Seasonality, weather and climate affect home range size in roe deer across a wide latitudinal gradient within Europe

Authors


Summary

  1. Because many large mammal species have wide geographical ranges, spatially distant populations may be confronted with different sets of environmental conditions. Investigating how home range (HR) size varies across environmental gradients should yield a better understanding of the factors affecting large mammal ecology.
  2. We evaluated how HR size of a large herbivore, the roe deer (Capreolus capreolus), varies in relation to seasonality, latitude (climate), weather, plant productivity and landscape features across its geographical range in Western Europe. As roe deer are income breeders, expected to adjust HR size continuously to temporal variation in food resources and energetic requirements, our baseline prediction was for HR size to decrease with proxies of resource availability.
  3. We used GPS locations of roe deer collected from seven study sites (EURODEER collaborative project) to estimate fixed-kernel HR size at weekly and monthly temporal scales. We performed an unusually comprehensive analysis of variation in HR size among and within populations over time across the geographical range of a single species using generalized additive mixed models and linear mixed models, respectively.
  4. Among populations, HR size decreased with increasing values for proxies of forage abundance, but increased with increases in seasonality, stochastic variation of temperature, latitude and snow cover. Within populations, roe deer HR size varied over time in relation to seasonality and proxies of forage abundance in a consistent way across the seven populations. Thus, our findings were broadly consistent across the distributional range of this species, demonstrating a strong and ubiquitous link between the amplitude and timing of environmental seasonality and HR size at the continental scale.
  5. Overall, the variability in average HR size of roe deer across Europe reflects the interaction among local weather, climate and seasonality, providing valuable insight into the limiting factors affecting this large herbivore under contrasting conditions. The complexity of the relationships suggests that predicting ranging behaviour of large herbivores in relation to current and future climate change will require detailed knowledge not only about predicted increases in temperature, but also how this interacts with factors such as day length and climate predictability.

Introduction

The distribution of many animal species is currently changing over many parts of the world (Parmesan & Yohe 2003; Root et al. 2003). The ability of a species to track the shifting climate relies on its dispersal capability and life-history responses to novel environmental constraints (Anderson et al. 2009). Because many large mammals have wide geographical ranges, populations of a given species may be confronted with different suites of limiting factors, causing the geographical range to contract or expand. For example, body growth of red deer (Cervus elaphus) is limited by drought during summer, occurring during autumn and winter at the southern fringe of its distributional range, in Spain, while body growth occurs in summer and ceases during fall and winter towards their northern distributional limit, in Scotland and Norway (Martínez-Jauregui et al. 2009). Life history, individual trajectories and social status are usually mediated through interactions between an animal and its habitat (Stephens, Brown & Ydenberg 2007), and this, in turn, is influenced by an animal's pattern of space use (e.g. Stien et al. 2010). Investigating how climate affects individual- and population-level ranging behaviour under contrasting conditions of seasonality and availability of food resources, that is, across wide distributional gradients, should help us to obtain a better understanding of how climate change may affect animal ecology.

An important concept describing animal space use is the home range (HR), an area of limited spatial extent over which an animal repeatedly travels in search of food and mates to survive and reproduce (Börger, Dalziel & Fryxell 2008). Understanding why HR size varies among and within species has received considerable attention for decades (reviews in Börger, Dalziel & Fryxell 2008; Kie et al. 2010). For an individual, HR size is closely linked to the balance between its metabolic requirements and its food acquisition abilities. The positive allometry of HR size with body mass is the best well-known evidence for a link with energetic requirements at the interspecific level, such that the average HR size of a species increases with increasing body size because of the greater energy needs of larger species (McNab 1963; Harestad & Bunnell 1979).

Intraspecific variation in HR size is less well understood (Nilsen, Herfindal & Linnell 2005; Börger et al. 2006a; Börger, Dalziel & Fryxell 2008). Within a population, HR size varies in relation to temporal-, spatial- and individual-level processes (Börger et al. 2006a; Rivrud, Loe & Mysterud 2010; van Beest et al. 2011). The most commonly reported mechanisms underlying HR size variation among populations are variation in food availability and distribution through foraging (Kjellander et al. 2004; Anderson et al. 2005), movement energetics (e.g. due to differences in topography Kie et al. 2002; Walter et al. 2009) and predation avoidance (Anderson et al. 2005). Within populations living in temperate regions, variation in HR size may be related to predictable seasonal variations in environmental conditions (i.e. related to the regular photoperiodic pattern) and to stochastic variability caused by local weather (Börger et al. 2006a; Rivrud, Loe & Mysterud 2010). Furthermore, the distribution and characteristics of habitat features, such as plant biomass (in roe deer: Tufto, Andersen & Linnell 1996; moose (Alces alces): van Beest et al. 2011), may generate further variation. Social organization and life histories are also important determinants of an animal's movement patterns, for example, females with a calf at heel usually have smaller summer HR sizes than barren females (van Beest et al. 2011).

Across Europe, climate change should lead to shifts in weather regime of differing magnitude depending on latitude, with forecasted consequences on seasonality, average (e.g. general increase in temperature) and variance (e.g. higher frequency of extreme events) of weather variables (IPCC 2007). For herbivores, climate change should affect the phenology, the duration of vegetative growth and the spatial distribution of plants (Root et al. 2003) that, in turn, should affect seasonal (predictable) and inter-annual (un-predictable) variation in forage quantity and quality (Craine et al. 2010). Large herbivores should therefore respond to climate change by reducing variation in foraging processes through space use, so as to minimize consequences on fitness (e.g. Ferguson et al. 1999). We can therefore expect HR size to reflect the ecological constraints faced by animals, and thus, large-scale analysis of how range size varies in relation to variation in these constraints (e.g. along environmental gradients) should help us to understand how large herbivores may be able to cope with climate change.

Here, we used the EURODEER data base (www.eurodeer.org) to analyse patterns of variation in weekly and monthly HR size of roe deer at inter- and intra-population levels, from the southern part of their range, in Italy, to the northern part of their range, in Scandinavia. The study sites cover widely contrasting environmental conditions, from lowland to mountainous areas, yielding a gradient of climatic and seasonal resource limitation and providing the opportunity for an unusually comprehensive analysis of the factors affecting ranging behaviour of a single species across its geographical range. Earlier broad-scale studies have determined the importance of factors such as habitat heterogeneity (Kie et al. 2002; Anderson et al. 2005; Walter et al. 2009 on ungulates) and how productivity and seasonality affect spatial variation in HR size across sites (McLoughlin, Ferguson & Messier 2000; Herfindal et al. 2005; Nilsen, Herfindal & Linnell 2005 on carnivores), but no study has yet tried to disentangle in a single spatiotemporal analysis how weather, climate and seasonality affect HR size across a wide range of latitudes, that is, a full spatiotemporal analysis using detailed GPS data. Indeed, such large-scale studies of HR size should lead to a better understanding of how climate affects space use behaviour in contrasting environmental contexts and are clearly important for generating predictions on the impact of climate change for the distribution and resilience of species. Our analysis was based on a large, high-quality data set of GPS-monitored roe deer and, for the first time to our knowledge, compared variation of HR size among populations across Europe while controlling for temporal variation of HR size within populations. Indeed, HR size variation depends on the temporal scale and responds to spatiotemporal changes in environmental conditions (see van Beest et al. 2011). We first tested, at the within-population scale, for temporal variation in HR size in relation to day length (as a cue of seasonality) and to available forage (indexed by temperature and the NDVI) (Table 1). We then tested for large-scale gradients in HR size related to latitude; then we evaluated a suite of hypotheses concerning the mechanisms underlying these broad patterns related to the degree of seasonality and primary productivity (Table 1), while controlling for temporal variation in HR size.

Table 1. Summary of the different expectations in terms of variation in roe deer HR size within and across latitudes at weekly and monthly scales, and the explanatory variables used to test them
Spatial scaleExplanatory variableBiological effectExpectation
Among populationsThe predictable variation of temperature (Cosw and Sinw)High predictability of seasonalityHR size increases with increasing seasonality
The unpredictable variation of temperature (RSE)Low predictability of seasonalityHR size increases with the increasing level of stochasticity in seasonality
Latitude/slopeThese geographical variables are linked to seasonalityHR size increases with increasing seasonality (Latitude: Ferguson, Higdon & Lariviere 2006, Slope: Anderson et al. 2005)
Snow cover durationHigh seasonality and reduction in available forageHR size increases with increasing snow cover duration (Snow depth: Rivrud, Loe & Mysterud 2010)
Index of landscape diversity, Percentage tree coverThese variables are linked to landscape heterogeneityHR size decreases with increasing landscape heterogeneity (Saïd & Servanty 2005)
Night light indexThis index is linked to the level of human activityHR size increases with increasing human disturbance (Grignolio et al. 2011)
Roe deer densityReduction in available forage – increase in social interactionsHR size decreases with increasing roe deer density (Kjellander et al. 2004)
Temperature, Leaf area index and INDVIThese variables are proxies of available forageHR size decreases with increasing available forage (temperature: Börger et al. 2006a,b, NDVI: Hansen et al. 2009)
Within populationsDay lengthThis is a cue for seasonalityHR size decreases with increasing photoperiod (Börger et al. 2006a,b)
Temperature and NDVIThe temperature and NDVI are proxies of available forageHR size decreases with increasing available forage (Tufto, Andersen & Linnell 1996; temperature: Börger et al. 2006a; NDVI: Hansen et al. 2009)

Materials and methods

Data collection and study areas

To analyse patterns of variation in HR size of roe deer, we used the EURODEER data base (Cagnacci et al. 2011). We focussed on females monitored with GPS collars and selected all study areas with at least seven individuals. Thus, we obtained location data for 190 females (juveniles and adults) from seven study sites located along a wide latitudinal and altitudinal gradient (Electronic Appendix S1, Supporting information and Fig. 1).

Figure 1.

Map of Western Europe with the locations of the seven study sites (see Electronic Appendix S1, Supporting information) from which HRs were estimated. The study sites were numbered from 1 to 7 according to latitude, from the south to the north (1 = I:38°N; 2 = I:41°N; 3 = F:43°N; 4 = I:46°N; 5 = D:49°N; 6 = S:58°N; 7 = N:61°N).

Estimation of home range size

We first selected and standardized location data among the seven study sites. We removed all locations recorded during the first week of monitoring after an animal's capture and release because of potential behavioural alterations following capture and handling (Morellet et al. 2009). The roe deer monitored in the two southern Italy populations (sites I:38°N and I:41°N) were reintroduced animals which might therefore display unusual postrelease explorative behaviours (Calenge et al. 2005), so we removed the entire first month of monitoring postrelease for these sites to ensure that ranging behaviour had stabilized (see Electronic Appendix S9, Supporting information). Finally, because the sampling regime of GPS locations differed among and within study sites, we selected a comparable number of locations per unit of time for each individual (Börger et al. 2006b). We split the day into six periods of 4 h ([0–4], [4–8], and so on) and retained the closest fix to the beginning of a given interval for each period. Thus, we retained a maximum of six locations per day (mean = 4·1 ± 1·1 locations per day), that is, at most, one fix every 4 h.

