I have noticed that DMI's model and the simulations by Fettweis are using different tresholds: 1mm/day vs. 5mm/day. That could explain the difference between DMI (left, 1mm treshold) and Fettweis' MAR model (right, 5mm treshold):
It makes sense, since the melt area >1mm is larger than the area >5 mm.
However, it's also a bit puzzling, because the area with melt >1 mm is above the 1981-2010 average whereas the area with melt >5mm is below the same climatological mean. That should imply a wider but weaker melt than average. It looks somewhat strange.
In addition, it becomes even more puzzling it we look at NSIDC's data. Instead of using a model, NSIDC's data are based on an analysis of passive microwave observations. Their algorithm is sensitive to the presence of wet snow on the surface, in any amount, whereas the threshold for DMI's mapping is 1mm liquid water produced -- this is actually a significant amount of melting, surely more than the amount needed to trigger the passive microwave signal.
In summary, NSIDC data should be closer to DMI (1mm) than to Fettweis' MAR model (5mm). But their graph shows just the opposite: NSDIC data are much more closer to Fettweis' results than to DMI's ones.
Moreover, it's even more puzzling if we compare DMI's 'daily contribution to the surface mass balance' map (right) with their own 'areas where melting has taken place' map (left):
The divergence at NE Greenland is apparent even if we compare DMI data vs. DMI data!
Taking all this into account, I stand by my previous statement: it seems that something is wrong with DMI's 'area where melting has taken place' graph:
And NSIDC's and Fettweis' graphs look like the correct ones: