Arctic Sea Ice Resampling

[WilliamOlsen]

Hello All,

I am using some Landsat 8 imagery of Arctic Sea Ice and am trying to create polygons of the individual ice chunks. The problem is the small sections connecting  the larger pieces.  They act as an "ice isthmus" and I can't seem to sample them out. 

Does anyone have a good method of re-sampling the raster to remove these ice connectors while preserving the basic shape of all ice chunks?  I have created an 11 band composite image using the image analysis tools and then an ISO Unsupervised Classification binary raster consisting of Ice and Water.  The Image classification tools do a great job of classifying cells. (Below is a swipe screenshot showing ice as blue)  I've then tried a number of the raster generalization methods but most will remove the smaller ice chunks or reduce the size of the larger ice floes.  Any help would be appreciated!  I can send the TIFF if it helps!  Thanks, Bill.

[larryzhang]

willianm,

To effectively detect individual ice chunks from Arctic Sea Ice with LANDSAT 8, some special bands should be used carefully. For example, band 1 is a new coastal/aerosol band, which can be used with two other bands for closer investigations of coastal waters and estimating the concentration of aerosols in the atmosphere. However, in most cases, models will use the radiance values of this band to differentiate ice chunks from water as well.

After 'traditional' spectral classification in combination with false color imagery (654, 764, and/ or 753), you can get mostly what you need (80-90%).

And then, convert to vector polygons for generalization of post-processing, including aggregate, region group, boundary clean….

[larryzhang]

In practice, RADARSAT images are widely used to effectively detect ice objects, because C-band is useful for ocean and ice features' detection.

Pls refer to Satellite Characteristics - Canadian Space Agency

[WilliamOlsen]

Thanks so much for your reply Larry...  I will check out the RADARSAT images. 

And just to clarify my issue, I am having very good luck differentiating ice from water.  It is the raster generalization process (Boundary Clean, Aggregate, etc) that I'm having trouble with.  I can resample the ice pixels and eventually get them to split up into individual pieces but I'm losing a lot of  detail in the processing.  Do you have a good suggestion for the the order of the cleanup operations?

Thanks again,

Bill

[larryzhang]

Not to lose the small details (i.e., to keep small details or moved into larger polygons), pls use the gp tool 'Aggregate Polygons' first with proper parameters like  'The minimum area & The minimum size', etc at ArcGIS Help (10.2, 10.2.1, and 10.2.2) . Easy to control Python;

And then run 'Smooth Polygon', which should use 'Smoothing Tolerance' with over-100 -m (30-m resolution X3, at least), BEZIER_INTERPOLATION (not PEAK), etc