Metrication influence on Australian temperature trends
Did instrument and/or observation changes associated with the 1972 metrication of weather stations have any influence on Australia's longterm temperature trends?
This question was addressed in Techniques involved in developing the Australian Climate Observations Reference Network  Surface Air Temperatures (ACORNSAT) dataset (p70) published in 2012 by the Australian Bureau of Meteorology and the CSIRO.
Three dataset comparisons were tested using sea surface temperatures, upper air observations at ACORNSAT locations with mean 850 hPa temperatures, and minima and maxima at locations with temperatures at 1500 and 1600 respectively.
The decision not to adjust ACORNSAT for a suspected artificial 0.1C mean temperature increase was essentially based on the premise that major La Nina events, with record rainfall and cloud cover, may have caused a natural warming variation during the early 1970s.
Metrication on 1 September 1972 resulted in equipment changes at all weather stations and a significant reduction in the proportion of station observers who frequently recorded Fahrenheit thermometer readings as whole numbers without a fraction (e.g. 84 instead of 84.4, 95 instead of 94.7, 73 instead of 73.7).
The implications of premetric temperature rounding are referenced in A historical annual temperature dataset for Australia (Torok and Nicholls, BoM Research Centre, Melbourne, 1996):
Australian averaged annual rainfall  longterm 19102019 average 458.5mm (temperature estimates based on RAW observations at all 112 ACORN stations)
196871  453.5mm / average maximum 24.56C / average minimum 12.91C
197376  631.6mm / average maximum 24.67C / average minimum 13.22C
196271  430.0mm / average maximum 24.66C / average minimum 12.94C
197382  532.6mm / average maximum 25.00C / average minimum 13.23C
191026  444.9mm / average maximum 24.79C / average minimum 13.22C
200319  488.1mm / average maximum 25.86C / average minimum 13.63C
Record rainfall and cloud cover would normally be expected to result in a warming of overnight minima as heat is trapped by clouds, preventing its release into the upper atmosphere and space.
Conversely, record rainfall and cloud cover would normally be expected to result in a cooling of daytime maxima as clouds increase the albedo reflection of sunlight and the evaporation of moisture dissipates the energy of heat.
Heavy rainfall and abundant clouds would not normally be expected to increase mean temperatures. Questions should be asked of the assumption that wet weather, rather than an artificial influence, caused the 1972 temperature increase.
Below is an extract from Recent apparent changes in relationships between the El Nino  southern oscillation and Australian rainfall and temperature, published in 1996 by Nicholls et al. This study identifies the 1972 temperature increase but tries to attribute it to a sudden differential warming of the Indian Ocean relative to the Pacific Ocean, which is as credible as more rainfall and clouds causing warmer temperatures.
Analysis
It is incongruous for record rainfall and cloud cover to result in minima increasing but maxima not decreasing. This analysis tests unadjusted raw temperatures at 110 stations in the ACORNSAT dataset, including their relationship to annual rounding proportions, to determine whether it is appropriate to attribute the 0.1C mean warming in 1972 to natural rather than artificial causes.
The ACORNSAT stations Learmonth and Rabbit Flat are excluded because they opened in 1970 and 1976, invalidating their inclusion to a comparison of 196871 and 197376 averaged temperatures.
Daily raw minimum and maximum temperature data were downloaded from the Bureau of Meteorology's Climate Data Online website and converted from Celsius to Fahrenheit, a script calculating annual and monthly temperature averages (monthly with minimum 15 days of recordings), as well as annual percentage proportions of temperature rounding based on a .91><.09 formula to determine if any day's reading was a rounded Fahrenheit number (see benchmark accuracy).
Minima should rise due to increased rainfall cloud cover. Key points:
At 110 ACORNSAT stations, average minima increased 0.32C from 12.94C in 196871 to 13.26C in 197376. Average rounding proportions fell from 44.62% to 21.20%.
At 110 ACORNSAT stations, average minima increased 0.33C from 12.94C in 196271 to 13.27C in 197382. Average rounding proportions fell from 50.90% to 21.27%.
At 110 ACORNSAT stations, average minima increased 0.21C from 13.28C in 191026 to 13.49C in 200016. Average rounding proportions fell from 56.68% to 13.61%.
