This September was warm, and unusually so. Averaged over ten Swiss ski areas and calculated at 2,000 metres it comes to 9.89 degrees. That is the highest figure since 1991, just ahead of 2023 on 9.80 and 1991 itself on 9.77.
The question hanging on that, and a lot of people are asking it: does it mean anything for the winter?
We ran the numbers. 35 winters, ten areas, the same altitude. For each year, the mean September temperature and the total snowfall from 1 December to 31 March that followed.
The answer in three bars
That is the answer, and it is uncomfortable for both camps. More snow fell after the ten warmest Septembers than after the ten coldest, 316 against 267 centimetres. Anyone inferring a poor snow year from a warm autumn is not merely wrong by these figures but wrong in the other direction.
The opposite conclusion should not be drawn either. The relationship is weak: the correlation is 0.20, the coefficient of determination 0.04. Four per cent of the differences between winters can be explained by the September temperature; ninety-six per cent cannot. With 35 winters, a value of 0.20 is also not cleanly distinguishable from chance.
What the winters actually look like
Show the figures as a table
| Entry | Value |
|---|---|
| 91 | 345 cm |
| 92 | 194 cm |
| 93 | 358 cm |
| 94 | 436 cm |
| 95 | 213 cm |
| 96 | 216 cm |
| 97 | 259 cm |
| 98 | 420 cm |
| 99 | 361 cm |
| 00 | 358 cm |
| 01 | 239 cm |
| 02 | 218 cm |
| 03 | 318 cm |
| 04 | 226 cm |
| 05 | 292 cm |
| 06 | 306 cm |
| 07 | 300 cm |
| 08 | 362 cm |
| 09 | 249 cm |
| 10 | 190 cm |
| 11 | 383 cm |
| 12 | 325 cm |
| 13 | 330 cm |
| 14 | 302 cm |
| 15 | 300 cm |
| 16 | 232 cm |
| 17 | 472 cm |
| 18 | 346 cm |
| 19 | 302 cm |
| 20 | 351 cm |
| 21 | 252 cm |
| 22 | 229 cm |
| 23 | 423 cm |
| 24 | 234 cm |
| 25 | 246 cm |
Two winters in that series deserve a second look, because together they finish off the assumption.
The snowiest winter in the series is 2017/18 with 472 centimetres. The September before it, at 4.24 degrees, was one of the coldest of all. Anyone who in autumn 2017 had inferred heavy snow from the cold would have been right.
The least snowy is 2010/11 with 190 centimetres. The September before it, at 5.73 degrees, was also cool, well below the mean. The same starting point, the opposite outcome.
And the two Septembers closest to this one? 2023, at 9.80 degrees, was followed by 423 centimetres, the second-snowiest winter of the series. 1991, at 9.77 degrees, by 345 centimetres, also above average.
What a warm autumn does do
It pushes back the start. That is a different question from how much snow falls, and there the link is direct: as long as the wet-bulb temperature does not drop below minus two degrees, no snow gun runs, and without snowmaking hardly any area opens. That is exactly what is happening now. Over the sixteen days from today, ten Swiss areas have 95 of 3,840 hours cold enough, which is 2.5 per cent. The figures are in our snowmaking analysis.
A late start and a poor snow year are two different things. The winter of 2023/24 showed both: one of the warmest Septembers in the series, and then 423 centimetres of snow.
What this calculation cannot do
It is not a forecast. It says how autumn and winter related to one another over 35 past years, and the answer is: barely. Nothing follows from that for this winter except an invitation not to pretend to know. What can responsibly be said about the coming winter is collected in our winter forecast.
It uses model data, not measurements. ERA5 is a reanalysis, a model fitted to observations. For comparisons across decades that is the clean basis, because the same method applies across the whole series. A measuring station on the slope would give more accurate values for a single year.
Total snowfall is not snow depth. How much of that snow stays depends on temperature, rain and wind. A winter with 300 centimetres of snowfall and four warm spells skis worse than one with 250 in a single run.
35 winters are few. For a weak relationship of the kind found here, that is too short a series either to confirm it or to rule it out. That is why the headline says “almost nothing” and not “nothing”.
Read on: What will winter 2026/27 be like? · Too warm to make snow · Will the Sölden World Cup go ahead? · Weather or climate?
Sources and method
- Data: Open-Meteo, archive based on the ERA5 reanalysis, retrieved 27 September 2026. Ten areas: Zermatt, Verbier, Andermatt, Davos, Arosa Lenzerheide, St. Moritz, Grindelwald, Adelboden, Engelberg, Airolo. The altitude is fixed at 2,000 metres for all of them so the series stays comparable.
- September figure: for the ranking, which includes 2026, the mean is taken over 1 to 26 September, because the month is not over. For the correlation with the following winter the full month is used, for the years 1991 to 2025.
- Winter figure: total snowfall from 1 December to 31 March, averaged over the ten areas. A winter is assigned to the September that preceded it.
- Statistics: Pearson correlation at the yearly level over 35 pairs, r = 0.201, coefficient of determination r² = 0.040. At the level of individual areas the values range from minus 0.10 in Zermatt to plus 0.36 in Arosa Lenzerheide, so weak throughout and inconsistent in sign.
- Checking the figures: the query script is in the repository at
scripts/herbst-winter-vergleich.mjs.
