28
L.-K. SOH, C. TSATSOULIS, AND B. HOLT
another data set (Fig. 13 c,d) obtained on July 4. In Fig. 13a, the floes are primarily dark
separated by bright ridges, except that there is a small circular region on the border
between frames 3 and 4 that is brighter. The floe results in Fig. 13b are largely acceptable, noting that within the brighter region there is a reduction of measurable floes and
increased water/ice mixture. On July 4, Fig. 13C shows that the brightened area has
extended northward and the floes in frames 1 and 2 are also less dark and distinct. In
the brightened areas, there has been an almost complete loss of floe discrimination. We
attribute this large-scale change in backscatter in Fig. 13C to one or a combination of
these possible explanations: (1) There was an overall cooling trend which has caused
refreezing of liquid water in the snow and ice. This could be augmented by the refreezing of water that has surrounded large snow and ice grains, which may form a collection of nodules at the ice/snow interface (Onstott and Gogineni 1985). This explanation
is probably not correct since the process was found to occur before extensive melt. (2)
A cloud cover is present containing ice particles, which would contribute to the overall scattering levels and reduce the detectability of ice surface features. This has been
noted in several ERS-l investigations and is attributed to the sensitivity of C-band frequency to atmospheric conditions (Johannessen et al. 1994). (3) Massive drainage of
melt ponds and surface water has occurred, leaving behind more porous surface ice
grains, which would increase scattering (Holt and Digby, 1985). In Fig. 4C ofWinebrenner et al. (1994), the buoy closest to these images shows steadily increasing air temperatures above 0 0 C after June 18 (melt onset and a steep decrease in backscatter from about
-9 dB to -18 dB) and a sharp upturn in backscatter of +4 dB after June 25.
The resulting floe size measurements are dramatically different on July 4 (Fig. 13d)
as compared to June 30 (Fig. 13b). Fewer floes are measured in frames 1 and 2, frame 3
indicates one very large floe, and frame 4 shows a mixture of floes, discarded floes, and
water/ice mixture. The results of frame 3 are clearly erroneous and are due to the comparatively uniform, featureless backscatter returns. The differing results cannot be
attributed to a different field of ice moving into view of the sensor since tliere are overlapping features between the two dates. Summarizing, large and rapid variations in the
ice backscatter in response to changing ice surface and environmental conditions limit the capability of the algorithm to produce meaningful results.
A late summer image and results are seen in Fig. 14 (August 26,1992). Note in the SAR
image the bright open water along one edge. Compared with Figs. 11, 13, the image and
output are characterized by fewer large floes, with an increased relative fraction of
smaller floes and water/ice mixture as well. This follows the intuitive trend over the
summer season of floes becoming smaller with melting, which is also reflected by the
relative increase in water/ice mixture or subpixel size floes. Also, the bright open water
provides good contrast with the surrounding ice and in this case is also segmented quite
accurately.
The algorithm has particular difficulty in at least two situations, when the radar signature of the ice pack is bright, and when analyzing imagery from the marginal ice zone.
When temperatures drop steadily below freezing, which occurred during August
29-September 7, 1992, in the Beaufort Sea (Winebrenner et al. 1996), multiyear ice
becomes brighter than both first year ice and new ice. In general, results during cold
conditions are not satisfactory, even using the 'cold weather' switch in the algorithm,
with the floes being largely undersampled. This is likely due to the reduced overall contrast between the multiyear ice and the other ice types, which means that the segmen-
L.-K. SOH, C. TSATSOULIS, AND B. HOLT
another data set (Fig. 13 c,d) obtained on July 4. In Fig. 13a, the floes are primarily dark
separated by bright ridges, except that there is a small circular region on the border
between frames 3 and 4 that is brighter. The floe results in Fig. 13b are largely acceptable, noting that within the brighter region there is a reduction of measurable floes and
increased water/ice mixture. On July 4, Fig. 13C shows that the brightened area has
extended northward and the floes in frames 1 and 2 are also less dark and distinct. In
the brightened areas, there has been an almost complete loss of floe discrimination. We
attribute this large-scale change in backscatter in Fig. 13C to one or a combination of
these possible explanations: (1) There was an overall cooling trend which has caused
refreezing of liquid water in the snow and ice. This could be augmented by the refreezing of water that has surrounded large snow and ice grains, which may form a collection of nodules at the ice/snow interface (Onstott and Gogineni 1985). This explanation
is probably not correct since the process was found to occur before extensive melt. (2)
A cloud cover is present containing ice particles, which would contribute to the overall scattering levels and reduce the detectability of ice surface features. This has been
noted in several ERS-l investigations and is attributed to the sensitivity of C-band frequency to atmospheric conditions (Johannessen et al. 1994). (3) Massive drainage of
melt ponds and surface water has occurred, leaving behind more porous surface ice
grains, which would increase scattering (Holt and Digby, 1985). In Fig. 4C ofWinebrenner et al. (1994), the buoy closest to these images shows steadily increasing air temperatures above 0 0 C after June 18 (melt onset and a steep decrease in backscatter from about
-9 dB to -18 dB) and a sharp upturn in backscatter of +4 dB after June 25.
The resulting floe size measurements are dramatically different on July 4 (Fig. 13d)
as compared to June 30 (Fig. 13b). Fewer floes are measured in frames 1 and 2, frame 3
indicates one very large floe, and frame 4 shows a mixture of floes, discarded floes, and
water/ice mixture. The results of frame 3 are clearly erroneous and are due to the comparatively uniform, featureless backscatter returns. The differing results cannot be
attributed to a different field of ice moving into view of the sensor since tliere are overlapping features between the two dates. Summarizing, large and rapid variations in the
ice backscatter in response to changing ice surface and environmental conditions limit the capability of the algorithm to produce meaningful results.
A late summer image and results are seen in Fig. 14 (August 26,1992). Note in the SAR
image the bright open water along one edge. Compared with Figs. 11, 13, the image and
output are characterized by fewer large floes, with an increased relative fraction of
smaller floes and water/ice mixture as well. This follows the intuitive trend over the
summer season of floes becoming smaller with melting, which is also reflected by the
relative increase in water/ice mixture or subpixel size floes. Also, the bright open water
provides good contrast with the surrounding ice and in this case is also segmented quite
accurately.
The algorithm has particular difficulty in at least two situations, when the radar signature of the ice pack is bright, and when analyzing imagery from the marginal ice zone.
When temperatures drop steadily below freezing, which occurred during August
29-September 7, 1992, in the Beaufort Sea (Winebrenner et al. 1996), multiyear ice
becomes brighter than both first year ice and new ice. In general, results during cold
conditions are not satisfactory, even using the 'cold weather' switch in the algorithm,
with the floes being largely undersampled. This is likely due to the reduced overall contrast between the multiyear ice and the other ice types, which means that the segmen-