ElasticTruth

  The Three BoM Defences, and Why None of Them Work

There are three arguments being used to dismiss the audit. None of them survives contact with the documents. Here they are, one by one.

 Defence 1: "The peer reviews covered this."

On 23 October, Media Watch wrote to Dr Jennifer Marohasy that the coverage was missing "key context, including the past reviews into BoM's temperature record."

The reviews are the 2011, 2015 and 2017 assessments of ACORN-SAT. They are real, and they are substantial. But read what they actually cover.

They assessed the homogenisation methodology — the statistical algorithms used to adjust for site moves and equipment changes. They asked: is the adjustment method sound? They answered: yes.

They did not audit the raw data for duplicated blocks. None of them ran pattern-matching searches for whole-month replications or multi-day identical sequences. That was not their scope. It is not a criticism of them to say so; it is simply what they did.

So when the reviews are cited as "context that was missing," the context being supplied is a review of one thing, presented as though it were a review of another. A review that confirms the adjustment method is sound tells you nothing about whether June 1943 was copied into June 1944.

The question of whether a block of raw data is a duplicate is not answered by any of those reviews, because none of them asked it.

That is not a technicality. It is the whole point.

 Defence 2: "You need a climate science qualification to see this."

The second defence is that the audit requires expertise that the auditor does not have.

It does not. What the audit requires is a file and the ability to compare two columns.

Here is what finding a duplicate involves:

1. Open the CSV file the Bureau publishes.

2. Sort by date.

3. Compare the June 1943 column to the June 1944 column.

4. Notice that all thirty values are identical.

That is it. That is the whole technique. It is called data deduplication, it is a standard operation in computer science, and it is used in every backup system in the world. It does not require a climate science degree. It requires a spreadsheet.

The same is true of counting days without decimal precision. The file has a column. The column has values. Some have a decimal point and some do not. You count them. That is the technique.

The same is true of the three-day patterns and the value traps. You look for repetitions. You count them. You compare the frequency to what random variation would produce.

These are computer techniques, not climate science. They have been standard for decades. Every data analyst uses them. Every quality-control process relies on them.

The question is not whether the auditor has a climate science qualification. The question is why the Bureau's own quality control did not catch these patterns, when a spreadsheet and an afternoon would have.

 Defence 3: "The conclusion was reached, and the audit is exaggerated."

On 23 October — before I submitted my statement, before I provided the CSV files, and before I had answered any of the questions put to me — Media Watch wrote to Dr Marohasy:

"Our research suggests the accusations levelled at BoM, while accurate in part, have been exaggerated and are largely disputed."

That is a conclusion reached before the subject responded. It is difficult to describe a conclusion reached before the evidence as an assessment of the evidence.

It is also the reason the peer-review argument matters. If you have already decided the accusations are "exaggerated," the peer reviews are useful — they sound like a rebuttal. If you have not decided, you read them, and you notice that they cover the adjustment method and not the raw data.

The sequence matters. The conclusion came first. The reviews were cited afterwards, as support.

That is not how an assessment works. It is how a defence works.

What the files show

The files are public. They are in the Bureau's Climate Data Online, and they can be downloaded by anyone.

-June 1943 and June 1944, Sydney: all thirty days identical, both flagged Quality = Y.

-December 1935 and December 1936, Sydney: fourteen identical days, a two-day insert, a malformed record, fourteen more identical days, both flagged Quality = Y.

-Georgetown, January–February 1962: 34 of 39 days at exactly 37.8°C, at 0.1°C precision, flagged Quality = Y.

-1995–2011: 51,249 station-days across 23 stations where decimal precision collapsed to 90–100% whole numbers.

The Bureau's statement says it has "detected no further duplicate months in the entire bureau network other than those quoted in the report and then confirmed by us."

June 1943 and June 1944 are further duplicate months. December 1935 and December 1936 are a further duplicate month with a seam in it. Georgetown is not a duplicate — it is a carry-forward, which is a different mechanism and arguably a worse one.

None of these are covered by the peer reviews, because the peer reviews did not look at the raw data.

The files are public. Anyone can open them and see. That is the point.