Sutton Bonnington Part 2 – Why absurd readings occur, what the potential effects are and whether they occur all the time unseen.

In my part 1 investigation into data recorded at Sutton Bonington in the first two months of overlapping manually observed and automatic readings, I noted some huge discrepancies between the two systems working side by side at the same time. In theory there should not be any difference between simultaneous readings from the two instrument sets but therein lies the problem – they are most certainly NOT simultaneous readings that are being archived. Understanding how this lack of simultaneity arises is key to realising the natures of all those identified errors and why these problems are continuing around the country to this day – furthermore resolving the problem can artificially simulate the appearance of a warming where none may actually exist.

{Author note:I apologise for any repetitions of points in my posts – these are required for the benefit of those reading for the first time.}

I have discussed the issue of reading protocols on many posts (notably here) however, I will briefly reiterate the situation in simplified terms.

Traditional Observations

The UK “Meteorological Day” runs from 09:00 to 09:00 GMT the following calendar day. Readings thus do NOT directly correspond to the conventional calendar day taken.

Readings were taken by human observers at most standard climate reporting sites at 09:00 GMT just once daily. A minority of sites, (principally those used for immediate forecasting purposes) were observed more frequently even up to hourly, though these are not the main source of historic data. These sites were typically manned sites such as airfields, horticultural research centres, coastguards, port authorities and so on.

Instruments were standard Liquid in Glass Thermometers (“LIGT” originally filled with Mercury, latterly alcohol) of a design by James Six which allowed the maximum and minimum temperatures that occurred in the period between “resetting” to be noted.

Modern Automatic Observation

The traditional system required human eye observation which would be difficult to automate and transmit electronically. To overcome this the thermometers were changed from visually observed expansion/contraction of a liquid within a tube to measuring the varying resistance to an electric current passed through a wire probe as temperatures rise or fall. This electronic circuit’s data is easily transmissible over telecommunications links. Electronic Resistance Thermometers (ERTs) were subsequently standardised to a specified thickness/length of a platinum wire probe and became known as Platinum Resistance Thermometers (PRTs)

Multiple frequent readings were now possible to the extent that these are now taken every 15 seconds with 4 readings averaged every minute for potentially 1,440 readings per day.

Archiving of data.

The Met Office protocol was simply to record the traditional manual readings against the meteorological date they were taken. Thus manual readings taken on the 2nd January at 09:00 are recorded on that day for maximum and minimum. However, there is no certainty that either the highest or lowest actually occurred on the 2nd only that they both occurred after the last “reset” of the thermometer. Had that reset been the previous day at 09:00 both the maximum and minimum could have actually occurred on the 1st January. Conversely they could both have occurred on the 2nd January – the Six design cannot offer that delineation. A basic assumption is made for normal calendar day attribution – the maximum is attributed to the day before i.e. 1st January and the minimum is attributed to the 2nd January.

Automatic readings are archived on a twice daily basis at 09:00 and 21:00 and day attributions now change with an either/or applied reasoning whichever is the higher/lower applied to the two readings. This largely (but not quite entirely) corrects the days reading to the appropriate calendar day and is thus frequently different to readings attributed by the traditional system.

Inherent Problems

The inability to confidently compare readings by the exact correct day is not as significant an issue in itself as it may initially seem. Though it may look awkward in readings not conforming by real date these would simply average out over time if day attribution was the sole concern – but it actually is not. What is of concern is that the traditional recording system can effectively incorrectly record readings twice or not give an accurate representation of the day’s maximum or minimum.

Example

2nd January: Minimum recorded at 09:00 GMT at Skye of minus 3°C (below freezing point.) Riders are reset. (n.b.sunrise was just 3 minutes prior to observation and it is known that the coldest time of day is generally within 30 minutes of sunrise.) A warm front moves in from the west significantly raising the temperature throughout the day which does not fall overnight.

3rd January: Minimum reading noted from the bottom of the “rider” is still showing minus 3°C from the previous day even though the air temperature is now reading 11 °C above freezing. The observer still notes the minus 3°C reading despite the real time temperature being 14 °C warmer. Yes this sort of event happens quite frequently and from the historic records of a manually observed site there is no way of knowing whether it actually was the case or not. All that would be visible is the same or a similar temperature recorded on two consecutive days.

This reading carry over is generally termed the “time boundary effect” and relates to a “Time of Observation bias”. Taking readings at a naturally colder time of day tends to lead to a cold bias in traditional readings. The reverse effect of the carry over of maximum readings is also possible though far less common due to this early morning reading tradition.

The automation of readings (and/or the taking of multiple manual readings during the day) largely overcomes this problem thus in doing so effectively creates an appearance of warming when a site transfers over.

Relevance to Sutton Bonington

The transfer from manual to automatic has occurred at literally hundreds of weather stations. The process started in the latter 1970’s but with very few sites involved. The latter 1990’s started a sudden surge of automations for a few years, this then slowed for a while before a further increment around 2009 to 2012. Even at manually observed sites the LIGT were replaced with PRTs and data loggers in 2017 and all new manual sites from 2017 employ them.

Rationally the change over period should run comparative readings from the two systems concurrently in order to establish conformity in readings. A few sites did indeed record these (such as Sutton Bonington) in the very early transition period but this practise stopped soon after (- or rather the information was not publicly recorded if taken.) This is where “Quality Control” should come in.

Above is the manuscript of readings selected at random at Sutton Bonington for February 1935 to demonstrate the point. All that red ink is the work of quality controllers who checked every single individual reading (not just basic temperature readings rather every type of observation) for accuracy to weed out any probable human errors and wrong calculations. On this basis the readings from Sutton Bonington from December 1999 through to March 2000 should have still been subject to this level of scrutiny. So how on earth does all this below occur and then get placed into the official archives seemingly unchecked?

Sutton Bonington: January 2000

Date	DLY max	AWS max	Max diff.	DLY min	AWS min	Min diff.	DLY mean	AWS mean
1 Jan	—	8.7	—	—	2.4	—	—	5.55
2 Jan	—	11.1	—	—	−0.7	—	—	5.20
3 Jan	—	9.7	—	—	5.4	—	—	7.55
4 Jan	11.1	6.0	−5.1	—	−0.8	—	—	2.60
5 Jan	11.9	11.9	0.0	−0.6	−0.7	−0.1	5.65	5.60
6 Jan	11.9	10.2	−1.7	5.4	5.0	−0.4	8.65	7.60
7 Jan	—	9.6	—	2.8	2.6	−0.2	—	6.10
8 Jan	—	7.6	—	—	4.5	—	—	6.05
9 Jan	9.6	5.8	−3.8	—	0.3	—	—	3.05
10 Jan	8.2	8.2	0.0	−2.3	−2.7	−0.4	2.95	2.75
11 Jan	10.2	10.2	0.0	2.6	2.3	−0.3	6.40	6.25
12 Jan	—	9.2	—	8.0	8.0	0.0	—	8.60
13 Jan	—	5.3	—	—	2.2	—	—	3.75
14 Jan	—	5.2	—	—	0.4	—	—	2.80
15 Jan	—	6.4	—	—	3.1	—	—	4.75
16 Jan	—	5.7	—	—	−2.4	—	—	1.65
17 Jan	—	9.1	—	—	−1.5	—	—	3.80
18 Jan	—	8.3	—	—	1.1	—	—	4.70
19 Jan	—	7.5	—	—	2.5	—	—	5.00
20 Jan	9.2	5.8	−3.4	—	−0.9	—	—	2.45
21 Jan	—	7.7	—	−2.1	1.5	+3.6	—	4.60
22 Jan	—	7.1	—	—	2.4	—	—	4.75
23 Jan	7.4	6.8	−0.6	—	1.5	—	—	4.15
24 Jan	5.4	5.5	+0.1	−1.3	−1.3	0.0	2.05	2.10
25 Jan	6.2	6.3	+0.1	0.4	−0.3	−0.7	3.30	3.00
26 Jan	6.4	6.4	0.0	0.1	−0.1	−0.2	3.25	3.15
27 Jan	6.9	6.8	−0.1	1.5	1.0	−0.5	4.20	3.90
28 Jan	—	11.6	—	2.4	2.4	0.0	—	7.00
29 Jan	—	12.9	—	—	4.6	—	—	8.75
30 Jan	13.6	13.6	0.0	—	7.8	—	—	10.70
31 Jan	12.4	12.4	0.0	3.2	7.9	+4.7	7.80	10.15

Maximums

4th January 2000 5.1°C higher manual reading. No reset had been done from whenever before. Completely worthless junk manual reading knowingly archived.

6th January 1.7°C higher manual reading. Classic carry over of manual maximum reading from previous day – this is only visible due to the automatic comparison to check. This is an event that would not have been noticed prior under sole manual observation and could occur at any time under LIGT reading protocols but would not using a PRT at an automatic site or manual PRT/Data Logger since 2017.

9th January 3.8°C higher manual reading. No reset – junk.

20th January 3.4°C higher manual reading. No reset – junk.

23rd January 0.6°C higher manual reading. Possible carry over of maximum or more likely no rest junk.

Minimums

21st January 3.6°C lower manual reading. No reset – junk

25th January 0.7°C higher manual reading. No known reason.

27th January 0.5°C higher manual reading. No known reason.

31st January 4.7°C lower manual reading. No reset – junk.

Given that there were only 14 days of manual readings taken, 9 of them produced unreliable readings by margins up to 5.1°C. The incredibly alarming issue is that these junk readings passed through quality control (or should have but possibly did not go there ) and were subsequently archived into storage. Who can confidently state what the maximum temperature on, say, 4/1/2000 actually was?

The Historic Station data for the month of January does NOT indicate any estimated readings nor omissions but neither the averages of manual nor automatic readings derive the quoted monthly figures. Only the hybrid system I used came to the quoted monthly averages. Did the Met Office really use obvious JUNK – looks that way so far!

How bad was December?

Despite no missed days of manual readings there were still 6 days where readings were well in excess of being explained by variances of instrument readings accuracy. Margins of 1.5 °C are great enough to noticeably move even monthly averages.

Implications

What becomes even more concerning are the associated implications.

1. Prior to the joint systems running there was no way of identifying “carry over” events caused by the limitations of the traditional reporting method.
2. Without checking in extreme detail there is no obvious way of identifying the lack of reset due to missing days readings. The Historic stations data is clearly not genuinely showing omissions as demonstrated here by Sutton Bonington..
3. A single days missing reading could add in double digit individual reading discrepancies large enough to skew an entire months average.
4. The substitution of readings (modified or otherwise) from other sites to infill missing days does NOT eliminate the bogus higher/lower readings from failure to reset. It actually specifically appears to include these junk readings.
5. Readings substituted into, out from, or homogenised from other stations seem able to include junk readings. A gap between two clearly wrong readings may easily be filled by a wrong reading from elsewhere.

The Met office makes numerous claims about the quality of its stations siting that are quite obviously blatantly untrue when those sites are actually visualised and studied in detail as I have repeatedly demonstrated in my reviews. At Sutton Bonnington (and many others) the same is true of the data quality. Demonstrably junk readings are routinely archived and can propagate throughout the averaging systems with no simple way of spotting them. The scale of these errors runs into two orders of reporting magnitude and thus can have substantially corrupt even large homogenised bases on a completely random basis.

In part three I will further analyse the months of February and March 2000 data where the scale of errors is substantially greater still despite manual readings returning to better frequency. This 3 part report will demonstrate the catastrophic failures in Met Office datasets and how often these have occured.

Source: https://talkshop.blog/2026/09/19/sutton-bonnington-part-2-why-absurd-readings-occur-what-the-potential-effects-are-and-whether-they-occur-all-the-time-unseen/