Type a place. You get its snow briefing: the National Blend's snowfall and ice for ten days with the blend's own range around it, the ECMWF ensemble two weeks out, the weight on the ground, and how good this forecast has been this winter — as one static page and card you can save, print and send.
Every 6-hour block the NWS grid carries: snowfall and ice accumulation as bars, temperature as the line. The freezing mark is drawn across it, because that is the line the whole forecast turns on.
Snowfall and ice to ten days, the blend's 10th–90th percentile range, density and plow load, snow level, what changed since the last run — on the map and at 1,538 places.
— Static · share it, print itOne page and one card per place — 971 cities, 527 ski areas, 40 passes — rebuilt every cycle. No map to drive, nothing to click: the forecast, its range, and its track record.
— Forecast · downscaledOne ECMWF run put on the real terrain at 2.5 km, and ~60 m in the big ranges, with base and summit totals for every ski area and pass.
— Observed · eighteen wintersThe NWS National Snowfall Analysis on one map: the average season, every winter since 2008-09, month by month, snow-day frequency and the biggest storms.
2.5 km · 193 sectors · 15,000 city valuesThe places with the most snow in the latest National Blend cycle, each linked to its briefing. Empty columns mean the blend has nothing there yet — which, in November, is the honest answer more often than not.
Past a week a single forecast stops meaning much, so this range is run as a 50-member ensemble, read straight from ECMWF's open data twice a day. The band is where 80% of the members put the snow so far, the line is the median, and the bars underneath are the share of members with an inch or more, and six inches or more, on the ground by then.
What this is. ECMWF's own snowfall field (water equivalent) at the nearest 0.25° land cell, drawn at a fixed 10:1 ratio because the ensemble publishes no snow-to-liquid ratio and this site will not invent one from a single level. It is a chance, not a total: read the band, not the line. The temperature plume and the freezing share are in every place's briefing.
Every morning the Situation Room scores the previous climate day — 12Z to 12Z, the window the NWS analysis publishes — against what the blend said one, two and three days before, on the analysis's own 2.5 km grid. No resampling, no cherry-picked storms. This is that ledger, this winter so far.
Mean absolute error and bias in inches over every scored cell, and the critical success index at six inches — did the blend draw the six-inch area where it fell? 1.0 is perfect; a coin would score near zero.
Two storms from opposite ends of the winter's problem space, each with every National Blend of Models run from four and a half days out, the blend's own percentile range around it, and the NWS snowfall analysis it was aiming at — on the same grid, in the same 24 hours.
The NWS grid stops at the border, so Canada runs on Environment and Climate Change Canada — — city points with current conditions and the public forecast.
Every number below is observed — November through March against the 1996–2025 normals, with each state's December–February mean ranked against every winter back to 1951.
National temperature departure for each month of the season. Blue is colder than normal, red is warmer; the number is the departure in °F.
Seasonal departure, where the Dec–Feb mean ranks among all winters since 1951, heating degree days against normal, and days at or below freezing. Click a column to sort.
| State▲ | Nov–Mar vs normal▲ | Dec–Feb rank▲ | Heating degree days▲ | Precip▲ | Days ≤32 °F▲ |
|---|
A winter forecast is several forecasts at different ranges, each from a different model. This is exactly which ones are wired into this site, which are published but not yet read here, and which still wait on a feed — no range is filled in with a model that is not actually running.