Early this month the ACORN-SAT dataset was updated to 31 December 2014. A quick check shows that despite much public criticism, nothing has changed since March 2012 when it was first released. There have been no corrections.
There are still thousands of days of missing data.
There are still gross errors.
Adjustments remain, and so still have the effect of making many individual sites LESS comparable with neighbours, and of making national trends much greater than those for raw data.
However, in this post I am merely focussing on one of the many erroneous features in Acorn, that of the instances of minimum temperatures exceeding maximum temperatures.
Why bring this up again, as it was first spotted in 2012? Because it is more evidence of unjustified adjustments, and Acorn’s authors have done nothing about it.
Although a specific check for errors in recording raw maxima and minima was conducted before homogenising, this check could not have been done with the homogenised data. It might be claimed that this feature is normal and due to a cold change arriving after 9.00 a.m. This would be especially evident in winter at high altitudes such as at Cabramurra, with 212 occurrences. If so, it would have been evident in the raw data. However, a check of 22 of the sites, comprising 84% of the instances (803 of 954) reveals there are NO instances of maximum less than minimum in the raw data for Cabramurra, or anywhere else. All instances occur in the adjusted data.
Fig. 1: Min > Max in Acorn (803 of 954 instances)
Fig. 2: The same test for raw data at the same sites as Fig. 1
Further, despite the Bureau being aware of the problem since at least 1 July 2013 when Blair Trewin, lead author of ACORN, assured readers of the blog Open Mind (sic) at https://tamino.wordpress.com/2013/06/29/a-clue-for-willis/#more-6693 that “in the next version of the data set (later this year), in cases where the adjusted max < adjusted min, we’ll set both the max and min equal to the mean of the two” (which merely hides the fault caused by adjustments), the problem still exists- 212 occurrences are still in the ACORN record for Cabramurra, and there are 954 in total. Trewin has done nothing about it- mainly because he would have to redo the whole dataset to do it properly. The problem exists because adjustments have been too vigorous and too erratic. The algorithm produces bogus data. Merely averaging maxima and minima on those 954 days hides the underlying problem.
Here is one example, the worst I have found, (circled in Fig. 1), from Alice Springs, 15 June 1932.
Fig. 3: Raw vs Acorn at Alice Springs
Notice there are NO adjustments in Acorn to raw maxima, but Acorn minima mostly runs along one or two degrees less than raw- except on 1 and 2 June and most notably on 15 June, when Acorn minima shoots to 4.8 degrees above the maxima for the day! This is caused by an adjustment of 6.4 degrees to the raw temperature. The Bureau lists the following sites as neighbours used for this adjustment: Boulia, Tibooburra, and Tennant Creek. Ignore for the moment that these are all hundreds of kilometres from The Alice, just visually compare their data.
Fig. 4: Alice vs ‘neighbours’
Tennant Creek and Boulia, hundreds of kilometres north and northeast, are usually several degrees warmer than The Alice, except when Acorn makes those huge adjustments. Which is the outlier?
This Alice Springs instance is a rogue adjustment due to a leading “1” being inserted; 18.1 entered instead of 8.1. This is not uncommon in Acorn and crops up repeatedly.
While this is annoying and farcical, it can be easily explained and easily fixed. The real problem is with data that is not so grossly erroneous, such as at Cabramurra. This graph shows raw minima and maxima for May and June 1962. Note that often maximum is not much above minimum: there is little room for error.
Fig. 5: Cabramurra raw
But see what happens with adjustments.
Fig. 6: Cabramurra raw and adjusted
Maxima have been adjusted down, minima up, without any thought for the consequences, so four times in May and June 1962 minima exceed maxima. Merely averaging maxima and minima for those days will hide the symptom of the underlying disease. The adjustments cannot be justified because they result in nonsense data. Exactly how do you explain negative Diurnal Temperature Range?
The problem of a tiny percentage of the record featuring this error is not minor: it indicates that many adjustments are too great, and also that no quality assurance checks have been performed, and therefore many other types of errors are likely to exist. And they do.
Now how is it that my laptop can find these errors, whereas the Bureau’s supercomputer can’t?
Here is the full list of sites and instances of bogus data:
Station, Number of days with minimum temperature exceeding the maximum temperature.
Adelaide, 1. Albany, 2. Alice Springs, 36. Birdsville, 1. Bourke, 12. Burketown, 6. Cabramurra, 212. Cairns, 2. Canberra, 4. Cape Borda, 4. Cape Leeuwin, 2. Cape Otway Lighthouse, 63. Charleville, 30. Charters Towers, 8. Dubbo, 8. Esperance, 1. Eucla, 5. Forrest, 1. Gabo Island, 1. Gayndah, 3. Georgetown, 15. Giles, 3. Grove, 1. Halls Creek, 21. Hobart, 7. Inverell, 11. Kalgoorlie-Boulder, 11. Kalumburu, 1. Katanning, 1. Kerang, 1. Kyancutta, 2. Larapuna (Eddystone Point), 4. Longreach, 24. Low Head, 39. Mackay, 61. Marble Bar, 11. Marree, 2. Meekatharra, 12. Melbourne Regional Office, 7. Merredin, 1. Mildura, 1. Miles, 5. Morawa, 7. Moree, 3. Mount Gambier, 12. Nhill, 4. Normanton, 3. Nowra, 2. Orbost, 48. Palmerville, 1. Port Hedland, 2. Port Lincoln, 8. Rabbit Flat, 3. Richmond (NSW), 1. Richmond (Qld), 9. Robe, 2. St George, 2. Sydney, 12. Tarcoola, 4. Tennant Creek, 40. Thargomindah, 5. Tibooburra, 15. Wagga Wagga, 1. Walgett, 3. Wilcannia, 1. Wilsons Promontory, 79. Wittenoom, 4. Wyalong, 2. Yamba, 1.
(From Willis Eschenbach at http://wattsupwiththat.com/2013/06/28/australia-and-acorn-sat/ . Butlers Gorge has a further 37 days.)
Here is a plot of their occurrence- the big lump after 1960 is due to Cabramurra.
Fig. 7: Running 365 day count of min > max (803 of 954)