Theories of habitat selection assume that individuals of a species occupy a restricted and favourable part of their potential habitats at low population density and thereby optimise their individual fitness (Wiens 1989, Rosenzweig 1991). The potential habitats are filled with territories in a deterministic order, often in a density-dependent fashion, as suggested by the models of Fretwell & Lucas (1969). Therefore, favourable habitats will more often be occupied than marginal habitats. In studies where the difference between the compared habitats is large, the modelled predictions often are realised quite well (O'Connor 1986, 1987). However, where the habitats are more similar and all are rather favourable, the spatial distribution of animals can be more stochastic (Rotenberry & Wiens 1980, Orians & Wittenberger 1991). The distribution of habitats in landscapes has been described as heterogeneous or homogeneous (Addicott, Aho, Antolin, Padilla, Richardson & Soluk 1987), a division clearly related to scale and the focal animals' ranges of movements (Forman & Godron 1986, Wiens 1989). A similar classification of landscapes is the separation between fine-and coarse-grained habitat patterns (Levins 1968), which by definition is related to the behaviour and landscape perception of the species under study.

The response of animals to the distribution of habitats has been shown to differ according to spatial scale (Rotenberry & Wiens 1980), in large-scaled studies due to the matrix (Åberg, Jansson, Swenson & Angel-stam 1995, Saari, Åberg & Swenson 1998), as well as on the smaller scales due to habitat composition, e.g. within home-ranges (Jokimäki & Huhta 1996). Thus, multi-scale approaches in studies of animal/habitat relationships are of great importance (Wiens, Rotenberry & van Horn 1987).

The importance of stochastic versus deterministic factors in influencing the selection of territories by forest birds was studied by Haila, Nicholls, Hanski, & Raivio (1996). In their study of resident birds' territories over six years in a heterogeneous forest, an element of stochasticity was apparent. Haila et al. (1996) found that avoidance of certain areas by birds was predictable, whereas a specific habitat preference was not. They concluded that, in relatively suitable areas, individual birds might not seek to optimise their territory but rather to find an acceptable one, and that the influence of stochastic events must be evaluated at several spatial scales.

We have studied the habitat selection of the hazel grouse Bonasa bonasia, which is considered to be a habitat specialist (Eiberle & Koch 1975, Scherzinger 1979, Swenson & Danielsen 1991, Swenson & Angel-stam 1993), in a forest reserve. In this paper we examine the role of stochasticity on habitat selection on several spatial scales using a 10-year data set. Two types of digitised vegetation data were used; firstly, a detailed vegetation description obtained especially for hazel grouse and this study, and secondly, the common forestry maps of the study area. The study aimed to answer two questions: 1) are stochastic or deterministic factors more important in habitat selection in a generally good environment, and 2) how does the spatial scale of measurement influence the observed pattern of habitat selection? Also, we discuss methodological problems in measuring habitat selection to predict the occurrence of the species, and to evaluate the applicability of forest stand descriptions and census techniques for the management of the species.

Study area

The study area covers 522 ha and is located in south-central Sweden (59–60°N, 15–16°E). Forest dominates the area (72%), but bogs are also common (18 %; Cederlund 1981). In 1914 most of the forest in the study area was burned, but between 1914 and 1972 modem intensive forestry was conducted. After 1972, the forestry activities ceased, and the area became a nature reserve in 1993. The western part of the study area (Fig. 1) was dominated by Norway spruce Picea abies (57% of trees), white birch Betula pendula and pubescent birch B. pubescens (together 22%), and Scots pine Pinus sylvestris (17%). Other deciduous trees (4%) included aspen Populus tremula, black alder Alnus glutinosa, goat willow Salix caprea, rowan Sorbus aucuparia and grey alder A. incam. The eastern part had a much lower proportion of deciduous trees (Swenson 1991b), but spruce was still the dominating coniferous species. The proportion of deciduous trees in the stands surrounding the reserve was very low, < 7% (Anon. 1999:66) and in stands more than 60 years old the deciduous component was < 1 % (Swenson & Angelstam 1993). The field layer throughout the reserve was dominated by bilberry Vaccinum myrtillus, cowberry V. vitis-idaea, and wavy hairgrass Deschampsia flexuosa.

Figure 1.

The forest reserve and immediate surroundings with indication of the study area. Grey circles represent the 197 hazel grouse census points and black dots the sampling plots on which the detailed vegetation description was performed. Thin lines define the different forest stands. Dark grey areas are lakes and lighter grey areas are bogs. The dotted line in bold divides the western from the eastern part of the forest reserve.

