The Hours a Glass Ball Could Burn

For most of the last century and a half, a "sunshine hour" in a weather record meant a length of scorch on a strip of card, burned there by a solid glass ball. The ball is a strange and exact little machine. It is also, it turns out, a unit of measurement, and when Britain began replacing it with electronic sensors the sunshine went down.

1. A lens with an axis in every direction

A glass sphere focuses sunlight like any lens, but it has a property no ordinary lens has: every straight line through its centre is an optical axis. Light arriving along any of them passes the centre undeviated and comes to a focus on the far side, at the same distance, whichever way it came from. So the sphere never has to be pointed at the sun. As the sun moves, the focus moves by exactly the same angle, the other way round, on a bowl concentric with the ball. Put a card in the bowl and the sun writes its own track.

John Francis Campbell of Islay first let a glass ball scorch the sun's path into a wooden bowl in the 1850s (the Met Office says 1858; other accounts say 1853). George Gabriel Stokes replaced the wood with "a slip of card, which can be renewed from day to day", in a paper for the Royal Meteorological Society's journal dated 1880, and that is the instrument still in use.

The WMO's specification for the reference grade of recorder (the "interim reference sunshine recorder") asks for a sphere 10 cm across, refractive index 1.52 ± 0.02, sitting in a spherical segment of radius 73 mm. The paraxial focal length of a ball lens, measured from its centre, is f = nR / 2(n − 1); for a 50 mm radius at n = 1.52 that is 73.08 mm. The bowl is simply the sphere of focus. (The same table also gives a focal length of 75 mm for sodium light, which is what the formula returns at n = 1.50: the two numbers sit at either end of the stated glass tolerance.)

2. Why three cards, and why the middle one is straight

Because the burn is the sun seen straight through the centre of the ball, it keeps the sun's geometry exactly. The sun's hour angle turns at 15° an hour, so the burn does too, and the hour lines on the card are evenly spaced: unlike a flat sundial, whose hour lines crowd together at noon, the sphere needs no correction at all. The sun's declination stays nearly fixed through a single day, so the burn runs round a circle of fixed "latitude" on the bowl. At the equinoxes that circle is a great circle and the strip of bowl it crosses is a band of a cylinder, which flattens without distortion into a straight card. In summer and winter the circle is smaller, the strip is a band of a cone, and a cone unrolls into a curve.

Those shapes are forced, not chosen. A narrow band at declination δ on a sphere of radius f lies on a cone whose slant height is f / tan δ, so a summer card laid flat is an arc of radius about 16.8 cm at the solstice, and it bows the opposite way in winter. The instrument's bowl carries three pairs of grooves for the three cards, and the one straight card serves both equinoxes. The Met Office's Observer's Handbook gives the northern-hemisphere calendar: long curved cards from 12 April to 2 September, short curved cards from 15 October to the end of February, straight cards in between.

3. How bright is "sunshine"?

A card only scorches when the focused beam is strong enough, and nobody decided what "strong enough" was: the recorder decided. When meteorologists went looking for a physical definition that electronic instruments could share, the WMO Guide records that the threshold at which the card burns had been found to range from 70 to 280 W/m² of direct beam at different stations, and that further work with the reference recorder in France gave a mean of 120 W/m². Today's definition of sunshine duration is the time the direct beam exceeds 120 W/m²: a number chosen to match a piece of card.

On a perfectly clear day the three thresholds disagree only in the first and last minutes of sunlight, when the beam is climbing out of the thick air near the horizon. The real disagreements come on broken days, and in the other direction. A glass ball does not switch off the instant a cloud edge crosses the sun: the card is already hot, the scorch spreads sideways, and a string of short bright spells burns as one long mark. The Met Office's internal comparisons of the two instruments give this spreading of the burn in broken cloud, worst when the sun is high, as the main reason the ball reads high; on days of unbroken sun the two agree closely. The Met Office Observer's Handbook puts the width of a strong burn at up to 3 mm.

4. The day the sunshine went down

The Met Office publishes monthly records for 37 long-running UK stations as plain text files, free to download. In the sunshine column, a value from the Campbell–Stokes ball is unmarked and a value from an automatic Kipp & Zonen sensor carries a #. Sixteen of the stations switched, between November 2000 (Manston) and January 2019 (Ross-on-Wye), and none of them switched back for more than a few months. Others, Oxford and Armagh among them, carry no # anywhere, to the end of 2025.

That makes a natural experiment. For each station that switched, compare its sunshine with each neighbour that kept the ball, as a ratio, in the five years before its own switch date and the five years after. The weather cancels in the ratio, because neighbours share it; the instrument does not.

The drop is not a trend that happened to pass by. Lined up on each station's own switch date, which ranges over nineteen years, the median ratio stays within 2% of 1.00 through the eight years before, then falls by about a tenth in the first year of the sensor and stays down. Where nothing changed, the same arithmetic run at invented switch dates at stations that kept the ball scatters around zero.

Is this known? Yes, and that is the point of checking it from outside. The Met Office's public page of UK climate records says it plainly:

“On average, KZ sensors record slightly lower values than co-located CS instruments, so an upward adjustment of KZ totals is made to give a monthly ‘CS equivalent sunshine’.”

Behind that sentence are side-by-side comparisons at stations where both instruments ran at once. Its internal memoranda record a rough correction factor of 1.1, first applied in May 2002, later replaced by factors that vary with the season; the 2007 update (Perry and Legg, NCIC Climate Memorandum 27, revised 2011, from 489 station-months at 24 sites) puts the ball above the sensor by between 3.7% (December) and 15.2% (May), a mean over the twelve months of 1.096. From nothing but the public files and their # marks, with no overlap data at all, the natural experiment gives 1.10. The seasonal shape is harder to see from outside: summer steps are larger than winter ones, as the Met Office finds, but month by month the outside estimate is too noisy to confirm its curve.

What the step also shows is that the # values in the public station files are the sensor's own readings, not the adjusted "Campbell–Stokes equivalents" the Met Office uses for its summaries and grids (the files do not say which; if they were adjusted, there would be no step). Anyone who adds up the column without reading the footnote sees each of those stations lose about a tenth of its sunshine on the day its instrument changed.

And the direction is not a law of nature. It is a fact about these two instruments in this climate. In Poland, a CSD3 sensor at Borucino recorded 15% more sunshine than the glass ball over 2015–2021 (Owczarek and Malinowska, 2023), while Germany's weather service, with yet another kind of sensor, found the glass ball higher as Britain does, most of all in summer (Hannak et al., 2019). Each network has to measure its own seam.