Wondering Willis Eschenbach wonders about Australian temperature data

Today on WUWT, Willis Eschenbach is wondering about ACORN-SAT.  He is bemused by the fact that within the Bureau of Meteorology (BoM) temperature data, there are some records in which the maximum temperature is lower than the minimum temperature.  Willis writes:

What happened was that while researching the ACORN-SAT dataset, I chanced across a website with a post from July 2012, about four months after the ACORN-SAT dataset was released. The author made the surprising claim that on a number of days in various records in the ACORN-SAT dataset, the minimum temperature for the day was HIGHER than the maximum temperature for the day … oooogh. Not pretty, no.

Well, I figured that new datasets have teething problems, and since this post was from almost a year ago and was from just after the release of the dataset, I reckoned that the issue must’ve been fixed …

… but then I came to my senses, and I remembered that this was the Australian Bureau of Meteorology (BoM), and I knew I’d be a fool not to check. Their reputation is not sterling, in fact it is pewter … so I wrote a program to search through all the stations to find all of the days with that particular error. Here’s what I found:

Out of the 112 ACORN-SAT stations, no less than 69 of them have at least one day in the record with a minimum temperature greater than the maximum temperature for the same day. In the entire dataset, there are 917 days where the min exceeds the max temperature.

How maximum and minimum temperatures are recorded by BoM

I think Willis has it all wrong.  Here is how the maxima and minima are determined, from the Australian Bureau of Meteorology:

Air Temperature:
Air temperature is measured in a shaded enclosure (most often a Stevenson Screen) at a height of approximately 1.2 m above the ground. Maximum and minimum temperatures for the previous 24 hours are nominally recorded at 9 am local clock time. Minimum temperature is recorded against the day of observation, and the maximum temperature against the previous day. If, for some reason, an observation is unable to be made, the next observation is recorded as an accumulation. Accumulated data can affect statistics such as the Date of the Highest Temperature, since the exact date of occurrence is unknown .

The chart below is how I understand it.  On day one (in blue) the maximum is greater than the minimum, which is usual.  On day two (dark red) however, a change came through and the minimum temperature as recorded on the day of observation to 9:00 am was greater than that of the maximum of the previous day (click the chart to enlarge it).

This is the span of temperature recordings for a 24 hour period,
the maximum and minimum of which are recorded as the max and min temperatures
on the date of "the previous day to the day of observation".

Say for the Day Two temperature record (dark red columns), the left part of the above chart is the 1st December and the right part after midnight is the 2nd December.  The temperature record is for the 1st December.  The observation is taken at 9:00 am on the 2nd December.  The maximum is the highest temperature reached on 1st December.  The minimum is the lowest temperature recorded to 9:00 am on the day of observation.  That is, between midnight and 9:00 am on the 2nd December and recorded as the minimum for the 1st December.

Consider a day in Kyancutta, the first day of December in 1966 when the minimum recorded to 9:00 am on the day of observation was 13.4 degrees, whereas the maximum (observed on the previous day to the day of observation) was only 13 degrees.  The maximum for the second of December was 21.1 degrees as the weather warmed up again.  From the newspaper of the day, it looked as if a cool change swept across the country around that time.

So that explains why I believe Willis is wrong (again).  Feel free to tell me if it's me that's got it wrong.  However based on the past history of Wondering Willis, I'd go with BoM any day.

How often does this happen?

Back to Willis's wonderings.  He found that there were 917 days where the recorded minimum was greater than the recorded maximum.  Each data set runs from a different start date, but let's conservatively set the average recording period at 50 years long.  There are 112 stations.  That would mean that at most, the minimum is greater than the maximum recorded on only 4.5 out of every 10,000 measurements.

That's right.  It doesn't happen very often.

Comments from Willis and others on WUWT

Wondering Willis assumes he's correct but not everyone thinks so.  Instead of doing a bit more investigation, Willis is only too willing to believe that BoM has it wrong.  He writes:

The issue is that the authors and curators of the dataset have abdicated their responsibilities. They have had a year to fix this most simple of all the possible problems, and near as I can tell, they’ve done nothing about it. They’re not paying attention, so we don’t know whether their data is valid or not. Bad Australians, no Vegemite for them …I must confess … this kind of shabby, “phone it in” climate science is getting kinda old …
w.

Finally after 25 comments, a WUWT reader, Johanna, has piped up to correct Willis:

June 29, 2013 at 2:49 am  The BOM clicks on to a new “day” at 9am (presumably when the sleepyheads roll into the office). It is quite possible, but quite misleading, for minima to exceed maxima for a 24 hour period given this. All it takes is a fast moving weather system, of which we get plenty on this vast continent.

I wonder how Willis will react.  The fact that johanna rips into BoM at the end of her comment will count in her favour as far as Willis is concerned.

Another update: After Johanna's comment correcting Willis' mistake, Nick Stokes chimes in and says the same thing.  Guess what happens?  Does anyone complain about Johanna?  No, because she rejects climate science.  Does anyone complain about Willis misleading them?  No.  What do they do? In among the stupid comments, they complain about Nick Stokes of course, because he doesn't reject 97% of climate science in the way the WUWT illiterati do.

Update 3: No, Willis.  It's not an error it's a convention.

This is truly weird.  Quite a number of people now have pointed out that Willis made a mistake and that the above description is how daily temperatures are recorded in Australia.  Willis is having none of it.  He says it's an "error" because Australia doesn't do it the way he wants it done.  I have a suspicion that he doesn't understand the method.  He writes:

Regarding the “explanation”, I don’t care about the explanation. Whatever the circumstances and assumptions might have been, it’s an error.

You seem to think that they are somehow prohibited from fixing an error because they’d be “rightly criticized” … are you serious? Do you know how many times these guys have “adjusted” and otherwise changed the data, without any such obvious error?

Now, I don’t care how they fix it. They can throw out the bad data. Or they can flag it and leave it in. My point is that doing nothing to an admitted error, in a supposedly scientifically quality controlled dataset, does not give me confidence in their other actions.
w.

No, Willis, it's not an error.  You can call it a "practice" if you like or a convention.  A way of doing things here in Australia.  Whatever you call it, it beats the practice in the USA hands down, where time of observation free-for-all causes real problems that has to be corrected for a lot.  As noted in Hansen et al (2010):

Temperature records in the United States are especially prone to uncertainty, not only because of high energy use in the United States but also because of other unique problems such as the bias due to systematic change in the time at which observers read 24 h maximum‐minimum thermometers.

Victor Venema has written an excellent article describing the problem of changes in time of observation as can typically occur in the USA.

Here is how it's done in the UK:

Why are the daily temperature maxima and minima for different 24-hour periods?

Conventionally, maximum and minimum temperatures are recorded for 24-hour periods ending at 0900 GMT each day. Maximum temperatures tend to occur during mid-afternoon, so the relevant maximum for a given calendar day is the one recorded between 0900 on the day in question and 0900 on the following day. However, minimum temperatures generally occur around dawn, so the relevant minimum temperature for a given calendar day is the one recorded between 0900 on the previous day and 0900 on the day in question.

The way I read it is that in Australia, the minimum temperature is assigned to the day prior to the day of observation.  In the UK the minimum temperature is assigned to the day of observation.  In both cases there will be some (rare) days on which the readings can appear anomalous.