ACORN-SAT 2.0: Western Australia- A State of Confusion

(UPDATE 17/02/2019:

I have corrected a glitch in trend calculations which are now as shown.  I have deleted all Diurnal Temperature Range plots and discussion as well.)

This is the first in a series of posts in which I directly compare the most recent version of Australia’s temperature record, ACORN-SAT 2, with that of the previous version, ACORN-SAT 1.  Daily data are directly downloaded from the Bureau of Meteorology. I do not analyse against raw data (available at Climate Data Online), except for particular examples, as I am interested in how different Acorn 2 is from Acorn 1.  The basis for the new version is in the Research Report.

I start with Western Australia, and must thank Chris Gillham for his outstanding work and for allowing me to use data from stations he has used for his annual analysis.

Introduction:

The Bureau of Meteorology has released its latest revision of the Australian temperature record back to 1910.  Previous versions of our historic temperatures included “High Quality”, which I revealed in 2010 to have major flaws, not least being the strong warming bias; and ACORN-SAT 1, released in March 2012, proudly touted as being “World’s Best Practice”, which I (along with others) found to have very many severe problems.  (If you like, check these posts, here, here, here, and here.  There are many others.)

Stung by the public and media criticism which this generated, the Bureau set up a supposedly independent Technical Advisory Forum, which met on one day per year for three years and basically rubber-stamped Acorn.  They did, however, make some recommendations, particularly about transparency.  In the light of this recommendation, this latest release without any publicity at all is perplexing.

Nearly all of Australia’s climate analysis and modelling is based on the previous version, Acorn 1, including monthly, seasonal, and annual means, extremes, and trends.  Sometime in the near future, this will be based on Acorn 2 data.

As this an upgrade to an existing dataset, we might expect there would be a few small tweaks of maybe a few tenths of a degree in some records and any changes to temperature trends would be fairly small.  Perhaps there might be some extra stations in remote areas to improve the density of the sparse network, perhaps some records starting earlier because of newly digitized data, hopefully a sensible fix for the dreadful situation of many daily minimum temperatures being higher than the maximum.

Not so.

No wonder the Bureau has released Acorn 2 so quietly- it is a confusing mess, and completely alters Acorn 1.  Trends are vastly different, some temperatures altered by more than 10 degrees Celsius, and new records established.

The Context – Western Australia

Figure 1 is a map of Australia showing all of the Bureau’s ACORN-SAT climate monitoring stations.  Western Australia occupies the western third of the continent.  Most of it is desert, and there are vast distances between settlements and thermometers.

Figure 1:  Australian ACORN-SAT stations

Acorn map WA

There are 25 Acorn stations in the Western Australian BOM database.  One (Kalumburu 001019) has the latest version data for minima but not for maxima, so complete analysis is not possible.  Differences between Acorn 1 and Acorn 2 are summarized in the following sections.

Trend changes

Trends in maximum temperature have increased by an average of +0.25 degrees Celsius per 100 years (from +1.17C to 1.42C), which is an increase of 21.7% over the trend produced by Acorn 1.  (Click on each graphic to enlarge.)

Figure 2:  Maxima trend changes from Acorn 1 to Acorn 2

WA Max trend chart

The largest increase in trend is at Wittenoom.

Trends in minimum temperature have increased by an average of nearly +0.22 degrees Celsius per 100 years (from +1.04C to +1.27C), which is an increase of 21.53%.

Figure 3:  Minima trend changes from Acorn 1 to Acorn 2

WA Min trend chart

The largest increase  (+1.06C per 100 years- from +0.55C to +1.61C).  The largest decrease in trend was at Halls Creek: -1.31C per 100 years.

Largest temperature differences

In maxima, changes to Acorn 1 daily data were often very large.  Wandering gets the gong for greatest adjustments, ranging from -10.9C to +10.9C applied to individual daily figures, but only on a few days.  Eucla has many large changes made to Acorn 1 data.

