Temperature change in Australia
The rates of temperature change in Australia at rural and non-rural weather stations are compared. Over those periods in which there were increasing temperatures, the rural stations appear to have warmed at about 60-70% of the warming rate of the non-rural stations.
Data from the Australian Bureau of Meteorology (BOM) is used, and is assumed to be sufficiently accurate for the purpose. It is not known whether any original measurements have been adjusted in the BOM records.
The data values downloaded from BOM are monthly mean max and monthly mean min. These are used in preference to monthly max and monthly min, as being more representative of each month’s temperature.
47 weather stations were active from 1940 to 2000, ie. from 1940 or earlier to 2000 or later. Of these, 10 were rural.
|Station Number||Station Name||Latitude||Longitude||Start Year||Last Year||Is Rural?|
|4020||MARBLE BAR COMPARISON||-21.1756||119.7497||1901||2006||Yes|
|12071||SALMON GUMS RES.STN.||-32.9869||121.6239||1932||2009||Yes|
|29004||BURKETOWN POST OFFICE||-17.7425||139.5475||1890||2009||No|
|30018||GEORGETOWN POST OFFICE||-18.2922||143.5483||1894||2007||No|
|30045||RICHMOND POST OFFICE||-20.7289||143.1425||1893||2009||No|
|33047||TE KOWAI EXP STN||-21.1642||149.1192||1908||2009||Yes|
|39039||GAYNDAH POST OFFICE||-25.6258||151.6094||1893||2009||No|
|46037||TIBOOBURRA POST OFFICE||-29.4345||142.0098||1910||2009||No|
|66037||SYDNEY AIRPORT AMO||-33.9411||151.1725||1939||2009||No|
|66062||SYDNEY (OBSERVATORY HILL)||-33.8607||151.205||1859||2009||No|
|74128||DENILIQUIN (WILKINSON ST)||-35.5269||144.952||1858||2003||No|
|86071||MELBOURNE REGIONAL OFFICE||-37.8075||144.97||1855||2009||No|
|90015||CAPE OTWAY LIGHTHOUSE||-38.8556||143.5128||1864||2009||Yes|
|91057||LOW HEAD (COMPARISON)||-41.0567||146.7883||1895||2001||Yes|
|91104||LAUNCESTON AIRPORT COMPARISON||-41.5397||147.2033||1939||2009||No|
|94029||HOBART (ELLERSLIE ROAD)||-42.8897||147.3278||1882||2009||No|
|200288||NORFOLK ISLAND AERO||-29.0389||167.9408||1939||2009||No|
Whether each station was rural or non-rural was determined using Google Earth. The decisions were necessarily subjective. The Google Earth views of the stations are shown in RSelectedStationsGoogleMaps (10.7mb PDF)
Note : Station 9581 Mount Barker is obscured by cloud in the Google Earth view. The population of Mount Barker is over 4,000, and the weather station appears to be in or very near the built-up area, so it is classified as Non-Rural.
The monthly mean max and monthly mean min temperatures for all these stations were downloaded from the BOM website http://www.bom.gov.au/. The temperatures were downloaded in January and February 2011, but unfortunately the December 2010 temperatures for some of the stations had not then been posted. For those stations, data was in fact complete to November 2010. Note that “Last Year” for stations that are still open is given as 2009, and that 10 of the stations closed between 2000 and 2009. A further 3 stations (29004, 39039, 91104) appear to have closed in 2009.
The data as downloaded is in http://members.westnet.com.au/jonas1/QMeanTemps.pdf
Product code “IDCJAC0002″ is Monthly Mean Maximum Temperature, and “IDCJAC0004″ is Monthly Mean Minimum Temperature.
Temperature trends (least-squares linear fit) over various periods were calculated separately for each calendar month, for each station, using only the monthly temperatures that were given. There was no attempt to fill in any missing data.
For each period, only the stations active for the whole of the period were used (there could be some missing data, but the test was that there was data for the first year of the period or for an earlier year and that there was data for the last year of the period or for a later year).
The overall trend for each period was then taken as the average of all trends for that period – ie, the average across all calendar months and all active stations.
The reasons for using this method are:
- If an attempt was made to fill in missing data, then there would be a risk that the filling-in method used could introduce errors or biases.
- Each calculated trend is for one calendar month only (eg. all Januaries), so has no seasonal bias.
- If trends were calculated on annual averages, instead of by calendar month, then any year with a missing month’s data could not be used, because that would introduce a selection bias.
- If trends were calculated on all months, instead of by calendar month, then the least-squares algorithm would bias the result towards the hottest and coldest calendar months. The previous issue – a missing month’s data affecting a whole year – also applies.
- Each calculated trend is for one station only. If station data was averaged before calculating the trend, then it would not be possible to use any month for which any station’s data was missing, because it would introduce a selection bias.
- The station / calendar month trends for each period were only calculated for stations active in that calendar month over the whole of the period (individual months’ data may still be missing). Thus when trends are averaged for any given period, there is no station selection bias. Note that the number of selected stations can be different for different periods.
It is possible that there may be a geographical bias in the data, ie. the stations used may have different geographical distributions for different periods, or that the geographical distribution of stations is not representative of the region. It is also possible that missing data could affect results, especially if missing values are clustered (eg. same calendar month missing over consecutive years, for a given station).
The full results, where there were at least 5 stations used in each trend average, are given in this table:
and are as follows:
Trends are given in deg C per year. So, for example, the first number, 0.0036, for the period 1910-2000, means that the average trend of Monthly Mean Max temperatures from 1910 to 2000 for Non-Rural stations was 0.0036 deg C per year. (All results are given to 4 decimal places, so there may be rounding errors).
For stations whose Dec 2010 data was missing, the average trend shown to 2010 is the average of the Jan-Nov trends. Stations are only included for periods over which they have both “Max” and “Min” temperature data.
These figures indicate that over those periods in which there were increasing temperatures, the rural stations only warmed at about 60-70% of the warming rate of the non-rural stations.
There are some curious numbers in the results, which could deserve further analysis – for example, the rural trends for 1980-2000 and 1990-2000 are much lower than the non-rural, yet for 2000-2010 they are about the same as the non-rural. Also, there is no warming trend from 1910-1940, a period of quite rapid global warming (see, eg.,
Where there were at least 2 stations used in each trend average, the results are given in
and all trend averages are given in
All of the individual station / calendar-month trends (rounded to 4 dec places) are given in this document:
This too could deserve further analysis, eg. there may be significant summer / winter trend differences.
For further information and discussion, see
This document includes:
- List of the 47 weather stations.
- Google Earth maps used for Rural/Non-Rural identification. [link]
- BOM website (data source). [link]
- Source data. [link]
- Individual station / calendar-month trends. [link]