Via Jo Nova, an Australian Surface Stations Project has just reported its results.
The BOM say their temperature records are high quality. An independent audit team has just produced a report showing that as many as 85 -95% of all Australian sites in the pre-Celsius era (before 1972) did not comply with the BOM’s own stipulations. The audit shows 20-30% of all the measurements back then were rounded or possibly truncated. Even modern electronic equipment was at times, so faulty and unmonitored that one station rounded all the readings for nearly 10 years! These sloppy errors may have created an artificial warming trend. The BOM are issuing pronouncements of trends to two decimal places like this one in the BOM’s Annual Climate Summary 2011 of “0.52 °C above average” yet relying on patchy data that did not meet its own compliance standards around half the time. It’s doubtful they can justify one decimal place, let alone two?
We need a professional audit.
A team of independent engineers, scientists, statisticians and data analysts (brought together by the joannenova blog) has been going through the Australia Bureau of Meteorology records (BOM). They’ve audited some 8.5 million daily observations across 237 High Quality and other close sites in Australia. Shockingly, while the BOM calls their database “High Quality” and instructed observers before 1972 to record in tenths of a degree Fahrenheit, the auditors started finding sites with long stretches of records where the weather suspiciously rose and fell only in Fahrenheit quanta, like 72.0, 73.0, 72.0, 71.0, 73.0, 72.0. After 1972, the BOM went metric, and oddly, so did parts of the Australian climate. Numerous sites started warming and cooling in pure Celsius integers.
The bottom line:
- The BOM records need a thorough independent audit,
- It’s possible that a significant part of the 20th Century Australian warming trend may have come from something as banal as sloppy observers truncating records in Fahrenheit prior to 1972.
- Many High Quality sites are not high quality and ought to be deleted from the trends.
- Even current electronic equipment is faulty, and the BOM is not checking its own records.
- Even climate scientists admit that truncation of Fahrenheit temperatures would cause an artificial warming effect.
Ken Stewart has the whole in-depth report at his site: “Near Enough For a Sheep Station”
They have done a huge amount of data crunching. Ken has all the graphs of maxima and minima (people were extra lazy on the minima).
Then there is the wierd effect of rounding Fahrenheit to Celcuis and back and getting results of 0.1 and 0.9 when the regenerated Fahreheit records are used instead of the original.
Hopefully the final word, from me on rounding:
Unrounded temperature series example:
1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 = 14.5
Average = 1.45
Now rounded calculations follow of the above series:
1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 = 15
Average : 1.5
In time more recordings are taken, with the same distribution, i.e. the average temperature has not changed:
No rounding:
1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 + 1.4 + 1.5 = 29.0
Average: 29/20=1.45
Rounding the above series:
1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 + 1.0 + 2.0 = 30
Average: 30/20=1.5
—————-
As you can see, whether you use rounding or no rounding, no warming bias is introduced.
I hope I rest my case.
garymount;
the issue is “trends”, as you point out, since that’s the basis of “warming or not”, by definition. However, the absolute errors can’t be so easily sidestepped, since necessarily huge collections of measurements from many global (and national) locations and environments are being merged, averaged, compared, subtracted, etc. Your “consistency” constraint is un-meetable. So the only hope is best absolute accuracy available across and within sites and regions. And that requirement is massively violated.
_________
Ian, your point can also be discussed in terms of “extensive/intensive” variables. Intensive variables like temperature and density can’t be summed or averaged; they are, however, the outcome of other (extensive) variables which can be summed and averaged. Temperature requires heat, humidity, and density, at least, be known for both locations to be added or averaged, etc. It is meaningless, e.g., to average the temperature of a cu’ of air and a cu’ of water. Or a cu’ of air on a mountain top and a cu’ of air at sea level. Yet the latter is done routinely by climastrology.
Philip O’Neill:
As one of the authors of the report, I do apologize if any current or past observers feel this is a ciriticism of them. Far from it, it is a criticism of BOM management fair and square, and of course various Commonwealth governments for not giving them the resources to do the job properly. It is
laughable to expect untrained part time observers to read Fahrenheit thermometers to the nearest tenth in any weather conditions. Townsville, not being one of the Annual HQ sites, was not part of the survey. If it was, from what you say I’m sure it would be near the top rankings- as are other
Regional Offices, Met Offices, and airports mostly. However, the worst 10% of sites we looked at had rounded Fahrenheit data very probably more than 89% of the time. These were mostly country post offices and a couple of lighthouses. Very probably half of all Fahrenheit data surveyed were
rounded, but there were less than 10% of extra unexpected whole degrees in the Celsius era. (I say ‘very probably’ because the conversion from F to C to F introduces its own uncertainty which we allowed to be +/- 5%). This means that if truncation occurred, it would cause an artificial warming trend.
This report, which I urge you to read for a description of our methods and findings, is not a criticism of observers but is a criticism of current Bureau quality control, homogenisation methods, and climate analyses.
I’m sure other members of our team would reply to you in similar vein.
Ken Stewart
Phillip O’Neill wrote:
Somewhat insulting to Australian Observers of which I was one for 30 years, and over the last few years in charge of Townsville station quality control. In my last year our station was the best performing in Queensland with audited errors showing NIL ERRORS – the only such one in the State. I also Instructed for a period before and after 1972. And at no time were Observers taught nor did they in the Field – round upwards to whole degree. It was to the nearest tenth before and after metrification. If there was a need to round it was done to the nearest odd tenth, for instance 7.55 deg went to 7.5 deg and 10.05 deg went to 10.1 deg.
Any rounding that may have occurred would have been in porocessing data much further down the track, if at all.
As a co-author of the audit, I can assure you that we in no way intended offence against the observers who recorded temperatures on behalf of the BoM.
The Townsville station does indeed have an excellent record in logging accurate temperatures and you should be congratulated for your precise quality control of that station. The available BoM raw minima data for Townsville Aero 32040 from the first recording 19 Oct 1940 to 31 Dec 2011 shows that in the Celsius era since September 1972 there were no blocks at all of 10 consecutive days or more with temps rounded to .0 or .5. Excellent and I wish that was the case in all Australian stations as presented on the BoM’s national raw temp database.
In the Fahrenheit era before September 1972, 14.8% of all Townsville temperatures were rounded to .0F. That is, 1,713 out of 11,610. However, the mathematical bias of rounding revealed by proportions of .1F and .9F shows this is almost certainly an underestimate:
F degrees
0.0 1713
0.1 1383
0.2 1112
0.3 1087
0.4 987
0.5 911
0.6 1002
0.7 1012
0.8 1133
0.9 1270
.9+.0+.1 = 4,366 which is 37.6% of the total. This infers a .0F proportion slightly above 20% – only twice the expected proportion of 10%. Phillip, as you say, Townsville has one of the most accurate historical records in Queensland (possibly in Australia) – but it seems some of your trainees still occasionally rounded inaccurately.
You contend that the F data recorded by observers was in fact almost all accurate to the tenth degree, but the BoM may have manipulated it to round as .0F when converting to C. I hope you’re wrong because that would be a final nail in the coffin of BoM credibility.
Since September 1972 within which there is no scale conversion, the record shows 12% at .0C and 23.3% at .0C+.5.C This is within the margin of error and although hinting at a bit of rounding it is not significant.
The evidence within our audit shows that since 1972 almost all the C rounding in Australian weather stations has been up or down (to the odd whole if .5 in accordance with BoM protocol) rather than truncated down, and thus has little influence on the long-term rounded or precise averages since Celsius metrication.
Our audit does not claim pre-72 Fahrenheit temperatures were truncated rather than rounded but does provide evidence from the data that down rounding was more likely than up, and calculates the effect based on different possible proportions.
The audit does not state a definitive impact on temperature trends since 1910 but again based on the available data displayed, speculates that there may have been artificial warming between .1 and .4C over 100 years. However, the audit does state and prove that a large proportion of the raw temperatures upon which the HQ series is based is not accurate.
An accurate impact may be estimated through a thorough professional and independent audit of all BoM records.
Again, Phillip, no offence was meant to yourself or other observers who may have rounded or truncated temperatures. It would hardly be a sin if some (many, based on the audit results) thought 72.7, for example, is sort of correct as 72 but 73 is obviously wrong – particularly if it’s raining, stinking hot or there’s a mob of sheep that needs to be put on the trucks. That’s honest human nature and we are not criticising the practice itself. However, the data cannot be ignored.
The Excel macros built to conduct this audit are freely available at http://www.waclimate.net/round/rounded-background.html if anyone wants to replicate above or any other Australian station.
Brian H says:
March 16, 2012 at 12:41 am
Your “consistency” constraint is un-meetable.
———
As long as the ratio of the number of recordings below the .5 remain roughly equal to those .5 and above from past measurements, then my “consistency” will be met, within a close enough tolerance which should get even closer with more measurements. For example if the ratio of below .5 to .5 and above remains say 2 to 3 (any ratio will do), then there will be no introduced bias.
There certainly can be a bias of a warmer temperature, but your whole temperature set will start out with this little bit too warm than actual when using rounding than you would have produced from not rounding, but unless you start to have an increasing ratio of numbers in the .5 and above range compared to the below .5 range, you will not get a warming bias. That is the consistency I am talking about. It is just as likely a chance to have more recordings compared to previous measurements in the below .5 range and see a cooling bias set in, even when you use a rounding that records temperatures warmer than they actually are.
One would have to deliberately or I suppose some how unintentionally be recording an ever increasing quantity of temperatures in the range that gets rounded up in order for a warming bias to become present in the measurements as time passes, for the rounding to cause a warming bias. This could of course happen by chance, but seems higly unlikely, and by chance a cooling bias could also take place, but again, highly unlikely (with sufficient quantities of measurements, which was mentioned very early on in this thread).
So, once again, it is the ratio that must remain consistent for no bias to be introduce, to either warming or cooling, no matter which rounding method is used.
As a chemistry lab instructor at the University of Texas many years ago I found one of the most difficult concepts for the students, especially the non-majors, was “significant digits.” When asked to report the results of a calculation they would routinely report every last digit their calculators would display. Thus a solution of 1 g of salt in 70 ml would be reported to have a concentration of 14.28571428 g/l.
While I successfully enlightened many, I am sure many others, including some who now work in fields considered ‘scientific’, have never grasped the concept.
Not all numerical uncertainties are statistical.
Science is uncertain – everywhere.
“Consensus’ is a political concept.
Phillip O’Neill, I can understand your chagrin as one of the professional observers caught up in this audit of all High Quality sites in Australia. I was aware that this would happen and was hoping that some observers would come forward and comment and I thank you for that.
I did some observing myself as a schoolboy before 1972 at my amateur home weather site and put my own F readings through the program and was not surprised to see a bias towards some decimal fractions over others. I was particularly reluctant to record the .5 fraction. I was untrained but I’d call myself a diligent observer. I badly wanted to becoem a BOM observer but alas it did not happen.
What is so surprising is that before this year none of us even suspected there were these question marks about the quality of the observations generally. We knew there were some places of doubtful quality at some stage of their existence, but not that this extended to most stations.
As is often the case this study more or less happened by chance rather than design, with people making casual remarks and then others chipping in. I became involved when Chris said it took him two hours to work out the results for one station in Western Australia and I knew I could write a program to get the job done in less than a minute.
This was an audit waiting to happen.
Ian Hill
yamaka says:
March 15, 2012 at 3:05 pm
I think you’ll find there is a bit of a misunderstanding in your idea and any alleged Boeing “stories” are likely to be apocrophal at least. 0.000001 rounds to 0, 0.999999 rounds to 1. 0.499999 rounds to 0 and 0.5 rounds to 1. No need for extra rules and no bias introduced. Rounding produces NO bias, truncation does.
—-
Where did you get the idea that people using slide rules would calculate out to 6 digits?
Even in your example, there is still a bias, the magnitude of the bias decreases as the number of digits of accuracy increases.
LazyTeenager says:
March 15, 2012 at 3:34 pm
Personally I wouldn’t trust JoNova’s fan base to be able to add up a shopping list without stuffing up.
The truncation issue will only have an effect on the trend under certain conditions. I am betting the effect is minor.
——
Personally, I wouldn’t trust you to go to the bathroom without supervision.
Truncation vs. not truncation always produces a bias.
The effect is minor, then again the claimed warming is als minor, and off the same magnitude.
Brian H says:
March 16, 2012 at 12:41 am
……..Ian, your point can also be discussed in terms of “extensive/intensive” variables. Intensive variables like temperature and density can’t be summed or averaged; they are, however, the outcome of other (extensive) variables which can be summed and averaged. Temperature requires heat, humidity, and density, at least, be known for both locations to be added or averaged, etc. It is meaningless, e.g., to average the temperature of a cu’ of air and a cu’ of water. Or a cu’ of air on a mountain top and a cu’ of air at sea level. Yet the latter is done routinely by climastrology.
But if you look at this thread the majority of posters have been drawn into the ‘climastrology’ argument. Or to use the logic aphorism “they are searching for their lost keys under the lamp post as its lighter there”.
I thought this was a scientific blog 😉
We also see the same in satellite metrics where the microwave sounders are used to look at temperature – but if you look at GOES EAST imagery you can see the huge amounts of IR being radiated from water changing state in fronts and storms – and the over quoted Stefan-Boltzmann formula has no relevance to latent heat release which happens on change of state at any temperature.
What should happen is that _every_ time a climate scientist starts discussing the average global temperatures they should be stopped and told that atmospheric temperature is the incorrect metric for measuring heat absorption; and it is heat absorption and imbalance that they claim to be the problem. They should provide a value in Kilo-Joules and perhaps the break down of energy budgets from ERBE and CERES. Don’t hold your breath waiting for this as if they were to use the correct metric there would be no ‘catastrophic values’ to trumpet.
garymount says:
March 15, 2012 at 6:04 pm
Rounding up from 5 is correct because 0 (zero) is also a number, and therefore there are 5 numbers below 5 and 5 numbers from 5 to 9.
—-
Zero doesn’t round. It stays the same.
garymount says:
March 15, 2012 at 7:15 pm
ergo I am correct.
—
On this sequence of numbers. Add a few more numbers to the series, then you stop being correct.
markx said @ur momisugly March 15, 2012 at 6:58 pm
Not so. Some years ago, after hearing Phil Jones being interviewed by Robyn Williams on the Science Show I had occasion to check Australian record temperatures. Jones claimed that the recent spate of record highs was a clear indication of Global Warming. The majority of Australian record highs were in the 1920s, 30s and 40s, not the 1940s only.
I have hardly bothered listening to the Science Show since.
Further to my previous remark, markx is correct in noting that the records prior to 1957 appear to have been disappeared from the interweb. Based on the now missing records, I wrote the following on Monday 3 February 2003:
@MarkW
Please continue to read the comments. You will see where I acknowledge my mistake. However, there are no biased trends with any rounding method consistently used, as I have shown in my further comments.
johanna says: March 15, 2012 at 4:50 pm :Hopefully he (Geoff Sherrington) will be along at some stage to clarify any questions people raise.
Thank you, Johanna, Four points,
1. There is continuing work that will show more adverse consequences that could arise from the work so far reported.
2. If there are any questions, please direct them to the primary authors. They did the hard yards. I merely added an idea now and then. I would not like to misprepresent them by answering for them.
3. Collectively, with a few hundred man-years of combined experience, we would not have written the above if the conclusion was that there was no error of concern. Conversely, the more we dug, the more problems we found. They have not all been reported and indeed a number of people have not read them closely enough to understand them, as judged by comments here.
4. People are encouraged to conduct similar exercises in their respective countries.