3 Role of SAR in Surface Energy Flux Measurements Over Sea Ice
55
expected to produce a variable scattering return due to the atmospheric forcing on the
surface. Evidence for this relationship has been documented in the preceding sections
and is discussed in detail elsewhere (Barber et al.1994, 1995).
Pairs of ERS-1 scenes within the winter period can be used for ice segmentation using
the principle of variable scattering from first-year ice and stable scattering from multiyear ice. One can expect a detectable change in first-year ice scattering with changing
atmospheric temperatures and/or presence of cloud cover versus clear skies over the
sea ice surface (Barber and Thomas 1997), even when atmospheric temperatures are
well below the freezing point.
Difference images over the seasonal transition period also appear to have potential
for segmenting the sea ice surface into first-year, multiyear and rubble forms of sea ice.
The techniques of change detection outlined above were applied to repeat-pass image
pairs spanning the transition from winter to melt onset conditions (Thomas 1996). The
images in each pair were mosaics of repeat-pass images covering the SIMMS'95 study
area. Each pair is separated by 35 days with an interval of approximately 3 days between
image pairs (each successive pair overlapping the coverage of the previous pair by
between 40% and 80%). Mosaics were calibrated to a scattering coefficient according
to Olmsted (1993). The first image in each pair was acquired during the winter season.
Consequently, variation in ice signatures between these winter images is minimal (less
than 1 dB for multiyear ice) and the decibel values for multiyear ice can be considered
to represent the winter mean.
Once image calibration to decibels had been achieved, change detection images were
generated, one for each pair of repeat-pass images covering the melt period of 1995.
Image subtraction was used to create change detection images, with the spring period
mosaic being subtracted from the winter period mosaic in each case. The resultant
change detection images were classified into three change classes: ~O"0 > +1 dB, represented by red; negative change ~O"0 > -1 dB, represented by blue; and "no change:" ~O"0
between +1 and -1 dB, represented by green (Fig. 12). This three-way classification was
chosen as a simple way of illustrating seasonal progression of ERS-1 0"0 as a function of
ice type.
Bivariate histograms of the two image mosaics in each repeat-pass set were generated to illustrate seasonal ~O"0 as a function of magnitude in 0"0. An agreement line bounded on either side by a 1 dB envelope was superimposed over each bivariate histogram
(Fig. 12). This provided a graphical representation of 0"0 changes outside the "no
change" ~O"0 class (-ldB to +1dB).
Difference images and the corresponding bivariate histograms in Fig. 12 show the
segmentation classes which are available through different portions of the seasonal evolution (Thomas 1996). The magnitude in the difference classes has been restricted to
segmentation of first-year, multiyear and perhaps rubble ice classes. The T1 contrast
set (April 24 and May 29, 1995) shows that considerable change occurs in first-year ice
forms which have a scattering magnitude less than about -19 dB. The higher-scattering first-year surfaces (i.e. rough) and the multiyear forms show no significant change
(difference in scattering is within the ±ldB threshold) (Fig. 12). Unlike first-year ice, significant fluctuations in multiyear ice 0"0 do not occur during the winter period because
multiyear ice 0"0 is controlled by the bubble structure in hummock ice.
The May 2 and June 6 contrast set (T2) is the first pair where the second image in the
repeat-pass pair is clearly within the early melt period. This is evident in the dramatic
55
expected to produce a variable scattering return due to the atmospheric forcing on the
surface. Evidence for this relationship has been documented in the preceding sections
and is discussed in detail elsewhere (Barber et al.1994, 1995).
Pairs of ERS-1 scenes within the winter period can be used for ice segmentation using
the principle of variable scattering from first-year ice and stable scattering from multiyear ice. One can expect a detectable change in first-year ice scattering with changing
atmospheric temperatures and/or presence of cloud cover versus clear skies over the
sea ice surface (Barber and Thomas 1997), even when atmospheric temperatures are
well below the freezing point.
Difference images over the seasonal transition period also appear to have potential
for segmenting the sea ice surface into first-year, multiyear and rubble forms of sea ice.
The techniques of change detection outlined above were applied to repeat-pass image
pairs spanning the transition from winter to melt onset conditions (Thomas 1996). The
images in each pair were mosaics of repeat-pass images covering the SIMMS'95 study
area. Each pair is separated by 35 days with an interval of approximately 3 days between
image pairs (each successive pair overlapping the coverage of the previous pair by
between 40% and 80%). Mosaics were calibrated to a scattering coefficient according
to Olmsted (1993). The first image in each pair was acquired during the winter season.
Consequently, variation in ice signatures between these winter images is minimal (less
than 1 dB for multiyear ice) and the decibel values for multiyear ice can be considered
to represent the winter mean.
Once image calibration to decibels had been achieved, change detection images were
generated, one for each pair of repeat-pass images covering the melt period of 1995.
Image subtraction was used to create change detection images, with the spring period
mosaic being subtracted from the winter period mosaic in each case. The resultant
change detection images were classified into three change classes: ~O"0 > +1 dB, represented by red; negative change ~O"0 > -1 dB, represented by blue; and "no change:" ~O"0
between +1 and -1 dB, represented by green (Fig. 12). This three-way classification was
chosen as a simple way of illustrating seasonal progression of ERS-1 0"0 as a function of
ice type.
Bivariate histograms of the two image mosaics in each repeat-pass set were generated to illustrate seasonal ~O"0 as a function of magnitude in 0"0. An agreement line bounded on either side by a 1 dB envelope was superimposed over each bivariate histogram
(Fig. 12). This provided a graphical representation of 0"0 changes outside the "no
change" ~O"0 class (-ldB to +1dB).
Difference images and the corresponding bivariate histograms in Fig. 12 show the
segmentation classes which are available through different portions of the seasonal evolution (Thomas 1996). The magnitude in the difference classes has been restricted to
segmentation of first-year, multiyear and perhaps rubble ice classes. The T1 contrast
set (April 24 and May 29, 1995) shows that considerable change occurs in first-year ice
forms which have a scattering magnitude less than about -19 dB. The higher-scattering first-year surfaces (i.e. rough) and the multiyear forms show no significant change
(difference in scattering is within the ±ldB threshold) (Fig. 12). Unlike first-year ice, significant fluctuations in multiyear ice 0"0 do not occur during the winter period because
multiyear ice 0"0 is controlled by the bubble structure in hummock ice.
The May 2 and June 6 contrast set (T2) is the first pair where the second image in the
repeat-pass pair is clearly within the early melt period. This is evident in the dramatic