Monthly Archives: July 2025

Part 9a. Jervois, Northern Territory Australia

Data Homogenisation

Bureau of Meteorology (BoM) ID 15602, Latitude -22.95o, Longitude 136.1442o; temperature from January 1966 to December 2024.

Dr Bill Johnston

Keywords: BoM; homogenisation reference sites; data quality

Having completed analysis of Tmax data for the Jervois weather station using BomWatch protocols, (Report available here), I asked several online AI-reviewers to provide a short- and medium-length independent overview and summary.

I specifically searched for online AI-reviewers requiring no login and no upfront $$$.  Selfish I know, but I receive no reimbursement for the many hours, weeks and months it may take to research and analyse a weather station. BomWatch protocols are transparent, objective and replicable, which is the hallmark of the scientific method.

AI-reviewers were:

https://notegpt.io/

https://www.portfoliopals.net/summascript/

https://docxsummarizer.com/

https://noiz.io/free-ai-tools/doc-summarizer/

While all Reviewers provided similar overviews, the summary provided by NoteGPT was less accurate and missed the Report’s main focus.

The medium version of the review by https://www.portfoliopals.net/summascript/ seemed accurate and well structured. Highlighting specific aspects in logical order (context and background, key findings and analysis, implications and reflection, and conclusions) helped organise one’s thoughts.

Their 106-word summary can be quoted directly, and only requires light editing:

This report by Dr. Bill Johnston evaluates the reliability of maximum temperature (Tmax) data from the isolated Jervois weather station in Australia, used to homogenise ACORN-SAT records at Rabbit Flat, Alice Springs, and Boulia. Findings reveal the Jervois dataset is of poor quality, marked by missing observations, probable watering, and unexplained step-changes in 1979 and 2017 causing a spurious warming trend of 0.225°C per decade. Dr. Johnston argues, “The overall trend… was thus spuriously due to non-climate factors,” challenging the validity of adjustments based on Jervois data. This critique raises broader concerns about data integrity in climate monitoring and the scientific rigor behind national temperature records.

Some reviewers misinterpreted the role of BomWatch protocols in undertaking multi-faceted analysis of single-site datasets; failed to highlight the metadata problem and that maximum temperature (Tmax) and its relationship with rainfall was a key statistical metric.   

Due its clarity, the medium length summary from https://docxsummarizer.com/ also rated highly as it overviewed the objectives, provided a brief outline of the site, the analysis, and main conclusions. Their unedited report is as follows:

Assessment of Jervois Temperature Data and Its Role in Australian Climate Homogenisation

This report investigates the quality and suitability of maximum temperature (Tmax) data from the isolated Jervois weather station in the Northern Territory, Australia, for use in homogenising key sites within the Australian Climate Observations Reference Network – Surface Air Temperature (ACORN-SAT). Jervois, positioned some 246 km northwest of Alice Springs, is not itself an ACORN-SAT site but its data have been employed to adjust Tmax records for Rabbit Flat, Alice Springs, and Boulia. This practice prompted scrutiny due to existing concerns regarding the accuracy and completeness of Jervois temperature observations.

Analysis of annual Tmax and rainfall data from 1966 to 2024 revealed significant shortcomings. The dataset was segmented into three distinct periods marked by abrupt step-changes in 1979 and 2017, with temperature increases of approximately 0.52°C and 0.83°C respectively. Notably, these changes are attributed to non-climatic factors such as missing observations, site watering, and instrument changes rather than genuine climate trends. The reported overall warming trend of 0.225°C per decade thus appears spurious, driven by data inconsistencies rather than atmospheric change. As the report states, “as no trend remained in MLR residuals, it could not be claimed that the Jervois climate has warmed due to CO₂, coalmining, electricity generation or anything else.” Rainfall exerted a significant cooling effect on Tmax, reducing temperatures by 0.294°C per 100 mm of rain, explaining over half of the Tmax variability alongside the identified step-changes.