Methods

Statistical analysis

The SNFE forest stand descriptions were used to analyse the habitat selection of hazel grouse males using the vegetation parameters of the stand. A correlation matrix was formed for the original 12 parameters measured. As several of the parameters were correlated, a principal component analysis (with the varimax rotation normalised; see Statistica 1997) was performed to reduce the number of habitat variables and the correlation among the four continuous variables. By this method the habitat variables were reduced to two independent habitat variables (Appendix I). These variables were related to coniferous and deciduous forests, respectively. The coniferous component was negatively correlated to pine and positively correlated to spruce, and the deciduous component was positively correlated to deciduous trees. A reduced number of habitat variables (PC-scores <0.1) were incorporated in the PCA-components. The number of male hazel grouse occurrences at each census point, which could vary within 0-20 for all seasons combined, was summarised, as were the number of hazel grouse occurrences separately during spring and autumn. Moreover, the first four seasons (autumn 1987 - spring 1989) were defined as seasons of high density (Fig. 2). Multiple stepwise regression was used to analyse the influence of the habitat variables at the census point on the number of times male hazel grouse were present during each period. Also, multiple logistic regression was used to analyse the effect of habitat on the presence or absence of hazel grouse at a census point for the different periods.

Figure 2.

Number of male hazel grouse found in the study area per season. Years on the x-axis denote the autumn and unlabeled ticks the spring for the respective year. Each season (autumn-spring) is defined as showing either high (H) or low (L) density based on the mean number of males in autumn year 1 and in spring year 2.

The range of the census method (75 m) was used to define step one in the scale analyses. Thereafter a factor of 1.5 was used to create the other steps, which were 33, 50, 112 and 169 m. Thus, the largest area analysed (radius 169 m) using the detailed vegetation material constituted approximately half a normal hazel grouse male home-range in the study area (Swenson 1991b). The vegetation for each census plot was described using the detailed vegetation data for the five spatial scales. At the smallest scale, only one vegetation point was used per census plot, whereas at the scale of 50 m most often two vegetation plots fell within the range, and on the scale of 75 m four vegetation plots usually fell within the range and so on. When more than one vegetation point was within the radius, the mean of the vegetation parameters was used. Thus, some of the vegetation variables were common for some census points, especially on the larger scales. Variables indicating structures of very low abundance or not normally distributed were excluded from the analysis. As several of the parameters were correlated, a principal component analysis (with the varimax rotation normalised; see Statistica 1997) was used to reduce the number of habitat variables and the correlation among them. Of the original 76 habitat variables, several variables were measured as categories, such as type of soil, or were too rare (mostly field layer species) to use in a PCA. No classes of variables were left out. Thus, 15 variables were used in the PCA where four independent habitat components were left (Appendix II). The new habitat components created were mainly the same at all scales, but the relative importance differed between scales. The components were interpreted as follows: 1) ‘Cover’, constituting the horizontal cover at several heights, 2) ‘Coniferous’, constituting pine and spruce, 3) ‘Age’, constituting tree height, canopy cover and standing volume, and finally 4) ‘Shrubiness’, constituting moisture and shrub cover. Only variables with PC-scores <0.1 were incorporated in the PCA-components in order to reduce the number of habitat variables. In an additional analysis, the original variables were grouped into four groups, based on the relationships among the original variables. The groups were cover, tree species composition, field layer and finally a group with variables being related to tree age. For each of these groups, a PCA was performed which resulted in one principal component for each of the groups. There are, of course, difficulties in grouping the variables correctly, because most of the parameters are very correlated by nature, and the earlier impacts of forestry practises in the area, but grouping was done to obtain a greater descriptive power in the analysis. Multiple stepwise regression was used to analyse the influence of vegetation parameters on the number of times a male hazel grouse was present at a census point. Absence or presence of hazel grouse during both seasons, springs and autumns, and during low and high densities was analysed using a stepwise multiple logistic regression.

Table 1.

Habitat variables included in the multiple logistic regressions having a significant (P < 0.05) effect in explaining the occurrence of hazel grouse in the entire study area (172 census plots), and in the western (108 census plots) and eastern area (64 census plots) separately, based on the forest stand descriptions.

To visualise the occurrence pattern of hazel grouse males over the study period, a 20-ha grid was imposed on the study area (i.e. the whole reserve), with a grid cell size similar to hazel grouse territories in the study area, i.e. 15–25 ha (Swenson 1991b). That is, in this analysis, the occurrence per grid and season was normally based on nine census points. However, as no censusing was performed on bogs or lakes (see Detailed vegetation data) and due to the shape of the reserve, the number of census points varied. Therefore the dependent variable, the number of male hazel grouse occurrences in a grid per season for all seasons (21) and all years (10) were divided by the number of censused points in each grid. The characteristics of the grid cells were described using the mean of the forest stand description variables (stand age, proportion of pine, spruce and deciduous trees) for each census point in each grid, respectively, transformed into PCA-variables for each grid. The relationship between the number of occurrences of hazel grouse in a grid cell and the PCA-variables were analysed, as were the original habitat variables, using multiple regression.

