Posts Tagged ‘Australia’

How Accurate Is Australia’s Temperature Record? Part 2

January 19, 2021

In my last post I showed that Australia wide the Tmax ~ rainfall relationship has remained constant for the past 110 years (as it should) but Tmax reported in the Acorn dataset has increased by more than 1.5 degrees Celsius relative to rainfall.  Consequently, the ACORN-SAT temperature dataset is an unreliable record of Australia’s maximum temperatures.

Of course there are other aspects of climate besides rainfall.   In this post I will compare annual ACORN-SAT Tmax data with:

Rainfall

Sea Surface Temperatures (SST)

The Southern Annular Mode (SAM)

Cloudiness

Evaporation

all for the Australian region.

I have sourced all data from the Bureau of Meteorology’s Climate time Series pages

except for SAM data from Marshall, Gareth & National Center for Atmospheric Research Staff (Eds). Last modified 19 Mar 2018. “The Climate Data Guide: Marshall Southern Annular Mode (SAM) Index (Station-based).” 

Tmax, Rainfall, and SST data are from 1910; SAM and Daytime Cloud from 1957, and Pan Evaporation from 1975.  Cloud observations apparently ceased after 2014, and Evaporation after 2017, possibly because of staffing cuts.

Because Pan Evaporation data are only available from 1975 and are reported as anomalies from 1975 to 2004 means, I have recalculated Tmax, Rainfall, SST, SAM, and Daytime Cloud anomalies for the same period so all data are directly comparable.

As in the previous post, I have calculated decadal averages for all indicators to show broad long term climate changes.  Decadal averages show how indicators perform over longer periods.  Each point in the figures below shows the average of the 10 years to that point.  This can then indicate times of sudden shifts or questionable data. (For example in Figure 1 SAM (the green line) makes a sudden jump in 2015.  Was this a climate shift or a data problem?)

Figure 1 shows the 10 year means for all climate indicators.  I have scaled Rain and SST to match Tmax at 2019, Cloud and SAM to match Tmax at 1966, and Evaporation to match Tmax in 1984.  Rain and Cloud are inverted as they have an inverse relationship with temperature.

Figure 1:  10 Year Means of Climate Indicators

Tmax has stayed close to SST until 2001.  Clearly Tmax has increased far more than any of the others, especially since 2001.

The next plots show the difference between decadal averages of Tmax and the other indicators.  Zero difference means an excellent relationship with Tmax.

Figure 2:  Difference: 10 Year Means of Tmax minus Rain and SSTs.

Starting from 1919 (zero difference), Rainfall is close to Tmax until 1957, after which Tmax takes off until it is 1.6 degrees Celsius greater than expected in the 10 years to 2020.  Tmax diverges from SST values in 2001 and in 2020 is 0.7 degrees greater than expected.

In Figure 3, Rain, SST, SAM, and Cloudiness are scaled to match Tmax at 1966.

Figure 3:  Difference: 10 Year Means of Tmax minus Rain, SST, SAM, and Cloud

Figure 3 shows how closely Rain and Cloud are related: differences from Tmax are almost identical.  Compared with 1966, Tmax is 1.3 degrees more than rainfall would suggest in the 10 years to 2020.  SST and the SAM index are less different from Tmax but Tmax divergence is still clear.  You may notice that Tmax differences from all climate indicators seem to change at similar times, apart from SAM in 2015.

In Figure 4, all indicators are scaled to match Tmax at 1984.

Figure 4:  Difference: 10 Year Means of Tmax minus Rain, SST, SAM, Cloud and Evaporation

Differences increase rapidly after 2001, so in Figure 5 indicators match Tmax at 2001.

Figure 5:  Difference: 10 Year Means of Tmax minus Rain, SST, SAM, Cloud and Evaporation

There appears to be a problem with SAM in 2015, and it’s a shame that the BOM have discontinued Cloud and Evaporation observations.  In the last 20 years, it is obvious that Tmax has diverged from other indicators.

Conclusion:

All factors- Rainfall, SAM, SST, Clouds, and Pan Evaporation- point to a clear divergence of temperature nationwide, especially in the last 20 years.  In other words, ACORN-SAT, our official record of temperatures, is unreliable.

How Accurate Is Australia’s Temperature Record? Part 1

January 7, 2021

In my last post I showed that maximum temperature (Tmax) as reported by ACORN-SAT (Australian Climate Observations Reference Network-Surface Air Temperature) appears to be responsible for the growing divergence of the difference between Tmax and tropospheric temperatures from Australia’s rainfall.

In this post I show how Tmax is related to rainfall, and show that while this relationship holds for discrete periods throughout the last 110 years, Tmax has apparently diverged from what we would expect.  In other words, the Acorn Tmax record is faulty and unreliable.

For much of this analysis I am indebted to Dr Bill Johnston who has posted a number of papers at Bomwatch using the relationship between Tmax and rainfall.

At any land based location annual maximum temperature varies with rainfall: wet years are cooler, dry years are warmer.  More rainfall (with accompanying clouds that reflect solar radiation) brings cooler air to the ground; provides more moisture in the air, streams, waterholes, and the soil which cools by evaporation; causes vegetation to grow, the extra vegetation shading the ground and retaining moisture, with transpiration providing further cooling; and in moist conditions deep convective overturning moves vast amounts of water and heat high into the troposphere- especially evident in thunderstorms.  Less rainfall means the opposite: more solar radiation reaches the ground with fewer clouds and less vegetation; there is less moisture available to evaporate; less vegetation growth and transpiration; and much less heat is transferred to the troposphere through convective overturning.

While more rainfall than the landscape can hold results in runoff in rivers and streams, thus removing some moisture from the immediate area, this affects large regions only in tropical coastal catchments- the Kimberleys, the Gulf rivers, the Burdekin and Fitzroy.  Across the bulk of Australia there is very little discharge of water to the oceans.  In the Murray-Darling Basin, on average less than 0.005% of rainfall is discharged from the Murray mouth. (BOM rainfall data and 1891-1985 discharge data from Simpson et al (1993))

This temperature ~ rainfall relationship is particularly evident in desert areas far from any marine influence.  Alice Springs provides a good example.  Figure 1 shows how annual maximum temperatures at Alice Springs Airport vary with rainfall since 1997.  Data are from ACORN.