We considered both weekly and monthly temporal scales in our analyses rather than using site-specific seasons because we really needed to estimate HR size over the same time frame everywhere (see Börger et al. 2006a,b). Because day length influences the ranging behaviour of roe deer (Börger et al. 2006a), we arbitrarily set the winter solstice (December 22nd) as the common starting day for all sites for studying the temporal variability of HR size. We then divided the year into 52 and 12 time periods of equal time length for the weekly and monthly scales, respectively. At the weekly scale, we considered 15 locations per deer and per week as a minimum to obtain a representative estimate of the HR (Börger et al. 2006b). Similarly, at the monthly scale, we set the minimum to 60 locations per deer combined with the 15 locations per week rule to ensure that sampling covered the whole month uniformly. We estimated HRs using fixed-kernel methods at both temporal scales considering (i) an ad hoc approach to select the optimal smoothing parameters h for each HR estimate and (ii) a constant h equal to the overall median of the smoothing parameters estimated by the ad hoc approach for each individual HR estimate. HRs were estimated with the adehabitatHR package (Calenge 2006) for r (version 2.11.1; R Development Core Team 2010) at the 90% and 50% isopleths (Börger et al. 2006b). Finally, to avoid analysing very atypical ranging behaviour, in particular, nonsedentary individuals, we excluded one deer who had a HR of more than 1000 hectares at the weekly scale.

Site-specific environmental variables

For the within-population approach, we calculated the day length for each Julian day at each study site using the package ‘pheno’ in r (Schaber 2012). As day length is a function of latitude, we used the average latitude of each study site for calculations and assumed that within-site variation in day length was negligible. We obtained the daily average temperature for each study site using one or several meteorological stations per site situated within, or very close to the study site (11·9 ± 11·5 km). For mountainous areas (sites I:38°N and I:46°N), we used at least two meteorological stations to characterize weather conditions for HRs at low and at high elevation. That is, each weekly and monthly HR was matched with data from the closest weather station depending on its average altitudinal position. We used SPOT Vegetation NDVI data for each study site as a proxy of forage availability during the year (Pettorelli et al. 2005, 2011). To reduce the effect of cloud contamination and atmospheric variability (Pettorelli et al. 2005), we smoothed the NDVI time series with a mobile average filter and extracted the seasonal component of NDVI (see Santin-Janin et al. 2009 for a similar approach). Because the NDVI is only measured once every 10 days, the raw data values were too sparse to generate an estimate for each week or each month. We therefore fitted a nonparametric model (with a spline) to the NDVI time series and then predicted the value of NDVI for the median date of each week and each month, respectively. We retained NDVI values between April and November only because the presence of cloud or snow at high latitudes often leads to erroneous and poorly comparable NDVI measures (Pettorelli et al. 2005).

For the among-population analyses, we characterized each study site by a single value for each of the tested environmental variables (Table 1, Electronic Appendix S2, Supporting information). Each site-specific value for each of the variables was taken as the average of the variable's values for all pixels which fell within the study sites’ boundaries (defined as the 100% MCP of all roe deer locations for that population). Note that, due to the low spatial resolution of the environmental variables analysed relative to the study site, we could not quantify spatial variation of these variables within study sites. We extracted five remotely sensed measurements for our analyses from http://spatial-analyst.net with a cell size of 0·05 arcdegrees (5·6 × 5·6 km2): the average snow cover duration from MODIS, that is, the average number of days per year with some snow on the ground (http://modis.gsfc.nasa.gov/), the average slope from the Global SRTM Digital Elevation Model (with a resolution of 30 m), the Leaf Area Index (LAI) that indexes photosynthetic primary production, evapo-transpiration and plant growth (Pettorelli et al. 2011), the percentage tree cover index calculated as the proportion of tree cover at that site to describe overall landscape openness, and a night light index (Small, Pozzi & Elvidge 2005) as a measure of human activity. We calculated the mean average annual temperature across years for each study site from daily measurements. Fourier transforms were used to quantify the predictable and unpredictable components of within-year variation in temperature. Coefficients of the cyclic functions (cos and sin) index the predictable variation of seasonality, while the square root of the estimated variance of the random error (RSE) indexes the unpredictable variation of temperature. We used the integrated annual normalized difference vegetation index (INDVI) averaged across years from the SPOT Vegetation NDVI as a proxy of overall food resource availability at a given site, especially useful for comparison at broad scales (Pettorelli et al. 2011). We used an index of landscape diversity based on Corine land cover categories (Rocchini et al. 2011) as a proxy of within-site spatial heterogeneity in food resources. Finally, we generated a value for latitude at each site using the average latitude of all GPS locations of roe deer at that site.

Statistical analyses

We used linear mixed models (LMM) of log-transformed HR size, with individual identity as a random factor to control for repeated observations per individual (Börger et al. 2006a). Our basic model included (i) age class (juvenile, that is, <18 months old at the winter solstice vs. adult, that is, more than 18 months old at the winter solstice) to account for potential age-related changes in HR size; (ii) the distance between the geometric centre of successive HRs (we fixed the first distance to zero as a starting location) to account for possible variation in HR location due to specific ranging behaviour or landscape structure; and (iii) the square root of the number of locations used to estimate the HR to index the precision of the HR estimate (Gautestad & Mysterud 1993). Finally, (iv) we controlled for the birth period as a two modality factor (birth period vs. nonbirth period), including its interaction with age (Bongi et al. 2008) because juvenile do not reproduce in roe deer. We defined the birth period as running from the second week of April to the third week of June at the weekly scale and from the third week of April to the third week of June at the monthly scale. All more complex models systematically included these variables. We fitted all models with the ‘lme’ function in the ‘nlme’ package (Pinheiro et al. 2009) in r. As we needed to control for within-population variation to compare average HR size among populations, we first present within-population statistics.

Within-population variation of home range size

A previous study in central Italy (Börger et al. 2006a) found a consistent seasonal pattern of roe deer HR size variation which was closely related to day length variations. We therefore first tested the generality of this temporal pattern across study sites by fitting a model including time as a sine/cosine function to describe the temporal pattern of HR size variation over the year. This parametric model assumes that HR size variation over time is cyclic and symmetrical. In this model, time was included as a time–study site interaction term to test whether the temporal pattern in HR size variation was consistent among study sites. We then replaced the time variable by the weekly or monthly descriptors of the environment to attempt to explain the observed temporal pattern of HR size variation. First, we assessed the generality of the negative relationship between day length and HR size (Börger et al. 2006a). We included the interaction between study site and day length to evaluate whether this effect of day length was consistent across sites. Second, we assessed the influence of temperature and NDVI as a proxy of food availability for roe deer on HR size variation. As higher temperatures may improve plant productivity at northern latitudes, but reduce it at southern latitudes, we included the interaction between study site and temperature or NDVI. We used Akaike Information Criterion (AICc; Burnham & Anderson 1998) and AIC weights (wi) to select the model best supported by the data for each environmental descriptor separately. In order to compare the relative effects of the time-dependent variables, we scaled the predictive variables (by subtracting the sample mean and dividing this by their sample standard deviation) to standardize each of them to units of standard deviation (Schielzeth 2010). To quantify the amount of temporal variation in average HR size accounted for by each of the time-specific variables day length, temperature and NDVI, we used the ANODEV procedure (Grosbois et al. 2008). The R2 of the ANODEV measures the proportion of variation in average home range size among sites or time periods that is accounted for by site- or time-specific covariates. The ANODEV compares the deviance of three models: the basic model (simplest model), the basic model with a time-specific variable (intermediate model) and the basic model with time entered as qualitative variables (most complex). By comparing the deviance of these three nested models, the R2 provides a measure of the relative importance of each of the time-specific factors.

Among-population variation of home range size

At the among-population level, we compared the averages of site-specific HR size, controlling for the most important factors (i.e. the temporal variation, the two-way interaction between age class and birth period, the distance between the geometric centre of successive HRs and the square root of the number of locations) to attain comparable HR sizes for each site. We controlled for seasonal variation of HR size using a generalized additive mixed model (GAMM), including roe deer identity as a random factor to control for repeated observations per individual. We used GAMM rather than standard linear models because GAMMs can more efficiently capture nonlinear temporal variation. The smoothed effect of time (week or month) was based on cyclic P-splines, which differed among study sites, to take into account the cyclic pattern of HR size variation over the year. Finally, we added study site as a factor (considered as the reference model) to the base model and then replaced this factor by a given environmental variable, one at a time, to test our predictions (see Table 1). We considered the 11 environmental variables separately (i.e. rather than using a multiple regression-type approach, we built a separate model for each environmental variable): the predictable and unpredictable components of within-year variation in temperature, latitude, slope, snow cover duration, temperature, landscape diversity, LAI, night light index, percentage tree cover and INDVI (between April and November). To quantify the amount of variation in average HR size among sites that was explained by a given environmental variable, we again used the ANODEV procedure (as above). We fitted models with the ‘gamm’ function in the ‘mgcv’ r package (Wood 2006).

Results

The observed mean HR size of all individuals was 55·4 and 76·5 ha at weekly and monthly scales, respectively, and varied among study sites from 38·6 to 77·4 ha and from 51·4 to 136·0 ha at weekly and monthly scales, respectively (average HR size at weekly/monthly scales: I:38°N = 38·6/64·5 ha; I:41°N = 51·8/86·2 ha; F:43°N = 55·7/72·7 ha; I:46°N = 42·3/51·4 ha; D:49°N = 57·2/79·3 ha; S:58°N =54·2/68·3 ha; N:61°N = 77·4/136·0 ha). Our overall results were robust with respect to the choice of HR estimation method (see Electronic Appendix S4, Supporting information) and the HR isopleth used (see Electronic Appendix S5, Supporting information); thus, we only present results using the 90% fixed-kernel method based on the constant h in the following.

Within-population variation in home range size

HR size of roe deer females showed strong seasonal variation over time that differed in amplitude and timing among study sites (Fig. 3, Electronic Appendix S8, Supporting information). The model describing temporal variation in HR size with simple cyclic and symmetrical parametric functions of time explained 67·5% and 76·8% of the observed temporal variation in HR size at the weekly and monthly scales, respectively. In general, HR size was at a minimum during spring and a maximum during winter (Electronic Appendix S8, Supporting information). For the southernmost and northernmost study sites, however, the peaks tended to occur earlier. The seasonal pattern of HR size variability was almost identical at the weekly and monthly time-scale at all sites, except for one southern study site (I:38°N) for which minimum HR size at the monthly scale occurred later than at the weekly scale.

Overall, HR size was negatively related to day length, but this relationship was highly variable among study sites (Fig. 3, Table 3; ΔAICc > 100 and AIC weight = 1 for models describing a different response to day length among study areas compared with a model with a constant response, at both temporal scales) explaining 63·4% and 71·8% of the observed temporal variation in HR size at weekly and monthly scales, respectively. The effect of day length was very weak and negative for the northernmost study site (N:61°N) and positive at the weekly scale for the southern site I:41°N. Controlling for the effects of the other covariates, the predicted change in HR size in response to a reduction of 50% in day length ranged between −1·7 and +14·2 ha among study sites at the weekly scale and between +2·3 and +19·8 ha at the monthly scale (weekly/monthly scales: I:38°N = 2·5/16 ha; I:41°N = −1·7/2·5 ha; F:43°N = 14·2/19·8 ha; I:46°N = 3·5/3·6 ha; D:49°N = 7·4/13·5 ha; S:58°N = 5·7/11·3 ha; N:61°N = 3·8/2·3 ha).

Variation in temperature also markedly affected HR size at both spatial scales, but to a different degree among study areas (Fig. 3, Table 3, ΔAICc > 90 and AIC weight = 1 for models describing a different response to temperature among study areas, compared with a model with a constant response, at both temporal scales), explaining 41·0% and 57·7% of the observed temporal variation in HR size at weekly and monthly scales, respectively. According to the best model, the predicted change in HR size in response to a reduction of 50% in temperature varied between −4·7 and +14·8 ha among study sites at the weekly scale and between −13·5 and +22·9 ha at the monthly scale (weekly/monthly scales: I:38°N = −4·3/16·2 ha; I:41°N = −4·7/−0·8 ha; F:43°N = 14·8/22·9 ha; I:46°N = 4/7 ha; D:49°N = 5·7/14·4 ha; S:58°N = 3·3/11·5 ha; N:61°N = −1·9/−13·5 ha). As expected if we consider temperature as a proxy of food resource availability for roe deer, HR size decreased with increasing temperature at the weekly scale for intermediate latitudes, but a reverse pattern was found for the southernmost (I:38°N and I:41°N) and northernmost latitudes (N:61°N). At the monthly scale, the results were similar except for the southernmost study site (I:38°N), for which HR size decreased with increasing temperature, in contrast to our expectation.

Similarly, as expected if we consider the NDVI (from April to November) as a proxy of food resource availability for roe deer, HR size decreased with increasing values of NDVI, but the relationships differed among study sites (Table 3; ΔAICc > 45 and AIC weight = 1 for models describing a different response to NDVI among study areas, compared with a model with a constant response, at both temporal scales), explaining 51·1% and 44·3% of the observed temporal variation in HR size at weekly and monthly scales, respectively. According to the best model, the predicted change in HR size in response to a reduction of 50% in NDVI ranged between −1·8 and +10·4 ha among study sites at the weekly scale and between −12·1 and +10·9 ha at the monthly scale (weekly/monthly scales: I:38°N = −1·8/9·6 ha; I:41°N = 2·3/2·7 ha; F:43°N =10·4/9·8 ha; I:46°N = 7·2/9·4 ha; D:49°N = 7·4/8·1 ha; S:58°N = 8·7/10 ha; N:61°N = −0·9/−12·4 ha). However, contrary to our predictions, NDVI had a positive effect on HR size at the northernmost (N:61°N) and the southernmost latitudes (I:38°N and I:41°N).

Among-population variation in home range size

At the among-population level, HR size increased with the predictable and unpredictable components of variation in temperature, with latitude and with snow cover duration at both weekly and monthly scales (Fig. 2, Table 2). Unexpectedly, HR size decreased with slope at both scales although the effect size was very low at the monthly scale (Table 2). As expected, the proxies of forage availability (INDVI, LAI and average annual temperature) were negatively correlated with HR size (Fig. 2). However, the relationship between HR size and the INDVI was strongly influenced by N:61°N, the northernmost study site. In contrast, variation in HR size among populations was not influenced by the index of landscape diversity, the night light index or the percentage tree cover.

Table 2. Candidate general additive mixed models describing the logarithm of HR size variation at weekly and monthly scales, including a smoothing factor (based on cyclic P-splines) per study site, an environmental variable (see the list in the table), the interaction between age and birth period, the distance between the geometric centre of successive HRs and the square root of the number of locations used to estimate the HR, with the roe deer's identity as a random factor (N = 7510, and 1529 at weekly and monthly scales, respectively). The R2 describes the proportion of variation in HR size among sites which is accounted for by the given environmental variable in an ANODEV procedure. The lines of the table are sorted in relation to the values of R2 at a monthly scale
Environmental variablesWeekly scaleMonthly scale
Estimate (SE) R 2 Estimate (SE) R 2
INDVI−0·0669 (0·0137)0·58−0·0573 (0·0151)0·38
Snow0·0018 (0·0006)0·210·0024 (0·0007)0·33
Unpredictable variation of temperature0·149 (0·0528)0·200·1907 (0·0586)0·29
Latitude0·0177 (0·0047)0·350·016 (0·0051)0·27
Predictable variation of temperature−0·147 (0·0521)0·20−0·2423 (0·1063)0·14
Leaf area index−0·007 (0·0023)0·23−0·0056 (0·0025)0·13
Temperature−0·0185 (0·0077)0·15−0·0179 (0·0084)0·12
Slope−6·1233 (1·4491)0·44−2·5593 (1·6183)0·07
Index of landscape diversity−0·0157 (0·0119)0·040·0176 (0·0127)0·05
Night light index−0·0064 (0·009)0·010·0073 (0·0096)0·02
Percentage tree cover index−0·0017 (0·001)0·08−0·0003 (0·0011)0·00
Figure 2.

