The essentials in two minutes
A forecast is the output of a model: a program that divides the atmosphere into small boxes (their size is the grid spacing) and calculates how they change. The finer the grid spacing, the truer the terrain. AROME (the high-resolution model of Météo-France, the French national weather service: about 1.3 km) and ICON-D2 (the high-resolution model of the German weather service, DWD: about 2 km) see the major valleys; ECMWF (the model of the European Centre for Medium-Range Weather Forecasts: about 9 km) and GFS (the US global model: 13 to 25 km depending on the product) see planed-down Alps, with summits lowered by several hundred metres and valleys filled in. The direct consequence: the wind on the ridges, valley inversions, breezes and the exact time of thunderstorms are the weak points of any mountain forecast.
Reliability falls with lead time, and not at the same rate for everything: the general situation can be forecast five days ahead, the ceiling two days ahead, and a thunderstorm over a given massif at a given hour almost never with certainty. To measure this uncertainty, you look at several models side by side or, better, at an ensemble: thirty or fifty versions of the same forecast. If they are tightly grouped, you can rely on them; if they spread apart, you wait.
The method fits into a funnel: at D−3, “is there a window?”; the day before, “which site, what time?”; in the morning, “is the forecast holding?” — you compare what was forecast for 09:00 or 10:00 with what the wind stations are measuring; at launch or at the trailhead, you look at the sky. Finally, you calibrate yourself: by noting for each flight what was forecast and what was observed, you learn in one season by how much your favourite model is wrong on your sites. For hiking, the same logic applies: Météo-France’s mountain bulletin (a forecast written by a forecaster for each massif), the freezing level, the likely time of thunderstorms, and a turnaround time set before you leave.
Read on
The whole chapter is open to subscribers, with its animated figures: €3 a month or €25 a year. The book as an ePub, €6.99, comes with a month’s subscription. Contents:
- 10.1What a model is, in one image
- 10.2What models see poorly in the mountains
- 10.3Lead time, reliability and ensembles
- 10.4The toolbox
- 10.5Reading a windgram
- 10.6Reading a wind station
- 10.7Is the forecast holding? Comparing forecast and measured
- 10.8The funnel: from the day before to launch
- 10.9Calibrating yourself: the forecast-versus-observed log
- 10.10For hikers: bulletin, freezing level, timing
- 10.11Field checklists
Already a subscriber? Log in: the link brings you back here.
Chapter glossary#
- Grid spacing
- Side of the boxes into which a model divides the atmosphere. A model correctly describes only phenomena larger than five to eight grid spacings.
- Model run
- One complete calculation by a model, named after its start time in Universal Time (“the 06:00 UTC run”). Published several hours after that time.
- Lead time
- Time between the start of the calculation and the forecast time. Reliability falls as the lead time increases, more or less quickly depending on the parameter.
- Ensemble (ensemble forecast)
- Series of forecasts from the same model, starting from slightly different initial states. The spread between the members measures the uncertainty.
- Boundary layer (friction layer)
- Layer of air mixed by the ground (thermals, friction). On a windgram, its height gives the thermal ceiling.
- Windgram
- Time × altitude graph for a site: wind, ceiling, cloud at each level, rain. The central tool of the pilot the day before.
- Grid point
- The model box that contains the requested place, at the altitude of the smoothed terrain — not necessarily that of the launch or the summit.
- Plan B
- Fallback route, site or plan decided in advance, usable if the day turns bad.