Ken Stewart, May 2010
(This is the first in a series of posts, and therefore will need to include a lot of background information and explanation. Future posts will not be as long.)
Until about a month ago, I thought that the analysis of climate change in Australia, and information given to the public and the government, was based on the raw temperature data.
I was wrong.
- It is based on data that has been subjectively and manually adjusted.
- It makes no allowance for Urban Heat Island (UHI) effects.
- The methodology used is not uniformly followed, or else is not as described.
- In Queensland, it produces a trend in mean temperatures that is nearly 0.2 degrees Celsius greater than the raw data does.
- It does NOT give an accurate record of Queensland temperatures over the last 100 years.
For many years I have closely followed the weather forecasts, synoptic charts, and discussions of local weather, given by the Australian Bureau of Meteorology (BOM). The BOM has a place of honour in Australia and performs a sterling service in times of cyclone, flood, bushfire, and other crises. Naturally their forecasts are taken with a grain of salt, but generally BOM is respected as an impartial scientific organisation.
In previous posts I have contrasted the treatment of data by the Goddard Institute of Space Studies (GISS) with the published data provided on the BOM website. However, the 15th March State of the Climate Report by BOM and CSIRO pricked my interest with its claims of warming in some places up to 2 degrees C in 50 years.
The Trend Map shown was derived from the Reference Climate Sites, as compared with the Temperature Trend Maps on the BOM Climate Change site, which use the Australian High Quality Climate Site dataset. The alarmist tone of the State of the Climate Report (“Australia will be hotter in coming decades” contrasts with the more measured and cautious statements BOM usually makes, such as:
“Analysis periods starting after 1970 are considered too short to calculate meaningful trend values.”
“The trend maps are a useful way to compare how the temperature has changed in different regions of Australia over time. However, they need to be interpreted with caution. Trend values have been determined from a linear (straight line) fit to the data, but the change indicated may not have been gradual. For example, a calculated trend could be due to a relatively rapid “step” change, with the remainder of the series being fairly flat (see some of the timeseries graphs)”
“In addition, the trend values calculated here using past observations should not be used to imply future rates of change.”
The temperature record which can lead to these two interpretations is the subject of this study.
I should also acknowledge janama who alerted me to the 1996 adjustments by Torok and Nicholls (1996) and provided me with quantities of information, including a copy of the Torok paper.
The Australian High Quality Climate Site Dataset
The HQ page gives some useful background information about the dataset.
“The temperature trend maps are calculated from homogeneous or “high-quality” temperature datasets developed for monitoring long-term temperature trends and variability (see maps below). Where possible, each station record in these datasets has been corrected for data “jumps” or artificial discontinuities caused by changes in observation site location, exposure, instrumentation or observation procedure. This involves identifying and correcting data problems using statistical techniques, visual checks and station history information (metadata).”
Basically, the method(s) used are as follows: candidate stations are assessed for data quality, i.e currently open with more than 80 years of data (often created by combining sites), then surrounding “reference” stations are selected for comparison. Preferably at least 10 are required that are highly correlated with the candidate station, not at risk of becoming urban (>10,000 population), within 8 degrees latitude and longitude and 500m difference in elevation- and, Torok specified, in the same climatic area e.g. not over a mountain range. In 1996, this produced 224 stations. By 2004, this had been reduced to 138, of which 39 were urban.
“ The median reference series was created by taking the median of the ranked interannual temperature differences from each suitable reference station series for each year. This series was then added to the annual mean temperature of the first year of the candidate series to create the final median reference series. Trewin (2001) uses a distance weighted mean of highly correlated reference station anomalies to create the reference series.” (Della-Marta et al)
The results of the two methods were compared. The resulting series was manually checked for discontinuities (abrupt changes between the reference series and candidate data) and subjective decisions were made whether to make adjustments and by how much, by referring to the metadata (information about station history) which is not always available. One example of metadata (Roma) is given below.
“Overall, the updated dataset includes an average of approximately six adjustments per record since 1910.” (Della-Marta et al) Every station record was adjusted.
From this dataset, temperatures for areas up to 740km away are assumed and used to create the trend maps.
