Access to mates in a territorial ungulate is determined by the size of a male's territory, but not by its habitat quality

Authors

  • Cécile Vanpé,

    1. Laboratoire Comportement et Ecologie de la Faune Sauvage, Institut National de la Recherche Agronomique, BP 52627, F-31326 Castanet-Tolosan cedex, France;
    2. Grimsö Wildlife Research Station, Department of Ecology, Swedish University of Agricultural Science (SLU), S-73091 Riddarhyttan, Sweden; and
    Search for more papers by this author
  • Nicolas Morellet,

    1. Laboratoire Comportement et Ecologie de la Faune Sauvage, Institut National de la Recherche Agronomique, BP 52627, F-31326 Castanet-Tolosan cedex, France;
    Search for more papers by this author
  • Petter Kjellander,

    1. Grimsö Wildlife Research Station, Department of Ecology, Swedish University of Agricultural Science (SLU), S-73091 Riddarhyttan, Sweden; and
    Search for more papers by this author
  • Michel Goulard,

    1. DYNAFOR UMR 1201, Institut National de la Recherche Agronomique, BP 52627, F-31326 Castanet-Tolosan cedex, France
    Search for more papers by this author
  • Olof Liberg,

    1. Grimsö Wildlife Research Station, Department of Ecology, Swedish University of Agricultural Science (SLU), S-73091 Riddarhyttan, Sweden; and
    Search for more papers by this author
  • A.J. Mark Hewison

    Corresponding author
    1. Laboratoire Comportement et Ecologie de la Faune Sauvage, Institut National de la Recherche Agronomique, BP 52627, F-31326 Castanet-Tolosan cedex, France;
      Correspondence author. E-mail: mark.hewison@toulouse.inra.fr
    Search for more papers by this author

Correspondence author. E-mail: mark.hewison@toulouse.inra.fr

Summary

  • 1Territoriality is commonly associated with resource defence polygyny, where males are expected to gain access to females by anticipating how resources will influence female distribution and competing for resource-rich sites to establish their zone of dominance.
  • 2We tested this hypothesis in European roe deer (Capreolus capreolus) by simultaneously assessing the influence of resources on female distribution and the influence of female distribution on male distribution and breeding success using paternity analyses.
  • 3Females did not fully distribute themselves among male territories in relation to resources. As a result, relative female abundance in a male's territory depended on territory size, but not on its habitat quality. In turn, relative female abundance in a male's territory determined, at least partially, his breeding success.
  • 4Interestingly, male territory size, and hence access to females, was partly determined by male body mass (all males) and by residual antler size (subadults only). The latter result suggests that large antlers may be important to young males for establishing their first territory, which is then usually retained for all subsequent reproductive seasons.
  • 5To conclude, although territoriality of male roe deer has certainly evolved as a tactic for ensuring access to mates, our results suggest that it does not really conform to a conventional resource defence polygyny strategy, as males seem to gain no obvious benefit from defending a territory in an area of high habitat quality in terms of enhanced access to mates.
  • 6This may explain the stability of male territories between years, suggesting that male territoriality conforms to an ‘always stay’ and ‘low risk–low gain’ mating strategy in roe deer.

Introduction

In mammals (especially in ungulates), where males usually provide no parental care and have the higher reproductive potential, female reproductive success is directly constrained by access to the resources necessary to breed and to meet the energy requirements of gestation and lactation, whereas male reproductive success is limited more by access to potential partners (Trivers 1972). As a result, female distribution should depend primarily on resource dispersion (modified by predation pressure and the costs and benefits of associating with other individuals), while males should distribute themselves in relation to females (modified by male density), at least during the breeding period, in order to gain access to as many mates as possible (Emlen & Oring 1977).

Males can either directly compete for mates, or indirectly compete for resources that influence female distribution (Emlen & Oring 1977). The economics of these two mating tactics depends mainly on the defensibility of females, which is directly related to female density, distribution, range size, group size and stability, and indirectly to the distribution of resources (Clutton-Brock 1989). When females are spatially and temporally predictable (e.g. when they occupy small stable ranges containing clumped and high-quality food resources), the most economic mating tactic should be to adopt a ‘resource defence polygyny’ tactic, where males gain access to females by anticipating how resources will influence female distribution and competing for resource-rich sites to establish their zone of dominance (Emlen & Oring 1977).

Male territoriality is commonly associated with the resource defence polygyny tactic (Emlen & Oring 1977). In ungulates, in which male territoriality is widespread (see Owen-Smith 1977), the evolution of male territoriality as a mating tactic analogous to resource defence polygyny has been suggested for various species (e.g. in Grevy's zebra Equus grevyi, Klingel 1974; European roe deer Capreolus capreolus, Putman 1988; Wahlström 1994; Johansson 1996; Liberg et al. 1998; hippopotamus Hippopotamus amphibius, Nowak 1991; wildebeest Connochaetes gnou, Estes 1969; Reeves's muntjac Muntiacus reevesi, Dubost 1970; sika deer Cervus nippon, Miura 1984). However, no study has as yet tested this hypothesis by simultaneously assessing the influence of resources on female distribution and the influence of female distribution and resource dispersion on male distribution and reproductive success.

The aim of this study was to investigate this issue in the European roe deer [C. capreolus (Linnaeus 1758)], an ungulate in which only males are territorial and the spatial system of the two sexes is independent, so that a female's home range may overlap several male territories (Hewison, Vincent & Reby 1998; Liberg et al. 1998). First, we expected female abundance within a male's territory to increase with territory size (prediction 1). Second, if male territoriality evolved as a resource defence polygyny mating tactic, females should be distributed in relation to resources. The ‘ideal free distribution’ hypothesis (IFD; Fretwell & Lucas 1970) states that, in the absence of constraints on movement, individuals are expected to be distributed so that differences in local densities reflect differences in habitat quality. As a result, resources are equally shared among individuals and fitness is equal in all habitats (Fretwell & Lucas 1970). While Wahlström & Kjellander (1995) suggested that roe does fit the IFD, Nilsen, Linnell & Andersen (2004) found evidence for fine-scale spatial variance in fitness components induced by environmental heterogeneity. Similarly, Pettorelli et al. (2001) reported higher fawn body weight in the richer northern part of the Chizé forest compared to the poorer south, despite higher local density in the north. Assuming that females are free to distribute themselves in relation to resources, we expected female abundance within a male's territory to increase with habitat quality for a given territory size (prediction 2). Third, territoriality should secure mating access to resident females for the territorial male. Hence, we expected a male's breeding success to depend on female abundance within his territory (prediction 3). Also, males should attempt to enhance their mating success by either enlarging their zone of dominance, or by occupying areas particularly favoured by females, or both. Hence, we also expected male breeding success to increase with territory size and/or territory habitat quality (prediction 4). Finally, we should expect substantial male–male competition to occupy the ‘best’ territories (providing access to the highest numbers of females). Hence, the best territories should be occupied by the best competitors in male–male contests. We therefore expected a positive correlation between the suitability of the territory in terms of access to females, and male attributes which reflect ability to fight and dominate, such as age, body mass and antler size (prediction 5).

Materials and methods

study species and study site

The roe deer is a selective feeder, preferentially feeding on forbs, seeds and deciduous browse rich in soluble nutrients (Tixier & Duncan 1996), adapted to exploit the early stages of forest succession (Linnell, Duncan & Andersen 1998; Hemami, Watkinson & Dolman 2004). It is an income breeder (Andersen et al. 2000), relying on food intake rather than body reserves to cope with the marked increase in energy requirements during late gestation–early lactation. Females are considered as nonterritorial, living solitarily or with their fawns during summer in overlapping home ranges (Bramley 1970; Strandgaard 1972; Wahlström & Kjellander 1995). Adult males are considered as fully territorial, but only from early spring (March–April) until late August–early September, just after the July–August rut (e.g. Bramley 1970; Strandgaard 1972). Although males become sexually mature as yearlings, they usually do not defend territories until 2, 3 or 4 years of age (see Liberg et al. 1998 for more details).

The study was carried out in the 1420-ha western part of Bogesund (59°23′ N, 18°15′ E), a 2600-ha mainland peninsula situated on the coast of the Baltic Sea on the inner portion of the Stockholm Archipelago, within the hemiboreal zone of east-central Sweden. The habitat is fragmented, with approximately 65% woodland, 25% fields and 10% bed rocks and bogs (Kjellander 2000; see also Supplementary Appendix S1). The only natural predator of roe deer fawns is the red fox (Vulpes vulpes, Kjellander & Nordström 2003).

data collection and analyses

The Bogesund population has been monitored intensively since 1988. Each winter, roe deer were caught in box traps, sexed and individually marked with plastic ear tags (see Kjellander 2000 for more details). A variable number of caught individuals were also equipped with radio collars (TXT-2Sm, 151 or 152 MHz transmitters, Televilt International AB, Lindesberg, Sweden). Age was either known definitively (animals first caught < 1 year old) or estimated from tooth eruption and wear (Cederlund, Kjellander & Stålfelt 1992) during capture or after death. As age estimation based on tooth wear is inaccurate in roe deer (Hewison et al. 1999), we considered two age classes only, subadults (2 years old; all known-age) and adults (> 2 years old; both estimated and known-age males). Body mass was measured to the nearest 0·1 kg, while antler size was measured to the nearest 0·5 cm and standardized to the 14th of February (see details in Vanpéet al. 2007).

Radio tracking and estimation of territory size

All deer equipped with radiocollars and older than 1 year of age were intensively radio tracked during the ruts (i.e. from mid-July to mid-August) of 2004 (N = 25 males and 16 females) and 2005 (N = 23 males and 22 females, of which 15 and 9 were the same as in 2004, respectively). Radio tracking was conducted from the ground with Televilt RX-810 and RX-98 receivers and a four-element Yagi antenna. Animals were located by triangulation from at least three points along roads or hills. An experimental study of the accuracy of our radio-tracking fixes showed that the mean error ± SD of localizations was 41·0 ± 41·6 m (see Appendix S1). Based on graphs of area against 90% Kernel probability values (see Appendix S1), we chose to collect 30 fixes per individual and per rut, which is the recommended standard proposed by Kenward (1987) and Seaman et al. (1999). The 30 positions for each individual were collected over the same number of days, with up to a maximum of three fixes per 24-h period, both during daytime (8 positions/individual), and during night (2 positions/individual), dusk (10 positions/individual), and dawn (10 positions/individual), and with a minimum of 3 h between successive locations to limit temporal autocorrelation.

Radio-tracking positions were digitized in a geographical information system (arcview 3·2) and analysed using the spatial analyst and animal movement extensions of arcview (Hooge & Eichenlaub 1997). Range size was estimated using the fixed Kernel method (Worton 1989) with both the 90% and the 50% isoclines and the least-squares cross-validation (LSCVh) calculation for the smoothing parameter (see Appendix S1), as recently recommended by Börger et al. (2006). Based on measures of range overlap, we showed that a territorial male had relatively exclusive use of his 90% Kernel area (see Appendix S1), hence 90% Kernel during the territorial period can be considered as a reliable proxy of territory size.

Pellet-group counts for estimating an index of relative female abundance in each territory

A faecal pellet-group count census (Neff 1968; Cederlund & Liberg 1995) was carried out over the entire 2600-ha Bogesund peninsula, both in 2004 and 2005, in early April, immediately after snow melt. The 604 sample plots (10 m2) were circular and distributed every 100 m along north-south transects, with 400 m between neighbouring transects. We considered that no pellet degradation occurred during winter due to snow cover and low temperatures. Because new pellets that have deposited since defoliation could be distinguished from old pellets in relation to dead leaves and needles that covered the latter but not the former (Kjellander 2000), a standing crop design was used. The total number of pellet groups (minimum of 10 pellets per group, which is slightly more than half the mean number of pellets in a group) was recorded on each plot.

The number of pellet groups on each plot was interpolated using a statistical prediction (see Appendix S1). The total number of pellet groups was estimated within each male's territory (based on both 90% and 50% Kernels) for 2004 and 2005 as the sum of the point estimates falling within each territory using arcview 3·2 software. This value was used as a relative index of local roe deer abundance. It was corrected using an estimation of the local sex ratio, defined as the proportion of mature females caught in each trap zone (N = 7) during winter captures, in order to obtain an index of relative abundance of mature females in each territory.

Vegetation sampling and identification of the specific resource attributes that reflect habitat quality for females

During both ruts, a systematic vegetation sampling was performed over the 1420 ha of the study site, comprising 357 point locations. For each point, the habitat was classified and vegetation cover was evaluated (see Appendix S1). Vegetation sampling was carried out using a wooden frame of 25 × 25 cm. All vegetation inside the frame was identified to the species level and cut at the following heights: 0–20 cm, 20–50 cm and 50–150 cm (maximum browsing height for a roe deer). Each sample was labelled and stored frozen in a paper bag. All samples were later thawed, dried at 60 °C for 3 days and then weighed to the nearest 0·01 g. For each sampling point, the biomass of crop plants (i.e. oat, wheat, and barley), berry plants (i.e. Vaccinium spp. and Ribes spp.), pasture plants (i.e. Trifolium spp. and Vicia spp.), various grasses, and the total biomass of consumed plants were then estimated. Plants consumed by roe deer (see Appendix S1) were determined based on the analysis of 11 rumen contents from two females and nine males shot during the 2005 rut at Bogesund (Cécile Vanpé and Petter Kjellander, unpublished data).

Each of the 10 habitat variables (i.e. five factors: habitat, cover 0–20 cm, cover 20–50 cm, cover 0–150 cm, cover of bilberry; and five continuous variables: biomass of crop plants, biomass of berries, biomass of pasture plants, biomass of grasses, total biomass of consumed plants) was entered in a geographical information system (arcview 3·2), mapped, and interpolated over the whole study area. To identify the specific resource attributes that best reflect habitat quality (in terms of attractiveness for females), habitat selection by females during the rut was evaluated with an ecological-niche factor analysis (ENFA; Hirzel et al. 2002), using the enfa function of the adehabitat package (Calenge 2006) for the r 2·4·1 statistical software (R Development Core Team; Ihaka & Gentleman 1996). Details on the ENFA are provided in Appendix S1. To determine the specific resource attributes that best reflect the suitability of an area for females, we interpreted the marginality axis of the ENFA (see Appendix S1), which measures the strength of habitat selection (i.e. the mean difference between habitat use and availability; see Hirzel et al. 2002 for further details). This analysis indicated that females tended to avoid the intermediate age forest habitat with high cover of bilberry (negative values), while they selected crop field habitats with high biomass of crop plants, and high vegetation cover (positive values). A habitat suitability map was derived from the results of the ENFA by assigning a numerical value to each raster map unit of the study area based on its position in this ecological space. This map was then used to derive mean habitat quality within each male's territory (both 90% and 50% Kernels).

Paternity analysis and estimation of male breeding success

From 1988 onwards, tissue samples were collected for DNA genotyping from all individuals caught for the first time and unmarked shot roe deer. From 1997 onwards, tissue samples were also collected from neonates caught by hand in springtime, just after birth (May–June). This allowed us to sample a total of 1757 different individuals. A small (approximately 4 × 4 mm) piece of ear skin tissue was removed using sheep ear-notching pliers. Mother–offspring relationships were elucidated by direct observations of marked fawns with their mothers just after birth or during autumn. Genotyping of the DNA samples was carried out using 21 microsatellite markers (see Vanpéet al. 2008). In this study, we focused on fawns from the cohorts 2005 (N = 29) and 2006 (N = 29). We aimed to identify the father of these fawns in order to estimate yearly breeding success of the males radio tracked during the 2004 and 2005 ruts. Parentage was assessed using a likelihood-based approach with the program cervus 2·0 (Marshall et al. 1998) and the user-defined input parameters detailed in Vanpéet al. (2008). For the 2005 and 2006 cohorts, we identified 57 and 73 sampled candidate mothers respectively and 51 and 63 sampled candidate fathers respectively. In total, we were able to assign 26 fawns to the 31 different radio-tracked males. Yearly breeding success (YBS) was defined as the number of fawns assigned to a male (at the 80% confidence level) in a given year.

statistical analyses

To first describe the spatial scale of variation in pellet-group counts and habitat resources, we constructed Poisson empirical covariograms (Christensen, Moller & Waagepetersen 2000) for the 2004 and 2005 pellet-group count distributions, as was described earlier for interpolation of pellet-group count distributions, and a classical variogram for resource distribution using the geor package (Ribeiro & Diggle 2001).

To test our predictions, we used linear mixed-effects models (either Gaussian or generalized) implemented in the r 2·4·1 statistical software (R Development Core Team 2004), where male identity was entered as a random factor to control for repeated measures of the same individual. Model selection was performed using the small sample size corrected Akaike information criterion (AICc) as recommended by Burnham & Anderson (2002) (see Appendix S1). The territory size (both 90% and 50% Kernels) and the index of relative female abundance in a male's territory were log-transformed to normalize data in models where they were the dependent variables.

To test predictions 1 and 2, we used the lme function (in the nlme r package) for fitting linear mixed-effects models (Pinheiro & Bates 2000), using the maximum likelihood estimation procedure. To identify territory characteristics (territory size, habitat quality) that could reflect territory suitability in terms of access to females for males, the model relating the index of relative female abundance in a male's territory-to-territory size was compared with the constant model. The model describing the relationship between the index of relative female abundance in a male's territory (corrected for territory size) to the mean habitat quality index within the territory was also compared with the constant model.

To relate YBS (discrete response variable) to individual-based measures describing the males’ territory (i.e. relative index of local female abundance in the territory, territory size, mean habitat quality index within the territory) to test predictions 3 and 4, we used a generalized linear mixed model (GLMM) implemented in the glmmADMBR module (H. Skaug, D. Fournier & A. Nielsen, unpublished). YBS was modelled as a Poisson distribution (see McLoughlin et al. 2006 for a similar approach) and we used a zero-inflated model (see Appendix S1). Each model including the simple effect of the index of relative female abundance in the territory, territory size, and the mean habitat quality index within the territory was compared with the constant model.

Finally, to examine prediction 5, we tested the additive effects of male age class (subadults vs. adults), residual antler size (antler size corrected for the allometric relationship between antler size and body mass; see Vanpéet al. 2007) and body mass (standardized using the mean body mass of the corresponding age class), as well as their two-way interactions, on territory characteristics (territory size and/or habitat quality index in a male's territory) which reflect the suitability of territories in terms of access to mates for males. Two separate models (one for antler size and one for body mass) were performed, as both data were not always available for all males.

Results

scale of spatial variation in female abundance and habitat resources

The spatial autocorrelation of pellet-group counts was weak in both years and occurred over about 200 m (225 m for 2004 and 235 m for 2005) (Fig. 1a and b). For resources, the spatial structure was stronger, with spatial autocorrelation occurring over 400–600 m. Hence, the spatial structure of resources occurred over a greater distance than that of pellet-group counts, but both varied at a spatial scale which is of the same order of magnitude as the average home range size of a male roe deer on this study site [mean (± SD) territory size was 23·1 ha (± 13·8) and 29·5 ha (± 12·0) in 2004 and 2005, respectively, based on 90% Kernel; N = 24 for both years].

Figure 1.

Poissonian covariograms of pellet-group count distributions in 2004 (a) and 2005 (b) computed using the georglm package. The minimum and maximum limits were computed from generating 99 simulations by permutating data values.

characteristics determining the relative local female abundance in a male's territory

The models including an effect of territory size on the log-transformed index of relative female abundance in the territory described the data much better than the constant models (ΔAICc = 26·34 and 23·62, for 90% and 50% Kernels respectively). Hence, as expected (prediction 1), as territory size increased, relative female abundance in the territory also increased (mean slope ± SE = 0·040 ± 0·006 and 0·184 ± 0·032, for 90% and 50% Kernels respectively) (Fig. 2a and b).

Figure 2.

Effect of the territory size (a and b) and of the habitat quality index in a male's territory (c and d) on the log-transformed index of relative female abundance in the territory corrected for territory size, based on both the 90% Kernel (a and c) and the 50% Kernel (b and d).

In contrast, the models including an effect of the habitat quality index in a male's territory on the log-transformed index of relative female abundance in the territory corrected for territory size performed less well than the constant model (ΔAICc = 2·43 and 0·85, based on 90% and 50% Kernels respectively). However, AICc values were very similar, especially for the 50% Kernel, indicating some support for a negative effect of habitat quality on relative female abundance in a males’ territory (Fig. 2c and d), contrary to prediction 2. Note, however, that this negative effect was likely due to two out-lying data points, since the ΔAICc was greater than 2 once these points were removed (ΔAICc = 2·36). These results indicate that the suitability of a male's territory in terms of access to females seems to be best reflected by territory size, but not by its habitat quality.

effect of relative female abundance in a male's territory on male yearly breeding success

The constant model performed slightly better than the models including the effect of relative female abundance in a male's territory on male yearly breeding success (ΔAICc = 1·31 and 1·25, based on 90% and 50% Kernels, respectively), but values of AICc were similar, indicating some support for an effect of female abundance on male YBS. There was, thus, a trend for an increase in male YBS with an increase in the index of relative female abundance in their territory described by the equations: E(YBS) = (1–0·40) exp(–0·4624 + 0·0497 index of relative female abundance) based on the 90% Kernel and E(YBS) = (1–0·41) exp(–0·3523 + 0·2201 index of relative female abundance) based on the 50% Kernel (Fig. 3a and b).

Figure 3.

Variation in male YBS with the index of relative local female abundance in the territory based on both the 90% Kernel (a) and the 50% Kernel (b). Note: black open circles, observed data. Grey stars, values predicted by the zero inflation generalized linear mixed model.

effect of territory size and habitat quality on male yearly breeding success

The constant model performed better than the model including the effect of territory size on male yearly breeding success, based on both 90% and 50% Kernels (ΔAICc = 2·34 and 3·67, respectively), indicating the lack of a marked effect of territory size on male YBS. However, there was a slight trend (2 < ΔAICc < 4; see Burnham & Anderson 2002) for an increase in male YBS with male territory size described by the equations: E(YBS) = (1–0·3417) exp(–0·8362 + 0·0266 territory size) based on the 90% Kernel and E(YBS) = (1–0·3456) exp (–0·4801 + 0·0744 territory size) based on the 50% Kernel (Fig. 4a and b).

Figure 4.

Variation in male YBS with male territory size based on both the 90% Kernel (a) and the 50% Kernel (b). Note: black open circles, observed data. Grey stars, values predicted by the zero inflation generalized linear mixed model.

In contrast, the constant model performed much better than the model including an effect of territory habitat quality on male yearly breeding success (ΔAICc = 4·81 for the 90% Kernel and 4·87 for the 50% Kernel, respectively). Therefore, our index of territory habitat quality had no effect on male YBS.

effect of male phenotypic attributes on territory size

The best linear mixed models testing the effect of male body mass (corrected for age class) and age class (subadults vs. adults) on log(territory size) were the constant models (ΔAICc = 0·21 for the 90% Kernel and 2·38 for the 50% Kernel, respectively). Note, however, that the model including the simple effect of body mass had a very similar AICc to the constant model especially for the 90% Kernel, suggesting substantial support for this model too. Hence, body mass had a positive effect on male territory size when measured with the 90% Kernel (mean slope ± SE = 0·05 ± 0·04), but little effect on male territory size when measured with the 50% Kernel (0·01 ± 0·04) (Fig. 5a and b).

Figure 5.

Effect of residual body mass in all males (black open triangles, black line) (a and b, subadults and adults combined) and residual antler size in subadults (grey filled circles, grey line) and adults (black open circles, black line) (c and d) on territory size measured with both the 90% Kernel (a and c) and the 50% Kernel (b and d). Note: residual antler size is calculated as antler size standardized to the 14th of February and then corrected for the allometric relationship between antler size and body mass (see Vanpéet al. 2007). Residual body mass is calculated as body mass corrected for the mean body mass of that age class.

In addition, the best models testing the effect of male residual antler size and age class on log(territory size) were the models including both the additive effects of residual antler size and age class, as well as their interaction, for both 90% and 50% Kernels (ΔAICc = 1·74 and 0·61, respectively). Hence, territory size increased with male residual antler size, but only in subadults (mean slope ± SE = 0·29 ± 0·01 and 0·04 ± 0·01 for 90% and 50% Kernels, respectively), and not in adults (0·002 ± 0·002 and 0·003 ± 0·002 for 90% and 50% Kernels, respectively; see Fig. 5c and d). Hence, these results therefore support, at least partly, prediction 5. Note, however, that the data set included only three subadult individuals and that the effects of antler size and age class were not marked, since the constant models had very similar AICc as the best models.

Discussion

Our results show that female abundance within a male's territory depends on the size of his territory, as expected from prediction 1, but not on its habitat quality, contrary to prediction 2. In support of predictions 3 and 4, territory size and female abundance within the territory, in turn, seem to determine male breeding success, at least partially. Hence, our results do not conform to expectations if male territoriality in roe deer was a conventional resource defence polygyny tactic (sensu Emlen & Oring 1977), because males seem to aim to defend a large territory, irrespective of its attractiveness to females in terms of habitat resources.

We showed that relative female abundance in a male's territory was not related to territory habitat quality, which seems to contrast with predictions under the resource defence polygyny hypothesis. These results also differ from the observations of Johansson (1996) who reported that male mating success in roe deer was related to territory quality, but not territory size: males who occupied a territory which included clover fields were more successful than other males. Strandgaard (1972) also suggested that the quality of crops influenced the number of females on a given territory. However, Johansson (1996) used behavioural estimates of mating success which are certainly poor indicators of true male breeding success, particularly as multiple mating commonly occurs and roe deer are cryptic and often occupy dense forest habitat (Danilkin & Hewison 1996; Liberg et al. 1998). In addition, both studies used a subjective categorical variable to describe habitat quality based on the assumption that clover fields were the preferred resource selected by females. To our knowledge, our study is the first to use a niche analysis to determine the specific resource attributes that reflect the habitat quality of a male's territory in terms of access to females. Several other studies of territorial ungulates have demonstrated a direct relationship between the quality of male territories and access to potential mates (e.g. in American pronghorn Antilocapra americana: Kitchen 1974; puku Kobus vardonii and topi Damaliscus lunatus antelopes: Balmford, Rosser & Albon 1992; impala Aepyceros melampus and waterbuck Kobus defassa: Jarman 1974; in a territorial population of red deer Cervus elaphus: Carranza 1995). However, none of these studies had access to a reliable estimate of male breeding success to fully test the resource defence polygyny mating hypothesis.

The absence of a relationship between the index of habitat quality and relative female abundance in a male's territory observed in our study may be due to several factors:

First, our estimation of relative female abundance within a male's territory was based on the assumption that local female density in winter was a good estimate of local female density in summer, since female home ranges are globally stable between seasons (Hewison et al. 1998). However, some seasonal migrations can occur in yearling and 2- to 3-year-old females (see Wahlström & Liberg 1995). This may potentially result in a redistribution of young females between the winter and rut periods, but concerns only a small proportion of the total number of females. In addition, snow cover may restrict roe deer space use in winter (Cederlund & Lindstroem 1983), but snow cover in our study area is certainly too low to constitute a problem for roe deer ranging behaviour (normal snow cover rarely exceeds 10 cm at Bogesund). To test the reliability of our assumption, we estimated the degree of overlap between winter (January–March) and summer (July–August) home ranges (90% Kernels) of 12 females of 1 year old and older, based on radio-tracking data collected from 1992 to 1995. We found that female home ranges were slightly larger in winter (mean = 42·15 ha, SD = 15·10) than in summer (mean = 25·36 ha, SD = 17·51), and overlap between seasons was variable and quite high, ranging from 17% to 99% of the area of the smallest home range, with a mean of 48·2% (SD = 24·6). This suggests that winter female distribution should be a quite good proxy of female distribution during the rut. Note that in our study, it was not possible to obtain significant behavioural information because of the difficulties of observing roe deer in a fragmented habitat with a large proportion of forest, such as at Bogesund. However, collecting behavioural observations on female distribution, when possible (e.g. in a more open landscape; see Börger 2006), would provide complementary information for understanding male mating tactics.

Second, the use of pellet-group counts to reflect spatial variation in relative deer abundance has not been fully tested at the present time. However, it has been suggested that pellet-group counts provide an informative index of relative animal abundance (e.g. Forsyth et al. 2007), especially in northern areas (such as Scandinavia), where pellet decay rate should be negligible and spatially constant during winter.

Third, although our index of habitat quality integrated a number of habitat variables, it may still not adequately reflect the suitability of an area for females. Fox predation, for example, is known to be a strong determinant of summer fawn survival and variation in female reproductive success at Bogesund (Kjellander et al. 2004a). Hence, predation avoidance may swamp the role of food resources and vegetation cover in determining female distribution. However, a preliminary analysis suggested that the number of fox dens present within a male's territory has no effect on the index of relative female abundance in a male's territory or on male breeding success (Cécile Vanpé, unpublished data).

A fourth possibility is that females may not be free to distribute themselves in relation to resources because of social constraints. Indeed, the IFD model is not appropriate when some individuals can monopolize resources by securing the highest quality areas while forcing inferior competitors into suboptimal habitats (e.g. territoriality or dominance hierarchies). Female roe deer are considered as nonterritorial, except maybe during fawning site defence (Liberg et al. 1998), because they commonly live in overlapping home ranges. However, this overlap often concerns close relatives (see Liberg et al. 1998), with matrilineal associations common due to extensive female philopatry (see Hewison et al. 1998). Hence, we cannot rule out the possibility that dominance relationships between unrelated females impose a constraint on female ranging behaviour. Indeed, female home range size seems to decrease with increasing population density (Kjellander et al. 2004b) and aggressive behaviours between females seem to be common (Börger 2006).

If females do not fully distribute themselves in relation to resources, males will be unable to ‘predict’ the distribution of potential mates by anticipating how resource distribution influences female dispersion. Hence, in contrast to the expectations under resource defence polygyny (Emlen & Oring 1977), males would then show no interest in competing for resource-rich sites to establish their zone of dominance. We showed that, to some degree, local female abundance in a male's territory increased with territory size and, as a result, had a positive influence on male YBS. This suggests that, to maximize fitness, males should attempt to enlarge their territory in order to maximize their chance of encountering receptive females. Interestingly, a male's territory is remarkably stable from year to year (Liberg et al. 1998), with little change of borders and size (e.g. Bramley 1970; Johansson 1996; Linnell & Andersen 1998), even when the distribution of resources or females changes markedly (Strandgaard 1972). Territory borders may even persist after the death of an owner and the takeover of a new occupant (Strandgaard 1972; Cederlund, Kjellander & Liberg 1994), usually an immigrating buck or a nonterritorial subadult (Cederlund et al. 1994; Johansson 1996). These observations suggest that males do not constantly attempt to enhance their mating success by either enlarging their zone of dominance or by occupying areas particularly favoured by females (Hewison et al. 1998), contrary to expectations for a conventional resource defence polygyny strategy.

It should be noted, however, that we did not find very strong relationships between male YBS and territory size and between male YBS and relative female abundance in a male's territory. This could be due to low sample size, but may also be explained by the fact that males probably do not have full mating monopoly inside their territory. Indeed, in roe deer, 30% to 50% of females make short excursions outside their normal home range during the rut, traversing several territories, and potentially mating with other males, before returning (see Liberg et al. 1998; Richard et al. 2008 for more details). In addition, females whose range overlaps several male territories move frequently between them during oestrus, being courted and mated repeatedly by one or more territory holders (Liberg et al. 1998). The above suggests that female roe deer may take an active role in mate searching and mate choice, and this may explain why male breeding success is not more closely linked to the number of females present within the territory.

Our results also showed that, in support of prediction 5, body mass tended to have a positive effect on territory size (90% Kernel) in all males, and interestingly, that residual antler size had a positive effect on territory size in subadults only. Hence, among territorial males, heavy individuals with large antlers are the best competitors, able to defend the largest territories, with the highest access to mates. This also suggests that male phenotypic traits, and especially antler size, which reflect the ability of males to fight are particularly important in subadults, during the key period when males establish their first territory. Because of the strong fidelity of males to their first territory and the positive relationship between territory size and YBS, the establishment of a male's first territory may therefore play a crucial role in determining fitness over his lifetime (note, unfortunately, we were not able to conduct an analysis of the determinants of lifetime breeding success because we had lifetime breeding information and ranging data for too few males).

To sum up, although territoriality of male roe deer has certainly evolved as a tactic for ensuring access to mates (Hewison et al. 1998), our results suggest that it does not entirely conform to a resource defence polygyny strategy (see also Börger 2006 for a similar suggestion). We speculate that social constraints may govern, at least partly, the distribution of female roe deer at the local scale, and that variation in the number of females within a male's territory probably depends mainly on the resultant annual variation in female distribution (Bramley 1970; Strangaard 1972). Our results are therefore in agreement with the view that male territoriality conforms to an ‘always stay’ and ‘low risk–low gain’ strategy in roe deer (see Linnell & Andersen 1998), where males do not compete intensely for territories and mating (at least once they have established their first territory). Rather, males gain the benefits of site familiarity and reduced frequency of dangerous fights with rival males by remaining in the same area where they are dominant (Owen-Smith 1977). As suggested by Liberg et al. (1998) and Hoem et al. (2007), the adaptive function of male territoriality in roe deer is thus not the monopolization of females present within territories for mating, but rather the delimitation of an area in which a buck is dominant and so can court and mate without interference from other males. Roe deer territoriality seems therefore to differ from the territorial systems of most other ungulates, for which previous studies have demonstrated a direct relationship between territory quality and mate access (e.g. in impala and waterbuck: Jarman 1974; American pronghorn: Kitchen 1974; puku and topi antelopes: Balmford et al. 1992), suggesting a better fit with the conventional resource defence polygyny model than for roe deer. We therefore suggest that resource defence polygyny should not be considered as a homogeneous mating tactic. There are actually no discrete distinctions between the different types of polygynous mating tactics described by Emlen & Oring (1977) and, as Jarman (1974) pointed out, ‘any attempt to subdivide a continuum creates problems’. The variability of polygynous mating tactics would be probably more realistically described as a continuum extending from resource defence polygyny to female defence polygyny, with roe deer occupying an intermediate position.

Acknowledgements

We are grateful to all the field-workers who helped in fawn capture, radio tracking and pellet-group counts at Bogesund. Vegetation sampling was performed by Maria Tybergsson and Per Grängstedt. We also thank Maxime Galan for his work with the molecular analyses, Clément Callenge for his advice concerning spatial statistics, and Hélène Verheyden for her advice concerning habitat quality analyses. We thank two anonymous referees for constructive comments on an earlier version of this manuscript. The study was financed by the Swedish Association for Hunting and Wildlife Management, the private foundations of ‘Oscar och Lilli Lamm stiftelser’, ‘Olle och Signhild Engkvist stiftelser’ and a FORMAS-INRA grant.

Ancillary