Processing with Aug-Oct instead of Jun-Oct

These images compare MODICE output for our Karakoram test scene, using our persistent ice algorithm, but beginning with August 1 instead of June 1. The idea is that we could speed things up a lot (reduce MODSCAG time by 40%) if we don't have to process June and July data, and we can get satisfactory results.

Differences between these processing options will only result in more permance ice using the August option than the June option. So the differences in row 3 are all places where August processing left ice, but June processing removed it. Based on these two years of processing this scene, August processing yields something on the order of 10% more persistent ice than when we run processing starting in June.

But the real question: how much of this "extra" ice is real? So I compared it to the GLIMS, and it looks like it depends. For just the 2001 processing, 1045 of the 11,262 "extra" ice pixels (9.3%) correspond to GLIMS ice. But it looks even better for the 2001+2002 picture: 1875 of the 8,802 "extra" ice pixels (21.3%) correspond to GLIMS ice. Take a look at the "gold" pixels in the difference image. The percentages are deceiving, because they are taken as a total of the whole scene, and we don't have full GLIMS coverage for the whole scene. I think the match to GLIMS is actually far more improved than these numbers may indicate.

Based on this comparison, I'd like to suggest that we only run Aug-Oct, for 2 reasons: 40% shorter processing required, with significant improvements in the MODICE corrspondence to GLIMS ice.

MODICE, June-Oct 2001	MODICE, June-Oct 2001+2002
MODICE, Aug-Oct 2001	MODICE, Aug-Oct 2001+2002
MODICE difference, 2001 (11,262 pixels, 10.2% of persistent (jun-oct) ice)	MODICE difference, 2001+2002 (8,802 pixels, 9.8% of persistent (jun-oct) ice)