Why the forecast says 29 °C / 84 °F and sunny and your garden says 24 °C / 75 °F and raining

Your phone said this afternoon would reach 29 °C / 84 °F, sunny all day. You go out at three, and the thermometer screwed to the fence says 24 °C / 75 °F while it rains on you. One of them has to be wrong.

Neither is. They are answering slightly different questions, and the gap between those two questions turns out to be most of what makes flying ants hard to predict.

Weather models don't think in places. They chop the atmosphere into a grid of boxes, work out what each box is doing, and hand you the answer for whichever box you happen to be standing in. The service this app takes its weather from gathers models from national forecasters all over the world and picks the finest one that covers wherever you've asked about. Those run from around 1 to 2 km / 0.6 to 1.2 miles a side for the local models, out to 9 to 11 km / 5.6 to 7 miles for the ones that cover the whole planet.

So the forecast for your postcode is really the average state of a slab of air a few miles wide. You are one point in it. So is the industrial estate, the reservoir, and the wood on the north side of the hill.

Past weather is blurrier still. Anyone trying to work out what the weather was doing on a day a flight happened, us included, ends up leaning on ERA5, the reconstruction that covers the whole planet hour by hour back to 1940. It does that in boxes about 31 km / 19 miles across, which is wider than most cities. One number for the town centre, the ring road, the valley, and the farmland past it, all of it on the same afternoon.

Temperature is the gentle version of this. Rain is where it gets silly, because a summer shower is very often smaller than the box it lands in.

The kind that builds on a hot afternoon and soaks one end of town for twenty minutes leaves the model with a single rainfall figure to hand you for the whole box, and what it hands you is the average. A downpour over a tenth of the box comes out as a light sprinkle across all of it. You read 2 mm / 0.08 in; what actually fell was 20 mm / 0.8 in on one housing estate and nothing at all on the rest.

You can see forecasters working around this if you look at how the models are built. The Met Office runs its global model at about 10 km / 6 miles, which is nowhere near fine enough to grow an individual shower, so over the UK and Ireland it runs a second model on a 1.5 km / 0.9 mile grid inside a 4 km / 2.5 mile one. That finer grid exists more or less entirely because showers are smaller than the coarse boxes. Not every country has one.

Rain cuts both ways for ants, which the temperature gap doesn't. A few days of it beforehand softens the ground and is part of what sets a flight up. Rain on the afternoon itself usually ends one. So a shower narrow enough to be invisible to the forecast is still wide enough to shut down your street while the next village flies.

Air gets colder as it rises, by something like 5 to 10 °C / 9 to 18 °F for every kilometre / 3,300 feet of height, damp air cooling more slowly than dry. You don't need a mountain for that to bite.

The map draws the world as a mesh of hexagons. In hilly country, one hexagon at the size we used to look weather up at could comfortably hold 1,000 m / 3,300 ft between its lowest ground and its highest. Valley floor and ridge top, same box, same forecast. They are obviously not having the same day, and no amount of cleverness in the model repairs that, because the problem is the size of the box rather than the physics happening inside it.

Which way a slope faces does much the same job with no height involved at all. In Britain a south-facing bank collects hours more direct sun than the north-facing one across the valley from it, and the two are often in the same box.

Worth telling, because it is the same mistake in miniature and it was ours.

Until recently the coloured hexagons were cheaper to draw than they should have been. Instead of looking up the weather for every hexagon, we looked it up once for a larger patch and painted seven neighbours from that one reading. On flat ground nobody would ever notice. In hills, the reading could be coming from a point up to 10 km / 6 miles away, and at a completely different height, from the hexagon it was colouring.

Then somebody tapped one. Tapping a hexagon opens a panel that goes off and fetches the weather for that exact spot, properly, no shortcut. So the colour on the map and the detail underneath it were quoting two different afternoons, and a user wrote in to say seven hexagons were all showing good conditions while five of the panels behind them said middling.

They were right and the map was wrong, in the specific sense that the map was describing somewhere else. Every hexagon now gets its own weather reading and nothing is painted from a neighbour's data.

It wasn't free. Zoomed in close, filling a screen now takes twenty to fifty times as many weather lookups as it did before, which is a real bill, and we watch the meter on it. Still an easy trade. I would much rather explain a bill than explain why the map contradicts itself when you touch it.

Even one reading per hexagon, we are still describing a few square kilometres / a couple of square miles with a single set of numbers. Below that it all goes dark: the brick wall that has been in full sun since eleven in the morning, the tarmac drive still handing heat back to the air at six in the evening, the shaded corner of a lawn that stays damp two days longer than the rest of it.

Those are differences of metres / feet. No public weather model gets down there, and honestly I don't expect one ever will. So your garden carries a permanent offset from the forecast, and about the most useful thing you can do with a forecast is learn roughly what your own offset is. Mine runs hot, and the park at the end of the road is a frost pocket every October.

You could file microclimate under trivia if ants ignored it. They don't.

In 1981 two Dutch researchers, Boomsma and Leusink, published three summers of flight watching from a Wadden island and the country around Amsterdam. Four species, and one tidy idea out of it: queens go up when the air has warmed to about the temperature of the soil their nest sits in. They measured it that literally, air at 20 cm / 8 inches above the ground against soil at 5 to 7 cm / 2 to 3 inches down.

The detail that sticks with me is what they found next door. The two temperatures matched in the habitat of whichever species was flying, and didn't match in the neighbouring patches where the other ants lived. Same island, same afternoon, a short walk apart.

People still argue about how widely that idea holds, and it may well be one cue among several. But the shape of it is the awkward part. Whatever the queens are checking before they launch, a good deal of it lives at a scale no grid of boxes can see.

So two houses on the same street get the same forecast, the same number on the same screen, and completely different afternoons. There's a biological half to that answer as well, and why your neighbour's ants flew and yours didn't covers it.

None of this is a reason to distrust a forecast, incidentally. It's a reason to know what you're holding. We publish the reasoning behind the score for the same purpose: a number that turns up with no account of where it came from invites more faith than it has earned, and one built out of boxes a few miles wide has earned a fairly specific amount.

Boomsma, J.J. & Leusink, A. (1981). Weather conditions during nuptial flights of four European ant species. Oecologia 50(2): 236–241. DOI 10.1007/BF00348045. Where the soil-temperature idea comes from, along with a set of per-species flight temperatures that overlap far more than you would guess.

Hart, A.G., Hesselberg, T., Nesbit, R. & Goodenough, A.E. (2018). The spatial distribution and environmental triggers of ant mating flights: using citizen-science data to reveal national patterns. Ecography 41(6): 877–888. DOI 10.1111/ecog.03140. Open access. Strong weather associations at the scale of a whole country, and silent, as it must be, about your fence.

For the neighbouring arguments, ants don't read the barometer covers which bits of weather carry the signal, and what is a nuptial flight covers the event itself.