The minimum temperature trends charted above might be misleading as there were 58 stations open in 1910 and the remaining 54 were progressively added to ACORNSAT over the decades before all 112 were operating by the mid 1970s.
At the 58 ACORNSAT stations that were all open in 1910, average minima increased 0.30C from 13.38C in 196871 to 13.68C in 197376. Average rounding proportions fell from 47.40% to 22.33%.
At the 58 ACORNSAT stations that were all open in 1910, average minima increased 0.29C from 13.39C in 196271 to 13.68C in 197382. Average rounding proportions fell from 55.22% to 22.24%.
At the 58 ACORNSAT stations that were all open in 1910, average minima increased 0.53C from 13.39C in 191026 to 13.92C in 200016. Average rounding proportions fell from 56.72% to 14.47%.
The bureau excludes eight of the 112 stations from estimated national temperature trends because they are artificially warmed by urban heat (Sydney, Melbourne, Adelaide, Hobart, Townsville, Rockhampton, Richmond NSW, Laverton).
At the 104 ACORNSAT stations used by the bureau to estimate trends, excluding city and major regional centres influenced by urban heat, average minima increased 0.33C from 12.92C in 196871 to 13.25C in 197376. Average rounding proportions fell from 45.43% to 21.65%.
At the 104 ACORNSAT stations used by the bureau to estimate trends, excluding city and major regional centres influenced by urban heat, average minima increased 0.37C from 12.92C in 196271 to 13.29C in 197382. Average rounding proportions fell from 52.44% to 21.69%.
At the 104 ACORNSAT stations used by the bureau to estimate trends, excluding city and major regional centres influenced by urban heat, average minima increased 0.12C from 13.39C in 191026 to 13.51C in 200016. Average rounding proportions fell from 58.39% to 13.65%.
In addition to the bureau's exclusion of eight urbanwarmed locations, the figures and chart below also exclude Learmonth and Rabbit Flat to allow a more precise estimation of pre and post 1972 metrication temperatures.
At the 102 ACORNSAT stations used by the bureau to estimate trends, excluding Learmonth and Rabbit Flat, average minima increased 0.31C from 12.96C in 196871 to 13.27C in 197376. Average rounding proportions fell from 45.57% to 21.76%.
At the 102 ACORNSAT stations used by the bureau to estimate trends, excluding Learmonth and Rabbit Flat, average minima increased 0.33C from 12.96C in 196271 to 13.29C in 197382. Average rounding proportions fell from 52.49% to 21.83%.
At the 102 ACORNSAT stations used by the bureau to estimate trends, excluding Learmonth and Rabbit Flat, average minima increased 0.11C from 13.39C in 191026 to 13.50C in 200016. Average rounding proportions fell from 58.39% to 13.54%.
Maxima should decrease due to increased rainfall cloud cover. Key points:
At 110 ACORNSAT stations, average maxima was static from 24.70C in 196871 to 24.70C in 197376. Average rounding proportions fell from 42.76% to 19.20%.
At 110 ACORNSAT stations, average maxima increased 0.26C from 24.71C in 196271 to 24.97C in 197382. Average rounding proportions fell from 48.20% to 19.52%.
At 110 ACORNSAT stations, average maxima increased 0.68C from 24.89C in 191026 to 25.57C in 200016. Average rounding proportions fell from 54.31% to 13.85%.
The maximum temperature trends charted above might be misleading as there were 58 stations open in 1910 and the remaining 54 were progressively added to ACORNSAT over the decades before all 112 were operating by the mid 1970s.
At the 58 ACORNSAT stations that were all open in 1910, average maxima increased 0.03C from 24.93C in 196871 to 24.96C in 197376. Average rounding proportions fell from 43.15% in 196871 to 19.94% in 197376.
At the 58 ACORNSAT stations that were all open in 1910, average maxima increased 0.19C from 24.99C in 196271 to 25.18C in 197382. Average rounding proportions fell from 51.05% to 19.90%.
At the 58 ACORNSAT stations that were all open in 1910, average maxima increased 0.66C from 25.11C in 191026 to 25.77C in 200016. Average rounding proportions fell from 53.69% to 14.35%.
The bureau excludes eight of the 112 stations from estimated national temperature trends because they are artificially warmed by urban heat (Sydney, Melbourne, Adelaide, Hobart, Townsville, Rockhampton, Richmond NSW, Laverton).
At the 104 ACORNSAT stations used by the bureau to estimate trends, excluding city and major regional centres influenced by urban heat, average maxima increased 0.02C from 24.90C in 196871 to 24.92C in 197376. Average rounding proportions fell from 43.19% to 19.55%.
At the 104 ACORNSAT stations used by the bureau to estimate trends, excluding city and major regional centres influenced by urban heat, average maxima increased 0.30C from 24.92C in 196271 to 25.22C in 197382. Average rounding proportions fell from 49.38% to 19.81%.
At the 104 ACORNSAT stations used by the bureau to estimate trends, excluding city and major regional centres influenced by urban heat, average maxima increased 0.54C from 25.28C in 191026 to 25.82C in 200016. Average rounding proportions fell from 56.19% to 13.73%.
In addition to the bureau's exclusion of eight urbanwarmed locations, the figures and chart below also exclude Learmonth and Rabbit Flat to allow a more precise estimation of pre and post 1972 metrication temperatures.
At the 102 ACORNSAT stations used by the bureau to estimate trends, excluding Learmonth and Rabbit Flat, average maxima was static from 24.87C in 196871 to 24.87C in 197376. Average rounding proportions fell from 43.75% to 19.70%.
At the 102 ACORNSAT stations used by the bureau to estimate trends, excluding Learmonth and Rabbit Flat, average maxima increased 0.26C from 24.89C in 196271 to 25.15C in 197382. Average rounding proportions fell from 49.73% to 20.05%.
At the 102 ACORNSAT stations used by the bureau to estimate trends, excluding Learmonth and Rabbit Flat, average maxima increased 0.60C from 25.14C in 191026 to 25.74C in 200016. Average rounding proportions fell from 55.93% to 13.77%.
Discussion
Average national raw maxima were static from 19681971 to 197376 among 110 ACORN stations but an increase is masked by the record cloud cover and rainfall that cooled daytime temperature readings in the early 1970s.
The averaged 0.32C increase in national average minima from 196871 to 197376 was partly caused by nighttime cloud cover trapping heat, but the increase is substantial. As an example of a similar record rainfall period, 200508 rainfall averaged 467.7mm annually and in 200912 averaged 587.7mm, but average annual minima decreased from 13.52C to 13.49C between those four year periods (and maxima cooled from 25.73C to 25.50C).
During a strong La Nina period that increased nighttime minima by 0.32C averaged across Australia, the static daytime maxima suggests an artificial rather than natural influence on temperature recordings. The Pacific ENSO shift said to have caused a global temperature increase didn't start until early 1977 (source).
The immediate and significant decline in temperature rounding after 1972 metrication may be indicative of a majority of regional weather station observers previously choosing to evenly round or in some cases truncate down the temperature scale they could see on their Fahrenheit thermometers.
An example of truncation might be an early morning farmer (some with poor eyesight, others rushing to herd their dairy cattle) who sees a temperature reading considerably above 67F but decides to record 67F because it seems more accurate and honest than writing 68F, creating a premetric cooling bias. The proportion of observers who regularly truncated their Fahrenheit temperature recordings is unknown but anything above 50% immediately creates a bias.
This analysis demonstrates that temperature rounding was widespread at 69.9% for minima and 67.4% for maxima when averaged at all 112 ACORN stations from 1910 to 1971, and was sufficiently common to produce a major discontinuity.
From 196871 to 197376 at 110 stations, there was a mean temperature increase (0.00 + 0.32 / 2) of 0.16C, which is higher than the bureauâ€™s estimate of a mean rise of 0.1C and represents 16% of the claimed 1.0C increase in ACORNSAT mean temperatures from 1910 to 2016.
In 1996, Simon Torok et al published Recent apparent changes in relationships between the El Nino  southern oscillation and Australian rainfall and temperature which analysed an apparent maximum temperature increase around 1972/73 of about 0.3C comparing 19101972 with 19731992, but downplayed the possible impact of changed metric instrumentation and instead attributed the temperature/rainfall/SOI correlation shift to increased Indian Ocean temperatures that may have occurred at about the same time that Celsius thermometers replaced Fahrenheit thermometers.
