The BEST and worst of ACORN-SATv2 Tmins

Guest article by Dr Michael Chase

clip_image002

Map above: Changes in minimum temperatures (Tmin) in Australia since 1910, according to the Australian Bureau of Meteorology (BoM)

“It was the BEST of tmins, it was the worst of tmins” … apologies to Charles Dickens.

SCOPE

This article is about why the BoM map shown above has warm and cold spots, including a somewhat implausible cooling region near Halls Creek in Western Australia. A four-pronged validation analysis is being mounted against version 2 of ACORN-SAT, intended by the BoM to indicate how air temperatures in Australia have varied from 1910 to the present time. The four prongs are as follows:

· Consistency with near neighbours

· Absence of inhomogeneities

· Adjustments that match non-climatic changes in the raw data

· An adjustment procedure that is not prone to errors

ACORN-SATv2 fails all these checks, and the highly non-uniform map above is one of the results, the hot and cold regions reflecting errors.

The adjustment procedure used by the BoM is to detect and correct non-climatic influences on the measured temperatures, such as from station moves and equipment changes. Errors arise in this procedure from erroneous size estimation and missed detections of genuine non-climatic shifts, false detections, and from the time-varying nature of some non-climatic influences. Going backwards in time from the present the errors accumulate as in a random walk, but the walk is not entirely random, there is a bias towards excessive cooling of early data. The reason for this bias may be that sudden cooling is much more prevalent than sudden warming, but whatever the reason, the bias is definitely a thing.

FIVE OF MANY EXAMPLES

The following figure shows [ACORN – BEST] Tmin data, as 12-month moving averages, for the five towns shown on the map above. BEST stands for Berkeley Earth Surface Temperatures, used here as “reference series”. The BEST data locations used for each town are given in the appendix.

clip_image004

The figure above illustrates the accumulation of ACORN-SATv2 errors, leading to excessive cooling of early data by around 1C, except for Halls Creek, which has excessive warming of early data, explaining the cold spot on the map above.

Adjustment Error Examples

At this point some readers will question the validity of BEST data. The following figure deals with that question for the first two example towns, Rutherglen and Wagga Wagga, both near the ACORN-SAT town of Deniliquin:

clip_image006

In the figure above the data shown are as follows:

· BLUE = RAW – ACORN-SATv2. This shows the adjustments that have been made to raw data.

· RED = RAW – BEST (Albury). This shows the variations of the non-climatic influences on the raw data, such as station moves, equipment changes, and observer errors.

If the ACORN-SATv2 adjustments were correct the blue data would match the moving average of the red data, which it does with great success for the example of Deniliquin Tmin data. In effect ACORN-SATv2 Deniliquin Tmin, and BEST (Albury) Tmin, have jointly validated each other, but things went wrong for ACORN-SATv2 for Rutherglen and Wagga Wagga, as illustrated in the following two analysis figures:

clip_image008

clip_image010

The figures above show examples of ACORN-SATv2 making invalid adjustments, and failing to make adjustments that were needed. The analysis plot for Halls Creek is as follows, an example of incorrect step-size estimation:

clip_image012

The full consequences of the errors in ACORN-SATv2 are not yet known, but they include exaggerated 20th century warming in many locations, and probably the generation of fake temperature records.

APPENDIX

Technical notes are given below, further details and examples can be found at: https://diymetanalysis.wordpress.com/

BEST data is used in the validation tests as a “reference series”. A reference series has to have a good approximation to the regional average weather fluctuations, so that its subtraction increases the signal (steps/trends) to noise (weather) ratio. Ideally a reference series must have no more than “small” inhomogeneities. By design as regional averages over many stations, ready availability, and near global coverage, BEST is a very convenient source of reference series, at least for the post 1910 period in Australia. BEST Tmax data for New Zealand appears to fail to match raw data weather fluctuations before around 1942, the extent of this problem is unknown.

The BEST data locations used for the plots in this article are as follows.

· Rutherglen/Wagga/Deniliquin: BEST-Albury, 36.17 S, 147.18 E

· Halls Creek: BEST-Halls Creek, 18.48 S, 128.45 E

· Palmerville: BEST-Palmerville, 16.87 S, 144.00 E

· Boulia: BEST-Mount Isa, 20.09 S, 139.91 E

ACORN-SATv2 daily data (to May 2019) from CSV files was converted to monthly averages, requiring no more than 6 missing days of data in a month. Missing months of data were automatically infilled, up to a maximum gap size of 3 years, using BEST data, and the raw data either side of the gap, for the month in question. The infilling is not essential, but it makes the plots easier on the eye.

Advertisements

76 thoughts on “The BEST and worst of ACORN-SATv2 Tmins

  1. I got the jest of this, but after “This article is about why the BoM map shown above has warm and cold spots, including a somewhat implausible cooling region near Halls Creek in Western Australia.” I was technically lost. I know I am not a scientist–just a soft science researcher, but for us not in the know on the fine details–could we please have just a layman’s summary? Please please?

    • As a senile ex-auditor, I find that it’s not the “how” or “how much” the recorded temps or sea levels or any other original data has been altered that casts doubt on the AGW construct, it’s the “why” of these interventions.
      The answers to the “why” are usually redolent of hand-waving and deflection.

        • Nick, Why are there “statistical adjustments” for ACORN-2?

          How can one tell where the measurements were wrong?

        • Yeah, but the whole map would still be pink (Warm coloured) with a warming trend no more than five hundredths of a degree per decade!

          And while you’re on a winner Nick, you should mention that the BoM map would be completely blue (Cool coloured) – showing a zero trend – if it had included the warmest part of the record; before 1910!

          You know, those earlier times like the infamous Black Thursday of 1851, the heat waves of 1896*, that Red Tuesday of 1898 or that devastating drought so important to history that it merited three capitalised names – the Long Drought, the Great Drought and the Federation Drought of 1895 -1903 (and even longer in some regions).

          *Conditions were particularly bad in outback New South Wales where there were reports of heat up to 52°C. The town of Bourke had an average over three weeks of 44°C, including four consecutive days of 48°C. Those who could had fled Bourke by train, but some 160 people died of heat and disease. (Russell, “On Periodicity of Good and Bad Seasons”; Melbourne Age and Argus, 6 Jan 1896 ff, esp Age, 24 & 29 Jan 1896, and Argus 22 Jan 1896.)

        • Nick leaves out that Australia has been warming well before the current era and a couple of the current warm hot spots on the data are obvious data artifacts. That isn’t denying the overall trend of warming but what is shown is hard to evaluate without seeing the historic warming trend.

      • Nick Stokes “There is no information in this post about warming trends, The plots are of differences between measurements of temperature.”
        trend 1. a general direction in which something is developing or changing.
        information in this post “The plots are of differences between measurements of temperature.”
        ie trends
        “warming trends” are upwards differences in temp over time, which is all the changes in temp on this biased recording

  2. “At this point some readers will question the validity of BEST data.”
    Of course. All that is shown here is that the ACORN data differs from the BEST data (somewhere). That doesn’t make ACORN wrong.

    But they aren’t even measuring in the same place. BEST Albury is subtracted from ACORN Wagga and Deniliquin. But Wagga is 120 km north of Albury, and Deni is 200 km away. Boulia is 300 km from Mt Isa.

  3. BTW, the BoM has adjusted all annual mean temps since at least 1995 to 2017 by at least 0.1C, thanks to ACORN2.
    Check the Annual Climate summaries against the Timeseries graph for comparisons.
    For example, the 2011 CS says:
    ‘The Australian area-averaged mean temperature in 2011 was 0.14 °C BELOW the 1961 to 1990 average of 21.81 °C. This was the first time since 2001 (also a wet, La Niña year) that Australia’s mean annual temperature was below the 1961–90 average.’
    Now check the official Timeseries data.
    http://www.bom.gov.au/climate/change/#tabs=Tracker&tracker=timeseries

    Note 2011 now right on average (and 2001 above average).
    So now they can proclaim that Aus hasn’t had a below average year this century.

    Why would the BoM adjust the most recent temps upwards? I think we all know the answer.

    • Are the recorded temps in one-hundredths of a degree? If not where is the additional precision coming from? An average perhaps?

      • Nope. Added precision via averaging is valid only for multiple readings of the same thing! And by “same thing” that means steady state. A large vat of something that has been mixed and stirred until it has reached, as close as possible, the same temperature throughout, and no more heat addition or leaving, is the idea. Taking multiple readings and averaging can increase the precision, by the number of readings, to the tune of 1/N. So to get a hundredth of a degree precision, requires 100 readings.

        But in the atmosphere, how are you going to take 100 readings of exactly the same parcel of well-mixed air? By the time you have taken the reading convection happens and air molecules have left or entered your environment. And/or the sun shined (don’t fault my made up word), or the black body night skyed, or heat moved from our parcel to the neighboring parcel. Averaging the reading taken in Houston with the reading taken in Dallas, that’s not measuring the same thing either. Thus, when you get done your average has the same precision as the least precise of all your readings. If you have 99 thermometers of automatic recording digital instruments capable of recording to +/-0.01 deg F (such an instrument with a range capable of measuring the full range of any location’s extremes does not exist) and one thermometer alcohol-in-glass manually read and recorded when the junior floor sweeper can get around to it, let’s charitably call it +/- 2 deg C and it’s probably not even that close, then after you average all 100 of your readings, you have a precision +/-2 deg C.

        In other words, everything to the right of the decimal, and a good part of the first digit to the left of the decimal, is made up. Pure bunkum, in other words.

        • I agree completely. Temperature measurements are no different than any other laboratory physical measurement and should be subject to the same treatment as such.

          We are dealing with mathematicians and Excel programmers who have no experience in physical measurements nor having to justify their results.

          Temperature databases were not sent by God on stone tablets. Every study that uses an existing database should treat it as if they were the creators. Uncertainty budgets should be created (by quizzing the managers if necessary) and the precision of the measurements should be described and justified. The uncertainties should be assessed and their effect on conclusions should be stated.

        • I went back and reread your comment. Let me correct something. Multiple readings of the same thing DOES NOT increase the precision or uncertainty of a measurement. You simply can not add a digit or two of precision through averaging.

          What you are talking about is the central limit theory/law of large numbers. As you take more and more samples from a population, find the mean of each sample, then plot the sample means, you will see a normal distribution. As you increase the number of samples the distribution becomes steeper and steeper, in other words, the standard deviation keeps getting smaller and smaller. This is where the (1/N) comes from.

          What you are getting is a better and better prediction of the real mean of the population. If you assume that each measurement consists of the “true value” plus a random “error”, what you end up with is the random errors canceling out and leaving the “true value”. In other words, an exact normal distribution where each point above the mean has a corresponding value less than the mean and the mean is the “true value”.

          However the mean of the sample means is neither more accurate nor more precise than the original measurements. A lot of people equate the “error of the means” with the value being more accurate and more precise and less uncertain than any of the original data. This is not the case. It is simply describing how closely the mean of the sample mean is to the actual mean of the whole population.

  4. Mean annual temperature warming rate 1910-2016
    Raw 0.080 °C/decade
    V1 0.100 °C/decade
    V2 0.123 °C/decade
    That is a big difference.
    Excuse given –
    “The stronger warming trends in the homogenised datasets, relative to AWAP, largely reflects the
    tendency over time for sites to move from in-town to out-of-town locations”
    http://www.bom.gov.au/research/publications/researchreports/BRR-032.pdf

    Even though some of the stations are in big cities and experience considerable urban heat island increase.
    And most are in remote rural towns where the change between heat island of one location to another must be negligible IMO.

  5. As far as I know, temps in at least the first half of the 20th century were recorded with integer values and an uncertainty of +/- 0.5 degrees.

    This means the best temps you can plot are also integer values. The graphs shown has values down to the tenths if not the hundredths. Perhaps someone can explain how the scientific use of significant digits in physical measurements can allow this.

    The use of “anomalies” does not absolve a scientist from using correct procedures in portraying the correct accuracy and precision of measurements. If the temps are recorded in integers, any average and any corresponding baseline should be in integers also.

    Why round measurements? Here is an explanation from the Chemistry Department in Washington University at St. Louis that says it better than I could. “Significant Figures: The number of digits used to express a measured or calculated quantity. By using significant figures, we can show how precise a number is. If we express a number beyond the place to which we have actually measured (and are therefore certain of), we compromise the integrity of what this number is representing. It is important after learning and understanding significant figures to use them properly throughout your scientific career.”

    This is an extremely important concept: expressing a number beyond the precision to which it was measured, compromises the integrity of what the number is representing. You should be cognizant of this throughout your scientific career!

    In addition “error bars” should also be portrayed on the graph. I don’t even see any reference as to uncertainties and how they affect the conclusions.

    • Getting higher than integer resolution averages (and other metrics) from a set of integer temperature readings is done using basic statistics.

      Quick description:

      Consider a very large number of sample temperature readings with only integer designations of the estimations of the temperatures (ALL measurements are estimations – no such thing as a perfect measurement). So, in actuality, 100% of readings will have an associated amount of error. If the set of error values are compiled and studied (e.g. plotting how many readings in the set have a defined amount of error), knowledge of the amplitude and distribution of the errors will usually form a well defined pattern. Frequently the error amplitudes will have a “normal” bell curve distribution. Other times there will be skewed distribution of the measurement error amplitudes, and that skew can be characterized and quantified. Statistical analysis that employs knowledge of the amplitude and skew and distribution of the errors of the Integer temperture readings (in this discussion) will be able to derive Averages of the many readings (and many other metrics besides Averages) with far far higher resolution than the original integer readings…with the resolution increasing with the number of readings.

      Almost no analytical science can be “done” without employing statistics. It is a reliable and powerful tool when used appropriately.

      • But does that not apply only to measurements of the “same thing” as described by RED94 above? Seems to me that a measured temperature in New York City is not the “same thing” as a measured temperature in Chicago even if identical instruments are used.

        • Exactly. A temperature reading taken and recorded is the only value you will ever have for that point in time at that location. You can not take a reading either later in time or in another location and say that you can correct the original reading to a more precise or accurate temperature.

      • I discussed some of this in a post above. What you are describing is the central limit theory/law of large numbers. These deal with how closely the means of samples can predict the true mean of the whole population.

        As an example, if you break a population of 1000 things down into 50 samples of 10 things each, find the mean of each sample and then graph the value of each sample mean, you will likely end up with a normal distribution. The mean of this normal distribution will closely approximate the mean of the whole population. The more samples you take that are independent and that have an identical distribution to the population, the more accurate the mean of the sample means will be. If you think about it, if you take 100 samples of 10 each you basically have the whole population and the means of the sample means and the mean of the whole population should agree exactly.

        As I said above, none of this has anything to do with the precision, accuracy, or uncertainty of the measurements. Calculating the mean of a sample of temperatures consisting of integers should be rounded to an integer following significant digit rules. The mean of the normal population resulting from plotting a bunch of sample means should also never have more digits of precision than the original measurements.

        This doesn’t mean that the “error of the mean” of the sample means distribution won’t be very small but remember what it is telling you. It is telling you that the mean of the sample means distribution is very close to the mean of the entire population. It doesn’t mean that the mean itself has more accuracy, precision or less uncertainty than the actual measurements.

        • … the entire population.

          And this is why I disagree. By “population” it should mean a somewhat homogeneous group, that will have a normal distribution around some mean. Temperature readings of a bunch of different places are not a homogeneous population. It’s a bunch of different populations and therefore getting a normal distribution would be mere happenstance. I’m not a statistician but I am an engineer, and part of engineering is knowing how to measure stuff. I repeat, when reporting averages of temperatures of air, anything to the right of the decimal, and a good percentage of what’s left of the decimal is pure fiction.

          • I don’t disagree with your last point about what is right of the decimal point being pure fiction.

            However, here is a description of the central limit theory from

            https://guides.fscj.edu/Statistics/centrallimit

            Central Limit Theorem
            The central limit theorem states that the sampling distribution of the mean of any independent,random variable will be normal or nearly normal, if the sample size is large enough.

            How large is “large enough”? The answer depends on two factors.

            Requirements for accuracy. The more closely the sampling distribution needs to resemble a normal distribution, the more sample points will be required.
            The shape of the underlying population. The more closely the original population resembles a normal distribution, the fewer sample points will be required.
            In practice, some statisticians say that a sample size of 30 is large enough when the population distribution is roughly bell-shaped. Others recommend a sample size of at least 40. But if the original population is distinctly not normal (e.g., is badly skewed, has multiple peaks, and/or has outliers), researchers like the sample size to be even larger.

            In essence, you don’t need a population that is near to a bell curve in order to get a normal curve from a distribution of sample means. Take it from me, you will get at least a somewhat normal curve depending on the size of samples and how far from normal the population is. The more samples you have the smaller the variance, standard deviation and error of the mean you’ll have for your sample means graph. However, too many misinterpret to mean that a more accurate and precise mean. It doesn’t mean that. It means the normal curve gets tighter and tighter as you home into the mean of the population. If you sample your entire population, the mean of the sample means and the real means of your population should be the same. That means the normal distribution of the sample means would have a variance, standard deviation, and error of the mean of zero.

            A corollary is that the standard deviation and variance of the original population has not changed one iota. Neither has the accuracy, precision, and uncertainty of the original measurements.

            This is all foo-fa-rah from sampling theory where you have no idea what the mean is. If you have all the temperatures, simply add them together and divide by the number of items. This is what computers are for. There is no need to go through the statistics to find the mean, just do it. You can then concentrate on accuracy, precision, uncertainty, significant numbers, etc.

  6. Nick

    ““At this point some readers will question the validity of BEST data.”
    Of course. All that is shown here is that the ACORN data differs from the BEST data (somewhere). That doesn’t make ACORN wrong.”

    I wish skeptics would understand simple logic.
    but they dont.

    ACORN and BEST use entirely different adjustment proceedures

    BY DESIGN !!!

    ACORN ( as I recall) uses or has tested a variety of bottoms up methods that basically follow the
    same proceedure:

    1. Compare stations
    2. Make adjustments to stations
    3. Average adjusted series.

    The approach at BEST is entirely different. We do it entirely differently as a CROSS CHECK on the standard
    accepted methods. That was the WHOLE POINT of our approach. To show that a completely data driven statistical approach would give you the same GLOBAL ANSWER. Not the same regional answer. not the same Local answer.

    in our method
    1. A regional average of raw data is created.
    2. Stations segments are compared to the regional average.
    3. Dpending on their agreemwnt with the regional raw data the station Segments are give a quality
    weight.
    4. A regional average of WEIGHTED raw data is created.
    5. repeat 2&3 until errors are minimized.

    After the global average of WEIGHTED RAW DATA is created we can then predict what a given station
    SHOULD HAVE RECORDED.
    We call this predicted series “Adjusted data” but in actuality the data was never adjusted.

    Comparing our “adjusted” series with other methods on a station by station basis really doesnt
    understand the purpose to the proceedure. We know that these predicted series ( what we call adjusted data) wont match series that are individually adjusted. That wasnt the aim. The aim was to show
    that an entirely new method that was purely statistical ( no human thumbs on scales) would give you
    the same GLOBAL ANSWER.

    • “Steven Mosher December 12, 2019 at 6:13 pm

      ACORN and BEST use entirely different adjustment proceedures”

      Yes, we know this and you confirm what most believe that both datasets are not fit for purpose.

      • As Nick Stokes hinted at above, it is the TREND that is important not the absolute values. However I have niggling concerns about the use of erroneous absolute “corrected” temperatures especially if they are used as real temperatures in a different role to simple trend analysis. I expect that this is done frequently by some.

        • Agreed. However, we know the BoM is a data fiddler. The BoM adjusts past data taken on early instruments, then adjusts them as if they were taken on a modern device. That’s not good data management. The data, with constant adjustments, becomes unfit for purpose. There is no statistically significant trend before adjustments.

    • That was the WHOLE POINT of our approach. To show that a completely data driven statistical approach would give you the same GLOBAL ANSWER.

      The “WHOLE POINT of” your approach was to show a particular answer? I thought the idea behind data analytics was to show whatever the data showed?

    • The same global answer? Is that the sound of goalposts being moved? Both ACORN and BEST purport to give local answers, you should be pleased that your local answers are right for Oz, not so for ACORN V2.

    • LOL. Disagreeing local data leads to agreement when compiled to global…that is not “how science is done.” It is GIGO.

    • Mosher –>

      I have several questions.

      1. A regional average of raw data is created.

      How do you define a region? Do all the stations in the region experience the same weather simultaneously? Are they all at the same altitude, barometric pressure, humidity, etc. all the time?

      2. Stations segments are compared to the regional average.

      Same questions as 1., plus, why do you assume each station should be following the average? Have you done research to confirm that this is a valid assumption.

      3. Dpending on their agreemwnt with the regional raw data the station Segments are give a quality
      weight.

      How do you prevent assigning a poor quality weight to a station where it is perfectly legitimate that they do not agree with your regional average?

      4. A regional average of WEIGHTED raw data is created.
      5. repeat 2&3 until errors are minimized.

      Why does this operation give you a more accurate temperature? You are somehow assuming that making everything agree with a moving average makes the temperatures more accurate.

      • Jim,
        Here is an example of what I think what you are getting at. Maybe Steve can explain.
        Below is a table showing the mean Jan temps for four towns in NW NSW (Australia); the Jan mean temps recorded for January, 1939; and the ACORN1 adjustments.
        All time Jan39 Jan 39 adjusted
        Town Jan Mean Raw ACORN1
        Bourke 36.3 40.4 40.04
        Cobar 35.0 40.1 40.19
        Walgett 35.4 39.15 40.16
        Tibooburra 36.2 40.1 40.08
        See what happens?
        Under the raw data, Bourke has the highest Jan mean and the highest Jan 39 mean (as it should).
        After a 0.36C adjustment, Bourke now has the lowest Jan 39 mean in ACORN1.
        The highest adjusted Jan 39 mean now goes to Cobar which has the lowest Jan mean.

