Independent climate researchers were surprised by Bureau of Meteorology claims that Monday, 7 January 2013, was the hottest day ever in Australia based on an area-averaged calibration of temperatures.
The anomaly calibration procedure was applied to an unpublished but soon to be released BoM dataset of historic daily weather station temperatures in which 7 January at 40.3C is 5.25C above the day's average absolute maximum and 0.3C above the area averaged maximum in the much vaunted ACORN network of stations.
As stated in the BoM's interim Special Climate Statement on the extreme January heat: "Australia set a new record for the highest national area-average temperature, recording 40.33C and surpassing the previous record set on 21 December 1972 (40.17C)".
Or as the BoM put it in its media release for public consumption: "On Monday the average maximum daily temperature record for Australia was broken at 40.33C. The previous record, 40.17C on 21 December 1972, was held for 40 years".
From that it seems reasonable to assume that the national average maximum daily temperature was 40.33C on 7 January. Media reports quoted the BoM saying the temperature was estimated from the maxima of between 700 and 800 Australian weather stations and in its Special Climate Statement 43 the bureau corrects the national average maximum to 40.30C.
The actual maximum average
This analysis examines the average maximum at 721 of these stations where temperatures are publicly available.
Excluded were stations with missing 7 January temperatures in the BoM web database as well as those in locations such as Antarctica. Their results are contained in this Excel spreadsheet.
It turns out the average maximum of the 721 stations was 35.1C on 7 January 2013.
The BoM's Special Climate Statement includes a breakdown of each state's heatwave including 7 January, and the BoM's area-averaged estimate can be compared with the actual average maximum of all stations:
There's about a 5C difference, the same as the BoM's estimate of how much the January 2013 heatwave pushed national temperatures above average.
Climate researcher Ian Hill investigated what's necessary to achieve the BoM's national and state averages based purely on all station maxima, and it's interesting looking at how many of the the highest maxima stations would be needed out of the total in each jurisdiction:
Australia 339 out of 721
NSW 91 out of 172
Northern Territory 27 out of 54
Queensland 54 out of 125
South Australia 45 out of 80
Tasmania 57 out of 57
Victoria 68 out of 94
Western Australia 60 out of 139
In other words, if you chose the hottest 339 weather stations in Australia on 7 January 2013 and ignored the other 382, you'd find an average maximum of 40.3C.
It's noteworthy that the BoM's state area-averaged maxima can be multiplied by its area fraction of the Australian landmass, and then summed to obtain the national area average of 40.3C on 7 January:
Qld - 36.82 * 0.224829 = 8.2782
NSW - 38.87 * 0.104656 = 4.0680
Vic - 37.80 * 0.0296265 = 1.1199
Tas - 27.05 * 0.0088255 = 0.2387
SA - 43.40 * 0.128087 = 5.5590
WA - 42.29 * 0.328743 = 13.9025
NT - 40.73 * 0.175234 = 7.1373
AUS = sum of the states = 40.3036C
A simple averaging of all Australia's maxima to 35.1C on 7 January 2013 cannot be compared with the BoM's calculated record maximum because it is not area weighted.
A different measure
So does this mean the BoM's estimate of Australia's hottest day on 7 January is wrong?
Maybe. Maybe not.
In response to written questions, the BoM has explained that the estimate of an area average on 7 January 2013 could have been 40.00C or 40.33C, dependent upon which measuring stick is used.
The BoM uses various measuring sticks and it seems the much vaunted ACORN network of grid area weighted temperatures at 112 weather stations had an anomaly on 7 January that was 5.36C above the 1961-90 climatological average of 34.64C.
That adds up to 40.00C. The BoM decided instead to add the ACORN anomaly that day to its Australian Water Availability Project (AWAP) network of stations to achieve an Australian area average of 40.30C on 7 January 2013.
The BoM believes that to gauge the extent of heat across Australia, the most appropriate dataset to use is AWAP which uses daily gridded data from all available and unhomogenised temperature measurements from around 700 stations daily.
The grid file for the BoM's archived daily maximum RMSE Australian temperature map on 7 January 2013 suggests 718 weather stations were monitored, almost the same as the 721 stations within this analysis.
The AWAP network commenced in 1911 and its real-time daily gridded data monitoring system for daily maximum and minimum temperatures is generated at 1pm EST (2pm EDST) each day for the previous day. The BoM had previously calculated AWAP measures from 1996 but the methodology was not extensively published in peer-reviewed literature until 2009.
The BoM explains that the AWAP network is not fixed in time due to network changes and basic quality control but sensitivity analyses show stability in the calculations since 1950.
It is difficult to determine exactly how many station records are used in the BoM analysis as, in response to questions from the Quadrant website, the bureau stated:
"We estimate the AWAP network definition alone (which stations are used at each time-step) comprises around 11 million records over 103 years of data."
