To collect additional data for a better estimation model, all ATM flights (7) over Greenland during May 2007 were surveyed, and 21 sites/orbits for which there were good corresponding MISR image data in all 3 cameras were selected.
Data models were made from each of these sites, 2 of which are shown below. The images shown are the average of the Cf, An, and Ca camera views.
All 21 models were then combined into a global model which could accept a much wider range of input pixel values.
A total of 5292 MISR images were processed to produce 1764 2048x512 blocks of estimated roughness data covering Greenland during April-July of 2007. A gridded subset of the combined data model was used for estimation, which increased the speed of calculations by about 4x. Even so, the entire image processing of these results took about 5 days of cumulative computer time on a 2x2.4 GHz Macintosh running optimized C programs. Below is an example of 9 contiguous blocks of an orbit over northern Greenland. At left is a visible image consisting of the average of the Cf, An, and Ca camera views. At center and right are the estimated roughness values in cm.
Shown next are the distributions of all pixels of all 9 blocks in image space (grey). In color are the roughness values of estimated pixels which were within 0.1 unit of the combined data model. About 85% of all pixels of all 1764 blocks were estimated, although for many blocks 100% of all pixels were estimated.
Shown next are the monthly average roughness values over Greenland for April-July. Note that due to weather and the orbital geometry of the spacecraft, not all parts of Greenland could be covered each month.
Shown next is the 4 month total average. The color plot scale has been altered to reflect the cumulative distribution of values, so that there are equal map areas per color.
Finally, the 4 month averages for a single site (the Jakobshavn glacier on the central western coast of Greenland, the bright white/red feature at lower left of each plot). The bright (rough) area north of the glacier during April is due to the high variability of weather (e.g. snowfall). Otherwise, systematic changes in the roughness of the coast, the edge of the interior ice sheet, the glacier, and its outflow can clearly be seen over 4 months.
Future work will apply the technique to other snow and ice surfaces, and will automatically track the motion of features having low image contrast by virtue of their calculated 3d surface properties. The technique developed here is of general applicability, presuming that it can produce results of similar accuracy on a variety of surfaces other than snow and ice.
©Sky Coyote 2009