        See what happens with homogenisation when one hand doesn’t know what the other is doing.
        Bourke gets adjusted down to ‘smooth’ the other towns and the other towns get adjusted up to ‘smooth’ Bourke.

        BTW, Bourke has been adjusted even further down to 39.8C for the Jan 39 mean in ACORN2.

        • It is certainly a concrete real world example.

          I stand by my statement, and your evidence is proof, that you cannot adjust the measurements from one device with the measurements from another device that is measuring something different. That is what homogenization does.

          The numbers you end up with are not measurements! It is basically numerology where you end up with UNREAL descriptions of temperature. That isn’t science.

          If you have a question about about the accuracy of a device, then check the calibration of the device. If you want a real regional average, then average the actual recorded temperatures and live with it.

          • It makes sense to me Jim, you need to put in what the actual figures are to get an accurate answer. Otherwise it’s the equivalent of creative accounting. Coming up with a figure to suit the circumstances.

            I’m still waiting for the answer to my question regarding data being left out. The minimum temperatures are being randomly adjusted up at the end of the day from the early morning reading on my weather app. They are also leaving out big chunks of the data readings for minimum, maximum and rainfall. The figures for last month left out six non consecutive days of data. The thing is, they go on to average the figures that were recorded as the monthly figures. This has been happening for a few years now.

            What I’m getting at here is that is as a layman I don’t trust the figures that my app claims comefrom BoM. Even I know that if data is left out, then the figures that are claiming to be the average Min, Max or rainfall for my area are pretty much rubbish.

            Why would I believe any claim as to whether or not temperatures are doing anything unusual when I don’t know if the figures given are true?

            I know that this post is talking ‘big picture’ stuff from a scientific viewpoint that is beyond me, but does common sense not come into it at all? Sure if you were all on the same page, you could have a meaningful discussion. As far as accurate data, truth goes in, truth comes out, surely.

  7. I’d like to know how many devices were in Australia in 1910 given only 112 are used today to calculate a national average (Meaningless).

  8. I live about fifty kilometres from Rutherglen and have been to Halls Creek three times , there is no way that BOM have this right .

  9. I’ve been following the ‘Weatherzone’ app for many years now. If it was hot I liked to see just how hot it got that day. If it was cold I liked to see just how cold it was. If we had heavy rain…you get the picture.

    I liked to see how the averages changed from month to month and how much they did or didn’t change.

    That is, until I couldn’t trust the data anymore. I can’t pinpoint when it changed, maybe two or three years ago gaps started appearing in the data. Sometimes it was only a day or two, but after a while it increased.

    In winter particularly I liked to check the minimum temperature early in the morning. Sometimes I’d feel a sense of confirmation, like, no wonder I felt so cold. At the end of the day when I checked the minimum and maximum temperatures, I sometimes found that the minimum had been raised, sometimes a few degrees.

    I thought that maybe the weather station at Terry Hills had problems of some sort, but then I moved to the Central West of NSW.

    Same issues here with missing data and changed minimums. I just checked last months data and monthly averages. Waste of time really, 10 maximum and 9 minimum temperatures are missing from the month of November! The averages are meaningless. The rain data was missing 5 days too but given that we are in a drought here that may not be so critical, for now anyway.

    Why would I ever believe the hype about ‘increasing temperatures’? How could anyone believe it’s true when we are being blatantly lied to?

    • I also use the weatherzone app.

      I’ve noticed for years that the historical data (only for the current month unfortunately) in graphs showed about the long-term average most times, but the figures below always showed higher than average temperatures for the month.

      When your look at the fibe print, the graphs are from our local racecourse, and the figures are from the airport. This tells me that real rural temperatures are not increasing, but airports are. Who’d’ve thunk it, eh?

