2 Identifying Ice Floes and Computing Ice Floe Distributions in SAR Images
31
in the marginal ice zone (72 oN and southward) have extremely variable and inaccurate
segmentation due to the highly variable ice-ocean conditions and resulting backscatter; within the central pack, backscatter variations are localized and strongly affect segmentation. For the latter condition, note the solidly red third frame from the top in Fig.
15a, where the SAR image is basically featureless, and the frame centered at 78°N, where
there is a large area of substrate visible, and where the SAR image is comparatively bright
and more winter-like than surrounding images. For this data take, the solid red frame
was not included in the final selection, but the frame near 78°N was included, since it
was felt that the measured floes were accurately segmented. All the frames from 72 oN
and southward were eliminated from further analysis.
To briefly summarize these preliminary results, the floe algorithm generates good quality accurate measurements of thicker floes of all sizes during the summer when the ice is
dark from wetness, the open water is bright, and the floes are readily identifiable to the human
eye. Environmental factors which alter the ice and ocean backscatter away from this condition make it problematic for the algorithm to accurately segment the ice/ocean cover.
2.6
Summary and Conclusions
We have presented the restricted growing concept algorithm that establishes separateness among objects while preserving their original sizes and shapes. We have also
described an implementation based on probabilistic labeling and mathematical morphology, aimed at and tested successfully on synthetic aperture radar (SAR) sea ice
imagery. Our implementation is (1) extensible, since it can include any number of classes of objects in an image, (2) flexible, since it can be altered such that the roles of object
and background pixels can be switched to focus the user's interest on a certain class of
pixels, (3) economical, since it does not sacrifice small ice floes for separateness, (4) fully automated except for the requirement of the user to specify the object class, and (5)
computationally efficient.
Moreover, we have designed a technique for computing ice floe size distribution and
ice coverage statistics. This technique is able to segment a SAR sea ice image into floes
and non-floes, it is able to achieve floe identification and separation using our implementation of the restricted growing concept, it is able to analyze and discard irregular
floes, and it is able to distinguish zones of subresolution ice from open water regions.
We have shown preliminary results from a study of summer floe size distribution in
the Beaufort and Chukchi Seas and how the floe results are strongly dependent on ice
and environmental conditions during this dynamic season.
Acknowledgments: All the authors acknowledge the support of Robert Thomas, Manager of Polar Programs (Code YSG) at NASA. The work by L.-K. Soh and C. Tsatsoulis
was funded in part by NASA grant NAGW-3043. Soh and Tsatsoulis would like to
acknowledge the assistance provided by Donna Haverkamp in developing the floe filtering procedure. Holt performed this work at the Jet Propulsion Laboratory, California Institute of Technology, under contract to NASA. Holt would also like to acknowledge the excellent programming support of Remi Roques and Marie-Helene Rio, both
graduate students from the Ecole Nationale Superieure de l' Aeronautique et de l'
Espace in Toulouse, France, who performed their end-of-year study projects at JPL.
31
in the marginal ice zone (72 oN and southward) have extremely variable and inaccurate
segmentation due to the highly variable ice-ocean conditions and resulting backscatter; within the central pack, backscatter variations are localized and strongly affect segmentation. For the latter condition, note the solidly red third frame from the top in Fig.
15a, where the SAR image is basically featureless, and the frame centered at 78°N, where
there is a large area of substrate visible, and where the SAR image is comparatively bright
and more winter-like than surrounding images. For this data take, the solid red frame
was not included in the final selection, but the frame near 78°N was included, since it
was felt that the measured floes were accurately segmented. All the frames from 72 oN
and southward were eliminated from further analysis.
To briefly summarize these preliminary results, the floe algorithm generates good quality accurate measurements of thicker floes of all sizes during the summer when the ice is
dark from wetness, the open water is bright, and the floes are readily identifiable to the human
eye. Environmental factors which alter the ice and ocean backscatter away from this condition make it problematic for the algorithm to accurately segment the ice/ocean cover.
2.6
Summary and Conclusions
We have presented the restricted growing concept algorithm that establishes separateness among objects while preserving their original sizes and shapes. We have also
described an implementation based on probabilistic labeling and mathematical morphology, aimed at and tested successfully on synthetic aperture radar (SAR) sea ice
imagery. Our implementation is (1) extensible, since it can include any number of classes of objects in an image, (2) flexible, since it can be altered such that the roles of object
and background pixels can be switched to focus the user's interest on a certain class of
pixels, (3) economical, since it does not sacrifice small ice floes for separateness, (4) fully automated except for the requirement of the user to specify the object class, and (5)
computationally efficient.
Moreover, we have designed a technique for computing ice floe size distribution and
ice coverage statistics. This technique is able to segment a SAR sea ice image into floes
and non-floes, it is able to achieve floe identification and separation using our implementation of the restricted growing concept, it is able to analyze and discard irregular
floes, and it is able to distinguish zones of subresolution ice from open water regions.
We have shown preliminary results from a study of summer floe size distribution in
the Beaufort and Chukchi Seas and how the floe results are strongly dependent on ice
and environmental conditions during this dynamic season.
Acknowledgments: All the authors acknowledge the support of Robert Thomas, Manager of Polar Programs (Code YSG) at NASA. The work by L.-K. Soh and C. Tsatsoulis
was funded in part by NASA grant NAGW-3043. Soh and Tsatsoulis would like to
acknowledge the assistance provided by Donna Haverkamp in developing the floe filtering procedure. Holt performed this work at the Jet Propulsion Laboratory, California Institute of Technology, under contract to NASA. Holt would also like to acknowledge the excellent programming support of Remi Roques and Marie-Helene Rio, both
graduate students from the Ecole Nationale Superieure de l' Aeronautique et de l'
Espace in Toulouse, France, who performed their end-of-year study projects at JPL.