1 Burning and mowing ground vegetation for capercaillie 2 Tetrao urogallus conservation: an experimental test. 3 4 MARK H. HANCOCK*, RON W. SUMMERS*, ANDY AMPHLETT*, ROBERT 5 PROCTOR*, PETER HARVEY+ and STIJN BIERMAN‡ 6 * Royal Society for the Protection of Birds Scotland, Etive House, Beechwood Park, 7 Inverness, IV2 3BW, UK. 8 ‡ 9 UK + 32 Lodge Lane, Grays, Essex, RM16 2YP, UK. Biomathematics & Statistics Scotland, The King's Buildings, Edinburgh EH9 3JZ, 10 11 Correspondence: Mark Hancock, email mark.hancock@rspb.org.uk, phone 01463 12 715000, fax 01463 715315. 13 14 Running title: Burning and mowing for capercaillie 15 16 Word count: text (excluding Supplementary Material): 6604, tables: 372, total 6976. 17 1 1 Summary 2 3 1. Populations of the capercaillie Tetrao urogallus, a forest grouse of conservation 4 and economic importance, have declined over much of its range, often due to poor 5 breeding success. Burning or mowing of ground vegetation could increase breeding 6 success, if followed by increases in bilberry Vaccinium myrtillus and associated 7 arthropods, important to capercaillie chicks. We carried out an experiment testing 8 these management techniques at Abernethy Forest, Scotland. 9 2. Twenty-five experimental blocks were established within semi-natural pinewood 10 with a heather Calluna vulgaris and Vaccinium spp. shrub layer. Each block held 11 three 700 m2 plots, randomly assigned to control, mow and burn. Vegetation, 12 arthropods and capercaillie usage were monitored for one year before, and three years 13 after treatment. 14 3. Bilberry cover increased in mown and burnt areas, but there were also increases in 15 controls, linked to unusual natural heather die-back. Consequently, there was no 16 treatment effect on bilberry cover. Modelling a hypothetical scenario without heather 17 die-back, suggested that, had it not occurred, there would have been significant 18 treatment differences. For this ‘no die-back’ scenario, bilberry cover in burnt and 19 mown plots increased after three growing seasons to 23% (95% confidence intervals 20 13-35%), compared to control estimates of 13% (7-23%). 21 4. There were treatment effects on the biomass of some arthropod groups important 22 in the diet of capercaillie chicks, but, when modelling the ‘no die-back’ scenario, most 23 of these effects disappeared. Detection-corrected counts of summer grouse dung 24 (mainly capercaillie) were 4.3-6.3 (confidence intervals 2.5-9.8) times higher in 25 treated plots than controls. 2 1 5. Synthesis and applications. In forests with a heather-Vaccinium shrub layer, in the 2 absence of unusual natural heather die-back, burning or mowing ground vegetation is 3 likely to improve capercaillie habitat quality, by increasing the cover of a key plant 4 (bilberry) within three years. However, increases in arthropod food abundance for 5 capercaillie chicks, and longer-term vegetation responses, are uncertain, and will need 6 to be re-assessed after longer-term study of the experimental areas. 7 8 Keywords: arthropods, bilberry Vaccinium myrtillus, disturbance, dung counts, grouse 9 (Aves: Tetraonidae), heather Calluna vulgaris, pine forest, prescribed fire, Scotland. 10 11 Introduction 12 13 The capercaillie Tetrao urogallus L., a forest grouse highly valued by conservationists 14 and hunters across much of temperate/boreal Eurasia, has suffered widespread 15 population decline (Storch 2001), often linked to poor reproduction (Moss et al. 2000; 16 Wegge et al. 2005). In Scotland, rapid decline led to fears of regional extinction 17 (Moss 2001). Habitat management is considered key to capercaillie conservation 18 (Storch 2001). An important habitat element is bilberry Vaccinium myrtillus L., with 19 which capercaillie broods are strongly associated, probably because it supports 20 abundant arthropod food (Storch 1994; Summers et al. 2004; Wegge et al. 2005). 21 Capercaillie at more bilberry-rich sites, have higher breeding productivity (Baines, 22 Moss & Dugan 2004). 23 24 The ground vegetation at capercaillie sites commonly comprises bilberry and other 25 ericaceous shrubs, like heather Calluna vulgaris L., and cowberry Vaccinium vitis- 3 1 idaea L. (Storch 2001). In Scottish semi-natural pinewoods, heather often dominates 2 (Steven & Carlisle 1959; Summers et al. 1999), particularly in the high light 3 environment typical of these areas (Parlane et al. 2006), which is often linked to past 4 human impacts (Summers et al. 1999). 5 6 Post-fire succession on heather moorland sometimes includes a prolonged phase of 7 increased Vaccinium abundance (Ritchie 1955; Hobbs & Gimingham 1984). We 8 wished to test whether this might also occur in pinewoods. If so, it could improve 9 capercaillie habitat quality, and support the introduction of prescribed fire into 10 pinewood conservation management. This would support the contention that 11 ecological boreal forest management should include the emulation of some forms of 12 natural disturbance, such as fire (Angelstam 1998; Perera, Buse & Weber 2004). 13 Forest management with fire is novel in the UK (Bruce & Servant 2003). The 14 Scottish semi-natural pinewood resource is severely depleted (Anon 1994), and there 15 were concerns that prescribed fire could be unsafe or impractical in this habitat. 16 Therefore, we also tested mowing, which can produce similar effects to fire (Cotton & 17 Hale 1994; Schimmel & Granström 1996). Our aims were to determine whether 18 burning and/or mowing, within a pinewood with Calluna-Vaccinium ground 19 vegetation, would lead to (i) increased cover of bilberry; (ii) increased biomass of 20 arthropods important to capercaillie chicks; and (iii) increased usage by capercaillie. 21 22 Methods 23 24 STUDY AREA AND EXPERIMENTAL DESIGN 25 4 1 The study took place at Abernethy Forest nature reserve (3°37’W, 57°14’N), within 2 the Cairngorms National Park in the Scottish Highlands (Fig. 1a). The reserve 3 includes c 4000 ha of Scots pine Pinus sylvestris L. forest, on predominantly peaty 4 soils (Steven & Carlisle 1959; Summers et al. 1997). The only large herbivores are 5 roe deer Capreolus capreolus L. and red deer Cervus elaphus L., each at densities 6 around 8 km-2 (unpubl. data). At the nearest weather station (12 km to west, 228 m 7 altitude) during 1994-2003, mean annual rainfall was 1060 mm. January and July 8 mean temperatures were 2.4 and 13.9°C respectively. 