Résumé
L’influence de la densité de population sur la répartition du cerf de Virginie dans l’espace en relation avec la biomasse et le couvert disponibles a été mesurée dans 3 blocs formés de 2 enclos contenant respectivement 7.5 (faible densité) et 15 cerfs/km2 (haute densité). La biomasse et le couvert latéral et vertical ont été interpolés par krigeage et les localisations télémétriques des cerfs ont été divisées en 3 périodes quotidiennes. Pendant l’aube et le crépuscule, aux deux densités, l’utilisation de l’espace était positivement reliée à la biomasse de la végétation. À faible densité, la répartition des cerfs était aussi positivement reliée au couvert vertical à l’aube et au crépuscule. Pendant le jour, la distribution des cerfs était positivement reliée à la biomasse mais seulement à haute densité où les cerfs réduisaient aussi l’utilisation d’endroits avec un couvert latéral dense. Cette étude souligne l’influence de la densité de population et de la période du jour sur la répartition des cervidés dans l’espace ainsi que la valeur des expériences en densités contrôlées et de la géostatistique pour mieux comprendre les facteurs qui influencent le comportement d’approvisionnement des grands herbivores.
Abstract
Deer population densities are increasing in many areas of North America and Europe and are strongly affecting species composition and structure of plant communities. The effects of increased density on deer foraging behavior, however, have received little attention. We assessed the influence of population density on deer space use in relation to vegetation abundance and cover. We quantified space use in 3 blocks with 2 enclosures each containing 3 radio-collared deer at densities of 7.5 (low density) and 15 deer/km2 (high density), respectively. Vegetation biomass, canopy and lateral cover were interpolated by kriging and deer observations were divided into 3 periods: dawn/dusk, day, and night. At dawn and dusk, in both densities, space use was positively related to forage abundance. During the day, deer space use was also positively related to forage abundance but only at high density. At low density, habitat use was positively related to canopy cover during dawn/dusk. Deer decreased the use of areas with dense lateral cover during the day at high density, but no relationship was found at low density. Contrarily to our prediction, deer did not use open habitats more frequently at night than during the day. This study underlines the effect of population density and diel periods on the utilization of available resources and the value of controlled density experiments and geostatistics to disentangle the factors affecting foraging behavior of large herbivores.
Introduction
In summer, northern ungulates typically devote most of their time to finding and consuming food (Beier and McCullough 1990). Foraging decisions are thus important to an herbivore’s energy and time budget. When food acquisition is the primary determinant of patch selection, use of habitat patches should be positively related to forage availability (MacArthur and Pianka 1966, McNamara et al. 1993). There are, however, several other constraints such as heat or wind exposure (Belovsky 1981) and predation risk (Berger 1991) that can affect foraging behavior independently of forage characteristics. For cervids, a good foraging site is usually represented by a trade-off between proximity to protective cover and abundance of vegetation (Kotler et al. 1994).
Cover may be divided into two components: (1) the canopy cover corresponds to the projection of the tree crowns to the ground and (2) the lateral cover which is made up of concealing understory vegetation or topography. Lateral cover reduces predation risk and thus time devoted to vigilance (Altendorf et al. 2001) and, in the absence of predators, has been considered to play a “psychological” role in habitat selection related to the ghosts of predators past (Byers 1997, Mysterud and Østbye 1999). Animals are also typically exposed to milder weather conditions (e.g. temperature, wind, precipitations) in closed habitats than in open habitats (Mysterud and Østbye 1999). Open sites generally offer more abundant forage in summer (Hanley 1984), but present a higher risk of predation (Tufto et al. 1996) and higher thermoregulatory costs due to an increased heat load (Beier and McCullough 1990). Cervids thus often prefer feeding near edges of forests and open habitats because edges minimize the trade-off between exposition to predators and/or harsh weather, and forage abundance (Keay and Peek 1980, Tufto et al. 1996).
Clear-cuts provide open habitats that offer abundant food resources to deer and they are usually interspersed with forest stands that present dense canopy cover but low forage availability (Masters et al. 1993). Deer select cutblocks when clearings produce higher food resources than forest stands and if they provide sufficient hiding cover (Lyon and Jensen 1980). Clearings have been used many times to enhance forage production and thus improve deer habitat conditions (Masters et al. 1993). Tierson et al. (1985) found that deer
stopped migrating to traditional winter ranges to feed in recently logged areas but summer home ranges were not modified by the presence of cutblocks.
