The study was carried out in 14 areas of forest in Scotland (Fig. 1), including at least one within each of the main centres of capercaillie population (parts of Perthshire, Deeside, Strathspey and Morayshire; Catt et al. 1998). Choice of study areas (‘forests’) was constrained by the requirement to find a sample of 10 hens in each forest each year, so providing a reasonably reliable estimate of breeding success (Moss 1986). This was impracticable in many forests due to the generally low numbers of birds present, hence counts were made in forests where initial questionnaire surveys indicated moderate or high densities of capercaillie (Catt et al. 1998). Even so, due to the decline observed during the study, 10 or more hens were found in only six of the chosen study areas in 1991, 1992 and 1993, in four in 1994, 1995 and 2000, in two in 1996 and 1997 and in only one in 1998, 1999 and 2001.
Figure 1. Approximate locations of the forest areas searched for capercaillie broods. 1, Findhorn; 2, Grantown; 3, Alford; 4, Nethybridge; 5, Abernethy; 6, Carrbridge; 7, Aviemore; 8, Cairngorm; 9, Aboyne; 10, Ballater; 11, Aberfeldy; 12, Crieff; 13, Perth; 14, Auchterarder.
Download figure to PowerPoint
We categorized each study area, according to the predominant structure and species composition of its trees and ground vegetation, into one of three types: (i) open-canopied forest with well-developed ericaceous ground layers typical of native pinewoods dominated by Scots pine Pinus sylvestris; (ii) mature Scots pine plantation with canopy sufficiently open for the development of ericaceous ground layer shrubs; and (iii) mixed species plantations, often of spruce Picea spp. and larch Larix decidua, with closed canopy and few ericaceous shrubs, often having clear-felled and restocked areas (Table 1).
Table 1. Location and characteristics of the 14 forest study areas. Forest type: 1, open canopy, as in native pinewoods; 2, mature Scots pine plantation, canopy sufficiently open for some dwarf shrubs; 3, mixed species plantation with closed canopy, often with some clear-felled areas and restocked ground
|Forest area||Region||Forest type||Count years||Count area (km2)|
forest structure and ground vegetation composition
In April–July 1995, measurements of forest structure and ground vegetation were made at 20 plots evenly spread within each searched area, at the intersections of a 300-, 400- or 500-m grid, the size of the grid depending on the size of the area. Walking a random distance of ≤ 50 m in a random compass direction randomized the position of each plot.
The plot was sufficiently large to include 10 main canopy-forming trees. Four quadrats of size 4 m2 were placed at random, one within each quadrant of the plot. Within each quadrat the separate vegetation layers (moss, brash, dwarf shrubs and scrub with young trees) were identified and the percentage species composition of each layer below 2 m in height was estimated by eye to the nearest 10%. The height of each vegetation layer was measured at five places evenly spaced within the quadrat. Using a key from Picozzi, Catt & Moss (1992), each plot was given a box number, which classified the stand according to its structural characteristics. Each box has two scores derived from a principal component analysis of 14 forest habitat variables. A high GRANNY score, typically good for capercaillie (Picozzi, Catt & Moss 1992), is typified by well-spaced ‘granny’ trees (Steven & Carlisle 1959), with an open canopy and a well-developed ericaceous shrub layer. A high PLANTATION score, often poor for capercaillie, is characterized by tall, closely spaced mature plantation trees, especially spruces, with little ground vegetation.
Little is documented about predators of capercaillie in Scotland, but both fox Vulpes vulpes and pine marten Martes martes are known predators of capercaillie elsewhere (Marcstrom, Kenward & Engren 1988; Schroth 1991; Storch 2001) and our experience in Scotland (D. Baines, R. Moss & D. Dugan, unpublished data) confirms this. Within each forest area, approximately 10 km of tracks were searched for mammal scats. Where practicable, vehicle tracks were used rather than footpaths. The route formed a circuit that passed through representative parts of the area that was searched for capercaillie hens and chicks. Scats were classified as ‘fox’,‘pine marten’ and ‘others’.
The tracks were walked five times in spring 1995, initially in April to count and remove all scats, twice in May and twice in June The number of scats found per kilometre on the initial clearance round in April in each forest (y) was correlated with the total number of scats (x) found per kilometre on the four subsequent rounds in May and June (y = 0·71x + 5·15, r12 = 0·88, P < 0·001). In six of the 14 forests scats were again counted in spring 1997. No significant differences were found between years ( = 0·02, P = 0·90), showing that mammalian scat abundance did not change significantly within these forests between 1995 and 1997. In short, scat-based indices showed consistent differences between forests and we assumed that they reflected predator abundance.
