Baltasound No 2 WMO03002 – Spliced datasets, absurd comparatives and an AI hallucination of a 1971 Space Port.

Baltasound is (possibly) the Met Office’s most northerly official weather station and situated on the Island of Unst. I use the term “possibly” because there is a site further north also on Unst but its precise status is equally possibly either an AI hallucination or not one used for climate reporting purposes even though it probably/possibly is/was – who knows?

Below is the current location of Baltasound Number 2 located alongside the service road/taxiway to the now defunct airport only used for emergency purposes. Unst has a total population of around 650, less than half the roll of the Grammar school I attended and yet Unst manages its own secondary education school – life must be very different growing up here. The airport site is rather a bleak spot though from all images available it has to be one of the best maintained compounds in the Met Office network. Whoever is responsible for the grass cutting is remarkably diligent.

60.748413 -0.855426 Met Office CIMO assessed Class 4 Archived temperature records from 1/1/1994

CIMO guidelines become almost irrelevant in these sorts of circumstances. There is no vegetation likely to cause shading however low an angle the sun may regularly get. The fuel tanks may cast some shadow, the small pond is artificial to control drainage to the airfield and large areas of tarmac are not the best neighbours. Whether or not any of this is significant enough to affect readings seems improbable. Temperature readings here will likely represent the largely coastal dominated climate of an island which is over 100 miles nearer to Bergen, Norway than Edinburgh. Class 4 is correct but readings should be representative of the wider area of Unst – but then again! If there is a number 2 station there must have been a “number 1” that was deemed to have sufficient climatological difference to require renaming. This was simply known as Baltasound and was further north.

60.75857, -0.88800 No Known CIMO assessment. Digital archived temperature records from 1/1/1914 to 16/10/1995

The original Baltasound site was a very long term site. Temperature records are in digital format from 1914 but manual records are available to view online back to 1904. Unst lighthouse has known records back to 1866 and it is very likely that there are much earlier records stored somewhere from this extreme UK outpost. It is even possible that the Norwegian authorities kept their own records for similar strategic purposes. The original site was at what is now that island charmingly small secondary school. The important thing to note is that the two sites are close together.

Realistically it is very difficult to see how such an outpost can contribute to the overall climate record of the UK being so distinctly separate. Naturally it is a key location for forecasting but climate recording from a site so distant seems faintly ridiculous – using Dutch, Belgian, French stations would make much more sense – the Polish border is nearer to London than Unst. However, nothing seems to deter the Met Office – not even using stations over 140 miles distant!

I find this sort of contrived representation more an act of Civil Servants filling vacant tick boxes rather than anything resembling “climate science” or meteorology. Does anyone really think a “Gridded Cell” on Unst is a realistic climate representation?

Interestingly though this “record” goes back to 1960 which obviously predates the installation of the No 2 site so I felt it worth a brief comparison of the two sites as there was an observations overlap period from 1/1/1994 to 16/10/1995. As ever there are a lot of interesting facts discovered when studying RAW data and not simply believing adjusted, amended, estimated or homogenised derivatives.

Baltasound number 1 was a manually observed site with human taken readings once daily at 09:00 hours. Such an activity in a leafy back garden in Stratfield Turgis, Hampshire is one thing but 09:00 in Baltasound is still dark night time with sunrise not for nearly 20 minutes in deepest December. Taking readings by artificial light must have been much more difficult in the very early 20th Century and not easy even now. Time of Observation bias will be very evident in this northerly location close to the Arctic Circle. Conversely the number 2 station was automated and archiving readings for 09:00 and 21:00. This changeover causes a change of recording protocol and all that entails.

Firstly for clarity, the number 1 site has just one maximum and minimum reading per day. The UK meteorological day is deemed to run from 09:00 one day to 09:00 the following day. At such once daily manually observed sites (the historic norm) the minimum is allocated to the day of reading, whilst the maximum is allocated to the PREVIOUS day. This is based on the known false assumption that the highpoint is always during daylight hours – in reality that is often not the case. For stations which archive readings twice daily the minimum is whichever is the lowest reading of that day’s 09:00 OR the previous day’s 21:00 reading if it was lower – it often actually is. Maximum readings follow the same principle. No “assumptions” are made.

Changing from once daily reading protocol to multiple readings immediately introduces variations – I opted to initially compare just the first 2 weeks of the initial overlap period of the two sites i.e. 1/1/1994 to 15/1/1994 to ascertain if there were any significant issues.

Baltasound Original. The two right hand colums indicate maximums and minimums

Then Baltasound Number 2

For the avoidance of any doubt {and to put down those people who send me offensive emails trying to tell me I am an idiot who does not know what I am talking about} I asked Grok AI to confirm which readings represented the maximum and minimums for 9/1/1994 at Baltasound No 2.

I then asked the same question for Baltasound number 1 i.e. comparing the same days readings.

This shows the minimums vary from 3.5°C to -1.5°C, a variance of 5°C between the two sites. I will analyse these differences further, but it is quite clear this change of recording protocol is inducing statistical warming where none actually exists in exactly the same way noted at many other sites were overlapping data was available such as Dalwhinnie, Cassley and probably hundreds of other sites on conversion to automatic reporting. None of those first 14 days readings matched each other.

To quote AI analysis of this issue

“This is a classic and important example of inhomogeneity in weather station records. Baltasound and Baltasound No. 2 are two very closely located stations (hundreds of metres apart at most) on Unst in Shetland, but they operated under different observation protocols:

• One (likely the traditional climate station) used the standard once-daily 09:00 observation. This captures the maximum and minimum over the previous 24 hours (09:00 previous day to 09:00 current day).
• The other (likely a synoptic station) used twice-daily observations at 09:00 and 21:00, from which daily max/min are derived by combining the two periods.

Why the figures differ on 9 Jan 1994

• Once-daily (09:00) dataset: Max 6.4°C, Min -1.5°C
• Twice-daily dataset: Max 6.8°C, Min 3.5°C

This discrepancy is not surprising.”

If Grok AI understands this problem, I can 100% guarantee that every qualified meteorologist at the Met Office is fully aware of it……whether all “Climate Scientists” are aware, or even care, is much more debatable. What is certain is that indications of ongoing “warming” are very much artefacts of changing recording methodology rather than actuality. Just one reading per month with this scale of discrepancy will significantly slew figures upwards, several of them make a complete mockery of all the data.

For my part, I intend to produce an alternative GB historic temperature record but will concentrate only on mainland England Scotland and Wales – stations such as Baltasound will not feature as they are both too unreliable and unrepresentative outliers.

In conclusion, though, a typical quirk – not all AI is equal. Google dominates search engines but its Gemini AI seems very poor indeed and I avoid using it as a first point of call, though I do occasionally use it to cross check other sources. I asked if Baltasound genuinely was the most northerly Met Office weather station. It got very confused over digital coordinates and needed to be corrected (by me) that Lerwick is not further north. However, it then decided that it knew better and decided that the “Spaceport” had a “new” Met Office weather station.

“The Status of Saxa Vord

• A Dedicated Spaceport Station: The Met Office collaborated with the ⁠SaxaVord Spaceport to install an automated weather station specifically to provide data for rocket launch go/no-go decisions.“

I pointed out that the Saxa Vord weather station (at the RAF Remote Radar Head) had its weather station installed in 1971, was not at the spaceport despite what it thought or could reference, and it was never a climate reporting site. {It does, however, claim to have a legendary UK wind speed record before the anemometer blew away!}

The amusing Gemini response was an admission that it had made it all up and “hallucinated”!

“You have got me dead to rights on this, and I completely hallucinated the context about the spaceport to justify my previous answer.

You are completely correct: Saxa Vord has been a Met Office station (WMO 03002) since 1971. It has absolutely nothing to do with a modern spaceport initiative; it was established decades ago to support RAF Saxa Vord, a vital military radar and early-warning facility operating on the hill during the Cold War. “

A cautionary codicil that perhaps the younger generation may need to consider – AI can actually invent information simply to justify itself.

Source: https://tallbloke.wordpress.com/2026/06/13/baltasound-no-2-wmo03002-spliced-datasets-absurd-comparatives-and-an-ai-hallucination-of-a-1971-space-port/