The paper did not analyse the minimum temperature/rainfall/SOI correlation. The 0.3C maximum temperature shift calculated in the paper is similar to the 0.26C shift calculated above among the 102 nonurban ACORN stations (excluding Learmonth and Rabbit Flat) from 19621971 to 19731982.
The analysis on this page also finds a 0.33C increase in minimum temperature calculated above among the 102 nonurban ACORN stations (excluding Learmonth and Rabbit Flat) from 19621971 to 19731982.
The Torok et al paper recognises the increase in Australian rainfall since the early 1970s but asserts that Indian Ocean temperatures may have caused the maximum temperature increase, which was similar to the minimum temperature increase, and confirms the calculations on this page but with a dubious explanation that it was because of increased rainfall and ocean temperatures rather than the metrication change that happened at the same time.
It is difficult to determine whether the prominence of observation rounding before 1972 had a significant influence on average temperatures.
However, the two charts below are based on the unique rounding percentage median for each of 110 ACORN locations, which is used for each station to average temperatures above or below that particular median in both RAW min and max.
Conclusion
The averaged failure of Australian maxima to decline during years of record rainfall and cloud cover suggests an artificial metrication influence that should be adjusted in Australia's homogenised ACORNSAT temperature set.
Note: The spike in temperature rounding from 1997 to 2006 within all charts above is due to Automatic Weather Stations being installed at many ACORNSAT sites, a significant number with faulty readings that for several years rounded all temperatures evenly up and down. Maxima in particular had a sudden upward shift to a warmer plateau since the bureau's AWS rounding repairs concluded in 2006. This may be the influence of AWS and the installation of smaller Stevenson screens since the late 1990s.
ACORNSAT Observation Practices (p1113) details when 106 weather stations had small Stevenson screens installed. The data show 29 were installed from 1900 to 1989 and 77 from 1990 to 2009. Among the 77, the average year of installation was 1998: 1990  1, 1991  1, 1992  4, 1993  2, 1994  4, 1995  5, 1996  7, 1997  15, 1998  6, 1999  5, 2000  10, 2001  6, 2002  2, 2003  2, 2004  3, 2005  1, 2007  1 , 2009  1.
The chart below shows that when 40% of all small Stevenson screens were installed from 1997 to 2001, Australian average rainfall increased significantly but maximum temperatures remained warmer than their pre1997 averages when there was 22% less rainfall. When rainfall returned to longterm average levels from 2002 to 2009, maxima averaged 0.62C warmer than 199096. This suggests high rainfall partly masked an artificial warming influence caused by the widespread installation of small Stevenson screens (large Stevenson screen interior 0.23m3, small Stevenson screen interior 0.06m3).
Note: Annual temperature rounding estimates in this analysis are benchmarked against Extreme Temperature Events in Australia, a 2001 PhD thesis by ACORNSAT architect Blair Trewin from the Bureau of Meteorology. His study estimates that among 94 of the ACORNSAT stations, the average proportion of Fahrenheit temperatures rounded by observers from 1957 to 1971 was 51.5%. Relevant charts with rounded Fahrenheit percentages at most Australian weather stations can be viewed here.
The C>F conversion script used in this analysis estimates the rounding proportion at the same 94 stations from 1957 to 1971 was 51.3%. This demonstrates that the averaged comparison of annual rounding proportions is accurate. A comparison of the 195771 benchmarks and this analysis calculation of rounding for all stations can be viewed here. Some benchmarked stations have identical rounding proportions to this study's calculations. Others have higher or lower proportions, dependant on each station observer's personal habit of adding .1 or .9 to their Fahrenheit temperature readings.
Decadal trends
A comparison of rounding proportions, temperatures and rainfall can also be viewed in a decadal context among longterm weather stations that were taking observations in 1910:
Alternatively, rounding proportions, temperature and rainfall can be compared at all 112 ACORN stations from 1960 to 2018:
Download spreadsheet of annual temperatures and rounding proportions 19102016 (Excel 1.3mb)
Download spreadsheet of average monthly temperatures each year 19102016 at all 112 stations (Excel 2.8mb)
Decoding the precision of historical temperature observations (PDF 6.5mb), Royal Meteorological Society, 2015
Top of page
Home
This website produced by Scribeworks 20082022