  10. Steven Mosher
    “the ACORN data differs from the BEST data (somewhere). That doesn’t make ACORN wrong.”
    ACORN and BEST use entirely different adjustment procedures
    BY DESIGN !!!
    ACORN ( as I recall) uses or has tested a variety of bottoms up methods that basically follow the
    same procedure:
    1. Compare stations 2. Make adjustments to stations 3. Average adjusted series.
    The approach at BEST is entirely different. We do it entirely differently as a CROSS CHECK on the standard
    accepted methods. That was the WHOLE POINT of our approach. in actuality the data was never adjusted. Not the same regional answer. not the same Local answer.
    in our method 1. A regional average of raw data is created.
    2. Stations segments are compared to the regional average.
    3. Depending on their agreement with the regional raw data the station Segments are give a quality
    weight.
    4. A regional average of WEIGHTED raw data is created.
    After the global average of WEIGHTED RAW DATA is created we can then predict what a given station
    SHOULD HAVE RECORDED.
    We call this predicted series “Adjusted data” but in actuality the data was never adjusted.

    “I wish skeptics would understand simple logic. but they don’t.”

    Too funny for words.

    1. simple logic. Let’s try. Two statements from above.
    “Depending on their agreement with regional raw data the station Segments are give a quality
    weight.”
    “but in actuality the data was never adjusted.”
    FAIL.
    Try a third statement
    “ACORN and BEST use entirely different adjustment procedures”
    So the data is adjusted, is not adjusted and then is adjusted in totally different ways, by DESIGN!!!, but not adjusted.
    Good ole simple logic.

    2. “ACORN and BEST use entirely different adjustment procedures
    BY DESIGN !!!”

    So not an accident?
    Perhaps more of a nature trick?
    Maybe a marine scientist or someone from the fashion industry

    Anyway
    Nick Stokes spells it out
    “At this point some readers will question the validity of BEST data.
    Of course.
    All that is shown here is that the ACORN data differs from the BEST data (somewhere). That doesn’t make ACORN wrong.
    But they aren’t even measuring in the same place. BEST Albury is subtracted from ACORN Wagga and Deniliquin. But Wagga is 120 km north of Albury, and Deni is 200 km away. Boulia is 300 km from Mt Isa.”

    What Nick is trying hard to tell Steve is that the data sets do not match.
    They are taken from different places at times.
    Acorn uses adjustments which BEST cannot as it uses a different system with different weightings.

    And miracle of miracles
    The aim was to show that an entirely new method that was purely statistical ( no human thumbs on scales) would give you the same GLOBAL ANSWER. ” presumably as “the standard accepted methods.”

    Lets try that one in slow motion . Ignoring the ( no human thumbs on scales) claim that Statistics alone can produce an entirely new method without human involvement.
    Acorn and other standard accepted methods use adjusted data and can be used to give a GLOBAL ANSWER.
    Multiple data sets all using different subsets of the the greatly adjusted raw data all come up with the same GLOBAL ANSWER.
    BEST comes along, uses the raw data not the adjusted data, makes it’s own mysterious hand waves and voila, like magic, also gives the same GLOBAL ANSWER.

    Simple logic would say that the chances of two such different data sets and methods would give different answers.
    Simple logic would say that the chances of ending up with the same GLOBAL ANSWER is virtually impossible.
    Of course motivated reasoning, human nature, thumbs on scales [AKA adjustments] could all be posited with much higher probabilities.
    “I wish skeptics would understand simple logic.”
    Beware of having your wishes granted.

    Notes
    ACORN-SATv2 is the latest (2018) version of homogenised Surface Air Temperatures (SAT) produced by the Australian Bureau of Meteorology (BoM), intended to indicate how air temperatures in Australia have varied from 1910 to the present time. BEST stands for Berkeley Earth Surface Temperatures.

    ACORN-SAT is an adjusted temperature dataset. [BOM]

    • Sorry agech. If you dont get the difference between adjusting data… change 1 to 1.5 and weighting the data i cant help you.

      • Interesting. In engineering, if there is a tolerance of 1 micron and you change that to 1.5 you will find your work to be out of tolerance by 50%. I am glad you don’t build aircraft.

      • Steven. If you fire off an email trying to explain at how you arrive at your results, it will lead to confusion. I don’t think you can explain the statistical methods you use in an email.

      • It’s OK.
        “Adjusting” is a pejoritive word and has many sinister undertones.
        Like Gliek.
        Adjusting means changing.
        I emphasis with your dilemma.
        “Weighting the data”
        “Adjusting the data”
        You seem to be adjusting the meaning of weighting to assert it has no change on the data.
        Because you are stuck.
        Does not wash.
        Weighting data is adjusting it.

        Your words, not mine.

        “ACORN and BEST use entirely different adjustment procedures
        BY DESIGN !!!”

        You have been in denial on this point for so long it no longer hurts me to see a proud man like you in this position.