Fig. 1: Tmax and Rainfall, Alice Springs

The slope of the trend line shows that for every extra millimetre of rain, Tmax falls by 0.0047 of a degree Celsius, which is about half a degree less for every 100 mm.  The R-squared value shows that there is a good fit for the data- 79% of temperature change is due to rainfall.

I said above that this relationship holds for land locations.  An island, with a little land surrounded by water, is mostly affected by sea temperature and wind direction.  Locations near the coast are also affected by marine influence.  At Amberley in south-east Queensland daily maximum temperature can be moderated by the time of arrival of a sea breeze or whether it arrives at all.  (Site changes also can change Tmax recorded.)

Fig. 2: Tmax and Rainfall, Amberley

Further inland, the relationship is strong: at Bathurst, there is 0.4C temperature variation per 100mm of rainfall and 61% of temperature change is due to rainfall alone.

Fig. 3: Tmax and Rainfall, Bathurst

The BOM has sophisticated algorithms for area averaging temperature and rainfall across Australia and provide national climate records back to 1900 for rainfall and 1910 for maxima.  Averaged across Australia individual station idiosyncrasies are submerged so that the 1997 to 2019 relationship between Tmax and rainfall is very strong (and similar to that of Alice Springs):

Fig. 4: Tmax and Rainfall, Australia 1997-2019

However, the relationship is not strong throughout the whole record:

Fig. 5: Tmax and Rainfall, Australia 1910-2019

The relationship from 1910 to 2019 is poor.

In the next figure I compare the Tmax – rainfall relationships for the first 10 years of the record with the last 10 years.

Fig. 6: Tmax and Rainfall, Australia, first and last decades

The trendlines are almost exactly parallel, with tight fits, showing strong relationships 100 years apart- but the trendline for 2010 to 2019 is about 1.7 degrees above that for 1910 – 1919.  How can that be?

It is possible to compare rainfall and temperature throughout the last 110 years.  In the next figure, rainfall is inverted and scaled down so as to match Tmax at 1910.

Fig. 7: Tmax and Inverted Scaled Rain, Australia

Running 10 year means allow us to see long term patterns of rainfall and temperature more easily:

Fig. 8: Tmax and Inverted Scaled Rain, Decadal Means, Australia

Rainfall has increased over the last 110 years (despite what you might hear in the media), and so apparently have maximum temperatures.  In the above figures Tmax and rainfall track roughly together until the mid-1950s, then Tmax takes off.

I calculated an “index” of temperature ~ rainfall variation by subtracting scaled, inverted rainfall from Tmax, commencing at zero in 1919.  This allows us to identify when temperature appears to diverge markedly from inverted rainfall:

Fig. 9: Index of Temperature ~ Rainfall Variation: Tmax minus Inverted Scaled Rain, Decadal Means, Australia

There is a small increase from the mid-1950s, but the really large divergence commences in the 1970s, with the decade from 1973 to 1982 about 0.6 units above the decade to 1972.  The index decreases to 1995, then there is another steep increase to 2007, and a final surge to 2019.

This index alone shows how poorly the official temperature record represents the temperature of the past.

 While there are other times, in the next figures I compare four periods: 1910 to 1972, 1973 to 1995, 1996 to 2007, and 2008 to 2019.  Here I use annual data.

Fig. 10: Tmax and Rainfall, Australia, four periods

Again, trendlines are almost parallel with similar slopes, showing that the Tmax ~ rainfall relationship is fairly constant for all periods- (about 0.5C per 100mm after 1995 and about 0.4C per 100mm before 1995).  However, the lines are separated.  Temperature for each later period is higher than the ones preceding, such that the temperature recorded now is about 1.5 degrees Celsius higher than it would have been for similar rainfall before 1973. And rainfall has increased in that time.

Global Warming Enthusiasts and apologists for the BOM will claim that these breaks between separate periods are real and caused by changes in circulation patterns due to climate change- in particular the Southern Annular Mode.  That will be the subject of Part 2.

Whatever the reasons, Australia wide the Tmax ~ rainfall relationship has remained constant for the past 110 years (as it should) but the temperatures reported in the Acorn dataset have increased by more than 1.5 degrees Celsius relative to rainfall.

Conclusion:

The ACORN-SAT temperature dataset is an unreliable record of Australia’s maximum temperatures.

Surface and Satellite Temperatures: 2020 Update

December 19, 2020

What’s gone wrong?

In November 2015 in my post “Why are Surface and Satellite temperatures Different?” and two follow up posts I showed that the difference is very largely due to rainfall.  You are urged to read these posts in full.

I repeat a key paragraph:

Firstly, surface temperatures are supposed to be different from atmospheric temperatures. Both are useful, both have limitations. The TLT is a metric of the temperature of the bulk of the atmosphere from the surface to several kilometres above the whole continent, in the realm of the greenhouse gases- useful for analysing any greenhouse signals and regional and global climate change. Surface temperature is a metric of temperature 1.5 metres above the ground at 104 ACORN-SAT locations around Australia, area averaged across the continent- useful for describing and predicting weather conditions as they relate to such things as human comfort, crop and stock needs, and bushfire behaviour.

Here are three plots from my 2015 post.

Fig.1:  Tmax and Scaled, Inverted Rain (from Figure 7 from my 2015 post)

Dry periods are hotter, wet periods are cooler.

Fig. 2:  Surface maxima minus atmospheric temperatures and inverted rain (Figure 10 from my 2015 post)

Fig. 3:  Temperature difference compared with rainfall (from Figure 12)

The difference between Australian surface and satellite temperatures was very largely explained by rainfall. However, after five more years of satellite and surface data there is a problem (and I thank Chris Gillham for alerting me to this.)

Fig. 4:  Surface maxima minus atmospheric temperatures and inverted rain

Since about 2013 the difference between surface Tmax and satellite data has visibly increased above rainfall.

Now I have a confession to make.

In previous analyses I used running 12 month means in calculating correlation.  This can lead to inaccuracy as the means can be highly auto-correlated.  From now on I will use annual data, either with calendar years or, as in this post, annual means from December to November (so that summer months and most of the northern Wet season are included in the one datapoint).

I downloaded data from:

Monthly maxima

Monthly rainfall

Temperature of the Lower Troposphere- Australia Land

As with my 2015 post, I have recalculated rainfall and maxima from 1981-2010 means to match UAH.

In the past five years there have been changes:  the Australian maximum temperature record is now based on ACORN-SAT Version 2 instead of Version 1, including new adjustments and some station changes.  No doubt UAH has been tweaked a little as well.

