There was a new factor in the mix - GHCN has a new version 3.3. But its changes are mainly methodological, and made little difference to previous months. Both TempLS and GISS use that version. There is also a new version of ERSST, v4, but neither TempLS nor GISS (AFAIK) use that. TempLS will soon.
The difference does seem not very robust. TempLS Grid did show a significant rise, of about 0.05°C, which does indeed match February for that measure. This is working from the same data but with different (grid-based) weighting with lesser coverage. Since NOAA and HADCRUT share that characteristic, I think they might show May somewhat warmer.
HADSST3 also showed a rise in SST. So I wondered, if so much is rising, there must be something one could identify in going from April to May which would particularly affect the GISS/TLS indices. So, below the fold, I'll show a map and table
Here is an active WebGL plot of the differences, May anomaly - April, by stations (red dots) with readings in both months. The spacing between is linearly interpolated and shaded. The color scale is highly non-linear, with the extremes being quite large.
The cool spots are in the Arctic, Antarctic, and particularly NE Siberia. The ocean warms by a small, fairly uniform amount. The colors don't give a particularly good quantification of the temperature differences at the extremes. Here is a table of the top 10 cooler GHCN locations, by anomaly difference:
|-5.82||77.72||104.3||GMO IM.E.K. F|
|-4.38||36.32||-119.64||HANFORD 1 S|
It's heavily weighted toward N and NE Siberia. I've added a facility, which I'll formalise, to the NCEP/NCAR map. The table on the right allows you to choose a day, by mousing over and tracking what it says just NE of the globe - click when you have what you want. It now allows you to choose the 0'th on the month, for Jan-May 2015, and you get the map of the monthly average anomaly. Selecting 2015-5-0, I get the May average map. Here's a plot of the Arctic region:
The reanalysis certainly shows cold anomaly in the Arctic, but nothing special in NE Siberia, and it was cold there in April too. That seems to be the main difference between GHCN and reanalysis.
Update. Here's another interesting table. This time I've weighted the anomaly differences to show their contribution to the TempLS sum. So when it says Qiqihar -0.00149, that means Q contributed that amount to the May-April anomaly difference. The weighted order is different to the absolute order, and the amounts are fairly small. The Antarctic now look more significant than the Siberian. The discrepancy between NCEP and GISS/TLS is of order 0.05°C, so none of these if changed would make more than 1/30 of that.