Have automatic weather stations corrupted Australia's temperature records?
Following Australia's Glasgow commitment to net zero CO2 emissions by 2050, and having previously had a close look at the extraordinary distribution of rounded .0 Fahrenheit decimals before 1972 metrication, let's pay attention to another major equipment change that might have influenced temperature trends - automatic weather stations (AWS).
AWS installations have replaced most but not all liquid-in-glass thermometers that used to be read manually. They use an electronic probe to measure temperatures every second, constantly logging the results and relaying the data back to the Bureau of Meteorology.
Bureau data downloads at the time of writing suggest that 105 of the 112 weather stations in the Australian Climate Observation Reference Network (ACORN) are AWS, their average start year 1996. Of those 105 AWS installations, 14 were from 1986 to 1991, 66 from 1992 to 1999, and 25 from 2000 to 2018.
Wait a second
One second AWS observations are obviously far more sensitive than the preceding liquid-in-glass thermometers that either didn't or took longer for the mercury or alcohol to react to a brief gust of hot wind or the remnants of jet exhaust at an airport.
The BoM insists that it compensates for this by averaging three of the one second readings each minute to correlate with how thermometers used to read air temperatures. It's worth noting that in America they prefer to average over five minutes.
Skeptics claim there's evidence that because of the one second sensitivity, an AWS will more frequently record upper range temperature spikes and exaggerate maxima more than minima. Furthermore, most 230 litre Stephenson screen shelters were replaced with 60 litre shelters from the 1990s and these smaller screens altered the dimension of thermal influence on the new electronic probes.
Daily temperature percentile analysis
The extreme temperature influence of AWS one second sensitivity can be measured with trends in the 90th percentile (hottest 10%), 95th percentile (hottest 5%) and 99th percentile (hottest 1%) of all daily maxima during the 20 years before and after electronic probes replaced manual thermometers at all but seven of the 112 ACORN weather stations used by the BoM to calculate Australia's national average temperatures.
The seven excluded locations had not been converted to AWS and continued with manual thermometer observations in 2021.
Percentiles are calculated from every unadjusted daily RAW maximum observation (over 12 months in each year) since 1910 at each station (1910 the first year of estimated annual Australian temperatures within ACORN), with collective annual percentile averages calculated over the 20 years before and after AWS installation at each station.
Such analysis shows 56 ACORN AWS stations (exactly half the total number) had an immediate increase in the average annual frequency and temperature of these hot, very hot and extremely hot days when the electronic probes were installed.
The four stations charted below illustrate changes in 90th, 95th and 99th percentile temperature days 20 years before and after an AWS was installed (black line frequency, red line temperature).
Among the 49 remaining ACORN AWS stations, there was either little change or a reduction in the average annual frequency and temperature of 90th, 95th and 99th percentile days, their collective averages charted below.
The collective annual average 90th, 95th and 99th percentile frequency and temperature at all 105 ACORN AWS locations are charted below.
There are seven remaining ACORN stations that are not AWS and continue to have manual thermometer observations, and their annual average frequency and temperature of 90th, 95th and 99th percentile days since 1962 are charted below (Barcaldine opened in 1962), with rainfall trends likely to have influenced both frequency and temperature over the past 20 years at both the manual stations and all 105 AWS locations analysed above.
In the 90th percentile (10% hottest days), the seven manual ACORN weather stations without one second electronic temperature probes were 0.06C warmer in 2000-2021 than 1962-1999, compared to a 0.11C increase at the other 105 locations in the five years following AWS installation.
Hot, very hot and extremely hot days have a strong influence on monthly and annual average temperatures calculated by the bureau, partly due to their frequency and partly because more extreme temperatures have a greater influence than cooler temperatures on overall averages.
Apart from a majority of ACORN stations (56 v 49) experiencing an increase in the frequency of 90th, 95th and 99th percentile days, analysis results show that although conversion from manual to AWS observations caused a frequency and temperature increase at some stations and a decrease at others, the AWS extreme temperature influence is biased toward heating rather than cooling.
90th percentile (10% hottest days)
< Five years > AWS at 56 stations : 24.5% frequency increase / 0.33C temperature increase
< Five years > AWS at 49 stations : 10.5% frequency decrease / 0.12C temperature decrease
< 10 years > AWS at 56 stations : 28.6% frequency increase / 0.27C temperature increase
< 10 years > AWS at 49 stations : 4.7% frequency decrease / 0.12C temperature decrease
95th percentile (5% hottest days)
< Five years > AWSat 56 stations : 35.6% frequency increase / 0.30C temperature increase
< Five years > AWSat 49 stations : 16.3% frequency decrease / 0.03C temperature decrease
< 10 years > AWS at 56 stations : 41.5% frequency increase / 0.25C temperature increase
< 10 years > AWS at 49 stations : 9.6% frequency decrease / 0.07C temperature decrease
99th percentile (1% hottest days)
< Five years > AWS at 56 stations : 69.7% frequency increase / 0.18C temperature increase
< Five years > AWS at 49 stations : 27.0% frequency decrease / 0.04C temperature increase
< 10 years > AWS at 56 stations : 80.5% frequency increase / 0.06C temperature increase
< 10 years > AWS at 49 stations : 19.4% frequency decrease / no temperature change
The frequency and temperature of hot, very hot and extremely hot days continued to increase rather than plateau in the 20 years following AWS installation, possibly because 19 stations have converted since 2001 (20 years before 2021) with an accumulating influence, and possibly because many locations experienced a rainfall decline after the year 2000 which increased the likelihood of hot days. A majority of AWS electronic probes have also been replaced since the year 2000.