The BOM homogenisation process is thus different to GISS in a number of ways:
GISS uses stations up to 1200km away; BOM uses stations up to 8 degrees latitude and longitude and 500m difference in altitude, but not from a different climate type eg over a mountain range, and not if at risk of urbanisation. No list of stations used is supplied. In other words, they use a grid that is 16 degrees x 16 degrees centred on each station. GISS say their process is automatic; the BOM adjustments are made manually after visually checking for discontinuities, then deciding whether and how much to adjust.
BOM says it does not include urban sites in formulating the HQ climate record whereas GISS does.
Urban sites are sites that have or have had at some time since 1910 a population over 10,000. However the HQ list of sites has some anomalies: Torok and Nicholls, and Della-Marta et al both listed Cairns AMO, Gladstone Radar, and Rockhampton AMO as urban sites, but on the BOM HQ website they are shown as rural sites, and thus are included as reference sites and in producing the official climate record. Moreover Innisfail is listed as urban, but Innisfail has a population (2006) of 8,262. BOM’s response to these anomalies is:
“Urbanisation does not simply relate to population. It also can relate to the nature of the site’s surrounds.”
and
“Each of these is well outside of the town centre. We find no evidence in our analyses urban warming at these sites – Cairns and Rockhampton are at the airports and Gladstone is on a high hill outside of the urban influence.”
Actually, Gladstone Radar is approximately 150m from houses to the east and sheds and industrial sites to the south east. It is in fact surrounded by urban areas in all directions. BOM’s response is in spite of Torok et al’s 2001 investigation of UHI which found that :
“The urbanrural temperature difference was found to increase with increasing population via the equation
DeltaTu-r(max) = 1.42 log(population)-2.09”
(Torok, Morris, Skinner, and Plummer. “Urban heat island features of southeast Australian towns” in Australian Meteorological Magazine 50 (2001): 1-13)
Moreover, the papers cited specifically state that sites are urban and therefore not to be included, if they have had a population over 10,000 at some time in the last 100 years. This applies to the records from Cairns, Rockhampton, and Gladstone Post offices.
Therefore, the methods used by Torok et al and Della-Marta et al are not uniformly applied by BOM.
A further potential difficulty is that in both processes (1996 and 2004) reference stations i.e. sites used in the adjusting process, could have as little as 10 years of monthly data.
Another statement from Della-Marta et al (2004) raised my interest:
Despite the suitability of the dataset for national and regional-scale analyses, any individual station record within the dataset should still be treated with caution. The subjectivity inherently involved in the homogeneity process means that two different adjustment schemes will not necessarily result in the same homogeneity adjustments being calculated for individual records. However, if the overall biases of the different approaches are neutral then spatial averages should be highly consistent across a large number of homogenised records. Also, even though a subjective assessment of the likelihood of urban contamination within these records has been made, further work needs to be undertaken to better understand the relative contribution of urban warming to the temperature records of small Australian towns. (My emphasis)
Not having access to the list of stations, the metadata, the software used, or the expertise of BOM, the average citizen would normally accept the published results as they stand. However I wanted to have a closer look. Surely the results of any adjustments should be easy to compare with the previous record. In theory, as a BOM spokesperson replied in response to my query:
“On the issue of adjustments you find that these have a near zero impact on the all Australian temperature because these tend to be equally positive and negative across the network (as would be expected given they are adjustments for random station changes).”
This statement will be the yardstick for this study, starting with Queensland. As the survey progresses to cover all of Australia this will be updated.
Analysis
I downloaded annual mean maxima and minima for each of the sites from BOM Climate Data Online, calculated annual means and plotted these. Frequently, two or three stations (some closed) were needed for the entire record from 1910-2009, and even then there sometimes were gaps in the record- e.g. from 1957 to 1964 many stations’ data has not been digitised. (But 8 years of missing data is nothing- Amberley and Atherton have over 30 years that have been “filled in” to create the High Quality series). I also downloaded the annual means from the High Quality page, and plotted them. I then added a linear trend for each.
Because there are 100 non-urban sites across Australia, I plan to analyse the network in stages, beginning with Queensland. Future posts will look at the Northern Territory and northern Western Australia, and then other states.
Here are the BOM 1910-2009 maps and timeseries charts for Queensland:
From 1910 to 2009, the linear trend is 1 degree Celsius per 100 years.
This is the map of Queensland High Quality sites. I have added the trend per 100 years for the raw data (green) and the adjusted data (blue). Note that Urban sites are not included in developing time series graphs or trend maps.