However correlation between the difference between the surface maxima as recorded by Acorn and temperature of the lower troposphere (TLT) as recorded by UAH, and rainfall, has decreased.

Fig. 5:  Temperature difference compared with rainfall

The close connection between the temperature differences and rainfall became broken from about 2005, as can be seen in Figure 4.  Another step up occurred in 2013.

So there appear to be three distinct periods: 1979 to 2004, 2005 to 2012, and after 2013.  Plotting temperature differences against rainfall allows us to compare each period.

 Fig. 6:  Temperature difference compared with rainfall

From 1979 to 2004 and from 2005 to 2012 slopes are identical at 0.4 degrees lower temperature for each 10 mm of rain, with 76% and 93% of temperature variance explained by rainfall. The trend lines are parallel but offset by 0.26 degrees indicating either atmospheric temperatures have reduced or surface maxima have increased in the middle period.  From 2013 the relationship is different with closer to 0.5 degrees lower temperature per 10mm of rainfall, with rainfall explaining 78% of the variance.  Again, the offset shows either UAH has suddenly decreased or Acorn has suddenly increased.

Conclusion:  Something has gone wrong with the relationship between rainfall and temperature in Australia.  In recent years, and certainly since 2013, the surface- atmospheric temperature difference has rapidly increased relative to rainfall.  That should not have happened.

My suspicion is that Acorn’s maxima are to blame.   Figure 1 showed Acorn appeared to step up relative to rainfall in 2001 or 2002, or perhaps earlier in 1997, and again in 2013.  There can be no meteorological explanation for this.

The accuracy, and therefore usefulness, of the ACORN-SAT adjusted temperature record will be the topic of my next post.

Stay tuned.

Acorn Mish-Mash Part 2: Scone

December 13, 2020

In Part 1 we saw that Scone in NSW has the fastest increase in 120 month mean maximum temperatures of all 112 Acorn stations.  The Station Catalogue shows a recent photo of the site with long grass at least 60cm high surrounding the screen- not a very good advertisement for compliance with siting specifications.

Fig. 1:  BOM photograph of Scone site

However the Metadata for this site reveals how much the site has changed.  Before 2005 the screen was close to the runway and a service road, and there was considerable earthworks nearby in 2001.  By April 2005 the screen had been moved to its current location.  In 2012 the grass was 60cm high as in the above photo, and was whipper-snipped during the annual inspection.  In 2015 and 2019 the grass around the instruments was “sparse” as weed control had been used i.e. it had been sprayed out with herbicide.  Temperature data for the airport may be questionable based solely on site information.

The Acorn record has been created by merging data from 01/01/1995 to 31/12/1995 from the present site at the airport with that of a Soil Conservation Research Station (SCS) 10 km away from 1965 to 1994.

Data before 1975 were adjusted downwards because of a change or repair to the screen.

Fig. 2:  Adjustments to annual data at Scone

This resulted in an increase in trend of +0.43C per 100 years.

Fig. 3:  Scone raw and Acorn annual data

However, comparison with the average of the Bureau’s nominated neighbouring stations used to make this adjustment shows the adjustment was much too great.  While the raw record from 1965 to 1973 shows Scone warming 0.29C per 100 years faster than the neighbours, the Acorn record is warming at 1.46C per 100 years- much faster than the neighbours.

Fig. 4:  Difference between Scone and average of neighbours, 1965 – 1973

While that alone is enough to cast doubt on the Acorn adjustments, an analysis of the relationship between maxima and rainfall shows that little reliance can be placed on temperature data before 1974, and after 1995.

At every well maintained site there is a relationship between maximum temperature and rainfall: periods of dry weather are hotter and periods of wet weather are cooler, because of the effects of more or less cloud cover, evaporation and transpiration.  (Wind direction will also have an influence, especially in dry seasons.)  At a well maintained station much more than half of temperature variation is due to rainfall. Therefore, if this relationship varies markedly we can deduce that either temperature or rainfall data are questionable.  This is shown by Dr Bill Johnston at his website, BomWatch, which I urge you to visit, and my analysis is loosely based on his methods.

I calculated 12 month running means of temperature and rainfall for the Airport and the Soil Conservation (SCS) sites.  Figures 5 to 7 show 12 month average temperature plotted against 12 month average rainfall for the three periods, 1965 – 1973 (in which Acorn temperatures are adjusted), 1974 – 1994 (when Acorn and raw are the same), and 1995 – 2018 (when the temperature record switches from the SCS to the airport).

Figure 5:  Scone adjusted maxima plotted against local rainfall

That is a very poor relationship: either temperature data or rainfall data are unreliable.

Figure 6:  Scone unadjusted maxima plotted against local rainfall

Here, more than half of temperature variation can be explained by rainfall.  It is not brilliant, but much better than what comes before and after.

Figure 7:  Scone Airport maxima plotted against local rainfall

While not as bad as pre-1974, less than 30% of temperature variation is explained by rainfall.  Either temperature or rainfall data, or both, are dubious.  Considering the site history and varying vegetation, this is not surprising.

It is unlikely that Acorn is a true record of temperatures at this location. Scone Acorn data are not reliable and should not be included in regional and national climate analysis.

Acorn Mishmash- Part 1: They can’t all be right

November 23, 2020

The Bureau of Meteorology (BOM) produces climate analyses and forecasts based on their best efforts at estimating long term climate trends around the nation- the latest being their suitably scary State of the Climate 2020. 

The main datasets used are ACORN-SAT (Australian Climate Observations Recording Network- Surface Air Temperature) Version 2.1, Daily and Monthly Rainfall Networks, Monthly Pan Evaporation Network, and Monthly Cloud Amount Network.  In future posts I hope to look at some of the BOM’s claims in more detail, however in this series of posts I will look at climate trends at individual stations.  In this post I will look solely at monthly maximum temperatures at all 112 ACORN-SAT sites.  This information is freely available at http://www.bom.gov.au/climate/change/index.shtml#tabs=Tracker&tracker=site-networks and is adjusted and homogenized Acorn V.2.1 data.

Like the Bureau, in order to compare data from all stations I calculate anomalies from monthly means for all months from 1981 to 2010.  I then calculate 120 month running means.  120 month (decadal) means allows us to see long term patterns and changes.  For example, Figure 1 shows decadal monthly means of rainfall that fell at Alice Springs since October 1900. 

Figure 1:  Decadal rainfall at Alice Springs

I would not use the term “cycles” to describe what we see, but clearly there are wetter and drier periods: rainfall is not random from year to year at Alice Springs.

