IntroductionAcross the world and particularly in Australia no weather station sites have stayed the same. Therefore its important that effects caused by site and instrument changes are not attributed to the climate.

DiscussionHistoric temperature data collected to monitor the weather were from thermometers of unknown quality exposed in the shade, hung on the wall of south-facing verandas, in Glaisher stands which were manually turned away from the sun, in ‘thermometer houses’ of varying design, inside iron-roofed Stevenson screens and in some cases inside buildings.

For most historic datasets like Sydney Observatory, Cape Leeuwin, Cape Otway; and post offices like Port Hedland, Bourke, Darwin, Barcaldine and Mildura, details of site changes and the conditions under which observations were made are scant or unavailable. For many it’s not known when Stevenson screens were installed or the state of the instruments or if the vicinity was watered. Homogenisation aims to adjust for non-climate effects, however as explained in this note, methods used by the Bureau are unscientific, biased and should be abandoned.

There is no evidence in their reports that independent peer-reviewers [1] investigated datasets or looked for problems in homogenisation methods, or in the way data are observed or the usefulness of historic data for tracking trends or changes in the climate.

The three obvious sources of bias ignored by peer reviewers are:

Non-existent or faulty metadata (data about data) is used as primary evidence of site changes and as noted by Della-Marta et al. (2004) [2] (p. 85) “The decision of whether or not to correct for a potential inhomogeneity is often .. subjective”. Subjective assignment of changepoints allows the process to be guided by pre-determined outcomes.

Comparative methods whereby target site data are adjusted relative to data for up to 10 correlated comparators (whose data may not be homogeneous) lacks rigour and is also biased. As correlated comparators likely embed similar faults the adjustment process coerces site-related step-changes into a trend.

Residual trends that result from arbitrarily identified changepoints and dodgy adjustments are passed-off as being due to the climate by default.

Of particular concern is that by ignoring network-wide changes that happened in the 1950s, 1970s and more recently, homogenised datasets embed trends and changes that have nothing to do with the climate.