Lasting effects of snow accumulation on summer performance of large herbivores in alpine ecosystems may not last



  1. One of the clearest predictions from the IPCC is that we can expect much less snow cover due to global warming in the 21st century, especially in the lower alpine areas. In alpine ecosystems, snow accumulation in depressions gives rise to distinct snow-bed vegetation types, assumed to play a key role in ecosystem function. A delayed plant phenology yields high-quality forage in late summer for wild and domestic herbivores. Yet, the mechanistic pathways for how declining snow may affect future performance of large herbivores beyond the effect of phenology remain poorly documented.
  2. Here, we link unique individual-based data on diet choice, habitat selection and performance of domestic sheep over a 10-year period to manually GPS-recorded spatial positions of snow cover in early summer (0·57% to 43·3% in snow beds on 1st of July) in an alpine ecosystem.
  3. Snowy winters gave a higher proportion of easily digestible herbs in the diet and a more variable use of snow-bed and meadow vegetation types resulting in faster growing lambs. These patterns were consistent between two density treatment levels although slightly more marked for diet at low density, suggesting that effects of simple mitigation efforts such as managing population numbers will be meagre.
  4. Our study thus yields novel insight into the strong impact of melting snow on ecosystem function in alpine habitats, which are likely to affect productivity of both domestic and wild ungulate populations.


Arctic and alpine ecosystems are regarded as particularly high risk to climate change (Schroter et al. 2005; Kausrud et al. 2008; Post et al. 2009). One of the clearest predictions from the Intergovernmental Panel on Climate Change (IPCC) is that we can expect much less snow cover due to global warming in the 21st century (Christensen et al. 2007). For example, an increase of 4 °C in mean temperature is expected to reduce the duration of the period with snow cover by 50% at 2000 m a.s.l. and by 95% at 1000 m a.s.l. in the Alps (Christensen et al. 2007). Ecosystem services such as the provisioning of water are therefore regarded at particularly high risk in mountains (Schroter et al. 2005). In general, alpine ecosystems are naturally grazed (Körner 1995). Lower elevation alpine zones are core areas in rural economies due to their provisioning of grazing areas for livestock and wild large herbivores (Körner & Ohsawa 2005).

A key theory in large herbivore foraging ecology is that forage maturation plays a crucial role for distribution and in turn performance (Fryxell & Sinclair 1988; Hebblewhite, Merrill & McDermid 2008; Bischof et al. 2012). Plant onset of growth and further development is strongly affected by local temperature and snow conditions (Mårell, Hofgaard & Danell 2006). Newly emergent plants are in turn of higher quality to herbivores than older plants that are allowed to flower and die (Mysterud et al. 2011). The general description of the forage maturation hypothesis is that herbivores utilize such spatial variation in the onset of plant growth to enhance the duration of access to newly emergent, high-quality plants (Hebblewhite, Merrill & McDermid 2008). This, in turn, is expected to have marked consequences for ruminant growth, both due to higher quality forage, but also as higher quality forage means shorter time for rumination and thus more time to forage, the so-called multiplier effect (White 1983; Cebrian, Kielland & Finstad 2008). Several studies have indicated that snow-rich winters may positively affect summer body growth of both wild (Mysterud et al. 2001b, 2008) and domestic (Nielsen et al. 2012) ungulates and that part of the mechanism is linked to forage maturation. Typically, studies aiming to provide mechanistic evidence for this chain have focused on the effect of snow deposition on quality of single plant species (Walsh et al. 1997), or overall quality (Albon & Langvatn 1992) or timing of onset of growth (Post & Forchhammer 2008). Yet, there has been little focus on determining whether also a shift in diet composition may play a role related to the forage maturation hypothesis. Different plant species flower at different times of the year (Körner 2003). Therefore, both spatial variations in single species onset of growth and/or different species may be the mechanism behind the general forage maturation hypothesis.