Compounding concerns are metadata inaccuracies, including erroneous site coordinates and undocumented instrument transitions. These undermine confidence in the station’s data quality and the validity of its use in homogenisation processes. The report’s author, Dr Bill Johnston, concludes that given the “poor” quality of Jervois data, its use for homogenising ACORN-SAT sites “has no statistical or scientific merit,” and calls for the abandonment of the ACORN-SAT project in its entirety.

This detailed examination illuminates challenges in climate data homogenisation where underlying observational inconsistencies distort perceived warming trends. As climate policy and public discourse increasingly rely on precise temperature records, the integrity of such datasets is paramount. The findings here serve as a cautionary note on the risks of incorporating flawed data in climate analyses and stress the importance of rigorous, transparent data quality assessments to maintain scientific credibility in climate monitoring efforts.

Points to note.

The Report itself was developed and constructed entirely using human intelligence.  Information about the site was sought from BoM metadata, the National Archives of Australia, Google earth Pro satellite images, and Google (see for example, https://walerdatabase.online/jinka-and-jervois-stations/, and of historical interest https://cifhs.com/ntrecords/ntcensus/jervois.html). Analysis was undertaken hands-on. Use of R packages, and detection of shifts in Tmax ~ rainfall residuals using Sequential Step-change Analysis of Regime Shifts (STARS), required human judgement, which is unlikely to be replicated by AI.

Is using AI to review a complex report ethical?

It is important to acknowledge all sources of information, programs etc. used in compiling research reports. Likewise, to acknowledge the contribution of peers and colleagues. In that it is adding value to words and complex analyses that some may find daunting to read, provided it’s use is acknowledged, summarising the main points using AI-tools is no different.

With the caveat that AI-reviews be replicated across multiple tools, they also provide a rapid form of ‘peer review’, even though they ought not be thought of as actual peers. AI-reviewers also cannot step far beyond the boundaries of the report on which reviews are based. Further, if the report itself or the methodology on which it is based is faulty, while seeming authoritative and authentic, reviews generated using AI are also likely to be faulty.

It would also be cheating (i.e., academic misconduct) to take a publication or report, create summaries using AI, and combine and publish the result as one’s own work. Most universities publish guidelines on the use of AI in academic studies that set clear boundaries around what is, and what is not acceptable (e.g., https://www.teqsa.gov.au/students/artificial-intelligence-advice-students).

(18 July 2025)

Part 9. Rabbit Flat roadhouse, Northern Territory, Australia

Data Homogenisation,

Dr Bill Johnston

Former NSW Department of Natural Resources research scientist and weather observer.

scientist@bomwatch.com.au

Situated east of the Great Sandy-Tanami Desert, 450 km along the Tanami Track south from Halls Creek in Western Australia and 597 km north of Alice Springs, the Rabbit Flat roadhouse is one of the most isolated of the 112 Australian Climate Observations Reference Network – Surface Air Temperature (ACORN-SAT) sites used to monitor Australia’s climate. The site was selected by the Bureau of Meteorology (BoM) as a weather reporting station in December 1969 and useful data commenced in 1970.

The Rabbit Flat roadhouse beside the Tanami Track, is the site of a fresh-water bore east of the Tanami Desert. Maximum temperature (Tmax) is used to homogenise ACORN-SAT sites at Halls Creek, Victoria River Downs and Tennant Creek (solid lines). Dashed lines indicate stations used to homogenise Tmax at Rabbit Flat.

According to the ACORN-SAT catalogue, there was “no indication of any moves between 1969 and 1998”. As the roadhouse weather station site (BoM ID 15548) had become overgrown, an automatic weather station (AWS) was established about 200 m west on 13 November 1996 (BoM ID 15666). Daily maximum temperature (Tmax) for both sites was merged as per ACORN-SAT, summarised as annual averages, aligned with annual rainfall and analysed using BomWatch protocols.

Step-changes indicative of site changes, were detected in the merged dataset in 1986 and 2016, but not in 1997 after datasets were merged.

The magnitude of the change in 1986 (+0.87oC (±0.17sem)) is consistent with the Stevenson screen having moved from the vicinity of the roadhouse, where surrounds were possibly watered, to a more open position where conditions were generally warmer. Google Earth Pro satellite images show the up-step in 2016 (+0.89oC (±0.23sem)) resulted from erecting a 2.4m-high, chain-wire security fence, probably in 2015. Neither change could be attributed to the climate.

While adjustments for site changes are justified and necessary, it is vital that changepoints be detected objectively, and that adjustments directly match the magnitude of the effect.

ACORN-SAT V1 (to 2017) and V2.5 (to 2023) detected a significant changepoint in 1987, and another in 1997 corresponding with the merge. However, AcV2.5 failed to detect the significant changepoint in 2016. Adjustments made by AcV1 and AcV2.5 in 1987 were not the same, and in both cases were disproportionate to the rainfall-adjusted magnitude of the effect. Adjustments made by ACORN-SAT for merging the data in 1997, were also not the same.

The failure of AcV2.5 to adjust for installation of the security fence, which, as they paid for it, they must have known about, compounds previous errors, thus creating non-significant trends in homogenised data of 0.18oC/decade and 0.14oC/decade respectively.

A Google Earth Pro satellite image of the Rabbit Flat AWS on 18 August 2023. The Stevenson screen in the upper-right of the enclosure appears to be less than 4m from the 2.4m-high security fence. Fencing the site in 2015, groundcover destruction using herbicide, and other disturbances have warmed the site independently of the climate.    

Homogenisation adjustments by AcV1 and AcV2.5 in 1987 were not the same as the magnitude of the change they aimed to correct. Also, data were adjusted for a changepoint in 1997 that was not statistically significant. In addition, the significant site change in 2015 the BoM must have known about was ignored as though the effect on data reflected the climate.

The Bureau’s homogenisation methods cannot be independently replicated and it seems doubtful that statistical tests of differences with highly correlated neighbours are sufficiently sensitive to discriminate between step-change effects that are real and verifiable, and site and instrument changes that have no effect.  

ACORN-SAT also undertakes no post hoc tests to verify outcomes and check assumptions underlying the trend verses step-change model. ACORN-SAT adjustments resulted in non-significant positive trends of 0.18oC/decade for AcV1, and 0.14oC/decade for AcV2.5. In contrast, BomWatch protocols, which include verification and assumption tests, found that adjusting for step-changes and rainfall simultaneously left no residual trend that could be attributed to CO2, coalmining, electricity generation or anything else.     

Implications

The Bureau’s incessant marketing of fake record temperatures, record-hot years, heatwaves and in particular, runs of warm days, ignores that since 1970, Tmax at Rabbit Flat increased 1.76oC (±0.23sem) due to site changes alone. However, ignoring that the site had warmed independently of the climate, it was reported in March 2019 that Rabbit Flat smashed the national heat-record previously held by Marble Bar.

It was claimed that maximum temperature of 38.4oC on Tuesday 26 March 2019 ended the longest-run of 115-days of maximum temperatures over 39oC in Australia’s history. However, the Rabbit Flat site was not operating when the record of 106-days was set at Marble bar between November 1921 and March 1922.

ACORN-SAT has exceeded its use-by date. Site-summary and ACORN-SAT metadata has been poorly researched. Statistical tests of differences between comparator datasets that are not homogeneous, and faulty ACORN-SAT-site data, can find changepoints that make no difference, while not detecting those that do. As the methodology is fundamentally flawed, not scientific and cannot be replicated the ACORN-SAT project should be abandoned in its entirety.   

Two important links – find out more

First Link: The page you have just read is the cover story for the full paper. If you are stimulated to find out more, please link through to the full Report in downloadable pdf format. Photographs, diagrams, graphs and analysis will make compelling reading for those interested in the issue.

Click here to download the Report

Second Link: This link will take you to the Excel workbook containing the data used in the analysis.

Click here to download the dataset

Note: Line numbers are provided in the Report for reference. If comments are of a highly technical nature, including Bomwatch protocols and statistical procedures, please email Dr Bill Johnston directly at scientist@bomwatch.com.au.   

Reference:

Trewin B.C., 2001. Extreme temperature events in Australia. PhD thesis, School of Earth Sciences, University of Melbourne.