Deviation from average HR size among the different study sites in relation to significantly informative environmental drivers from Table 2 (INDVI between April and November, snow cover, unpredictable variation of temperature, latitude, predictable variation of temperature, leaf area index, average temperature and slope), at weekly and monthly scales, derived from the GAMM of the basic model. The study sites were numbered from 1 to 7 from the south to the north (1 = I:38°N; 2 = I:41°N; 3 = F:43°N; 4 = I:46°N; 5 = D:49°N; 6 = S:58°N; 7 = N:61°N). The lines (and dotted lines) represent the models (with the confidence intervals) presented in Table 2, the gray line indicates an absence of deviation.

The difference in HR size at the weekly scale (monthly scale in brackets for each variable) for the minimum and maximum observed values for each of the environmental variables (see Electronic Appendix S2, Supporting information) was 10·9 ha (14·9 ha) for temperature, 8·3 ha (15·6 ha) for the stochastic variation of temperature, 8·1 ha (11·4 ha) for the average temperature, 13·9 ha (18·0 ha) for the latitude, 8·3 ha (16·0 ha) for the snow cover, 21·1 ha (25·7 ha) for the INDVI and 7·2 ha (8·3 ha) for the LAI. Similarly, the predicted difference in HR size over the range of observed values of slope and deer density was 10·1 ha for slope at the weekly scale and 9·5 ha for roe deer density at the monthly scale. Based on the proportion of explained variance (Table 2), the environmental factor with the strongest influence on HR size variation was the INDVI, although latitude was also influential at both temporal scales. Nevertheless, it was difficult to disentangle the effects of the different environmental factors, as most of the variables considered here were strongly correlated (see Electronic Appendix S6 and S7, Supporting information).

Discussion

This first pan-European report of the biogeography of HR size in a large herbivore highlights how local weather, climate and seasonality jointly affect space use patterns through time and space. As expected from the scaling effect of energetic needs on HR size, average HR size was negatively associated with proxies of forage abundance (INDVI, LAI and average annual temperature) determined by local weather conditions. These local weather conditions are expected to change in the future (IPCC 2007); hence, roe deer ranging behaviour should be expected to respond accordingly as a function of future climate change. HR size across both time-scales increased with an increase in both the predictable and unpredictable variation of local temperature and with snow cover duration (components of weather seasonality) and latitude. This suggests that roe deer living at high latitude compensated for the shorter growing-season by foraging over a larger area. Further, a strong seasonality in HR size was observed in all populations across the roe deer's geographical range, though its amplitude varied according to site-specific day length or proxies of forage abundance (temperature or NDVI). In contrast, landscape topography (slope) and landscape heterogeneity (index of landscape diversity, percentage tree cover and night light index) appeared to be of little additional importance in determining roe deer ranging behaviour.

Home range size: the fundamental role of energy from forage

Forage availability and spatial distribution are pervasive predictors of animal movement, space use and habitat selection (Fretwell & Lucas 1970) and, ultimately, of HR size (Tufto, Andersen & Linnell 1996; McLoughlin & Ferguson 2000; Anderson et al. 2005; van Beest et al. 2011). We have shown that the average HR size of roe deer increases with latitude across Europe (Fig. 2), while similar findings were previously reported for wolf (Canis lupus) (Okarma et al. 1998). For large herbivores, this trend is expected in relation to the general decrease in food availability due to the progressive decline in the length of the growing season and net primary productivity at northern latitudes (Huston & Wolverton 2009). Across our study sites, encompassing almost the entire latitudinal gradient of the geographical range of roe deer, the growing season ranges from 320 days in the south to 138 days in the north (using 5 degree days as a growth threshold i.e. the number of days with a temperature >5 °C). However, in the southernmost range, drought during summer may shorten the growing season (Börger et al. 2006a). Our interpretation of a general decrease in forage availability with increasing latitude was further supported by the negative association we observed between average HR size and correlates of forage availability (INDVI, LAI and average annual temperature) across study sites (Fig. 2). Overall, roe deer appear to compensate for the shorter growing-season they experience at high latitudes by foraging over larger areas, even though food resources may be either more abundant or of higher quality during the growth season than at lower latitudes.

Larger HRs at high latitude could also result from an allometric relationship with body size (McNab 1963), as Scandinavian roe deer are around 30% larger than Spanish deer (Andersen et al. 1998). Harestad & Bunnell (1979) estimated an allometric coefficient of HR size of 1·02 for herbivores, that is, close to linearity. If latitude equates to a proxy of roe deer body size, the allometric coefficient we obtained is 0·89 and 0·76 at the weekly and monthly scales, respectively, suggesting that factors other than body size shape HR size variation, for example food abundance (Tufto, Andersen & Linnell 1996). For instance, average HR size increased with increasing duration of snow cover at both the weekly and monthly scales (Fig. 2). Because female roe deer are income breeders, with low levels of stored fat (Andersen et al. 1998) and relying on current energy intake to offset the costs of reproduction, when environmental conditions are limiting they must increase HR size so as to access sufficient food resources to fulfil their energetic requirements.

Our results on temporal variation in HR size were consistent across sites, lending general support for the hypothesis of indirect effects of weather on the ranging behaviour of large herbivores, that is, through a modification of forage availability in relation to seasonal variation in temperature. This interpretation is supported by the similar response of HR size variation to a widely used proxy of food resources for herbivores, the NDVI (see Table 3; Pettorelli et al. 2005), in all sites, at least at a monthly scale. Comparable relationships of declining HR size with increasing productivity and decreasing seasonality have been demonstrated for several species of carnivores (McLoughlin, Ferguson & Messier 2000; Herfindal et al. 2005; Nilsen, Herfindal & Linnell 2005), although the opposite pattern occurs in some other species (Nilsen, Herfindal & Linnell 2005). Finally, landscape topography and landscape heterogeneity appear to be of lesser importance for determining roe deer ranging behaviour, even though several studies have documented effects of landscape attributes such as species richness or habitat heterogeneity on HR size (Kie et al. 2002; Walter et al. 2009). Although HR size decreased with increasing average slope at the weekly scale (Fig. 2), this environmental variable is likely also linked to forage availability (Electronic Appendix S6, Supporting information). Pronounced slope is clearly characteristic of mountain ranges where herbivores take advantage of the elevation gradient to forage on high-quality food for a longer period of time than on plains (Reitan 1988), thereby improving the average quality of individual HRs.

Table 3. Estimated slopes (estimate ± SE and associated P values) for the relationship between HR size and day length, temperature and NDVI for each study site at weekly and monthly scales, obtained from three separate models. Each variable was scaled to allow the comparison of slopes among models
ScaleStudy siteDay length ± SEP valueNDVI ± SEP valueTemperature ± SEP value
WeeklyI:38°N−0·09 ± 0·0440·0410·062 ± 0·0520·2300·093 ± 0·0380·015
I:41°N0·034 ± 0·0320·287−0·033 ± 0·0390·3870·077 ± 0·0330·022
F:43°N−0·289 ± 0·009<0·001−0·301 ± 0·015<0·001−0·239 ± 0·009<0·001
I:46°N−0·072 ± 0·017<0·001−0·113 ± 0·016<0·001−0·058 ± 0·0180·002
D:49°N−0·111 ± 0·014<0·001−0·102 ± 0·014<0·001−0·08 ± 0·015<0·001
S:58°N−0·057 ± 0·016<0·001−0·092 ± 0·023<0·001−0·041 ± 0·0250·097
N:61°N−0·023 ± 0·010·0280·005 ± 0·0120·6930·014 ± 0·0170·414
MonthlyI:38°N−0·296 ± 0·084<0·001−0·192 ± 0·0880·03−0·192 ± 0·0690·005
I:41°N−0·029 ± 0·060·632−0·028 ± 0·0690·6860·007 ± 0·0580·901
F:43°N−0·305 ± 0·018<0·001−0·22 ± 0·029<0·001−0·278 ± 0·017<0·001
I:46°N−0·06 ± 0·030·046−0·125 ± 0·029<0·001−0·079 ± 0·0340·019
D:49°N−0·146 ± 0·028<0·001−0·084 ± 0·0280·003−0·146 ± 0·032<0·001
S:58°N−0·089 ± 0·0280·002−0·107 ± 0·0450·017−0·112 ± 0·0440·01
N:61°N−0·009 ± 0·020·6550·042 ± 0·0240·0830·058 ± 0·0340·088

Breaking down seasonality: the day length component

Seasonal variation in HR size was observed in all of our seven study sites across Europe (Fig. 3). The amplitude of the seasonal variation and the timing of maximum and minimum HR size, however, varied considerably among sites and between time-scales (Fig. 3). Day length has been repeatedly reported to correlate with variation in HR size (Börger et al. 2006a; Rivrud, Loe & Mysterud 2010; van Beest et al. 2011). Day length may affect individuals directly, by modifying their activity (Bradshaw & Holzapfel 2007), or indirectly, via the lengthening or shortening of the growing season. Indeed, day length is the main environmental cue of seasonality for animals and is the driving force behind optimal timing of foraging or reproduction (Visser et al. 2010), particularly if temporal variation in forage availability or energy needs match day length variation. Photosynthetic activity of plants is strongly modulated by the duration of leaf exposure to light (Lawlor 1995). Day length may thus determine how much forage is available to herbivores and, hence, indirectly, their HR size. Indeed, McLoughlin, Ferguson & Messier (2000) used seasonality as a surrogate of habitat quality and showed that increases in food abundance or decreases in variation of food availability may directly affect HR size by allowing individuals to obtain sufficient energy to meet their requirements when ranging over smaller areas.

Figure 3.

Temporal variation of average HR size in female roe deer across seven study sites in Europe (see Fig. 1 for site locations) at (a) the weekly and (b) monthly scales. The time series starts on the 22nd of December. We represented the three best models of temporal variation in HR size, each with a simple cyclic and symmetrical parametric model including time as a sine/cosine function (black line), and then replaced time with two descriptors of the environment, the day length (gray line) or the temperature (dotted line) to explain the observed pattern of HR size variation over time. The gray band represents the summer solstice that occurs on June 21st.

In our study, day length explained a substantial part of the observed seasonal variation in HR size of female roe deer at both temporal scales (63·4% and 71·8% at weekly and monthly scales, respectively). However, although day length captures most of the variability in temperature, the predicted variation of HR size in relation to temperature lags several weeks behind the observed pattern (Fig. 3). Day length had virtually no effect on HR size at the extreme edges of the roe deer's geographical range in Europe, and the observed timing of the minimum HR size did not match the summer solstice, occurring slightly earlier at high and low latitudes compared with mid-range latitudes. The observed deviations of HR size from predictions based on day length suggest a role for plant phenology, resulting from a combination of temperature and precipitation, which does not strictly follow day length, or, alternatively, imply physical constraints on deer movements. This also fits with the interpretation of reduced activity of animals at high temperature (Beier & McCullough 1990). The positive association between temperature and HR size in southern Italy, where plant productivity is mainly limited by summer temperature and water stress (Minder 2012), is also consistent with an effect of primary production. Overall, our results could support the prediction that species are limited at the northern edges of their range by abiotic conditions (e.g. snow conditions), whereas at the southern range, they are limited by biotic interactions (plant productivity), including predation and competition (Ferguson 2002).

Seasonality and the climate amplitude

The marked difference in seasonality of HR size across sites (Fig. 3) seems to be the outcome of different cost-to-benefit ratios of movement for deer exposed to differing climatic regimes. We expected the seasonal pattern of HR size variation to show the greatest amplitude for the most seasonal climate, that is, in study sites located in the Alps at high altitude (Italy: I:46°N) and in northern Europe at high latitude (Norway: N:61°N, Sweden: S:58°N). Instead, we found that seasonality was stronger for study sites located at intermediate altitudes and latitudes (France: F:43°N; Germany: D:49°N) at both temporal scales. The constraints on animal movement imposed by deep snow in winter or the occurrence of artificial winter feeding may explain the low seasonality of HR size at the highest latitude (Guillet, Bergström & Cederlund 1996), as the energy costs of locomotion increase with snow depth (Parker, Robbins & Hanley 1984). Individuals, hence, minimize movements during snowy periods, which translates into smaller HRs in winter (Rivrud, Loe & Mysterud 2010; van Beest et al. 2011). Although we did not directly test for the effect of snow depth, our results suggest that the costs of locomotion for roe deer at high latitudes were such that the size of the winter and the summer HRs are similar, as seasonality in HR size was no longer apparent. Similarly, roe deer living at the southernmost latitude (I:38°N & I°41°N) showed seasonal variation in HR size of a very limited amplitude. The same compensation mechanism could also be at work here, but with a shift towards larger HRs in summer, when food abundance is limiting (due to drought) and female roe deer increase their foraging area.

The average HR size of roe deer increased with the predictable component of seasonality at both the weekly and monthly scales (Fig. 2), that is, with the amplitude of seasonal variation in temperature, even when seasonal variation in HR size was accounted for. A larger HR in response to predictable environmental variation is expected from theoretical models of foraging (Abrams 1991). Strongly seasonal environments at high elevation or latitude represent harsher conditions for roe deer. Remarkably, the average HR size of roe deer also increased with the unpredictable component of seasonality, that is, with the deviation of the observed temperature from its long-term average (Fig. 2). This suggests that roe deer may increase the size of their HR when faced with unpredictable variation of temperature (a proxy of plant productivity) to minimize temporal variation in food resources within their HR (Ferguson et al. 1999). Indeed, as HR size increases, the expected variance in habitat quality in space and time decreases (Wiens 1989). Possible fitness costs of maintaining large HRs may therefore arise in the face of increasing unpredictability of the environment, as predicted by current forecasts of climate change (IPCC 2007).

Conclusion

We showed that the effects of food availability and weather are pervasive on roe deer ranging behaviour across populations and spatiotemporal scales. Seasonal variation in HR size of roe deer was similar in all studied populations across the European continent, reflecting a consistent behavioural response to seasonality in the availability of food resources. Day length is arguably the most predictable and predictive proxy of seasonal variation in forage, with strong links to the deer's ranging behaviour. In comparing the HR size of roe deer from the northern to the southern limits of their distribution, we found that food availability, local weather and both predictable and unpredictable components of seasonality, all of which will be affected by global change, have marked effects on ranging behaviour. The European climate will become warmer and wetter in the North and wetter in the South, and the frequency of extreme events will increase in the near future (IPCC 2007). The expected increase in temporal variability of weather should, thus, lead to an increase in HR size for large herbivores such as roe deer, with possible increased energy needs and potentially deleterious effects on life histories.

Acknowledgements

This paper has been conceived and written within the EURODEER collaborative project (paper n. 02 of the EURODEER series; www.eurodeer.org). The co-authors are grateful to all members for their support for the initiative. The EURODEER spatial data base is hosted by Fondazione Edmund Mach. GPS data collection of the Fondazione Edmund Mach was supported by the Autonomous Province of Trento under Grant N. 3479 to F.C. (BECOCERWI – Behavioural Ecology of Cervids in Relation to Wildlife Infections). For Norway, we thank the Research Council of Norway and the Norwegian Directorate for Nature Management, as well as the Offices of Environmental Affairs from Buskerud, Telemark and Vestfold counties. For Sweden, we thank the land owner the Silfverschiöld family, the Swedish Environmental Protection Agency Wildlife Research fund, Swedish Hunters Association Research fund and the private fund ‘Marie-Claire Cronstedts stiftelse’. Financial support for the Bavarian-Czech study site was provided by EU-programme INTERREG IV (EFRE Ziel 3), the Deutsche Bundesstiftung Umwelt (DBU) and the Bavarian Forest National Park Administration, and the Czech part of this research was funded by grant ‘CzechGlobe’ – CZ.1.05/1.1.00/02.0073 of the Czech Ministry of Education. MH would like to thank Martin Gahbauer, Horst Burghart, Lothar Ertl, Michael Penn, Helmut Penn, Füdiger Fischer and Martin Horn for technical support and for capturing the roe deer. NM would like to thank the local hunting associations, as well as numerous co-workers and volunteers for their assistance and, in particular, Bruno Cargnelutti, Jean-Marc Angibault, Bruno Lourtet and Joël Merlet. We are also indebted to Mark Hewison for comments on an earlier version of this manuscript.

Ancillary