Snow cover from winter is thus expected to have a lasting effect on the ecosystem dynamics for the duration of the entire summer growth season, as the snow accumulated during winter melts slowly in depressions yielding a prolonged period of spring flush. In alpine ecosystems, snow-bed vegetation forms a distinct vegetation type with characteristic species composition (Galen & Stanton 1995; Bjork & Molau 2007; Schob et al. 2009; Hulber, Bardy & Dulinger 2011) and is assumed to play a key role for herbivore foraging (Moen, Lundberg & Oksanen 1993; Virtanen, Henttonen & Laine 1997; Virtanen 2000). At a detailed mechanistic level, there is no study linking explicitly annual variation in large herbivore diet, habitat selection and individual performance to spatially explicit information of snow cover. Factors affecting early growth are important for performance in ungulates, as an early start of growth in life often have lasting effects on performance (reviews in Lindström 1999; Gaillard et al. 2003).

We here present a unique data set with observations of a large herbivore, domestic sheep (Ovis aries), in an alpine ecosystem over a 10-year period (2002–2011) kept at high and low density in a fully replicated landscape level experimental facility. We have previously documented strong density dependence and annual variation in habitat selection (Mobæk et al. 2009), diet (Mobæk et al. 2012a) and body growth of lambs (Mobæk et al. 2013), and we here provide novel aspects by including explicit links to weather variables and quantify how habitat selection, diet and body growth are linked. In this ecosystem, grass snow beds account for 12·3% of the total area. The average snow cover in snow beds the first week of July was 14·3%, but varying from 0·57% (2006) to 43·3% (2005) among years. We here assess how this variation affected (i) habitat selection, (ii) diet composition and, in turn, (iii) performance of this large herbivore. For diet composition, we contrast proportional intake of multiple species of herbaceous plants (hereafter referred to as herbs) being of high quality and low availability and Avenella flexuosa being lower quality and high availability (cf. Mysterud et al. 2011). Our study highlights that changes in diet composition towards more palatable plant groups after snowy winters provide an overlooked mechanism of increased performance, in addition to the often reported changes in within or overall species quality due to delayed phenological development.

Materials and methods

Study area and experimental design

The study was conducted within a fenced experimental enclosure covering 2·7 km2 in Hol municipality, Buskerud county in southern Norway (60°40′ N, 7°55′ E). The area has a subcontinental alpine climate with annual precipitation c. 1000 mm (Evju et al. 2009). The enclosure is spanning an altitude range from 1050 m to 1320 m a.s.l., and the vegetation is mainly composed of low shrubs with scattered grass-dominated meadows and snow beds. The area is above the forest, although scattered birch (Betula pubescens) trees are found mainly in the lowest part of the altitudinal range (Speed et al. 2010). The enclosure was established in 2001 and split into nine subenclosures (Austrheim et al. 2008). The vegetation types and its assumed value to sheep were mapped in 2001 prior to the study (Rekdal 2001). Each of the subenclosures covered the same main altitudinal range and a comparable distribution of vegetation types. The treatments high density, low density or control (no sheep) were replicated three times randomized within three blocks. Sheep density levels were 80 and 25 sheep per km2 for high and low density, respectively. For each year 2002–2011, a total of 24–27 lactating ewes and 44–46 lambs were released into the experimental enclosure each summer (see Table S1, Supporting Information). Each grazing season lasted from late June to late August or early September. All the sheep belonged to the same sheep farmer and were of the most common breed in Norway (‘Norsk Kvit Sau’). All ewes had lambs (1–3) and were lactating at time of release. All sheep were weighed to the nearest 0·5 kg at the day of release and gathering.

Sheep data

The ewes and their lambs were individually marked and numbered with colour-coded tags. We used each ewe as the unit of within subenclosure replication, as this sheep breed does not form stable large herds. A total of 6929 direct observations (i.e. focal watches) of ewes from the entire grazing seasons (2003–2011) were used in the analysis of habitat selection. Number of position per year averaged 770 ranging from 687 to 956. Observations were made using two working cycles distributed randomly throughout the season; for a given day, either ‘early’ from 9.00 to 17.00 or ‘late’ from 14.00 to 22.00 (Mobæk et al. 2009, 2012b). Typically, only one (58·6%) or two (35·3%) positions were retrieved per ewe and day to avoid serial autocorrelation. The position (with GPS) and behaviour (active, inactive) of each ewe were recorded using binoculars from a distance of 20–50 m.

Samples of faeces were collected and put in plastic bags and frozen for microhistological analyses of diet for all years 2002–2011. This was carried out as part of the direct observations, and only samples from known individuals were collected. We aimed to ensure equal representation of densities (high vs. low), age classes (ewe vs. lamb) and three time periods (early, middle and late grazing season). We obtained 861 samples from 412 individual sheep. Microhistological analyses (Stewart & Stewart 1970; Cortés et al. 2003; Takatsuki 2003) were performed following a standard procedure boiling 1 mL of faeces in 4 mm of nitric acid (Kausrud et al. 2006; Mobæk et al. 2012a). For 2002–2006, each faeces sample was split into two parallel subsamples that were processed independently (for 343 samples). Due to high consistency among samples, this was not carried out for years 2007–2011. The mean number of faeces samples analysed per individual sheep was 2·09 (±1·89 SD) and 2·92 (±2·49 SD) if including the two parallel subsamples. Plant fragments were identified to species whenever possible, and otherwise genus or family names were determined.

Spatially explicit snow data and weather data

Snow cover was recorded each year with a handheld GPS and walking around all snow drifts taking waypoints that later was formed to polygons in ArcGIS. This was usually carried out on the 1st of July, and no later than 5th of July. We also retrieved meteorological data for our study area. These were provided by the Norwegian meteorological institute and are model outputs (interpolations) for the middle of our study area (UTM 33: East 115355.8 North 6749590.7) at 1150 m a.s.l. We retrieved mean monthly temperature and precipitation for June, July and August. We also summed these variables to get summer temperature and precipitation. For predicting climate impact on sheep in similar alpine areas, mean temperature has been found to be better than for example growing degree days (Nielsen et al. 2012).

Vegetation map

The distribution of vegetation types in the experimental area was mapped in 2001 (Rekdal 2001). Analysis of selection requires that all habitat types are available. A categorization into broader classes avoids issues that arise when a particular (rare) vegetation type does not occur in all subenclosures, or when some vegetation types were recorded as used, but not available, because vegetation mapping was conducted on a coarser scale (raster, resolution of 2 ha; Rekdal 2001) than activity observations (points). We therefore considered pooled vegetation classes. In addition to (i) grass snow beds, forming the main topic here, other vegetation types were grouped according to their productivity based on previous work (Mobæk et al. 2009); (ii) high productivity (tall herb meadow, low herb meadow); (iii) medium productivity (dwarf shrub heath); and (iv) low productivity (moss snow bed, lichen heath, bog, fen, stone polygon land) vegetation types.

Statistical analyses

The analyses consisted of three parts: analysis of (i) proportion of herbs and Avenella flexuosa being the main dietary items, (ii) habitat selection and (iii) body growth of lambs (performance) using linear mixed models with the ‘lmer’ function in library ‘lme4’ in r. In all models, random terms modelled as random intercepts were individual ID (motherID for lamb body mass analysis) and subenclosure. For all data sets, we used model selection with Akaike information criterion (AIC) to find the most parsimonious model (Burnham & Anderson 2002), but we started with models previously found to fit our data (Mobæk et al. 2009, 2012a, 2013), and then adding the terms of interest (Tables S2–S5). Some of the weather variables were weakly correlated, but not the variables entering the final models (Table S6).


Dietary data were proportions and therefore arcsinsqrt transformed prior to analysis. We focused on proportion of herbs, being of assumed high quality, and proportion of the grass A. flexuosa, being of lower quality and forming the most important staple food (cf. evidence presented in Mysterud et al. 2011). Based on previous modelling, the starting point was a model including density (high/low), Julian date and age class (ewe/lamb) (Mobæk et al. 2012a). We then added the snow cover on 1st of July and tested the weather variables as single terms before adding the best weather variables together and their potential interaction with density. For these data, the individual ID as a random term accounts for both repeated sampling of the same individual within a sample (the two parallels), within a season or between years. However, results are consistent if only using one parallel (results not presented). For the dietary data, we also performed an ordination analysis to quantify an overall dietary shift in relation to climate variables. Due to low standard deviation of the axes, we used redundancy analysis (RDA) in which the ordination axes are constrained linear combinations of the climatic variables. Each variable was tested for significance using a Monte Carlo permutation test with 999 unrestricted permutations. Ordinations and permutation tests were performed in the ‘Vegan’ library in r (Oksanen et al. 2010).

Habitat selection

We used generalized linear mixed-effects models (GLMM) for multivariate analyses of habitat selection (Gillies et al. 2006; Godvik et al. 2009). This approach compares use relative to habitat availability. For each individual ewe, we regarded its subenclosure to represent the available summer home range. Availability of each habitat in each subenclosure was calculated by distributing 20 times the number of random locations as there were used points and extracting the vegetation type in each point from a raster map (10 m2). Both used and random points falling on bare rock and on pixels that had not been classified (one pond in one subenclosure only) were excluded before analyses. Importing shape files (function readOGR in library ‘rgdal’), selection of random points (function csr in library ‘splancs’) and extraction of raster values (function join.asc in library ‘adehabitat’) were carried out in r (Mobæk et al. 2009). The response variable consists of use (given a value of 1) and availability (0) of a given habitat type and is thus expected to be binomially distributed. We only considered direct observations when sheep was actively foraging, thus excluding resting and rumination periods. The predictor variables were habitat type (4 classes), density and Julian date. ‘Individual ID’ and ‘subenclosure’ were included as nested random intercepts. To allow for model selection of fixed effects, models were fitted using maximum likelihood (ML) estimation (Pinheiro & Bates 2000) using the function lmer in the R library ‘lme4’. The predictions of this GLMM are a ‘population level’ estimate of the logarithm of the odds for using a map pixel with a certain combination of predictor variable values (Godvik et al. 2009).

Herbivore performance

The response variable was autumn body mass of lambs (ln-transformed). Based on previous modelling (Mobæk et al. 2013), we included covariates sex, (ln) spring body mass, (ln) ewe body mass, litter size (categorical 1–3), density (high/low), number of grazing days and year (fitted with spline using library ‘splines’) as a trend interacting with density (Mobæk et al. 2013). We used mother ID as a random term to account for dependence between twins or triplets, and/or if a ewe was included in several years.


The number of snow drifts within the study area varied from 25 to 91 on 1st of July, and the coverage of snow in snow beds ranged from 0·57% (2006) to 43·3% (2005). Over the decade, years 2005, 2007 and 2008 had much more snow than the other years (Fig. 1). The diet composition switched to a higher intake of assumed high-quality herbs (Fig. 2a) and lower intake of the bulk grass A. flexuosa (Fig. 2b) in years with a high coverage of snow on 1st of July. Selection for grass snow-bed and meadow vegetation types was negative in the early season and positive in the late grazing season in years with much snow (Fig. 2c). This resulted in higher performance in years with more snow cover (Fig. 2d).

Figure 1.

The spatial distribution of snow on 1st of July over a 10-year period (2002–2011) in an alpine ecosystem in Hol, Norway. Grass snow beds account for 12·3% of the total area.

Figure 2.

The effect of early snow accumulation on dietary proportion of (a) high-quality herbs, (b) the bulk forage Avenella flexuosa, (c) selection of snow-bed habitat in early and late summer and (d) performance of domestic sheep over a decade in an alpine ecosystem of Norway. (a–b) Predictions are made for ewes at low density on 1. July with July temperature of 10 °C, while the points for each year are not corrected for July temperature. Years are 2002–2011 (denoted 02-11).

