The data of CRUTEM3 extend far into the sea; and the data of HadSST2 extend far inland. Grid boxes with only a tiny fraction of land are included in CRUTEM3, and grid boxes with only a tiny fraction of sea are included in HadSST2. The temperatures of these grid boxes are not very reliable, especially for HadSST2. In the merged HadCRUT3 dataset these boxes are often weighed strongly due to the low variability.

To get correct trends in Europe (with a convoluted coast line) I have constructed a new combination of CRUTEM3 and HadSST2, in which the overlap is weighed according to the fraction of land and sea. This fraction has been computed from the 0.5° CRU TS 2.1 dataset land/sea mask. It extends a little too far into the sea, but this has the good property that it downplays the unreliable coastal HadSST2 data.

The combined dataset therefore consists of

• CRUTEM3 wherever HadSST2 is undefined
• HadSST2 wherever CRUTEM3 is undefined
• an area-weighed average where both are defined

Note that near islands this can give rise to spurious signals when ships pass by. Use with care.