It has been proposed that population density modulates the trade-off between using habitats rich in forage and habitats rich in cover (Mysterud and Østbye 1999) because population density is generally negatively related to forage abundance (Healy et al. 1997) and increases intraspecific competition (Clutton-Brock et al. 1982). Indeed, Lesage et al. (2000) found that white-tailed deer generally increased the use of forests over agricultural fields when competition for forage in forest stands was low but, as deer density increased, the use of open areas increased, likely because forage was more abundant in open habitats. Another study showed that because of the low availability of forage in the forest due to high population density, deer adapted to feeding in agricultural crops at night (Rouleau et al. 2002). In agricultural landscapes, population density and landscape composition may thus affect the degree to which deer feed on crops. An increase in the use of open habitats, such as agricultural fields, in summer may indicate the low abundance of forage in areas with adequate cover and the impacts of high density on space use (Mysterud and Østbye 1999). Population densities of many cervid populations in North America are rapidly increasing (Côté et al. 2004). Although of high interest, it is unknown whether increasing density is modifying deer behavior, especially in relation to trade-offs between selection for forage or cover in the context of landscapes composed of forests and clear-cuts.
Controlled-browsing experiments have been used for many years to study the foraging behavior of domestic animals and their use is now advocated for wild ungulates (Hester et al. 2000). Controlled-browsing studies generally investigate the effects of different herbivore densities on vegetation abundance and diversity (Tilghman 1989, Hester et al. 2000); however, they can also be very useful tools to understand the effects of population density on the foraging behavior of wild ungulates.
Deer use open habitats more often during the night than during the day (Beier and McCullough 1990, Rouleau et al. 2002) and since their activity peaks at dawn and dusk (Beier and McCullough 1990), the trade-off between using forage or cover patches may
also depend on diel periods (Mysterud et al. 1999). Therefore, our objective was to address the influence of population density on deer space use in relation to vegetation abundance and cover by experimentally controlling population density in large enclosures in which we obtained 2 replicated densities. We also examined how daylight and activity peaks influence the forage/cover trade-off by separating our radiolocations into 3 daily periods (dawn/dusk, day, night). We predicted that deer at high density would use open areas of the enclosures where forage is more abundant independently of cover characteristics. However, deer at low density should use areas of the enclosures in relation to available cover because competition for forage is less influential at low density.
Study area
Anticosti Island (Québec, Canada, 49° 28’ N, 63° 00’ W) is located at the northern fringe of the white-tailed deer range in North America and covers 7,943 km2. Forests are naturally dominated by balsam fir (Abies balsamea), white spruce (Picea glauca) and black spruce (P. mariana). White birch (Betula papyrifera) and trembling aspen (Populus tremoloides) are irregularly found on the island. About 220 deer were introduced on the island at the turn of the 19th century. In the absence of predation, the population spread and grew rapidly. Today, deer densities of >20 deer/km² are found in most areas on the island (Potvin and Breton 2005). Deer have modified the original forest and greatly reduced the abundance of deciduous woody vegetation on the island (Potvin et al. 2003, Tremblay et al. 2005). The climate of Anticosti is maritime and characterized by longer and milder winters compared to the white-tailed deer range on the continent (Huot 1982). Mean temperatures are -12°C in January and 15°C in July, snow precipitation averages 406 cm annually and rainfall 630 mm (Environment Canada 1993).
Methods
Experimental design
Our experimental design consists in 3 sets of enclosures (A, B, C) in which we introduced 24 deer during 2 different years. Enclosures were located in balsam fir dominated forests that were partially cut in the early summer of 2001. Water was easily accessible to deer at
many streams and artificial water holes in every enclosure. One block (A) was erected in the western part of the island near Simonne Lake and two blocks (B and C) were erected 130 km to the east in the central part of the island near the Jupiter River. Between 30 and 40% of residual forest stands of different sizes (0.19 ha - 21.6 ha) were left in the enclosures. In 2002, one site (A) was studied and in 2003, the 3 sites were studied. To test the influence of deer density, the blocks were divided in two enclosures to obtain densities of 7.5 deer/km2 (40 ha enclosure with 3 deer; LDE) and 15 deer/km2 (20 ha enclosure with 3 deer; HDE). We used different animals in 2002 and 2003 (Table 3-1).
Deer captures
In late June, we fitted 6 deer in 2002 and 18 deer in 2003 with VHF collars (LMRT series) equipped with sto-2a variable pulse activity sensors (Table 3-1). We used different methods to capture deer: dart guns, Stephenson box traps and cannon nets baited with cattle feed and balsam fir twigs. Deer were released in the study enclosures shortly after capture. The Animal Care and Use Committee of Université Laval, Québec, Canada (2005-008) approved all capture methods. We verified reproductive status of adult females by direct observation at capture and at the end of summer. Since only 2 monitored females had a fawn, we did not include reproductive status in the analyses.
