Analysis of Parallel Tmax Data from Brisbane Aero

Dr Jennifer Marohasy has recently stirred up the Bureau of Meteorology and their usual uncritical apologists with her analysis of three years of parallel data obtained for Brisbane Airport (after years of denial and obstruction by the BOM).  This relates to side-by-side recording of temperatures taken from the traditional mercury Liquid In Glass (LIG) thermometer and the repacement Automatic Weather Station (AWS) which show that “41% of the time the (AWS) probe is recording hotter than the mercury, and 26% of the time cooler.”

She also identified a step change in AWS values in December 2019 which she thought represented recalibration of the AWS system.  The BOM denied this, “explaining there was a fault in the automatic weather station that was immediately fixed and operating within specifications from January 2020 onwards.”

This post is a further analysis of the parallel data (kindly shared by Dr Marohasy).  It shows that:-

• The discontinuity in December 2019 is beyond doubt.
• There was indeed a fault in the AWS in December 2019, but the repair resulted in the “fixed” AWS reading on average 0.23 degree Celsius higher than it was before the fix (for equivalent LIG temperatures) for the next two and a half years.
• Before the AWS was “fixed” it was recording temperatures on average 0.2C cooler than the LIG.  After the “fix”, the AWS was on average less than 0.1C warmer.
• Before the “fix”, 3.6% of the AWS recordings were higher than the LIG on the same day, and 86.2% were lower.
• After the “fix”, 47 % of the AWS recordings were higher than the LIG on the same day, and 16.5% were lower.
• 26.7% of all readings before and after the “fix”were outside of the +/- 0.1C range.
• There were seasonal variations in the difference between AWS and LIG.
• Unexplained spikes continued randomly before and after the AWS was “fixed”.
• The “fixed” AWS may have begun to deteriorate again in mid-2022.

In analysing the parallel data I compared same day observations by calculating the difference between AWS and LIG readings.  Figure 1 shows daily differences in degrees Celsius as a time series.

Fig. 1: Daily differences (outliers removed)

A 31 day running mean is a good way to see changing patterns:

Fig 2: 31 day running average of differences

There was a sharp dip and sudden rise in values in December 2019 to above zero difference, at the time of the AWS fault and repair.  Note the dip below zero in June and July 2022.

Figure 3 shows the daily differences from 31 July 2019 until the “fix”.

Fig. 3: Daily values to 22/12/19.  Note large fluctuations from 16^thDecember:

I excluded values from 16^th to 22^nd December when the AWS was faulty.

BOM experts have repeatedly claimed that AWS systems report temperatures that are predominantly within +/- 0.1C of LIG readings.  From 31/7/19 to 15/12/19 there were 80 incidences (58%) of the difference of AWS minus LIG being less than -0.1C and 2 incidences (1.45%) of the difference being greater than +0.1C.  From 23/12/19 to 30/7/22 there were 30 incidences (3.2%) of difference < -0.1C and 178 incidences (18.8%) of difference > +0.1C.  26.7% of all readings before and after the “fix”were outside of the +/- 0.1C range.

The following plots analyse the data for differences outside this range.

Fig. 4:  Running count of days with difference more than +0.1C

The number of days with AWS reading more than +0.1C above LIG rose steadily from 23 December.

Fig. 5:  Running count of days with difference below -0.1C

The break in December 2019 is obvious.

A 31 day running count is a good way to see changing patterns:

Fig. 6:  31 day count of days with difference more than +0.1C

Before the AWS “fix”, differences over 0.1C were rare.  After this, the number immediately increased.  Note that the incidence of days with a difference over 0.1C fluctuated in large swings.  In early December 2020, 17 out of the previous 31 AWS readings were more than +0.1C higher than LIG, and in October 2021 there were 16.  The count continued fluctuating while gradually decreasing, with a drop to zero in early and mid-2022.  This may indicate a deterioration in the AWS system.

The fluctuations are also seen in differences below -0.1C.

Fig. 7:  31 day count of days with difference lower than -0.1C

There was clearly a discontinuity in December 2019, followed by fluctuations in the incidence.

Is there a pattern in the months after the AWS was “fixed”?

The next figures plot 31 day counts by month of the year from January 2020.

Fig. 8: Count of differences >+0.1C by month

Higher frequency of differences > +0.1C occur in early winter and summer.  Lower counts occur in March, April, July, and August.

Fig. 9: Count of differences <-0.1C by month

Note that there is a clear seasonal pattern: most differences occur in July; least from October to January.

Figures 10 and 11 are timeseries of monthly counts.

Fig. 10:  31 day counts of differences > +0.1C each month

The swings we saw in Figure 6 above are replicated with months more easily identified.  There were spring peaks in 2020 and 2021, but there was also a peak in May and June of 2020.  Differences were low from December 2021 to July 2022.  Was the AWS beginning to malfunction?

Fig. 11:  31 day counts of differences under -0.1C each month

This is (almost) the opposite of Figure 10.  There are more days with AWS less than LIG in winter months.  However winter of 2022 saw an unusually high number of days with AWS cooler than LIG.  In July 2022 up to 42% of AWS readings were more than -0.1C different from LIG. 