9 10 Twenty-five experimental blocks (Fig. 1b) were selected at random within ‘old, open 11 forest’ (‘Box 2-3’: Picozzi, Catt & Moss 1992), at altitudes of 250-400 m. These 12 comprised mainly P. sylvestris-Hylocomium splendens-Vaccinium woodland (W18b: 13 Rodwell 1991). Block-scale densities of mature trees (modal height 17 m) averaged 14 174 ha-1 (range 62-378). Each block consisted of three, 20 m x 35 m plots, separated 15 by 10 m (Fig. 1c). Within each plot, eight 5 m x 5 m quadrats were used for recording 16 (Fig. 1d). One plot per block was assigned at random to control, burning, and 17 mowing. Baseline measurements took place in 2002, treatments were applied in 18 spring 2003, and further monitoring took place in 2003-5. 19 20 EXPERIMENTAL TREATMENTS 21 22 Plots were burnt in strips, with water used to protect plot boundaries and features such 23 as pine saplings (Dugan 2004). Mowing was by handheld, metal-bladed strimmer. 24 Because the availability of suitable prescribed burning weather was uncertain, 25 mowing only took place at a block after the ‘burn’ plot had been burnt. 5 1 2 Treatment characteristics were expected to vary in ways that might influence 3 subsequent succession (Schimmel & Granström 1996). Some of this variation might 4 be amenable to management control. Therefore, various treatment characteristics 5 were measured. For fires, we estimated ‘fireline intensity’ (Byram 1959) from three 6 flame-length estimates per quadrat. To aid estimation, ‘fire canes’ (graduated 7 2.5 m x 2 cm steel tubes) were fixed vertically at quadrat centres before fires. ‘Depth 8 of burn’ (Schimmel & Granström 1996) was measured by marking the moss/litter 9 surface at four points per quadrat using small metal posts, and measuring any 10 reduction in height after the fire. Soil heating can strongly affect vegetation 11 succession (Schimmel & Granström 1996), with 10 minutes at c 55°C being lethal for 12 bilberry rhizomes (Granström & Schimmel 1993). Therefore, two simple measures of 13 heat duration were taken: firstly, time above 55°C was measured using timers linked 14 to thermocouples fixed on the moss/litter surface, 20 cm from each fire cane; and 15 secondly, by direct observation of the duration of flames touching fire canes. For 16 mowing, in the summer after treatment, we measured the height above the moss/litter 17 surface at which stems were cut, and the cover of severed material. 18 19 VEGETATION SURVEYS 20 21 Ground vegetation was surveyed in August-September 2002-5, by a single observer 22 (MH), using a 2 m x 25 mm x 5 mm graduated stick, used in other fire studies (Davies 23 et al. in press). At four sample points per quadrat, the stick was pushed vertically into 24 the moss/litter, down to the soil/humus surface. The maximum height within 5 cm of 25 the stick was measured for shrubs and moss/litter. Vegetation structure was 6 1 characterised by standing with the vertical stick at arm’s length, and estimating the 2 percentage of the stick visible in a series of 10 cm bands. This gave an index of 3 vegetation openness at various levels above the soil/humus surface. The value of this 4 index for a 10 cm band centred at the surface of the moss/litter, termed ‘ground-level 5 openness’, was estimated by interpolation. The total cover (including cover below 6 other species) was estimated for shrub species inside a 1 m radius circle centred at 7 each point. Young pines (under 1.5 m) were counted within the same circle. Cover 8 was scored separately for heather that was live (green), recently-dead (brown), long- 9 dead (grey), or shoot or seedling regeneration. Bilberry defoliation, which affects 10 cover estimates, was estimated as the proportion of the 10 bilberry shoots nearest the 11 vertical stick that were completely without leaves. 12 13 Before treatments were applied, we measured the positions and heights of all trees 14 over 1.5 m, within plots and adjacent 5 m buffer zones. For pines, we recorded the 15 proportion of foliage within three height bands: 0-2 m, 2-5 m and over 5 m. The 16 proportion of foliage that was brown was recorded within the same height bands, 17 before and after fires. Sub-canopy light level was calculated using tree densities and 18 heights, and the regression equation of Parlane et al. (2006). 19 20 ARTHROPOD SURVEYS 21 22 Arthopods were surveyed in June, when insectivorous capercaillie chicks are present 23 (Summers et al. 2004). Our main sampling technique was pitfall trapping, which 24 despite its potential biases, remains a practical and commonly-used technique 25 (Southwood & Henderson 2000; Saint-Germain et al. 2007), including for studies of 7 1 capercaillie diet (Summers et al. 2004). To supplement pitfall data, we also carried 2 out some direct counts of arthropods. Direct counts of caterpillars were found by 3 Atlegrim & Sjöberg (1995), to be positively correlated with foraging success of tame 4 capercaillie chicks. 5 6 Pitfall traps were 150 ml polyethylene containers, with a 46 mm opening diameter, 7 two-thirds filled with trapping solution (ethylene glycol diluted 1:3, with one drop of 8 detergent). Traps were set at each block, at all quadrat centres, on randomly-selected 9 dates during the first half of June, and collected 14 days later. Four arthropod groups 10 of importance in capercaillie chick diet (Kastdalen & Wegge 1985; Spidsø & Stuen 11 1988; Picozzi, Moss & Kortland 1999; Summers et al. 2004) were extracted and 12 counted: spiders (Araneae), beetles (Coleoptera), caterpillars (Lepidoptera), and ants 13 (Hymenoptera: Formicidae). Caterpillars were measured to the nearest mm. Adult 14 ants, beetles, and spiders were identified to species. For traps that caught many 15 Formica ants, a random sample of 20 was identified. Other Formica were counted 16 and assumed to comprise similar species proportions. 