Figure 4:  Daily changes in maxima from Acorn 1 to Acorn 2 at Eucla

Diff Tmax Eucla

Minima adjustments ranged from -10.8C at Esperance to +7.8C at Halls Creek for a few adjustments, but at most stations the range was much less, though still substantial changes to Acorn 1.  Here is Perth:

Figure 5:  Daily changes in minima from Acorn 1 to Acorn 2 at Perth

Diff Tmin Perth

(Remember, these are adjustments to Acorn 1, which was supposed to be “world’s best practice” seven years ago.  How did Blair Trewin get it so wrong the first time?  Has world’s best practice changed so much in seven years?)

Record temperatures

A new record maximum was established at Carnarvon, whose already homogenized record increased from 48.5C to 51C.  This is now the record for all of Australia, apparently (although I have 87 more stations to check).   Additional large adjustments are the cause:

Figure 6:  Three versions of maxima at Carnarvon 23 January 1953

Carnarvon Max

The previous “record”, held by Albany in the cool south, after much ridicule was reduced from 51.2C to 49.5C.  New records were also established at Bridgetown, Dalwallinu, Eucla, Kalgoorlie, Katanning, Marble Bar, Merredin, Perth, and Port Hedland.

New record low temperatures were established at Bridgetown, Cape Leeuwin, Cunderdin, Dalwallinu, Esperance, Eucla, Forrest, Geraldton, Halls Creek, Kalgoorlie, Learmonth, Marble Bar, Meekatharra, Perth, and Wittenoom.

Apparently the adjustments made to raw data in Acorn 1 weren’t good enough.

Quality Control: especially minimum temperatures higher than maximum.

In Acorn 1, 16 out of 25 stations had at least one example of minimum higher than maximum.  Blair Trewin has “fixed” this problem (which he concedes was “physically unrealistic”) by adjusting temperatures in Acorn 2 so that the maximum and minimum are the same, so that DTR for the day is zero.  In his words:

A procedure was therefore adopted under which, if a day had a negative diurnal range in the adjusted data, the maximum and minimum temperatures were each corrected to the mean of the original adjusted maximum and adjusted minimum, creating no change in the daily mean.

But that is not how he “corrected” the worst Western Australian example in Acorn 1 (minimum 2.1C above maximum) at Kalgoorlie.  Here is a plot of the raw data for 14th to 18th November 1914.

Figure 7:  Kalgoorlie Post Office data for 14-18 November 1914

Kalgoorlie raw

The 16th was a cold rainy day, with only 0.1C separating minimum (15.5C) and maximum (15.6C).  But temperatures in 1914 were read from a Fahrenheit thermometer.  Both 60F and 60.1F convert to 15.6C; 15.5C is 59.9F.  It is likely the temperature ranged from just under 60F to just over 60F.

Acorn 1 adjustments were made with brute force rather than finesse.  The maximum was reduced by 1.3C to 14.3C, and the minimum was raised by 0.9C to 16.4C, resulting in nonsense.

Figure 8:  Kalgoorlie Post Office and Acorn 1 data for 14-18 November 1914

Kalgoorlie Ac1

In Fahrenheit, 57.7F maximum and 61.5F minimum.

The solution in Acorn 2?  Even more brutal adjustments- and not to the mean of the Acorn 1 adjustments (which would have been 15.35C):

Figure 9:  Kalgoorlie Post Office and Acorn 2 data for 14-18 November 1914

Kalgoorlie Ac2

The Acorn 1 minima is decreased (by 3.4C) to 13C, and Acorn 1 maxima decreased by another 1.3C to 13C (or 55.4F), making it 2.6C below the raw temperature as read in 1914.  Now there is no problem with minimum exceeding maximum, but at the cost of raw data tortured beyond recognition.

“Square wave” pattern in adjustments

Bob Fernley-Jones first noticed a peculiar repeating pattern of adjustments to Perth in Acorn 1 monthly data.  I can replicate this in dailies.

Figure 10:  Perth Acorn 1 daily maxima differences 1983-1986

sq wave perth acorn 1

This pattern is still visible in Acorn 2, but is much reduced.  Adjustments are still applied month by month, but they are not as rigid.

Figure 11:  Perth Acorn 2 daily maxima differences 1983-1986

sq wave perth acorn 2

This is how it was changed:

Figure 12:  Perth Acorn 2 minus Acorn 1 daily maxima differences 1983-1986

sq wave perth acorn 2- acorn1

A new square wave- almost a mirror image of Figure 11.  It is good to see that the Bureau has taken notice of criticisms!

