High Quality weather station audit
Australia raw temperature trend audit of BoM HQ stations
The proportions have been identified through an independent data verification audit of 237 weather stations in all Australian states and territories within the Bureau of Meteorology's annual High Quality (HQ) temperature series used to establish Australia's official climate record since 1910.
The audit included 474 records of raw, unadjusted daily minima and maxima collected from the 237 stations since their opening. The stations include the 134 sites in the Annual HQ dataset plus associated neighbouring stations used by the BoM to homogenise temperatures for a reliable trend dating back to 1910.
The audit considered 4,351,125 temperatures recorded daily before September 1, 1972, and 4,229,458 temperatures recorded since that date, totalling 8,580,583 minima and maxima altogether.
The chart below of all temperatures recorded at the HQ weather stations before September 1972 shows that almost one in three was logged as a whole .0 Fahrenheit degree. With accurate recording, there should be a distribution close to 10% for each fraction.
A breakdown of this data shows that 32.02% of all Fahrenheit minima recorded before September 1972 were rounded to .0F, and 30.07% of all maxima were rounded to .0F.
The high proportions of .1 and .9 are due to the calculation of Celsius degrees that have been converted from Fahrenheit and rounded to the first decimal before conversion back to Fahrenheit, a mathematical conversion that is imprecise. The high proportions of .1 and .9 provide strong evidence that many of these were in fact .0 Fahrenheit recordings.
It is highly likely that more than 50% of all recordings at the 177 HQ and associated stations that logged temperatures before September 1972 were rounded by observers to the whole .0 Fahrenheit degree. Many of these observers were country folk who volunteered to log temperatures each day from the Stevenson Screen weather stations provided by the Australian Bureau of Meteorology, which instructed that Fahrenheit degrees should be recorded to the tenth fraction at all times.
This audit demonstrates that it was common for observers to ignore the instruction to record precise Fahrenheit fractions, with many rounding down or up to the whole .0F. Torok et al acknowledge that if the practice of truncating down to the whole degree was common, there would have been an "artificial warming" in the early 1970s due to the temperatures before metrication appearing to be lower than they actually were.
In The development of a high quality historical temperature data base for Australia (page 210), Simon Torok writes:
"Some doubts remain regarding possible discontinuities in the early 1970s, when Australian temperature recordings changed from the Fahrenheit scale to the Celsius scale. New thermometers were issued at that time. Thermometers before and after the change to metric units underwent calibration tests and a search through the documentation regarding the change does not suggest possible instrumental discontinuities. As well, measurements with thermometers manufactured before and after the changeover date have been compared (Alex Kariko, 1996 personal communication). This check revealed no instrumental reason for expecting discontinuities at the time of the changeover to metric units.
"Another possible source of a non-climatic discontinuity is observer practice in recording temperatures. Some observers, prior to the change to metric units, recorded temperatures in whole degrees Fahrenheit, instead of recording to the nearest tenth of a degree, as specified in directions to observers. If many observers truncated their measurements to the nearest whole degree below the actual measurement, prior to metrication, and after metrication 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 widespread, 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. If only a small number of observers regularly truncated the observations, this would have been detected as a discontinuity by the objective tests in this study."
Detailed displays of BoM observer field books at Eversleigh 056056 in NSW from 1877 to 1907 can be downloaded from the University of Newcastle (slow server, be patient). A typical year at Eversleigh was 1906, when there were 124 whole .0F degrees among 363 days of maximum recordings, representing 34.2% of the total. There were 134 whole .0F degrees among 364 days of minimum recordings, representing 36.8% of the total. The 1906 Eversleigh Fahrenheit temperature recordings have been digitised (Excel 49kb) and this field book is typical of whole degree rounding proportions at most Australian weather stations in the Fahrenheit era.
It is unknown and impossible to determine what proportion were rounded up or down by observers but a natural human tendency would be to round down, an assumption supported by the greater numbers of fractions recorded below rather than above .5F. In essence, observers are more likely to round down from 38.4 rather than round up from 38.6 simply because many feel 38 is more or less correct but most feel 39 is wrong.
Evidence that the tendency was to round down is supported by the frequency of timeline temperature trends in many weather station records to shift up as of metrication in 1972 (see mean, minima and maxima), a trend that could only occur if a majority of previous recordings were rounded down - unless Australia's weather coincidentally also warmed sharply and immediately as of 1972.
Whether the downward or upward drift predominated or was balanced is a matter for conjecture but the evidence remains clear that a large proportion of degrees recorded in Australia before 1972 were inaccurate or estimates of the actual temperatures (see truncation).
