How the forecast is made
Four steps, every hour: watch the snow fall, run every model worth running, carry each one onto the mountain band by band, and score the result at a gauge that sits at a surveyed height. Every number on a resort page is one of three things: a model's own output, a quantity derived here, or an observation.
The pipeline
Watch it fall
- MRMS precipitation rate and surface precipitation type, 1 km every 2 minutes
- SNOTEL hourly precipitation, snow water and temperature at surveyed heights
- NOHRSC snow depth and water analysis, nightly
- COOP and CoCoRaHS east of the Mississippi, where SNOTEL does not reach
Run every model
- 0–48 h: HRRR and RRFS at 3 km, the REFS ensemble for spread
- 0–11 days: NBM v5 calibrated percentiles, snow level and snow ratio
- Days 3–15: ECMWF IFS and the 51-member ENS at 9 km, the AIFS ensemble
- Northern tier: Canada's HRDPS at 2.5 km
Carry it onto the mountain
- Temperature read from each model's own vertical profile at each band's height, never a fixed lapse rate
- Precipitation reshaped by PRISM over the model's effective resolution, blended with an upslope term from its low-level wind
- Snow line from the wet-bulb profile, checked against the NBM snow level
- Snow ratio from band temperature and the model's own, kept as both water and depth
Weight by what verified
- Blend weights each model by its trailing error at the resort's own SNOTEL
- Percentiles come from the pooled ensembles, calibrated against past storms
- Scorecards per resort and per model, published every storm
The inputs
Snowfall from HRRR and RRFS is depth with the model's own snow ratio applied; ECMWF snowfall is water equivalent. Plotted on one axis they differ by the snow ratio, a factor of ten to twenty, so both are carried and every number says which.
| Source | Grid | Reach | Job here |
|---|---|---|---|
| MRMS | 1 km | Every 2 minutes | Where it is snowing right now, and where it is raining |
| SNOTEL | Station | Hourly | Truth at height; the score for every western resort |
| HRRR | 3 km | 18 h hourly, 48 h every 6 h | Lake-effect bands, timing of the front |
| RRFS v1 | 3 km | 18 h hourly, 84 h every 6 h | HRRR's successor, carried alongside it |
| REFS | 3 km ensemble | Short range | Spread for the first two days |
| NBM v5.0 | 2.5 km | 11 days, hourly to 48 h | Calibrated snow percentiles, snow level, snow ratio, snow depth |
| ECMWF IFS | 9 km | 10–15 days | The synoptic anchor |
| ECMWF ENS | 9 km, 51 members | 15 days | Ranges for days 3–10 |
| AIFS ENS | 0.25° | 15 days | Machine-learned check on the pattern |
| HRDPS | 2.5 km | 48 h | Northern-tier resorts and the Canadian side of the border |
| PRISM | 800 m | 1991–2020 normals | Where terrain puts the snow that a model cannot resolve |
The lifts define the mountain
An aerialway is the one object in OpenStreetMap whose two ends are, by construction, the bottom and the top of ground somebody rides. Reading the terrain under all 4,183 lower-48 lifts and taking each area's lowest and highest lift terminal reproduces published verticals to a median of about 50 feet on a 27 m DEM. Most of the residual is one-signed: a resort quotes its peak, and this measures its highest terminal.
Each resort is cut into 500-foot bands over the lift-served footprint, the convex hull of the uphill lift lines buffered by 400 m. Each band carries its area, mean slope, mean aspect and the PRISM winter normal at that height. A band's aspect is the direction of the mean unit vector, so 350° and 10° average to 0°, not 180°.
The lapse rate is not 6.5 K/km
Measured on 432 paired western SNOTEL stations (under 25 km apart, more than 300 m apart in height) over DJF 2025–26, 37,582 station-pair-days:
| K/km | Mean | Median | 10th | 90th | Inverted | Within ±1 of 6.5 |
|---|---|---|---|---|---|---|
| Daily mean T | 3.33 | 4.31 | −4.29 | 9.37 | 22.5% | 18.2% |
| Daily min T | 1.53 | 2.60 | −9.03 | 10.07 | 36.3% | 12.4% |
| Daily max T | 6.00 | 6.67 | −0.95 | 12.03 | 11.8% | 17.7% |
A fixed 6.5 K/km is right about one winter day in six and costs 7.7 °F at the summit of a 3,000-foot mountain before the model adds any error of its own. Overnight the mountain is upside down more than a third of the time. A better constant does not rescue it: the daily spread is enormous (σ = 5.87 K/km) and surface predictors explain r² = 0.07. So the profile has to come from the model's own resolved vertical structure, which is the only thing that knows an inversion is on.
SNOTEL sites sit in clearings that pool cold air, so these numbers likely over-represent inversions relative to an exposed slope; the pairs are western only; and some of each difference is horizontal, not vertical.
Terrain carries the precipitation
Precipitation uses PRISM 800 m normals the way the NWS Mountain Mapper does: as the shape of terrain precipitation a model cannot resolve. The ratio R = PRISM(x) / mean PRISM over the model's footprint is about 1 across the footprint, so it conserves the model's total while redistributing it the way four decades of gauges say the terrain does.
The footprint must be the model's effective resolution, roughly six to seven grid lengths, not its grid spacing. Over a 3 km box the HRRR ratio comes out 1.00–1.03 and appears to do nothing; over the ~20 km HRRR actually resolves it spans 0.80–1.30. A ski mountain is about 5 km wide, entirely inside the unresolved band. R is climatological, so it encodes the average storm's wind; it is blended with an upslope term from each model's own low-level wind and published per band rather than hidden inside a number.
What a band forecast cannot do
A band resolves 500 vertical feet of a mountain, not a named run. Wind loading, scouring and cornices are not modelled, and a 13 m DEM cannot see the gully that holds snow. Eastern verification is thin, because SNOTEL is a western network. Snow-to-liquid ratio remains the largest single error in turning water into inches.
- Winter mean lapse rate
- 3.33 K/km
- Station-pair days
- 37,582
- Overnight inversions
- 36.3%
- Effective resolution
- 6–7 Δx