Jennifer Marohasy,
My good friend Patrick Lloyd, enamoured with the new Artificial Intelligence ChatGPT, sent me maximum temperatures for Tasmania as individual annual averages of all the daily values for all 155 weather stations since 1882. He had downloaded these values from the Australian Bureau of Meteorology’s website with the help of ChatGPT. He wanted to know from me if they showed global warming.
After importing the numbers into my statistical program, I drew a scatter plot with a fitted regression line, Chart 1. It suggests that maximum temperatures in Tasmania are not as high now as they were back in the early 20th century, and that there was a period of cooling to about 1950.
The regression equation indicates that overall, there has been cooling at a rate of 0.0051 degrees Celsius per year. Converting this into climate change speak we get a rate of cooling of 0.5 C per 100 years. This doesn’t sound right, does it?
The official historical reconstruction for Tasmania also shows temperatures cooling to about 1960, and then temperatures increasing such that since at least the year 2000, it has been very hot, Chart 2.
Clearly the official maximum temperature reconstruction for Tasmania is not just an amalgamation of all the maximum temperature series. Charts 1 and 2 show very different overall trends, though both show cooling to about 1960.
Is the Bureau’s historical reconstruction (Chart 2), a more accurate representation of historical temperatures for Tasmania than the annual mean of all the daily values that ChatGPT sourced for Patrick (Chart 1)?
Every temperature series has its limitations, and sometimes issues are just compounded when many temperature series, beginning at different times and from different locations and altitudes, are just combined. That is what Chart 1 represents, a simple compilation of everything available. It needs checking.
Patrick has suggested a solution might be to just combine the longest and most complete maximum temperature series, perhaps the best 10 percent of the 155 weather stations, and throw the rest out.
But there is an additional potential problem.
Since 1 November 1996, the Bureau have been transitioning from mercury thermometers to probes in automatic weather stations for the measurement of this temperature data. Probes in automatic weather stations represents a fundamentally different way of measuring temperatures than the traditional mercury thermometer where the daily maximum used to be manually read at 9am each morning and then the mercury reset.
Could it be that the automatic weather stations are recording hotter for the same weather?
Considering just Larapuna (also known as Eddystone Point, station number 092045), as an example, this weather station has a long and relatively compete temperature record. Larapuna would represent one of the ‘best 10 percent’ from Tasmania.
Chart 3 shows the annual average maximum temperatures (unhomogenised) as measured at Larapuna first using a mercury thermometer, and since 1997 with a probe in an automatic weather station.
Considering maximum temperatures for this location there does appear to be a jump-up of more than 1 degrees Celsius after about 1997.
How can we know if this apparent step-change circa 1997 was caused by a change in the weather, or a change in the equipment used to measure the temperatures at Larapuna?
I’ve said to Patrick, that before we begin combining the individual temperature series I would like to sort out the issue of the change from mercury to probe. In fact, this issue has been preventing me continuing with a reconstruction of Australia’s temperature history for some years now.
The probes are generally considered to be more responsive to fluctuations in temperatures and, therefore, likely to record both hotter and colder for the same weather. This is stated in the Bureau’s Research Report No. 032 The Australian Climate Observations Reference Network – Surface Air Temperature (ACORN-SAT) Version 2 (October 2018) by Blair Trewin:
In the absence of any other influences, an instrument with a faster response time will tend to record higher maximum and lower minimum temperatures than an instrument with a slower response time. This is most clearly manifested as an increase in the mean diurnal range. At most locations (particularly in arid regions), it will also result in a slight increase in mean temperatures, as short-term fluctuations of temperature are generally larger during the day than overnight.” (Page 21)
Perhaps because Tasmania is neither particularly hot, nor particularly arid, we don’t need to worry about the change over? I wonder what ChatGPT thinks? I wonder if ChatGPT can solve the equipment change over problem?
How much hotter and how much colder the probes record relative to an old style mercury thermometer will depend on the time constant intrinsic to that instrument and also on how it has been calibrated. Probes in automatic weather stations can be calibrated so the shift is uniformly in one direction.