The next two plots show the frequency (as a percentage) of all differences before and after the “fix”.

Fig. 12: Frequency of differences between AWS and LIG: to 15/12/19

From 31/7/2019 to 15/12/2019 AWS was on average 1.7C cooler than the LIG, as shown. 

Fig. 13: Frequency of differences between AWS and LIG: from 23/12/19

The plot of difference values has shifted right.  The BOM is correct in that 75% of AWS readings after the “fix” were within +/- 0.1C of LIG, but the average difference increased from -0.17C to +0.05C: about +0.2C. 

Also, the range of differences remained large.

Fig. 14: Differences each month before and after the AWS was “fixed”

Apart from the circled values in December 2019 when the AWS was faulty, the range of differences between AWS and LIG was similar after the “fix”.   Spiles were even greater.

Greatest difference to 15/12/19:               +0.6C

Smallest difference to 15/12/19:               -0.5C

Greatest difference from 23/12/19:          +0.7C

Smallest difference from 23/12/19:          -0.7C

AWS values from 31/7/2019 to 30/7/2022 show a clear discontinuity in December 2019, but the AWS continued to “spike” above and below LIG values.  The only change was that the AWS was now reading about 0.2C warmer than before.    Figures 15 and 16 show this in a different way.

Fig. 15: Daily AWS values before and after the “fix” for corresponding LIG values

They are very close, but different.

Fig. 16: Zooming in: Daily AWS values before and after the “fix” for corresponding LIG values

The trendlines of the two plots are offset, and the difference is about 0.2C, as shown above.

In conclusion:

My analysis confirms that Dr Marohasy is correct: the AWS system at Brisbane Aero reads higher than the LIG thermometer, and there was a distinct step up after the AWS was “fixed” in December 2019.

The Bureau of Meteorology cannot claim there was no difference following December 2019.  One of the two records is accurate, the other is not.  If the AWS system was performing within specifications after 23/12/2019, it was definitely not before then.

One Minute Data and Extremes Part 1: Thangool

In 2017 I purchased from the Bureau of Meteorology (BOM) a slab of one minute data from 16 country Queensland stations with Automatic Weather Stations (AWS).  One minute data is the temperature of the final second of every minute- 1,440 of them each day.  I posted a few times about this, and now I return to it to check on some recent claims by the BOM.

They repeatedly assert that the difference between AWS temperatures and those measured by mercury thermometers (LIG) is less than 0.1 degree Celsius.

The one minute data, infuriatingly, is NOT published by the BOM for more than 72 hours, and is NOT used for any daily temperature recording.  The AWS reads the temperature every second in each minute, but only the highest, lowest and final second temperatures are kept.  The highest of those highest one second values, from 9:00 a.m. to 8:59 a.m. next day, becomes the maximum (Tmax) of the day, and the lowest (also 9:00 a.m. to 8:59 p.m.) becomes the minimum (Tmin).  Tmax and Tmin are freely available, published at Climate Data Online (CDO).  One minute data is available at a cost, and at the time of my purchase did not include one minute high and low values.  Therefore, I can only compare daily data for final seconds of 1,440 minutes with the one highest and one lowest seconds, and can only estimate their time of recording.  Grrr!

A further source of frustration is that daily temperatures at CDO for many places have not passed Quality Assurance checks more than six years later- but that doesn’t stop them from calculating monthly means for them, claiming the monthly means are quality controlled.

Therefore in this series of analyses I only use daily data that is quality controlled.

Thangool is a very small town about 120km south-west of Gladstone and has the airport for Biloela.   Figure 1 shows the difference of the daily Tmin (one second value) minus the lowest one minute (final one second value) for February 2017.

Figure 1:  Daily Minimum Difference

Note that no daily minimum value is more than 0.1C below the lowest one minute value on any day in February.  No apparent issue there.

Figure 2:  Daily Maximum Difference

Clearly the difference is greater for maximum temperatures.  On 11 out of 28 days (39.3%) the difference between the maximum temperature and the highest temperature in the final second of any minute was greater than +0.1C.  The greatest difference was on 19 February when Tmax was +0.7C higher.  And that is at least, as I will show.

That is not comparing AWS readings with the old mercury LIG thermometers- we need parallel data for that, which the BOM is extremely reluctant to release.

However, we can draw some inferences.

Figure 3 is a plot of 1-minute temperature at Thangool Airport between 11:00 a.m. and 2:00 p.m.  on 19 February 2017 as measured by the AWS, the maximum recorded by the AWS, and an illustration of what an LIG thermometer might have recorded.  If we assume the AWS accurately simulates a mercury thermometer, I have shown how the mercury would have risen in steps: it would not have fallen after these steps until reset at 9:00 a.m. next day.  The maximum was reached after 1:00 p.m. and was recorded by the AWS as 35.7C.