The same decadal averaging when applied to maximum temperatures should show how temperature changes over years, and because Acorn 2.1 is homogenized using neighbouring stations for adjustments, there should be similarities between stations in the same regions.  Let’s see.

I have made all means zero at December 2019 (except Boulia, which ends in June 2013, Point Perpendicular, ending in January 2017,and Gunnedah, ending in June 2019), so in the following plots all data points are relative to the most recent available.  Each data point is the mean of all monthly maxima of the previous 10 years.

Figure 2:  Running 120 month means, maxima anomalies (from 1981-2010 means), relative to most recent data (mostly December 2019), all 112 Acorn stations

That spaghetti plot shows decadal Tmax for all 112 Australian stations, with a few stations identified.  What a mess.  There is a range of 1.5 to 2.5 degrees between highest and lowest in most years before 2000.  We need to look at different regions to make more sense of it.  I will show a map for each region.

Figure 3: Tasmanian stations

Figure 4: Decadal anomalies, Tasmania

Tasmania is a small, compact region, and all stations appear to show the same decadal climate variations.  However, Grove seems to have much less increase than the others, and Larapuna has a much greater increase than its close neighbours, Low Head and Launceston.

Figure 5: East coast of Queensland

Figure 6: Decadal anomalies, east coast of Queensland

Similarity between stations barely extends back as far as 2005.  There is little sign of common climatic patterns except in very recent years.  Brisbane Aero and its closest neighbour Cape Moreton Lighthouse diverge between 1986 and 2007.  And Mackay in particular is an outlier: what reason can there be for Mackay to be more than one degree cooler in all decades to 1940 than Bundaberg to the south and Cairns to the north?

Figure 7:  North-east NSW stations

Figure 8: Decadal anomalies, north-east NSW

Again, while there are some similarities, there is much variety.  Inverell is more than one degree cooler than neighbouring Moree in the decade to the early 1920s, then their decadal means converge to within 0.3 of a degree in the 1950s.  And Scone has had a meteoric rise from 1.6 degrees less than now in November 2001- faster than anywhere else in Australia.

Figure 9:  South-west Western Australian stations

Figure 10: Decadal anomalies, South-west Western Australian stations

This climate region has fairly consistent records, at least back to the 1930s, when Perth’s diverges from the others.  Perth goes from coolest in the 1920s to warmest (relative to now) in the 1980s.

The northern part of Western Australia is messier.

Figure 11:  Northern Western Australian stations

Figure 12: Decadal anomalies, northern Western Australia

Halls Creek and Broome are much cooler than Port Hedland, Marble Bar, and Carnarvon in the decades before the 1930s.  There is a range of 1.3 degrees between decadal means of Marble Bar and neighbouring Karijini North (the former Wittenoom) in 1969, and there is a large divergence between Kalumburu and Carnarvon (at opposite ends of the coast), and the rest of the stations, between 2000 and 2008.

Central Australian stations, because of their remoteness, have a large impact on our climate signal.

Figure 13:  Central Australia

Figure 14: Decadal anomalies, Central Australian stations

While there are similar decadal patterns in maximum temperatures, you will note that Alice’s record rises from the coolest in the 1920s and 1930s to warmest from the 1940s to 1970s, in steps rather than rises and falls.

The Top End is subject to the annual north-west monsoon, with climatic seasons alternating between Wet and Dry.

Figure 15:  Top End stations

Figure 16: Decadal anomalies, Top End

Again we see in most stations rises in the 1970s and 1990s, and falls in the 1980s and early 2000s.  The exception is Darwin, with an almost linear increase with a small acceleration in the 1990s.  Normanton in the far east is an outlier before the 1980s, and VRD in the 1990s.

Inland New South Wales is another region showing common climate patterns, but a few surprises.

Figure 17:  Western NSW stations

Figure 18: Decadal anomalies, western NSW

Here is a good example of many stations showing common climate patterns, rising and falling almost in unison.  However there is still well over one degree between highest and lowest in nearly every year before 1990.  Further, it is not perfect unison: not all stations show similar responses to regional climate swings.  In 1956 and 1957 Canberra at 2.4 degrees cooler than now and Walgett at more than 2.5 degrees cooler are clear outliers, and are well below the pack from 1950 to 1972, and again from 1980 to 2002.  Walgett in particular shows little response to the 1980s surge shown by most other stations.  These two are joined by West Wyalong in the 1970s, and are just under 1.5 degrees cooler than now in 2000 before surging rapidly.

Finally, for comparison, the next plot shows some of the big movers in the Acorn stations, most of which we have seen before.

Figure 19: Decadal anomalies, big hitters

Linearly rising Darwin and recent rapid riser Scone we have met before.  Alice Springs and Perth are joined by Geraldton and Eucla, both in Western Ausralia, in rising from about 2 degrees cooler than now in the decade to the 1920s.  In the 1930s another WA station, Morawa, is almost 2.5 degrees cooler than now.  In the previous figure we saw Canberra in the 1950s 2.4 degrees cooler than now and Walgett more than 2.5 degrees cooler than now: the coolest of any station in Australia. 

Conclusion:

Decadal means show broad patterns of climate change in various regions but there are many examples of individual stations within these regions standing out from these patterns.  They can’t all be right.  The accuracy of the BOM’s ACORN-SAT dataset for maximum temperatures must therefore be called into question at a number of its stations.  This must then throw doubt on the Bureau’s climate analyses and future projections.

In future posts I will look more closely at some of these individual stations’ records.

The Mexican Wave: Covid19 in Australia to October

November 2, 2020

Postscript: For more detailed information and graphs that support/ augment/ supersede my analysis, see https://www.health.gov.au/news/health-alerts/novel-coronavirus-2019-ncov-health-alert/coronavirus-covid-19-current-situation-and-case-numbers

In Queensland we refer to people in the southern states as “Mexicans” (because they’re from “south of the border, down Mexico way” as sung by Gene Autry, Patsy Kline, Patti Page and many others.)

Read on to find why I describe the Australian Covid19 experience from June to October as the Mexican wave.

Worldometers has these plots illustrating the Australian experience:

Figure 1:  Daily new cases

There were (apparently) two waves in Australia.

Figure 2: Cumulative death toll

In four months the death toll increased by 803- more than 770 %! 

We know what went wrong, but the following plots might illustrate it more clearly.

