3 Role of SAR in Surface Energy Flux Measurements Over Sea Ice
51
3.4.1
Detecting Melt Onset with SAR
The evolution of the multiyear ice ERS-l signature offers significant possibilities for the
estimation of sea ice physical state changes. Unlike first-year ice, whose signatures are
highly variable during the winter period, the multiyear ice signatures are influenced
primarily by stable volume scattering from hummock ice (Onstott 1992). The multiyear
ice signature is an appropriate indicator of sea ice physical changes since the initial
stages of melting are accompanied by a steep, clean drop in 0"0 caused by liquid water
in the snow cover over the ice. The magnitude of the drop is dependent upon the snow
depth in which signal absorption occurs. A 2% water by volume in a lO-cm-thick snow
cover is sufficient to produce a 0"0 drop of 10 dB (Winebrenner et alI994). This response
creates an effective means of detecting the onset of melt within the marine cryosphere.
Multiyear ice 0"0 as an indicator of melt onset has been investigated using a combination of buoy temperature records and ERS-l data acquired over the Beaufort Sea
(Winebrenner et al. 1994). Analysis of multiyear ice 0"0 frequency histograms reveals
significant shifts near the onset of melt from high to low 0"0 magnitudes due to the
absorption of microwave energy in the damp snow cover. Melt onset has been estimated as the date on which the temporally interpolated location of the histogram peak
crosses a predetermined threshold from high to low 0"0 magnitude. The value of this
threshold is dependent on the scattering properties of multiyear ice and the thickness
of its snow cover. A threshold of -14 dB has been used successfully by Winebrenner et
al. (1994) to indicate melt onset from ERS-l imagery of the Beaufort Sea.
3.4.2
Scene Segmentation and Classification
An ongoing problem in the machine classification of SAR data from sea ice is an effective method of image segmentation and class labelling (i.e. classification). Various
researchers have investigated approaches that range from neural networks to texture
classification/segmentation approaches (see chapters in this text). Operational use of
machine-assisted segmentation and classification are critical if operational sensors
such as RADARSAT are to find effective application in areas such as tactical marine
navigation (Ramsay et al. 1993) and or global climate change (Kwok and Cunningham
1994).
Time series analysis illustrates that the temporal evolution of the scattering coefficient may be an effective tool for both image segmentation and classification. Although
the analyses presented here are limited by the temporal coverage of the ERS-l sensor,
the prospect of segmentation and classification are greatly enhanced with the repeat
coverage of RADARSAT ScanSAR products (i.e. daily coverage of the Arctic basin) that
are now available operationally.
3.4.2.1
Classification
During the SIMMS'93 experiment we investigated the utility of the time series evolution of the scattering coefficient for classification of various ice types. Six sites were
51
3.4.1
Detecting Melt Onset with SAR
The evolution of the multiyear ice ERS-l signature offers significant possibilities for the
estimation of sea ice physical state changes. Unlike first-year ice, whose signatures are
highly variable during the winter period, the multiyear ice signatures are influenced
primarily by stable volume scattering from hummock ice (Onstott 1992). The multiyear
ice signature is an appropriate indicator of sea ice physical changes since the initial
stages of melting are accompanied by a steep, clean drop in 0"0 caused by liquid water
in the snow cover over the ice. The magnitude of the drop is dependent upon the snow
depth in which signal absorption occurs. A 2% water by volume in a lO-cm-thick snow
cover is sufficient to produce a 0"0 drop of 10 dB (Winebrenner et alI994). This response
creates an effective means of detecting the onset of melt within the marine cryosphere.
Multiyear ice 0"0 as an indicator of melt onset has been investigated using a combination of buoy temperature records and ERS-l data acquired over the Beaufort Sea
(Winebrenner et al. 1994). Analysis of multiyear ice 0"0 frequency histograms reveals
significant shifts near the onset of melt from high to low 0"0 magnitudes due to the
absorption of microwave energy in the damp snow cover. Melt onset has been estimated as the date on which the temporally interpolated location of the histogram peak
crosses a predetermined threshold from high to low 0"0 magnitude. The value of this
threshold is dependent on the scattering properties of multiyear ice and the thickness
of its snow cover. A threshold of -14 dB has been used successfully by Winebrenner et
al. (1994) to indicate melt onset from ERS-l imagery of the Beaufort Sea.
3.4.2
Scene Segmentation and Classification
An ongoing problem in the machine classification of SAR data from sea ice is an effective method of image segmentation and class labelling (i.e. classification). Various
researchers have investigated approaches that range from neural networks to texture
classification/segmentation approaches (see chapters in this text). Operational use of
machine-assisted segmentation and classification are critical if operational sensors
such as RADARSAT are to find effective application in areas such as tactical marine
navigation (Ramsay et al. 1993) and or global climate change (Kwok and Cunningham
1994).
Time series analysis illustrates that the temporal evolution of the scattering coefficient may be an effective tool for both image segmentation and classification. Although
the analyses presented here are limited by the temporal coverage of the ERS-l sensor,
the prospect of segmentation and classification are greatly enhanced with the repeat
coverage of RADARSAT ScanSAR products (i.e. daily coverage of the Arctic basin) that
are now available operationally.
3.4.2.1
Classification
During the SIMMS'93 experiment we investigated the utility of the time series evolution of the scattering coefficient for classification of various ice types. Six sites were