For example, Karijini North AWS opened in 2018, replacing the manual thermometer observations at nearby Wittenoom.
This analysis raises questions about the accuracy of Australian temperature averages since the 1980s, as well as the veracity of claimed record heat days where the artificial influence of one second sensitivity in automatic weather stations may have caused the record rather than actual heat.
The analysis results also raise questions about temperature trends globally since most countries in the world began using automatic weather stations instead of liquid-in-glass thermometers during the 1980s and 1990s.
Note : Annual percentile calculations for all 112 ACORN weather stations in the AWS analysis above can be downloaded here (2.8mb).
Long-term weather station temperatures
Let's look at decimal frequency and temperatures within the 58 long-term ACORN stations that were open in 1910.
As per the claims that an AWS is more sensitive to extreme maxima rather than minima, the first maxima chart shows a clear temperature shift around 2000 and the second minima chart shows stability since 1980 with a slight cooling probably due to a drop in rainfall (although less annual variability than in those earlier years).
The average maximum at the 58 long-term ACORN stations was 25.24C in 1990-99 and 25.62C in 2000-09, an abrupt 0.38C increase. Average minima at the 58 stations was 13.94C in 1990-99 and 13.83C in 2000-09 - it cooled 0.11C!
However, annual average rainfall at the 58 long-term ACORN stations dropped from 717.5mm in 1990-99 to 647.0mm in 2000-09. Cloud cover is an influence but when rainfall previously dropped a similar amount from 1950-59 (772.9mm) to 1960-69 (703.9mm), maxima at the 58 stations increased .only 05C.
So as in the last 1972 metrication post (link), let's again look at the distribution of decimal observations to see if they detect any change in how temperature readings were influenced by the AWS instrument changes.
Here the waters are muddied because, as you can see in the tables if you look closely, the number of rounded .0C observations increased significantly from 1997 to 2004 in both maxima and minima.
This is because many of the automatic weather stations installed in the mid-1990s had a recording fault acknowledged by the BoM that caused them to only report .0C readings.
For example, every single day at Cape Leeuwin from 18 May 1998 to 8 January 2003 recorded a temperature with a decimal of .0, and not a single .1, .2, .3, etc. It took well over four years but the BoM eventually noticed the problem and fixed it, claiming it was an equal up and down decimal error so it didn't corrupt the average temperatures.
The table below shows the evolution of .0, .1 and .9 Celsius decimals at the 58 long-term ACORN stations in the decade before 1996, the following nine years inclusive of acknowledged AWS rounding errors, the decade to 2018 and then 2019, Australia's driest year on record.
The frequency of .0C declined 34.8% from 1986-95 to 2019, the frequency of .1C increased 24.3%, and the frequency of .9 increased 41.5%.
To be pedantic, from 1986-1995 to 2019, the frequency of .1, .2, .3 and .4 increased 11.1%, and the frequency of .6, .7, .8 and .9 increased 13.9%.
The average number of .9 compared to .1 maximum recordings increased by 175 from 1986-1995 to 2005-2018, while the average number of .9 compared to .1 minimum recordings increased by 25.
Might the AWS increase in higher decimals help to explain why maxima increased by 1.39C at the 58 stations between those periods, while minima increased by only 0.22C?
BoM testing of AWS response times
The BoM has tested one second variations within the minute at 6am and 3pm for replacement temperature probes at 98 ACORN stations, and acknowledges that at 17 of them there was a significant breakpoint likely to suggest a change in the instrument response time (source).
Due to area averaging that influences temperature homogenisation at neighbouring stations (hundreds of kilometes distant in the remote outback), Alice Springs is arguably the most influential station in Australia in terms of the ACORN calculation of national averages.
The table below displays maximum, minimum, rainfall and solar exposure totals and averages at Alice Springs Airport from 2009-2010 to 2012-2013.
There were 27 fewer days of rainfall in 2012-2013 than in 2009-2010, but 66 more days at or above 30C+, 65 more days at or above 35C, and 56 more days at or above 40C.
Solar exposure, which is influenced by cloud cover, was overall the same from 2009-2010 to 2012-2013, with more days above average in 2012-2013 than 2009-2010 but those days having a lower MJ/m2 intensity in the latter years.
Comparing 2009-2010 with 2012-2013 at Alice Springs Airport, average maximum at or above 30C warmed 1.0C, average maxima at or above 35C warmed 1.1C, and average maxima at or above 40C warmed 0.6C, with the average maxima for all days throughout the years increasing 1.9C.
There was less impact on minima from the AWS Almos probe replacement on 11 November 2011, although the average minimum for all days cooled 0.6C from 2009-2010 to 2012-2013.
