Chances, not promises
A forecast that says “12 cm of snow” is one guess dressed as a fact. This page runs the whole ensemble instead — every member is a complete, self-consistent week — and counts how many of them deliver the day you are hoping for. Nine out of thirty is a real answer. It is also a different one from “probably not”.
Where, and how high
The ensemble is run at the summit and moved to the height you pick, which decides where the snow line falls.
The odds, day by day
The share of ensemble members whose own day meets each description. Click a column to see what those members actually look like.
The week in numbers
Middle of the distribution first, with the range the members actually cover beneath it. A wide range is information, not a failure.
Every member, drawn
One line per member. Where the lines fan out evenly the forecast is simply uncertain; where they split into two bunches, two different weathers are on the table and the average of them will not happen.
What the learning actually changed
Trained once a day by the scheduled pipeline on the last weeks of archived forecasts against ERA5-Land reanalysis, and checked on a block it never saw. If it does not beat the raw ensemble there it is switched off, and this panel says so either way.
How much to trust a percentage
- An ensemble is a sample of one model’s doubt, not of reality’s. Every member comes from the same forecasting system with the same blind spots. Reality lands outside the members more often than the member count suggests — ensembles are systematically overconfident, and more so for precipitation than for temperature.
- Only the rain-or-not figure is calibrated. That one has been checked against what actually happened, and the panel above reports its Brier score against the uncorrected ensemble. Every other percentage on this page is the raw share of members, shown as exactly that. A calibrated number and a counted one are not the same kind of claim.
- Confidence falls off a cliff after about five days. Day six and day seven are in the outlook because leaving them out would not make them less tempting to guess at. Read them as the shape of the week, never as a plan.
- The members were run at one height. Moving them to another band applies the same lapse rate the forecast uses and decides rain against snow again at that height. That is the dominant effect near the snow line and it is not the only one — nothing here knows about aspect, shading or where the wind stacks the snow.
- A 30 % chance of a powder day is not a small number. It happens roughly one week in three. The mistake is not reading 30 % as unlikely — it is reading 70 % as certain, and then being surprised on the day it goes the other way.
Every threshold on this page — what counts as a powder day, a storm or a bluebird — is listed on the method page.