Hi Greg,
I think we might have to agree to disagree about the implications of your analysis.
The metadata are far more complete than your description of them allows. They show changes in the observing system with time scales measured from years to decades. I think that the ‘structure’ you see in the raw data is contaminated by artificial biases in the data, not just during the 4 year ‘glitch’, although that is their most obvious manifestation, but in the whole of the record.
The raw data can be thought of as a true climate signal (T) plus a bias term (B). i.e Raw data = T+B (for the sake of temporary simplification I’ll discount spatial sampling and other measurement errors although obviously they are important). What you see in the raw data is variations arising from T and from B. You say that the similarity between T+B and B “seems to be pushing the bounds of coincidental similarity”, but I think that by definition there ought to be a similarity. The proper comparison would be between T and B.
Unfortunately, we don’t have T and only estimates of B. The estimated B’s from HadSST3 (remember there are more than one) don’t look like the estimated T’s.
I can understand your uneasiness with the adjustments to the data: the adjustments are large and there are uncertainties. That is why I have been trying to persuade other SST data set developers to address the problem and why I’m keen to see how the Berkeley team approach the problem.
Best regards,
John