How PowCall Forecasts Work

PowCall generates probabilistic 15-day snow forecasts for 120 ski resorts using ensemble weather models, terrain-aware corrections, and daily ground-truth verification. Every forecast includes uncertainty ranges so you can make better powder day decisions.

Forecast Pipeline

Each forecast update runs through a 5-step pipeline, executed 4 times daily when new GEFS model runs become available.

1

GEFS Ensemble Ingestion

We download all 31 members of the NOAA Global Ensemble Forecast System (GEFS) at 0.25-degree resolution. Each member represents a slightly different plausible evolution of the atmosphere out to 16 days. By processing all 31 members (not just the control run), we capture the full range of forecast uncertainty.

2

Orographic Precipitation Correction

Global models smooth terrain to ~25 km grid cells, dramatically underestimating precipitation in complex mountain terrain. We apply per-resort, wind-direction-dependent multipliers calibrated against 30 years of PRISM (Parameter-elevation Regressions on Independent Slopes Model) climatology data. For example, Alta receives a larger precipitation boost from NW flow than from SW flow, matching its Little Cottonwood Canyon geography.

3

Snow-to-Liquid Ratio (SLR)

Corrected precipitation (in liquid equivalent) is converted to snowfall using the Kuchera method, which calculates SLR based on the temperature profile of the lower atmosphere. Colder storms produce higher SLR (fluffier snow), while warmer storms produce denser, wetter snow. Typical values range from 8:1 (heavy/wet) to 20:1+ (ultra-light powder).

4

Probabilistic Aggregation

The 31 ensemble members produce 31 different snowfall estimates for each resort and time step. We extract the 50th percentile (median/expected) and the 80th percentile (a realistic upper estimate) to form a one-sided probability estimate: low is a fixed 0", since snowfall can't be negative, and high is the 80th percentile. The gap between the two shows how far the members spread above the median. That spread is the model's own; how often reported snowfall actually stays under the high estimate is measured separately, per lead time (see below).

Forecast v2 for North America: the snow by each morning

Notice, planned to take effect no earlier than 1 December 2026, after a live test, and not at all if the test is not passed; the date is confirmed here when the switch is committed: from that date every North American resort's forecast is stated as new snow by a morning, not for a calendar day. Each card is the 24 hours ending at 12:00 UTC (5 am Mountain standard time, 7 am Eastern, 4 am Pacific), the window the resort's own morning report covers, and is labelled by the morning it ends: "Tmrw AM", "Sat AM", "by Saturday morning"; between the resort's midnight and 12:00 UTC the first card reads "This AM". The number is a whole inch ("4", up to "7"); a long-lead card whose expected amount is 0 reads "Likely none" with its up-to and chance. Temperatures read about 5 °F higher on average (1.5 to 15 °F, depending on the resort) than under the previous form: the new form shows the ensemble's own 2 m temperature, without the previous form's elevation (lapse-rate) cooling. Australia, New Zealand and Chile keep the local-day form.

The amounts come from the same 31 GEFS members, totalled over the window and adjusted to each resort's own reporting history, with the median as "Expected" and the 80th percentile as "up to" (still a model estimate, up to the 80th pct.). Each card also shows the model's chance of at least 1" and 6" by that morning, written as "N in 10" ("unlikely" for 0, "very likely" for 10) and labelled model chance: it is the model's own number, scored against resort reports only at the end of the season, and it is never a confidence grade. The number shown in the evening is the number frozen at 5 pm and checked against the next morning's report; "1 day out" is the morning after the forecast.

What is scored does not change: the resort's reported 24-hour snowfall, paired with the morning it was reported, whole inches. The season-to-date count for the new form starts with the first forecasts shown in the new form (later for longer lead times) and is shown separately from the previous form's; the two are never pooled. Cards 12 to 15 mornings out are shown for their up-to and chance: the model has no measured skill at those leads. No padlock or lock claim appears on any card unless a certified lead is announced here.

5

Verification & Accuracy Tracking

The range we show is scored against the 24-hour snowfall each resort reports, the number skiers actually see. The range we score is the one we showed the day before, frozen so it can't be revised after the fact. We publish the measured share of days the report stayed within our range, together with how many snowy days that measurement rests on.

Forecasts are also checked against weather stations: SNOTEL (SNOpack TELemetry) stations operated by the USDA Natural Resources Conservation Service and, where a resort has none, NOAA GHCN-Daily stations whose observer reads the gauge each morning. These are independent physical measurements that do not depend on how a resort reports. Each reading is matched to the local day it measured: a SNOTEL value dated tomorrow is the snow depth at midnight, and a morning observer's total covers the 24 hours before the reading, so both describe today. Resorts with no such station are not station-scored; a model estimate is never used as the actual. We compute accuracy metrics at 1, 2, 3, 4, 5, 10 and 15-day lead times and publish them on each resort's accuracy page.

Data Sources

SourceResolutionUpdate FrequencyPurpose
NOAA GEFS0.25° (~25 km)4x daily (00/06/12/18Z)31-member ensemble for probabilistic forecasts
PRISM800mClimatologyOrographic correction calibration
Resort snow reportsPer resortDailyWhat the range is scored against
USDA SNOTELPoint stationsDailyIndependent physical check
NOAA GHCN-DailyPoint stations (morning observers)Daily, 3-7 days lateIndependent physical check where a resort has no SNOTEL station

Accuracy Metrics

We track forecast performance using several complementary metrics:

MetricWhat It MeasuresIdeal Value
Skill vs climatology1 − (our total error / the error of the resort's climatological average), on the station track; the number the leaderboard ranks by1
Station scoremax(0, 100 − |forecast − station| / max(station, 1) × 100), averaged over every day, dry days included. A dry day forecast at 0.02" scores 98%, so this number never ranks resorts and is never a headline; it is kept for continuity100%
MAEMean Absolute Error in inches0
BiasSystematic over- or under-prediction tendency0
RMSERoot Mean Square Error (penalizes large misses)0
Range Containment% of days the resort's reported snowfall did not exceed our high (80th pct.) estimate80% (target, not a claim)
Skill vs Climatology1 − (our total absolute error ÷ the total absolute error of the historical average for those dates) over the window; not shown below 10 days, or when the historical average already misses by less than 0.1" a day> 0

Notice, effective 15 November 2026: a resort's "last 24 hours" snow report is a window that ends when it is posted, so a report posted in the morning describes mostly the previous day. From 15 November 2026 we pair each snow report with the day it describes when we measure Range Containment, and the season-to-date count on this site restarts on that date. Until enough North America reports have been scored, the figures show "not enough days" rather than a number.

How we state range containment: 80% is what the range is built to achieve, not something we claim until it has been measured. Until a lead time's range has passed our pre-registered calibration check, we show only the measured share so far this season, with the number of snowy days behind it. Below 10 snowy days we show no percentage at all, because a handful of storms can't support one.
Why a probabilistic estimate matters: Single-point forecasts give a false sense of precision. Our high (80th pct.) estimate shows how much more snow the ensemble members allow for beyond the median. A high estimate close to the median means the members mostly agree; a much higher one means some of them bring a lot more snow. Members agreeing is not the same as the forecast being right: whether reported snowfall stays under the high estimate as often as it should is the measured share described above, and we don't grade a forecast's confidence any other way. More snow than the high estimate can fall.

Orographic Correction Details

Each resort has a unique set of wind-direction-dependent precipitation multipliers. These account for the fact that mountains preferentially enhance precipitation when moisture-laden air flows from certain directions. The multipliers were calibrated by comparing GEFS raw QPF against PRISM's 30-year, terrain-aware precipitation climatology at each resort's coordinates.

For example, a resort in a west-facing canyon will see greater orographic enhancement from westerly flow (multiplier > 1.0) and less from easterly flow. These corrections are critical for mountain forecasting accuracy and are one of PowCall's key differentiators.

Covered Resorts

PowCall currently covers 120 Epic and Ikon Pass ski resorts across 31 states and provinces:

RegionResorts
Utah Alta, Brighton, Deer Valley, Park City, Powder Mountain, Snowbasin, Snowbird, Solitude
Colorado Beaver Creek, Breckenridge, Crested Butte, Keystone, Loveland, Monarch, Vail, Arapahoe Basin, Aspen Snowmass, Copper Mountain, Eldora, Winter Park, Steamboat, Telluride
California Big Bear Mountain Resort, Heavenly, June Mountain, Kirkwood, Mammoth Mountain, Northstar, Palisades Tahoe, Sierra-at-Tahoe, Snow Valley, Squaw/Alpine (Base), Sugar Bowl
Pacific Northwest Crystal Mountain, Mt. Bachelor, Mt. Hood Meadows, Stevens Pass, The Summit at Snoqualmie
Wyoming Grand Targhee Resort, Jackson Hole
Montana Whitefish, Big Sky, Bridger Bowl
Idaho Schweitzer, Brundage Mountain, Sun Valley
New Mexico Taos Ski Valley
Alaska Alyeska Resort
British Columbia Revelstoke, Whistler Blackcomb, Cypress Mountain, Fernie Alpine Resort, Kicking Horse, Kimberley Alpine, Panorama Mountain Resort, RED Mountain Resort, SilverStar Mountain Resort, Sun Peaks Resort
Alberta Banff Sunshine Village, Lake Louise Ski Resort, Marmot Basin, Mt. Norquay, Nakiska
New England Attitash, Crotched Mountain, Jay Peak, Killington, Loon Mountain, Mount Snow, Mount Sunapee, Okemo, Stowe, Stratton, Sugarbush, Sugarloaf, Sunday River, Wildcat Mountain
New York & Mid-Atlantic Big Boulder, Blue Mountain Resort, Camelback Resort, Hidden Valley, Hunter Mountain, Jack Frost, Laurel Mountain, Liberty Mountain, Roundtop, Seven Springs, Snowshoe Mountain, Whiteface, Whitetail
Midwest Afton Alps, Hidden Valley, Snow Creek, Alpine Valley, Boston Mills, Boyne Highlands, Boyne Mountain, Brandywine, Granite Peak, Lutsen Mountains, Mad River Mountain, Mt. Brighton, Paoli Peaks, Snowriver Mountain Resort, Wilmot Mountain
Eastern Canada Blue Mountain, Bromont, Le Massif de Charlevoix, Mont Sainte Anne, Stoneham, Tremblant
Australia Perisher, Thredbo, Falls Creek, Hotham, Mt Buller
New Zealand Mt Hutt, Coronet Peak, The Remarkables
Chile Valle Nevado

Transparency Commitment

PowCall is built on a commitment to forecast transparency. We:

© 2026 PowCall. Forecast data from the GEFS ensemble, checked against SNOTEL and GHCN weather stations where a resort has one.