does the skill of your method improve for shorter periods Using a skill metric similar to Schroder et al., (2014) over the period 1995-2013 for looking at Sept. mean ice extent, the skill of my method improves from near zero at about 100 days lead time to above 0.6 at 50 days lead time in a rather linear fashion. I expect skill would improve further as lead time gets shorter - the influence of initial conditions is greater the shorter your lead time. I have put in a pure model result (90 days lead time) for the July SPIN; it should be out soon. I have not looked at sensitivity to change in area as my result depends to some degree on where the ice concentration has changed.

Meanwhile, the Andrew Slater model as collapse in the last few days! I expect/hope many of you might have thought this to be an error - it was. As noted by Hans, the NSIDC ice concentration processing had an issue this morning, but was later fixed ... my forecast is back to where it should be.