Can a snow structure model estimate snow characteristics relevant to reindeer husbandry ?

Snow a"ects foraging conditions of reindeer e.g. by increasing the energy expenditures for moving and digging work or, in contrast, by making access of arboreal lichen easier. Still the studies concentrating on the role of the snow pack structure on reindeer population dynamics and reindeer management are few. We aim to !nd out which of the snow characteristics are relevant for reindeer in the northern boreal zone according to the experiences of reindeer herders and is this relevance seen also in reproduction rate of reindeer in this area. We also aim to validate the ability of the snow model SNOWPACK to reliably estimate the relevant snow structure characteristics. We combined meteorological observations, snow structure simulations by the model SNOWPACK and annual reports by reindeer herders during winters 1972-2010 in the Muonio reindeer herding district, northern Finland. Deep snow cover and late snow melt were the most common unfavorable conditions reported. Problematic conditions related to snow structure were icy snow and ground ice or unfrozen ground below the snow, leading to mold growth on ground vegetation. Calf production percentage was negatively correlated to the measured annual snow depth and length of the snow cover time and to the simulated snow density. Winters with icy snow could be distinguished in three out of four reported cases by SNOWPACK simulations and we could detect reliably winters with conditions favorable for mold growth. Both snow amount and also quality a"ects the reindeer herding and reindeer reproduction rate in northern Finland. Model SNOWPACK can relatively reliably estimate the relevant structural properties of snow. Use of snow structure models could give valuable information about grazing conditions, especially when estimating the possible e"ects of warming winters on reindeer populations and reindeer husbandry. Similar e"ects will be experienced also by other arctic and boreal species.


Introduction
Semi-domesticated reindeer (Rangifer tarandus tarandus) in northern Finland live in an environment where continuously changing weather and foraging conditions signi cantly a ect winter forage (Holleman et al., 1979). is in turn is a ected both by the amount of the main winter forage, (reindeer lichens Cladina spp.), and also by the snow conditions on pastures (Skogland, 1978;Helle & Tarvainen, 1984;Kumpula, 2001).Both reindeer and its northern American relative, caribou (Rangifer tarandus), are morphologically and behaviorally adapted to Arctic ecosystems (Telfer & Kensall, 1984).Reindeer herders acknowledge the e ects of weather and snow conditions on wellbeing of their herds, and husbandry has always been relatively adaptable to what comes to intra-and inter-annual variations in grazing conditions (Tyler et al., 2007;Roturier & Roue, 2009;Riseth et al., 2010;Vuojala-Magga et al., 2011).Despite this, the deep snow cover and late snow melt in spring can cause high winter mortality (Adamczewski et al., 1988;Kumpula & Colpaert, 2003;Helle & Kojola, 2008) and low calf production (Adams & Dale 1998;Post & Stenseth, 1999;Aanes et al., 2000;Kumpula, 2001) of both caribou and reindeer.
In addition to amount of snow, the structural properties of snow are also important.
e energy required for digging e ort is greater with increasing snow density and hardness (Fancy & White, 1985;Kumpula et al., 2004).Extensive ground ice (due to thawing-freezing at the snow-ground interface) has been observed to decrease the reproduction rates of Svalbard reindeer (Rangifer tarandus platyrhynchus) population (Hansen et al., 2011) or even cause population crashes (Helle, 1980;Kohler & Aanes, 2004).In addition, the number of warm days (mean T > 0 °C) during early winter or the winter time rain events, which is assumed to lead to dense or icy snow cover have been shown to decrease the calf production and winter survival of reindeer (Lee et al., 2000;Solberg et al., 2001;Kumpula & Colpaert, 2003;Helle & Kojola, 2008).Damages to reindeer by predation are partly connected to snow conditions.
Majority of previous research has been based on measurements on snow depth and meteorological observations that have daily or rougher time scales.It is di cult to identify winters with icy snow cover using this kind of observations only (Helle & Kojola, 2008;Vikhamar-Schuler et al., 2013).In Vikhamar-Schuler et al. (2013), a snow structure model SNOWPACK was successfully used to simulate the evolution of the snow cover, especially high-density layers, during years 1956-2010 in Kautokeino (Guovdageaidnu), Northern Norway.
SNOWPACK (Bartelt & Lehning, 2002;Lehning et al., 2002a and2002b) is a widely used model for describing the development of snow mass and energy balance during the winter.It is one of the few existing snow structure models and enables to estimate the layered structure within the snow cover and physical properties (e.g.density, hardness, grain size, grain type and bonding between the grains) of the layers.In this work we used combination of detailed meteorological information, snow structure simulations by the model SNOW-PACK and the annually made reindeer herders' reports to create a comprehensive view on snow conditions in a selected reindeer herding district in Muonio, northern Finland.
Due to an intensive management system relatively reliable estimates on annual mortality and productivity of Scandinavian reindeer population are available.Also winter conditions, including di cult snow condition, are annually reported by reindeer herders.Unfavourable snow and weather conditions a ect in a similar way to other northern ungulates, and more broadly, to several arctic and boreal species.e global mean temperature is predicted to increase by 1.4 -6.4 °C by the end of the year 2100 (IPCC, 2007). is warming will most likely be most extreme during winters in north-eastern Europe, and precipitation (consisting of rain on snow during warm winters) is expected to increase.ese changes will alter the amount and structure of snow cover, as well as in the length of the snow season, in many locations (Venäläinen et al., 2001;Räisänen, et al., 2003;ACIA, 2004;Rasmus et al., 2004;Kellomäki et al., 2010).Used together with climate model output data, SNOWPACK can work as a tool in climate impact studies.erefore, it is important to validate this modelling tool in present day conditions and to examine its development needs.
We aim to answer the following questions: evant for reindeer herding in northern boreal zone according to the experiences of reindeer herders?rate of reindeer in this area?to reliably estimate the relevant snow structure characteristics within the study area?and foraging conditions by reindeer compared to the conventional meteorological observations?

