[Peter Thorne] Top is Phil's data which is to nearest whole percent (x100), next is IGBP as a fractional coverage and then last is the difference field as a fractional measure (Phil minus IGBP). There are some fairly large differences throughout. In the poles these are because of different classification procedures I should think. Differences beyong the 3rd d.p are probably rounding (obviously). The others look kinda scary so I'm advocating adopting an Ostrich position ...
As Simon intimated earlier in this thread in HadCRUT3 the merge is going to be based upon estimated error rather than fractional coverage.
Path of least resistance is install Phil's dataset back in which we've been using for a long time and run and just ignore any error messages, send UEA everything (complete rather than solely post-1990 merged hadcrut2 and hadcrut2v files) this month and then keep running it with a mask we know to be sub-optimal because the whole thing is going to be replaced on a timescale of months.
Skeptics Deliver An Open Letter to Pope Francis
3 hours ago