If only we had parallel data for Larapuna, that is measurements taken from mercury thermometers and also probes in the same Stevenson screen, we could know the extent of the equivalence between the temperatures as measured with the different types of equipment. So far, the Bureau has been keeping the parallel data under wraps; the Bureau has a policy of not making the parallel data publicly available.
What are some of the other possible issues we might encounter as we combine the temperature series in search of the most accurate historical temperature reconstruction for Tasmania?
****
UPDATE
ChatGPT, via Patrick, just gave me a scatterplot for Larapuna, through until yesterday.
*****
This is the first of a new series of blog post about reconstructing Tasmania’s temperature history with the help of Patrick and ChatGPT.
If you would like to follow along as the three of us attempt to understand how and why the two temperatures series (Chart 1 versus 2) are so different and attempt our own historical temperature reconstruction for Tasmania then, consider subscribing at my website, specifically click the box ‘Temperature Trends’ and you will start receiving information on this topic.
The photograph is of Patrick and me in front of the weather station in Tewantin, Queensland.
Suggestions for investigation.
Look at the record of daily lows, and of daily mean temperatures for the period(s) and station(s) in question. If the trend lines for those plots are consistent with those of the daily highs, at least you know that if there is a problem, it is only in the difference in instrumentation and recording practice.Obtain and review engineering test data for the different instruments.Find someone still using an obsolete instrument and compare their data to the nearest available modern station. Alternately, obtain the loan of an example of each sort of instrumentation, then compare records over a 30 to 90 day period. Extend the test period if there are inconsistencies.Compare the existing records to UAH satellite data for daily low, mean, and high. Montly data for AUS from 1979 is at https://www.nsstc.uah.edu/data/msu/v6.0/tlt/uahncdc_lt_6.0.txt and should at least give a ‘sanity check’ over the Tasmanian data,but using different instrumentation. Perhaps Dr. Spencer can suggest a source at UAH for more localized data.Obtain and evaluate urbanization and land use records for a sample of the station/locations.
“…the Bureau has a policy of not making the parallel data publicly available.”
______________________________________________________________
Why am I not suppose to think, “What are they trying to hide?”
“… the Bureau has a policy of not making the parallel data publicly available“.
FOIA request needed (Freedom of Information Act).
“Probes in automatic weather stations represents a fundamentally different way of measuring temperatures than the traditional mercury thermometer where the daily maximum used to be manually read at 9am each morning and then the mercury reset.”
The above quote from the article seems a bit confusing. Maximum temperatures usually occur after noon, around 1-2 pm. What have I misunderstood?
“This doesn’t sound right, does it?”
No, it doesn’t. That’s because it isn’t done right.
For an exercise like this it is essential to use anomalies. The reason is that you have a varying set of stations reporting in each year. If some mountain stations were included in the later years then you will get cooling. No individual places cooled, but the set of stations changes from warmer places to cooler.
The other thing you need to do is use area weighting. Otherwise, if it happened, for example, that you had a lot of stations around Hobart, then the result is overly influenced by those stations.
“If only we had parallel data for Larapuna, that is measurements taken from mercury thermometers and also probes in the same Stevenson screen, we could know the extent of the equivalence between the temperatures as measured with the different types of equipment. So far, the Bureau has been keeping the parallel data under wraps; the Bureau has a policy of not making the parallel data publicly available.”
You could try some natural intelligence – ie looking. The BoM does make the parallel data available. It posts a station Eddystone Point Comparison, which runs from Oct 1996 to Oct 1998. They have daily and monthly data. The monthly averages are here.
Over ~30 years ago the famous Dr Phil Jones of Uni of Norwich found Hobart data from 1841. I just had a quick look on BoM,s CDO page but could not see it.
I know this is only anecdotal, but as a Tasmanian latchkey kid of the 70’s .. I can declare that those summers were hotter and winters were colder.
For those interested, my best go yet at an area reconstruction, and using area weighting and anomalies, was a few years ago. Link to blog post here:
https://jennifermarohasy.com/2016/12/temperatures-trends-southeast-australia-1887-part/
And the book chapter was published with a DOI number at bottom of that blog post.
Cheers,
Jennifer & Patrick,
Over the decade starting in 1980, Eddystone Point had these trends (expressed as ⁰C per decade.)
Trend.