Study area
e Muonio reindeer herding district (2670 km 2 ) is located in the northern boreal zone representing typical herding districts in middle parts of Finnish Lapland (Fig. 1).Snow conditions are rather homogenous through the district.Reindeer are mainly grazed on the natural pastures in Muonio, even though supplementary winter feeding has gradually increased.According to the reindeer pasture inventory conducted during 2005-2008, 27.5% of the land area is covered by ground lichen pastures, 38.7% by mature and old coniferous forests with arboreal lichen, 20.1% by dwarf shrub and graminoid vegetation and 27.5% by mires (Kumpula et al., 2009).Only small fraction of the land area is high elevation (>300 m.a.s.l), tundra vegetation.Ground lichen pastures in the Muonio herding district are mostly heavily grazed (lichen biomass < 300 kg ha -1 ) although the lichen biomass is higher in a winter range than in a summer range area (Kumpula et al., 2009).Arboreal lichen is found most abundantly in the old growth pine and spruce forests.Intensive land use forms in the area are forest harvesting in commercial forest area, and tourism in more local fell areas.e largest allowed number of reindeer within the district during winter is 6000; the mean number of reindeer has been 5579±419 during years 2000-2007.

Historical records and reindeer data
Reindeer herders' observations and experiences of winters were collected from the annual management reports during winters 1972/1973-2009/2010.Additionally, reindeer census data from the Muonio district consisting of the numbers of reindeer counted during the annual round-ups in the autumn/early winter slaughter season during the period 1972-2010 was used.Annual calf production percent in the slaughter season (autumn/early winter) after each winter was produced using information on number of calves per 100 female reindeer (calf production percentage, CPP) (data provided by Reindeer Herders' Association).

Meteorological data
A 37-year time series of winter weather conditions (1972-2010, except winter 1982/1983; from 1 October to 30 April for each winter) was available from a synoptic observation station in Muonio, operated by Finnish Meteorological Institute (Fig. 1).e following weather parameters were obtained: air temperature (°C), relative humidity (%), wind velocity (m s -1 ) and wind direction (°), all observed from 2 meter height above the ground level.In addition, daily precipitation (mm) and snow depth values (m) were available from the station.
Annual mean temperature measured in the Muonio meteorological station was -1.4 °C during years 1971-2000, and annual precipitation 484 mm.Mean annual maximum snow depth during the period was 81 cm, with permanent snow cover usually formed after mid-October and with melting during May.Maximum snow depth is normally measured in March.(Drebs et al., 2002) We assume that weather conditions observed at the Muonio FMI station represent relatively well the general conditions of the whole reindeer herding district, and that the between-year variability observed at the Muonio station can be used as an estimate of the between-year variability on a larger area around the station.
e SNOWPACK model e meteorological observations were used to run the SNOWPACK-model.SNOWPACK is a one dimensional model for snowpack mass and energy balance, developed by the Swiss Federal Institute for Snow and Avalanche Research (SLF).A complete description of the model can be found in Bartelt and Lehning (2002) and Lehning et al. (2002a;2002b).
As a physically based model, SNOWPACK has been used in several applications, e.g. in avalanche forecasting (Lehning & Fierz, 2008) and as a part of watershed scale hydrological modeling (Lehning et al., 2006).SNOWPACK can estimate the evolution of the layered structure in the snow cover and the physical properties of these layers (grain size, grain form and bonding between the grains, temperature, density and hardness of snow, fractions of ice, liquid water and air volume in snow).It has been used together with a regional climate model by inputting the climate model output data when future changes in snow cover in open area were evaluated during a 100 year time scale in the selected locations in Finland (Rasmus et al., 2004) and more recently when future snow cover and its runo in the Alps were simulated (Bavay et al., 2009).e ability of the model to simulate the snow mass balance and snow structural properties has been validated in several climate conditions (Lehning et al., 1998;Lundy et al., 2001;Rasmus et al., 2007) and it has proven to be reliable, especially in open areas.In snow structure simulations, snow temperature and density had highest correlations with observations (r=0.90 and 0.85, respectively) and grain size and type lower (r=0.30;contingency coe cient C=0.71) (Lundy et al., 2001).
SNOWPACK uses air temperature, relative humidity, wind velocity and wind direction, and incoming shortwave and longwave radiation with 0.5-6 hour temporal resolution as input data.Depending on data and the aim of the simulations, either observed snow depth or precipitation can be used in the mass balance calculations of the model.Use of snow depth is justi ed when the data is easily available and when it is more important to simulate the snow layer properties most reliably, and in the open areas.However, precipitation data is still needed to correctly simulate the rain events which lead to icy layer formation in the snow cover.

Model simulations on snow structure evolution
e SNOWPACK-model was used to produce a 37-year time series on the annual evolution of snow structure on the basis of the used weather input data.Recently a canopy module has been added to the SNOWPACK model, which allows simulations also below the forest canopies (Lehning et al., 2006).e canopy radiation transmission sub-model has been calibrated and evaluated by Stähli et al. (2009), but the ability of SNOWPACK to correctly simulate the snow structure below the canopies has yet to be validated.Additionally, the energy and mass balance calculations below the canopies are sensitive to correct estimates of forest parameters (forest height, LAI and sky view fraction; Rasmus et al., 2012).For these reasons we decided to run our simulations in open area conditions only as meteorological input data was only available for open areas.
Temperature, humidity and wind data were obtained from the Muonio FMI station with a three hours resolution.Incoming shortwave radiation (W m -2 ) was available from the Sodankylä FMI station (approximately 170 km away) with the same temporal resolution.Incoming longwave radiation (W m -2 ) was estimated using the di erence between potential and observed incoming shortwave radiation, air temperature and relative humidity in each time step (method described in Konzelmann et al., 1994).Daily snow depth observations from the Muonio FMI station were used as a given parameter in simulations, because it is assumed that more exact the snow depth, the better the quality of the structure simulations.
As a bottom boundary condition there is a standard soil assumed (Bartelt & Lehning, 2002) as well as a prescribed temperature pro le in the beginning of the runs.Simulations were started on 1 October and nished on 30 April for each winter.Model output included time series for the mass and energy balance components in the snow cover, as well as graphical and numerical time series of the snow structure.

Validation of the snow density simulations
In this study the model SNOWPACK was used to simulate the snow structure, not depth or duration of the snow cover.Grain type and bonding between the grains largely determine the density of the snow, so density simulations are suitable for testing the performance of the model.
For the validation of the snow density simulations made by SNOWPACK, we used the monthly mean snow density values measured in four permanent snow survey lines located around the Muonio weather station (Fig. 1).
ese long-term snow survey lines are operated by Finnish Environmental Institute, SYKE.Lines are four kilometres long with 80 snow depth and eight to ten snow density measurements, designed to include the typical terrain and biotypes (open areas, forest openings, bogs and di erent forest types) of the region.(Perälä & Reuna, 1990) Calculations Parameters from both meteorological observations as well as from simulation outputs were listed in each winter (Table 1).From simulation outputs the average values of parameters were calculated for the whole winter period (November-April) and for three winter periods separately -early winter (November-December), mid-winter (January-February) and late winter (March-April).If snow fell later than 1 November or melted before 30 April, the average values for each parameter have been  calculated for the snow covered period only.A layer was classi ed as icy if simulation indicated melt and refreeze of the layer, and either major or minor grain type of the layer was melt/refrozen grains.A bottom layer was classi ed as ground ice if both major and minor grain types were melt/refrozen grains, and layer had gone through melt and refreeze.Statistical analysis of the reindeer and snow data was done using the Systat 13 and the IBM SPSS Statistics 20 softwares.Trends and statistical signi cance of the observed trends in reindeer and snow data were examined using the Mann-Kendall test.Pearson correlation test was conducted between the studied snow related parameters.e unpaired two sample t-test was used to determine the di erences in snow parameters between the winters judged as di cult or easy according the reindeer herders.T-tests were done as two-tailed and assuming equal variance for the two samples.A principal component analysis (PCA) was made to extract the components accounting for most of the variance in our set of 14 observed or simulated snow related parameters.Extracted four principal components were included in the analyses of correlations and t-tests.ditions during the winters and their response to di cult snow conditions.We had access to 38 reports between 1972/1973-2009/2010. Altogether 22 of the winters were classed as easy; in 16 winters snow conditions were experienced di cult (Table 2) and 12 of these cases were explained by deep snow cover.During nine of the winters snow melted late.In four winters, problems were caused by icy snow or ground ice (1991/1992, 2004/2005, 2006/2007 and 2009/2010).During three autumns the snow cover was reported to be formed on unfrozen ground, which means favorable conditions for mold growth and mold growth on pastures was reported during two of these winters.According the t-test, mean and maximum snow depth as well as, consequently, ground surface tem-perature were signi cantly higher (p<0.001)during the winters with reported di cult snow conditions; and snow season was signi cantly longer (P=0.02).

Observed snow conditions and reindeer calf production
Snow depth and length of snow cover time varied greatly among the winters (Table 3).No signi cant trends were observed in the long time series of these.Large between-year variability was seen also in calf production percentage during the observation period (Fig. 2).Weak but statistically signi cant increase in CPP of 0.483 per year was estimated using the Mann-Kendall test on trend in a time series (P=0.002).
Relevance of snow depth and melt date, experienced by reindeer herders, was con rmed since CPP was negatively correlated to winter mean and maximum snow depth (R=-0.45;P=0.005 and -0.38; 0.02, respectively) and length of snow cover time (R=-0.37;P=0.02) (Fig. 3).Still, winters with reported di cult snow conditions did not clearly show in the time series of CPP (Fig. 2  CPP between winters with easy and di cult snow conditions did no di er signi cantly from each other according the t-test.

Validation of the model SNOW-PACK
Simulated values of mean monthly snow densities were compared to the monthly observations from the four survey lines of Finnish Environment Institute (Fig. 4).e densities simulated by the SNOWPACK were generally higher than the observed ones; however, inter-annual variation in snow density was well reproduced by the model.e Pearson correlation coe cients between the SNOWPACK model outputs and the snow survey observations ranged from 0.08 in Hormakumpu (P=0.745),0.49 in Kattilamaa (P=0.002),0.56 in Hetta (P=0.001) to 0.58 in Pulju (P=0.004).When mean value of these four surveys was compared  to the simulated one, correlation was highest (r=0.61;P<0.001).

Distinguishing di cult winters for reindeer herding using snow structure simulations
As observed snow parameters, simulated structural parameters showed considerable variation between the winters, with no signi cant trends seen in the long time series (Table 4).Simulated mean snow density was signi cantly higher (P=0.02)during the winters with reported di cult snow conditions.Relevance of snow structural parameters was con rmed as CPP was negatively correlated to winter mean snow density and mean thickness of layers with density > 350 kg m -3 (R=-0.39;P=0.02 and -0.36; 0.03, respectively).
We used historical records in annual management reports to evaluate if the model was able to capture the problematic snow conditions for the herders.Snow covering unfrozen ground / mold growth on pastures were reported during autumns 1972, 1991 and 1996.Also Kumpula et al. (2000) observed exceptional growth of microfungi on pastures during winter 1996/1997.Simulated ground surface temperature was 0°C on the formation date of the permanent snow cover during two of these autumns, but this was the case also during additional eight autumns.When looking at the simulated mean ground surface temperature for the whole winter period, these three reported winters had clearly higher ground surface temperature than any other winters, equal or higher than -0.5 °C (Fig. 5).Icy snow and/or ground ice conditions were reported during four winters (1991/1992, 2004/2005, 2006/2007 and 2009/2010).
ese winters were not possible to distinguish accurately, when comparing winters on the basis of meteorological observations (Fig. 6).ree of the four winters with reported icy conditions had mean air temperatures as well as precipitation sums above the long term means, but in addition, ten other winters had this kind of conditions with no problems with icy snow reported.
Weather events leading to icy conditions were seen in the meteorological data only when looking at detailed three hour time resolution.During most of the events the air temperature stayed above zero at least for 24 hours, and maximum observed air temperature was several degrees above zero (3 -6 °C).After the warm spell, temperature dropped within 24 hours clearly below freezing.Occurrence of rain was not necessary during these events, but in some cases rain on snow before the drop in the temperature led to thick icy layers.
When analyzing the mean values of the simulated snow structural parameters, only few proved useful for identifying the icy conditions.e highest quartile of simulated fractions of icy layers included three of the four winters with reported icy conditions -together with six winters with no icy conditions reported (Fig. 7a).Mean hardness of the snow cover (Fig. 7b) showed the similar result.
On the basis of graphical outputs we identi ed the winters with thick (> 10 cm) ground ice layer or exceptionally thick (> 20 cm) icy layer anywhere  in the snow cover during the early and/or mid-winter.ree of the four winters with reported icy conditions could be distinguished using this method.For example during winter 1991/1992 the snow cover was not exceptionally deep but conditions were experienced as problematic.On the basis of the simulated snow structure evolution (Fig. 8) it is seen that a thick icy layer was formed in the snow cover in the end of January, and it was preserved till the spring.is layer probably prevented reindeer from grazing the ground lichen, and it did not help them to reach arboreal lichen either.
For the 37-winters study period, simulation outputs showed icy conditions during ve winters even if these were not reported by reindeer herders.

Principal components of the snow parameters
In PCA, observed and simulated snow parameters were reduced to four principal components, which explain approximately 80% of the total variance in the variables.In the further analysis these were called 'Snow amount' (PC1; e.g.snow depth and ground surface temperature; explains 33.27% of the total variance), 'Quality of snow' (PC2; e.g.snow hardness and thickness of ice layers; 23.19%), 'Ground ice' (PC3; thickness of ground ice, bottom layer hardness; 15.12%) and 'Duration of the snow season' (PC4; 8.88%).CPP was positively correlated with the Quality of snow-component (R=0.47;P=0.003), and in the t-test the Snow amount-component got signi cantly higher values during the winters experienced as dicult by reindeer herders compared to the easy winters (P=0.003).Quality of snow-component, in turn, got signi cantly lower values during the winters with di cult snow conditions (P=0.005).

Discussion
We used combination of meteorological observations, snow structure simulations and reports by reindeer herders to study the relevance of snow properties on reindeer herding in Muonio reindeer herding district in northern Finland.Suitability and reliability of a snow structure model SNOWPACK was evaluated on the basis of the measured snow density observations and by comparing the simulated snow characteristics to annual reproduction rate by reindeer and to observations of herders concerning the snow conditions.According the observations made by the reindeer herders in Muonio, deep snow cover and late snow melt are the most common unfavorable conditions for reindeer herding.Other problematic snow related conditions, are winters with icy snow or ground ice and snow cover that forms on unfrozen ground, potentially leading to mold growth.In previous studies, deep snow cover and late snow melt have been observed to cause high winter mortality (e.g.Kumpula & Colpaert, 2003;Helle & Kojola, 2008) and low calf production (e.g.Aanes et al., 2000;Kumpula, 2001) of caribou and reindeer.Relevance of these snow characteristics for reproduction rate of reindeer was also observed in this work -reindeer calf production in the Muonio herding district decreased as maximum snow depth and length of snow cover time increased.
E ects of snow structure on reindeer and caribou populations are less extensively studied.Ground ice has been seen to limit Svalbard reindeer population growth rate or cause population crashes (Kohler & Aanes, 2004;Hansen et al., 2011).Weather conditions favoring ice formation have been shown to be related to calf production and winter survival of reindeer (e.g.Kumpula & Colpaert, 2003;Helle & Kojola, 2004;2008), but only small number of direct observations on ice formation has been made.
Our results con rm the view (Helle & Kojola, 2008;Vikhamar-Schuler et al., 2013) that the icy conditions in snow or on ground are di cult to distinguish on the basis of long term meteorological observations.In our study the icy snow and/or ground ice conditions could be estimated with some accuracy using combination of winter mean air temperatures and precipitation sums, but this method informed also false cases relatively often (ten times during the 37 winter study period).Graphical results of the SNOWPACK simulations could be used to distinguish the icy conditions in three of four reported cases, but there were also ve false cases during the study period.Simulated values of fraction of icy snow and mean snow hardness could also be used to detect the icy snow conditions.Ground ice reported in 2009/2010 could not be detected using any of these methods.Proportions of false alarms (ground ice simulated but not reported) and failed detections (ground ice reported but not simulated) were similar to those reported by Vikhamar-Schuler et al. (2013) in Kautokeino.
Even though some winters reported by the herders to have exceptional snow conditions (e.g. in 1972/1973 mold growth on pastures and late snow melt; in several winters during 1990s deep and/or icy snow cover) had also low reindeer calf production, these winters were not always clearly distinguished in the time series of CPP.Nevertheless, on the basis of the annual management reports winters experienced problematic led to changes in herding practices and caused more herding work and expenses for reindeer herders (e.g. in the form of supplementary winter feeding).Also losses due to predation were probably partly connected to di cult snow conditions (see Tveraa et al., 2003).
Reindeer herding is relatively adaptable to intra-and inter-annual variations in grazing conditions (Tyler et al., 2007;Roturier & Roue, 2009;Riseth et al., 2010;Vuojala-Magga et al., 2011).In the old intensive herding system the reindeer herders tried to nd most suitable grazing areas for reindeer and also aided reindeer to make craters during winters with difcult snow conditions.Tightly herded animals could also be released to graze freely on arboreal lichen pastures (Helle & Jaakkola, 2008).Cutting old trees to provide reindeer arboreal lichens has also been done in the past (Berg et al., 2011).Since the 1980s, supplementary feeding of reindeer has been carried out annually in Muonio reindeer herding district, and it has gradually become more common due to the reduction of the most important natural winter forage (ground and arboreal lichens).Reindeer are usually fed on pastures, certain part of herd is also gathered for feeding in pens.
Compared to Kautokeino region in northern Norway, where the model has been previously used to estimate the foraging conditions for reindeer, Muonio has considerably milder winter climate.Main di erence still is the presence of the forest pastures, which complicates the conclusions drawn from the model outputs.In forested areas snow can a ect foraging con-ditions positively by lifting the animal and giving better access to arboreal lichens (Alectoria, Bryoria ssp.) (Helle, 1984).
In addition to availability of winter forage, there are other factors a ecting the well-being of the reindeer population.Poor body condition of animals in autumn due to foraging conditions and insect harassment in previous summer a ect the successful preparation to winter (Holster, 1948;Helle, 1980;Dau, 2005).e annual reports of Muonio reindeer herding district listed also damages to reindeer by tra c and predation.ese damages had large yearto-year variability and ranged between 79-446 individuals.is corresponded to 1-7 % of annual number of reindeer in winter stock, but did not have signi cant correlation with the CPP of the district.
SNOWPACK is not a spatial model, and simulations made with input data from a certain point best represent the mean snow conditions of the herding district.Pasture areas with variable vegetation and topography will develop variable snow covers.In the future, developing SNOWPACK model more suitable for forest landscape conditions or modeling snow distribution e.g. by a distributed model Alpine3D (Lehning et al., 2006;Stähli et al., 2009) are noteworthy options.Canopy module development should be continued in di erent types of forest environments, including di erent forest vegetation types and canopy coverages.
Using the structural modeling of annual snow conditions can give valuable information to reindeer herding compared to standard meteorological observations if present models are developed to be more suitable for northern boreal environment and also more precise forest and vegetation type data is used as simulation inputs.Even though the most often experienced problematic snow conditions (deep snow cover and late snow melt) are easily distinguished from the meteorological time series, use of this type of model improves the detec-tion of icy snow or ground surface conditions.In our study, the winters which were reported to have conditions favorable for mold growth on pastures could be distinguished most reliably from the SNOWPACK outputs.
Climate change will lead to changes in the established winter weather patterns and snow conditions, to which large fraction of the arctic and boreal species are well adapted (ACIA, 2004;IPCC, 2007).e on-going and expected change will a ect in many ways the foraging conditions for reindeer in winter, in uencing the productivity and pro tability of reindeer husbandry as a livelihood (Heggberget et al., 2002;Post & Forchhammer, 2002;Tyler, 2010;Moen, 2008;Turunen et al., 2009).Combination of climate and snow modeling can be a valuable tool when estimating the degree and range of these changes.In the future not only reindeer populations and reindeer husbandry, but also many other arctic and boreal species will be a ected by changes in the length of the snow season, snow depth and frequency of the ground ice on vegetation.