        • You can not be scientific and “weight” data in order to force it to meet an average. That is changing a temperature measurement made by a specific device based upon a temperature reading from other device(s). That is going against everything I have ever heard or learned.

          This is saying that you can increase the accuracy of a device by using measurements of different things by other devices. I can find no references anywhere that says this is an appropriate method to use in scientific endeavors.

          This sounds like what mathematicians and programmers would do with counting numbers. It’s not what scientists do with physical measurements. If you have questions about the accuracy of a device, then you should investigate that device not work up some so-called procedure using other devices to change the data.

      • Steve

        Please explain what happened with the four NSW sites which I have set out above at 8:41.

        BTW, Bourke’s Jan 39 temps were adjusted each day in the ACORN1 version.
        All temps above 30C were adjusted down (by up to a degree C) and those below 30C adjusted up by 0.1C.

        ACORN1 was replaced by ACORN2 because the first version was hopeless. ACORN2 is just as bad.

  11. Michael,
    After that bit of fun.
    “This article is about why the BOM map shown above has warm and cold spots, including a somewhat implausible cooling region near Halls Creek in Western Australia”

    “A four-pronged validation analysis is being mounted against version 2 of ACORN-SAT,”

    You need to add a simple 5th prong.
    The temperature map as shown is fatally flawed.
    It has bands of concentric temperature changes that go all the way from North to South Coast.
    This is physically implausible, improbable and virtually impossible.
    Australia is a massive land with quite variable temperature conditions from North to South.
    The temperature bands of contiguous temperature should lie in a west to East direction.
    Not ever go the same all the way from North to South.
    It is virtually impossible for a Northern based e.g Darwin Temperature change to match a Southern eg Great Australian Bight Temperature change over any reasonable long time period.
    They simply should not be joined in this way.

    It possibly indicates a grouping and homogenization of stations that are on other sides of the continent to each other.
    It could indicate that these stations have been deliberately grouped together due to the paucity of actual weather stations in these areas in the past.
    One of the problems for both BEST and Acorn is that these guys like to try to get spatial gridding where possible, impossible in most of central Australia. So they grab known sites Kalgoorlie, Broome, Perth in WA I presume, Halls Creek, Darwin, Alice Springs hence the local variations and then link then up to badly placed stations which they then homogenize over long distances as if no one is looking.

    The patterns seem to indicate central major reference stations.
    1 cold spot Halls Creek, 1 mild warm spot around it, 6 lukewarm spots, 4 warm spots including a massive part of central Australia going to all 4 coasts ,6 or 7 very warm spots, 3 very hot spots

    Homogenization of records over such vast distances for land masses is fraught with danger of getting the regional areas wrong.

    I see you have included a little upside down Australia map The Australian Bureau of Meteorology Gets it Wrong July 4, 2013 at the bottom of your article. It is not at all the same thing, But,

    That layered change in temperatures should be similar to the way that temperature variations should appear on a map of Australia.
    Colder areas should rarely have the same temperature change over time as warmer regions. They should not match.

  12. At this point, what does it matter? I no longer trust any set of data for which the original raw data is not available. Too many people with agendas manipulating numbers to conform either their personal bias or the groupthink of their organization.

    I only look at what is happening now vs the panic of the alarmists. Satellite and balloon temps vs models. Actually numbers of polar bears vs wild stories. Amount of new green leaf coverage vs disaster predictions. Rivers being cleaned of pollution. All pointing to a healthy earth doing what it always has done, vary a bit in spots, but provide a nice warm, safe place to live. Or at least til the next ice age, which will be the real killer of civilization.

  13. ‘The original daily data is available at BoM’s Climate Data Online page.’

    Despite some errors, the BoM’s CDO data is the only data worth referencing. ACORN1 and 2 are useless data sets as they only use around 110 sites and are heavily adjusted (same as the BoM’s first homogenised data set, High Quality Data).

    Another method of adjustment I have noticed is ‘smoothing’ the temps between sites and using the temps from 1961-1990 as the average mean.
    Queensland’s min mean for Sep 2019 was -0.03C yet 80% of the individual sites were below each site’s average.
    How do you get an anomaly map like this and claim it’s an almost average min mean?
    http://www.bom.gov.au/web03/ncc/www/awap/temperature/minanom/month/colour/history/qd/2019090120190930.gif

    Full QLD Sep 19 report here.
    http://www.bom.gov.au/climate/current/month/qld/archive/201909.summary.shtml

Comments are closed.