Forecasting in practice

The essentials in two minutes

1. A model computes boxes; the finer the grid spacing, the more faithful the terrain: AROME (1.3 km) and ICON-D2 (2 km) the day before, ECMWF (9 km) and the ensembles for the trend.
2. A model only describes correctly what measures at least 5 to 8 times its grid spacing: thermals, slope breezes and rotors escape them all; the wind on the ridges is underestimated, valley inversions are misplaced.
3. Each parameter has its range: the general situation at five days, the upper wind at two or three, the ceiling at one or two, thunderstorms on the day itself, as a probability.
4. Ensembles (30 to 50 members) and model comparison measure the uncertainty: disagreement is information, and you keep the least favourable scenario for your margins.
5. The funnel: D−3 the window, D−1 the site and the time, in the morning the forecast-versus-measured check, at launch the sky; five questions before launching, and a single “yes” is enough to call it off.
6. Reading a wind station: the direction says who is in charge, the gap between the mean and the gusts gives the nature of the air (10 to 15 km/h or more: turbulent), the trend matters more than the value, and the location is everything.
7. If the wind station departs from the forecast at 10:00 (by 10 km/h or 45° or more), the afternoon forecast is suspect.
8. Hiking: the mountain bulletin, the freezing level (3,500 to 4,500 m in summer, 2,000 to 2,500 behind a cold front), the time of the thunderstorm, then a turnaround time and a plan B set before you leave; and a forecast-versus-observed log to calibrate yourself.