These plots are from statistics from State government websites, such as this one from Victoria: https://www.dhhs.vic.gov.au/victorian-coronavirus-covid-19-data .

All are correct as of 31 October.  They speak for themselves so I will keep my comments to a minimum.

The next figure compares seven day averages of Victorian and all Australian new cases from 25 July to 6 August, at the peak of the “second wave”.

Figure 3: National and Victorian new cases

Until 5 June, Victoria had 1,681 cases.  From then, the new cases began increasing, adding another 18,666 cases to 31 October.  92% of Victorian cases were in this period.

Comparing all states:

Figure 4:  Total cases

Figure 5:  Mortality:

I estimated population figures from March ABS figures.  With almost zero overseas net immigration and very little interstate migration, natural growth remains, which does not change the rates per million by very much at all.

If Victoria was a separate country, its case rate per million would rank it at 127th, just ahead of Bangla Desh.   

Figure 6:  Case Rate per million people

Its Death Rate per million would rank it at 76th, just ahead of Turkey.

Figure 7:  Mortality Rate per million people

The next figure shows Case Fatality Rate, the number of deaths per total cases, which is not complete until the pandemic is over.  These figures are for the CFRs to 31 October.

Figure 8: Covid19 Case Fatality Rate

CFR is affected by whether the virus gets into nursing homes and hospitals which have high proportions of vulnerable people.  There was an outbreak of Covid19 in hospitals in northern Tasmania which affected the Tasmanian CFR.

 4.03% of all Victorian cases so far resulted in death.

The figure for all of Australia is 3.29%.

The figure for Australia excluding Victoria is 1.22%.

The virus first entered Australia via overseas travellers, then spread by local transmission.  The next plot compares infections acquired overseas with those acquired locally in Victoria.

Figure 9: Victorian overseas and locally acquired infections

The contrast is stark.  Victoria compares most unfavourably with other states with over 95% of all cases locally acquired. (Data not available for Tasmania and Territories.)

Figure 10:  Percentage of local transmission in larger states

And Victoria has more than 90% of total national local transmission.

Figure 11:  Percentage of national local transmission

Therefore it can be clearly seen that Australia’s “second wave” was really all about Victoria.  This was easily avoidable with strict hotel quarantine and better contact tracing.  There was no second wave in other states, with small outbreaks mostly due to travellers from Victoria.

Perhaps “Mexican” should from now on describe the government of Victoria, but not their long suffering people, and not governments of NSW, Tasmania, or South Australia.

The Mexican Wave is not something we wish to see repeated.

First Wave Covid19 Mortality in Context

October 22, 2020

Key takeaway points:

  • It is likely that the real Covid19 death toll was at least double the official tally, and possibly hundreds more.
  • Despite this, there were 1,457 fewer deaths in the first six months of this year than last year.
  • The first lockdown worked- until the Victorian fiasco.

In this post I use the most recent Mortality data (released 1 October 2020) from the Australian Bureau of Statistics (ABS), up to 30 June 2020, and the most recent ABS Population data, to examine the effect of the Covid19 pandemic on Australian deaths.  This period covers the whole of the first wave of the pandemic and gives interesting insights.  Future data releases covering the second wave (with another 800 Covid19 deaths) will provide further illumination.

The ABS advises that the data are provisional and not complete as deaths subject to coroners’ inquests are not included, but with completeness percentages in the high 90s “meaningful comparison with historic counts” may be made.

Key statistics from the ABS:

  • 68,986 doctor certified deaths occurred between 1 January 2020 and 30 June 2020.
  • Numbers of deaths have been below historical averages since mid May and below baseline minimums since the week ending 9 June.
  • Deaths from respiratory diseases and heart diseases were below historical minimum counts throughout June.

Figure 1:  ABS chart of deaths and Covid19 infections

The peak of new coronavirus infections was in the week ending 31 March, with 2,428 new infections in that week (Week 13), and the peak in all mortality also occurred in that week.  The following plot shows official Covid19 mortality (from Worldometers) peaking in Week 14.

Figure 2:  Covid19 first wave deaths

The ABS says that the World Health Organisation (WHO) early in 2020 “directed that the new coronavirus strain be recorded as the underlying cause of death, i.e. the disease or condition that initiated the train of morbid events, when it is recorded as having caused death……..

……. Deaths due to COVID-19 are included in the total for all deaths certified by a doctor. They are not included in deaths due to respiratory diseases or any of the other specified causes.”

The first reported Covid19 death was on 1 March, (Week 9).  In Week 14, one week after the peak in new infections, the peak in the first wave deaths occurred.  In this post I define the first wave of the pandemic as Weeks 9 to 21.  (The second wave commenced in Week 24.)  Figure 3 shows Covid19 deaths in context.  The duration of first wave deaths is indicated by the horizontal red line.

Figure 3: Covid19 and total deaths

Note the increase in total deaths in Weeks 12 to 15, and the insignificance of official Covid19 mortality by comparison.  (Australia closed borders on 16 March- Week 11- and began restricting movement in the days following.)

The next graph compares 2020 mortality so far with the five previous years.

Figure 4: Total Australian deaths 2015 – 2020

This year’s peak in deaths also occurred in Weeks 12 to 15, at the height of the first wave infections.

You will also note Australia’s 2020 mortality levelled off well below previous years’ figures, which usually continue rising to peak in Winter and early Spring.  Mortality figures for Weeks 27 to 52 will be very interesting.  There was an unusual early surge in 2019, and a very large increase in deaths in Winter and Spring of 2017.

I now look at excess deaths.  The ABS says:

Measuring ‘excess’ deaths

Excess mortality is an epidemiological concept typically defined as the difference between the observed number of deaths in a specified time period and the expected numbers of deaths in that same time period. Estimates of excess deaths can provide information about the burden of mortality potentially related to the COVID-19 pandemic, including deaths that are directly or indirectly attributed to COVID-19.

… counts of deaths for 2020 are compared to an average number of deaths recorded over the previous 5 years (2015-2019). These average or baseline counts serve as a proxy for the expected number of deaths, so comparisons against baseline counts can provide an indication of excess mortality. “

However, Australia’s population has increased by nearly two million from March quarter 2015 to March quarter 2020 (from 23,745,629 to 25,649,985).  This has a large impact on calculations.  Mortality rate per 1,000 head of population is a better measure. Figure 5 shows mortality rates for recent years.

