« Scottish Proscription | Main | Slopping Out »

Forensic Climatology and the Central England Temperature (CET) record - UPDATED

A very welcome guest post by Willis Eschenbach which raises questions over the UK's long running temperature record.

Because of the difficulty in obtaining the underlying data and computer codes used for various studies and graphs, I often find myself involved in what I call "forensic climatology". This is the art and science of reconstructing data and procedures from published graphs and descriptions.

Thanks to Tim at "An Englishman's Castle", I recently got interested in the UHI adjustments to the Central England Temperature (CET) record. ....

These issues are discussed in Parker et al., " Uncertainties in the Central England Temperature series 1878-2003 and some changes to the maximum and minimum series" Here are Figures 8a and 8b from that paper:

The first step is to extract the data from the graphs. I use a combination of manual and automated methods. First I carefully trace over one of the lines in the graph. Then I mask out the rest of the lines, and use digitizing software to extract the values for that line. (I use "GraphClick" on the Mac, but there are equivalent programs for the PC). I repeat this for each of the lines. Then I copy the digitized data into Excel, and interpolate it to determine the values at regular monthly intervals for the period of record (January 1959 - December 2002).

The next step is to look for internal relationships in the data. In this case, the data are station anomalies relative to the CET average values. Since the CET average is defined in the Parker paper as being the average of Rothamsted + Malvern + 0.5*(Squires Gate + Ringway), one internal relationship should be:

Rothamsted anomaly+ Malvern anomaly+ 0.5*(Squires Gate + Ringway) anomaly = 0.

However, when we look at the data, this is not the case. Here are the Tmax anomaly figures:

Rather than all of them being zero as we would expect, there are fairly wide swings in the data. The exact reason for this is unknown, but it must relate to adjustments in either the station data or the CET average data. One clue comes from the Parker et al. paper, which says:

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 (Table 1). 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
Manley (1974). In this section, we estimate the uncertainties arising from these
adjustments. We also estimate the uncertainties stemming from the adjustments
applied by Parker et al. (1992) to the CETmean data from 1980 onward to
compensate for urban warming. These adjustments differed between calendar
months and have been increased in magnitude to reach -0.2°C in all months by
2003. In Section 6 we consider the biases arising from the application of double
these adjustments to CETmin and no urban adjustment to CETmax.

Since there is a big change in 1974, it is tempting to think that the above mentioned changes were the cause. But if this is the case, it invalidates the reason for using the values in Tables 8a and 8b. Since the average values have been adjusted so much (e.g. more than three quarters of a degree from 1964-1967), the resulting comparison of averages with the station values loses meaning. In fact, having to make this size of adjustment suggests that there may be deeper problems with their data.

Next, here is the corresponding station Tmin minus CETmin graph:


The difference in this graph is that the CET Tmin values have been adjusted downwards for urban warming (Urban Heat Island, or UHI), while the CET Tmax values were not. Because of this, the difference between the station temperature and the CETmin increases over time. Again, however, we find the jumps downward in 1974 and upward in 1980.

Finally, we can determine how much the UHI adjustments are by subtracting the data in the first graph from the data in the second graph. This will zero out the common differences in the CETmin and CETmax data, leaving only the UHI adjustments. Here is that graph:


Once again, questions arise. You would expect the UHI adjustment (which began in 1980) to be relatively smooth. Instead, there is an adjustment of about 0.3°C around 1980, and then no further adjustment until 1997. At that point, there is an abrupt adjustment of about 0.65°, followed by a steep climb. These adjustments seem quite odd.

What can we conclude from all of this? Unfortunately, we end up with questions rather than conclusions.

1) Why don't the post 1974 Tmax station data agree with the CETmax averages (first graph)? Prior to 1974, the CETmax and CETmin averages were adjusted to fit the Manley data. But after that, the station data average should have been stable with respect to the CETmax average. Instead, the station - CET TMax data jumped upwards in 1980, and down again in 1997.

2) Why did the station Tmax - CETmax values change in 1980 (first graph)? The Parker et al. paper says that the UHI adjustment was applied solely to the Tmin data.

3) Why did the station Tmax - CETmax values rise from 1990-1995 after nearly a decade of stability? (first graph)

4) Why the change in adjustments in 1980? (Third graph)

5) Why the huge change in the adjustments post 1997? (third graph)

6) The Parker et al. paper lists the following adjustments made in 1974 in order to match the earlier (Manley) data:


Given these adjustments to match the pre-1974 data, why is there such a large drop in both Tmax and Tmin post 1974 (graphs one and two)?

7) Why is the UHI adjustment (post 1980, third graph) consist of a jump in 1980, no adjustment for seventeen years, and then a radical adjustment?

8) Finally, it appears that rather than adjusting the individual station averages for urban warming, by comparing them with nearby rural sites, the average of the individual sites is adjusted. Why adjust the averages rather than the individual stations?

Some or all of these questions may have perfectly simple answers. However, I could not find them in the description by Parker et al., which may be my poor reading skills, or may be because they were not explained in the paper, or may be because they are actually errors in the processing of the data.

I look forward to any answers to these questions.

My best to everyone,


UPDATE David Parker responds:

The adjustments Willis Eschenbach shows up to 1974 are as expected the
mirror image (approximately, owing to the smoothing in the plots) of the
adjustments plotted in Figure 1 of Parker and Horton (2005) for 1958-74.
These adjustments account for station changes, ensuring a consistent
series whenever stations close and have to be replaced. They remove the
average difference in climate between an old station and its
replacement. The paper estimates the uncertainties resulting from this
adjustment process. The Table 5 Willis Eschenbach reproduces shows these
uncertainties, not the actual adjustments!

However Figures 8 a and b also suggest an error in Willis Eschenbach's
extraction or subtraction. Visual inspection of the Figures does not
indicate that [Rothamsted anomaly+ Malvern anomaly+ 0.5*(Squires Gate +
Ringway) anomaly] deviates much from zero for Tmin or exceeds about 0.4C
(the maximum urban warming adjustment) for Tmax. Note also that near the
end of the plot the filter has to assume imaginary values (equal to the
average value in the last 31 months because it is a 61-month smoother)
beyond the end of the plot so the data-extraction method could be
misleading there. If we didn't do this we would be unable to show
anything within 2.5 years of the end of the series (which were available
through 2003 when the paper was being written).

It is true that, rather than adjusting the individual station averages
for urban warming, by comparing them with nearby rural sites, the
average of the individual sites is adjusted. We did this as the
combination of sites reduces "noise" in the adjustment process because
local-scale weather and micro-meteorological influences are thereby
cancelled out somewhat more effectively.

I hope this helps


David Parker

UPDATE - Willis responds to David:

David, thank you for your prompt reply. You say that the adjustments up to
1974 are the mirror image of the adjustments made to the station data. If
this is the case, perhaps I misunderstood your graph. You say that you are
showing the stations (presumably adjusted for station closure, etc) and
their anomaly about CET average. If that is the case, then why is the sum of
the anomalies about the average not always zero?

On the other hand, it is possible that the graph shows the unadjusted
stations versus the CET average. But what would be the use of that? Why show
unadjusted data, which is presumably inaccurate, rather than adjusted
station data?

Post 1974, you say that there is an error in my extraction or subtraction. I
have rechecked my figures and cannot find it (which of course does not mean
that it is not there!). If you would be so kind as to post on the web (or
send me directly) the data upon which the graph is based, we could determine
if I have made an error.

Next, you say that you adjust the station average (rather than the
individual stations) for the UHI. While this makes sense, why is the
adjustment done in a step fashion, with a jump around 1980, no further
adjustment until 1997, and then a jump after that?

Finally, if all of the individual station anomalies are filtered using the
same filter as the CET average, the sum of the filtered individual anomalies
minus the CET filtered anomaly should equal zero. In other words, the
filtering should not affect the calculation.

Access to the underlying data would resolve all of these questions. Your
assistance in providing the data to settle all of these matters would be
much appreciated.

All the best,



It's a bit late at night for this kind of thing. I'm struggling to get my head round the sign of the adjustment and the old grey matter refuses to respond.

A positive UHI adjustment implies depopulation, right? Is this Central Park all over again? Or am I missing something?

Bishop Hill, thanks for the question. It's actually the other way around. These are the station data (unadjusted) minus the CET average data (adjusted). A positive value indicates that the average data has been adjusted downwards, in the correct direction.


I've just come across this interesting stuff - minutes ago!
I am /very/ interested in CETs, and have been for years. In fact, am considering writing something for a well known journal. My analysis techniques are completely different from Willis's, and I generally stick to the monthly data published by the Hadley Centre, believing that they have already made any useful adjustments. Am I naive in this?

I could post numerous diagrams, and loads of text and explanations of what I do, but as a newcomer I'll wait to see if anyone is interested. Happy to exchange email addresses for detailed communication.

Regards, Robin


Of course. Amazing how simple things look after a good night's sleep.


The adjustments are where most of the action is.


Even the numbers as published don't show what is claimed for them. I don't think you are naiive, it's easy to get too bogged down by the station data. Would be pleased to communicate on the subject.

Based on unanswered questions about their "data management", I believe some caution is warranted with regard to the Hadley data. As you probably know, the issue of "data management" and resultant erroneous claims and forecasts was highlighted by what recently transpired with regard to the US temperature record at GISS/NASA [thx to Steve McIntyre at ClimateAudit.org]. The "adjustments" outlined above by Willis Eschenbach go to heart of this issue.
As we have learned from the GISS/NASA case, it is interesting to observe the implications when it turns out that more recent temperatures have in fact been overstated. 2006 will likely be adjustede downwards even further, in which case none of the years post-1998 make the 1880-2006 top 10 for continental North America.

Post a comment