....
But what we are interested in is how the March extent compares to the September extent. That is, for any two years, if the extent is lower in March in one year, is it also lower in September in that year? That is true only 36% of the time. (I wrote a program to compute these numbers for me. This was so low that I didn't believe it the first time and checked it again by hand.) This is low enough to suggest a negative correlation between March extent and September extent, meaning that after accounting for the fact that on average the extent decreases every year, a low extent in March actually implies a high extent in September.
....
That does seem counterintuitive at first, though it makes sense if you think of the open water losing more heat than it would if ice-covered. By the same reasoning, I would expect a (weak positive) correlation between the date of the maximum and the September minimum (i.e., later max leads to higher min).
What's interesting about the latter relationship is the possibility of negative feedback under certain conditions. Think of a point in space and time where the ocean, on average, gains as much heat from insolation as it loses from radiation (and evaporation etc - the main idea being a balance of heat energy over the diurnal cycle). Join all such points together to form a loop around the seas of the northern hemisphere. It's neither a latitude line nor an isotherm, though influenced mainly by insolation and water temperature. During the spring, the loop gradually moves northward: the sun's increasing warming power leads gains in SSTs.
Now, where is the ice edge in relation to this moving line? Let's say it's farther north, and extent is anomalously low. Less insulation from ice cover, more heat radiated from water to air: negative feedback; faster ice growth (or slower melt). Time passes, the equilibrium loop continues farther north and overtakes the ice edge. Now we have either of two conditions. If extent anomaly has remained low, there is more open water than normal to absorb the solar energy. Positive feedback, the familiar albedo flip. So we would expect - other things being equal - that anomaly measures continue downward and a low minimum obtains.
But what if the ice has grown rapidly, taking extent anomaly to the high side? That means more ice than normal to reflect sunlight back into space - under conditions where heat gain from insolation would otherwise exceed radiative heat loss. So negative feedback would continue later into the season. That might have happened in 2014, when extent rose rapidly to a late peak. Not knowing where the equilibrium loop is and how it moves, this is only speculative. But it could have been a contributing factor to the slow melt last year.