Figure 5:  Australian mortality rates, 1st 26 weeks, 2015 – 2020

The method I have used is different from the ABS methodology because of the population increase and is based on mortality rates rather than absolute numbers of deaths. 

I have calculated the mortality rate per 1,000 people for each of the 2015-2019 years (using the population for the March quarters of those years), and similarly for the 2020 data.  I then multiply the average of the 2015-2019 mortality rates by the 2020 March quarter population to obtain an estimate of predicted deaths for 2020.  Subtracting this from the actual 2020 number gives an estimate of excess deaths.  An excess death figure of zero indicates the mortality rate is no different from previous years.  The next figure shows plots of actual and expected deaths for the first half of 2020.

Figure 6:  Predicted and actual deaths

Figure 7 is my plot of excess deaths to 30 June.

Figure 7: Estimated Excess Mortality

Excess and actual deaths peaked in Weeks 12 to 15, with weeks 13 and 14 nearly 200 above the expected level- but there were only 56 official Covid19 deaths in those weeks.  Officially, Covid19 was involved in 29 deaths in Week 14, 12 each in Weeks 13 and 15 and only 3 in Week 12.  It is possible that Covid19 deaths were being vastly under-reported in March. 

By the end of June estimated excess deaths were at minus 349, 11.5% below the expected number for Week 26.  Actual deaths in the first half of the year were 1,457 fewer than for the same period in 2019.

States and Territories:

Figure 8 shows actual numbers of deaths for all states and territories.

Figure 8:  2020 mortality numbers for each state

Mortality figures are dominated by New South Wales, followed by Victoria and Queensland.  Figure 9 shows excess deaths.

Figure 9:  Excess mortality by states

Smaller states had smaller changes in excess mortality, although Western Australia had a peak of 54 excess deaths in Week 13.   Figure 10 shows excess deaths for the larger states only.

Figure 10:  Excess deaths in the large eastern states

Peaks in excess deaths occurred between Weeks 9 to 17, but note earlier peaks in New South Wales and Queensland 7 or 8 weeks before the pandemic peak, with Queensland much higher than New South Wales, largely counteracted by Victoria, and a peak in Victoria in Weak 11, counteracted by New South Wales.  There was a third peak in Weeks 17 to 19, coinciding with another peak in Covid19 deaths.  Remember these numbers are additional to Covid19 deaths.  And officially Queensland had only seven Covid19 deaths, almost certainly due to under-reporting.

Age at death

Figure 11 shows the ages at which excess deaths occurred.

Figure 11:  Excess mortality by age

People aged from 0 to 44 years were not affected by the large changes in death rates in older age groups, but there was an increase in excess deaths in the 45 to 64 age bracket in Week 13, at the height of Covid19 infections, as Figure 12 shows.  That looks suspicious, but may be chance.

Figure 12:  Excess deaths for younger cohorts

The majority of excess deaths were in older age groups, as Figure 13 shows.

Figure 13:  Excess deaths for older Australians

There was a peak of 132 excess deaths in those 85 years and over in Week 14, but in Week 13 there were 146 excess deaths in those aged 65 to 84.  There were additional substantial peaks in earlier weeks as well.  It was not a good first half of the year for senior citizens, but excess deaths for all age groups were well below expected numbers by June.

Cause of death

  A death certificate lists all causes of death, and with elderly people these can be three or more.  It is very likely that a person over 85 may die of pneumonia (classified as a respiratory illness), but may also have any or all of dementia, diabetes, cerebrovascular disease, ischaemic heart disease, and cancer.  However, the ABS asks doctors to report the (one) underlying cause of death, and since earlier this year, Covid19 as the underlying cause “when it is recorded as having caused death.

 Figure 14 compares all respiratory deaths with Covid19.

Figure 14:  Covid19 and respiratory deaths

Influenza and pneumonia are subsets of respiratory illness, and the next figure shows interesting excess mortality data for 2020.

Figure15:  Excess deaths due to respiratory causes

Note the peak in respiratory deaths at the height of pandemic infections, but an earlier peak some four weeks previously.  It is likely that Covid19 was not correctly reported to the ABS by all doctors until Week 14 or 15- doctors are human too.  Since the first wave and the increase in personal hygiene, social distancing and little travel, deaths have remained well below previous years.

Figure 16:  Ischaemic heart and cerebrovascular disease excess deaths

This plot illustrates the advances in medicine:  ischaemic heart disease in 2020 had fewer deaths than expected for all of the first six months apart from a peak in Week 7.  Cerebrovascular disease (chiefly strokes) also had fewer deaths than expected except for Week 14 (so was potentially related to Covid19), and another peak in Week 24. 

Figure 17 plots excess deaths caused by the common co-morbidities of Covid19, dementia and diabetes.

Figure 17:  Excess deaths caused by dementia, diabetes, and Covid19

Diabetes and Dementia excess deaths were also higher than expected during the first wave, but there was another large surge in excess deaths with dementia as a cause weeks earlier.

Conclusions:

With the caveat that the ABS mortality figures are provisional, and putting together figures for various states, ages, and causes of death, some conclusions may be drawn:-

Either a mystery respiratory illness or undiagnosed Covid19 was widespread in the eastern states amongst elderly people weeks before the peak of first wave deaths, possibly arriving from cruise ships.

There were probably many more Covid19 deaths and infections than reported.  It is likely that the real Covid19 death toll was at least double the official tally, and possibly hundreds more.

Social distancing, good hygiene, and travel restrictions have caused a large decrease in mortality in May and June by restricting the spread of many common illnesses.  The first lockdown worked- until the Victorian fiasco.

The net effect of the first wave of the Covid19 pandemic on Australian mortality was negative.  Covid19, and public health responses to it, resulted in a lower death toll in the first half of 2020.  This lower death toll was not just in relative (mortality rate) terms but also in absolute terms: there were 1,457 fewer deaths in the first six months of this year than last year.

ABS data for the second half of the year will be released around April 2021 and will provide much better information about excess mortality for all states (and Victoria in particular), for all age groups, and for all causes.

I include an appendix with raw mortality data for 2015 -2020.

Appendix:  Raw mortality data for all causes 2015 – 2020.

Figure 18:  Respiratory mortality

Note the typical winter and spring surge in respiratory deaths, mainly due to influenza outbreaks in cold months.  There was an early surge in 2019 and a very large surge in 2017 which will skew means for those weeks.  Median mortality rate may be more appropriate than means.

Figure 19:  Ischaemic heart disease mortality

Heart disease mortality has been below previous years for most of the first 26 weeks.

