The warming bias in Australia’s ACORN-SAT maximum dataset is largely due to just two adjustments.
Last week’s Report of the Technical Advisory Forum’s review of the ACORN-SAT temperature reconstruction produced some rather bland, motherhood type statements. However, hidden in the public service speak was a distinct message for the Bureau of Meteorology: lift your game. Two of the areas I have been interested in are (a) whether individual adjustments are justified, and (b) the effect of these adjustments on national and regional temperature trends. In this post I look at adjustments at just two sites, which are responsible for the single largest increase in national trend.
On page 17 of the Report we find the following graphic:
Fig. 1: Scatterplot of difference between AWAP and Acorn annual mean temperature anomalies.
This is a clear statement of how much Acorn adjustments have cooled past temperatures, as AWAP is regarded as being only “partially homogenised”, and close to raw temperatures. There is a considerable difference- more than 0.2 degrees- between the two interpretations of temperatures 100 years ago.
Mean temperature is the average of maximum and minimum. In this post I shall look at just maximum temperatures, from 1911 to 2013. The following graph is a plot of the difference between monthly Acorn and AWAP maximum anomalies, which I think is much more informative:
Note there is a trend of +0.22 degrees / 100 years in the differences, indicating a predominance of cooling of earlier data; there is a very large range in the first 50 years, from about -0.7C to +0.3C, and one outlier at +0.4C, reducing to a much narrower band in the 1960s before increasing in the last 20 years; and the bulk of differences are negative before 1970.
Now let’s look at what has been happening in the past 35 years- in fact, in the satellite era:
Fig. 3: Monthly differences between AWAP and Acorn before and after December 1978
The trend in differences for the first 67 years is 0.33C / 100 years, but there is a very small tendency for Acorn to be cooler than AWAP recently- and the range of differences has been increasing.
That’s an interesting find, but I want to examine in more detail the effect of the adjustments which cause those differences. Here are annual maxima in AWAP compared with Acorn.
Fig. 4: Annual mean of monthly maximum anomalies: AWAP and Acorn
Again we see that Acorn has increased the warming trend from +0.59C to +0.81C per 100 years, an increase of +0.22C, or 37.3%.
However, the difference appears more marked before the mid 1950s. The next graph shows the trends from 1911 to 1955 compared with the trends from 1956 to 2013.
Fig. 5: Comparison of trends in maxima before and after the middle of the 20th Century.
Note: while the trends of AWAP and Acorn are very similar (+1.32 to 1.4C per 100 years) since the 1950s- which the Bureau never tires of proclaiming- before then the plot tells a different story. Acorn reduces the cooling trend by 0.44C per 100 years, a reduction of 86%.
How was this achieved?
On page 44 of the technical paper CTR-050 we find this explanation:
Returning now to maximum temperature, the differences between the AWAP and ACORN analyses show a marked drop in the early 1930s, with a sudden decrease of about 0.15 °C. This is most likely attributable to substantial negative adjustments between 1929 and 1932 in the ACORN-SAT dataset, indicating substantial discontinuities (expressed as artificial cooling) at a number of individual locations with a large influence on national analyses, because of the sparsity of data in their regions in that period. These discontinuities are mostly related to site moves that are associated with concatenated records for single locations. These include Alice Springs, Kalgoorlie and Meekatharra. Alice Springs, where the adjustment is associated with a site move in late 1931 or early 1932 from the Telegraph Station to a climatologically cooler site in the town, has a notably large “footprint”; at that time there were only two other locations within 600 kilometres (Tennant Creek and Charlotte Waters) which were observing temperatures, while the nearest neighbours to the west (Marble Bar and Wiluna) were more than 1200 kilometres away.
This large change between AWAP and Acorn is shown in the next graph.
Fig. 6: 12 month mean difference in monthly maxima anomalies
As I explained in my post in September 2014, Acorn sites are homogenised by an algorithm which references up to 10 neighbouring sites. A test for the validity of the adjustments is to compare the Acorn site’s raw and adjusted data with those of its neighbours, by finding the differences between them. Ideally, a perfect station with perfect neighbours will show zero differences: the average of their differences will be a straight line at zero. Importantly, even if the differences fluctuate, there should be zero trend. Any trend indicates past temperatures appear to be either relatively too warm or too cool at the station being studied. My aim is to check whether or not individual adjustments make the adjusted Acorn dataset compare with neighbours more closely. If so, the trend in differences should be close to zero.
I have tested the three sites named above. I use differences in anomalies calculated from the mean of maxima for the 30 year period centred on 1931, or for the period of overlap if the records are shorter. The neighbours are those listed by the Bureau on their Adjustments page.
Fig. 7: Meekatharra differences from neighbours (averaged)
Note that the Acorn adjustment (-0.77C at 1/1/1929- the adjustment of +0.54C at 1/1/1934 does not show up in the national signal) is indeed valid: the resultant trend in differences is close to zero, indicating good comparison with neighbours. However, since Meekatharra’s record starts only in 1927, two years of the Meekatharra adjustment cannot have had a large influence on the national trend as claimed.
Fig. 8: Kalgoorlie differences from neighbours
Kalgoorlie’s steep cooling compared with neighbours (from 170 km to 546 km away) has been reversed by the Acorn adjustment (-0.62C at 1/1/1930- the adjustment of -0.54C at 1/12/1936 does not show up in the national signal), so that Kalgoorlie now is warming too much (+1.02C / 100 years more than the neighbours). Kalgoorlie’s adjustment is too great, affecting all previous years.
I now turn to Alice Springs, which ‘has a notably large “footprint”’. Too right it does- its impact on the national climate signal is 7% to 10%, according to the 2011 Review Panel, p. 12.
Fig. 9: Alice Springs differences from neighbours
Alice Springs, cooling slightly compared with neighbours, has been adjusted (-0.57C at 1/1/1932) so that the Acorn reconstruction is warming (+0.66C / 100 years) relative to its neighbours. The adjustment is much too large.
And exactly where are these neighbours?
Tennant Creek (450 km away), Boulia (620 km), Old Halls Creek (880 km), Tibooburra (1030 km), Bourke (1390 km), and Cobar (1460 km)!
The site with the largest impact on Australia’s climate signal has been “homogenised” with neighbours from 450 km to 1460 km away- except the adjustment was too great, resulting in the reconstruction warming too much (+0.66C / 100 years) relative to these neighbours. The same applies at Kalgoorlie. Meekatharra’s record only starts in 1927 so its effect can be discounted. These are the only remote Acorn sites that had large adjustments at this time. All other remote Acorn sites open at this time either have similar trends in raw and Acorn or had no adjustments in this period.
The 37.3% increase in the trend of Australian maxima anomalies in the “world’s best practice” ACORN-SAT dataset compared with the “raw” AWAP dataset is largely due to just two adjustments- at Kalgoorlie and Alice Springs- and these adjustments are based on comparison with distant neighbours and are demonstrably too great.
If it wasn’t so serious it would be laughable.