Figure 3: One minute and Maximum Temperature at Thangool

Note I show the “theoretical” temperature a mercury thermometer might have recorded as following the peaks of the one minute values.  It may well have been higher than these steps, but below 35.7C- but we don’t know because those previous Tmax values were discarded.  It is most likely near one of the two spikes between 1:30 and 2:00 p.m.  In any case, Tmax of 35.7C is 0.7C above the highest one minute temperature of the day.  But the change is supposed to have been up  by 0.7C (at least) and back down again in one minute- it is not just one step up.

By the way, the BOM do quality checks on 1 second data, discarding any value that differs from those either side of it by more than 0.4C.  So the AWS could record a temperature increase of 4 degrees in 10 seconds without causing any alert.

Figure 4 shows the likely times when the AWS would have measured 35.7C.

Figure 4: One minute and Maximum Temperature at Thangool, 1:30 p.m. to 2:00 p.m.

Figure 5 shows temperatures from 1:30 p.m. to 1:35 p.m.- the time when the minute to minute temperature change is less..

Figure 5: One minute and Maximum Temperature at Thangool, 1:30 p.m. to 1:35 p.m.

Tmax was probably in either of the minutes indicated.  If it was at B (between 1:33 and 1:34) the difference was 0.7C.  If it was at A (between 1:32 and 1:33) the difference was 0.8C.  That’s why I say the real difference between highest 1 minute temperature and Tmax on any day is a minimum estimate. At any previous or later minute (such as the second peak at 1:52 p.m. in Figure 4) the difference would be much greater.  The important difference is between Tmax and the next highest 1 minute temperature: that is in this case the previous minute.

BOM apologists assert that the difference between LIG and AWS is negligible.  They also assert that each 1-second reading, because of the probe design, is really an average of the previous 40 to 80 seconds.

If that is true, then for the minute from 1:32:01 p.m. to 1:33:00 p.m. the running smoothed average of all the fluctuations between 1:31:01 and 1:33:00 rose from 34.9C to 35.7C then fell to 35C.  Therefore the real (unsmoothed) temperature must have fluctuated very rapidly to values much higher and much lower in that 120 second period. 

Further, could any human or animal detect such changes in less than one minute, and would it matter to anyone?  For example, would aircraft preparing for take-off need such precision?

That is why we say that AWS temperature data is over-precise and inaccurate.

However, only parallel observations will prove whether AWS simulates LIG to within +/- 0.1C.

The next post will look at Sunshine Coast Airport.

Extreme Weather Events 3: Sydney

Are extreme weather events showing up in Australia’s largest city?

Floods and bushfires might affect smaller areas, but droughts, heatwaves, and very heavy rainfall from large weather systems affect large areas. All of the above have occurred near Sydney in the past few years: surely there should be visible signs in temperature and rainfall.
First, rainfall.

In July and October 2022 flooding affected the western Sydney region again, with The Conversation of course saying “climate change is projected to bring far worse extreme rain events than in the past.”

For long term rainfall I look at Sydney’s longest rain records, at Observatory Hill and the Botanic Gardens. Figure 1 shows their location.

Figure 1: Central Sydney, courtesy of Google Maps

Observatory Hill rain records start in July 1858, but the original data ends in August 2020. I choose not to splice data from old and new gauges. Botanic Gardens start in 1885 but there is a large gap, with continuous data from late 1909 to the present. Figures 2 and 3 plot daily rainfall for each:

Figure 2: Observatory Hill daily rain

Figure 3: Botanic Gardens daily rain (1910 to 2022)

Long term means:

Figure 4: 10 year running means of rainfall at Observatory Hill and Botanic Gardens

Note that the means are similar until about 2010 when they start to diverge. Reasons might include changes to the sites. Rainfall was clearly higher in several previous decades.

Figure 5: 10 year running Standard Deviations

There was much greater variability in Sydney’s rainfall for most of the 50 years from 1950 to 2000. To show Standard Deviation relative to mean rainfall:

Figure 6: 10 year running Standard Deviations divided by 10 year means

Which shows there is little daily variability in rainfall in recent years, and both sites are comparable.

I will now analyse Botanic Gardens data in more detail.

Figure 7: Running 365 day means

2022 was the wettest year on record, followed by 1950.

Rainfall accumulated over several days is a factor in large scale riverine flooding such as occurred in Sydney’s west.

Figure 8: Four day total rainfall

Clearly there were many much greater 4 day rain events in the past than in the latest floods.

I measure “droughts” by counting the number of days with less than 4mm of rain in running 365 day periods.

Figure 9: Running 365 day counts of days with under 4mm of rain

2022 was by far the most consistently wet. There is no sign of increased drought in Sydney.

Conversely, do recent years have more days with high rainfall?

Figure 10: Running 365 day counts of days with over 100mm of rain

No. Only 3 days in 2022, while 1999 had 5, and many others in previous years had more than 2022. It seems that the Sydney region, going by the Botanic Gardens rain gauge, has less extreme rainfall than the past.

I now analyse temperature at Sydney Observatory Hill, using the latest version of Acorn to 2021, and Climate Data Online for 2022 and January 2023 up to Australia Day.

Figure 11: Daily Maxima Sydney Observatory Hill 1910 to 26/1/2023

Maximum temperatures in Sydney, according to the best the Bureau can provide, have warmed at 0.9 degrees Celsius per 100 years. Decadal means show an almost identical trend.

Figure 12: 10 year mean Tmax

Standard Deviation measures daily variability, and 10 year mean Standard Deviations show some interesting patterns:

Figure 13: 10 year running Standard Deviation, Sydney Tmax

Variability is greater with higher temperatures and less with lower temperatures, and temperatures should be related to rainfall- because a dry period will have hotter days and usually cooler nights. Temperature adjustments might interfere with this.

Whatever, there were several past periods with higher Standard Deviations than the past decade, and when divided by the 10 year means the contrast is even greater:

Figure 14: 10 year running Standard Deviations divided by 10 year means

Are days getting hotter? Well, years are, mostly:

Figure 15: 365 day running means of Tmax

Highest and lowest daily maxima in 365 day periods are not co-operating:

Figure 16: Highest Tmax in 365 day periods

The hottest day was back in 1939, and 2022 had the lowest “hottest day” in a 365 day period on record, with the hottest day being 31.9 degrees.

Figure 17: Lowest Tmax in 365 day periods

Several past winters had cooler maxima.

But is Sydney getting more frequent hot and very hot days?

Figure 18: Running 10 year counts of days over 34.9 degrees

Figure 19: Running 10 year counts of days over 39.9 degrees

The last 10 years have had fewer hot and very hot days than in the past.

What about heat waves, when there are strings of hot days? The definition appears to have changed, but if we consider three hot days in a row to be a heat wave:

Figure 20: Running 10 year counts of 3 consecutive days over 34.9 degrees

There is a very small trend (0.8 in 100 years) but there were many more 3 day heatwaves in the past.

Figure 21: Running 10 year counts of 3 consecutive days over 39.9 degrees

There is a decreasing trend of very hot heat waves (more than 3 less per 100 years), with nearly three times as many 3 day heatwaves of 40 degrees or more in the 10 years to 1982 as in the past 10 years.

Conclusion:

Contrary to popular belief encouraged by politicians and the media, in Australia’s largest city it is clear that:

Rainfall and temperature variability is LOWER than in the past

Droughts are NOT increasing

Extreme rainfall is NOT increasing

Dry years are NOT increasing

Very hot days are DECREASING in frequency

Heatwaves are NOT increasing and are very much LESS COMMON than 40 years ago.

If anything, Sydney’s weather is becoming less extreme and more benign. That should be good news.

We are still waiting for the “projections” of more extreme weather to arrive.

Extreme Weather Events: 2

Further to my post yesterday about the Climate Council’s recent fear mongering, with my look at whether the recent flooding at Fitzroy Crossing could be due to increasingly severe rain events, here are two more locations.

I calculate the 10 year running standard deviation of daily rainfall, the 10 year mean, and because the standard deviation must change as the mean changes, I divide the 10 year standard deviation by the 10 year mean.

Early this year there was sever flooding in northern New South Wales. Brays Creek is near Mt Warning about 40 km north of Lismore. Here is the standard deviation divided by average rainfall:

Rainfall over the past 10 years is less extreme than it was 40 to 50 years ago.

The Bruce Highway to north Queensland was blocked for several days, as it normally is every Wet season, by flooding at Goorganga Plains just south of Proserpine. Is rainfall becoming more extreme? Here is the raingauge at Lethebrook, using the same technique.

Nothing exciting to see there either.

Extreme Weather Events: 1

Last night On Wednesday night 18 January, the Climate Council released their latest doomsday publication, with the support of Beyond Blue (they’re now off my list of charities to donate to.)

“HIDDEN MENTAL HEALTH TOLL OF WORSENING CLIMATE DISASTERS ON AUSTRALIANS REVEALED WITH NEW NATIONAL POLL”

Climate Councillor, climate scientist at the Australian National University and author of Humanity’s Moment: a Climate Scientist’s Case for Hope, Dr Joelle Gergis said: “The results of this poll are confronting. It’s heartbreaking to realise that many Australians are living with significant levels of distress related to the reality of our changing climate. It shines a light on this invisible mental health crisis that is undermining the stability of our local communities all over the country.

“We need to have a national conversation about climate change adaptation and listen to the experiences of people who have lived through these disasters.

“Extreme weather events are going to escalate as our planet continues to warm, so the impacts we have witnessed in recent years are really just the tip of the iceberg. We urgently need to develop plans that protect and support our local communities as climate change-fuelled disasters continue to upend the lives of countless Australians.”

Time for a reality check:

Is there evidence of increasing climate extremes?  Rainfall and temperature are easily measured and data is freely available from the BOM.

First example:  The recent flooding at Fitzroy Crossing. 

A useful measure of extremes is Standard Deviation.  For this technique I am indebted to Willis Eschenbach whose recent post at WattsUpWithThat sparked my interest.

I calculate the 10 year running standard deviation of daily rainfall, the 10 year mean, and because the standard deviation must change as the mean changes, I divide the 10 year standard deviation by the 10 year mean.

The nearest rain gauge with a reasonably long record is Fossil Downs.  Here is the 10 year average daily rainfall:

As you can see average daily rainfall (which nearly all falls in the Wet) has nearly doubled since the decades to the 1960s.

10 year standard deviation:

No wonder people are anxious!  The 10 year figure is very high (but not as high as the 1980s!  Was it more extreme 40 to 50 years ago?)

But here is the standard deviation divided by average rainfall:

This shows that relative to the average, rainfall extremes are actually getting smaller.

Over the next few days I will show rainfall and temperature plots for several Australian cities.  Stay tuned.

Australian Temperature- Satellites or Surface Stations?

For years we have been very sceptical about the official Bureau of Meteorology (BOM) temperature record which is based on 104 surface stations in the ACORN-SAT (Acorn) network.  In this post I look at one of the main reasons for doubting the veracity of the surface record- the increasing divergence from the satellite record.

First up I should say that the two records should not necessarily agree, because they measure two completely different things.  Surface stations measure the temperature of the air 1.2 metres above the ground and report the highest and lowest one second samples each day at 104 locations.  These are combined in a grid average to give monthly, seasonal, and annual temperatures.  Satellites measure temperatures of the atmosphere from the ground to many kilometres up, every second, over a wide area for each pass.  These are similarly combined by algorithms to calculate a monthly average for (in this case) the land area of Australia’s Temperature of the Lower Troposphere (TLT). 

They are both useful for different purposes but are not easily compared.  Because minimum surface temperatures poorly match TLT, mean surface temperature is also a poor match.  Maxima are a better match, but still not perfect.

For this post I use data from the University of Alabama (Huntsville) (UAH) which calculates anomalies from 1991 to 2020 means.  I have converted Acorn data from anomalies from 1961-1990 means, to anomalies from 1991-2020 means, to match.

Figure 1 shows monthly Acorn maxima data and UAH means from December 1978.

Figure 1: Monthly Surface Tmax and UAH data

Although surface maxima have a much larger range than TLT anomalies, they plainly follow similar trajectories.  12 month running means smooth the data and allow easier visual comparison.

Figure 2: Running 12 Month Means: Surface Tmax and UAH data

Similar, but different at several times.   Annual means show that in some years Tmax and TLT are close to identical, while in other years they have large differences.

Figure 3: Annual Means: Surface Tmax and UAH data

In 2015 I showed the reason for these differences (but not the difference in trends).  The differences between the two datasets are very largely due to variations in rainfall.  In wet years surface maxima are relatively much cooler than TLT; in dry years surface maxima are much warmer.  In Figure 4 I have calculated rainfall anomalies scaled down by a factor of 20 and inverted, to compare with the difference between Tmax and TLT.

Figure 4: Running 12 month Means: Surface Tmax minus UAH and Inverted Rainfall

The match is close.  Figure 5 shows annual values, and trend lines.

Figure 5: Annual Means: Surface Tmax minus UAH and Inverted Rainfall

While annual rain has been slightly increasing (it’s inverted, remember) the relative difference between surface temperature and atmospheric temperature has been increasing at a rate of one degree per hundred years.  That’s odd.  Figure 6 shows the relationship between the temperature difference and rainfall.

Figure 6: Annual Surface Tmax minus UAH versus Scaled Rainfall

For every extra 20mm of rainfall, the difference between surface maxima and TLT decreases by 0.85 degrees Celsius.  The trend lines in Figure 5 should be close to parallel, not diverging.

As well, as rainfall increases, Tmax should decrease, as Figure 7 shows.

Figure 7: Surface Tmax as a Product of Rain

But as we saw in Figure 3, Tmax is increasing faster than UAH.

Furthermore, as surface Tmax increases, TLT should be increasing as well, which it is, but at a slower rate.

Figure 8:  Atmospheric Temperature as a Product of Surface Tmax

Is the atmospheric temperature lagging behind surface temperature?  Figure 9 shows the last two years of monthly values.

Figure 9:  Monthly Atmospheric Temperature and Surface Tmax, January 2020-March 2022

The values are mostly synchronous, with sometimes a delay in one or the other of one month.  (Remember, we are comparing data from 104 stations scattered across the continent, with that of the atmosphere with constantly changing and circulating winds).  When the land warms, the atmosphere warms with it; when the land cools, so does the atmosphere.

Conclusion:

Tmax should not be increasing faster than atmospheric temperature.  There is no real delay in any temperature change, as the atmosphere is heated each day by the land.  Therefore it appears that there must be some fault with the maximum temperatures reported by ACORN-SAT, which appears to be warming too rapidly.

Explanation of the mechanism for rainfall moderation of surface-atmospheric temperature differences:

In wet years more moisture carried upwards condenses, releasing heat, thus warming the atmosphere, while the surface is cooled by cloud cover, evaporation, and transpiration.  In dry years much less moisture is convected, so less heat is released in the atmosphere, while the surface is hotter because of less cloud cover and less evaporation and transpiration.  Thus dry years have a greater relative difference between Tmax and TLT than wet years.

The only energy source is solar radiation heating the land surface in daylight hours, which in turn heats the atmosphere by conduction and convection.  At night as radiation to space rapidly cools the earth, convection also rapidly decreases, so maxima, not minima, are responsible for the relationship with TLT. 

A complication is that in summer (and more so in very wet La Nina years) large volumes of very moist air from the tropical seas to the north converge over northern Australia and penetrate even into southern Australia.  This warm moist air cannot heat the surface but through condensation transfers heat to the upper atmosphere- therefore the difference between surface and atmosphere is even smaller.

More Problems With Australia’s Temperature Record: Part 3

We have seen in Parts 1 and 2 that every extra year of annual data can decrease the temperature trend at a weather station by from -0.02 to -0.03℃ per decade, and that less than half (47% actually) of Australia’s weather stations used for climate analysis have data from 1910, and three of them have insufficient data to calculate trends.

Figure 1 shows a map of non-urban Acorn stations with enough data to calculate trends, at 1910.  The others I have blanked out.

Figure 1: Acorn stations with data for 1910

The network is very sparse.  To estimate a national temperature for 1910 enormous weighting must be given to the values of a few remote stations like Alice Springs, Boulia and Kalgoorlie, so we hope they got the adjustments right!  Unfortunately, in 2015 I found adjustments at Kalgoorlie and Alice Springs were very problemmatic.

The Bureau explains the process of calculating average temperatures here.

Figure 2 shows the BOM map of trends from 1910 to 2020:

Figure 2:  Australian Tmean trends 1910-2020

Note that there a few “bullseyes” which surround stations whose temperature trends are out of phase with areas around them- e.g. Boulia is warmer, Marble Bar is cooler. 

Now here is a paradox.  As the years go by and more stations have data available, the area weighting for each station will decrease, however trends at the newer stations will show increased warming compared with the older ones.  However they will also have more variability.  This will result in oddities as I shall show, and reveals something of the difficulties with the BOM methods.

 Figure 2 is a plot of mean temperature from 1970 to 2020.

Figure 2:  Australian Tmean 1970-2020

The Acorn 2 trend is now +0.23℃ per decade or +2.3℃ per 100 years- a full degree more than the trend from 1910.

Now let’s look at the trend map for 1970 to2020:

Figure 3:  Australian Tmean trends 1970-2020

Note the little “bullseye” around Victoria River Downs, the little “balloon” around Halls Creek to the south-west of VRD, and the little surge to the south-southwest of VRD of 0.05 to 0.1℃ per decade.  Note also that north-eastern Arnhem Land, with no stations, has a warmer pocket.  Figure 4 is the BOM data for VRD.

Figure 4: Annual mean temperature at Victoria River Downs

VRD opened in 1965 and has too much data missing for BOM to calculate a trend.  The area weighting algorithm still gives it a cooling trend of between minus 0.05 and 0℃ per decade (Figure 3).  Que?

With more than 27% of data missing I wouldn’t calculate a trend either, but with only six of 43 years missing I can calculate a trend from 1978:

Figure 5: Annual mean temperature at Victoria River Downs

The trend is -0.09℃ per decade, which is a bit more cooling than the trend map (Figure 3) shows.  Now let’s look at trends from 1980 to 2020.

Figure 6:  Australian Tmean trends 1980-2020

There are more bullseyes, and I have shown temperature trends for some- Carnarvon, Meekatharra, Forrest, Thargomindah, and Gayndah.  But remember Figure 3’s little surge to the SSW?   It now has its own bullseye, and that is Rabbit Flat.

Figure 7: Annual mean temperature at Rabbit Flat

Rabbit Flat opened in 1970 and has a trend of +0.08℃ per decade, which agrees with the trend map in Figure 3.  Now from 1980:

Figure 8: Annual mean temperature at Rabbit Flat

What a difference a few years make in a short timeseries.  The trend of -0.06℃ per decade also agrees with the 1980-2020 trend map.

However, just 328km away Halls Creek shows a warming trend of +0.17C per decade from 1980 – 2020:

Figure 9: Annual mean temperature at Halls Creek 1980-2020

But from 1970 to 2020 Halls Ck is warmer still at +0.19C per decade:

Figure 10: Annual mean temperature at Halls Creek 1970-2020

And at Tennant Creek 441km away the 1970-2020 trend is +0.19C per decade:

Figure 11: Annual mean temperature at Tennant Creek 1970-2020

From 1980 it is +0.06C per decade.

Figure 12: Annual mean temperature at Tennant Creek 1980-2020

Temperatures are trending in different directions and wildly different rates at the closest stations: they can’t all be right!

The method of drawing trend maps is to use anomalies of temperatures of all years of all stations whether or not an individual trend can be calculated, then calculate a gridded average, and from that calculate trends, then spread those trends hundreds of kilometres in every direction- even across the Gulf of Carpentaria from Horn Island to Arnhem Land, as seen in Figures 3 and 6- averaged with the trends propagated by other stations.  If a site has data missing, the grid is infilled with the weighted data from other sites.  

In recent decades this causes great variability because of the short records, which leads to grave doubts about the reliability of some records.  Further back in time, there is less variability because there are more stations, and the longer records smooth and decrease the trends- however the weighting has to be much greater because of the large areas with no data at all for many years. 

The problem is: we can have either a long record, or an accurate record, but not both.

This leads to the obvious conclusion:

The official temperature record since 1910 is just a guesstimate.

More Problems With Australia’s Temperature Record: Part 2

My colleague Chris Gillham at WAClimate uses 58 long term weather stations for his analyses.

And with good reason.  Here’s why.

Figure 1 is a screenshot of the annual mean temperature record at a typical Acorn station, Longreach (Qld) with the linear trend shown.

Figure 1: Annual mean temperature at Longreach

The linear trend is +0.12℃ per decade.  Nine (9) of the 111 years of data from 1910 to 2020 are missing, leaving 102 years.

Australia’s official climate record is based on 112 sites like Longreach.  Of those, 8 are not used for seasonal and annual analyses because they are affected by Urban Heat Island (UHI) effect.  Five (5) of the non-urban stations have more than 20% of their data missing, so the BOM does not calculate trends for them. Of those remaining, only 50 started in 1910, and another 8 before 1915.  What is the effect of different length records on our understanding of how temperatures have changed over the years?

Figure 2 is a plot of the trends of mean temperatures per decade as a factor of the number of years of annual temperature data on record at those 107 Acorn stations with enough data to calculate trends.

Figure 2:  Trend as a factor of amount of data

Stations with  longer data records have lower trends.  The trends at stations with shorter records vary wildly, with some obvious outliers. 

At those stations with UHI effect, the relationship is even stronger.

Figure 3:  Trend as a factor of amount of data at sites with UHI

These sites are in larger towns and cities, possibly with better maintenance and observation practices (although not necessarily better siting).

The slope of the trendlines in the above two figures show that for every additional year of data, temperature trend decreases by about -0.02 to -0.03℃ per decade. In 100 years that could make a difference of as much as three degrees Celsius 0.3C at a well maintained site.

Figure 4 is a map of trends across Australia from 1910 to 2020.  I have shown the years of available data at each site (locations only approximate) and I have circled in blue those 5 sites that have insufficient data.

 Figure 4:  Years of data contributing to 1910 to 2020 trend map

Trends in different regions vary from less than 0.1C per decade to up to 0.3C per decade.  As you can see there is a large variation in the amount of available data in each different coloured band.  That’s for 1910 to 2020.  Note that there are only three (3) non-urban stations with no missing years- Carnarvon, Esperance, and Mt Gambier- which I have circled in red.  There are some big gaps.

In Part 3 I will look at some individual stations and how trends vary in the 51 years from 1970 to 2020.

More Problems With Australia’s Temperature Record: Part 1

Since 2010 I have been documenting problems with different versions of Australia’s official temperature record as produced by the Bureau of Meteorology (BOM).  Since the High Quality (HQ) dataset was quietly withdrawn in 2012 we have seen regularly updated versions of the Australian Climate Observation Reference Network- Surface Air Temperature (ACORN-SAT or Acorn).  We are now up to Version 2.2.  In this Part I shall show the effect of these changes on temperature trends.  In Part 2 I will show how record length affects trends, and in Part 3 I will look at the record since 1970 at some individual stations.

Figure 1 is from the BOM Climate Change Time Series page.

Figure 1:  Australian Official Temperature Record 1910 to 2021

The linear trend is shown as +0.13℃ per decade, or 1.3C per 100 years.  My colleague Chris Gillham of WAClimate has provided me with archived Acorn 1 annual mean temperature data to 2013 which allows this comparison:

Figure 2:  TMean: Acorn 1 and Acorn 2

The result of introducing Acorn 2 has been a much steeper trend:  Acorn 1 trend to 2013 was 0.9℃ per decade.  The trend has now become 0.13℃ per decade. (The extra 9 years have added an extra 0.017C per decade to the trend.)

Figure 3 shows when and how large the changes were:

Figure 3:  Difference between Acorn 1 and Acorn 2

Acorn2 is cooler than Acorn 1 before 1971 and warmer in all but three years since.  Since these were based on the same raw temperatures (with some small additions of digitised data and a couple of changes to stations) the changes were brought about entirely by adjustments to the data.

I calculated running trends from every year to 2013 for both datasets.  As trends shorter than 30 years become less reliable I truncated the running trends at 1984.  Figure 4 compares thre trends to 2013 of Acorn 1 and Acorn 2.

Figure 4:  Acorn 1 and Acorn 2 running trends per decade to 2013

The weather fluctuations of the mid-1970s to 1980s played havoc with trends.

Figure 5 shows the difference between the trends.

Figure 5:  Difference between Acorn 1 and Acorn 2 Trends

The difference ranges from +0.024C per decade for 1910 to 2013, to +0.039C for 1950 to 2013.  Having increased warming by from 0.25C to 0.4C per 100 years (just by making different adjustments) Acorn 2’s trend is much more alarming than Acorn 1’s.

Conclusion:

This is from the BOM’s explanation for Acorn:  

“A panel of world-leading experts convened in Melbourne in 2011 to review the methods used in developing the dataset. It ranked the Bureau’s procedures and data analysis as amongst the best in the world. ‘The Panel is convinced that, as the world’s first national-scale homogenised dataset of daily temperatures, the ACORNSAT dataset will be of great national and international value. We encourage the Bureau to consider the dataset an important long-term national asset.’” ACORN-SAT International Peer Review Panel Report, 2011.

 Acorn 1.0 was apparently such an important long-term asset that it was quickly superseded by Acorn 2 with a much more alarming trend.

Diurnal Temperature Range and the Australian Temperature Record: More Evidence

In an earlier post, I demonstrated through analysing Diurnal Temperature Range (DTR) that the Bureau of Meteorology is either incompetent or has knowingly allowed inaccurate data to garble the record.

A couple of readers suggested avenues for deeper analysis. 

Siliggy asked, “Is the exaggerated difference now caused by the deletion of old hot maximums and or whole old long warmer records?”

Graeme No. 3 asked, “Is there any way of extracting seasonal figures from this composition?”

This post seeks to answer both, and the short answer is “Yes”.

Using BOM Time Series data (from the thoroughly adjusted Acorn dataset) I have looked at data for Spring, Summer, Autumn, and Winter (although those seasons lose their meaning the further north you go).

DTR is very much governed by rainfall differences as shown by this plot.

Figure 1:  Winter DTR anomalies plotted against rainfall anomalies- all years 1910-2020

This shows that in winter DTR decreases with increasing rainfall.  The R squared value of 0.79 means that for the whole period, rainfall explained DTR 79% of the time on average.  However, the average conceals the long term changes in the relationship.

To show this, I simply calculated running 10 year correlations between DTR and Rainfall anomalies for each season, and squared these to show the “R squared” value.  This is a good rule of thumb indicator for how well DTR matches rainfall over 10 year periods.  A value of 0.5 indicates only half of the DTR for that decade can be explained by rainfall alone.  As you will see in the following figures, there are plenty of 10 year periods when the relationship was 0.9 or better, meaning it is ideally possible for 90% of DTR variation to be explained by rainfall.  Here are the results.

Figure 2:  Spring Running R-squared values: DTR vs Rain

There was a good relationship before 1930.  In the decades from then to the mid-1970s it was much worse, and very poor in the decade to 1946. It was poor again in the decade to 2001, and the 10 years to 2020 shows another smaller dip, showing something not quite right with 2020.

Figure 3: Summer Running R-squared values: DTR vs Rain

Summer values were very poor before the 1960s, especially the decades to 1944 and 1961, and dipped again in the 1990s.

Figure 4:  Autumn Running R-squared values: DTR vs Rain

The DTR/Rain relationship was very poor in the decades to 1928, and again before 2001.  The recent decade has also been poor- less than half of DTR to 2020 can be explained by rainfall.

Figure 5:  Winter Running R-squared values: DTR vs Rain

The DTR/rainfall relationship was fairly good, apart from two short episodes, until the 1990s.

I now turn to the northern half of the continent.

A large area of Northern Australia is dominated by just two seasons, wet and dry.  Here is the plot of northern DTR vs Rain for the wet season (October to April).

Figure 6:  Northern Australia Wet Season Running R-squared values: DTR vs Rain

Apart from the 1950s, the late 1970s-early 1980s, and 1998 to 2020, the DTR : Rainfall relationship is very poor, with a long period in the 1930s and 1940s in which rainfall explains less than half of DTR variation (only 13% in the decade to 1943). 

Because the northern half of Australia accounts for the bulk of Australian rainfall, and the wet season is from October to April, this perhaps explains the problems in spring, summer, and autumn for the whole country.

We can get some clues as to the reasons by comparing long term average maximum temperatures with inverted rain (as wet years are cool and dry years are warm).

Figure 7:  Northern Australia Wet Season Decadal Maxima and Rain

The divergence before 1972 and after 2001 is obvious.

The above plots show how poorly DTR (and therefore temperature, from which it is derived) has matched rainfall over the past 111 years.  Low correlations indicate something other than rainfall was influencing temperatures.

In reply to Siliggy, who asked “Is the exaggerated difference now caused by the deletion of old hot maximums and or whole old long warmer records?” the answer appears to be: both, however Figure 7 shows old temperatures (before 1972) appear incorrect, but recent temperatures are at fault too.

The mismatch shows that the Acorn temperature record is not to be trusted as an indicator of past temperatures- and even recent ones.