17 18 Direct arthropod counts were carried out at every quadrat during June in the last two 19 years of the study. The observer crouched near the quadrat centre and, for one 20 minute, recorded any arthropods observed within 50 cm of the centre marker. 21 Arthropods were counted by group, and sized by eye to the nearest mm. 22 23 Different arthropod groups may have different trapping bias (Southwood & 24 Henderson 2000), attractiveness to capercaillie (Kastdalen & Wegge 1985), or effects 25 on chick survival (Picozzi, Moss & Kortland 1999). Therefore, biomass was 8 1 estimated separately by group, using the approach of Saint-Germain et al. (2007). 2 Counts by species, or size class for caterpillars, were determined for each plot. 3 Median adult body lengths were collated from published keys. The dry mass for an 4 individual of each species (or caterpillar size class) was estimated using published 5 log(length)-log(mass) regression coefficients (Rogers, Buschbom & Watson 1977; 6 Gowing & Recher 1984; Sample et al. 1993; Hódar 1996; Ganihar 1997), with back- 7 transformation correction (Sprugel 1983). The individual body mass of each species 8 was estimated as the mean of values given by all available regressions. These were 9 multiplied by count to give the biomass of each species in each plot. Species 10 biomasses were then summed to give the biomass for the whole group. For sexually- 11 dimorphic spider species, calculations were done separately by sex. 12 13 MEASURING CAPERCAILLIE USAGE 14 15 Dung counts were carried out in May and October each year, to provide a measure of 16 capercaillie usage. Dung found in May was cleared, so that October counts primarily 17 represented summer accumulation. This is the period when ground vegetation is most 18 used by capercaillie (Storch 2001; Summers et al. 2004). Dung counts involved 19 searching a 2 m radius area centred at the quadrat centre for five minutes. The 20 number and diameter of pellets in each dung group were recorded. We could not 21 assume that dung groups would be perfectly detected, and therefore quantified 22 detection rate as a function of vegetation openness using a trial with dummy dung. 23 Detection rate was then estimated for each plot in each year from vegetation openness 24 data (Supplementary Material). Black grouse Tetrao tetrix L. were also present and 9 1 their dung overlaps in size with that of capercaillie (Brown et al. 1987). Therefore, 2 we collated incidental grouse sightings, to see how commonly they occurred. 3 4 DATA ANALYSIS 5 6 Statistical analyses were used to investigate treatment effects on changes in bilberry 7 cover, arthropod biomass, and capercaillie usage, between the base-line year (2002: 8 pre-treatment) and post-treatment years. In all cases, we used linear mixed models in 9 SAS (SAS Inst., 2000), assuming that errors were normally distributed (after 10 transformation of the response), except for capercaillie usage (dung count) data where 11 a generalized linear mixed model with Poisson errors was used. The significance of 12 explanatory variables was estimated using F-tests with the denominator degrees of 13 freedom estimated by the Satterthwaite approximation. Model assumptions were 14 assessed visually by examining probability plots and plots of residuals against 15 predicted values. The normality assumption was checked for random effects and 16 model residuals. In some cases the response variables were transformed in order to 17 achieve normality (as indicated below). 18 19 Differences between treatments in the change in bilberry cover from the baseline year 20 (2002) to three years after the application of treatments (2005), were estimated using 21 the following linear mixed model: 22 23 Yi,j = a0 + a1*Xi,j + a2*X2i,j + a3*Si,j + a4*Vi,j + a5*Li,j + Bi + Tj 24 10 (1) 1 with Yi,j and Xi,j the total bilberry cover (arc-sine fourth-root transformed to achieve 2 normality of model residuals) in 2005 and 2002 respectively, as measured in the plot 3 with treatment j (j=1(control), 2(mown), or 3(burnt)) in block i (i=1,2,…25). Si,j and 4 Vi,j are the difference in mean defoliation score and visit date between these years 5 (2005 minus 2002) and Li,j is the light index. These were included because these have 6 been found to affect bilberry cover by Parlane et al. (2006) (Vi,j and Li,j) or were 7 assumed to do so (Si,j). Bi is a random effect representing potential block effects (a 8 spatial factor). The main parameters of interest are the treatment effects Tj. We 9 included baseline bilberry cover squared as a covariate, as we expected, and observed, 10 greater change in bilberry cover where initial cover was further from its minimum and 11 maximum possible values of 0% and 100%. 12 13 The experiment was affected by an unusual natural heather die-back event (Hancock 14 in press), roughly synchronous with the application of experimental treatments in 15 spring 2003. This was probably caused by exceptional weather conditions, combined 16 with the maturity of heather at the site. We expected heather die-back to lead to 17 increases in bilberry cover, due to evidence of competitive effects (Parlane et al. 18 2006). Therefore, as well as the observed experimental results, we wished to estimate 19 treatment effects for a hypothetical situation where this die-back event had not 20 occurred, termed the ‘no die-back scenario’. Die-back was calculated as the 21 proportion, in control plots, of heather cover in summer 2003, that was brown 22 (recently dead). To model bilberry response under the hypothetical ‘no die-back 23 scenario’, the heather die-back scores were added as covariates to eqn 1: 24 25 Yi,j = a0 + a1*Xi,j + a2*X2i,j + a3*Si,j + a4*Vi,j + a5*Li,j + Bi + Tj + a6*Di + bj*Di 11 (2) 1 2 where the coefficients a6 and bj are the parameters for the slopes for Di, the arc-sine 3 square-root transformed heather die-back score at the control area of block i in 2003. 4 Other terms are as in eqn 1. 5 6 Differences between treatments in change in arthropod biomass in pitfall traps, from 7 the baseline year (2002) to each of the three post-treatment years, were estimated 8 using the following linear mixed model: 9 10 Zi,j,t = a0 + a1*Mi,j + Yt + Bi + Tj + TYj,t (3) 11 12 where Zi,j,t and Mi,j are respectively the post-treatment and pre-treatment biomass 13 estimates for the invertebrate groups of interest in plots with treatment j in block i, 14 and post-treatment year t (t=2003, 2004 or 2005). Yt is a categorical variable for the 15 post-treatment year-effect, TYj,t is a nine-level categorical variable representing the 16 interaction between treatment j and year t, and other variables are as eqn 1. Visual 17 inspection of residual plots from these models suggested that errors could be assumed 18 to be normally distributed after a loge transformation for the spider and ant biomass 19 data, and a fourth-root transformation for beetle and caterpillar data. Therefore we 20 used these transformations for the variables Zi,j,t and Mi,j. 21 22 We also expected natural heather die-back (see above) to affect arthropod responses, 23 due to consequent changes in vegetation structure, impacts on populations of 24 herbivorous arthropods feeding on heather, and increases in dead plant material 25 available to detritivores. Therefore, analogous to the bilberry cover analysis (eqn 2), 12 1 we adapted eqn 3 to estimate differences between treatments in change in arthropod 2 biomass under the ‘no die-back scenario’, by adding the heather die-back scores as 3 covariates: 4 5 Zi,j,t = a0 + a1*Mi,j + Yt + Bi + Tj + TYj,t + a2*Di + dj*Di + et*Di + fj,t*Di (4) 6 7 The coefficients dj, et and fj,t are the treatment-specific, year-specific, and the 8 treatment-by-year-specific slopes for the heather die-back scores Di, respectively. 9 10 Direct arthropod counts were analysed in the same way except that no baseline data 11 were available. 12 13 To investigate differences between treatments in capercaillie usage, we estimated 14 differences in autumn grouse dung counts between plots with different treatments, in 15 the three post-treatment years. Baseline (pre-treatment) data were not included as 16 these showed an unexpected pattern of high counts in control plots (Results) which 17 could have led to estimates of treatment effects that were unduly positive. As there 18 were many zero counts, only blocks in years with at least one grouse dung group 19 recorded were included in the analysis. We estimated relative differences in dung 20 frequencies between treatments by fitting the following model: 21 22 Loge(DGi,j,t) = Bi,t + Pi,j + Loge(Ri,j,t) + Tj (5) 23 24 where Bi,t are intercepts (fixed effects) for block i in year t (to account for clustering 25 of counts of plots within the same block), and Pi,j random effects for plots with 13 1 treatment j in block i (to account for clustering of counts within plots across years). 2 Ri,j,t (included as an offset in the model) are detection rates of dung groups of median 3 size in plot i in block j in year t, as estimated from a model which related dung- 4 detection rates to visibility scores, parameterised using data from the dummy dung 5 trial (Supplementary Material). 6 7 Results 8 9 TREATMENT CHARACTERISTICS AND IMPACTS ON TREES 10 11 The method of treatment was similar at all blocks, but because of variations in 12 vegetation, terrain, and weather, certain treatment characteristics also varied 13 (Table 1). Mean mowing date was three weeks later than that for burning. Fires were 14 generally low intensity, with mean flame heights of 0.6 m. On average, only 1 cm of 15 the initial mean moss/litter depth of 15 cm was consumed by fires. 16 17 There were some losses of young (under 1.5 m) pines due to treatments. Before and 18 after treatment, the total number of young pines fell from 45 to 39 (179 to 155 ha-1) in 19 burnt plots and from 27 to 22 (107 to 88 ha-1) in mown plots. However, newly- 20 established pine seedlings were later found in both burnt and mown plots. Counts of 21 young pines, three years after treatment, were 192 (764 ha-1) and 80 (318 ha-1) in 22 burnt and mown plots respectively. 23 24 There were localised impacts of fires on mature pines. Recently dead (brown) needles 25 were rare before burning, averaging 0%, 0.4% and 0% of foliage in height classes 14 1 0-2 m, 2-5 m and over 5 m. However, after burning, plot-scale values in these height 2 classes rose to 43%, 17% and 1.2% (means) and 79%, 58% and 10% (maxima), with a 3 weighted overall mean of 8.2%. Six of the 289 mature (over 1.5 m) pines on burnt 4 plots were killed, the tallest being 3.5 m. 5 6 EFFECTS ON GROUND VEGETATION 7 8 Median bilberry cover was around 10% before treatment, and remained similar in the 9 first summer after treatment (Fig. 2a). By three years after treatment, there were 10 increases in cover, not only in burnt and mown areas (median 23-24%), but also in 11 controls (median 19%). Median live heather cover was initially around 75%, then 12 declined to less than 6% in mown and burnt areas after treatment (Fig. 2b). There was 13 a smaller decline in controls, to around 47%, linked to unusual natural heather die- 14 back (Hancock in press). Most surviving heather in treated areas was mature shoots 15 that had avoided treatment, having been part-buried in moss, or too damp to burn 16 during fires. Two years after treatment, when different types of heather regeneration 17 were most clearly distinguishable, regeneration by vegetative sprouting and seedlings 18 averaged only 0.03-0.04% and 0.1-0.3% cover respectively, while surviving mature 19 heather averaged 5-8%. Heather frequency within 1 m radius sample points was 98% 20 initially, then fell in burnt and mown plots to 60% after treatment, and recovered to 21 88% by three years after treatment. The height of ground vegetation fell from a mean 22 of 53 cm (s.e. 0.86) initially, to means of 38 and 28 cm (s.e. 1.9, 1.0) in burnt and 23 mown plots respectively, the year after treatment. 24 15 1 The effect of treatment on bilberry cover, three years after treatment, was not 2 statistically significant (eqn 1, Table 2a, Fig. 3a). However, significant treatment 3 differences were estimated under the ‘no heather die-back scenario’ (eqn 2, Table 2b, 4 Fig. 3b). The influence of heather die-back on estimates of treatment effects is caused 5 by the interaction between treatment and die-back (term bj in eqn 2; Table 2b). In 6 particular, the positive values of the control vs. mown and control vs. burnt contrasts 7 indicate how bilberry cover was affected by heather die-back relatively more 8 positively in control plots, than in mown or burnt areas. Estimating treatment effects 9 at a heather die-back value of 0%, gave the following ‘no die-back scenario’ results: 10 from an initial mean bilberry cover of 10%, modelled cover in mown and burnt plots 11 increased after three growing seasons to 23% (95% confidence intervals 13-35%), 12 when that of control areas, had there been no heather die-back, was estimated at 13% 13 (7-22%) (Fig. 3b). Although variance was high, the lower confidence limit of bilberry 14 cover in treated areas exceeded the control mean, implying that a positive treatment 15 effect was likely. This was also the case when treatment effects were estimated for 16 hypothetical moderate heather die-back of 5%. 17 18 Plotting the residuals of the ‘no die-back’ bilberry model, by treatment, against the 19 treatment characteristic variables listed in Table 1, suggested there was little evidence 20 that burning or mowing management with a particular subset of characteristics was 21 linked to a more positive bilberry response. 22 23 ARTHROPOD RESPONSES 24 16 1 Arthropod biomass estimates, from pitfalls and direct counts, were significantly 2 positively correlated for spiders, beetles and ants (rs=0.25, 0.16 and 0.76; P(one- 3 tailed)=0.001, 0.026, and <0.0001 respectively; N=150), but not for caterpillars 4 (rs=0.049; P(one-tailed)=0.28). 5 6 Spider biomass in pitfalls (Fig. 4a) was higher in the burnt and/or mown plots, than 7 controls, two and three years after treatment (Table 3a, Fig. 5a). Mown and burnt 8 areas, after three years, had 2.3 (confidence intervals 1.8-2.6) and 1.8 (1.4-2.1) times, 9 respectively, the spider biomass of controls. Conversely, pitfall beetle biomass in 10 mown and burnt plots (Fig. 4b) was lower than controls, two years after treatment 11 (Table 3a, Fig 5b) by 49% (confidence intervals 2-73%) and 54% (12-76%), 12 respectively. Caterpillar biomass (Fig. 4c) did not differ significantly between 13 treatments in any post-treatment year. After the first post-treatment year, ant biomass 14 (Fig. 4d) in burnt plots was higher than that of control and/or mown plots (Table 3a, 15 Fig. 5c), exceeding that of control plots 3.2-fold (confidence intervals 1.9-5.2) and 16 2.0-fold (1.2-3.3) in the second and third post-treatment years respectively. 17 18 For the ‘no heather die-back’ scenario, most significant between-treatment differences 19 in arthropod biomass disappeared: only spider biomass in the second post-treatment 20 year remained significantly different between treatments (Table 3b, Fig. 5d), with 4.6 21 times (confidence intervals 1.8-12) greater biomass in mown plots than controls. This 22 implies that, to some extent, the treatment differences found in analyses of the real 23 results, reflected an avoidance of (spiders, ants) or preference for (beetles) control 24 areas affected by heather die-back. 25 17 1 Direct count data suggested that ant biomass in the second post-treatment year 2 differed between treatments (treatment x year interaction, treatment contrasts within 3 year two: F2,120=6.67, P=0.0018). In this year, ant biomass seen in mown and burnt 4 plots, per count, was 3.1 and 4.7 times, respectively, that seen in control plots (95% 5 confidence intervals: 1.2-6.7; 2.0-9.4). However, in the ‘no-dieback’ model, this 6 difference disappeared (F2,115=0.87, P=0.42). No other significant within-year 7 treatment differences were found in the direct count data. 8 9 CAPERCAILLIE USAGE 10 11 The detection rate of dummy dung groups was around 0.3 in control and cut 12 vegetation, and around 0.5 in burnt vegetation, three years after treatment (Fig. 6a). 13 The preferred model (Supplementary Material) of detection probability included 14 dummy-dung group size and pellet diameter, and their interaction, and vegetation 15 openness. This model gave estimates of the detection probability of a median dung 16 group (a single 9 mm diameter pellet), as around 0.11 in untreated vegetation, and 17 0.16 in burnt or mown vegetation (Fig. 6b). Real dung data showed a strong shift 18 from greater counts in controls, in the pre-treatment year, to the opposite pattern in 19 post-treatment years (Fig. 6c). Modelling the post-treatment data, with detection rate 20 included as a covariate, showed that counts were much higher in burnt and mown 21 plots in the post-treatment period, than controls (Fig. 6d), by factors of 4.3 and 6.3 in 22 burnt and mown plots respectively (95% confidence intervals: 2.5-7.3 and 4.0-9.8). 23 Residuals were higher in controls where heather die-back was greater; thus treatment 24 effects might have been even more pronounced in the absence of die-back. 25 18 1 In total, there were 28 recorded sightings of grouse within 50 m of experimental plots. 2 All but one of these were capercaillie. Ninety-two percent of grouse droppings found 3 were within the range of diameters of full-grown capercaillie (Gjerde 1990). Together 4 with anecdotal knowledge of the usual areas inhabited by grouse species at our study 5 site, this suggested that most recorded grouse dung was that of capercaillie. 6 7 Discussion 8 9 This study provides the first experimental evidence of the potential effectiveness of 10 burning and mowing heather-rich ground vegetation in woodland, as tools for 11 improving capercaillie habitat. Interpretation was complicated by a natural heather 12 die-back event, but when modelling a hypothetical scenario without die-back, we 13 estimated that burnt and mown areas would have bilberry cover scores that were 14 approximately double those of control means after three growing seasons. Although 15 wide confidence intervals highlight the variability in the response, an increase of some 16 degree is likely. The importance of bilberry to capercaillie (Storch 2001), particularly 17 for broods (Storch 1994; Summers et al. 2004), means that increases are likely to 18 benefit the species. 19 20 Our study supports the idea that conservation managers should consider introducing 21 some forms of disturbance, for biodiversity objectives (Angelstam 1998). However, 22 we found no evidence that the more natural of the two disturbance methods tested, 23 fire, had greater immediate advantages in terms of our objectives. It was not the 24 naturalness of a disturbance, but rather its impact, particularly in terms of ‘dominance 25 reduction’ (Wohlgemuth et al. 2002), that was more important. 19 1 2 Schimmel & Granström (1996) found that intense fires could eliminate Vaccinium 3 species. However, our fires never approached the peak intensities they tested, of 75 4 minutes heating, compared to a quadrat-scale maximum of 10 minutes in this study. 5 Similarly, our fires had lower fireline intensities than those of Bruce & Servant 6 (2003), whose peak values were over ten times ours, with correspondingly two to 7 three times more severe canopy browning of mature trees. Other studies have shown 8 that heather can recover rapidly after burning (Hobbs & Giminham 1984) or mowing 9 (Cotton & Hale 1994). However, we found little vegetative heather regeneration, 10 probably due to the advanced age of plants at the site. Despite high post-treatment 11 heather frequency, mainly due to surviving shoots of mature heather, we found that 12 heather cover, especially of young seedlings, remained low in the first few years after 13 treatment. These factors may have contributed to the positive bilberry response. 14 15 Bilberry-rich areas are usually richer in arthropods important to capercaillie, than 16 other vegetation (Kastdalen & Wegge 1985; Summers et al. 2004). However, 17 modelling a scenario with no natural heather die-back, we found few biomass 18 increases in key arthropod groups following treatment. Of four groups, counted in 19 two ways, on two treatments, there was only one increase. 20 21 Increases in capercaillie usage, as measured by dung deposition, were striking. 22 Summer dung accumulation was 4.3-6.3 times higher in treated areas than controls, 23 and might have been higher still in the absence of heather die-back. Storch (1993) 24 found that adult capercaillie were strongly associated with bilberry in summer, and 20 1 preferred vegetation heights of 30-40 cm, lower than that of our controls: either or 2 both factors may explain capercaillie usage of treated areas. 3 4 The results of this study are applicable to similar semi-natural and production forests. 5 However, it coincided with an unusual heather die-back event (Hancock in press). 6 While we have statistically modelled a ‘no die-back scenario’, this may not fully 7 reflect what would have happened had heather die-back not occurred, reducing the 8 certainty with which our findings can be applied to more typical years. Unexpectedly, 9 however, heather die-back highlighted the potential for major short-term vegetation 10 changes without management intervention, and illustrated the possibility that natural 11 disturbance events may periodically, but unpredictably, deliver vegetation changes of 12 similar magnitude to those obtained by management. 13 14 No other studies have evaluated these management techniques in similar habitats for 15 similar aims. Bruce & Servant (2003) showed the feasibility of using prescribed fire 16 within a similar Scottish pinewood, but replication was limited, with no detailed 17 measures of capercaillie habitat quality. 18 19 Our results show that land-managers can expect mowing or burning of the ground 20 vegetation, in open forests with heather-Vaccinium ground vegetation, to deliver 21 benefits for capercaillie conservation, via increased bilberry cover after three years. 22 These results help support the EU-funded mowing management already being carried 23 out in Scotland for capercaillie. However, the following caveats remain. Firstly, the 24 striking increase in capercaillie usage following burning and mowing does not imply a 25 population effect. Such an effect is more likely following improvements in brood 21 1 habitat and breeding success. While increased bilberry is likely to benefit broods, 2 general increases in key arthropod groups have not yet been demonstrated. Secondly, 3 results reported here apply to the first three growing seasons after treatment. Longer- 4 term vegetation development could lead to long-lasting areas valuable to capercaillie 5 broods, such as bilberry stands, or, potentially, quite different communities, such as 6 thickets of Scots pine seedlings, or rejuvenated heather stands. Any benefits could be 7 short-lived if heather rapidly recovers in treated areas. Finally, this study has 8 unexpectedly highlighted the potential for rapid vegetation changes, towards a 9 composition more favourable to capercaillie, to occur without management 10 intervention. Long-term monitoring of the experimental areas will be needed to 11 assess the importance of these caveats. 12 13 Acknowledgements 14 15 We are grateful to J. Wilson, C. Legg, N. Cowie, J. Roberts and K. Duncan for ideas 16 and support; D. Dugan, C. McClean, B. Moncrieff, A. McAskill and R. Watson for 17 performing the experimental treatments; J. Willi, S. Rao, G. Nisbet, H. Swift, A. 18 Macfie and I. Hutson for data collection; P. Hammond and P. Kirby for assistance 19 with arthropod species determinations; G. Lyons, G. Nisbet, and P. and A. Sinclair for 20 sorting pitfall catches; M. Telfer and I. Dawson for specialist arthropod advice; and K. 21 Kortland, S. Taylor, and J. Dunsmore for organising funding, which was provided by 22 RSPB, Scottish Natural Heritage and the EU LIFE fund. 23 22 1 References 2 3 Angelstam, P.K. (1998) Maintaining and restoring biodiversity in European boreal 4 forests by developing natural disturbance regimes. Journal of Vegetation 5 Science, 9, 593-602. 6 Anon, (1994) Caledonian Pinewood Inventory. Forestry Commission, Edinburgh. 7 Atlegrim, O. & Sjöberg, K. (1995) Lepidoptera larvae as food for Capercaillie chicks 8 (Tetrao urogallus): a field experiment. Scandinavian Journal of Forest 9 Research, 10, 278-283. 10 Baines, D., Moss, R. & Dugan, D. (2004) Capercaillie breeding success in relation to 11 forest habitat and predator abundance. Journal of Applied Ecology, 41, 59-71. 12 Brown, R., Ferguson, J., Lawrence, M. & Lees, D. (1987) Tracks and Signs of Birds 13 14 of Britain and Europe: an Identification Guide. Christopher Helm, London. Bruce, M. & Servant, G. (2003) Fire and pinewood ecology in Scotland: a summary 15 of recent research at Glen Tanar Estate, Aberdeenshire. Scottish Forestry, 57, 16 33-38. 17 18 19 Byram, G.M. (1959) Combustion of forest fuels. Forest Fire: Control and Use (ed K. P. Davis), pp. 61-89. MacGraw-Hill, New York. Cotton, D.E. & Hale, W.H.G. (1994) Cutting as an alternative to burning in the 20 management of Calluna vulgaris moorland: results of an experimental field 21 trial. Journal of Environmental Management, 40, 155-159. 22 Davies, G.M., Hamilton, A., Smith, A. & Legg, C.J. (in press) Using visual 23 obstruction to estimate heathland fuel load and structure. International Journal 24 of Wildland Fire. 23 1 2 3 4 5 Dugan, D. (2004) Field layer management trials at Abernethy Forest Reserve: an update. Scottish Forestry, 58, 17-19. Ganihar, S.R. (1997) Biomass estimates of terrestrial arthropods based on body length. Journal of Biosciences, Bangalore, 22(2), 219-224. Gjerde, I. (1990) Determination of sex in Capercaillie Tetrao urogallus by means of 6 winter dropping size. Fauna Norvegica Series C Cinclus 13: 31-32. 7 Gowing, G. & Recher, H.F. (1984) Length-weight relationships for invertebrates 8 from forests in south-eastern New South Wales. Australian Journal of Ecology, 9 9, 5-8. 10 Granström, A. & Schimmel, J. (1993) Heat effects on seeds and rhizomes of a 11 selection of boreal forest plants and potential reaction to fire. Oecologia, 94, 12 307-313. 13 14 15 Hancock, M.H. (in press) An exceptional Calluna vulgaris die-back event. Plant Ecology and Diversity. Hobbs, R.J. & Gimingham, C.H. (1984) Studies on fire in Scottish heathland 16 communities: II. Post-fire vegetation development. Journal of Ecology, 72, 17 223-240. 18 19 20 Hódar, J. (1996) The use of regression equations for estimation of arthropod biomass in ecological studies. Acta Œcologica, 17(5), 421-433. Kastdalen, L. & Wegge, P. (1985) Animal food in Capercaillie and Black Grouse 21 chicks in south east Norway – a preliminary report. Proceedings of 22 International Grouse Symposia, 3, 499-513. 23 24 Moss, R. (2001) Second extinction of Capercaillie (Tetrao urogallus) in Scotland? Biological Conservation, 101, 255-257. 24 1 Moss, R., Picozzi, N., Summers, R.W. & Baines, D. (2000) Capercaillie Tetrao 2 urogallus in Scotland – demography of a declining population. Ibis, 142, 259- 3 267. 4 Parlane, S., Summers, R.W., Cowie, N.R. & van Gardingen, P.R. (2006) 5 Management proposals for bilberry in Scots pine woodland. Forest Ecology 6 and Management, 222, 272-278. 7 8 9 10 11 12 13 14 15 16 Perera, A.H., Buse, L.J., & Weber, M.G. (2004) Emulating natural forest landscape disturbances: concepts and applications. Columbia University Press. Picozzi N., Catt D.C. & Moss R. (1992) Evaluation of capercaillie habitat. Journal of Applied Ecology, 29, 751-762. Picozzi, N., Moss, R. & Kortland, K. (1999) Diet and survival of Capercaillie Tetrao urogallus chicks in Scotland. Wildlife Biology, 5, 11-23. Ritchie, J.C. (1955) Biological flora of the British Isles. Vaccinium myrtillus L. Journal of Ecology, 44, 291-299. Rodwell, J.S. (1991) British Plant Communities. Cambridge University Press, Cambridge. 17 Rogers, R.L., Buschbom, R.L. & Watson, C.R. (1977) Length-weight relationships of 18 shrub-steppe invertebrates. Annals of the Entomological Society of America, 70, 19 51-53. 20 Saint-Germain, M., Buddle, C.M., Larrivée, M., Mercado, A., Motchula, T., Reichert, 21 E., Sackett, T.E., Sylvain, Z. & Webb, A. (2007) Should biomass be considered 22 more frequently as a currency in terrestrial arthropod community analyses? 23 Journal of Applied Ecology, 44, 330-339. 25 1 Sample, B.E., Cooper, R.J., Greer, R.D. & Whitmore, R.C. (1993) Estimation of 2 insect biomass by length and width. American Midland Naturalist, 129, 234- 3 240. 4 SAS Inst., 2000. SAS/STAT Users’ Guide, Version 8. SAS Institute, Cary. 5 Schimmel, J. & Granström, A. (1996) Fire severity and vegetation response in the 6 7 8 9 10 11 12 13 14 15 16 boreal Swedish forest. Ecology, 77, 1436-1450. Southwood, T.R.E. & Henderson, P.A. (2000) Ecological Methods. Blackwell, Oxford. Spidsø, T.K. & Stuen, O.H. (1988) Food selection by Capercaillie chicks in southern Norway. Canadian Journal of Zoology, 66, 279-283. Sprugel, D.G. (1983) Correcting for bias in log-transformed allometric equations. Ecology 64, 209-210 Steven, H.M. & Carlisle, A. (1959) The Native Pinewoods of Scotland. Oliver & Boyd, Edinburgh. Storch, I. (1993) Habitat selection by Capercaillie in summer and autumn: is Bilberry important? Oecologia, 95, 257-265. 17 Storch, I. (1994) Habitat and survival of Capercaillie Tetrao urogallus nests and 18 broods in the Bavarian alps. Biological Conservation, 70, 237-243. 19 Storch, I. (2001) Tetrao urogallus Capercaillie. BWP update, 3(1), 1-24. 20 Summers, R.W., Proctor, R., Raistrick, P. & Taylor, S. (1997) The structure of 21 Abernethy Forest, Strathspey, Scotland. Botanical Journal of Scotland, 49, 39- 22 55. 23 Summers, R.W., Mavor, R.A., MacLennan, A.M. & Rebecca, G.W. (1999) The 24 structure of ancient native pinewoods and other woodlands in the Highlands of 25 Scotland. Forest Ecology and Management, 119, 231-245. 26 1 Summers, R.W., Proctor, R., Thornton, M., & Avey, G. (2004) Habitat selection and 2 diet of the Capercaillie Tetrao urogallus in Abernethy Forest, Strathspey, 3 Scotland. Bird Study, 51, 58-68. 4 Wegge, P., Olstad, T., Gregersen, H., Hjeljord, O. & Sivkov, A. V. (2005) 5 Capercaillie broods in pristine boreal forest in northwestern Russia: the 6 importance of insects and cover in habitat selection. Canadian Journal of 7 Zoology, 83, 1547-1555. 8 9 10 Wohlgemuth, T., Burgi, M., Scheidegger, C., Schutz, M. (2002) Dominance reduction of species through disturbance - a proposed management principle for central European forests. Forest Ecology and Management, 166, 1-15. 11 27 1 2 28 1 2 3 29 1 2 3 30 1 2 a b 3 4 5 6 7 8 9 10 cc 11 Figure 1. Experimental location and design. (a) Part of Scotland, showing 12 Cairngorms National Park (black line) and Abernethy Forest reserve (black shading). 13 (b) Central part of Abernethy Forest reserve, showing reserve boundary (black line), 14 the forested area (grey), and study blocks (black circles). (c) Air photograph of one 15 experimental block in open forest, after treatment, showing the three plots. (d) 16 Schematic diagram of one plot, showing the quadrats used for recording (grey). 17 18 31 1 Figure 2. Vegetation changes during the experiment: (a) bilberry cover; (b) heather 2 cover. Boxplots show median (central line), quartiles (box), and 5th and 95th centiles 3 (whiskers). Black dots indicate means. White bars: control plots; grey bars: burnt 4 plots; striped bars: mown plots. An arrow indicates the timing of the experimental 5 treatments. 6 32 1 2 3 Figure 3. Fitted bilberry cover, by treatment, three years after treatment (back- 4 transformed means and standard errors). (a) As recorded in the experiment (eqn 1), 5 including the effects of unusual, natural heather die-back in the control areas. (b) As 6 modelled under a ‘no die-back scenario’ (eqn. 2), estimating treatment means for a 7 hypothetical situation without natural heather die-back. Key as Fig. 2. 33 1 2 3 4 Figure 4. Estimated biomass caught in pitfall traps, of arthropods important in 5 capercaillie chick diet, by treatment, in June each year. (a) Spiders; (b) beetles; (c) 6 lepidopteran caterpillars; (d) ants. Note log scale for ants. Key as Fig. 2. 7 34 1 2 3 Figure 5. Significant treatment effects on post-treatment pitfall trapped biomass of 4 arthropods important to capercaillie chicks (back-transformed fitted means and 5 standard errors). (a)-(c): Results as recorded in the experiment (eqn 3); (d) results 6 from the ‘no heather die-back’ scenario (eqn 4). (a) and (d) spiders; (b) beetles; (c) 7 ants. Key as Fig. 2. 8 35 1 2 3 Figure 6. Grouse (mainly capercaillie) dung counts and dung detection estimates. (a) 4 Detection rate of dummy dung groups, three years after treatment. (b) Estimated 5 detection probability of a median grouse dung group (see text). (c) Grouse dung 6 counts per plot. (d) Detection-compensated modelled grouse dung counts, post- 7 treatment (eqn 5). (b) and (d) are for non-zero block-years only. Means and standard 8 errors. Key as Fig. 2. 9 10 36 1 Supplementary material: method for estimating dung- 2 detection rates 3 4 Here, we describe how dung detection rates Ri,j,t in plots with treatment j in block i in 5 year t (as used in eqn 5 in the main paper) were estimated. This was necessary 6 because we could not assume that dung groups were detected perfectly. Dung 7 detection rates could be expected to be a function of vegetation openness, which in 8 turn is influenced by the treatments. Thus, it was necessary to correct for differences 9 in detection probabilities between treatments in order to avoid biases in estimates of 10 treatment effects on dung densities. 11 12 We obtained independent estimates of detection rates, by carrying out a detection 13 trial. Dummy dung pellets, made of brown-stained pine dowel, were placed in 14 quadrats, and their detection rate measured. The number and diameter of pellets in 15 each dummy group were those of a randomly-selected real dung group, found on 16 previous surveys. Shortly before the May 2005 dung survey, 300 dummy groups 17 were placed at random locations within the search areas of randomly-selected 18 quadrats. The vegetation openness (see above) of each dummy group location was 19 measured. The details of any dummy dung groups located during subsequent standard 20 searches were recorded. 21 22 Let Yi be the response variable, indicating whether dung group i was detected (Yi=1) 23 or not detected (Yi=0). We modelled the probability that a dummy dung group i was 24 detected (P(Yi=1)), as a function of the covariates that we expected to influence 25 detection probabilities, namely: the size of dung group i (number of pellets in the 37 1 group, loge-transformed), the pellet-size of dung group i (diameter of a typical pellet 2 in the group, loge-transformed), and vegetation openness (see methods section of the 3 main paper), as measured at the location that dung group i was placed. 4 5 The relationship between detection probabilities and these covariates was estimated 6 using generalized linear models with a binomial distribution for the Yi, and a logistic 7 link function for the covariates. Nineteen models were fitted, including a null model, 8 and all possible combinations of the three covariates and their first and second order 9 interactions. The model with the lowest AICc (the ‘best approximating model’: 10 Burnham & Anderson 2002) was used to estimate detection probability, in all plots 11 and years, of a ‘median dung group’. This was defined as a dung group with the 12 median number and diameter of pellets of recorded real dung groups. The best model 13 was as follows: 14 15 Logit(P(Yi=1)) = a0 + a1*Ni + a2*Pi + a3*Vi + a4*NPi, 16 17 with Pi being the (loge-transformed) pellet-diameter of dung group i, Ni the (loge- 18 transformed) number of pellets in group i, Vi the vegetation openness score in the 19 neighbourhood of group i, and NPi the product of pellet number and diameter. 20 21 We used this model to estimate detection rates in each of the plots and years, using the 22 median observed dung pellet-diameter P_med (9 mm), the median group size N_med 23 (one pellet), and the measured vegetation openness at each of the plots in each year 24 Vi,j,t, as follows: 25 38 1 Ωi,j,t= a0 + a1*N_med + a2*P_med + a3*Vi,j,t + a4*N_med*P_med. 2 3 Ri,j,t =exp(Ωi,j,t) / (1 + exp(Ωi,j,t)) 4 5 REFERENCES 6 7 8 Burnham, K. P. & Anderson, D. R. (2002) Model Selection and Multimodel Inference. Springer, New York. 9 39