I've been playing with a slightly different idea - which is to try to systematically correct for ice movement backwards over time from the current date, then incorporate knowledge from previous looks, using the same source, at the same ice that now occupies each pixel. I'm hoping that it might yield a picture that has more information content than a single look from a single angle, with a single surface scatter/cloud pattern, etc.
Even working from low granularity data over a small area there's a large amount of processing involved - I've been trying to compute the day-to-day transformation + remapping using an FPGA + modified machine-vision logic - but so far all I've got from it is a flashing LED and a headache. Will let people know if it ever comes to anything...