A really good paper has been published online a couple of days ago on Nature, called September Arctic sea-ice minimum predicted by spring melt-pond fraction. It's really good because it's interesting, short, and it confirms what I've been suspecting for a while now. And when a paper confirms what one is suspecting, it must be really good, right?
All joking aside, the paper by Schröder et al. presents evidence that melt ponds play a very important role at the start of the melting season, to the point that it can heavily influence the September minimum. The last two melting seasons actually proved to be a great lesson in this respect. 2012 had a really good* start to the melting season, so good that when bad weather showed up, it didn't really slow down sea ice loss, the trend lines just kept dropping (low sea ice volume also played a role, of course). The reverse was true in 2013: cold and cloudy weather during the first half of the melting season caused a lagged response during the short periods when the Sun and higher temps finally got to the ice.
What Schröder et al. did, was develop a melt pond model and incorporate it into the larger Los Alamos sea-ice modelcalled CICE. Here's what they came up with: