The snowfall risk assesses how often the site gets, in one day, a snowfall able to overload roofs and block access and operations: at least 30 cm in the lowlands or 60 cm in the mountains, the MeteoSwiss danger level 4 threshold. It is estimated from the heaviest snowfall of every winter since 1940, measured by the nearby station or by the ERA5 reanalysis: the more often such a snowfall comes back, the higher the risk.
Indicator. Return period, in years, of a one-day snowfall of MeteoSwiss danger level 4 ('high danger', phenomena of unusual intensity): 30 cm in 24 hours in the lowlands, 60 cm in the mountains above 800 m (above 1,600 m south of the Alps). The threshold used, the 50-year snowfall, the highest observed daily snowfall and the data source are also shown.
Data. Daily fresh snow of the nearest MeteoSwiss station, when it is within 25 km (the ERA5 grid spacing) and 200 m of height and has at least 30 complete winters, the length of a WMO climate normal (MeteoSwiss open data, CC BY 4.0, Source: MeteoSwiss). Otherwise daily snowfall and mean temperature of the ERA5 reanalysis (Hersbach et al. 2020) from the Open-Meteo archive (ERA5 only): the water equivalent is turned into centimetres of fresh snow with the density of Hedstrom and Pomeroy 1998, computed from the temperature of the day and capped at 150 kg/m³. The height of the site comes from the Open-Meteo 90 m digital elevation model. Period: winters from 1940/41 to 2024/25 (August to July; calendar year in the southern hemisphere). Scenarios: SSP1-2.6 / SSP2-4.5 / SSP3-7.0 / SSP5-8.5, with the same level.
Calculation.
(1) For every winter the highest daily snowfall is taken.
(2) A GEV (Generalized Extreme Value) distribution is fitted on the winters with snow (at least 1 cm in a day, the resolution of the MeteoSwiss measurements) and multiplied by the share of winters with snow (Evin et al. 2025); with fewer than 30 snowy winters, the winters above the threshold are counted directly.
(3) The return period is the inverse of the yearly probability of exceeding the threshold; the 50-year snowfall is the one exceeded with a 2% yearly probability, the return period of the snow loads of Eurocode EN 1991-1-3.
(4) For now the level is the same in every scenario and year: with warming, extreme snowfall decreases much less than mean snowfall (O'Gorman 2014) and at 900 m in the French Alps the 100-year snowfall falls by about 15% between +1 and +4 °C of global warming (Le Roux et al. 2023), so today's level is slightly conservative. Projections from climate models are the next step.
Note: ERA5 cells are about 25 km and underestimate local peaks in complex terrain (in southern Ticino a one-day record of 27 cm against 55 cm at the Lugano station), which is why a nearby station is used where there is one.
Thresholds → level A–F.
• level-4 snowfall at least once every 3.2 years → F
• every 3.2–10 years → E
• every 10–22.4 years → D
• every 22.4–50 years → C
• every 50–1,000 years → B
• less than once every 1,000 years → A.
ThinkHazard! bands (GFDRR): high hazard within 10 years, medium within 50, low within 1,000, translated as for landslides (high → E/F, medium → C/D, low → B, beyond → A); within the high and medium bands the more frequent half (split at the geometric middle: 3.2 and 22.4 years) takes the higher class, a xNatura choice.
The MeteoSwiss thresholds (30 cm in the lowlands, 60 cm in the mountains) are defined for Switzerland and are applied worldwide: this is a xNatura choice. A return period beyond 10,000 years is not shown as a number.
From physical to impact. The physical level (A–F) is combined with the production process's sensitivity through the risk matrix (6 physical levels by 5 sensitivity bands) → impact level. The euro amount of this risk is the one described in the "Economic damage" paragraph (Version 2026): the damage factor per class of Version 2025 is no longer used for this risk.
Economic damage. Besides the class, the loss in euro from roof failure under the year's maximum one-day snowfall is estimated as a share of the asset value. Basis: "xNatura estimate".
• Load: the snowfall comes from the same GEV distribution used for the class; ground load = 1.0 kN/m³ × snow depth (fresh snow, EN 1991-1-3, Annex E: the upper end of the 0.7-1.0 kN/m³ measured by Judson and Doesken 2000); roof load = 0.8 × ground load, 0.4 for roofs pitched over 30 degrees (EN 1991-1-3).
• Roof resistance (xNatura choice): the resistance of a designed roof is the one that makes it fail once in 10,000 years under the site's snow, the middle of the range from 1 in 100,000 to 1 in 1,000 per year of reliability studies of real roofs (Takahashi and Ellingwood 2005); never less than twice the 0.4 kN/m² imposed load for roofs (EN 1991-1-1), the resistance of roofs where snow is rare. Light (sheet metal, panels) and timber roofs have 0.8 times that resistance (light roofs give way first). The resistance is calibrated on the roof's own load, so the pitch changes the loss only where that minimum applies.
• Roof type: the one entered in the site data ("Roof type"), otherwise from the structure type (light steel and timber as a light roof), otherwise the mean of a light and a heavy roof.
• Damage: when the roof fails, 50% of the value is lost (xNatura choice: there is no published loss ratio for industrial roofs).
• Expected annual loss (mean of the loss over the distribution of the yearly maximum snowfall, over the whole tail, as for the resistance calibration) and loss at the snowfall that comes back once in 100 years. Same values in every scenario, as the class.
Simplifications: only one-day fresh snow, not the accumulated or wet snow that causes most roof failures; no wind drifts, no rain on snow.
Sources.
• [MeteoSwiss, Beschreibung der Gefahrenstufen (May 2021)](MeteoSwiss, Beschreibung der Gefahrenstufen (May 2021))
• [MeteoSwiss Open Data](MeteoSwiss Open Data)
• [Hersbach et al. 2020, The ERA5 global reanalysis](Hersbach et al. 2020, The ERA5 global reanalysis)
• [Hedstrom and Pomeroy 1998, Measurements and modelling of snow interception in the boreal forest](Hedstrom and Pomeroy 1998, Measurements and modelling of snow interception in the boreal forest)
• [WMO Guidelines on the Calculation of Climate Normals (WMO-No. 1203)](WMO Guidelines on the Calculation of Climate Normals (WMO-No. 1203))
• [Evin et al. 2025, extremes of snow water equivalent and global warming levels](Evin et al. 2025, extremes of snow water equivalent and global warming levels)
• [Le Roux et al. 2023, Projection of snowfall extremes in the French Alps](Le Roux et al. 2023, Projection of snowfall extremes in the French Alps)
• [Contrasting responses of mean and extreme snowfall to climate change](Contrasting responses of mean and extreme snowfall to climate change)
• [Eurocode EN 1991-1-3, snow loads](Eurocode EN 1991-1-3, snow loads)
• [Eurocode EN 1991-1-1, densities, self-weight and imposed loads for buildings](Eurocode EN 1991-1-1, densities, self-weight and imposed loads for buildings)
• [Judson and Doesken 2000, Density of freshly fallen snow in the Central Rocky Mountains](Judson and Doesken 2000, Density of freshly fallen snow in the Central Rocky Mountains)
• [Takahashi and Ellingwood 2005, Reliability-based assessment of roofs in Japan subjected to extreme snows](Takahashi and Ellingwood 2005, Reliability-based assessment of roofs in Japan subjected to extreme snows).
Indicator. Return period, in years, of a one-day snowfall of MeteoSwiss danger level 4 ('high danger', phenomena of unusual intensity): 30 cm in 24 hours in the lowlands, 60 cm in the mountains above 800 m (above 1,600 m south of the Alps). The threshold used, the 50-year snowfall, the highest observed daily snowfall and the data source are also shown.
Data. Daily fresh snow of the nearest MeteoSwiss station, when it is within 25 km (the ERA5 grid spacing) and 200 m of height and has at least 30 complete winters, the length of a WMO climate normal (MeteoSwiss open data, CC BY 4.0, Source: MeteoSwiss). Otherwise daily snowfall and mean temperature of the ERA5 reanalysis (Hersbach et al. 2020) from the Open-Meteo archive (ERA5 only): the water equivalent is turned into centimetres of fresh snow with the density of Hedstrom and Pomeroy 1998, computed from the temperature of the day and capped at 150 kg/m³. The height of the site comes from the Open-Meteo 90 m digital elevation model. Period: winters from 1940/41 to 2024/25 (August to July; calendar year in the southern hemisphere). Scenarios: SSP1-2.6 / SSP2-4.5 / SSP3-7.0 / SSP5-8.5, with the same level.
Calculation.
(1) For every winter the highest daily snowfall is taken.
(2) A GEV (Generalized Extreme Value) distribution is fitted on the winters with snow (at least 1 cm in a day, the resolution of the MeteoSwiss measurements) and multiplied by the share of winters with snow (Evin et al. 2025); with fewer than 30 snowy winters, the winters above the threshold are counted directly.
(3) The return period is the inverse of the yearly probability of exceeding the threshold; the 50-year snowfall is the one exceeded with a 2% yearly probability, the return period of the snow loads of Eurocode EN 1991-1-3.
(4) For now the level is the same in every scenario and year: with warming, extreme snowfall decreases much less than mean snowfall (O'Gorman 2014) and at 900 m in the French Alps the 100-year snowfall falls by about 15% between +1 and +4 °C of global warming (Le Roux et al. 2023), so today's level is slightly conservative. Projections from climate models are the next step.
Note: ERA5 cells are about 25 km and underestimate local peaks in complex terrain (in southern Ticino a one-day record of 27 cm against 55 cm at the Lugano station), which is why a nearby station is used where there is one.
Thresholds → level A–F.
• level-4 snowfall at least once every 3.2 years → F
• every 3.2–10 years → E
• every 10–22.4 years → D
• every 22.4–50 years → C
• every 50–1,000 years → B
• less than once every 1,000 years → A.
ThinkHazard! bands (GFDRR): high hazard within 10 years, medium within 50, low within 1,000, translated as for landslides (high → E/F, medium → C/D, low → B, beyond → A); within the high and medium bands the more frequent half (split at the geometric middle: 3.2 and 22.4 years) takes the higher class, a xNatura choice.
The MeteoSwiss thresholds (30 cm in the lowlands, 60 cm in the mountains) are defined for Switzerland and are applied worldwide: this is a xNatura choice. A return period beyond 10,000 years is not shown as a number.
From physical to impact. The physical level (A–F) is combined with the production process's sensitivity through the risk matrix (6 physical levels by 5 sensitivity bands) → impact level. The euro amount of this risk is the one described in the "Economic damage" paragraph (Version 2026): the damage factor per class of Version 2025 is no longer used for this risk.
Economic damage. Besides the class, the loss in euro from roof failure under the year's maximum one-day snowfall is estimated as a share of the asset value. Basis: "xNatura estimate".
• Load: the snowfall comes from the same GEV distribution used for the class; ground load = 1.0 kN/m³ × snow depth (fresh snow, EN 1991-1-3, Annex E: the upper end of the 0.7-1.0 kN/m³ measured by Judson and Doesken 2000); roof load = 0.8 × ground load, 0.4 for roofs pitched over 30 degrees (EN 1991-1-3).
• Roof resistance (xNatura choice): the resistance of a designed roof is the one that makes it fail once in 10,000 years under the site's snow, the middle of the range from 1 in 100,000 to 1 in 1,000 per year of reliability studies of real roofs (Takahashi and Ellingwood 2005); never less than twice the 0.4 kN/m² imposed load for roofs (EN 1991-1-1), the resistance of roofs where snow is rare. Light (sheet metal, panels) and timber roofs have 0.8 times that resistance (light roofs give way first). The resistance is calibrated on the roof's own load, so the pitch changes the loss only where that minimum applies.
• Roof type: the one entered in the site data ("Roof type"), otherwise from the structure type (light steel and timber as a light roof), otherwise the mean of a light and a heavy roof.
• Damage: when the roof fails, 50% of the value is lost (xNatura choice: there is no published loss ratio for industrial roofs).
• Expected annual loss (mean of the loss over the distribution of the yearly maximum snowfall, over the whole tail, as for the resistance calibration) and loss at the snowfall that comes back once in 100 years. Same values in every scenario, as the class.
Simplifications: only one-day fresh snow, not the accumulated or wet snow that causes most roof failures; no wind drifts, no rain on snow.
Sources.
• [MeteoSwiss, Beschreibung der Gefahrenstufen (May 2021)](MeteoSwiss, Beschreibung der Gefahrenstufen (May 2021))
• [MeteoSwiss Open Data](MeteoSwiss Open Data)
• [Hersbach et al. 2020, The ERA5 global reanalysis](Hersbach et al. 2020, The ERA5 global reanalysis)
• [Hedstrom and Pomeroy 1998, Measurements and modelling of snow interception in the boreal forest](Hedstrom and Pomeroy 1998, Measurements and modelling of snow interception in the boreal forest)
• [WMO Guidelines on the Calculation of Climate Normals (WMO-No. 1203)](WMO Guidelines on the Calculation of Climate Normals (WMO-No. 1203))
• [Evin et al. 2025, extremes of snow water equivalent and global warming levels](Evin et al. 2025, extremes of snow water equivalent and global warming levels)
• [Le Roux et al. 2023, Projection of snowfall extremes in the French Alps](Le Roux et al. 2023, Projection of snowfall extremes in the French Alps)
• [Contrasting responses of mean and extreme snowfall to climate change](Contrasting responses of mean and extreme snowfall to climate change)
• [Eurocode EN 1991-1-3, snow loads](Eurocode EN 1991-1-3, snow loads)
• [Eurocode EN 1991-1-1, densities, self-weight and imposed loads for buildings](Eurocode EN 1991-1-1, densities, self-weight and imposed loads for buildings)
• [Judson and Doesken 2000, Density of freshly fallen snow in the Central Rocky Mountains](Judson and Doesken 2000, Density of freshly fallen snow in the Central Rocky Mountains)
• [Takahashi and Ellingwood 2005, Reliability-based assessment of roofs in Japan subjected to extreme snows](Takahashi and Ellingwood 2005, Reliability-based assessment of roofs in Japan subjected to extreme snows).