The origins of the audit
The independent audit of raw temperature records at 237 WA and eastern states High Quality and associated weather stations was conducted over seven weeks from January to March 2012.
The rounding of degree fractions was first suspected in 2008 when climate researchers were commenting on "data not looking right" within temperature records at the BoM's 103 Reference Climate Stations around Australia.
Papers written by scientist Geoffrey Sherrington noted incongruities in the spread of fractional tenths while investigating instrumental difficulties preventing accurate temperature measurements.
The issue of temperature fractions received little attention for the next three years, although former ABS statistician Ian Hill noted in an email dated 3 August 2010:
One very interesting remark appearing in "Discussion" of the Torok paper is the possibility of a "non-climatic discontinuity" of observer practice in the days of fahrenheit (before September 1972) of only recording the temperature to a whole degree and that being truncated (eg 58.8 is recorded as 58 and not rounded to 59). The paper says:
"If many observers 'truncated' their measurements to the nearest degree below the actual measurement, prior to metrification, and after metrification recorded to tenths of a degree, this would result in an artificial warming in the early 1970s. Discontinuities caused by such a practice, if it was common, would not be detected by the statistical programs used here. Examination of field books does not suggest that this practice was sufficiently common to produce a major discontinuity."
The later paper by Della-Marta which refined the earlier work does not mention the issue at all, that I could find, admittedly from a quick scan only.
I think this could potentially explain some of the warming by itself at some stations, by as much as 0.3 degrees C. Even at Mount Gambier Aero which was given the quality rating of 1 (very good) by Della-Marta, had many days where the temperature was reported in whole degrees. In January 1939 for example (then the Post Office) only about half the days are reported as fractions of a degree F. No doubt the task of reading the thermometers fell to more than one person.
Also in 2010 but in relation to issues other than temperature rounding and metrication, researchers associated with this study requested that the Australian National Audit Office independently audit BoM climate records to ensure their accuracy and reliability.
A truncation test demonstrates the likelihood of rounding down, particularly by weather station observers who were unaware that thermometer readings of .5 should be rounded both up and down - not just down if they were ignoring the BoM instructions to record all Fahrenheit fractions.
The accuracy of all historic temperature records from manual observations is also questionable due to other influences such as thermometer resolution and mercury meniscus, and the impact of temperature screen design corrupting the accuracy of recordings.
A comparison of average annual minima and maxima at all ACORN sites in Western Australia also shows a rapid temperature increase starting in 1972.
The issue remained dormant for a further 18 months until research journalist Chris Gillham noticed that 19 months of 1995/96 at the Perth Metro 9225 weather station showed consecutive .0C roundings. This prompted a query to the BoM with the following response on 9 January 2012:
For the months in question, the Automatic Weather Station at Perth Metro was only sending 9am-to-9am minimum temperatures at a precision of 1 degree.
My initial investigation suggests we do not have higher-precision data for the months in question. I am looking into why some months (and the maximum temperatures) do show 0.1-degree precision.
This response prompted further questions from Chris Gillham with the following reply from the BoM on 11 January 2012:
If the Mt Lawley AWS was only sending 9am-to-9am minimum temperatures at a precision of 1 degree, why didn't anybody notice at the time? Did the AWS produce obviously wrong minima results for 19 months and nothing was done about it?
In the early days of the Bureau's automatic weather stations, they were programmed to report in an international code, which was limited to reporting maximum and minimum temperatures to the nearest degree Celsius. High precision data from the AWS was monitored in the Perth Regional Office and manual intervention forced some of the data to be sent to the database with its full precision, however this was a manually intensive task and for a period from 31 March 1995 to 1 December 1996, the less precise data was automatically forwarded to the climate database, though there are some exceptions during that period. After this time, the software was altered to allow the full precision to be automatically transmitted. Thus the Bureau was fully aware of the inconsistency at the time, as you would expect. It is important to note that minimum temperature data to the nearest degree was not wrong, but simply less precise than the raw data.
Accurate min temps must have been produced in 95/96 as decimals were in the broadcast minimum temps at the time. Were these sourced from the 9225 AWS or from a different thermometer? Can these precise minima produced at the time be provided?
The precise minimum temperatures were recorded in a local database from the 009225 AWS and these form the basis of the official Perth Metro climate record that is reported via our climate monitoring products. The precise minimum temperatures can be provided if required.
Have the erroneous minimum data been in Perth Metro 9225 temperature records since 95/96 and why haven't they been previously corrected?
As I pointed out above, the data are not erroneous, just less precise. A process is underway to have the more precise data entered into the national climate database, from our local database, however this is a slow process, particularly given the resources that are available for such tasks. Our desire is to have the most accurate data available to the public via our website.
Is there an estimate of how much these 19 months of incorrect minima influence Perth's Metro's mean temperature trend since 1995?
