ElasticTruth

Australia's temperature record lost a decimal place for years. The Bureau knew.

Between 1995 and 2011, at more than twenty stations across Australia, the Bureau of Meteorology's temperature record stopped recording temperatures to 0.1°C.

It recorded them to the nearest whole degree.

51 separate loss windows. 23 stations. Over 51,000 station-days.

At Cape Otway, minimum temperatures were recorded as whole numbers for 3,533 consecutive days — nearly ten years. 

At Katanning, maximum and minimum temperatures lost their decimal for more than 2,800 days. 

At Cabramurra, 2,403 days. At Deniliquin, 2,071 days. 

At Palmerville, 699 days — every single one of them a whole number.

These are not isolated glitches. They are multi-year windows, and the recovery dates cluster tightly around 2002–2004, which tells you the cause was common rather than independent.

Why a lost decimal is not harmless

The Bureau's defense is that rounding cancels out over long averages. For a simple mean, that is largely true.

But nobody uses the record as a simple mean. What people use it for is thresholds.

- A day that truly reached 35.4°C is recorded as 35°C. A day at 34.6°C is also recorded as 35°C. Counts of days above 35°C — the metric behind heatwave and fire-danger warnings — become quantised and unreliable.

- A true 44.4°C and a true 43.6°C both become 44°C. "Hottest day" rankings and record lists depend on the difference.

- At 1°C resolution, true values of 30.6°C and 31.4°C both become 31°C. That is where the flatlines come from. They are artefacts of the recording system, not the climate.

- The Bureau's own homogenisation compares neighbouring stations. When many stations lose precision at the same time, the reference network is contaminated too. The problem becomes harder to detect and harder to correct.

The World Meteorological Organization specifies 0.1°C as the standard reported resolution for air temperature. During these windows, the product falls an order of magnitude short of its own benchmark.

The period that matters

The loss windows overlap almost exactly with the Millennium Drought and the 2002–03 heat and fire season.

The Bureau's own climate statements record 2002 as Australia's hottest year on record for maximum temperatures — national mean maximum 1.22°C above normal, far exceeding the previous record. In January 2003, Melbourne and Adelaide both reached 44.1°C. The summer brought fires that killed four people in Canberra and burned more than three million hectares.

At Deniliquin, Cabramurra, Eddystone and Cape Otway — stations identified in this audit — the daily values during those very years were being recorded to the nearest whole degree.

That does not mean the heat was not real. It was. It means the precision of the official record during the heat was lower than the product claims, and nobody was told.

What the Bureau said, and what it did not say

In the 2011 ACORN-SAT technical report, the lead author wrote that whole-degree rounding "affected some ACORN-SAT sites at times in the 1990s and early 2000s but no longer affects any of them."

That is the full extent of the public acknowledgment.
Over 51 000 days of missing decimal values is not some or at times.

The Bureau has never published a station-by-station inventory. It has never said how many stations, how long the windows lasted, or how tightly they clustered. "Some sites" and "at times" are qualitative, and they are minimising. The data shows multi-year, near-total loss of precision at more than twenty stations, with recovery bunched into a twelve-month period around 2003.

The loss is present in the raw CDO source with identical dates. It originated upstream of ACORN-SAT. Homogenisation neither introduced it nor corrected it. It sits in the final product, unflagged.

The question

The Bureau knew the precision was gone. It wrote a sentence about it in a technical report. It did not tell the users who calculate fire-danger and heatwave metrics. It did not flag the affected periods in the public product. It did not publish the inventory.

If you knew the precision was gone, why was it never flagged where users could see it?

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