Faceshots

Computer-written from model data, updated twice a day.

How to use this site

Raw computer-model snowfall for Colorado ski resorts, plus an elevation-adjusted estimate and a tuned estimate for each resort's base, mid and summit. This is model output, not a human forecast.

Quick start

  1. Start with the Briefing. A short write-up of the next storm: when, where and how much, and how sure the models are.
  2. Read the map. It opens on the newest run's storm total. The colors are snow the model piles up from the start of the run to the hour on the slider (legend top left, in inches).
  3. Scrub or play. Drag the slider or press ▶ to watch the storm build step by step.
  4. Select a resort. Its panel shows raw, adjusted or tuned snow at base, mid and summit, a chart over time, and whether recent runs have trended up or down.
  5. Compare models. Switch between HRRR, GFS and ECMWF. Agreement between them, and runs trending the same way, is worth more than any single big number.

The map

Resort pins
Raw summit snow at the slider hour. ▲/▼ shows how much the adjusted summit total (Kuchera, over the same stretch of time) changed from the previous run to the newest one. It appears when the change is 0.5″ or more, and doesn't follow the slider, the ratio toggle or the run picker.
Dark bubbles
Several resorts close together: how many, and the deepest raw summit total among them. Select one to zoom in and split it into pins.
Gray, dashed area
Outside the area this site maps. It means “no data,” not “no snow.”

The controls

HRRR · GFS · ECMWF
HRRR is NOAA's high-resolution model and goes out 48 hours: best for the next day or two. GFS (NOAA) and ECMWF (European) are global models that see further out at coarser detail.
Consensus
An equal mix of the models, for comparing them. The briefing leads with the best-verified model instead.
Kuchera · 10:1
How melted water that falls as snow becomes snow depth. 10:1 turns every 1″ of it into 10″ of snow. Kuchera uses the warmest temperature between the ground and about 18,000 ft, so cold storms make fluffier, deeper snow and warm ones less. Hidden under Tuned, which uses its own fitted ratio.
Run picker
When a model run started, in UTC (“18Z” is noon MDT, 11 AM MST). The last four runs are kept; pick an older one to see what it said.
Slider and ▶
The forecast time, in Denver time. HRRR steps every hour, GFS and ECMWF every 3 hours. Totals add up from the start of the run.

Raw, adjusted and tuned

Models smooth the mountains, so their terrain sits well below real summits and above some valleys. Raw is the model's snow on its own smoothed terrain. Adjusted re-estimates snow at the real elevation using the model's air temperatures. It never changes how much water the model drops, only how much of it falls as snow and how fluffy that snow is. The panel notes the model's terrain height so you can see the gap. Tuned is Adjusted with settings fitted to six past winters of SNOTEL (mainly a fitted snow-to-water ratio; for HRRR also a water correction). It calls 6″+ days less often, to cut busted big-storm calls, at the cost of missing a few real ones. The Scorecard shows which has been closest.

The briefing

Computer-written
Computer-written from model data, updated twice a day. Every number and place in it comes from the model runs on this site; nobody edits it by hand.
Lead model
For each storm the briefing leans on the model that has verified best this season at that lead time (from the Scorecard), and its numbers are the summit totals in the resort list.
Models agree · Models split
Agree: the models are within a few inches of each other, so the range is solid. Split: they disagree, so the range leans low on purpose and the badge names the model with the bigger number. Hover over or tap a badge to see it.
Consensus
An equal mix of the models, for comparing them. The briefing leads with the best-verified model instead.
The map image
The lead model's storm total. Tap it to open that model on the Map at the end of the storm.

The tabs

Briefing
Opens first: the storm write-up, the resorts to watch and a "Share with the crew" button.
All resorts
Every resort in one table: raw / adjusted total through the end of each model's newest run (HRRR 2 days, GFS and ECMWF 5 days, so the columns cover different stretches), deepest first by raw summit for the model selected on the map.
Scorecard
How each model has actually done this season (or a past season, from the season picker; past seasons use tuned settings fitted without that season), checked against SNOTEL snow stations near the resorts (resort snow reports as a cross-check). Pick a lead time (how far ahead the forecast was made). Bias above zero means the model called for too much; typical miss is the usual size of the error. Select a storm to see every station. Water numbers are the most reliable, since measured snow depth settles and blows around (depth jumps above 40:1 snow to water are sensor errors and are left out).
Big picture
Upper-air (500 mb) maps are coming later.

Reading a storm 2–3 days out

  • Look at the trend over several runs, not one run's total.
  • Check the Scorecard at that lead time (HRRR only reaches 2 days): a model that has been running high this season probably still is.
  • Big totals that only one model shows deserve a wait-and-see.

Data: NOAA (HRRR, GFS), ECMWF open data (CC-BY-4.0), USDA NRCS SNOTEL, resort snow reports. Map: Esri. Reopen this guide any time with the ? button.