A Forensic Analysis of Systematic Data Anomalies in the Bureau of Meteorology's Official Dataset
The record reveals profound data-quality failures that persisted through quality control pipelines and all versions of homogenization in the ACORN-SAT record.
Executive SummaryThis audit examines one fundamental question:
Can the Bureau of Meteorology's temperature dataset be trusted as a faithful record of what thermometers actually recorded?The answer is no.The audit has identified systematic, widespread data integrity failures that persist from the "raw" Climate Data Online (CDO) record through every version of the ACORN‑SAT homogenised product. These are not subtle statistical anomalies—they are obvious patterns of human intervention that should have been flagged by any competent quality control system.
3. Massive data voids – 647 consecutive days with no observations at an ACORN‑quality station, and 408 consecutive days with no decimal precision at another.4. 110 years of one‑way adjustments – Sydney's maximum temperature adjustments have been positive in every single year from 1910 to 2020. Not once in over a century has a net cooling adjustment been applied.5. Machine‑learning predictability – adjustments are predictable from calendar information alone (R² ≈ 0.70) and from data at a station 800 km away in a completely different climate zone (R² ≈ 0.71). This proves the adjustments are following a systematic, network‑wide pattern, not responding to individual station events.6. Forensic “human fingerprint” patterns – 879 Value Traps and over 500 structured repeat patterns (3+1+3, 4+1+4, 5+1+5) across 224 station‑variables. These are the unmistakable signatures of manual data manipulation.Crucially, these artifacts are not corrected by homogenisation. They are carried forward, unchanged, through every version of ACORN‑SAT (v1 to v2.6). The algorithm built its reference correlations on top of corrupted data.None of these findings require any knowledge of climate science – they are purely data integrity failures that would disqualify any dataset in any other field of science.
Key findings:
1. Large‑scale copy‑paste operations – entire months of daily temperatures duplicated identically across years, including one case spanning 40 years and another spanning 82 years. A “broken seam” pattern – 14 days copied, a 2‑day gap at the exact point where source data was missing, then 15 days copied – proves manual editing.2. Statistically impossible flatlines – 17 consecutive identical daily maximum temperatures at an ACORN‑quality station (probability ≈ 10⁻³⁴). Extended periods using only 3‑5 distinct temperature values.3. Massive data voids – 647 consecutive days with no observations at an ACORN‑quality station, and 408 consecutive days with no decimal precision at another.4. 110 years of one‑way adjustments – Sydney's maximum temperature adjustments have been positive in every single year from 1910 to 2020. Not once in over a century has a net cooling adjustment been applied.5. Machine‑learning predictability – adjustments are predictable from calendar information alone (R² ≈ 0.70) and from data at a station 800 km away in a completely different climate zone (R² ≈ 0.71). This proves the adjustments are following a systematic, network‑wide pattern, not responding to individual station events.6. Forensic “human fingerprint” patterns – 879 Value Traps and over 500 structured repeat patterns (3+1+3, 4+1+4, 5+1+5) across 224 station‑variables. These are the unmistakable signatures of manual data manipulation.Crucially, these artifacts are not corrected by homogenisation. They are carried forward, unchanged, through every version of ACORN‑SAT (v1 to v2.6). The algorithm built its reference correlations on top of corrupted data.None of these findings require any knowledge of climate science – they are purely data integrity failures that would disqualify any dataset in any other field of science.
The full audit is in PDF format and documents all the above anomalies and more:
No comments
Post a Comment