Conclusion:

Comparison of Acorn2 versus Acorn 1 data for Western Australia does not encourage confidence in the Bureau’s methods:-

There are no additional stations, so the network is still extremely sparse.

There is a very small amount of additional digitized data.

The average trend in maxima for WA has been increased by 21.7%, and in minima by 21.5%.

Differences between Acorn 1 and Acorn 2 daily data can be up to nearly 11 degrees Celsius.

New record maximum temperatures have been set.

The issue of instances of minima being higher than maxima caused by too vigorous adjustments has been “fixed” by further vigorous adjustments.

The “square wave” pattern in adjustments in Perth has been largely rectified.  The square wave is now in the difference between Acorn 1 and Acorn 2.

It beggars belief that a dataset that was proudly described as “world’s best practice” just seven years ago has needed to be adjusted by so much.  Has “best practice” changed so much?  How was Acorn 1 so wrong?  How can we be sure that the new version is better, and will itself not be changed again in a few years?

There are now four versions of WA temperature:  Raw; High Quality (no longer available); Acorn 1; and Acorn 2.  All are different.

The record for Western Australia reveals a state, not of excitement, but of confusion.

 

Next: the Northern Territory.

Tags: , , , , ,

7 Responses to “ACORN-SAT 2.0: Western Australia- A State of Confusion”

  1. ACORN-SAT 2.0: Queensland: Welcome to Dreamworld | kenskingdom Says:

    […] my previous posts for Western Australia and the Northern Territory for a general […]

  2. ACORN-SAT 2.0: South Australia- Science Fiction | kenskingdom Says:

    […] my previous posts for Western Australia, the Northern Territory and Queensland for a general […]

  3. ACORN-SAT 2.0: Tasmania- May the Farce be with you | kenskingdom Says:

    […] my previous posts for Western Australia, the Northern Territory, Queensland, and South Australia for a general introduction.  An […]

  4. ACORN-SAT 2.0: Victoria- A comedy of errors | kenskingdom Says:

    […] my previous posts for Western Australia, the Northern Territory, Queensland,  South Australia, and Tasmania for a general […]

  5. ACORN-SAT 2.0: New South Wales- What a mess | kenskingdom Says:

    […] my previous posts for Western Australia, the Northern Territory, Queensland,  South Australia, Tasmania, and Victoria for a general […]

  6. John Patrick Says:

    I’ve just started looking at acornsatv2 and am trying to get my head around their statistical changes. I passed the “method” across a statistician and a physicist and let’s just say they were unable to find any mathematical basis for this treatment of data.

    The only site I’ve looked at is Kalumburu, WA, which had a single thermometer through the course of its operation as a non AWS station from 1941 to 2005.

    I’ve also read Cowtan’s paper on homogenisation from nearby stations. That didn’t inspire confidence. Areas of concern are:
    No definition of “nearby” though the paper uses 500km in the text which is a rather extreme definition of nearby.
    The analysis for correlation seems to provide some basis for two sites to behave similarly “on average” ie. That they have similar climates, but looking at the data for a “match” shows daily departures that are in keeping with the data they are wanting to “fix”. So effectively they are using a climate average to change a daily temperature which is by no means an outlier. That’s pretty much a no-no.

    With my very simple analysis of Kalumburu, I noticed that the “nearby” stations are 225, 445, 621 and 755km away, with nearer stations not used. More concerning was their apparent adjustment of temperatures in 1941, despite none of the “nearby” stations operating earlier than the 50s.

    I stopped looking after this, so there may be some subtleties I’m missing, like weird no operating stations not accessible through their near search interface, or that aren’t visible in the 4 stations they have listed as determinants of the statistical changes, but were antecedents of these stations with data listed elsewhere. I’ve instead started the effort of mining the data from the site, inclusive of raw, to conduct a more complete analysis. It’s sad that after so many years they still haven’t responded to the recommendations regarding single format easy access to all 3 data sets and metadata and code, particularly since the first of these would be a good half day starter project for an intern.

Comments are closed.


%d bloggers like this: