transitions on Arctic sea ice should be viewed.
A SAR image of sea ice is essentially a snapshot of the spatially varying backscattering cross section of the ice. Ground-based and airborne measurements (Onstott et al.
1987; Livingstone et al.1987; Barber et al.1992; Carlstrom and Ulander 1993; Beaven and
Gogineni 1994) show that like-polarized backscattering cross sections of sea ice, at frequencies from 5 GHz upward, change at the beginning and again at the end of the melt
season. In the case of autumn freeze-up, the measurements of Carlstrom and Ulander
(1993) and Beaven and Gogineni (1994) provided fine temporal resolution of the 6- to
lO-dB change, with continual observations of the same ice samples. Their simple but
convincing modeling indicated the physical cause of the change to be an increase in
volume scattering from air bubbles in the upper 10-20 em of ice as dielectrically lossy
liquid water froze into low-loss, relatively fresh ice. The timing of the freeze-up signature change followed within a week the onset of sustained air temperatures below freezing. In the case of melt onset, fine temporal resolution observations of individual ice
floes were rare or uncalibrated, but the observations did indicate a decrease for multiyear ice of roughly 7-12 dB between periods before and after the beginning of melting.
Temporal changes in first-year ice signatures are evidently smaller and more subtle
(Barber et al. 1992).
Winebrenner et al. (1994) used temperature records from drifting buoys and nearly
coincident SAR imagery from the first European Remote Sensing (ERS-l) satellite to
show that sharp, 9-dB drops in multiyear ice backscattering cross sections (at 5.3 GHz,
VV-polarization) coincided, to within 2-4 days, with rises in air temperature to o°C at
locations in the Beaufort Sea between 72 oN and 82 oN during June 1992. Theoretical
analysis by these authors showed that the cross-section drop is due solely to the appearance ofliquid water in the snow overlying the multiyear ice. Essentially, the strong 5.3GHz backscattering from the bubbly upper layer of multiyear ice dominates the SAR
return while the snow is dry, but moist snow strongly attenuates illumination of, and
scattering from, the bubbly layer. Thus the physical event corresponding to the melt
D.P. WINEBRENNER, D.G. LONG, B. HOLT
interpret the observations in a quantitative physical sense, and we must implement the
interpretative operations on machines that can keep pace with the torrent of raw data
that satellites provide.
The fine spatial resolution provided by synthetic aperture radar (SAR) can aid quantitative, physical interpretation of observations in cases where, as in the case of sea ice,
the geophysical medium of interest varies greatly on horizontal scales of tens of meters
to hundreds of kilometers. Spatial resolution allows one to focus attention on sea ice of
a particular type (e.g., leads, thick first-year ice, multiyear ice, ridges, etc.) for which a
theoretical understanding of signatures may be available. The use of fine spatial resolution avoids mixing signatures of various ice types into an average which cannot, in
general, be uniquely unmixed, or which can be unmixed only by a laborious analysis
using ancillary information.
There are, however, prices for to be paid the advantages of SAR. One of these has been
a limitation in swath width and temporal coverage; various design and information
throughput constraints limit the width of the SAR image strip and the interval between
revisits of a given geophysical target. Thus what is gained in interpretability may be lost
in lack of context over the largest relevant spatial scales or shortest time scales. There
is, in addition, a danger of drowning in data while trying to extract a relatively small
volume of information. This is the background against which recent work to utilize SAR
to observe seasonal transitions on Arctic sea ice should be viewed.
A SAR image of sea ice is essentially a snapshot of the spatially varying backscattering cross section of the ice. Ground-based and airborne measurements (Onstott et al.
1987; Livingstone et al.1987; Barber et al.1992; Carlstrom and Ulander 1993; Beaven and
Gogineni 1994) show that like-polarized backscattering cross sections of sea ice, at frequencies from 5 GHz upward, change at the beginning and again at the end of the melt
season. In the case of autumn freeze-up, the measurements of Carlstrom and Ulander
(1993) and Beaven and Gogineni (1994) provided fine temporal resolution of the 6- to
lO-dB change, with continual observations of the same ice samples. Their simple but
convincing modeling indicated the physical cause of the change to be an increase in
volume scattering from air bubbles in the upper 10-20 em of ice as dielectrically lossy
liquid water froze into low-loss, relatively fresh ice. The timing of the freeze-up signature change followed within a week the onset of sustained air temperatures below freezing. In the case of melt onset, fine temporal resolution observations of individual ice
floes were rare or uncalibrated, but the observations did indicate a decrease for multiyear ice of roughly 7-12 dB between periods before and after the beginning of melting.
Temporal changes in first-year ice signatures are evidently smaller and more subtle
(Barber et al. 1992).
Winebrenner et al. (1994) used temperature records from drifting buoys and nearly
coincident SAR imagery from the first European Remote Sensing (ERS-l) satellite to
show that sharp, 9-dB drops in multiyear ice backscattering cross sections (at 5.3 GHz,
VV-polarization) coincided, to within 2-4 days, with rises in air temperature to o°C at
locations in the Beaufort Sea between 72 oN and 82 oN during June 1992. Theoretical
analysis by these authors showed that the cross-section drop is due solely to the appearance ofliquid water in the snow overlying the multiyear ice. Essentially, the strong 5.3GHz backscattering from the bubbly upper layer of multiyear ice dominates the SAR
return while the snow is dry, but moist snow strongly attenuates illumination of, and
scattering from, the bubbly layer. Thus the physical event corresponding to the melt