Figure 20:  Cerebrovascular disease mortality

Cerebrovascular disease (stroke) deaths peaked during the first wave of Covid19 but have been mostly near the bottom of the range of previous years, with a second peak in June.

Figure 21:  Dementia mortality

Deaths with dementia as a cause have increased over the years.  A peak in dementia deaths coincided with Covid19 but deaths have been in the normal range since then.

Figure 22:  Diabetes mortality

A peak in diabetes deaths coincided with the peak in Covid19 infections and deaths, and was much higher than expected.  At the end of June deaths were in the range of previous years.

Figure 23:  Cancer mortality

Cancer deaths have increased over the years and 2020 remains within the expected range.  You may note there is no winter increase in cancer mortality.

Distance Records for Temperature Adjustments

October 6, 2020

Trigger Warning:  ridicule of the Australian Bureau of Meteorology below!

The official Australian climate record is developed from ACORN-SAT– the Australian Climate Observation Reporting Network- Surface Air Temperatures.  This is relied on by governments and industry and so should be completely trustworthy and free from any problems that might lead to lack of confidence.

The Acorn stations have had their temperature records adjusted to account for any discontinuities or irregularities.  This is done by comparing Acorn stations’ data with those from a selection of comparative stations. 

The Bureau says:

The process of homogenisation seeks to answer a very simple question: what would Australia’s long-term temperature trend look like if all observations were recorded at the current sites with the current available technology? Homogenisation means we can compare apples with apples when it comes to temperature trends.

One might expect that, with the aim being to “compare apples with apples”, stations used for comparison and making adjustments would be physically not too distant- ideally, neighbouring.  

Not so.

For many stations, not even remotely so.

Australia is a very big country with vast areas of sparsely inhabited desert.  There are very large distances between towns in the outback, so it is not surprising that it is often difficult to find suitable comparative stations.

But the Acorn Station Catalogue, which has helpful lists of comparative stations used for adjustments, has some absolute doozies.  Here are some for your amusement.  (Obviously most stations have many comparative sites.)

Carnarvon, in Western Australia, has been adjusted by reference to a number of stations hundreds of kilometers away, including Southern Cross, only 897km away.

Camooweal, Queensland, “ “ “  Thargomindah, 1,067km away.

Boulia, Qld, “ “ “  Walgett, New South Wales, 1,130km

Halls Creek, WA, “ “ “  Boulia, Qld, 1,370km

Tennant Creek, Northern Territory, “ “ “   Charleville, Qld,  1,443km

Mount Gambier, in South Australia, has been adjusted with the help of Lismore in northern New South Wales, 1,526km away.  (And it’s not as if there is a shortage of sites in this well populated part of South Australia.)

But the gong, the gold medal, the record breaking achievement for the Bureau, goes to…….

Alice Springs, in the Northern Territory, which has been adjusted using data from Collarenebri in New South Wales,  1,590 kilometres away.

And they want the public to trust them.

More Questionable Adjustments- Cape Moreton

October 5, 2020

Here’s another Acorn station with interesting adjustments- Cape Moreton (40043) minimum temperatures.

Cape Moreton Lighthouse is on the north-eastern tip of Moreton Island, 65 km north-east of Brisbane.  It is not compliant with siting specifications. 

Figure 1 is the adjustment summary shown by the Bureau in its Station Catalogue.

Figure 1:  Adjustment summary for Cape Moreton

Two points to note:  The Bureau has TWO adjustments applied to the same date- 1/01/1946; and there are four comparative stations used to make these adjustments at this Acorn station.

Figure 2 shows the neighbours the BOM used for comparison. 

Figure 2:  Google Maps image showing Cape Moreton and its neighbours

There are many neighbouring stations the Bureau could have used for comparison, but the Bureau chose those with the “best correlation” during the comparison period (the late 1940s):  Brisbane Regional Office 65km away, and probably affected by Urban Heat Island effect; Yamba, also coastal but 267km south; Dalby Post Office on the Darling Downs 220km west; and Miles Post Office 330km west.

Figure 3 shows the annual average minima for these weather stations.

Figure 3:  Annual minima, Cape Moreton and neighbours

UHI at Brisbane is visible as the plot line rises faster than the others after 1950.

The next figure shows Acorn’s adjustments have increased the rate of warming from +1.2 degrees Celsius per 100 years to more than +1.5 C.

Figure 4:  Cape Moreton Minima

Figure 5 shows the difference between the original raw record and Acorn.

Figure 5:  Cape Moreton adjustments

It is plainly obvious that the Acorn adjustment summary shown in Figure 1 is wrong.  The first adjustment was applied from 01/01/1948 (not 1946) and decreased the annual minima for 1946 and 1947 by -1.2C.  The second adjustment was applied from 01/01/1946 and increased previous annual minima by +0.8C or +0.9C. The raw minima were decreased by -0.3C or -0.4C, but that is not how the Bureau describes the adjustment process:

Date applied: data prior to this date was adjusted for the reason (cause) cited. Adjustments are superimposed on each other; for example, if two adjustments are shown, one for 1/1/2000 and one for 1/1/1988, data prior to 1/1/1988 has both adjustments applied to it, data between 1/1/1988 and 1/1/2000 only has the first adjustment applied, and data after 1/1/2000 is not adjusted at all.”

The documentation of Acorn is a mess.

In order to compare data from stations with varying temperatures we need to calculate their anomalies from their means for the same period.  Figure 6 shows Cape Moreton’s and comparison stations’ anomalies from their 1931-1960 means.

Figure 6:  Minima anomalies, Cape Moreton and “neighbours”

Hard to follow, there is too much variability.  You may note that by comparison with the periods before 1948 and after 1960, the 1950s show much agreement.  The next figure shows the period from 1930 to 1960.

Figure 7:  Minima anomalies, Cape Moreton and “neighbours” 1930-1960

Notice that in 1946 and 1947 (indicated by the arrow) Cape Moreton is far too warm- the reason for the adjustment; however Yamba’s record is just as erratic or more so, being too low in 1933, 1934, 1940-1944,  and 1947; and too high in 1950.  This suggests firstly that the 1946 and 1947 adjustments were justifiable for those two years, and secondly that Yamba is not a good comparison station.  The next figure, with Yamba excluded, clearly illustrates this point.

Figure 8:  Cape Moreton and comparison stations, excluding Yamba

Apart from 1946 and 1947 Cape Moreton’s record is not greatly dissimilar from the three remaining stations. 