Telemetry
In July and August 2002, we radiotracked 6 deer in block A and in July and August 2003, we radiotracked 16 deer in the 3 blocks (Table 3-1). One adult male lost its collar and 1 yearling male had a malfunctioning collar. Deer were located with telemetric receivers (SRX-400 version W9, Lotek Engineering, Newmarket, Ontario, Canada and a TR-2 scanner/receiver, Telonics, Meza, Arizona, USA), unidirectional antennas and compasses. Telemetry stations were positioned with a GPS Garmin (Garmin international, Olathe, Kansas, USA; precision of <5 m) on trails adjacent to the enclosures. To limit human disturbance, trails were generally located more than 100 m away from the enclosures. At least 3 azimuths differing by a minimum of 30º were obtained by moving between stations with a vehicle (White and Garrott 1990). To reduce location error, positioning had to be completed within 15 minutes (White and Garrott 1990). 24-h days were evenly divided
Table 3–1. Number of locations recorded in each diel period for radiocollared white-tailed deer tracked in controlled-density enclosures on Anticosti Island, Québec.
Number of locations
ID Year Block Density Age Sex Dawn/dusk Day Night
(deer/km²) 1 2002 A 7.5 Adult Male 71 124 55 2 Yearling Male 68 125 56 3 Yearling Female 52 108 37 4 15 Yearling Female 66 117 62 5 Yearling Female 69 123 58 6 Yearling Female 73 125 56 7 2003 A 7.5 Adult Female 24 46 15 8 Yearling Female 20 43 21 9 Adult Female 22 50 18 10 15 Adult Female 27 47 19 11 Adult Male 25 38 18 12 Yearling Male 0 0 0 13 2003 B 7.5 Yearling Female 21 44 16 14 Yearling Male 19 48 17 15 Adult Female 20 42 17 16 15 Adult Female 19 46 21 17 Yearling Female 17 48 16 18 Adult Female 22 46 16 19 2003 C 7.5 Yearling Female 24 41 22 20 Adult Female 28 43 21 21 Yearling Male 22 34 26 22 15 Adult Male 24 43 21 23 Yearling Female 21 46 21 24 Adult Male 0 0 0
into 3 periods of 8 hours (8h00–16h00, 16h00-0h00 and 0h00–8h00). These 8-hour periods were evenly sampled and rotated between 2 observers and between groups of enclosures every 3 days. During sampling periods of 8 hours, deer were positioned about every 2 hours.
LOAS software (Location Of A Signal Version 2.07, Ecological Software Solutions, Schwägalpstrasse, Urnäsch, Switzerland) was used to estimate positions and error polygons. Error polygons were calculated with “Andrews” estimator. All locations were plotted with LOAS software on maps and were assigned Universal Transverse Mercator (UTM) coordinates. The average error from plotted to actual locations was determined by using control transmitters set at known locations throughout the enclosures and was estimated at 107 m (SE = 88 m; n = 88 trials). We removed locations with error polygons greater than 0.01 ha. After processing, we kept 2,916 locations from the 3,251 original locations. Positions were assigned to 3 diel-periods: dawn and dusk (1h30 before sunrise to 1h30 after sunrise, and 1h30 before sunset to 1h30 after sunset), day (1h30 after sunrise to 1h30 before sunset), and night (1h30 after sunset to 1h30 before sunrise).
Biomass and cover sampling
To characterize uniformly the vegetation of the enclosures, 5 sampling points were drawn with the « Generate-randomly distributed points » extension of ArcView GIS (ArcView GIS Version 3.1, Environmental systems research institute, Redlands, California, USA) in every 2 ha squares of a grid superposed to each enclosure. We found the sampling stations in the field with a GPS Garmin. At each sampling point, percent of plant cover was estimated in two 1-m2 quadrats randomly chosen in a 10×10 m quadrat centered at the sample point. In block A, the same sampling points were used in 2002 and 2003.
Plant biomass was estimated for every major plant component of deer diet and the most abundant species on Anticosti (Huot 1982) using regressions between percent of plant cover and mass of the corresponding dried plant biomass (Bonham 1989). We analysed the following species: Abies balsamea, Betula papyrifera, Cirsium spp., Coptis groenlandica, Cornus canadensis, Epilobium angustifolium, Grass sp., Hieracium sp., Maianthemum
canadense, Picea glauca, Rubus idaeus, Rubus pubescens and, Trientalis borealis. The number of samples needed for regressions was estimated empirically by plotting regression coefficients with number of samples until an asymptote was reached (Frontier 1983). We summed the biomass values of all plants for each quadrat and used the mean value of the two quadrats for each sample point in the analyses.
At each sampling point, canopy cover was estimated by vertically projecting foliage (>4 m trees) to 20 points distributed equally on the ground every 3 m in four directions (east, southeast, southwest and west) from the center of the sampling unit. Each point was judged as with cover or not and canopy cover corresponded to the sum of all sampled directions (value of 1 for each point with cover). Lateral cover was measured with a cover board (2.5 m×0.3 m divided in 0.5 m sections) in 2 opposite directions by attributing board concealment to 4 classes (1: 0-25; 2: 26-50; 3: 51-75; 4: 76-100%; Nudds 1977). We used the mean value from the first two sections of the board (0-1 m) and values from both directions were averaged.
Analyses
Data points could not account for spatial relationships and spatial correlation between biomass abundance and cover. We thus estimated biomass and cover abundances with a geostastistical software (Geostatistical analyst; ArcMap 9.0, Environmental systems research institute, Redlands, California, USA). Geostatistics is a branch of applied statistics that focuses on the detection, modeling, and estimation of spatial patterns in spatially correlated data (Rossi et al. 1992). Spatial autocorrelation occurs because samples collected closer to each other are more similar than samples collected farther apart. This particularly occurs when the variable sampled is spatially structured (e.g. in patches). In their simplest form, geostatistics involve 2 steps: 1) characterizing the spatial structure of the variable with a variogram, thus defining the degree of autocorrelation between the data points; and 2) predicting values between measured points based on the estimated degree of autocorrelation (Robertson 1987). Semivariance is the average measure of the variance associated to any two sampled points in a given distance class. For example, with a lag size of 50 m, a mean variance value would be obtained for each distance class (0-50 m, 51-100
m, 101-150 m, etc.). In spatially structured data, changes of semivariance represented in a variogram usually augment with increasingly separated points (Cressie 1993). Variograms may also include a directional value to weight for directional changes in the factor (e.g. slope direction; Cressie 1993). The semivariogram models provide the following parameters: 1) the nugget effect which indicates the residual spatial variability below the lag size that cannot be modeled with the current sampling resolution, 2) the sill, which defines the asymptotic value of semivariance; and 3) the range, defined as the distance over which autocorrelation is present (Cressie 1993).
Semivariograms were prepared individually for forest stands and cuts of each enclosure because vegetation and cover drastically change between these two habitats (Masters et al. 1993). Biomass values were log-transformed to normalize data. By trial and error, we determined the best fit of spherical variograms until a maximum lag distance of 125 m (i.e., half of the minimum enclosure dimension) in 5-m increments and in all directions (Jurado- Expòsito et al. 2004). The best-fitted values, determined by cross-validation results, of nugget, sill and range were calculated and recorded for further analysis in ordinary kriging (Cressie 1993). Ordinary kriging is an interpolation technique that uses observed values associated with X and Y coordinates and estimates values for all locations within the sampled coordinates with the help of a variogram describing how values change through space (Cressie 1993).We used kriging values to validate the fitted variogram through cross- validation. This procedure is based on the systematic removal of observations, one by one, from the raw data set, which is then estimated by kriging (Isaaks and Svriastava 1989). Kriging provides an error term for each estimated value, thus giving a measure of reliability for the interpolations. Biases in estimation errors were evaluated using the standardised root mean squared error (Appendix 3−1; RMSE; Isaaks and Svriastava 1989). The nugget value divided by the total variance (sill) gives an estimation of the spatial dependence (Appendix 3−1; Jurado-Expòsito et al. 2004).
Rettie and McLoughlin (1999) recommended the use of buffers to account for telemetry error in habitat selection studies. In our study, the error of locations obtained by telemetry was 107 m on average and thus larger than most forest stands present in the enclosures. In
70% of the trials with hidden collars at known locations, collars were located less than 100 m away from the real position. Furthermore, up to 100 m, the probability of obtaining a position closer to the real position was not a function of distance (F1,9 = 0.11; P = 0.75). We thus took into account the mean location error by placing a 100 m buffer around each location. Since the buffers were quite large compared to the size of the enclosures, they overlapped considerably and were thus dependent on one another. To take this into account, we randomly placed 1 point in every 150×150-m square of a grid placed over each enclosure with the extension “Simple random sample” (ArcView GIS, Version 3.1, Environmental systems research institute, Redlands, California, USA). We used 150×150- m squares because they were sufficiently large compared to the buffer size and allowed for a reasonable number of sampling points for the regression (20-30 points). For each of the 3 diel periods, we counted the total number of overlapping buffers at each point for each deer with the extension “dissect overlaps” (ArcView GIS, Version 3.1, Environmental systems research institute, Redlands, California, USA). We then divided the total number of overlapping buffers for each random point by the total number of locations for this deer during each diel period to describe a relative use of the enclosures that was independent of the number of positions taken on an individual.
Resources and relative use (mean relative number of overlapping buffers per diel-period) were evaluated at random points distributed across the total surface of the enclosures. We estimated relationships between relative use and resource abundance by means of regression analyses. We did not consider resource availability in the models because a random and independent sample of points were already considered in the analysis. For each diel period, the relationship between relative use for all deer in a particular enclosure and biomass, lateral and canopy cover was quantified using a linear mixed model with block and year as random factors (Proc Mixed, SAS Version 9.1, SAS Institute Inc., Cary, North