Decade 1980 first line, then Decade 2000 second line.
1980 Trend Tmax 0.0075 Trend Tmin -0.0008
2000 Trend Tmax -0.0046 Trend Tmin -0.049
All of these trends over the decade were negative or very small positive. Negative is cooling.
The 1980 decade was measured with Liquid in glass thermometers. The 2000 decade used Pt resistance thermometry.
Mean and Standard Deviation.
Decade 1980 line first, then Decade 2000 second line
1980 data Tmax Mean 16.9 SD 3.35. Tmin Mean 10.65 SD 3.44
2000 data T max Mean 18.0 SD 3.64 Tmin mean 10.62 SD 3.45
Mainly, the later decade saw a big Tmax movement to being 1.1⁰C hotter and more variable.
This fits with higher counts of extreme high values observed in the latter decade of Tmax.
This is all consistent with a change attributable to the Pt thermometry that entered service between those 2 decades, said to be starting on 1st November 1996. This effect could be artificial and superimposed on two decades which otherwise saw cooling or next to no warming.
Right now, I am analysing the decade of instrument change, starting in year 1990.
Geoff S
I am VERY surprised that they did not parallel run for ten years, to get a decent comparison between traditional and automatic thermometers.
It is almost as if they welcomed the artificial warming.
R
“Probes in automatic weather stations represents a fundamentally different way of measuring temperatures than the traditional mercury thermometer where the daily maximum used to be manually read at 9am each morning and then the mercury reset.”
Why would the daily maximum be read at 9 AM? Isn’t the daily maximum more likely to occur in the afternoon, such as 2 PM in winter or 4-5 PM in summer?
From my experience, you will find that chart 1 is mostly correct. However, you can clearly see that you will get a higher correlation if you do a non-linear plot, like a bi-nominal or sinewave. 1952 is the deep end, on average. I found that the speed of warming in K/year (use at least 4 stretches of time, as per the relevant Hale-Nicholson solar cycles) when set out against time, iow the acceleration or deceleration of warming, is on a quasi sine wave with wavelength 86.5 years. This is the so-called Gleissberg cycle.
If you want to go back in time, that far, Tmax is probably your best bet, as they read (past tense) the temperature for the day and then re-set the thermometer. Provided there was someone to read every day, it is probably the most reliable reading that gives an indication of the strength of the sun.
It is true that there is a reaction of glass with mercury and depending on the quality of the glass, this did affect the daily readings. It was only in the 70’s that they started with re-calibrations but not long after that they introduced the thermo-couples.
You can eliminate most of the problem due to the error with the mercury thermometers by looking at the speed of warming. See my comment earlier up the thread.
https://breadonthewater.co.za/2021/11/25/an-inconvenient-truth/
It was so good to receive an email just now from John Christy with the satellite data for Tasmania.
He writes: “I read your piece on Tasmanian temperatures and wanted to see what the atmosphere above Tasmania has been doing. Attached are the results from our Lower Tropospheric dataset. It looks like the peaks in 1998 and 2017 are indistinguishable. The upward shift in temperature from 1996 to 1997 looks real, but not much has happened since then. The overall trend is +0.15 C/decade with the trend from 1997 to present being +0.02 C/decade. The upper air is representative of the broader circulation in the region, so may not be able to represent the surface temps well. However, since it is a bulk atmospheric measurement, it will have a better capability to capture the influence of extra greenhouse gas forcing.
From 1861 a clue to what document to search for in the Tasmanian state records.
Some extensive details from it are quoted.
EG:”It may bo observed, by Tables 9 and 11,
when compared with Tables 10 and 12, that the
City Register gives a higher temperature than
the register made at the Royal Observatory ;
and the latter, from Table 2-1, when compared
with Tablo 25, gives a greater degree of hu-
midity than the former ; but this circumstance
alone will not absolutely prove what has been
currently reported, that the temperature of the
Colony is increasing; for if all the instruments
aro supposed to assimilate, the situations of the
two Observatories are very different. Ross
Bank Observatory was some distance from the
City, and situated on the banks of the Derwent ;
while tho private Observatory is in, tho centre
of tho town, with no great amount of water
near it.
https://trove.nla.gov.au/newspaper/article/8800163