A comparison of the monthly mean minimum temperatures between the precise values and those rounded to the nearest degree for the 1995/6 period suggest differences of between 0.0 C and 0.3 C. These values will have a negligible impact on the Perth mean temperature trend, though as I said above, the most precise data is used in our analysis and reporting of Perth Metro climate statistics.
If the Perth Metro AWS was providing rounded minima and there is no precise data, how can the BoM be sure that the error has been a degree rounding up or down instead of a uniform rounding down (e.g. 14.9 = 14)?
As I said above, the precise data has been kept and rounding follows the standard procedure of rounding to the nearest degree, except at the 0.5 increment when values are rounded to the nearest odd whole number.
If the BoM has previously been unaware of this 19 month data error, will a notice be posted on the web or via the media to alert researchers that monthly and/or annual temperature estimates for Perth over the past 16 years are possibly inaccurate?
The data on the CDO web pages is provided to the general public so they can access national climate records freely. However there are issues with a small proportion of the data in the database, as you have identified, and these are explained in general terms in the link at the top of the temperature data pages
For the reasons stated on this web page, it is very important for researchers to understand the backgroyund to the data they are using and they need to consult the Bureau to make sure that the data is suitable for their purpose. The Bureau maintains a high quality dataset for research purposes at http://www.bom.gov.au/climate/change/
The precise data (csv download, 16kb) was requested and sent by the BoM on 19 January 2012, showing identical rounded average temperatures at Perth Metro 9225 in 1995/96 whether or not the degrees have fractions or are rounded to .0C. This data, if representative of all other Australian weather stations with similar consecutive blocks of rounded .0 Celsius since 1972, indicates that almost all rounding since then has been to the nearest integer rather than truncated down from above .5C to .0C. The same evidence is not available for pre 1972 Fahrenheit temperatures to determine whether up/down rounding or down truncation was most prevalent during that era.
This dialogue was debated by members of the audit group as they began looking into the fractional breakdown of temperature recordings. On 15 January 2012, Chris Gillham emailed the following:
While scanning through the maximum tables for Eucla (opened 1957), I could swear that I was seeing the temps 21.1C and 21.7C all the time. Then I thought I was only seeing the 21.1 / 21.7 hallucination up to about 1970. So instead of turning to a psychiatrist, I whipped out Excel and simply descended the max temps from raw data csv so I could group which years they fell into:
1957-1970 (14 years)
So in the first 14 years of record, 378 days or an entire year happened to have a max of either 21.1C or 21.7C. Remarkable since the Eucla max temp record swings from a lowest 10C in 2005 to a hottest 47.9C one day in 1979. And rounded days of 21C only happened once from 1957 to 1970 but 203 times from 1971 to 2011. And .5C days went from 1 to 168.
Later that day, Geoffrey Sherrington responded with the following:
Your old but likeable alcoholic thermometer reader can see 70 deg F on a cool day when he forgets his spectacles. Mathematical conversion gives 21.1111 deg C. rounds to 21.1. 71 deg F on a slightly warmer day gives a math conversion to 21.666667 deg C, rounds to 21.7.
Then decimal conversion came on 14 Feb 1966 and new thermometers were issued graduated in deg C. Have you made a count of 22.3 = 72.1 deg F in the earlier period? I’ve seen station data like this before but I can’t remember where. Several places. Hence my comment about absent at Church on Sundays. Ah retract that, saw no church at Eucla when last there. That’s why David S wrote the routine to look at adjustment via counting the last place after the decimal.
Ian Hill also responded:
re Eucla, looks like many days were simply recorded as 70 or 71 degrees fahrenheit and these convert to 21.1 and 21.7 respectively. Don't forget fahrenheit was used in this country until September 1972.
These clues were sufficient to recognise that "data not looking right" was somehow linked with the rounding to whole integers of Fahrenheit and Celsius degrees and the metrication of Australia's temperature scale in September 1972.
From 24 January 2012, it was collectively agreed that the BoM's raw temperature records should be audited to determine how widespread these distortions are throughout Australia, not just in WA.
The primary research was conducted by Ken Stewart, Chris Gillham, Geoffrey Sherrington, Ian Hill and Ed Thurstan, the latter two providing invaluable assistance by building macro applications for the Excel spreadsheet program that allowed rapid and precise analysis of 8,580,583 lines of minimum and maximum temperature data that was audited from 237 High Quality and associated weather stations across Australia.
None of the participants in this audit has received any financial reimbursement or other assistance from any third party, corporate or political, and all results are accurate and verifiable. The tabular audit results are available upon request.
The full audit results for all Australian HQ and associated weather stations are within:
The Excel macro applications built to conduct this audit are:
Both macros and the Table of Sites should be run simultaneously. Full instructions are included within the tenth fraction distribution calculator macro in the sheet titled Instructions.
The accuracy of the tenth fraction distribution Excel macro script can be compared with actual 1908 daily Fahrenheit recordings in Perth.
The results of the audit and analysis are detailed on this website and at Kenskingdom for a national perspective examining and detailing the various data limitations that may influence the audit results.
HI QUALITY audit|