Examples of recording errors ignored or overlooked within the BoM HQ data series include Swan Hill Post Office (1/1/60-31/8/72), Cowra Airport (1/10/66-31/8/72) and Moree Post Office (1/1/57-31/8/72), charted below to show the actual numbers of maximum Fahrenheit fractions recorded before September 1972.
The examples charted above demonstrate how the proportions of whole .0 Fahrenheit detected in this audit are probably a gross underestimate.
This audit has designed a Compliance Index wherein the percentage of .0, .1 and .9 readings are added before subtracting 30% (three times the expected 10% frequency), with the results inverted and figures below zero indicating possible non-compliance. To allow for 5% uncertainty within each tenth, the non-compliance benchmark is -15 and stations are ranked according to how they reflect the theoretical distribution of tenths in their observations.
All Australian Fahrenheit records from HQ stations before September 1972 can thus be charted to illustrate the compliance of their .9, .0 and .1 fractions with the expected 10% proportion for each:
High proportions of .1 and .9 are an artefact of converting rounded Celsius degrees to rounded Fahrenheit degrees. More accurate approximations of the .9, .0 and .1 fractions can be calculated by assigning to .1 and .9 the mean of values for .2 to .8, and adding the excess from .1 and .9 to .0 - as charted below:
To reduce the bias of errors from stations with low counts of .0F, total inferred values for .0 to .9 for all Australian HQ and associated weather stations can be calculated by only considering those with greater than 1,500 within minima and maxima combined ... see chart.
The inferred proportion of all Australian HQ and associated weather stations with greater than 1,500 recordings of rounded .0F is 52.66%. This demonstrates that the size of the database creates little bias within the results (see Kenskingdom for more).
Extreme Temperature Events in Australia is a 2001 PhD thesis by ACORN architect Blair Trewin from the Bureau of Meteorology in which he estimates that 50.8% of all Fahrenheit temperatures recorded from 1957 to 1971 were rounded (.0) by observers. The relevant charts with rounded Fahrenheit percentages at most Australian weather stations can be viewed here.
The bureau argues that although more than half of Australia's temperature history before 1972 was rounded, there is no evidence that truncation (rounding down rather than up) was common so the practice had no effect on averages.
The chart below of all temperatures recorded at the HQ weather stations since September 1, 1972, shows that almost twice as many whole 0. Celsius degrees were recorded than the average of other fractions. With accurate recording, there should be a distribution close to 10% for each fraction.
A breakdown of this data shows that 20.11% of all Celsius minima recorded since September 1972 were rounded to .0C, and 18.82% of all maxima were rounded.
The high proportion of whole .0 Celsius recordings is partly due to a large number of consecutive blocks of .0 Celsius fractions that appear over months and sometimes years in more than half the temperature records of all Australian weather stations.
The BoM's post 1972 raw recordings also show periods with blocks or high proportions of .5C in certain time periods, usually in conjunction with the blocks of .0C, and this is reflected in the graph above.
As noted above, a proliferation of .5C recordings are in the BoM records since 1972, often associated with blocks of .0C. All Australian Celsius records from HQ stations since September 1972 can be charted to illustrate the compliance of their .5 and .0 fractions with the expected 10% proportion for each:
The results of the High Quality raw temperature audit and analysis from a national perspective can be found at Kenskingdom, examining and detailing the various data limitations that may bias the audit results.
In March 2012, the Bureau of Meteorology introduced the ACORN-SAT temperature dataset which supersedes the High Quality dataset, replacing various weather stations and applying new adjustments to estimate Australia's temperature trends since 1910.
An audit of the ACORN-SAT stations in Western Australia shows raw temperatures in the new weather stations are influenced by rounded .0 and .5 patterns almost identical to their influence in the High Quality dataset.
In technical manuals released with Australia's ACORN-SAT temperature dataset, the Bureau of Meteorology states that tests show metrication in 1972 caused a .1C warming breakpoint that year, the precise anomaly unknown and unadjusted due to cool La Nina weather conditions from 1973 to 1977 which masked the data influence of Celsius conversion.
An audit of New Zealand's Seven Station raw temperature dataset shows Fahrenheit .0 rounding patterns similar to audit results in Australia, suggesting a common unrecognised and unadjusted anomaly in countries which have converted their temperature scales from Fahrenheit to Celsius starting in the 1960s.
These findings suggest that much of Australia's temperature record before metrication in September 1972 is based on averaged estimates and contains numerous errors that should be further investigated through a professional independent audit of the BoM's records.
HI QUALITY audit|