Yesterday, in a spot check, we observed that Torok and Nicholls had acknowledged a substantial inhomogeneity in the Low Head lighthouse series, used in Jones et al 1990. Given this problem in one of the 49 series, this begs the question of what the other 48 series look like. Here are quite a few plots of individual stations plus some spaghetti graphs.
I’ve identified 37 of the 49 Jones series in the Torok and Nicholls 1996 archive. Through some patient cross-checking I’ve made a codex linking the Australian BOM ids to GHCN ids. I was able to match location and name for 36 of the 37 series (I couldn’t locate a GHCN number for Maryborough, Queensland although there is one for Maryborough, Victoria). I have 4 series available for each series:
the annual average from monthly data in the Torok and Nicholls archive (“BOM monthly average”)
the adjusted Torok and Nicholls value (annual)
the annual average from monthly data in the GHCN v2 archive
The Jones et al 1990 version ties almost exactly to the annual average of BOM monthly data for the period of overlap – the Jones period of 1930-1988 being shorter than available (raising the question of why Jones selected exactly this period – this is the Team, after all). In turn, both generally match GHCN data. When you look at the graphs, notice that the data all ends in the early 1990s – why has GHCN been unable to organize itself to update “rural” Australian data – another interesting question.
When you start looking through the various station versions, it’s hard not to ponder where one finds firm ground amidst the various adjustments – there’s a reason for all the adjustments, but the adjustments are so large relative to the size of the presumed “signal” that Step 1 in all of these studies surely needs to be a systematic description in an organized form of the adjustments. You’ll see what I mean better by looking at some examples.
Here’s a plot of the 4 versions for Omeo – wherever that is. You can see that the GHCN data (red) closely matches the average taken from BOM monthly data (blue), which matches the Jones version (black) during the 1930-1988 overlap. Torok has made a substantial downward adjustment of 19th century data, which is not reflected in the GHCN raw version supposedly used NOAA gridded (I’ve not checked against the GHCN adjusted version yet.) Torok and Nicholls 1996 describe a variety of plausible reasons for adjusting 19th century data, pointing out, in addition to the usual problems of changing instrumentation, problems distinctive to rural Australia: thermometers being seized by dingoes, taken by crows and being smashed by angry wives.
Here’s another one from Richmond Post Office where the Torok adjustment has a little different effect. In this case, the Torok adjustment increases values in the 1930s somewhat. Looking at the graphic as a whole, one does not leave with a sense of a strong trend. Nonetheless, this site has a strong positive trend in the Jones et al study, accentuated somewhat by the fact that it starts with an exceptionally low value in 1930 and ends with a high value in 1988. Again one of those providential coincidences so characteristic of Team climate science.
Here’s another series with a similar pattern – again no particular overall trend. However, in the Jones version, there is a positive trend, to which the downspike at the opening of the series contributes substantially. Note that the Torok adjustment here eliminates the opening downspike in the Jones version.
Now for something even punchier. Here’s a series with a very strong upward trend. Torok and Nicholls adjusted this series a bit, but, even after the adjustment, there was still a strong upward trend. Since other series don’t have this trend, how can one assume that this particular trend is not due to some unannotated inhomogeneity? I don’t know what the inhomogeneity is here, but it sure looks to me like there’s something wrong with this “High Quality” series.
Here’s another series with no particular trend, in which GHCN, Jones, Torok and BOM average versions are all virtually identicl in the overlap period, while the Torok adjustment lowers the 19th century values.
While many have no trends, here’s another series with a strong trend – is the trend real? Right now, I don’t see how you can assume that it’s real.
What do things look like an overall basis? Here I’ve made a spaghetti graph of the above 36 stations from the Torok archive (the GHCN plot looks very similar). Tree rings start to look pretty good. The Jones et al network period is shown by dashed lines and pretty much coincides with this graphic.
After adjustment, the “High Quality” Torok and Nicholls network looks like this – the main difference being the elimination of higher late 19th century values due to adjustments for measurement changes.
Here’s a plot of the average of the data in the 4 networks, showing clearly the effect of the Torok and Nicholls adjustments on 19th century values and what seems to be a somewhat opportunistic choice of period in Jones et al 1990.