However, in response to questions for this analysis, the bureau responded:
"We estimate the AWAP network definition alone (which stations are used at each time-step) comprises around 13 million temperature records over 102 years of data."
These may be typographical errors by the BoM in its written answers but reflects poorly on the bureau's explanation for Australia's hottest day and does not instill confidence.
For many years the BoM developed what it calls the High Quality network of 237 weather stations around Australia that meet strict criteria to ensure accurate temperature comparisons back to 1910, with an anomaly baseline from 1961-90 (see research published January 2013 re bias in the HQ network).
In early 2012, the HQ network was superseded by the Australian Climate Observations Reference Network (ACORN) comprising 112 stations around the country with an anomaly baseline from 1981-2010 and a slightly warmer trend than HQ according to the CSIRO and the National Climate Centre.
The accuracy of HQ and ACORN datasets can be questioned but they are accepted as the official yardsticks and ACORN feeds global temperature indices. ACORN's 112 stations are located strategically around Australia to represent an accurate spatial average of Australia's national temperature, accounting for numerous factors such as historic location shifts, equipment changes, UHI, etc.
The BoM cites 21 December 1972 as the previous hottest Australian day at 40.17C.
Climate researcher Ken Stewart looked closer and found the averaged mean ACORN maximum on 21 December 1972 was 35.91C, somewhat warmer than the raw average max of 35.1C at 721 stations on 7 January 2013 but nevertheless a lot cooler than the 40.17C estimated by the BoM's AWAP national daily procedure.
Uncertainties created by the application of a daily ACORN homogenised anomaly at 112 stations to 718 stations with unhomogenised maxima since 1911 are highlighted in the Jones et al paper referenced by the BoM and in Report 3b for the Independent Peer Review of the ACORN-SAT dataset which shows less warming in the AWAP network:
Differences between the AWAP and ACORN weather station networks have also been highlighted by Central Queensland University Adjunct Research Fellow David Stockwell within Is the temperature or the temperature record increasing?
Age of weather stations
Climate researcher Ed Thurstan analysed all 1,886 BoM weather stations (from BOM Product IDCJMC0015) to identify mainland locations that remain open with more than 30 years of temperature data, and that have now been introduced to the BoM's derivation of daily maxima.
His research found there were 269 excluding the 112 ACORN stations (download Excel spreadsheet).
The ACORN stations have a mean age of 86.5 years whereas the new ones introduced for the BoM's new AWAP daily area average calculations have likely operated for an average of only 46 years. That is, since 1966.
Who knows what extreme events might have been recorded at those stations prior to 1966? For example, these records generally do not include the 1939 heatwave.
What's going on?
The oddity you may have noticed is that Tasmania was the only place where the estimated and actual maxima agree, and all available stations in that state contributed to the average.
This is a hint at the procedure used by the BoM to declare Australia's hottest ever day.
The BoM's estimate is area-averaged, whereby the entire Australian continent is divided into grid cells of latitude and longitude, and the weighted average of these cells is calculated. The BoM has used cell sizes as small as .25x.25 degrees in their evaluation of ACORN data.
All methods achieve average temperatures based on complicated variants such as land area proportions, meridional convergence and a distance weighted interpolation over multiple cells.
Valid extrapolation of temperatures can only be over limited distances but Australia historically had many areas with no temperature estimates at all. It is these unknown elements that can produce different results and demand the application of an accepted and consistent methodology.
But the BoM has introduced a new, unpublished procedure that apparently calculates area weighted means within states, then an area weighted mean across states based on their relative land areas.
Few inland thermometers
It sounds complicated because it is. The adjustments are needed partly because most weather stations in the 1800s and early 1900s were established in populated areas but there was a dearth of isolated inland thermometers and temperatures to build a true national average or historic comparison.
In March 2012, the BoM stated there were 761 weather stations in Australia, 16 with 100 years or more of data, 184 with 50 or more years, 506 with 30 or more, and 29 with less than five years of data.
For example, the maps below show how many more weather stations are feeding Australia's temperature record from 1930 to 2010, with scant recording possible of the interior where most of the January 2013 heatwave occurred.
A breakdown of contributing ACORN stations for area weighted mean minima and maxima, using a 5x5 degree cell grid since 1910, shows the percentage of the Australian continent with temperature data input by decade:
How is it possible?
It is difficult to imagine how an area-averaged, day-by-day comparison of station recordings as early as 1910 or 1911 is possible, regardless of cell grid weighting, when in many vast areas there were no stations to weight.
A similar problem persists today and to compensate for the sparsity of weather stations in many inland regions, often with totally empty grid cells between them, the ACORN database applies a weighted average which proportionally multiplies their occurrence, at the same time dividing the occurrence of temperatures among more densely located stations in populated regions mostly on the coastal fringe.
