Roman has an interesting mathematical post here. Last night I finished a gridded GHCN global dataset based on his work but wasn’t able to write it up. That late at night it wouldn’t make sense anyway, but there will be something new on it shortly. His new post explains some of the reasoning behind this approach.
the Air Vent 
Jeff, the use of the word “anomaly” in the title of your post is a bit misleading. It is the misconception that Tamino seems to have. Using monthly offsets does NOT remove any seasonal effects from the combined temperature sequence. They are eliminated from the residual station series whereas the single offset leaves some of the seasonality still present in those residuals terms from which the standard errors are calculated for error bar estimates.
What it does do is to prevent the missing values from spuriously passing on seasonal effects to the end result.
#1, I’ve struggled with a proper name for your method. Knitted anomaly, monthly temperature offset is pretty dry. Any other ideas?
How about “seasonal offset matching”?
The calculations method depend on matching stations pairwise by month to put comparisons between them on the same level.
That works. I hope people take the time to look closely at this method. It would be a good CA post.
Minimal-Seasonality-Residue Aggregation Method for Incomplete Station Records
Or Spliced Spaghetti Sans ExcesS Seasonality Artifacts
Or SemiOptimal Splicing for Similar Semicomplete Station Records
or
Roman’s Holiday (from Seasonality Residue in Merged Temperature Records with Missing Months)
Anomalies Optional…
Hard to capture the traits accurately with only 2 or 3 or even 4 word name…
Heck, I don’t even ‘get it’ yet.
RR
A miracle has happened!
🙂
#6 Holy crap- Guh, yup!!
——–
tamino // February 26, 2010 at 1:00 am | Reply
I’ve looked at the annual cycle for different stations in the grid containing Skikda, as used by RomanM in his example of the difference between computing a single offset for each station, and computing 12 separate monthly offsets for each station.
The differences in the sizes of the annual cycles were larger than I expected. Hence on that basis, I now think RomanM is right, that using 12 separate monthly offsets is a better way to combine station records for a gridwide average to incorporate into a global average, than using a single offset. That doesn’t mean the single-offset method is bad, just that the separate-monthly-offsets method is better.
And it in no way invalidates the results of the analysis, which still shows conclusively that station dropout did not create a false warming trend, and GISS adjustments did not exaggerate warming they reduced it.
———
When this becomes the standard, they better give credit.
Good for him.
Re: steven mosher (Feb 26 04:03),
I also give him credit for doing that reasonably quickly.
I would presume that he ran my script so maybe the fact that I posted the script and data allowing others to check the work will also rub off on him.
Just desired to say that you’ve some awesome content on your blog. If it’s OK I would like to use some with the information you provided on my web site. If I link back again to your website would it be OK to do so?