The probe replacement influence on maxima can be illustrated with charts:
Decreased rainfall is an influence but this data contradicts the BoM claim that the November 2011 probe replacement at Alice Springs Airport only caused a 3pm increase of 0.16C and a 6am increase of 0.03C, with resultant warming of neighbouring ACORN stations due to area averaging.
Alice Springs Airport had 76.8mm of rainfall in 2009 with an average maximum of 30.0C, but 194.2mm in 2013 with an average maximum of 30.8C. There were 201 days above 30C in 2009 and 223 in 2013.
An independent analysis examines the two years before and after the original installation of AWS probes at ACORN stations, finding that average minima warmed 0.06C and average maxima warmed 0.11C. If you want a detailed breakdown of AWS installations, replacements and their influence on temperatures, pop over to a web page quite aptly titled Climate change or instrument change?
The analysis also looked at the two years before both the original installation and replacement of probes at the stations, finding that average minima warmed 0.07C and average maxima warmed 0.31C.
Also analysed are the number of rainfall days, very hot 40C+ days and solar exposure in the two years before and after.
Many of the probes were replaced over time and, measuring these, average maxima were 0.46C warmer in the two years after probe replacement compared to the two years before, with an average -0.01 decrease in solar exposure.
Parallel non-AWS and AWS weather stations
Automatic weather stations are said to exaggerate the number of maxima which spike rapidly due to brief gusts of hot air, with a similar influence on minima with brief gusts of cold air, compared to liquid-in-glass thermometers with a slower reaction time.
The table below shows the total number and average temperature of daily observations within cool and hot categories at 12 parallel weather stations.
To minimise environmental influences, the 12 are selected only where non-AWS and AWS shelters are positioned less than a kilometre from each other at sea level elevations the same or with no more than three metres difference, and only with days when observations were recorded at both stations over the same time periods (i.e days with missing observations at either station are excluded from both).
The average of all maxima above 30C recorded at AWS stations was 0.1C warmer than in parallel non-AWS stations, and the average of all minima below 20C recorded at AWS stations was 0.2C cooler than in parallel non-AWS stations.
All 12 AWS stations are manufactured by Almos. There is no consistency among the 12 but the overall averages suggest minima are recorded cooler than maxima are recorded warmer in AWS compared to their nearby non-AWS stations. However, the results do not suggest more frequent or hotter temperatures are recorded in automatic weather stations on extreme heat days above 40C.
Similar results can be seen when comparing the hottest and coldest days and nights, rather than averages, recorded at the 12 parallel stations.
These results suggest AWS response times might partly explain why Australia's national average maximum has increased at a significantly higher rate than the average minimum, as illustrated in the temperature charts higher on this page.
Instrument change = temperature change
That's all quite a jumble of decimals, time periods, one second recordings, maxima, minima and electronic probes, but there's a consistent pattern along the trail of breadcrumbs - the introduction of automatic weather stations appears to have caused a fundamental change in the measurement of air temperatures and the changes (particularly higher decimals) are likely to affect maxima more than minima.
And what happened to temperatures at the 58 long term ACORN stations? From 1986-95 to 1996-2005, maxima at the 58 stations increased 0.28C and minima cooled 0.03C. There was an annual average 703.4mm of rainfall in 1986-95 and 687.5mm in 1996-2005.
From 1979-95 to 1996-2012, maxima increased 0.31C and minima increased 0.01C. Annual rainfall at the 58 stations averaged 700.7mm in 1979-95 and 703.3mm in 1996-2012. There was a tiny bit more rainfall cloud cover but maxima jumped while minima behaved as it should.
From 1986-1995 to 1996-2018, maxima at the 58 stations increased 0.47C and minima increased 0.04C. Average annual rainfall at the stations averaged 703.4mm in 1986-1995 and 689.0mm in 1996-2018.
These comparison suggest either that climate warming affects daytime maxima but not overnight minima, or one second averages from AWS probes are more sensitive to high temperatures than low temperatures.
However and aside from 1972 metrication and AWS corruption, it's worth noting that the 2005-2018 average temperatures might also be influenced by something that happened in 2013, and the next article will analyse what that influence was.
Note : Annual Fahrenheit and Celsius decimal counts for minima and maxima from 1910 to 2019 at the 58 long-term ACORN stations are contained in an Excel file that can be downloaded here.
Note : The accuracy of pre-metric .0F decimal estimations in this post are validated through comparison with a 2001 PhD thesis titled Extreme Temperature Events in Australia by the BoM's Blair Trewin in which he calculates the .0F proportion of all observations at 94 ACORN weather stations from 1957 to 1971, presumably using original digitised Fahrenheit observations held by the bureau. A study extract of his calculations can be viewed here, showing that 51.5% of the observations were .0F. His calculations can be compared with the decimal calculation formulas used in this post here, showing they average 0.F proportions at the 94 stations from 1957 to 1971 at 51.3%, compared to 51.5% calculated in the Trewin thesis.
Note : See BomWatch for a thorough analysis of four weather stations in its analysis of are Australia's automatic weather stations any good.
This website produced by Scribeworks 2008-2022