There is sterling work going on in the USA by Citizen Journalists auditing the temperature recording network at www.surfacestations.org. Some unbelievably badly sited or maintained stations are being used to feed data into the system - see for instance Detroit Lakes.
So what about the UK?
The official record is the Central England Temperature (which) is representative of a roughly triangular area of the United Kingdom enclosed by Bristol, Lancashire and London. The monthly series begins in 1659, and is the longest available instrumental record of temperature in the world.
... The stations used to compile CET are chosen from the UK surface station network to be consistent as possible with those used historically. The data is then adjusted to ensure consistency with the historical series.
The Met office released a paper on corrections to the CET series here:
Uncertainties in the Central England Temperature series 1878-2003 - pdf
Quantification of uncertainties in climatic data records is a prerequisite for the
interpretation of trends and extreme values. Here we estimate the uncertainties in
the most recent and reliable 125 years of the longest instrumental record in the
world: Central England Temperature (CET)...
CETmax and CETmin are constrained to have an average equal to CETmean,
as published by Manley (1974)
What? They have decided that Manley in 1974 got the mean right so all they are going to adjust are the max and min, in equal amounts to keep the mean the same. Isn't that deciding on the answer before looking at the evidence?
Measurement errors include thermometer calibration errors, errors from reading
and recording temperatures (precision errors), and errors arising from the
method of housing the thermometers.
Calibration errors in the late 19th century cited by the Met Office (1879, 1880) have a
standard deviation of about 0.2°C in magnitude. So we assume a single
thermometer standard error of 0.15°C throughout and assume the errors are
random between thermometers, ...
Results: Calibration standard error for CETmax and CETmin
1959 on 0.079
I know my statistics are rusty but somehow they just made a thermometer which may be 0.2°C out only give an error of 0.08..hmmm.
They do a similar calculation on the reading errors. They continue...
For thermometers housed in Stevenson screens, the type and condition of the
screen affects the accuracy of the measured value. A comparison of temperatures
taken from digital thermometers housed in various screens in Sweden (Andersson
and Mattisson, 1991) revealed much bigger extreme differences between the
screen thermometers and an aspirated thermometer (the 'true' value) than those
given in the World Meteorological Organization (WMO) Guide to Meteorological
Instruments and Methods of Observation (WMO, 1983)....Andersson and Mattisson calculated RMS
errors of 0.26°C, 0.31°C.. for daily Tmax, Tmin......
....the 1980s. The subsequent increase in diurnal range then appears to be
exceptional in a historical context. The known increase in sunshine in the United
Kingdom in recent decades (Parker et al., 2004) is consistent with an increase in
diurnal temperature range.
Owing to the availability of additional digitized daily data, Parker et al (1992)
used different stations for daily CETmean than Manley (1974) had used for monthly
CET ). Because of these differences in stations, the areal average
temperature at Parker et al’s stations differed slightly from Manley’s values. So
Parker et al. (1992) adjusted their daily CETmean values to make their monthly
averages consistent with Manley (1974). For similar reasons, when we created
daily CETmax and CETmin series, again using a different sequence of stations (Table
1), we adjusted the values so that each day’s average of CETmax and CETmin
equaled that day’s adjusted CETmean and was therefore also compatible with
....no urbanisation adjustments were made to the CETmax series because
urbanisation is known to affect minima much more than maxima...
The total uncertainty for a given time-scale is the square root of the sum of all of
the individual error variances on that time-scale.
...This shows that temperature differences between at least the 10 warmest years
are not statistically significant given (especially) the areal sampling and
calibration errors. Therefore statements such as “2003 was the 7th warmest year
in the CET record” must be qualified with reservations regarding the uncertainty....
And so it goes on.. It seems to my inexpert eye they have decided that the mean temperature is correct and all they are doing is estimating how far off the extremes are -I don't get any feeling they have asked the question what if there is a systematic bias in one direction at one or more of the measuring sites. But then they are Brtitish so they must be reliable, mustn't they?