In addition to snow cover, temperature in July had a positive effect on the proportion of herbs in the diet, while other weather variables did not enter the best model (supporting information). Lambs had a higher proportion of herbs in the diet than ewes, a higher proportion of herbs were eaten at low compared with high density and use of herbs declined over the season (Table 1). For A. flexuosa, there was no density effect, but use increased over the grazing season, and was lower in lambs than in ewes. Overall, plant taxa composition in the diet was affected by both temperature in July (F = 51·991, = 0·01) and the coverage of snow in snow beds (F = 30·377, = 0·01), in addition to population density (= 69·084, = 0·01), age class (= 96·278, = 0·01) and Julian date (= 191·417, = 0·01). More herbs and less A. flexuosa were linked to warmer July, more snow cover, lambs, early date and low population density.

Table 1. Estimates of the effect of early summer snow cover on (a) proportion of herbs, (b) Avenella flexuosa in the diet and (c) body mass of lambs over a decade (2002–2011) in an alpine ecosystem, Norway
ParametersEstimateSELower limitUpper limit
(a) Herbs
Density (low vs. high)0·0950·0350·0250·166
Julian date−0·0030·000−0·003−0·002
Agecat (lamb vs. ewe)0·0670·0100·0470·087
Snow cover0·1060·0370·0320·179
Density × Snow cover0·1700·0590·0510·289
(b) A. flexuosa
Density (low vs. high)−1·2882·954−7·1954·619
Julian date0·3970·0160·3650·429
Agecat (lamb vs. ewe)−2·1450·264−2·673−1·617
Snow cover−10·8353·558−17·952−3·719
Density × Snow cover−7·0435·631−18·3064·220
(c) Body mass
Density (high vs. low)0·0470·033−0·0190·113
Year (spline, 1)−0·0100·076−0·1610·142
Year (spline, 2)0·1360·0550·0260·247
Year (spline, 3)0·0550·036−0·0170·128
Sex (m vs. f)0·0500·0080·0340·066
Littersize (triplet)−0·0650·014−0·093−0·036
Littersize (twin)−0·0380·012−0·063−0·013
Ln (spring mass)0·6180·0190·5800·656
Ln (ewe mass)0·0900·0340·0220·158
Snow cover0·0940·0350·0250·164
Density: year, 1−0·2210·086−0·393−0·048
Density: year, 2−0·1590·061−0·281−0·036
Density: year, 3−0·1630·041−0·245−0·080

The analysis of body mass confirmed a positive effect of snow cover on growth of lambs, after controlling for that the body mass declined at high density over the first 2 years before stabilizing at a lower level, while body growth increased slightly at low density over time. We controlled for the interaction between density and year as a trend (cf. Mobæk et al. 2013). Males were heavier than females, singletons were heavier than twin and triplet lambs, spring body mass of both lamb and ewe positively affected autumn mass, and mass increased with the number of grazing days (Table 1). Habitat selection was density-dependent, with sheep at low density selecting more productive habitat types, and use being more even at high density (Table 2).

Table 2. Estimates of the effect of early summer snow cover in interaction with date affect habitat selection of sheep in an alpine ecosystem, Norway. Baseline for habitat is ‘high-productive meadows’
Snow bed−0·3350·066−0·466−0·203
st. snow−0·0330·027−0·0880·021
Density (low vs high)0·3860·1210·1440·628
st. julian date−0·1100·027−0·164−0·055
st. snowcover.julian date0·1310·0270·0760·186
Snow bed :density−0·4090·094−0·597−0·221
Heaths :density−0·6560·076−0·808−0·504
Impediment :density−0·472232·663−465·799464·854
Snow bed :st. julian0·1620·0450·0720·251
Heaths :st. julian0·1670·0360·0950·238
Impediment :st. julian0·109114·329−228·550228·768
Snow bed :st. snowcover0·0370·045−0·0530·127
Heaths :st. snowcover0·0840·0360·0130·156
Impediment :st. snowcover0·029114·357−228·685228·743
Snow bed :st. snow.julian−0·0530·045−0·1430·038
Heaths :st. snow.julian−0·2390·036−0·312−0·167
Impediment :st. snow.julian−0·137115·635−231·407231·132


The melting of snow and ice in arctic and alpine regions following global warming receives considerable attention. Sea ice providing critical habitat for polar bears (Ursus maritimius) has become the symbolic case for a warming globe (Bonn 2003; Stirling & Derocher 2012). The effect of earlier snow melting in alpine areas is less spectacular, but might be very important for the provisioning of ecosystem services from alpine ecosystems. Earlier studies have highlighted how earlier melting advance timing of reproduction in plants (Cleland et al. 2007; Smith et al. 2012; Duparc et al. 2013) and the direct positive effect of reduced winter snow on reindeer survival (Moen 2008). It is also well-established that annual variation in snow depth may have delayed positive effect on growth of large herbivores (Mysterud et al. 2001b; Nielsen et al. 2012). The suggested mechanism has been on how late snow melt delay plant phenology improving late season plant quality (Mårell, Hofgaard & Danell 2006), and in turn spatial distribution of large herbivores (Fryxell & Sinclair 1988; Hebblewhite, Merrill & McDermid 2008; Bischof et al. 2012). Clearly, the mechanism of late season plant phenology is important for forage quality. We here highlight how the diet composition changed depending on snow deposition in early summer (1st of July), an additional mechanism not previously reported, and how this affected ungulate performance.

Many studies report a link between the effect of snow deposition on quality of single plant species (Walsh et al. 1997), or overall quality (Albon & Langvatn 1992) or timing of onset of growth (Post & Forchhammer 2008). In Alaska, C:N ratios for specific species such as dwarf birch Betula nana and the tussock cottongrass Eriophorum vaginatum were lower on late melting plots that were created by increasing snow deposition (Walsh et al. 1997). Similarly, in another study in Alaska, manipulations of snowmelt affected the timing and the chemistry of E. vaginatum. Other approaches have looked at quality more in general (Albon & Langvatn 1992; Mårell, Hofgaard & Danell 2006; Hebblewhite, Merrill & McDermid 2008). Using indirect measures, such as timing of the onset of growing season for the pooled community of plants, has also been carried out either directly (Post & Forchhammer 2008) or using indirect measures such as the derived-NDVI time-series values (Bischof et al. 2012). It is well-established that different plant species flower at different times of the year, and this in turn may be differently affected by annual variation in snow deposition. We argue that the literature of large herbivore foraging has put too little emphasize on that a shift in diet composition may play a role, forming an additional component of the forage maturation hypothesis. For example, some species such as roe deer (Capreolus capreolus) seem to follow the phenology of more than 100 plants during summer even within home ranges in flat environments (Selås et al. 1991). Therefore, both spatial variations in single species onset of growth and/or different species may be the mechanism behind the general forage maturation hypothesis. It has been argued that spatial variation in plant phenology is more important for grazers than for browsers (Mysterud et al. 2012), and our study also suggests it might be interesting to look further into whether stationary animals might utilize different species phenology within a given area.

Snow deposition in mountains is well known to positively affect summer growth for migratory wild ungulates such as red deer (Mysterud et al. 2001a,2001b; Pettorelli et al. 2005; Mysterud et al. 2008). In a comprehensive analysis of growth of domestic sheep lambs in alpine ranges in south of Norway, the only consistent effect across alpine ranges was the positive effect of annual variation in snow depth on growth (Nielsen et al. 2012), while effects of summer temperature and precipitation differed largely in effect between ranges. We here provide the first detailed account of the role of habitat selection and diet composition adding to our understanding of the benefits of accumulated snow for the duration of the entire summer growth season. During snow-rich springs, sheep avoided the snow beds early in the growing season when high-quality fodder is available elsewhere, as indicated by a low proportion of staple fodder plants such as A. flexuosa. Grazing in snow beds increased over the summer as snow melted away and new vegetation growth started in snow beds. Snow-bed vegetation forms a distinct group (Austrheim et al. 2008; Baptist et al. 2010; Hulber, Bardy & Dulinger 2011), which may explain why not only phenological development, but also species composition of diet changes in snowy years. Consistent with earlier work in alpine ecosystems, we focused our attention mainly to grass snow-bed vegetation that is assumed to play a key role for herbivore foraging (Moen, Lundberg & Oksanen 1993; Virtanen, Henttonen & Laine 1997; Virtanen 2000). However, the habitat selection pattern of less use early and more use late in the season in years with much snow was found also for meadows. Meadows are in general the most productive vegetation types in mountains (Mobæk et al. 2009), and we cannot therefore separate the relative contribution of delayed phenology of grass snow-bed and meadow vegetation types. Also, warm July seems linked to higher production of herbs and more favourable foraging conditions.

Even the current changes of snow deposition may have wide-ranging cascading ecosystem effects (Ims & Fuglei 2005). Less snow will change vegetation composition of snow beds (Galen & Stanton 1995; Bjork & Molau 2007; Schob et al. 2009; Hulber, Bardy & Dulinger 2011), but also on surrounding vegetation being reliant on water from these (Bjork & Molau 2007). Reduced snow would increase chances of drought, which is known to have contrasting effects on grasses and herbs and in turn altering carbon fluxes in alpine snow-bed ecosystems (Johnson et al. 2011). Snow beds are regarded as particularly important grazing areas for lemmings (Lemmus lemmus) and reindeer (Rangifer tarandus) (Virtanen 2000). Rain-on-snow events ruining the subnivean space for lemmings were found to cause a lack of population cycles, which in turn reduced numbers of grouse and migratory birds by effects of shared predators (Kausrud et al. 2008). The key role of snow for population cycles has firmly been linked spatially to altitudinal variation in snow structure and deposition (Ims, Yoccoz & Killengreen 2011). These changes in the vertebrate fauna may slowly transform vegetation towards more shrub dominance, as the regular ‘mowing’ of the tundra by lemmings are important to remove shrubs (Moen, Lundberg & Oksanen 1993). Sheep also play a vital role in keeping tree line below the climatic limitation (Speed et al. 2010, 2011). Wolverines are predators on domestic sheep in Norway (Landa et al. 1999), but whether there will be trophic cascading effects of reduced snow cover, and growth of sheep lambs is less clear. The number of sheep is mainly influenced by management practices (Skonhoft, Austrheim & Mysterud 2010). Wolverines denning are, however, also affected by snow pack development related more to direct effect of snow (Copeland et al. 2010; McKelvey et al. 2011). One should also be aware that wild herbivores may potentially be able to utilize spatial heterogeneity at much large scales than domestic herbivores, but wild herbivores may to a stronger degree be limited by other conflicting processes such as antipredator behaviour.

In high-elevation alpine areas, more snow is the initial response to warming, as there is also increased precipitation and sufficiently cold for most of the precipitation to come as snow during winter, like in the Norwegian mountains (Mysterud et al. 2000; Kausrud et al. 2008) and in the Colorado Rocky Mountains (Inouye et al. 2000). Therefore, initial positive effects on snow accumulation in alpine ecosystems may not last when tipping points are reached so that further warming also melts the snow at higher altitudes. If so, lasting effects of winter snow on summer foraging conditions may not last. This thus adds to the list of ecosystem services in alpine habitats around the globe likely to be strongly impacted of melting snow.


The study was funded by the Research Council of Norway, Miljø 2015 programme (Project 183268/S30) and the Directorate for nature management. We are grateful to Kyrre Kausrud, Camilla Iversen, Kristina Ehrlinger, Lars Korslund, Steve Parfitt, Harald Askilsrud, Kim Magnus Bærum, Lars Qviller, Malo Jaffre, Anna Blix, Randy Lange and Ragnhild Mobæk for help with field work and to Barbro Dahlberg for microhistological analysis of faeces. We are grateful to two anonymous referees for very useful comments to previous drafts and to Leif Egil Loe for very helpful and substantial statistical advice related to habitat selection.