The vote
Every forecast on the site is made three times, once with each of the ECMWF, GFS and ICON models. For any period we count how many of the three forecast meaningful snow. Three is Likely. Two, or a split, is Possible. When none does, the label says how confidently the models agree it will stay dry. The methodology page gives the thresholds for each period.
The label is shown as dots next to the headline, and the same count runs through the maps, the rankings and the day-by-day table on each page, where each day shows how many models put snow on it.
Why not a percentage
A percentage would look more precise, and it would be less honest. Three models are not enough to build a probability from. With three votes the only possible answers would be 0, 33, 67 and 100 per cent, and calling a two-to-one split “67 per cent” would dress up a disagreement as a measurement.
The precipitation chances that most apps show come from one forecast centre running its own model many times with small changes to the starting conditions. That is a real and useful thing, but it measures how sure that one model is of itself. It cannot tell you that a different centre, using different physics, sees the storm differently. Comparing three independent models can, and a plain count of how many agree is the simplest true statement of what that comparison found. The guide to the three models explains how they differ.
Watching a label move
Forecasts are made every three hours, and a day’s label often changes as the day approaches. Here is what happened to three days between the run of 18 September 2026 at 1pm UTC and the run of 21 September at 6am UTC.
Mt Hutt, Sunday 27 September. On 18 September the day was nine days out, in the 14-day outlook, where only two models reach. One of the two showed snow, so the page said “Chance of snow, 0–4 cm”. By 21 September, six days out and with all three models in play, all three had snow, 10–25 cm of it, with rain mixed in. The day went from a faint chance to Likely.
Valle Nevado, Friday 25 September. On 18 September, one of two models showed snow, “0–10 cm”. On 21 September, all three did, “5–15 cm”. A Possible that firmed up.
Cardrona, Saturday 26 September. On 18 September, one of two models showed snow, “0–5 cm”. On 21 September the day read rain, with no snow from any model. A chance that evaporated.
That is the normal pattern. Beyond a week a single model’s snow is a hint. Inside a week, the number of models on it is the thing to watch. A day that moves from Possible to Likely across successive runs is one to plan around. A day that stays Possible as it gets closer is one where the models are split on something real, usually the track of the storm or the freezing level.
What to do with each label
- A day or two out: Likely is as sure as a snow forecast gets. Possible means look at the range and the freezing level; the disagreement is usually about how much or whether it turns to rain, not whether something is coming.
- Three to five days out: Likely is worth a plan. Possible is worth checking tomorrow.
- Six days and beyond: both labels are about the pattern, not the day. Where the snow lands and on which day still moves. Look at the whole week’s total and the trend across runs.
Rain or snow
The same vote settles the question of rain against snow when the temperature is near freezing. When at least two models show rain and at least two show snow in the same hour, the page says “Rain or snow” rather than picking one. The guide to village rain and mountain snow covers what that means for a resort.
We keep every forecast the site has published. When there is enough history, this guide will say how often Likely and Possible days actually brought snow.