Note: There are 8 east coastal sites. Only 4 sites are semi coastal, i.e. between the east coast and the Great Dividing Range (Amberley AMO, Gayndah, Rockhampton AMO, Charters Towers). The remaining 16 are inland of the Dividing Range, and Normanton, which is close to the Gulf.
Results:
Graphs for all sites in Queensland are in the Appendix in alphabetical order .
This is a table showing for each site the unadjusted trend, the adjusted trend, the difference, and the mean for each series.
Note: The adjusted trends average at 0.95C. Round to 1 degree C per 10 years, which is that shown on the timeseries graph.
The raw data trend is 0.78C- a not insignificant difference. The total difference is +4.75C.
The above table plotted:
Plot the frequency of the magnitude of adjustments in 0.2 increments:
- Barcaldine is an obvious cooling outlier.
- There is an obvious warming bias.
- 11 sites have a warming of 0.4C or more.
What about error?
Yes, my trend estimates must have a possibility of error- say + or – 0.1 degrees for raw data and for adjusted data. The total error could be +/- 0.2 C. There are 5 sites with cooling > 0.2C, and 12 sites with warming > 0.2 C.
I am particularly interested in the outliers- Cooktown , Ayr DPI, Barcaldine, Longreach, Goondiwindi, Sandy Cape Lighthouse, Roma.
I have made a manual “splice” of the raw data in a number of these. I make no claims of superiority- these splices are not better, but they show an alternative interpretation.
Here are graphs for each of these “interesting” sites:
Cooktown has had 3 stations- the Post Office, Mission Strip, and the Aerodrome. The HQ site is Mission Strip, so they have obviously spliced the three, which I did. I also compared with the nearest coastal town (not over the range), Port Douglas:
Cooktown Mission Strip adjusted: from flat to 0.9 C warming. 
In reply to my query about this apparent inconsistency, BOM state:
“You are merging data from different stations. That is why your trends are inconsistent with the data.”
That’s the explanation!
Ayr DPI. Ayr had a station at the Shire Council until 1984 and the DPI research station just out of town from 1952. I spliced the two. Compare with HQ.
Once again, data up to 1952 has been seriously adjusted down.
Barcaldine– huge cooling adjustment!
But Longreach, just 110km west, has a huge warming adjustment!
First, Longreach’s record- Post Office and Aerodrome.
Now, spliced compared to HQ:
Compare Longreach with Barcaldine and nearby Isisford:
Why do Longreach and Barcaldine differ by so much? It looks like they have been averaged. My money is on the raw Longreach record.
Goondiwindi raw:
Goondiwindi splice vs HQ: from 0.3 to 0.9C/100 years!
This needs explaining, as other stations don’t show this much warming:
Moree: what warming?
Mungindi and Warialda about 120km SW and S: 0.8 and 0.5C
Cooler earlier, warmer later.
Sandy Cape Lighthouse: from 0.45 to 1.2C warming.
But the nearby stations of Maryborough, Gympie, and Childers are closer to the raw data!
Maryborough:
Gympie:
Childers:
BOM’s response: “I cannot see the data you are describing.”
And worst of all, Roma:-
Roma raw:
Roma HQ compared to a splice: 0.6 to 1.5C
Nearby towns (see Appendix) of Surat, Mitchell, Injune, Miles, and Dalby don’t show this: Mitchell’s trend is at most 0.9C. Surat and Injune (from 1938 and 1968 respectively) mirror Roma’s temperatures fairly well, a little above and below. Miles shows a trend in raw temperatures of 0.5C, and Dalby (urban and therefore not included) 0.55C. And to the west, Charleville’s adjusted data shows less than 0.6C.
So why the huge adjustment to Roma? BOM’s response:
“Here is the raw and corrected data for Roma. The site has moved a number of times and the screen has changed.
Here is the list of changes to the site which are related to the corrections. The main change relates to a major site move which accounts for a large part of the inhomogeneity.
METADATA Summary
Roma,043091,043030
10/1897: Stevenson screen supplied.
02/1908: First correspondence.
12/1908: Screen moved due to building.
11/1912: New large screen.
10/1916: Poor observations during the 1910s.
11/1929: Screen moved 50 feet due to new shed and incinerator.
10/1941: Screen needs to be repaired.
10/1962: Site has deteriorated due to bitumen and buildings.
01/1971: Temporary site during building.
09/1972: Move to new PO.
