Skip to content

Forecasts

The daily cycle

Every day, each board runs this sequence:

  1. Analysis. The snowpack is carried forward from yesterday using the best estimate of past weather, ERA5 reanalysis (Hersbach et al., 2020). ERA5 is published with a delay of about five days. The most recent days are filled with short-range forecast data.
  2. Forecast. From the latest ECMWF IFS run, the downscaled weather drives the snow model forward about 10 days.
  3. Blending. Analysis and forecast forcing are cross-faded over 6 hours, so there is no jump in temperature or snow at the join.
  4. Products. Maps, anomalies, warnings and point panels are rebuilt and published.
gantt
    dateFormat  YYYY-MM-DD
    axisFormat  %d %b
    section Forcing
    ERA5 reanalysis         :a1, 2026-11-01, 4d
    Recent forecasts (gap)  :a2, after a1, 4d
    IFS forecast (today →)  :a3, after a2, 10d

Deterministic forecast

The single high-resolution IFS forecast. Depending on the region it is the 0.1° (≈ 9 km) product or the 0.25° open-data product (see Domains). It is downscaled exactly like the analysis.

Ensemble forecast

The ECMWF ensemble (ENS) has about 50 members, each a full forecast from slightly perturbed initial conditions and model physics. SnowMapper downscales the members. It then reduces them per terrain unit to:

  • percentiles P10, P50, P90 for each hour (snowfall, precipitation, temperature);
  • accumulations of new snow over 1, 3, 5 and 10 days, also as P10/P50/P90;
  • the probability of snow: the share of members with snowfall.

The percentiles are computed at each unit independently.

Why forecasts are uncertain in mountains

  • Where a storm hits is uncertain beyond a few days; the ensemble spread shows how much.
  • Global models smooth mountains, so orographic precipitation is placed too broadly. Downscaling corrects temperature and phase, but not the amount (see Known limitations).
  • Convective (showery) precipitation is poorly timed in all global models.

Data assimilation in the forecast

In regions with data assimilation (see Satellite and data assimilation), the forecast starts from the corrected snowpack, not the open-loop one.