A minimum of 5 km of the same circuit was also used as a transect along which raptors and crows Corvus corone, potential avian predators of capercaillie eggs and chicks (Cramp & Simmons 1980; Storch 2001), were counted. Transects were walked just after dawn, twice monthly in May and June, and sightings and calls of crows and raptors were recorded.
Some predators are routinely killed by gamekeepers. Hence the number of gamekeepers and deer-stalkers actively involved in killing predators, on either a full-time (one person) or part-time (half a person) basis, was expressed as a density (gamekeepers per 100 km2) for the estate or forest block as a whole, not for just the area of forest studied. In three forests there was either very little or no organized predator control and gamekeeper density was entered as zero. Data on the effort made by individual keepers and on the range and number of each species killed on each estate were not collated.
We analysed the three measures of capercaillie breeding success in two stages. First, we checked whether breeding success differed consistently between forests over years. Secondly, we investigated whether aspects of habitat or indices of predator abundance explained differences in breeding success between forests.
To assess the effects of explanatory variables on breeding success, we used generalized linear models (SAS Institute 1996; genmod procedure, release 8.02) in preliminary exploration of the data and generalized linear mixed models (SAS glimmix macro; Littell et al. 1996) to enact the analyses presented here. The main difference between the two approaches was that the category year had to be entered as a fixed effect in generalized linear models, but could be treated as a random effect in generalized linear mixed models, so increasing the generality of the conclusions. We investigated variations in chicks per hen by setting the number of chicks seen in each forest each year as the dependent variable and forest descriptors as explanatory variables, in Poisson regressions (Poisson distribution, log link, adjusted for overdispersion; see Appendix 1) with the logarithm of the number of hens as an offset. We allowed for the fact that counts were repeated annually at the same forests by defining forest as a repeated measure. Brood size was analysed in the same way but excluding hens with no chicks and setting the logarithm of the number of broods as the offset. Broods per hen was modelled using logistic regression (binomial distribution, logit link), categorizing hens as successful if they had at least one chick, or as unsuccessful.
The statistical significance of the results was assessed in two ways. Sequential (SAS type 1) analyses estimated the significance of each of the fixed effects as they were entered one by one into the model. Partial (SAS type 3) analyses estimated the significance of each effect, after controlling for all other effects, giving the same result as if each effect was entered last in a sequential analysis. Partial analyses were appropriate for models with main effects only, not for models that included interactions between explanatory variables.
Ideally we should have made measurements of habitat and predator abundance throughout the study, but this was not done for financial reasons. We had planned to make these measurements in 1995 and to compare them with capercaillie data from 1995. However, again for financial reasons, we did not count capercaillie in all forests in 1995. Also, by 1995 numbers of capercaillie were generally low and in only four of the forests did we see the target of 10 hens, hence estimates of breeding success were unreliable. Therefore we increased sample size by continuing the study until 2001, used the 1995 measurements as continuous variables that described each forest, and modelled year as a categorical effect.
The indices of predator abundance (fox scats day−1 10 km−1, pine marten scats day−1 10 km−1, carrion crows km−1 and raptors km−1, where ‘day’ is the number of days since the previous visit) were negatively skewed, with a few big measurements that would have disproportionately influenced the results. We therefore used the natural logarithm (+0·1) of each index, so approximately normalizing them.
At Abernethy forest, most crows and some foxes were killed in some years but not in others, as part of a management trial to discover whether the killing of these predators improved capercaillie breeding success. Full details of this trial will be given elsewhere (Summers et al. 2004). For present purposes we categorized data from Abernethy into years when crows and foxes were killed (1992–96 and 2000–01) or not killed (1991 and 1997–99). Data from other forests were arbitrarily categorized as ‘killed’ to indicate no experimental change in predator management.
There were too many habitat measurements to include them all in the results. We selected the most useful ones by plotting scatter diagrams of capercaillie breeding success against summary statistics of the measurements. These included scores from two principal component analyses (SAS Institute 1996; princomp procedure), one representing three-dimensional forest structure and one ground vegetation. The breeding success data were too sparse to give reliable estimates for each forest in each year (above) and so, for the scatter diagrams, we postdicted (predicted after the event) breeding success from a generalized linear mixed model that had forest as the sole fixed effect and year as a random effect.
The summary statistics investigated by scatter diagrams included the mean, the median and the upper and lower quartiles of each habitat measurement and principal component score. We analysed the data for simple correlations and for quadratic associations between the three measures of breeding success and each summary statistic. In the event, only mean values gave significant results.