This study focuses on an assessment of the future snow
depth for two larger Alpine catchments. Automatic weather station data from
two diverse regions in the Swiss Alps have been used as input for the
Alpine3D surface process model to compute the snow cover at a 200 m
horizontal resolution for the reference period (1999–2012). Future
temperature and precipitation changes have been computed from 20 downscaled
GCM-RCM chains for three different emission scenarios, including one
intervention scenario (2
The interannual snow volume is highly variable in the Alps. This is mainly caused by the combined effect of temperature and highly variable precipitation amounts (Bartolini et al., 2009). Consecutive winters with little snow or short snow duration can endanger the livelihood of tens of thousands because up to 90 % of the economy in Alpine villages depends on winter tourism (Abegg et al., 2007), whereas huge amounts of snow can cause destructive avalanches like in winter 1999 (Wilhelm et al., 2001). The Alps are already affected by climate change, mainly by increasing temperatures (Serquet et al., 2011). Several studies demonstrate the effect of these changes on the snow depth and snow duration (Scherrer et al., 2004; Durand et al., 2009; Marty, 2008), on snowfall (Valt and Cianfarra, 2010; Serquet et al., 2011) or snow water equivalent (SWE) (Marty et al., 2017).
Projections of the future winter climate reveal that temperatures will further increase, whereas the precipitation signal in the Alps is less clear (Gobiet et al., 2014; Zubler et al., 2014). On the point scale Schmucki et al. (2015) have shown that the depth of the future snow pack is clearly controlled by increasing temperatures and that the projected small increases in precipitation can only partly compensate the effect of the dominating warming signal and only at the highest elevations. On the regional/catchment scale so far the impact of these changes on the snow cover has mostly been investigated by either using GCM-RCM output directly with the limited ability to model high elevation changes (Mankin and Diffenbaugh, 2015) given the coarse spatial resolution of such models or using a limited set of high-resolution RCMs and emission scenarios: for example, Steger et al. (2013) assessed the trend of future SWE in the Alps from direct RCM output. Laghari et al. (2012) investigated the change of SWE and runoff in a catchment in the Austrian Alps by applying the conceptual hydrological model PREVAH to direct projections of a larger set of RCMs from the PRUDENCE project. Rousselot et al. (2012) modeled snowpack scenarios across the French Alps with the snow model CROCUS using the French RCM ALADIN. Marke et al. (2014) used the model AMUNDSEN with three RCMs and one emission scenario to simulate the future snow cover and ski season length for a small region in the central Austrian Alps.
Elevations in the Aare (left) and the Grisons region (right). Weather stations used for simulations are marked with dots. The location of the two regions within Switzerland is shown in the red inset with the two grey areas.
The goal of this study is to investigate the impact of climate change on the
Alpine snow cover with the surface process model Alpine3D
(Lehning et al., 2006). In contrast to Bavay
et al. (2009, 2013), which used Alpine3D to analyze the changes in
runoff in eastern Switzerland, we use a full set of RCM simulations, mean
and extreme changes, three emission scenarios and two diverse regions to
investigate the difference in snow cover between two non-intervention
scenarios and a climate stabilization scenario that supposes global
emissions will be cut by about 50 % by 2050. This scenario likely limits
global warming to 2
Two different catchments (the Aare region in central Switzerland and the
Grisons region in eastern Switzerland) were chosen to assess the future
evolution of the Alpine snow pack (Fig. 1). The
Aare region covers 3190 km
The Grisons region is three times larger (10 373 km
The meteorological data were provided by 48 automatic weather stations (AWS) in the Aare region and 34 AWS in the Grisons region at hourly resolution. For both regions the large majority of the stations were located between 500 and 2500 m a.s.l. The stations were selected based on the requirement that they provide hourly meteorological data and are located in or close to the simulation region. The following meteorological parameters were used for model input: Air temperature, relative humidity, wind velocity, precipitation, shortwave radiation and (if available) snow surface temperature and ground surface temperature. Incoming longwave radiation (ILWR) was parameterized and precipitation was corrected for wind-induced undercatch as described in Schlögl et al. (2016). The years between 1999 and 2012 were selected as reference period in order to keep the data availability optimal. This 13-year period contains one of the most snow abundant (2009) and one of the most snow-scarce winters (2007) in the last 30 years. The fact that not all stations provided the same number of parameters was not a problem since each parameter was individually interpolated to the model grid (see next section).
The snow pack was simulated with the surface process model Alpine3D. Alpine3D is a spatially distributed (surface), three-dimensional (atmospheric) model for analyzing and predicting the dynamics of snow-dominated surface processes in mountainous topography. It consists of a snow cover (SNOWPACK) and optional modules like runoff, vegetation, soil and snow transport (Lehning et al., 2006). Essential input data for the successful simulation were the following different surface grids and time series of meteorological parameters: a digital elevation model (DEM) with a horizontal resolution of 200 m was used in order to represent the topography of the two regions. The necessary land cover data were taken from CORINE (Bossard et al., 2000) with a horizontal resolution of 100 m and converted into the PREVAH classification (Viviroli et al., 2009). The PREVAH classification is less detailed than the CORINE classification, but sufficient for our simulations. These data were interpolated to the 200 m grid of the DEM by the nearest neighbor method. Since this study focusses on snow on ground but not snow on glaciers, the few pixels with glacier surfaces were removed in the post-processing in order to reduce the uncertainty of our results.
The meteorological data from the stations described above were all spatially interpolated to the grid of the DEM by inverse distance weighting (IDW), except the radiation components, which are physically calculated in the energy balance module. Vertical gradients were calculated based on the AWS data for hourly time steps using IDW. If the correlation coefficient of the vertical gradient is lower than 0.7, Alpine3D omits the data of the AWS with the worst fit to the linear regression. This process is iterated until the correlation coefficient becomes higher than 0.7. If this threshold is not reached, a constant value (independent of elevation) is assumed. For details see Bavay and Egger (2014). Finally, the different snow pack parameters were calculated for each pixel of the DEM grid based on the interpolated meteorological input parameter. The different soil and vegetation types were also considered, as well as the boundary layer parameters. For example, the roughness length was set to 7 mm and the height of the wind measurement was set to 3.5 m. The condition of the boundary layer was assumed to be neutral. The surface heat fluxes were parameterized using the Monin–Obukhov similarity theory.
Projections of future climate are provided as an extension of the CH2011
climate change initiative. This initiative provides among others daily
change values of temperature and precipitation for Switzerland on a 2 km
grid (Zubler et al., 2014), which are based on the 20 GCM-RCM
model chains of the EU-ENSEMBLES project (Van der Linden and Mitchell,
2009). Three emission scenarios (A2, A1B and RCP3PD) are provided for three
different time periods (2020–2049, 2045–2074, 2070–2099), which for
convenience are labeled by their middle year 2035, 2060 and 2085,
respectively. The three emission scenarios can briefly be described as
follows: in the RCP3PD scenario, the only interventional scenario, the
emissions are supposed to be halved by the middle of the century and thus the
CO
The assessment of the changes in temperature and precipitation are based on
20 different GCM-RCM ensemble combinations. The focus of this work is related
to the median estimate of these 20 different combinations, which were derived
by Bayesian methodology. The upper and lower estimates (extremes) of this
dataset, which contains the 97.5 % and 2.5 % quantiles, respectively, of
the 20 member ensembles are also considered for some analyses in order to
obtain information on the range of the uncertainties of the temperature and
precipitation changes. It is thus important to know that this Bayesian
methodology in some cases contracts the uncertainty range directly derived
from the variation of the original RCM simulations. Therefore, the
uncertainty range in this paper should also be seen as being indicative only.
A simple delta change approach was used to compile meteorological time series
of future Alpine climate. This means that the time series from the reference
period were modified with the provided gridded daily change air temperature
(
Mean January to March temperature (left) and precipitation (right) changes including the uncertainty bars from the upper and lower estimates for the Aare (red) and the Grisons (green) region.
In Fig. 2 the median estimate deltas and their uncertainty range are shown for the means of the months January to March. Slightly higher temperature changes in Grisons than in the Aare region are projected, especially for the end of the century. In this scenario period precipitation increases according to the A2 scenario by 4.3 % in the Aare region and 7.3 % in Grisons. The influence of the precipitation change is negligible compared to the temperature changes, because the predicted changes in precipitation are very small in the winter half year (Schmucki et al., 2015).
Due to the fact that the parameterized ILWR is a function of temperature, we calculated the parameterization of the ILWR for each emission scenario separately. This implies an emission scenario-dependent ILWR, which is necessary because ILWR fluxes contribute significantly to snowmelt, especially in spring (Schlögl et al., 2016).
Changes of glacier coverage were provided by Linsbauer et al. (2013) and used in order to adapt the land cover data to future scenarios. The changes were calculated with an elevation-dependent ice thickness model (M2) for the three emission scenarios and the three different time periods. Future glacier free areas were assumed as pixels with rocks in the land cover data. Note that the ice thickness model is only forced by the temperature change. Changes in the precipitation as seen in Fig. 2 were neglected in the model because the uncertainties in the assessment of future precipitation are too high. The current relative amount of glacial areas in the Aare region (6.7 %) is higher than in Grisons (1.7 %). The future glacier-covered area will be halved until 2060 and only a few pixels will still be covered with glaciers towards the end of the century.
We present projected changes of snow depth and duration for two Alpine regions based on the difference between the simulated values of the reference period and nine different climate projections (three time periods and three emission scenarios). The results are mainly based on the median estimate of all 20 model combinations, but in the last paragraph the uncertainty based on the 95 % spread (upper and lower estimates) is also shown. We often show results for both Alpine regions, but sometimes we focus on the Aare region only since the results are quite similar and its area below 500 m elevation is larger and more homogeneous than the corresponding elevation zone in the Grisons region.
By comparing the modeled Alpine3D snow depths of the reference period with measured snow depths, the model fidelity is estimated by means of the RMSE. The nearest and all neighboring pixels (nine in total) were considered for comparison with the station values. The pixel which showed the best agreement with the station elevation was chosen for comparison. The agreement is generally good, but in a heterogeneous topography like the Alps such a comparison will always be limited by the fact that the observations are point measurements in a flat field and the pixel value represents an average over an area which is inclined and at a different elevation. Moreover, measured snow depth in high-alpine flat fields usually is higher than the spatially averaged snow depth, e.g., from a grid cell (Grünewald and Lehning, 2015) and therefore generally not representative of a larger area. The RMSE was calculated for each of the 13 years of the reference period for the observed snow depths above 0.01 m (Table S1 in the Supplement). Mountain stations generally show a higher RMSE due to the topographical effect described above. Figure S1 illustrates some typical cases, where the simulated snow cover is either too great or too small: at the high-elevation station Weissfluhjoch (2540 m) in the Grisons region the simulation underestimates snow depth, whereas several stations between 1000 m and 1700 m a.s.l., especially Disentis (1190 m) tend to start snowmelt later than observed. This is also the reason why the simulated snow depth was overestimated for the lowland station Bern (542 m). This could be partly caused by a known limitation of the albedo function as described in Schmucki et al. (2015). High RMSE values at high-alpine sites are also explained by the fact that the measured precipitation is often heavily affected by the uncertainty of the undercatch correction and often shows poor elevation dependence because the regression is sometimes calculated across mountains ranges with different climates on each side. A regression across smaller areas or across the same climate regions (as in the Aare region) would probably improve the linear regression (Schlögl et al., 2016).
A comparison with the station-based approach of Schmucki et al. (2015), which
also used some of the stations in our investigated regions, demonstrates that
the error in simulating the mean winter snow depth in the reference period at
the point scale (between
Decrease in annual mean snow depth (%) relative to the reference period (1999–2012) for the Aare region and the Grisons region for the three different emission scenarios and time periods based on the median estimate change of temperature and precipitation (bars). The lowest and highest estimates (Table 3) are only shown for the Aare regions and A2 scenario (dots).
The impact on the mean snow depth has been investigated by computing the
temporal (13 hydrological years) and spatial (area of the region) mean
relative changes for the nine different climate projections. The calculated
mean snow depth can also be seen as a proxy for the mean snow volume. The
analysis reveals similar relative decreases for the two different regions
(Fig. 3). The Grisons region, however, always shows somewhat greater decreases,
which can be explained by the slightly higher
The delta change method applies changes in temperature and precipitation,
which depend only on time period and emission scenario but are otherwise
constant. Therefore changes in future climate variability, which may be
present in the original RCM model predictions, are neglected. According to
climate model projections there are no clear signs as to how future
temperature and precipitation variability will evolve in winter in
midlatitudes (Deser et al., 2012), although a recent study
indicates a slight decrease in winter temperature variability
(Holmes et al., 2016). The analyzed interannual variability in
this study is therefore first of all determined by the interannual
variability of the underlying temperature and precipitation conditions
during the reference period. For the future scenario periods the
interannual snow variability shown is additionally influenced by the
nonlinear dependence of snow on temperature, which changes the variability
dependent on the size of the
We also analyzed the mean snow depth evolution (mean of 13 years of simulation) and its variability (minimum and maximum snow depth for each day) in six elevation zones for the reference and the last scenario period for the A2 simulation (Fig. 4). The snow depth maxima at the end of the century are lower than today's mean snow depth in all elevation zones, except the highest (3000 m a.s.l.), where the maxima correspond more or less to today's mean values. On the other end, the mean snow depth evolution at the end of the century is similar to today's minima for all elevation zones except the highest.
Mean (solid), maximum and minimum (dotted) snow depth for the
reference period (black) and the A2 2085 scenario (blue) in the Aare domain
for 6 elevation zones. The elevation zones are 100 m wide, i.e., the 1500 m
zone, contains all pixels between 1450 and 1550 m. Note: the scale of the
Relative decrease of snow depth for different elevation zones and months in the Aare region for the 2085 A2 emission scenario.
As mentioned earlier, future snow depth is mainly dependent on the increasing winter temperature since the precipitation change in the winter half year is small. The evolution of the mean winter temperature and the maximum snow depth are therefore correlated (Fig. S4). In the 500 m (450–550 m) elevation zone, 6 (3, 0) of the 13 years in 2035 (2060, 2085) show a higher maximum snow depth than the lowest maximum in the reference period. The same figure also reveals that in the elevation zones between 500 and 1500 m a.s.l. the winter with the lowest maximum snow depth in the reference period corresponds to about the winter with the highest maximum snow depth at the end of the century. At the 2500 m, 12 (11, 8) out of 13 years show a higher maximum snow depth in 2035 (2060, 2085) than the lowest maximum of the reference period. At these higher elevations more winters remain with maximum snow depths higher than the current minimal snow depth. This is caused by the fact that the colder baseline climate makes snow conditions at higher elevations less sensitive to warming. The same results are also valid for the mean snow depth (not shown).
The relative decrease of the snow depth is dependent on time and elevation zone (Fig. 5). The highest relative decrease can be found in the lower elevations. Below 1000 m a.s.l. the relative decrease is over 70 % for all emission scenarios and time periods. Elevations above 2000 m a.s.l. are less sensitive to climate change. Nevertheless, even at 3000 m the snow depths will be halved towards the end of the century according to the A2 scenario (Table 1). This in good agreement with a study by Rousselot et al. (2012) in the French Alps, who found a 69 % decrease at 1800 m a.s.l. (compared to 75 % in our study). The graphs for the RCP3PD scenario demonstrate that the benefit of interventions is only discernible after the first scenario period and then mainly above 2000 m, where the snow cover reduction is limited to about 20 %. The begin and the end of the snow season are more sensitive to climate change due to generally warmer temperatures than the mid-winter months January and February, which is especially obvious in higher relative decreases in the spring months (Table 1). This finding is in agreement with the study by Steger et al. (2013), which also observed the highest snow cover reduction in spring.
Relative decrease (%) of the snow depth in the Aare region for the three emission scenarios (top, middle, bottom) and the three different time periods (left to right), dependent on season and elevation. The white colors indicate a lack of data.
Due to the fact that daily mean and maximum snow depths are decreasing, the total volume of snow must also shrink. In contrast to the relative decrease, the absolute decrease is small below 1000 m a.s.l. for the end of the century, since the usual snow volume is small anyway in this elevation zone (Fig. 6). The absolute decrease is greatest between 1500 and 2500 m a.s.l., since this elevation band is currently always snow covered during the winter months and heavily affected by higher temperatures. This is not the case above 3000 m a.s.l., where absolute decreases are again small since it is usually still cold enough to prevent melting during the winter months. This is also true for the interannual variability of the January to March period (shaded areas in Fig. 6), which clearly decreases with increasing elevation due to the fact that the snow volume is mainly dependent on precipitation and much less on temperature at higher elevations.
Total volume of snow (January–March) in the Aare region for today (solid line) and the end of the century (dotted line). The shaded area for the reference period indicates half of the standard deviation (for readability) of the interannual variability. The shaded area of the 2085 scenario period indicates the range between the lowest and highest estimate based on the A2 emission scenario (Table 3).
The date of the first continuous snow (snow depth at least 0.01 m) and the
end of the snow season were calculated based on the longest snow-covered
period for each of the 13 years for all time periods. Finally, the median of
these 13 years was calculated for 100 m elevation bands. The results of this
process for the Aare region and the A2 emission scenario are shown in
Fig. 7. At 1500 m, for example, the snow season
starts about 2 (2035) to 5 (2085) weeks later on average and ends 2 (2035)
to 11 (2085) weeks earlier. The temporal retreat of the snow disappearance
is also dependent on elevation, especially for the end of the century, when
the most sensitive elevation zone is roughly at 1500 m a.s.l. This is probably
caused by this elevation zone being closest to the 0
Begin and end of a continuous snow cover for the A2 emission scenarios for the reference and three future time periods in the Aare region.
The snow season at 1000 m a.s.l. currently lasts about 4 months from December until the end of March. At the end of the century almost no snow is projected at this elevation. A similar reduction of 4.5 months can be observed at 1500 m a.s.l., where the duration of continuous snow cover is reduced to only 2 months, i.e., mid-December to mid-February. It should be noted that these numbers are based on an average winter in the corresponding time period and neglect the fact that future winters at this elevation will often be characterized by ephemeral snow cover, which is nowadays only typical for elevations below 1000 m. This result is in good agreement with the findings of Schmucki et al. (2017), who demonstrate that at 1500 m a.s.l. in the Swiss Alps the probability for the occurrence of a winter with a continuous snow cover is only 60 % at the end of the century. Generally, the decrease in snow duration is equal to an elevation shift of 200–500 m for the first scenario period and 700–1000 m for the last scenario period for the A2 scenario. This is in agreement with a study by Bavay et al. (2013), who only used three RCMs and found similar values for the Swiss Alps. The slight bump at 2800 m in the curve of the first scenario period (Fig. 7) results from the lower glacier coverage in the period 2020–2049. Originally deleted pixels (due to glacier coverage) are now snow-covered pixels in this time period (see Sect. 2.3).
Interannual variability of the number of snow days (snow depth at least 5 cm) at three stations (Bern 540 m, left; Grindelwald 1030 m, center; and Mürren 1650 m, right) in the Aare region. The little square in the box plots represents the mean value and the whiskers show the 2.5 and 97.5 % quantile value of the different model simulations.
The demonstrated decrease in snow depth and snow duration also affects the number of snow days. We define a snow day as a day with a least 5 cm snow on the ground, because with regard to winter tourism this is the minimum snow depth to generate a winter feeling, build a snowman or go sledding. The number of snow days was calculated for the four time periods for several towns in the two investigated regions. Table S2 shows the median number of such snow days for the A2 scenario. The results clearly show that the number of snow days on the Swiss plateau will reach zero in the final scenario period. A multi-day snow cover will therefore be a rare event towards the end of the century in this elevation zone. Stations at about 1500 m will lose ca. 100 snow days, especially in the melting season. Davos (1560 m), for example, will only have 10 snow days more at the end of the century than Chur (593 m) has today and Adelboden (1350 m) will have less snow days than Bern (542 m) has at present.
The interannual variability in snow days is shown in Fig. 8 for three selected stations in the Aare region. The range is highest for Bern (542 m) at present, for Grindelwald (1034 m) in the first scenario period and for Mürren (1650 m) in the last scenario period, which corresponds well with the findings described in Sect. 3.3, where the elevation with the highest variability increased with time. Note that the inter-model variability, from which the median estimate is calculated, is much lower than the interannual variability as shown by Schmucki et al. (2017).
Assessment of the elevation-dependent natural snow reliability in the Aare region for north-facing (left) and south-facing (right) aspects. Red: occurrence of snow not reliable – a minimum snow depth of 30 cm was reached over less than 40 days. Green: snow reliable – at least 100 days with more than 30 cm snow depth. Yellow: the cases between the green and red definitions.
The probabilities of a winter with 0 snow days, less than 5, 15 or 50 snow days depending on elevation and scenario period, are shown Fig. S5. As expected, the same probability in future would be found at higher elevation. For example, there is a 7 % probability that we experience less than five snow days at 500 m a.s.l. today. During the middle of the century and using the A2 emission scenario the same probability can be found at 850 m a.s.l.
A higher snow day threshold has to be taken into account when the natural snow reliability for a ski resort is analyzed. The snow reliability is an important factor for a profitable ski resort and is directly correlated with the expected costs for the additional production of technical snow. A minimum snow depth of 30 cm during the 100-day period between 1 December and 15 April is often used a threshold for this purpose (Elsasser and Bürki, 2002), because experience shows that this is the minimum requirement for an economically viable ski area operation. To illustrate the declining elevation- and time-dependent natural snow reliability, the median number of days where at least 80 % of the pixels have a snow depth of at least 30 cm were therefore calculated in 200 m elevation bands. Figure 9 shows the current and future snow reliability calculated thus for the A2 scenario in the Aare region in north- and south-facing aspects. Elevations and time periods with less than 40 days and with at least 30 cm snow on the ground are shown in red. In contrast, green dots indicate a snow guarantee for the ski resorts (over 100 days with at least 30 cm snow). The cases between 40 and 100 days are labeled yellow, which indicates elevations and time periods, where the natural snow reliability is marginal and local effects may be a dominating factor.
According to this approach the natural snow cover is already definitely insufficient below 1000 m today. This elevation limit shifts to 1800 (2000) m a.s.l. at the end of the century for north- (south-) facing slopes. In contrast, sufficient snow can only be guaranteed above 1400 (1600) m a.s.l. for north- (south-) facing slopes today. In 2085, however, natural snow can only be guaranteed above 2400 (2600) m a.s.l. for northern (southern) aspects. This upward shift of the snow reliability of 800–1000 m between the reference period and the last scenario period is within the elevation shift range found for the continuous snow cover (Sect. 3.5). Compared to observations, the 200 m elevation difference in snow reliability between northern and southern slopes seems to be on the low side but can be explained by the following fact. Even the high-resolution 200 m DEM produces a smoother topography, in which slopes are less steep and therefore southern aspects less exposed to the low winter sun. In addition, small-scale processes due to rough terrain (e.g., enhanced melting in rocky slopes of southern aspect, drifting snow) are not considered in our modeling setup.
Snow is not only an important economic parameter for winter tourism but also plays an important role for the evolution of permafrost in high Alpine regions. This permafrost will probably thaw in a warmer climate if it is not protected by a deep and long-lasting snow pack in spring and summer (Haberkorn et al., 2015). The retreat of the permafrost in the alpine regions can affect the stability of tourism infrastructure in these areas or cause debris flows, which threaten populated areas far downstream (Haeberli et al., 2010). For this purpose we analyze the number of days with at least 30/50 cm of snow at 3000 m a.s.l. in all four time periods for the A2 scenario in the Aare region. There are currently 308 such snow days with at least 50 cm snow at this elevation (Table 2). The reduction is 30 % in the first scenario period, 40 % in the second and 60 % in the last one, leaving only 133 snow days. Note that this is a much higher reduction than the corresponding decrease in snow cover duration at the same elevation, which only accounts to ca. 30 % in the last scenario period (Fig. 7).
Number of days with over 30/50 cm snow depth at 3000 m a.s.l. in the Aare region for the reference period and the three future scenario periods based on the A2 emission scenario.
Ideally the uncertainty analysis incorporates every step in the modeling process. The biggest uncertainty has been assessed by considering three different emission scenarios. Another source of uncertainty comes from the snow model with regard to the resolution, the parameterized processes, the choice of the boundary layer parameters and the available meteorological stations to verify the RCM runs. Schlögl et al. (2016) concluded in a recent study that the uncertainty of the simulated SWE from these factors is typically ca. 15 % but is negligible in climate change studies as long as only relative changes are considered. As described in Zubler et al. (2014) there is also uncertainty in the RCM downscaling procedure. One important point to keep in mind is the interpretation of the high-elevation results, because the highest point in the ENSEMBLES grid is only 2600 m a.s.l. Moreover, above this elevation increasing exposure of rubble and till from glacier melt augments the potential for deposition of dust on the glacier surface, contributing to a lowered surface albedo and a positive feedback (Oerlemans et al., 2009), which is not taken into account in our study.
In the following we focus on the uncertainties originating from the
different temperature and precipitation changes as projected by the 20 different GCM-RCM chains available from the CH2011 initiative. So far the
focus was on projected changes based on the median value from the ensemble
of these different models. To investigate this uncertainty the snow cover
has also been simulated for the upper and lower upper lower upper lower
Simulating only the first two cases is not sufficient because the upper
Change (%) in snow depth relative to the reference period for the three future time periods based on the A2 emission scenario in the Aare region.
Not surprisingly, in each scenario period the highest relative decrease was
found for the upper
Figure S6 demonstrates the effect of these possible
combinations on snow depth dependent on elevation for all three scenario
periods: the upper panels show the impact on the absolute snow depth and the
difference relative to the mean changes as considered in the above
paragraphs. The lower panels illustrate the relative difference in snow
depth between the upper and lower
The difference in snow depth between the two extreme precipitation scenarios
is greatest at about 3300 m a.s.l. for the
However, the difference in snow depth between the two extreme
temperature scenarios peaks at about 1000 m a.s.l. for the 2035 scenario
period, indicating that this elevation zone is most sensitive to temperature
changes in the near future. The 2060 and 2085 scenario periods show a peak
at 1800 m and 2300 m a.s.l., respectively, demonstrating that the most sensitive elevation
zone is increasing with time independently of the precipitation change
(
The large set of downscaled climate models used in this study demonstrates a
clear temperature increase for all time periods and emission scenarios. In
contrast, the precipitation signal diverges between the individual models.
Future seasonal temperature increase is projected to be highest in summer
and lowest in spring with only slightly higher changes in winter and autumn.
This is in contrast to the observed temperature increases during the last
decades, which are highest in spring, closely followed by summer. Median
estimate values of these projected changes together with a measured
climatology reference period were used as input for Alpine3D to analyze the
impact of these changes on the future snow cover for two different Alpine
regions. Our results corroborate the general findings of earlier studies,
but they quantify the uncertainty better because the median estimate and the
lower and upper bounds (2.5 and 97.5 % quantiles) from 20 different
GCM-RCMs were used as input. Moreover, in addition to the widely used A1B
and A2 emission scenarios, the benefit of an intervention scenario (RCP3PD)
was investigated, which allowed analyzing how much Alpine snow can be saved
if we manage to stabilize the global temperature increase below 2
The results demonstrate that the duration and mass of the snow cover in typical Alpine catchments such as the Aare and Grisons region will shrink until the end of the century, independently of the emission scenario and the climate model used. However, the magnitude of the decrease can be heavily reduced with an intervention scenario. Both regions show a similar clear reduction in the future snow volume (January–March) based on the median estimate values. For the A1B emission scenario the expected snow volume reduction averaged over both region will be about 25 % in the near future (2035), 50 % towards the middle of the century (2065) and 60 % towards the end of the century (2085). The higher A2 scenario differs only at the end of the century, where the reduction increases to ca. 70 %. The RCP3PD scenario, however, can limit the expected snow reduction to 30 % after the middle of the century.
Since the current emissions do not follow the RCP3PD track
(Peters et al., 2013), the following paragraph refers to
the A2 scenario: the expected snow volume reduction increases from ca. 50 % above 3000 m to almost 100 % at lowest elevations (500 m a.s.l.).
Similarly, the reduction increases, e.g., at 1500 m a.s.l. and for middle of the
century from ca. 50 % in mid-winter to almost 100 % in spring. A
detailed analysis of the interannual variability demonstrates that the snow
cover of a nowadays snow-scarce winter below 1500 m can be expected to be
the average snow cover at the end of the century. The elevation of the
largest absolute declines is increasing with time from 1000 m a.s.l. during the
first scenario period (2035) to 1800 m a.s.l. during the second period (2065)
to 2300 m a.s.l. during the third period (2085), because the elevation with the
conditions for melt and the maximum available snow are also increasing with
time. It is obvious that such reductions in snow volume also imply a
decrease in snow duration. Our analysis reveals that the snow duration at
2000 m decreases by 2 weeks in 2035 and by 11 weeks in 2085. Thus, in the
mid-century at low elevations between 500 and 1000 m a.s.l. there will already
only be a few days with snow cover. The generally shorter and thinner future
snow cover is equivalent to an elevation shift of ca. 400 m in 2035 and ca.
900 m in the last scenario period. These values demonstrate that projected
snow reduction is highly dependent on elevation and season. Considering the
lower and upper bounds of the projections reveals that the snow volume
reduction has an uncertainty of about
The clear decrease in future snow depth and snow duration as shown above negatively affects society by decreasing natural snow reliability for ski resorts and by changing the influence of the snow cover in high-elevation permafrost areas. Since the snow cover of the two investigated regions finally ends as meltwater in three of the major rivers in central Europe (Rhine, Danube and Po) the changing seasonal runoff might heavily impact water usage downstream (e.g., hydropower, irrigation, transportation), especially during the projected dryer summer months (Beniston and Stoffel, 2014). Furthermore, our results clearly demonstrate that at low elevations, where the majority of the population in the Alpine area lives, a multi-day snow cover will become a rare event after middle of the century. In these heavily populated areas the vanishing snow cover due to warmer winter temperatures may also have positive side effects on society because a decreasing number of frost and snow days are positively correlated with the number of road accidents (Norrman et al., 2000), airport closures (Hess et al., 2009), traffic interruptions as well as the costs for winter road maintenance (Schmidlin, 1993).
We want to reemphasize (1) that the projections are based on the Delta change approach, which implies that the variability does not change over time, and (2) that the presented results are mostly based on the expected median estimate changes and on a climatologically averaged snow cover. The uncertainty analysis demonstrates that the range of uncertainty in the simulated snow cover decrease is determined by the interannual variability and the uncertainties in the climate change signal of the different RCM projections.
The Alpine3D modeled snow projections used in this study are available at
The authors declare that they have no conflict of interest.
The meteorological input data used for the simulation of the reference period were provided by the Swiss Federal Office of Meteorology and Climatology MeteoSwiss. The snow depth data used for verification purposes were provided by SLF and MeteoSwiss. The climate change signal data were obtained from the Center for Climate Systems Modeling (C2SM) at ETH Zürich provided by the CH2011 community. We thank Marcia Phillips for proofreading the manuscript. The study was partly financed by the Swiss national science foundation (#200021_150146) and by a special grant of the Swiss Federal Institute for Forest, Snow and Landscape Research WSL. Edited by: R. Brown Reviewed by: two anonymous referees