Useful weather alerts

Useful weather alerts tells you when frost or rain is likely at an exact place such as your garden, an allotment, a paddock, rather than defaulting to the nearest town.

Why a general forecast is not enough

Most forecasts take an average for a grid square several kilometres across. For frost (in particular) this doesn't work very well: for instance cold air drains downhill and pools in hollows, a south-facing slope loses less heat overnight, and a sheltered corner behaves quite differently from an exposed one.

To produce the forecast that powers the alerts, I start from the Met Office forecast and then correct it for the ground under your pin — height, slope, which way it faces, how much open sky it sees, how cold air moves across the surrounding land, and the long-run climate of that square kilometre.

I also predict the grass minimum temperature, which is what matters if you are growing anything. Grass frost forms while the air is still above freezing, usually by a degree or two. This means a forecast can promise 3 °C tonight, but you'll still lose your seedlings.

Frost: one alert, on evenings when frost is likely overnight

Ground frost. My model applies a hyperlocal correction to the Met Office forecast and catches about 74% of ground frosts, and when it warns you, it is right about 85% of the time.

Air frost. My correction catches about 75% of air frosts and is right 85–86% of the time: this is roughly in line with the Met Office forecast, but with roughly half the false alarms.

Rain: two alerts on rainy days, morning and evening

Rain is not local the way frost is. It is driven by weather systems tens of kilometres across rather than by the shape of your garden, so there isn't much I can do (for now) to improve on the Met Office forecast, so that's what you get, just recalibrated so it catches about 82% of any rain and is right 75% of the time it warns.

It gets better the more it is used

After alerts are sent, I'll sometimes follow up and ask for feedback. That feedback is the most valuable information I have, because it is measured at the precise spot the prediction was for, which no weather station network can offer. This is hugely valuable data and helps to make the model even more accurate over time.

The effect is local. The more people use the service in an area and say what happened, the more accurate it becomes for everyone in that area. If you are the first person in your valley, your answers are doing a great deal of work.

So please give feedback when we ask for it, and tell as many people in your local area as you can about this site ... the more feedback I get, the better the forecast (and the alerts) will be.