Garbage In, Garbage Out- Horn Island

After last year’s major project of checking the compliance of 666 Australian weather stations with the guidelines set out by the Bureau of Meteorology, and finding that nearly half did not meet siting specifications, I decided to take a break from analysing BOM climate data.

It might now be time to re-enter the fray.

The pride of the BOM climate fleet, Version 2 of the Australian Climate Observation Recording Network- Surface Air Temperatures (ACORN-SAT, or Acorn 2) was launched with no fanfare at all in late 2018 and now is the basis for the Bureau’s climate claims and predictions.  I summarised many of its faults in May 2019.

Now I am going to look at some individual examples of Acorn 2 nonsense- firstly, Horn Island’s minimum temperatures.

Horn Island is our most northerly Acorn site.  It is the airport for Thursday Island and is closer to Port Moresby than any other Australian town except Weipa and of course TI.  Figure 1 shows the station’s only neighbours on Cape York Peninsula the BOM used for comparison. 

Figure 1:  Google Maps image of Cape York Peninsula showing Horn Island and its neighbours

Figure 2 shows the annual average minima for all of these weather stations.

Figure 2:  Annual minima, Horn Island and neighbours

Horn Island 27058 has a very limited data record commencing in March 1995.  To construct a longer record back as far as 1951, BOM merged Horn Island’s observations with those of Thursday Island Township 27021 and Thursday Island Meteorological Office 27022.  Figure 3 shows the result of the BOM’s merge.

Figure 3:  Horn Island and Thursday Island minima

The 27022 TI Met Office annual average Tmin has been reduced in Acorn by 0.5 or 0.6 degrees for all years before 1993.  (The Adjustment Summary claims there was an annual impact of -0.53 degrees.)  There were other adjustments at 01/01/1967 and 01/01/1958, but the 1993 adjustment has the largest effect.

Unfortunately, there are only 6 months of overlap between 27021 and 27022 (September 1992 to February 1993), and 11 months between 27021 and 27058 (March 1995 to January 1996), as figure 4 shows.

Figure 4:  Extent of overlaps in monthly minima

In September, October, and November 1992 TI Township 27021 recorded minima 1.1 to 1.4 degrees cooler than the Met Office site in “a relatively exposed location” on top of the hill 900 metres west.  However, Acorn tracks 27021 and approximately splits the difference from 27022.

Figure 5:  Extent of overlaps in monthly minima, including Acorn 2

To do this, the BOM uses two comparative stations, Lockhart River 28008 and Coen Airport 27006.  Coen is 385km from Horn Island. 

Figure 6 shows a plot of monthly anomalies from 1981 to 2010 means for these stations.

Figure 6:  Monthly minima anomalies from 1981-2010 means, all stations

Note that 27022 is close to 27021 from December 1992 to February 1993, but not before, while Coen is nothing like either.  I cannot see the justification for the adjustment.

I constructed a merge of annual data using 27021 for 1993, 1994, and 1995, joined to 27022 and 27058 with no adjustment to construct a “raw” record.  This is entirely artificial, but no more so than Acorn.  Figure 7 shows plots of both with annual trends.

Figure 7:  Horn Island minima, Acorn 2 and “raw”

Note Acorn has enormously increased the warming trend.  Figure 8 plots the differences between Acorn and my “raw” record.

Figure 8:  Horn Island minima adjustments

The earlier adjustments were also large and based on “statistical” breakpoints.

The object of adjusting temperatures using neighbours for comparison is to endeavour to produce a record that more truly reflects climate trends of the area.  The resulting record should be more like the neighbours than the original raw record.  We can test this by plotting the differences between Acorn and the raw record and the average of the neighbours.  If the comparison is good, while individual years’ differences may vary, the trend should be close to zero: the station should not be warming or cooling more than the neighbours.  Figure 9 shows the results for Horn Island minima for the period that the 1993 adjustment applied.

Figure 9:  Differences between Horn Island and two neighbours

You will note that the blue trend line, showing the trend of the difference between annual data merged with no adjustment and the average of Coen and Lockhart River, is almost flat, indicating the “raw” record already compares well with the neighbours.  In contrast the red trend line shows the adjusted data is warming faster than the neighbours, indicating a very poor reflection of the climate of the area.


The Horn Island record should never have been merged because of the lack of suitable overlap.

Once merged, it should never have been adjusted downwards so much.

 Lockhart River and Coen are far too distant to be suitable for comparison.

The result is nonsense.

Garbage in, garbage out.

Tags: , ,

6 Responses to “Garbage In, Garbage Out- Horn Island”

  1. GlenM Says:

    Do you think that statisticians at the Bureau get awards for “inventiveness” ? Gently massaging politics into it.

  2. Beverley Jackson Says:

    Thank you for this clarification.

  3. Lint Robb Says:

    Great work!

  4. climanrecon Says:

    I agree with this analysis. Reading the documentation for ACORN-SATv2 rings alarm bells for this retired algorithm analyst, the main problem likely to be the introduction of new methods of detecting inhomogeneities. Maybe Blair Trewin himself could have made it work, but he is now too senior for such things, and juniors are likely responsible for failing to stop false detections.

    ACORN-SATv2 fails even to get internal self consistency, there are many examples which are clearly inhomogeneous relative to their neighbours, often the result of erroneous adjustments.

    My analysis is relative to Berkeley Earth, which I believe gives a good approximation to regional averages for Australia, but you don’t have to accept that if you don’t want to:

  5. ngard2016 Says:

    Here’s an interesting exercise using the York UNI temp tool,comparing RSS V4 TLT and UAH V6 TLT.
    First start at about the top of the 1997.75 el nino and check the trends to the present day.
    For RSS v 4 we have a trend of 0.198 c/ decade and UAH V 6 0.111 c/ per decade. Or nearly twice the trend if you believe RSS V 4.
    Next check the trend from the 1997.75 el nino to the 2017.1 el nino. The start and ending are about as accurate as I can make it. See graph for trend.
    The trend for RSS V 4 TLT is 0.164 c / decade and the trend for UAH V 6 TLT is 0.053 c / decade. This trend is over 3 times the trend if you believe RSS V 4 over that period of 19 years.
    Little wonder that Dr Christy and Dr Spencer don’t accept the RSS V 4 TLT data. Yet the true believers STILL BELIEVE.
    Any comments Ken?

Comments are closed.

%d bloggers like this: