Two instruments, six inches apart, in the same screen, reading the same air. On overcast nights they agree to a tenth. On clear ones the glass reads about two degrees colder than the electronics.
I have kept a climate station in my garden for twenty nine years and it reports to the national network.
The specification is the point of it, so here it is.
A Stevenson screen: a white louvred box on legs, thermometers at one and a quarter metres above short grass, at least four times the height of the nearest obstruction away from it. Double louvres, ventilated, painted white, and repainted when it dulls.
Inside are a dry bulb, a wet bulb, a maximum thermometer and a minimum thermometer. The minimum is a spirit thermometer lying horizontal with a little dumbbell index inside it. As the temperature falls the meniscus drags the index down. As it rises the spirit flows past and leaves the index where it was. So the index marks the lowest the air got, and you read it in the morning and reset it with a magnet.
There is no electricity in any of that and nothing in it to go wrong: it is a piece of glass with a thread of spirit in it and a scrap of metal that gets pushed along, and the design has not needed changing since the eighteen forties.
Beside it, six inches away, in the same screen, is a platinum resistance sensor logging every minute to a box in my shed, which is the modern way and which the network prefers.
The two of them agree, which is the whole reason I have never had to think about either of them.
On an overcast night they agree to a tenth of a degree, which is as well as two instruments can agree, and they have for the eleven years the electronic one has been in.
On a clear night the glass reads lower.
Between one point eight and two point four degrees lower, always inside that band, and only ever on the nights with little or no cloud in them.
The electronic sensor logs a minute by minute trace, so I can see exactly what the air did. On a clear night it falls the way it should, bottoms out at about half five, and comes up with the light.
The glass says the air got two degrees colder than that trace ever shows. Not for a moment. All night.
The obvious answers, and there are three good ones.
The index has slipped. A minimum thermometer can be knocked and the index moves, which reads as a false low. It is the commonest error there is with them, and it is why you handle the thing carefully and why the screen door is on the north.
I have three minimum thermometers. Two in the screen for eight years and one kept as a check. All three agree with each other and disagree with the electronics.
Radiative cooling. This is the real physics and it is why I took four years to be sure.
On a clear still night an object radiates heat to the sky and can go colder than the air around it. That is why frost forms on a car roof when the air is at plus two. A thermometer in a screen is shielded from the sky, which is what the screen is for, but no screen is perfect.
If it were that, the glass and the platinum sensor would both be affected, because they are six inches apart in the same box. They would drop together.
And it would be worse in a screen that had gone thin or grey. Mine was replaced in 2019, new, to specification, and repainted last spring.
Response time. A glass thermometer is slower than a platinum sensor, so a sharp brief dip could be caught by one and smoothed by the other.
That works the other way round. The fast instrument catches a dip and the slow one smooths it away, and what I have is the opposite: a slow instrument finding a minimum the fast one, sitting six inches from it and reading sixty times a minute all night, never once saw.
What I do with the readings.
I report the glass, and I would go on reporting the glass if somebody came and told me not to. That is what the network wants: the observation is the observation, and the whole value of a twenty nine year record is that it was taken the same way for twenty nine years.
The electronic trace goes in as a separate stream and somebody at the other end quality controls the two against each other, and on clear nights my station is flagged, and has been for eleven years.
I have had two letters about it. The first suggested I check the index and the screen, which I did. The second, in 2021, from somebody more senior, said that a consistent offset under clear sky conditions was not unusual in older screens and that they would apply an adjustment.
So there is a correction factor sitting in the national record for my garden, applied to clear nights, and I know it is not a screen problem, and I have never said so.
The one thing that is not a measurement.
I read the minimum at nine in the morning, which is the standard hour.
For about the last three years, on the clear nights, there has been condensation on the inside of the glass at the bottom end of the tube, in the couple of degrees below where the index has stopped.
You do not get condensation inside a sealed spirit thermometer. It is sealed at manufacture with the spirit and a little inert gas and there is no water in it to condense.
All three. Only those nights. Gone by eleven.
I have photographed it perhaps forty times, with a hand lens and a phone, and the photographs show what I have described and nothing more, and I have not sent them to anybody.
What I have done instead is start reading at eight rather than nine on clear mornings, because at eight there is more of it, and I find that I want to know how much.