Figure 1 .
Figure 1. e reindeer management area and its 56 herding districts in northern Finland (the Muonio reindeer herding district shaded).Locations of the meteorological observation station operated by Finnish Meteorological Institute (FMI) in Alamuonio and the four Finnish Environment Institute's snow measurement lines (Hetta, Hormakumpu, Kattilamaa and Pulju) are marked on the map.

Figure 2 .
Figure 2. e annual and mean calf production percentage (CPP) in the Muonio reindeer herding district during 1972-2010.Winters experienced as di cult by the reindeer herders are marked with stars.

Figure 3 .
Figure 3. Calf production percentage (CPP) in relation to the annual observed maximum snow depth (a) and duration of the snow cover (b) in the Muonio reindeer herding district during 1972-2010.Pearson correlation coe cients and P-values given in the gures.

Figure 4 .
Figure 4. e mean snow density values (calculated from monthly values for whole winter) in open areas at four Finnish Environment Institute's snow measurement lines and in the SNOWPACK simulations for open area.

Figure 5 .
Figure 5. e mean simulated ground surface temperature for the whole snow cover period during 1972-2010.ree winters with conditions favoring mold growth reported by the reindeer herders marked with stars.

Figure 6 .
Figure 6. e measured annual mean air temperature during winter in relation to the corresponding mean precipitation sum in Muonio during 1972-2010.Winters with icy conditions reported by reindeer herders marked with stars.

Figure 7 .
Figure 7. e simulated mean fraction of icy layers (a) and hardness of the snow cover (b) during winters 1972-2010.Winters with icy conditions reported by the reindeer herders marked with stars.

Figure 8 .
Figure 8. Evolution of the snow structure during winter 1991/1992 according to the simulations made by the SNOWPACK model.Shadowed areas separate the layers with di erent snow structure.Areas with light gray shading indicate dry snow with grains recently bonded together, and dark shaded areas closer to the bottom of the snow cover represent layers with loosely bonded, porous depth hoar.Occasions with melt and subsequent refreeze of snow (formation of icy layers or ground ice) are marked with white arrows.Icy layers formed during the early or mid-winter are possible to distinguish throughout the winter until the snow melts.

Table 1 .
(Vikhamar-Schuler et al., 2013)ogical observations and from simulation outputs in each of the study winters.Snow density above a 350 kg m -3 threshold was considered as icy and problematic for reindeer grazing(Vikhamar-Schuler et al., 2013).

Table 2 .
Di cult snow conditions informed in the annual management reports of the Muonio reindeer herding district during 1972-2010.Calf production percentage (CPP), type of snow condition and reported impacts of snow conditions on reindeer populations (di culties in grazing/active movement of reindeer/winter mortality) as well as responses of reindeer herding practices (feeding) are listed.
) and

Table 4 .
Mean, minimum, maximum and standard deviation of whole winter mean values of the selected snow structural parameters simulated by the SNOWPACK model inMuonio during 1972Muonio during  -2010.  .