The European avalanche danger scale has five levels, and almost everybody reads it wrong. Not because the levels are complicated, but because a run of numbers suggests an evenness that is not there: level 3 looks like the midpoint between 1 and 5. In the accident statistics it is the opposite, namely the most dangerous band of all.

The scale

Forecast daysFatalities
  1. 1 low ~20 % ~5 %
  2. 2 moderate ~50 % ~30 %
  3. 3 considerable ~30 % ~50 %
  4. 4 high 2.1 % ~10 %
  5. 5 very high very rare ~1 %
On the left, how often the level is forecast in the Alps between 1 December and 30 April. On the right, the share of avalanche fatalities it accounts for. The percentages are rounded and therefore do not add up to exactly one hundred. SLF, explanatory notes on the danger levels in the avalanche bulletin.

Two sets of figures that do not match. Level 2 applies on half of all winter days and accounts for less than a third of the victims. Level 4 applies on barely two days in a hundred and accounts for one death in ten.

The number that makes it visible

Divide the share of fatalities by the share of forecast days and you get a measure of how lethal a single day at that level is. A value of 1.0 would mean: exactly as dangerous as its frequency would lead you to expect.

Level 1 low 5 % of deaths on 20 % of days
0.25
Level 2 moderate 30 % of deaths on 50 % of days
0.6
Level 3 considerable 50 % of deaths on 30 % of days
1.7
Level 4 high 10 % of deaths on 2.1 % of days
4.8
Share of fatalities divided by share of forecast days. Above 1.0 means: a day at this level carries more deaths than its frequency would imply. Level 5 is left out because it is too rare for the division to mean anything. Our own calculation from the SLF figures for frequency and share of victims by danger level.

A single day at level 4 is about nineteen times as lethal as a day at level 1, and getting on for three times as lethal as a day at level 3.

And that figure is flattering. It counts calendar days, not people in the terrain. Far fewer people go out on a level 4 day than on a level 2 day. Per person actually out there, the gap opens wider still.

So why does level 3 kill the most?

This is not a contradiction but the difference between probability and frequency. At level 4 most people stay home; at level 5 essentially everyone does. At level 3 they do not: it stands in the bulletin for a third of the winter, and anyone who only set out at levels 1 and 2 would hardly ever go at all.

Then there is the span. The SLF itself gives as its reason for introducing sub-levels that level 3 covers “a very broad spectrum”. Between a barely considerable day and a nearly high one lies what amounts to a full level of difference, and both appear as the same 3 in the bulletin.

The Swiss sub-levels

Which is why Switzerland has, since the winter of 2022/23, been doing something nobody else does.

  1. 1 low
  2. 2 moderate subdivided in Switzerland: 2−, 2=, 2+
  3. 3 considerable subdivided in Switzerland: 3−, 3=, 3+
  4. 4 high subdivided in Switzerland: 4−, 4=, 4+
  5. 5 very high
From level 2 upwards, Switzerland additionally states where within the level the danger sits: minus at the lower end, equals in the middle, plus at the upper end. No subdivision is given for wet-snow avalanches, where in the SLF's assessment it does not work. SLF, subdivision of danger levels in the avalanche bulletin, introduced for the winter of 2022/23.

“Level 3” thus becomes 3−, 3= or 3+. The five base levels are unchanged; all that is added is a statement of where inside the level you stand.

The reasoning is well evidenced. Assessed over six winters, the sub-levels reduced the discrepancy between forecast and field observation, and they correlated with objectively measurable quantities such as the frequency of whumpfing and the results of snowpack stability tests.

Important for anyone crossing a border: the sub-levels exist only in Switzerland. France, Austria, Italy and the other members of the European avalanche warning services still work with the plain five-point scale. They are watching the Swiss experience, but none has adopted it. Read a 3− in Davos on Monday and a 3 in Tignes on Wednesday, and you do not have the same information in front of you.

What the scale is not

Three misconceptions that persist.

It is not a percentage. Level 4 does not mean “80 per cent avalanche danger”. The levels describe a combination of how widespread the danger spots are, how likely a release is and how large the avalanches are likely to be. No probability figure appears anywhere in it.

It is not linear. The step from 3 to 4 is larger than the step from 1 to 2, and considerably so. That is not sloppiness but design: the scale reflects how avalanche danger actually increases, and that is not a straight line.

It applies to a region, not to the slope in front of you. The bulletin describes an area the size of a region, not an individual slope. It says what to expect, not where. Mapping it onto the ground in front of you is precisely the part nobody can read off a colour, and the part that requires training.

What it means in practice

The usable conclusion from these figures is uncomfortable, because it runs against intuition.

Most people treat the scale like a traffic light: 1 and 2 green, 3 amber, 4 and 5 red. The statistics suggest a different split. The entire serious range lies between 2 and 4, and the level at which most people die is the level at which most people set out.

For piste skiers none of this changes anything: open, marked pistes are secured and are, as a rule, safe even at level 4. The whole statistic above concerns the backcountry. In many places the transition is a single step off the edge of the piste, and that one step changes the risk class.

Sources and caveats

  • SLF, explanatory notes on the danger levels: frequency in the Alps from 1 December to 30 April, level 1 around 20 per cent, level 2 around 50, level 3 around 30, level 4 2.1 per cent, level 5 very rare. Share of avalanche fatalities: level 1 around 5 per cent, level 2 around 30, level 3 around 50, level 4 around 10, level 5 around 1.
  • SLF, subdivision of danger levels in the avalanche bulletin, introduced for the winter of 2022/23, for dry avalanches only, from level 2 upwards.
  • EAWS (European Avalanche Warning Services), the common European danger scale. The sub-levels remain a Swiss arrangement; the other members are following the evaluation but have not introduced them.
  • The percentages quoted are rounded and do not add up to exactly one hundred. The quotient of victim share and day share is our own calculation, not an SLF metric. It assumes the same number of people are in the terrain on every day, which is untrue and which makes the figure for the high levels an underestimate rather than an overestimate.

This piece explains how the scale is built and does not replace an avalanche course. Training comes from the SLF and the SAC.