The object of adjusting temperatures using neighbours for comparison is to endeavour to produce a record that more truly reflects climate trends of the area.  The resulting record should be more like the neighbours than the original raw record.  We can test this by plotting the differences between Acorn and the raw record and the average of the neighbours.  If the comparison is good, while individual years’ differences may vary, the trend should be close to zero: the station should not be warming or cooling more than the neighbours.  Figure 9 shows the results for Cape Moreton minima for the period before the 1946 adjustment, excluding Yamba from the average.

Figure 9:  Differences between Cape Moreton and Qld neighbours

You will note that the blue trend line, showing the trend of the difference between Cape Moreton’s annual data and the average of Brisbane, Dalby, and Miles, has a trend of about +0.3 degrees per 100 years, indicating Cape Moreton is already warming more than the others.  The “raw” record already compares fairly well with the neighbours, considering that they are inland stations, unlike Cape Moreton.  In contrast the red trend line shows the adjusted data is warming more than three times faster, indicating a poor reflection of the climate of the area.

Conclusion

The Bureau has not followed its own methodology in its adjustment summary.

Documentation of adjustments is incorrect.

Three comparison stations are hundreds of kilometres away and another is subject to Urban Heat Island effect.

One comparison station (Yamba) has a record more erratic than Cape Moreton’s and should not have been used.

The adjustments have increased the difference between Cape Moreton and its neighbours, and has increased the warming trend by 30%.

Garbage in, garbage out.

Sources for annual minima data:

Acorn: http://www.bom.gov.au/climate/change/hqsites/data/temp/minT.040043.annual.txt

Raw:

Cape Moreton:

http://www.bom.gov.au/jsp/ncc/cdio/weatherData/av?p_nccObsCode=38&p_display_type=dataFile&p_startYear=&p_c=&p_stn_num=40043

Brisbane Regional Office: http://www.bom.gov.au/jsp/ncc/cdio/weatherData/av?p_nccObsCode=38&p_display_type=dataFile&p_startYear=&p_c=&p_stn_num=40214

Dalby Post Office:

http://www.bom.gov.au/jsp/ncc/cdio/weatherData/av?p_nccObsCode=38&p_display_type=dataFile&p_startYear=&p_c=&p_stn_num=41023

Miles Post Office:

http://www.bom.gov.au/jsp/ncc/cdio/weatherData/av?p_nccObsCode=38&p_display_type=dataFile&p_startYear=&p_c=&p_stn_num=42023

Yamba Pilot Station:

http://www.bom.gov.au/jsp/ncc/cdio/weatherData/av?p_nccObsCode=38&p_display_type=dataFile&p_startYear=&p_c=&p_stn_num=58012

Australia’s Wacky Weather Station Comparison 4: Penrith (NSW)

February 20, 2020

After surveying 666 weather stations across Australia and finding nearly half (49.25%) are not compliant with Bureau of Meteorology siting specifications, in this series of posts I compare daily temperature data from pairs of compliant and non-compliant stations. Here’s the first in this series.

Penrith and Richmond RAAF

These stations are in western Sydney, 16km apart.

Fig. 1:  Penrith map location per Google Maps

Fig.2:  Penrith and Richmond

Penrith Lakes AWS 67113 is beside a large area of excavation and bare soil, and 200 metres from a large artificial lake.

Fig. 3:  Penrith (Google satellite image 2019)

Richmond RAAF 67105 is at an Air Force base. It is open, flat, and at least 50 meters from any concrete or tarmac.

Fig. 4:  Richmond RAAF site plan 2016

Fig. 5:  Richmond RAAF (Google satellite image 2020)

Richmond is 6 metres higher than Penrith.  Both are Automatic Weather Stations with electronic temperature probes transmitting data every minute. While there can be no observer error, as we shall see there can be instrumental error.

Richmond RAAF is an ACORN station. The Bureau says in its Station Catalogue: “The region is a major growth corridor for Sydney and there is evidence of anomalous warming of minimum temperatures in recent years.”

If we plot all daily maxima from 2010 to 2019 for Richmond against Penrith, we see that temperatures match quite closely:

Fig. 6:  Tmax at Richmond as a function of Tmax at Penrith

Richmond is on average cooler than Penrith. A time series of the 31 day centred mean of the daily difference between them shows more detail:

Fig. 7:  31 day mean daily difference Penrith minus Richmond Tmax

Values above zero mean Penrith is warmer than Richmond; below zero, Penrith is cooler.  Most summers Penrith is warmer, and winters slightly cooler, though the record appears to have breakpoints in early 2012 and early 2016, and some unusually high values.

This is a plot of mean differences by month:

Fig. 8: 31 day mean daily difference Penrith minus Richmond Tmax by month

Penrith is warmer in every month, especially in summer, though there are some cooler values in every month.

Minimum temperatures at Richmond are much cooler than Penrith:

Fig. 9:  Tmin at Richmond as a function of Tmin at Penrith

Fig. 10:  31 day mean daily difference Penrith minus Eichmond Tmin

Penrith is 2C to 2.5C warmer in cooler months and up to 0.5C warmer in summer.

Fig. 11: 31 day mean daily difference Penrith minus Richmond Tmin by month

A note on accuracy:

The centred 31 day running correlation is useful for detecting inconsistencies.

Fig. 12:  Centred  31 day running correlation between Penrith and Richmond maxima

Fig. 13:  Centred  31 day running correlation between Penrith and Richmond minima

The much poorer correlation in the summer of 2013-2014 shows in Figures 7 and 10. Here are the actual minimum temperatures recorded:

Fig. 14:  Daily minima at Penrith and Richmond Summer 2013 – 2014

It appears that the Richmond probe began malfunctioning in mid-December and failed completely in mid-January. It failed again a few months later.

In recent years, Penrith Lakes AWS 67113 has recorded generally warmer maxima than Richmond RAAF 67105 in summer and comparable or slightly cooler maxima in winters. Minima are always much warmer at Penrith. This may be due to the proximity to the large artificial lake.

In this example, siting non-compliance has a large effect on temperature.

***

This will be the last comparison, as it is very difficult to identify non-compliant stations with nearby compliant sites with similar environment. We can conclude however that non-compliance with siting specifications affects temperatures recorded, which varies between locations. Sometimes maxima are much warmer, sometimes minima. Temperatures at 329 non-compliant stations cannot be regarded as reliable for weather or climate analysis.