The mountain, day by day
Pick a day. Every part of Åreskutan gets its own forecast, and the freezing level, snow line and cloud base are drawn where they actually sit.
Worth looking up
Northern lights by night, a sea of clouds by day — when the ingredients line up, which is not the same as a promise.
Every zone, every day
Top of the mountain at the top. Highest and lowest temperature, strongest wind and the day's precipitation — snow in white, rain in blue.
Seven days, five heights
Hour by hour through the week, one line per zone. Where the lines cross, the summit is warmer than the village — an inversion.
Where the zero line sits
The freezing level, the snow line and the cloud base through the week, against the parts of the mountain. A zone above the yellow line is below freezing; above the grey line it is in cloud.
Elevation × time
Every 100 m of the mountain, hour by hour. The white line is the rain/snow boundary.
Keyboard: ← → step the selected hour.
Vertical profile
Sounding-derived temperature, cloud base and wind through the mountain.
Mountain intel
Snapshot for the selected hour.
Which side of the mountain
Wind strips one side and loads the other.
Is this normal?
Today against thirty years of the same date.
Reading thirty years of archive…
Ground truth
Live readings from the nearest SMHI stations, and how the forecast compares against them right now.
Reading the stations…
Hourly at summit
Ensemble spread (p10–p90) around the machine-corrected forecast.
Model agreement
Weighted by measured 45-day skill over this exact point.
Learning log
What the bias model learned on its last training run, and how much it actually helped.
How this forecast is madeNo black boxes — the summary is here, every formula and licence is on the method page.
- Six numerical weather models are fetched per mountain from Open-Meteo: MET Norway MET Nordic (1 km, the best model for Scandinavian terrain), DMI Harmonie AROME (2 km), DWD ICON, ECMWF IFS, UKMO and NOAA GFS.
- A real sounding, not a lapse-rate guess. Temperature, humidity and wind on pressure levels from 1000 hPa to 500 hPa with their geopotential heights give the true vertical structure above the mountain — including inversions, where the summit is warmer than the village.
- Elevation downscaling. The sounding is anchored to the surface ensemble at model height, then interpolated to every 100 m band. Wind is blended from surface to free atmosphere and multiplied by a terrain exposure factor. Precipitation gets an orographic enhancement with height.
- Phase from wet-bulb temperature, not air temperature — the physically correct way to place a snow line, then a temperature-dependent snow ratio converts millimetres to centimetres.
- Machine learning, retrained every day. The pipeline keeps every archived forecast from every model since January 2023, scored against ERA5-Land reanalysis at this exact point, and the record grows by a day each day. A softmax over measured error assigns each model its weight and a logistic model calibrates precipitation probability. For temperature and wind, three candidates compete — ridge regression, gradient-boosted trees and a neural network — on a block of hours none of them trained on, and the simplest one that clearly wins is used. The learning log shows the scores, including when nothing helps.
- Ensemble probabilities come from a 30+ member ICON-EU / GFS ensemble, giving the p10–p90 bands and the chance of precipitation.
This is not an avalanche forecast. Wind-loading and new-snow numbers are meteorological indices only. For avalanche danger in the Swedish fjäll always use lavinprognoser.se, and for official warnings SMHI. Mountain weather kills people who trust a screen over their eyes.