ElasticTruth

 

Special Report:
The Precision Collapse: 80,000 Days Where the Bureau's "High-Quality" Data Lost Its Decimal Points

The Bureau said "some stations at times" lost decimals.
The data shows 33 stations lost over 80,000 days—at exactly the same time.

The Bureau of Meteorology claims its ACORN-SAT dataset is the highest-quality climate record in Australia, accurate to 0.1 C. But a forensic audit has revealed a massive data-handling failure.

For years, the Bureau's automated weather stations were recording temperatures in whole degrees only, silently destroying the decimal precision. And the Bureau's quality control system didn't catch it.

 The Scale of the Loss

The Bureau has publicly acknowledged that "some stations at times" lost decimal precision in the early years of the automatic weather station era. But the data shows it was far worse than that.

Across 33 ACORN stations, there are 80,612 days where the temperature was recorded in whole degrees only. During these windows, the effective resolution drops from the promised 0.1 C to a blunt 1.0 C.

- Cape Otway (minimum): 3,533 days (nearly a decade).

- Katanning (maximum): 2,891 days.

- Cabramurra (maximum): 2,403 days.

These are not occasional rounded months. They are near-total loss of the decimal digit for years on end.

 The QC Failure

99.5% of these days are flagged as "Y" (quality checked). Why? Because the Bureau's QC checks for missing data, not missing precision. A value of 31 C passes the check, even if the thermometer is supposed to be recording 31.4 C.

The data loss begins exactly when the Bureau installed its Automated Weather Stations (AWS) in the late 1990s, and ends abruptly in 2003. This proves the data loss is a Bureau data-handling failure. Their systems were archiving whole degrees and throwing away the decimals for years, and they never told anyone.

 Why This Matters

BOM is correct that pure whole-degree rounding produces no systematic bias in the long-term mean. But it does affect:

- Threshold counts: Heatwave days, frost days, growing-degree days. This occurred during the 2002 droughts.

- Extremes and records: A 1 C grid cannot record a record-breaking hot day that is 0.3 C above the previous record.

- Day-to-day sequences: The flatlines and artificial runs that plagued the original audit.

 The Bottom Line

The Bureau's "high-quality" dataset contains multi-year periods where the effective resolution is 1 C, not the nominal 0.1 C. And it was never disclosed.

The Bureau's quality control system certified 80,000 days of lost precision as "quality checked." That is a fundamental failure of data integrity.

Thanks to Patrick Lloyd for the updated counts.