As an example, to achieve 40.3C you would have to add another 494 stations which all had Leonora's maximum of 47.8C (the hottest station in Australia on 7 January) to the other 721. Then you'd have 1,215 stations which altogether averaged 40.3C if the estimate was based simply on unadjusted maxima.
Australia's "area averaged" record hot temperature on 7 January was (probably) based on an algorithmic compensation for hot outback areas with few thermometers, and an estimation of how this procedure would apply to every day of temperature readings at each of the 718 weather stations since 1911.
It sounds tricky because many of the stations have shifted, usually to airports, or simply didn't exist back then. It would be interesting to know what are the error bands in the BoM's area averaged all-station procedure and whether those errors have any consistency back to the early 20th century.
A confusing heatwave
Confusion about the BoM's representation of the January 2013 heatwave is also caused by its explanation of what's causing Australia's heatwave in which a map shows over 70% of the continent recording temperatures in excess of 42C during the first two weeks of January 2013. The bureau then states that "it’s not like these sorts of days occur that often. The records set last week sit between two and three standard deviations above the long-term January mean of 35C".
The bureau seems to be confusing the daily maximums with the highest temperatures over 14 days. The mean and standard deviation would be the daily mean and standard deviation, whereas the actual figure needed to calculate is the mean and standard deviation for the maximum temperatures over a 14 day period, which would be much higher than the daily figure (read more).
Further, in claiming the records are two to three standard deviations above the 35C mean for January, it should be pointed out that January 2013 was far from over when the statement was made so a valid comparison was not possible.
The January mean from 1961-90 is 34.6C with a standard deviation of 0.91. Temperatures cherry-picked from a short time period are going to deviate from the mean value over a longer period.
There is no doubt the January 2013 heatwave was hot over an extended area of inland Australia. However, such misrepresentations of available data suggest an exaggeration of its intensity in an historical context.
Uncertainty about the accuracy of historic estimates is not clarified by BoM responses such as:
"The network is not fixed in time due to network changes and basic quality control, however sensitivity analyses (as documented in the published papers referenced in the links above) shows stability in the calculations post 1950."
"The calculation of daily area averages dates back to the mid-1990s, when computing power first allowed for these intensive calculations.
"The analysis can be applied retrospectively from 1910. It should be noted that calculations of monthly and annual temperatures are based on the technique for analysing daily temperatures and comprise the same data. In other words, the daily calculation is mathematically identical to that used for monthly and annual temperatures reported by the Bureau. Calculations of global-mean temperature by international research centres also use similar techniques, with the earliest estimates dating back into at least the 1980s."
"A detailed comparison of AWAP and ACORN-SAT datasets was performed as part of the international review of our practices, and is contained in the scientific papers linked to above. While there are differences between the two data sets at the local scale, the differences between the two datasets for large area-averages are typically small (in the order of a few hundredths of a degree for annual means) over the last 50 to 60 years."
New BoM daily dataset
In its answers to questions about the procedure used to estimate a record hot day on January 7, the BoM revealed that a more detailed analysis of the January heat event, including changes in daily average temperatures, has been submitted to the peer-reviewed Bulletin of the Australian Meteorological and Oceanographic Society (BAMOS).
The BoM has advised that upon publication of the research paper within a few months, the historical analysis of daily AWAP temperatures will be made publicly available.
This will hopefully provide a new, publicly accessible database of estimated historic daily temperatures since 1911 at more than 700 Australian weather stations.
With a far broader range of stations than the 112 in the ACORN dataset, the new AWAP gauge will arguably contain even more debatable estimates of anomalies at any given location on any given day since 1911.
The assurances of the BoM suggest its measuring stick may be accurate for Australian daily temperature estimates starting in the 1950s.
Significant questions still surround the validity of the HQ and ACORN anomaly adjustments or lack of adjustments for historic and recent temperature readings.
The ACORN estimated average anomaly at 112 weather stations is added to Australia's average maximum in 1961-90 and applied to a different dataset of more than 700 stations over 100 years, with identical mathematics producing "typically small" differences in the latter half.
Anomaly inconsistencies remain between AWAP, the ACORN absolute maximum of 35.9C on 21 December 1972 and the absolute maximum of 35.1C at 721 stations on 7 January 2013.
The claim of a 40.3C record hot day on 7 January 2013 is dubious, particularly when compared before the 1950s.
Contributory research by Ian Hill, Ed Thurstan, David Stockwell and Ken Stewart.
Postscript: See 17 January 1908: Was it Australia's hottest day? for a comparison of raw maxima at 221 weather stations.
Top of page
This website produced by Scribeworks 2013