LATERNSER M., SCHNEEBELI M. Long-term snow climate trends of the Swiss Alps (1931-99). International Journal of Climatology, 2003, Vol. 23 , p. 733-750.

Climate change, snow cover, spatial variability, trend analysis, Switzerland.

WSL Swiss Federal Institute for Snow and Avalanche Research SLF, Flüelastrasse 11, CH-7260 Davos Dorf, Switzerland. schneebeli@slf.ch

Trends of snowfall days and heavy snowfall events
As for most other snow parameters, from 1960 to 1980 the days with snowfall were mainly above average and then a decreasing trend began. This trend to less snowfall days becomes stronger with decreasing altitude. In all regions, the snowfall days in the 1980s are significantly higher than in the 1990s; in all except one region (R7) the 1960s are also higher, and in some regions (R1-R3, R6) even the 1970s are significantly higher than the 1990s.

Regarding the number of days with HN > 10, 20 and 50 cm (dhn10-dhn50), low-lying stations do not always, rarely or never reach daily new snowfalls greater than 10, 20 or 50 cm every winter. Long-term trends are similar to dhn, but they become weaker with higher thresholds. Whereas for dhn10 only the 1980s differ significantly from the 1990s, for dhn20 no decade differs significantly from any other; and, because of the very few events, for dhn50 no long-term changes can be determined at all.

The annual maximum of HN3 (HN3max) is taken as an indicator for heavy snowfall events. Compared with the HS mean, the annual HN3 maxima averaged over a decade show much more variability on a smaller scale and the range of possible extremes is considerably larger. It is hard to find decades with particularly high or low HN3 maxima; but, as for the HS mean, the 1980s were largely above average and the 1990s rather below average. Looking at individual winters confirms the pronounced small-scale variability. Hardly any changes can be seen in the long-term development, neither for the whole Swiss Alps nor for single regions. However, looking at different altitude zones reveals that stations above 1300 m a.s.l. show a very weak rising trend, whereas stations below this tend to fall slightly. Only the very low foreland stations (< 650 m a.s.l.) show a marked drop since the early 1980s.

Trends of the average snow depth
Looking at the relative deviation of the 10-year mean of HS compared with the long-term mean, it can be assumed that the 1930s had generally below-average snow (except in the south) and the 1940s had above-average snow at lower altitudes, whereas the higher altitudes show slightly negative deviations from the long-term mean.

After the 1950s, with mostly below-average snow, three decades with plenty of snow follow, and the last decade, the 1990s, is again, throughout, clearly below average. During the 1990s, when stations at low levels had clearly below-average snow and the higher the stations were, the more they approached average conditions. Rather the opposite is the case during the snowy 1960s, 1970s and 1980s, with low areas showing slight tendencies to higher relative values than high-altitude stations.

Looking at two-monthly part seasons, a much more diversified picture is attained. Snowy conditions prevailed mainly during the 1960s (mid winter), 1970s (late winter) and 1980s (mid and late winter), whereas little snow occurred during the 1940s (early and late winter), the 1950s (early and late winter again) and during the 1990s (particularly in mid and late winter).

During the 1960s, especially 1963, 1968 (and 1970) the winters were well above average, whereas 1964 was one of the poorest winters on record. 1980, 1981, 1982, 1986 and 1987 mainly contributed to the snowiness of the 1980s and the period with successively poor winters continuing through most of the 1990s had already started in 1988. It is striking that five winters of the last decade (1990-99) are within the 11 least snowy winters of the past 69 years, and 1989 also belongs to this group.

Looking at the annual relative deviation from the long-term mean, averaged over all stations, very typical are the large year-to-year variations. In more than 60% of all cases the direction towards increasing or decreasing snow changes from year to year. Clusters of successive years with above-average snow can be found in the late 1960s and with short interruptions from the late 1970s till the mid 1980s. Accumulations of years with below-average snow are recorded in the late 1940s, late 1950s till early 1960s, early 1970s and from the late 1980s through most of the 1990s.

Whereas the entire North Slope (R1-R3) and parts of Grisons (R5) behave very similarly, the interior areas of the Alps (R4 and R7) have some distinctive deviations, and the South Slope (R6) has its own line completely. This clearly demonstrates the climatological diversity of the Swiss Alps. In addition to regional variations, trends are not uniformly distributed with altitude. Since the 1930s, for the entire Swiss Alps and the whole winter period, high elevations (> 1600 m a.s.l.) show a very slight increase culminating in the early 1960s, followed by a 20-year period remaining on this high level and then a gradual decrease towards the end of the century ending somewhat below the initial start-up level. On the other hand, low elevations (< 1000 m a.s.l.) show much more variability: a very snowy period during the 1940s is followed by the lean 1950s, then again the snowy 1960s, about average 1970s and after a final climax in the early 1980s a marked drop towards unprecedented low values during the 1990s.

Trends of the duration of continuous snow cover and related parameters
Time series of the duration of snow cover (d0) reveal large variability for different stations, regions and altitudes. The general pattern for d0 is that high stations (> 1600 m a.s.l.) and low stations (< 1000 m a.s.l.) are fairly constant over the last 70 years, whereas at mid altitudes (about 1000-1600 m a.s.l.) a strong increase culminating in the early 1980s was followed by a marked decrease towards the end of the century. However, exceptions are plentiful, and in dry interior areas the boundary between mid- and high-altitude characteristics reaches up to 1900 m a.s.l. This is found for absolute values; looking at relative deviations from the long-term mean reveals a slightly different picture. As for the HS winter mean, the decreasing trend since the early 1980s gets more pronounced from high- to low-altitude stations and looks near identical. At low elevations, small and hardly visible absolute trends obviously turn into large and noticeable relative changes.

Associated with the duration is the time of beginning (b0) and end (e0) of snow cover. Altogether, b0 is fairly constant over the last 70 years. On the other hand, e0 shows a clearer trend towards earlier snow melting during the 1980s and 1990s. In fact, most stations at all altitudes reveal this behaviour, and there are only a few exceptions. This implies that the generally shorter duration of snow cover is mainly caused by earlier snow melting than later first snowfalls. In contrast to that, the relative deviations from the long-term mean indicate that b0 is rising more sharply (especially at high and mid altitudes) and e0 remains virtually constant (with a slight decreasing tendency at low altitudes). However, the 11-year moving-average curves remain within +/- 10% of the long-term mean, which is little variability on the absolute scale.

For this paper, an effort was made to select a maximum number of consistent long-term snow series all over the Swiss Alps, divided into seven snow climatological regions (R1-R7) and including the adjacent forelands. R1-R3: North Slope, R4 and R7: the interior areas of the Alps, R5: Grisons and R6: South Slope. Trends of the average snow depth, the duration of snow cover, the number of snowfall days and heavy snowfall events are evaluated for a maximum period of 69 years (1931-99). The analyses are based on daily, manually measured snow depth (HS) and new snow (HN) data from the observational networks.

No systematic homogeneity tests and corrections were performed (e.g. Peterson et al., 1998), but only stations with consistent long-term records were used. Consistency was checked by looking at possible station shifts during the entire observation period. During the 1930s and 1940s the station coverage was rather meagre; the sharpest rise took place during the 1950s. Although regional density differences and some missing according to elevation, averaged over the whole of Switzerland, from the 1950s onwards the altitudinal distribution is evenly spread, and thus the dataset can be regarded as continuous through time. Only consistent long-term (> 25 years) stations within the period of 1931-99 were considered. This results in a total of 140 HS stations and 120 HN stations.

The mean seasonal snow depth from 1 November to 30 April is analysed. Apart from the whole winter, two-monthly part seasons have been studied. Duration (d0), beginning (b0) and final dates (e0) of snow cover are examined. Only the time period during which the ground was continuously covered with snow (HS > 0 cm) is considered. The same parameters are also determined for higher HS thresholds greater than 20, 30, 50 and 70 cm (d20-d70, b20-b70, e20-e70). Trends on daily new snowfall are analysed for long-term changes. The seasonal sum of daily HN correlates (r = 0.88) with the mean seasonal snow depth and thus shows very similar trends to the HS mean. Therefore, the HN sum is not discussed further here. However, the number of days with snowfall above several HN thresholds > 0, 10, 20 and 50 cm (dhn, dhn10-dhn50) is investigated. Additionally, the annual maximum new snow sum of three successive days (HN3max) is taken as an indicator for heavy snowfall events and is often connected to periods with spontaneous climax avalanches.

All calculations are performed with S-Plus (StatSci, 1993). A linear interpolation within a triangulation scheme (S-Plus functions interp and image) has been used too. The lowess smoother applies a robust locally linear fit to scatterplot data. The fraction f of the data used for smoothing at each point determines the window inside which points are weighted so that nearby points get the most weight. The larger the value of f , the smoother the fit. For smoothing long-term trends the default value f = 0.67 was taken.