09/1983: Site moved 50 m south to better site.
01/1992: Move to composite site.
04/1994: Bubbles in Min Thermometer
02/1997: AWS (Almos) Installed (28/02/1997)
02/1997: SITE MOVE- Short distance West (@8/02/1997)
1997-2002: Routine AWS upgrades and maintenance.”
Sorry, I’m not convinced.
Conclusion:
The High Quality data for Queensland shows a warming bias of nearly 0.2 degrees Celsius per 100 years. Comparison of the HQ data for these sites with the raw data shows unexplained inconsistencies in a number of cases. Leaving out the adjustments of the sites with the most glaring inconsistencies brings the average HQ trend back to the raw data trend of about 0.8 degrees C /100 years. Furthermore, it is based on data that has been subjectively and manually adjusted, and it makes no allowance for Urban Heat Island (UHI) effects.
The methodology used either is not uniformly followed, or else is not used as described in the supporting papers, and appears to have flaws in quality control.
It does NOT give an accurate record of Queensland temperatures over the last 100 years. However, the trend maps and time series graphs are based on it, so it is no wonder that they use cautious language.
As I analyse data from other states, I will test BOM’s statement that
“On the issue of adjustments you find that these have a near zero impact on the all Australian temperature because these tend to be equally positive and negative across the network (as would be expected given they are adjustments for random station changes).”
As far as Queensland is concerned, the only “High Quality” in the Australian High Quality Climate Site data is in the title.
Appendix
Here are the plots of the remaining HQ stations.
Amberley Air Force Base– almost flat, but 30 years data filled in. A pity BOM doesn’t trust the RAAF before 1981.
Atherton – 29 years of data filled in by using Herberton 10km away, over a range.
Bollon– a tad warmer, and nearly every data point adjusted:
Boulia: not warming enough already?
Bowen– PO and Airport:
From this they make:
Cairns– I have already discussed in Climate Confusion in Cairns- a town PO and an airport given an extra warming boost:
Camooweal – not much change:
Cardwell – a little cooler (they don’t like 1971 to 1985, or the 30s and 40s)
Cape Moreton Lighthouse: not warm enough, so make it cooler before 1960:
Charleville –reasonable. Note the 1950s, similar to Cunnamulla to the south:
Cunnamulla – little change
Dalby raw and HQ: (urban, therefore not included by BOM)
Gayndah– about the same. I went to high school here in 68 & 69- a cold place in winter!
Georgetown 0.4 cooler
Gladstone– I previously discussed Gladstone in “Cherries, anyone? Another data trick in Australia”. Post Office, Airport, and Radar raw data:
Gladstone Radar HQ:
I still like my splice better!
Hughenden: a little cooler
Mitchell, Surat, Injune (not HQ but close to Roma)
Miles raw data:
Miles HQ (warming increased): (yet look at Roma and Dalby!)
Normanton: little change, but from mid 50s to 2000 a downwards correction.
Palmerville: no change
Richmond: ditto
Rockhampton: 
Rocky spliced:
References:
Della-Marta, P., Collins, D., Braganza, K. “Updating Australia’s high-quality annual temperature dataset” Australian Meteorological Magazine Vol. 53, no. 2, June 2004
Jones, D.A. and Trewin, B.C. 2000. The spatial structure of monthly temperature anomalies over Australia. Australian Meteorological Magazine, 49, 261-276.
Jones, D.A. and Trewin, B.C. 2002. On the adequacy of digitised historical Australian daily temperature data for climate monitoring. Australian Meteorological Magazine, 51, 237-250.
Torok, S.J. and Nicholls, N. 1996. A historical annual temperature dataset for Australia. Australian Meteorological Magazine, 45, 251-260.
Torok, S, Morris, C, Skinner, C, and Plummer, N. “Urban heat island features of southeast Australian towns” Australian Meteorological Magazine Vol 50, 2001.
http://www.bom.gov.au/climate/change/about/temp_trendmaps.shtml
http://www.bom.gov.au/cgi-bin/climate/change/timeseries.cgi
http://www.bom.gov.au/inside/eiab/State-of-climate-2010-updated.pdf
The Northern Territory has a warming trend of 1.1 degrees Celsius per 100 years, so a lot depends on those two sites, because as you can see from the table below, the trend has been increased from 0.98 degrees to 1.13 degrees per 100 years.
kenskingdom 
