Most studies show negative trends in snow depth and snow duration over the past decades (Table 1). These negative trends are well documented in the Alps due to the abundance of long-term observations. The changes are typically elevation dependent, with more (less) pronounced changes at low (high) elevations (Marty, 2008; Durand et al., 2009; Terzago et al., 2013). Decrease in spring snow water equivalent (SWE) is equally found for the Alps (Bocchiola and Diolaiuti, 2010; Marty et al., 2017b; Fig. 1) as well as for low elevations in Norway (Skaugen et al., 2012). Only the higher and colder regions of the Fennoscandian mountains exhibit positive trends of maximum snow depth and maximum SWE, although trends have recently become negative in these regions too (Johansson et al., 2011; Skaugen et al., 2012; Dyrrdal et al., 2013; Kivinen and Rasmus, 2015). In the Pyrenees, a significant reduction of the snowpack is reported since the 1950s (Pons et al., 2010), and in other European mountains – where observations are less abundant – studies also report declining snowpacks. The latter is particularly true for mountains in Romania (Birsan and Dumitrescu, 2014; Micu, 2009), Bulgaria (Brown and Petkova, 2007), Poland (Falarz, 2008), and Croatia (Gajić-Čapka, 2011).

The observed changes in snow depth and snow duration are mainly caused by a shift from solid to liquid precipitation (Serquet et al., 2011; Nikolova et al., 2013) and by more frequent and more intense melt (Klein et al., 2016), both resulting from higher air temperatures during winter and spring. In addition to a general warming trend, large-scale atmospheric circulation patterns such as the North Atlantic Oscillation (NAO) have been shown to influence the snow cover in Europe (Henderson and Leathers, 2010; Bednorz, 2011; Skaugen et al., 2012; Birsan and Dumitrescu, 2014; Buisan et al., 2015). For the Alps, 50 % of the snowpack variability seems to be related to the establishment of atmospheric blocking patterns over Europe, although in this case the correlation between the annual snowpack variability and the NAO is weak and limited to low elevations (Scherrer and Appenzeller, 2006; Durand et al., 2009). At higher elevations, the NAO influence can be detected a through a “cascade” of processes that include the manner in which the positive or negative NAO modes translate into surface pressure in the Alps, Pyrenees, or Scandes and thus influence temperature and precipitation according to the resulting pressure patterns in the different European mountain regions. Together with the effect of air temperature, this determines the amount of snowfall. In recent decades, this cascade has led to an increased number of warm and dry winter days, which obviously is unfavorable for snow accumulation (Beniston et al., 2011b). In addition, the Atlantic Multidecadal Oscillation – a natural periodic fluctuation of North Atlantic sea surface temperature – has been shown to affect the variability of alpine spring snowfall, contributing to the decline in snow cover duration (Zampieri et al., 2013).

The observed changes in snow amounts are often abrupt. Several studies have reported a step-like change for snow depth occurring in the late 1980s (Marty, 2008; Durand et al., 2009; Valt and Cianfarra, 2010) and for snow-covered areas of the Northern Hemisphere (Choi et al., 2010). This step-like development, also observed for other compartments of the environment (Reid et al., 2016), has been suggested to be linked to atmospheric internal variability (Li et al., 2015) and shrinking sea-ice extent (Mori et. al. 2014). Since that step change, the monthly mean snow-covered area in the Alps has not decreased significantly (Hüsler et al., 2014), and winter temperatures in large areas of the Northern Hemisphere (Mori et. al. 2014) and in the Swiss Alps (Scherrer et al., 2013) have been stagnating.

Studies analyzing high-magnitude snowfalls are rare, but they indicate that extreme snow depths have decreased in Europe (Blanchet et al., 2009; Kunkel et al., 2016), with the exception of higher and colder sites in Norway (Dyrrdal et al., 2013). The decrease in extreme snowfall rates is less clear, except for low elevations where the influence of increasing air temperature is predominant (Marty and Blanchet, 2012). Studies related to past changes in snow avalanche activity are scarce as well, but observations indicate that over the last decades (a) the number of days with prerequisites for avalanches in forests decreased (Teich et al., 2012), (b) the proportion of wet snow avalanches increased (Pielmeier et al., 2013), and (c) the runout altitude of large avalanches retreated upslope (Eckert et al., 2010, 2013; Corona et al., 2013) as a direct consequence of changes in snow cover characteristics (Castebrunet et al., 2012).

Regional climate model simulations show a dramatic decrease both in snow cover duration and SWE for Europe by the end of the 21st century (Jylhä et al., 2008). It has to be noted, however, that the projected increase in air temperature for coming decades is accompanied by large uncertainties in changes of winter precipitation. For mainland Europe, climate models show no clear precipitation change until the 2050s and slightly increasing winter precipitation thereafter. Projections for regional changes in snow cover are thus highly variable and strongly depend on applied emission scenarios and considered time period (e.g., Marke et al., 2015).

For the Alps at an elevation of 1500 m a.s.l. (above sea level), recent simulations project a reduction in SWE of 80–90 % by the end of the century (Rousselot et al., 2012; Steger et al., 2013; Schmucki et al., 2015a). According to the same simulations, the snow season at that altitude would start 2–4 weeks later and end 5–10 weeks earlier than in the 1992–2012 average, which is equivalent to a shift in elevation of about 700 m (Marty et al., 2017a). For elevations above 3000 ma.s.l., a decline in SWE of at least 10 % is expected by the end of the century even when assuming the largest projected precipitation increase. Future climate will most probably not allow for the existence of a permanent snow cover during summer even at the highest elevations in the Alps, with obvious implications for the remaining glaciers (Magnusson et al., 2010; Bavay et al., 2013) and the thermal condition of the ground (e.g., Marmy et al., 2016; Draebing et al., 2017; Magnin et al., 2017).

Projections for Scandinavia show clear decreases for snow amount and duration. Exceptions are the highest mountains in Northern Scandinavia, where strongly increasing amounts in precipitation could compensate the temperature rise and result in marginal changes only (Räisänen and Eklund, 2012). Simulations for the Pyrenees indicate a decline of the snow cover similar to that found for the Alps (López-Moreno et al., 2009). Again, the dependency on future emissions is significant: for a high emission scenario (RCP8.5), SWE decreases by 78 % at the end of the 21st century at 1500 ma.s.l. elevation are expected, whereas a lower emission scenario (RCP6.0) projects a decline of 44 %.

Extreme values of snow variables are often the result of a combination of processes (e.g., wind and topographic influence for drifting snow), making predictions of their frequency highly uncertain (IPCC, 2012, 2013). By the end of the 21st century, models suggest a smaller reduction in daily maximum snowfalls than in mean snowfalls over many regions of the Northern Hemisphere (O'Gorman, 2014). An investigation for the Pyrenees (López-Moreno et al., 2011), however, finds a marked decrease in the frequency and intensity of heavy snowfall events below 1000 ma.s.l. and no change in heavier snowfalls for higher elevations. Changes in extreme snowfall and snow depth are also likely to depend on compensation mechanisms between higher temperatures (Nicolet et al., 2016), more intense precipitation, and increased climate variability, rendering any prediction of future snow storm frequency, magnitude, and timing difficult. In addition, most available studies either deal with marginal distributions or postulate stationarity (Blanchet and Davison, 2010; Gaume et al., 2013), two approaches that are questionable in the context of a changing climate.

Changes in snow cover duration and snow depth, as well as other snow properties, will have an effect on various ecosystems: earlier snowmelt is associated with an anticipation of plant phenology (Pettorelli et al., 2007) which can potentially induce a mismatch between plant blooming and herbivore activity, similar to observations in Arctic regions (Post et al., 2009). In the case of alpine ibex, for example, snow has been found to have a dual effect: too much winter snow limits adult survival, whereas too little snow produces a mismatch between alpine grass blooming and herbivore needs (Mignatti et al., 2012). Abundance of alpine rock ptarmigan populations, in contrast, has been shown to depend on the onset of spring snowmelt and the timing of autumn snow cover (Imperio et al., 2013). The increasing number of hard (icy) snow layers due to higher temperatures, however, can have a significant effect on the life of plants and animals (Johansson et al., 2011).

In addition, human activities will be influenced by the anticipated changes. The reduction of the snow season duration, for example, will have severe consequences for winter tourism (Uhlmann et al., 2009; Steiger and Abegg, 2013; Schmucki et al., 2015b), water management (Laghari et al., 2012; Hill-Clarvis et al., 2014; Gaudard et al., 2014; Köplin et al., 2014), and ecology (Hu et al., 2010; Martz et al., 2016). Similarly, change in moisture content or density of snow will affect infrastructure stability under extreme loading (Sadovský and Sykora, 2013; Favier et al., 2014).

The effect of climate change on avalanche risk is largely unknown; although empirical relations between snow avalanche activity and climate exist (Mock and Birkeland, 2000), the knowledge is insufficient for sound long-term projections. With a few exceptions, existing studies on avalanche–climate interactions focus on recent decades and are very local in scope (Stoffel et al., 2006; Corona et al., 2012, 2013; Schläppy et al., 2014, 2016). Direct effects of climate change on avalanche frequency, timing, magnitude, and type mainly exist in form of changes in snow amounts, snowfall succession, density, and stratigraphy as a function of elevation. The trend towards wetter snow avalanches is expected to continue, although the overall avalanche activity will decrease, especially in spring and at low elevations (Martin et al., 2001; Castebrunet et al., 2014). In contrast, an increase in avalanche activity is expected at high elevations in winter due to more favorable conditions for wet snow avalanches earlier in the season (Castebrunet et al., 2014). Even if the expected rise of tree line elevation may reduce both avalanche frequency and magnitude, present knowledge on avalanche–forest interactions is incomplete (Bebi et al., 2009). Due to the highly nonlinear nature of avalanche triggering response to snow and weather inputs (Schweizer et al., 2003) and to the complex relations between temperature, snow amounts, and avalanche dynamics (Bartelt et al., 2012; Naaim et al., 2013), it remains unclear whether warmer temperatures will indeed lead to fewer avalanches because of less snow. The most destructive avalanches, moreover, mostly involve very cold and dry snow resulting from large snowfall, but they may also result from wet snow events whose frequency has increased in the past (Sovilla et al., 2010; Castebrunet et al., 2014; Ancey and Bain, 2015). Finally, mass movements involving snow often occur at very local scales, making them difficult to relate to climate model outputs, even with downscaling methods (Rousselot et al., 2012; Kotlarski et al., 2014).

Glaciers in mainland Europe cover an area of nearly 5000 km2 (Table 2) and have an estimated volume of almost 400 km3 (Huss and Farinotti, 2012; Andreassen et al., 2015). From historical documents such as paintings and photography (Zumbühl et al., 2008), it is clear that glaciers have undergone substantial mass loss since the 19th century (Fig. 2) and that the pace of mass loss has been increasing (Zemp et al., 2015). A loss of 49 % in the ice volume was estimated for the European Alps for the period 1900–2011 (Huss, 2012). Repeat inventories have shown a reduction in glacier area of 11 % in Norway between 1960 and the 2000s (-0.28 %yr-1) (Winsvold et al., 2014) and 28 % in Switzerland between 1973 and 2010 (-0.76 %yr-1) (Fischer et al., 2014). Periods with positive surface mass balance have, however, occurred intermittently, notably from the 1960s to the mid-1980s in the Alps and in the 1990s and 2000s for maritime glaciers in Norway (Zemp et al., 2015; Andreassen et al., 2016). Glacier area loss has led to the disintegration of many glaciers, which has also affected the observational network (e.g., Zemp et al., 2009; Carturan et al., 2016). In the more southerly parts of Europe, glacier retreat in Pyrenees has accelerated since the 1980s and the small glaciers are currently mostly in a critical situation (López-Moreno et al., 2016; Rico et al., 2017).

Glacier retreat during the 20th century has mainly been attributed to changes in atmospheric energy fluxes and associated air temperature changes. A good correlation between air temperature and melt exists, making long-term air temperature time series the favorite option to explain 20th century glacier retreat (Haeberli and Beniston, 1998). High melt rates in the 1940s have, however, also been associated to changes in solar radiation (Huss et al., 2009). Several studies used calibrated temperature-index methods to simulate snow and ice melt responses to atmospheric forcing (Braithwaite and Olesen, 1989; Pellicciotti et al., 2005) although the appropriateness of such approaches for long-term studies has often been debated (Huss et al., 2009; Gabbi et al., 2014; Réveillet et al., 2017). Glacier response to atmospheric forcing is driven by different factors, ranging from synoptic weather patterns to local effects enhanced by topography. The latter influences, among others, the distribution of precipitation, solar radiation, and wind. Several studies have shown that glacier mass balance can be influenced by the NAO. Glacier advances in Scandinavia during 1989–1995, for example, are attributed to increased winter precipitation linked to the positive NAO phase during that period (Rasmussen and Conway, 2005). In the European Alps, the relationship between the NAO and glacier surface mass balance is less pronounced (Marzeion and Nesje, 2012; Thibert et al., 2013). This is essentially because the Alps are often a “pivotal zone” between southern and northern Europe, where the correlations between the NAO index and temperature or precipitation tend to be generally stronger (i.e., in the Mediterranean zone and in Scandinavia).

Glacier evolution during the 20th century also highlights the importance of the surface albedo feedback, as albedo governs the shortwave radiation budget at the glacier surface, which is the dominant energy source for melting. The sensitivity of ablation to albedo has generally been assessed using energy-balance considerations (Six and Vincent, 2014) or degree-day approaches (e.g., Pellicciotti et al., 2005). Oerlemans et al. (2009) and Gabbi et al. (2015) investigated the influence of accumulation of dust or black carbon on melt rates for Swiss glaciers in the last decades, revealing annual melt rates increased by 15–19 % compared to pure snow. Monitoring, reconstructing, or modeling the surface albedo of glaciers is challenging (Brock et al., 2000) as its spatial and temporal evolution is linked to changes in surface properties (mainly snow grain size and grain shape) and to the deposition of impurities on the ice. Albedo changes are also determined by snow deposition (amount and spatial distribution), making the annual surface mass balance highly sensitive to snow accumulation (Réveillet et al., 2017). Properly quantifying the amount and distribution of accumulation over glaciers is therefore a key to better assess the glacier surface mass balance sensitivity to changes in climate and to simulate its future evolution (Sold et al., 2013).

Glacier dynamics are influenced by numerous variables such as mass change and basal hydrology for temperate glaciers and by ice temperature changes for cold glaciers. In temperate glaciers, ice dynamics is mainly driven by thickness changes and the basal hydrological system, which in turn affects basal sliding. The large decrease in ice thicknesses over the last three decades has led to a strong reduction in ice flow velocities (Berthier and Vincent, 2012). Increased water pressure, in contrast, reduces the frictional drag and thus increases the sliding rate. Sliding velocities are low when the water under glaciers drains through channels at low pressure and high when the water drains through interconnected cavities (Röthlisberger, 1972; Schoof, 2010). Although changes in seasonal ice flow velocities are driven by subglacial hydrology, it seems that, at the annual to multiannual timescales, the ice flow velocity changes do not depend on changes in subglacial runoff (Vincent and Moreau, 2016). A few temperate alpine glaciers, such as the Belvedere Glacier in Italy, have shown large accelerations due to a change in subglacial hydrology (Haeberli et al., 2002), whereby the mechanisms of this surge-type movement remain unclear. In some rare cases, the reduction of the efficiency of the drainage network followed by a pulse of subglacial water triggered a catastrophic break-off event as in the case of Allalingletscher in 1965 and 2000 (Faillettaz et al., 2015).

Studies of cold glaciers in the Monte Rosa and Mont Blanc area revealed that englacial temperatures have strongly increased over the last three decades due to rising air temperatures and latent heat released by surface meltwater refreezing within the glacier (Lüthi and Funk, 2000; Hoelzle et al., 2011; Gilbert et al., 2014). A progressive warming of the ice is expected to occur and propagate downstream. As a result, changes of basal conditions could have large consequences on the stability of hanging glaciers (Gilbert et al., 2014). Such changes in basal conditions are understood to be responsible for, e.g., the complete break-off of Altels Glacier in 1895 (Faillettaz et al., 2015).

Over the last two decades, various studies on potential future glacier retreat in Europe have been published. These can be broadly classified into site-specific (e.g., Giesen and Oerlemans, 2010) and regional studies (e.g., Salzmann et al., 2012). Methods range from simple extrapolation of past surface or length changes to complex modeling of glacier mass balance and ice flow dynamics. Similarly, applied models range from simple degree-day approaches (e.g., Braithwaite and Zhang, 1999; Radic and Hock, 2006; Engelhardt et al., 2015) to complete surface energy-balance formulations (e.g., Gerbaux et al., 2005) or from simple parameterizations of glacier geometry change (Zemp et al., 2006; Huss et al., 2010; Linsbauer et al., 2013), over flow line models (e.g., Oerlemans, 1997; Oerlemans et al., 1998), to three-dimensional ice flow models solving the full Stokes equations (Le Meur et al., 2004; Jouvet et al., 2011; Zekollari et al., 2014). All models indicate a substantial reduction of glacier ice volume in the European Alps and Scandinavia by the end of the century. Small glaciers are likely to completely disappear (Linsbauer et al., 2013), and even large valley glaciers, such as Great Aletsch Glacier, Rhône Glacier, Morteratsch Glacier (Switzerland), or ice caps such as Hardangerjøkulen and Spørteggbreen (Norway), are expected to lose up to 90 % of their current volume (Jouvet et al., 2009, 2011; Giesen and Oerlemans, 2010; Farinotti et al., 2012; Laumann and Nesje, 2014; Zekollari et al., 2014; Åkesson et al., 2017). Many glacier tongues will disappear, including the one from Briksdalsbreen, the outlet glacier of mainland Europe's largest ice cap Jostedalsbreen (Laumann and Nesje, 2009).

At the scale of mountain ranges, model studies relying on medium-range emission scenarios consistently predict relative volume losses of 76–97 % for the European Alps and of 64–81 % for Scandinavia (Marzeion et al., 2012; Radic et al., 2014; Huss and Hock, 2015) for the 21st century. Since the mountain glaciers in Europe are far out of balance with the present climate (e.g., Andreassen et al., 2012a; Mernild et al., 2013), such volume losses must be expected even with strong efforts to reduce CO2 emissions and to stabilize global warming at less than +2 ∘C as recommended by the Paris COP-21 climate accord (Huss, 2012; Salzmann et al., 2012). Due to their limited altitudinal extent, many glaciers are unable to reach a new equilibrium with climate even if air temperatures were stabilized by the end of this century. Furthermore, ice caps in Norway that contribute to a large part of the total ice volume in Europe (Table 2) are highly sensitive to mass balance–altitude feedback due to their hypsometry and large ice thicknesses. Model experiments suggest that Hardangerjøkulen will not regrow with its present mass balance regime once it has disappeared (Åkesson, 2017). However, uncertainties in projections of future glacier evolution are still considerable and improvements are required in both the quality of the input data and the physical basis upon which glaciological models are built (see Sect. ).

Permafrost borehole temperatures are monitored in many European mountain ranges (documented and available in the Global Terrestrial Network for Permafrost (GTN-P) database; Biskaborn et al., 2015), several of the sites being accompanied by meteorological stations and ground surface temperature measurements (Gisnås et al., 2014; Staub et al., 2016). However, as mountain permafrost is usually invisible from the surface, various indirect methods need to be employed to detect, characterize, and monitor permafrost occurrences. These methods include surface-based geophysical measurements to determine the physical properties of the subsurface, including water and ice content distributions (Kneisel et al., 2008; Hauck, 2013), and geodetic and kinematic measurements to detect subsidence, creep, and slope instabilities (Kääb, 2008; Lugon and Stoffel, 2010; Kaufmann, 2012; Kenner et al., 2014; Arenson et al., 2016).

The longest time series of borehole temperatures in Europe started in 1987 at the Murtèl-Corvatsch rock glacier in the Swiss Alps (Haeberli et al., 1998; Fig. 3), a period that is much shorter compared to the available ones for the other cryospheric components such as snow (see Sect. ) or glaciers (see Sect. ). The past evolution of permafrost at centennial timescales can to some extent be reconstructed from temperature profiles in deep permafrost boreholes (e.g., Isaksen et al., 2007), pointing to decadal warming rates at the permafrost table on the order of 0.04–0.07 ∘C for Northern Scandinavia. Permafrost has been warming globally since the beginning of the measurements (Romanovsky et al., 2010; Noetzli et al., 2016; Fig. 3). This warming was accompanied by an increase of the thickness of the seasonal thaw layer (hereafter referred to as the active layer; Noetzli et al., 2016). The considerable year-to-year variability can be linked to variations in the snow cover, as a reduction in snow cover thickness reduces thermal insulation. Latent heat effects associated with thawing mask the recent warming trend for “warm” permafrost sites (temperatures close to the freezing point), which is otherwise clearly visible in cold permafrost (see Fig. 3).

The increasing trend in permafrost temperatures and especially the deepening of the active layer has been hypothesized to lead to an increased frequency of slope instabilities in mountain ranges, including debris flows and rockfalls (Gruber and Haeberli, 2009; Harris et al., 2009; Bommer et al., 2010; Stoffel, 2010; Fischer et al., 2012; Etzelmüller, 2013; Stoffel et al., 2014a, b). The disposition conditions and triggering mechanisms of slope instabilities can be diverse and depend on subsurface material (e.g., unconsolidated sediments vs. bedrock), its characteristics (fractures and fissures, ice and water content, slope angle, geological layering), and changes of these properties with time (Hasler et al., 2012; Krautblatter et al., 2012; Ravanel et al., 2013; Phillips et al., 2017). Water infiltration into newly thawed parts of permafrost is often mentioned as a possible triggering mechanism (Hasler et al., 2012), but only few observational data are available to confirm this hypothesis. By means of tree-ring reconstructions (Stoffel et al., 2010; Stoffel and Corona, 2014), the temporal evolution of debris-flow frequencies has been addressed for a series of high-elevation catchments in the Swiss Alps. These studies point to increased debris-flow activity as a result of climate warming since the end of the Little Ice Age (Stoffel et al., 2008; Bollschweiler and Stoffel, 2010a, b; Schneuwly-Bollschweiler and Stoffel, 2012) and a dependence of debris-flow magnitudes due to instabilities in the permafrost bodies at the source areas of debris flows (Lugon and Stoffel, 2010; Stoffel, 2010).

Several studies have documented recent events of rock slope failures in the Alps (Ravanel et al., 2010; Ravanel and Deline, 2011; Huggel et al., 2012; Allen and Huggel, 2013). Some of these failures are clearly related to deglaciation processes (Fischer et al., 2012; Korup et al., 2012; Strozzi et al., 2010). Unusually high air temperatures have additionally been associated with these processes as the penetration of meltwater from snow and ice into cleft systems results in a reduction of shear strength and enhanced slope deformation (Hasler et al., 2012). Considering the multiple factors that affect rock slope stability, however, it is generally difficult to attribute individual events to a single one (Huggel et al., 2013). Improved integrative assessments are therefore necessary.

Further evidence of climatic impacts on high mountain rock slope stability comes from the analysis of historical events. For the Alps, inventories documenting such events exist since 1990 (Ravanel and Deline, 2011; Huggel et al., 2012) and indicate a sharp increase in the number of events since 1990. This makes the temporal distribution of rock slope failures resembles the evolution of mean annual temperatures. Given the fact that monitoring and documentation efforts have been intensified during the past decades, it remains unclear to which degree this correlation is affected by varying temporal completeness of the underlying datasets.

Data on the ice volume stored in permafrost and rock glaciers are still scarce. To date, hydrologically oriented permafrost studies have been utilizing remote-sensing and meteorological data for larger areas or have had a regionally constrained scope such as the Andes (Schrott, 1996; Brenning, 2005; Arenson and Jacob, 2010; Rangecroft et al., 2015), the Sierra Nevada (Millar et al., 2013) or Central Asia (Sorg et al., 2015; Gao et al., 2016). To our knowledge, no systematic studies exist on permafrost–hydrology interactions for the European mountain ranges to date.

In site-specific model studies, subsurface data are only available from borehole drillings and geophysical surveying. The models used often have originated from high-resolution hydrological models (e.g., GEOtop; Endrizzi et al., 2014), soil models (e.g., COUP model; Jansson, 2012; Marmy et al., 2016), or snow models (such as Alpine3D/Snowpack; Lehning et al., 2006; Haberkorn et al., 2017) and have been successfully extended to simulate permafrost processes. Recently, explicit permafrost models have been developed as well (e.g., Cryogrid 3; Westermann et al., 2016).

Because of its complexity, permafrost evolution cannot be assessed by thermal monitoring alone. Kinematic and geophysical techniques are required for detailed process studies. Kinematic methods are used to monitor moving permafrost bodies (e.g., rock glaciers) and surface geometry changes. Hereby, methods based on remote sensing allow for kinematic analyses over large scales (Barboux et al., 2014, 2015; Necsoiu et al., 2016) and the compilation of rock-glacier inventories (e.g., Schmid et al., 2015), whereas ground-based and airborne kinematic methods focus on localized regions and on the detection of permafrost degradation over longer timescales (Kaufmann, 2012; Klug et al., 2012; Barboux et al., 2014; Müller et al., 2014; Kenner et al., 2014, 2016; Wirz et al., 2014, 2016). Long-term monitoring of creeping permafrost bodies shows an acceleration in motion during recent years, possibly related to increasing ground temperatures and higher internal water content (Delaloye et al., 2008; Ikeda et al., 2008; Permos, 2016; Scotti et al., 2016; Hartl et al., 2016). The kinematic monitoring methods mentioned above, however, cannot be used for monitoring of permafrost bodies without movement or surface deformation (e.g., sediments with medium to low ice contents, rock plateaus, gentle rock slopes). Remote sensing has so far not enabled thermal changes in permafrost to be assessed.

Geophysical methods can detect permafrost and characterize its subsurface ice and water contents (Kneisel et al., 2008; Hauck, 2013). They also provide structural information such as active layer and bedrock depths. In recent years, repeated geoelectrical surveys have been applied to determine ice and water content changes, thus complementing temperature monitoring in boreholes (Hilbich et al., 2008a; Pellet et al., 2016). Results from such electrical resistivity tomography (ERT) monitoring show that permafrost thaw in mountainous terrain is often accompanied by a drying of the subsurface, as the water from the melted permafrost often leaves the system downslope and is not always substituted in the following summer (Hilbich et al., 2008a; Isaksen et al., 2011). A 15-year ERT time series from Schilthorn, Swiss Alps, shows for example a clear decreasing trend of electrical resistivity, corresponding to ice melt, throughout the entire profile below the active layer (Fig. 4). The corresponding temperature at 10 m depth is at the freezing point and shows no clear trend. ERT is increasingly used in operational permafrost monitoring networks to determine long-term changes in permafrost ice content (Hilbich et al., 2008b, 2011; Supper et al., 2014; Doetsch et al., 2015; Pogliotti et al., 2015).

Physically based models of varying complexity are employed for process studies of permafrost (for a review see Riseborough et al., 2008; Etzelmüller, 2013) and specifically for the analysis of future permafrost evolution. These models should not be confused with permafrost distribution models (Boeckli et al., 2012; Gisnaas et al., 2016; Deluigi et al., 2017), which are statistical and often based on rock-glacier inventories and/or topo-climatic variables such as potential incoming solar radiation and mean annual air temperature. Physically based site-level models are used in combination with regional climate models (RCMs) for studies of long-term permafrost evolution (Farbrot et al., 2013; Scherler et al., 2013; Westermann et al., 2013; Marmy et al., 2016), similar to land-surface schemes used for hemispheric permafrost modeling (Ekici et al., 2015; Chadburn et al., 2015; Peng et al., 2016). Physically based models are also used to explain the existence of low-altitude permafrost occurrences (Wicky and Hauck, 2017) and to analyze the dominant processes for the future evolution of specific permafrost occurrences in the European mountains (Scherler et al., 2014; Fiddes et al., 2015; Zhou et al., 2015; Haberkorn et al., 2017; Lüthi et al., 2017). Simulations for different mountain ranges in Europe suggest an overall permafrost warming and a deepening of the active layer until the end of the century (see Fig. 5 for four examples from the Swiss Alps; similar simulations from Scandinavia are found in Hipp et al., 2012; Westermann et al., 2013, 2015; Farbrot et al., 2013).

The projected increase in permafrost temperatures is mainly due to the anticipated increase in air temperatures. The latter also causes the snow cover duration to decrease, thereby reducing the thermal insulation effect (Scherler et al., 2013; Marmy et al., 2016). In spite of similar trends in RCM-driven permafrost studies, comprehensive regional-scale maps or trends for projected permafrost changes in Europe are not available to date. This is partly because of the insufficient borehole data, but mainly because of the large heterogeneity of the permafrost in European mountain ranges. The latter strongly depends on surface and subsurface characteristics (e.g., fractured and unfractured rock, fine and coarse-grained sediments, porosity), microclimatic factors (snow cover, energy balance of the whole atmosphere–active layer system, convection in the active layer, etc.), and topo-climatic factors (elevation, aspect, slope angle).

The largest and most important impacts related to permafrost thawing have yet to occur. Along with changes in precipitation, permafrost thawing is projected to affect the frequency and magnitude of mass wasting processes in mountain environments (IPCC, 2012). This is especially true for processes driven by water, such as debris flows (Stoffel and Huggel, 2012; Borga et al., 2014). Based on statistically downscaled RCM data and an assessment of sediment availability, Stoffel et al. (2011, 2014a, b) concluded that the temporal frequency of debris flows is unlikely to change significantly by the mid-21st century, but is likely to decrease during the second part of the century, especially in summer. At the same time, the magnitude of events might increase due to larger sediment availability. This is particularly true in summer and autumn when the active layer of the permafrost bodies is largest, thus allowing for large volumes of sediment to be mobilized (Lugon and Stoffel, 2010). Accelerations of rock-glacier bodies might play an additional role (Stoffel and Huggel, 2012). Providing projections for future sediment availability and release for areas that are experiencing permafrost degradation and glacier retreat is particularly important in the European Alps, where the exposure of people and infrastructure to hazards related to mass movements is high (Haeberli, 2013).

Finally, it should be noted that in contrast to glacier melting, permafrost thawing is an extremely slow process (due to the slow downward propagation of a thermal signal to larger depths and additional latent heat effects). As permafrost in European mountains is often as thick as 100 m, a complete degradation is therefore unlikely within this century.

Shifts in agricultural production are expected with climate change as a consequence of higher water demand from crops and less water available due to higher temperatures and longer dry spells but also as a consequence of earlier snowmelt and less glacier contribution (Jaggard et al., 2010; Gornall et al., 2010). Most studies for Alpine regions project reduced soil water content as a result of increasing evaporation, thinner snowpack, and earlier snowmelt (Wu et al., 2015; Barnhart, 2016). This will lead to increased water demand for irrigation (Jasper et al., 2004; Schaldach et al., 2012; Riediger et al., 2014) and will add to the changes in water availability resulting from changing snow and glacier melt (Smith et al., 2014). The effects of more frequent climatic and hydrological droughts in the future (Gobiet et al., 2014) will affect both croplands and grasslands. In Switzerland, the latter cover around 75 % of the agricultural land and sustain domestic meat and dairy production (Fuhrer et al., 2006). The majority of crops currently cultivated in the Alps have been shown to be very sensitive to precipitation deficits in the growing season (Fuhrer et al., 2006; Smith et al., 2014). High irrigation demands will thus likely put additional pressure on rivers, especially small ones as they suffer more from interannual variability (Smith et al., 2012). Together with generally decreasing summer discharge, this will more frequently create low flow conditions which largely favor increasing water temperatures in streams with negative consequences for water quality and the aquatic fauna. Long-term water-management strategies will be important to face these challenges and to ensure that future agricultural water needs can be met (Riediger et al., 2014).

Climate change is a key driver in electricity markets, as both electricity production and demand are linked to weather and climate (Apadula et al., 2012). As a consequence of earlier snowmelt and reduced water discharge from glaciers, hydropower production potential is expected to increase in winter and spring and to decline in summer (Hauenstein, 2005; Kumar et al., 2011). Currently, energy demand is higher in winter than in summer, but this may change as rising summer temperatures increase energy requirements for the cooling of buildings (López-Moreno et al., 2008, 2011; Gaudard and Romerio, 2014). A study conducted for the Mattmark dam in the Swiss Alps and for the Val d'Aosta, Italy (Gaudard et al., 2014), revealed that peak hydropower production has so far not been affected by climate change. This is possibly the result of the large existing reservoir volumes which enable to offset seasonal changes (Farinotti et al., 2016). Indeed, it has been suggested that no urgent adaptation of the hydropower infrastructure will be required in Switzerland within the next 25 to 30 years (Haunstein, 2005). For Austria, little changes in annual hydropower production (± 5 %) have been projected up to 2050 using A1B SRES, but there could be a significan reduction in summertime power generation (Wagner et al., 2017). Reservoir management, however, will become more challenging as a consequence of higher fluctuations in electricity demands linked to the intermittent production of new renewable energy sources such as photovoltaic and wind power or biogas (Gaudard and Romerio, 2014). Furthermore, the interannual fluctuations in water availability are expected to increase (Gaudard et al., 2014). Run-of-river power plants are expected to be less vulnerable to climate change, as they are usually installed on streamflows with small hydrological fluctuations (Gaudard et al., 2014). Hydropower plants can also be effective in attenuating floods (Harrison and Whittington, 2001). Additional safety concerns include the melting of permafrost and the possibility of more frequent heavy rainfall, resulting in both more frequent slope instabilities and potential flood waves that may endanger power plants (Peizhen et al., 2001; Schwanghart et al., 2016). Increased sediment loads from deglacierized surfaces may additionally affect power generation, in particular by affecting the wear of infrastructure or the silting of storage volumes (Beniston, 2003).

Increasing air temperatures are expected to result in shorter skiing seasons and a shift of the snow line to higher elevations (Abegg et al., 2007; Steiger, 2010). This will likely lead to smaller number of visitors and reduced revenues, and thus have important economic impacts on alpine winter tourism. Generation of artificial snow is designed to buffer the impact of interannual variability of snow conditions and is increasingly deployed in alpine ski resorts (Uhlmann et al., 2009; Steiger, 2010; Gilaberte et al., 2014; Spandre et al., 2016). In Switzerland, ski slope areas employing artificial snow-making equipment have tripled (from 10 to 33 %) from 2000 to 2010 (Pütz et al., 2011). In the French Alps, 32 % of the ski slope area was equipped with snow-making facilities in 2014, and this proportion is likely to reach 43 % by 2020 (Spandre et al., 2015). In Austria, this share is about 60 %, mainly due to the lower average elevations of the Austrian ski areas, and in the Italian Alps almost 100 % of the ski areas are equipped (Rixen et al., 2011). Water consumption for tourism in some Swiss municipalities is high compared to other uses. A study focusing on three tourism destinations in Switzerland, for example, found this consumption to be equivalent to 36 % of the drinking water consumption (Rixen et al., 2011). Water and energy demands of ski resorts will increase, which may in turn lead to higher prices for consumers (Gilaberte et al., 2014). Also, summer mountain resorts could be affected by water shortages in the future, thus calling for adapted water management (Roson and Sartori, 2012).

The snow pack is affected by a series of small-scale processes, including water transport in snow and firn (Würzer et al., 2017; Wever et al., 2014, 2016), phase changes (i.e., melt, refreezing, sublimation and condensation), drifting and blowing snow, and metamorphism (e.g., Aoki et al., 2011; Pinzer et al., 2012). While the mechanistic understanding of these processes at the point scale is rapidly increasing (Wever et al., 2014), challenges comes from quantifying their effects at larger scales. Examples of the interplay between large- and small-scale effects include the altered snow distribution after a storm (Lehning et al., 2008; Schirmer et al., 2009) or the change in snow albedo after a melt event. The underlying small-scale processes (such as snow-grain saltation and metamorphism in the case of wind redistribution and albedo changes, respectively) are reasonably understood at the point scale, but they are insufficiently represented or even neglected in large-scale models due to the lack of adequate upscaling schemes. Small-scale snow properties are also essential for the correct interpretation of remote-sensing signals. Ice lenses or the liquid water content in snow, for example, heavily influence the microwave backscatter, which could be used to infer large-scale SWE (Marshall et al., 2007). We argue that the challenges in snow modeling are currently scale dependent.

For modeling snow at the point scale and with a high level of detail, for example in snow stability estimates in the context of avalanche warning, two physically based models are mostly used: CROCUS (Vionnet et al., 2012) and SNOWPACK (Lehning et al., 1999). Nevertheless, many of the processes described in these models, such as metamorphism and mechanical properties, still have a high degree of empirical parameterization (Lehning et al., 2002). The main challenge for snow model development at this scale is the formulation of a consistent theory for snow microstructure (Krol and Lowe, 2016) and its metamorphism.

For hydrological applications, the catchment scale is the most relevant one (Kumar et al., 2013). Snow models of different complexity are used for this (Essery, 2015; Magnusson et al., 2015), and principal challenges are to (a) distinguish between uncertainties introduced by the model structure and uncertainties related to the input data (Schlögl et al., 2016) and (b) develop transferable, site-independent model formulations without the need of calibration. The latter is particularly important for reliable predictions of climate change effects (Bavay et al., 2013) and for model applications to ungauged catchments (Parajka et al., 2013).

Large-scale models, finally, use relatively simple, parametric snow schemes, as these require only a small set of input variables and are computationally less expensive (Bokhorst et al., 2016). Current numerical weather prediction systems generally use oversimplified single-layer snow schemes (IFS documentation, 2016; GFS documentation, 2016). Only in some rare cases do these models explicitly represent the liquid water content within the snowpack or incorporate a refined formulation for snow albedo variability (Dutra et al., 2010; Sultana et al., 2014). Climate models generally resolve the diurnal and seasonal variations of surface snow processes (i.e., surface temperature, heat fluxes) while they simplify the treatment of internal snow processes such as liquid water retention, percolation, and refreezing (Armstrong and Brun, 2008; Steger et al., 2013) or the evolution of snow microstructure due to metamorphism. Future research should clarify the degree of complexity required in snow schemes when integrated in large-scale climate models (van den Hurk et al., 2016).

Challenges that need to be addressed in permafrost modeling include the development of (1) static large-scale permafrost distribution models, (2) high-resolution and site-specific permafrost evolution models, and (3) transient hemispheric permafrost models or land-surface schemes of RCMs and global climate models (GCMs). Current state-of-the-art permafrost distribution models (e.g., Gisnås et al., 2017) are forced not only by statistical and topo-climatic variables such as mean annual air temperature and potential incoming radiation but also by operationally gridded datasets of daily air temperature and snow cover. Statistical distributions of snow and other surface characteristics (soil type, roughness) allow for the representation of sub-grid variability of ground temperature (Gubler et al., 2011; Gisnås et al., 2014, 2016). However, the lack of spatial data on subsurface properties (thermal conductivity, porosity, ice content, etc.) prohibits a refined assessment of the permafrost distribution on catchment or local scales, at least for the discontinuous permafrost zone. Acquiring spatial data on subsurface properties as input and validation data is hereby one of the greatest current challenges in permafrost research (e.g., Hauck, 2013; Etzelmüller, 2013; Gubler et al., 2013).

Model intercomparison studies using uncalibrated model setup show that due to the abundance of permafrost-relevant processes in the atmosphere and at the snow surface and subsurface, a detailed simulation of permafrost processes on local scales is impossible without the availability of surface and subsurface data (Ekici et al., 2015). This results from the difficulty of correctly simulating phase changes and corresponding latent heat transfer in permafrost when the initial ground ice content is poorly known. Challenges in regional or hemispheric permafrost modeling therefore include numerical aspects and process-oriented model improvements, as well as data availability and up-scaling issues (Fiddes et al., 2015; Westermann et al., 2015). Most land-surface schemes of current GCMs and RCMs now include soil freezing schemes (e.g., McGuire et al., 2016). However, neither reliable ground ice content maps as input nor ground temperature maps for validation currently exist. Combined with the need for reliable snow, soil moisture, and vegetation data, this lack of subsurface information poses the largest uncertainties in estimates of current and future permafrost temperature and spatial distribution.

To increase the accuracy of glacier mass balance and runoff estimates, the distribution of ice volume and ice thickness is of primary importance. Glacier volume can be estimated from their surface area using scaling approaches (e.g., Bahr et al., 1997). However, such approaches do not provide the ice thickness distribution, which together with the surface mass balance controls the ice dynamics and the response of the glaciers to climate forcing. As it is currently impossible to measure ice thickness distributions of all glaciers individually, model applications are necessary. Several existing glacier models have been compared within the Ice Thickness Models Intercomparison eXperiment (ITMIX; Farinotti et al., 2017), revealing that (i) results largely depend on the quality of the input data (glacier outline, surface elevation, mass balance or velocities) and (ii) model complexity is not directly related to model performance. New high-resolution satellite images make the required input data available, thus opening the way toward improved future global estimates of glacier thicknesses.

Another challenge in glacier modeling is the use of approaches that explicitly consider ice dynamics. Jouvet et al. (2009) showed that full 3-D Stokes models representing ice flow without approximation can be applied if the required input data are available (e.g., glacier thickness, surface velocities). At the regional scale, however, simplified approaches are still required (Clarke et al., 2015). For estimating future glacier evolution, ice dynamics models need to be coupled to adequate representations of glacier surface mass balance. The Glacier Model Intercomparison Project (Glacier-MIP, www.climate-cryosphere.org/activities/targeted/glaciermip) assesses the performance of regional- to global-scale glacier models to foster the improvement of the individual approaches and to reduce uncertainties in future projections. There are uncertainties in future meteorological variables and in their downscaling at the glacier scale. Some studies use degree-day approaches for long-term simulations (Réveillet et al., 2017) or account for potential radiation (Hock, 1999). However, with a change in the relative magnitude of energy fluxes at the glacier surface, calibrated degree-day factors are likely to change too. Application of models able to resolve the full energy balance are thus required (Hanzer et al., 2016). To be applicable, however, the accuracy and resolution of the corresponding input data need to be improved. More emphasis on the modeling of winter mass balances and the spatial distribution of snow accumulation (Réveillet et al., 2017), as well as the impact of supraglacial debris and related feedbacks (Reid and Brock, 2010), is required. The latter is particularly important as many glacier tongues become increasingly debris covered as they shrink. Feedback effects of black carbon and aerosols deposition on the glacier surface is also subject of study (Gabbi et al., 2015). Finally, more research on glacial sediment transport and erosion is needed as glacier retreat exposes large amounts of unconsolidated and erodible sediments that might – when entrained – represent a hazard downstream or reduce the efficiency of hydropower plants (Lane et al., 2016).

Predicting the future evolution of cryospheric components is challenging in many respects. On the one hand, these challenges stem from the incomplete understanding of the physical processes leading to given changes and from the lack of more comprehensive and consistent datasets (see Fig. 7). On the other hand, predictions are intrinsically affected by a range of uncertainties. Adequately evaluating and communicating such uncertainties is anything but a trivial task, both because the interplay between individual systems can be complex and because end users of projections are typically adverse to uncertainty. Outside the scientific community, “uncertainty” and “error” are two concepts often not sufficiently distinguished. This can lead to important misunderstanding and misinterpretations. Improving the way uncertainties are communicated is especially important when presenting scientific results to policymakers or stakeholders, as this can significantly affect the level of trust assigned to a particular finding.