You may have read Ken Stewart’s excellent blog on the official Australian temperature record. With the publication of the “adjustments.xls” file of the offical adjustments to the raw data in the ACORN-SAT dataset, as reported on JoNova’s blog, there has been a flurry of work behind the scenes, so to speak.
Jennifer Marohasy has been leading the charge also to audit or at least review the practises of the Bureau of Meteorology in adjusting raw temperature to produce the synthetic ACORN-SAT series. Rutherglen in particular has been in the news for massive warming adjustment to the minimum temperature trend.
The story has gone national in The Australian with articles like Climate records contradict Bureau of Meteorology. Even I have been quoted as saying that the BoM may be “adding mistakes” with their data modelling.
Well, all of this kerfuffle has been enough to get me out of hiding and start working on some stuff. I thought it would be good to have a quality assessment method that could reliably test the ACORN data. The idea I came up with is to test the trend of the ACORN-SAT series against the trends of the raw data neighbours. Ideally, if the trend of the synthetic series exceeds the overall trend of all the neighbours, then some thing must be wrong.
The difficulty is that the neighbours all start and stop at different times and so a slightly more complex test is needed than a simple ordinary least squares regression. The answer is a panel test, or POLS. Its all explained in the reports on the first three stations below.
The test seems to work remarkably well. The ACORN-SAT minimum temperatures for Rutherglen and Deniliquin fail the benchmark residual trend of 0.05 degrees C per decade – that is they warm at a much greater rate than their neighbours. ACORN-SAT at Williamtown, by contrast, is consistent with its neighbours and the raw CDO series are not, indicating that the single large adjustment applied around 1969 was warranted.
I intend to keep working through the stations one by one and uploading them to the viXra archive site as I go. I will also release the data and code soon for those who are interested. Its almost at the stage where you enter a station number and it spits out the analysis. Wish it would write the reports too, though latex goes a great deal towards that end.