Results

Using the forest stand descriptions, no single habitat variable measured in this study was significantly related to the occurrence of hazel grouse in the western part of the study area during spring, autumn, combined seasons nor periods of high or low hazel grouse density (Table 1). However, the coniferous component was significantly and negatively related to the occurrence of hazel grouse in the eastern part during spring and autumn combined, and in years of high and low hazel grouse density. This was presumably due to the fact that the PCA-component coniferous forest was composed of both pine and spruce, and an increasing proportion of pine negatively influenced the occurrence of male hazel grouse. Also for the entire reserve, a significant effect of coniferous forest was found for the occurrence of hazel grouse during autumn, combined seasons, and years of high and low hazel grouse density (see Table 1). Moreover, the total number of hazel grouse occurrences was negatively related to an increasing amount of coniferous forest in the eastern (P = 0.015) and entire reserve (P = 0.0003), but no habitat variable was significantly related to the number of hazel grouse occurrences in the western area using the stand descriptions. The original tree species variables incorporated in the principal components coniferous and deciduous for the entire reserve, the western and the eastern part are presented in Appendix III. Analysing the original habitat variables separately using multiple regression showed that the proportion of spruce was significantly and positively related to the occurrence of hazel grouse in the eastern part during all seasons, during low density and during high density (all Ps < 0.01).

When analysing the ungrouped detailed vegetation data and the occurrence of hazel grouse during spring, autumn, combined seasons, and years of high or low hazel grouse density, we found that shrubiness was significantly and positively (P = 0.025) related to the occurrence of hazel grouse during autumn at the largest scale (169 m) in the western part. The number of hazel grouse occurrences was significantly and negatively (P = 0.017) related to an increasing amount of pine at the smallest scale (33 m). The grouped principal components of the detailed vegetation data, analysed in the same way as the ungrouped, showed one significant relationship. That was a positive relationship between the number of hazel grouse occurrences at a census point and the proportion of field layer (P = 0.04) at the scale of 112 m.

The number of occurrences of hazel grouse males within the grid cells varied within 0–12 during the study period (Fig. 3). A significant relationship was found between the number of hazel grouse occurrences in a grid cell during all seasons and during low and high density using the forest stand description. The significant relationship was found for both the PCA-variable coniferous forest (negative) and the original habitat variable spruce (positive) on number of hazel grouse occurrences (all Ps > 0.01; Table 2).

Figure 3.

Location of the grid cells in the forest reserve showing the number of hazel grouse presences in each grid cell during the study period.

Discussion

We used a 10-year data set of hazel grouse censuses, collected using an inventory method with a high and known reliability, over a large area for which the habitat composition was carefully described. Yet, despite this, no clear relationships between the occurrence of hazel grouse males and the parameters of vegetation measured were found at any of the spatial scales analysed in the western part of the study area. One may have expected hazel grouse male occurrence to be influenced by habitat parameters such as tree species composition or tree age, as the hazel grouse has been considered to be a habitat specialist (Swenson & Danielsen 1991, Swenson & Angelstam 1993, Wiens 1989). Furthermore, the hazel grouse has been shown to recognise differences between source and sink habitats in natural forests (Beshkarev et al. 1994), as well as in a managed-forest dominated landscape (Linden & Wikman 1983). However, our study area constituted only a small portion of the landscape where Swenson & Angelstam (1993) performed their study. We suggest that the range of habitat variation in our study area was apparently too small to produce any clear patterns for habitat or patch preferences of hazel grouse males at the smaller scales studied and methods used. Instead of preferences, the only significant relationship we found was the negative influence of the PCA-component ‘coniferous’. This was probably because one variable of that component was the number of pine trees, and pine-dominated habitats are rarely used by hazel grouse (Swenson 1993a).

Table 2.

Habitat variables included in the multiple regression having a significant (P < 0.05) effect in explaining the occurrence of hazel grouse in the 20 grid cells, based on the forest stand descriptions.

The results of our study are comparable to those of Haila et al. (1996) in several aspects. In contrast to our study, Haila et al. (1996) used a rather small study area (36 ha) and included fairly generalised species in their study, for which some of the home ranges probably were as large or larger than the whole study area. Further, they analysed species occurrences in relation to habitat variables measured around the nest, even though the position of the nest may not be in the most typical or preferred part of the home range, as they mention themselves. Therefore, it may not be surprising that they did not find any clear preferences for habitat selection by the studied species. However, in spite of our long-term data on a specialist species and detailed habitat descriptions covering a study area that was more than 25 times larger than the average home range of a hazel grouse, we found neither clear preferences for specific habitat types nor structures. Instead, the avoidance of a certain habitat (the PCA-component ‘coniferous’) by the hazel grouse was predictable, which is similar to what Haila et al. (1996) found for some species. The occurrence of hazel grouse males was better explained in autumn than in spring. An increasing amount of cover and a decreasing percentage of pine positively influenced the occurrence of hazel grouse during autumn, a pattern in agreement with what Åberg et al. (2000) reported from a nearby intensively managed forest area, measured on a landscape level. We propose that the general lack of relationships was due to the following factors:

The importance of the proper spatial scale for studying the habitat selection of the hazel grouse, measured as presence or absence at a census point, is obvious. At the larger scales, several vegetation points were common for some census points, which may have diminished the differences in habitat structure for the census points at the larger scales, resulting in a low level of explanation. And, at the smallest scale the vegetation points may have been located at random in a fine-grained landscape, resulting in weak relationships. On the other hand, the occurrence of hazel grouse has been shown to be independent of scale regarding the presence of alder, which was important for hazel grouse at the four measured scales (Swenson 1993b). In a study of breeding birds in an urban landscape in southern California, the landscape ecological parameters at larger scales were found to predict the occurrence of some species better than the fine-scaled parameters (Bolger, Scott & Rotenberry 1997). In our study, the occurrence of hazel grouse was explained better at the largest scale, i.e. the eastern part of the forest reserve compared to the western part. This was expected because the proportion of hazel grouse habitat was considerably lower in the eastern part and because the difference between good and bad habitat was more distinct (Kareiva 1990). However, the total number of hazel grouse occurrences at a census plot was significantly influenced only by the habitat variables at our smallest scale, a scale where Haila et al. (1996) found bird abundance to be most linked to the heterogeneity of the habitat.

The overall density of hazel grouse (low and high) did not influence the predictability of hazel grouse occurrence in the western area, as opposed to the eastern more heterogeneous part, where the occurrence of hazel grouse was more predictable during years of low densities. The amplitude of the density variation, or the difference between low and high years, was perhaps too small to detect an effect in the more rich western area. The analysis of the grid cells, perhaps questionable due to the low N-value, showed no statistical relationship between the number of hazel grouse occurrences and the quality of the habitat. It is probable that some subtle habitat parameters that we did not measure, or did not measure adequately, in addition to habitat heterogeneity and stochasticity, were important in influencing this pattern of grid occupancy. Moreover, as the methodology in this study only considers the occurrence of hazel grouse males, the occurrence of males can be influenced by the distribution of hazel grouse females.

Our results suggest that at least the scale of territory size is a proper level to investigate possible patterns of density dependent habitat selectibn in a species. At smaller scales, i.e. within territories or home ranges, the occurrence of individuals is more stochastic. For example, in the extreme case of a certain 1-m² plot, the probability of the presence of an individual such as hazel grouse is always very low regardless of habitat. Therefore, to find associations between animal species and habitats, a suitable study area must include enough habitat variation and the relationships should probably be examined at a scale equal to or larger than the home range of the species in question, or considerably larger if population data are available.

Ecologists commonly agree that for an effective promotion of biodiversity in forests, there is a need for methods and measurements applicable in management (Harris 1984, Noss 1990, Angelstam 1997, Jansson & Angelstam 1999), as well as knowledge of the relevant spatial and temporal scales (MacArthur & Wilson 1967, Wiens 1976, 1989). The whistle census method seems to be suitable to estimate densities and the presence or absence of hazel grouse males in a certain area (Swenson 1991c). However, as an instrument to analyse small-scaled differences in hazel grouse habitat selection, caution must be taken regarding the heterogeneity or grain-size of the landscape. In this respect, telemetry is probably a better method, unless a repeated number of visits using the whistle at a census plot are conducted and the fact that birds may leave their original position and approach the whistler is taken into consideration. The forest stand descriptions we used did not give sufficient data to manage the forest for viable hazel grouse populations in this area, although a high proportion of pine negatively influence the presence of hazel grouse. In addition to horizontal stratification of habitats within stands, which has been shown to be important for the hazel grouse (Swenson & Angelstam 1993), there is also a substantial need for a landscape perspective in forestry planning, to preserve the hazel grouse in managed landscapes (Åberg et al. 1995, Saari et al. 1998, Åberg et al. 2000).