Study System and Design
The study was conducted in boreal forests in central Sweden in the counties of Västernorrland and Jämtland (mid point of the area: 62°40′ N 16°05′, see Hylander et al. 2005). To a large extent, the area is a managed landscape with coniferous forests on acidic bedrock (for more details see Hylander et al. 2005). The forests have been managed for timber production for more than 150 years. Clear-cutting did not become the main logging method until the 1950s and 1960s, which means that the forests subject to clear-cutting, in most cases, have not regenerated after clear-cutting (Östlund, Zackrisson & Axelsson 1997).
In 1998, we selected 15 stands, intersected by a small stream (0·5–1 m wide), which were planned for clear-cutting by the forest company SCA (Svenska Cellulosa Aktiebolaget). All the selected stands were multi-aged mature stands with a minimum time of 86–146 years since regeneration, measured by coring an average-sized tree in each stand. Dominant or frequent tree species were Norway spruce Picea abies, Scots pine Pinus sylvestis and birch Betula spp. Most stands had signs of human management such as old stumps from selective logging. However, all sites had a mature forest bryophyte community with no signs of recent disturbance. In each stand, two 1000-m2 plots (20 by 50 m) were established along the stream with the stream crossing the middle of the short side of the plots. We also established one plot each in 10 reference stands not intended for clear-cutting. The references were similar to the treatment stands in terms of tree species composition and no signs of recent disturbance, but were probably on average older, and a few of them had considerably more dead wood. This was not considered a major problem because the main aim with the references was to account for changes in inventory efficiency over the years. In the winter of 1998/99, the 15 stands were clear-cut according to our instructions, leaving a buffer strip (10 m on each side of the stream) along one half of each study site and removing trees from the other half of the site (see Hylander et al. 2005). These sites were re-inventoried in 2001 and again in 2009. In 2009, only 13 sites remained for evaluation because the buffer strip had been removed from two sites. Nine of ten reference plots remained, as one site had been logged.
In each plot, we recorded all bryophyte species within five subplots (10 × 20 m), giving us a frequency measure of 0–5 for each species in each plot. Similar to the first re-inventory in 2001, we had the previous species list with us in the field in 2009. In the two-first inventories, only one person (KH) carried out the bryophyte inventory; however, in the most recent inventory, the work was divided between two skilled bryologists (KH and HW), and there was no bias in the recorder effort between the plot types. As before, the inventory of one plot (20 × 50 m) by one person took half a day. We collected many small samples of species that needed microscopy for a correct identification. However, to ensure comparable species lists among localities and among years, we grouped certain species groups together owing to, for example, many young individuals (see Appendix S2, Supporting information). We followed the checklist of Swedish bryophytes from 2006 (Hallingbäck, Hedenäs & Weibull 2006).
General environmental data were collected in the first year (1998), and variables that were related to changes probably caused by the logging were collected in the first re-inventory (2001) (Hylander et al. 2005). In 2009, we followed the same protocol for recoding basal area of different tree species, cover of different vegetation layers [canopy, shrubs (woody species <5 m), understorey and bryophytes (on ground)], number of uprootings, cover of mineral soil and amount of woody debris. We measured the length of the logs and counted the number of stumps in two size classes (10–20 and >20 cm in diameter at the narrowest point) and two decay classes (soft or hard wood). In 2009, we estimated the cover and height of regenerating trees of different species.
The statistical software r (version 2.10.1) (R Development Core Team 2010) was used for all analyses except where noted. Differences in selected vegetation characteristics among years and/or treatments were tested with pairwise Wilcoxon signed-rank test because of non-normal distributions of values.
The extent to which the species composition had changed between the two re-inventories was analysed by comparing the Bray–Curtis dissimilarity index among treatments (clear-cuts, buffer strips and references) and years. We calculated these values using both the presence/absence data and species frequency data. The differences between the two post-logging years and among treatments were evaluated by two linear mixed-effect models (using the package nlme, version 3.1-103): one that included clear-cuts and buffer strips (with site and plot as random factors; plot nested within site) and one that included buffer strips and references (with only plot as a random factor because the references were not in the same streams). Assumptions of normality and equal variances were checked by inspection of residual plots.
To understand whether the species compositional change was related to changes in the environmental variables (mostly disturbance intensity) in the buffer strips, we also calculated the Bray–Curtis dissimilarity index between 1998 and 2009 for the buffer strips. Thereafter, we conducted Pearson's correlation tests between these values and environmental variables that described the changes that had happened since the buffer strips were retained (i.e. number of tip-up mounds, change in canopy cover, change in basal area and cumulative length of fresh logs).
To investigate the differences in main trajectories in the response pattern among the treatment groups, including local colonizations and extinctions and time-lags in such responses, we defined nine different response types (Fig. 3a). We assigned all species with a difference in occupancy of 0, +1 or −1 between the three inventories as belonging to the stable category. If a species had changed its occupancy with ≥2 sites between two consecutive inventories, it was assigned to increasing or decreasing categories between these years. However, in the case of a changed occupancy of one site in the first period and another site in the second so that the total change was two sites, it was also classified as increasing or decreasing. Species with that response patterns (+1, +1 or −1, −1) were divided in equal proportions between the categories 2 and 3 (for increasing species) and 7 and 8 (for decreasing species) when plotted (See Fig. 3 and Appendix S2, Supporting information). Species displaying the following response patterns (−2, +1; +2, −1; −1, +2, or +1, −2) were classified as stable because of unclear responses and a total change of only one site between the first and last inventory. This analysis is not used to calculate the absolute number of species with different response types, but rather to illustrate the different response patterns within and among the treatments. The overall difference in frequency distribution of response types among the three treatments was compared with two Chi-squared tests: one between buffer strips and references and one between buffer strips and clear-cuts.
The mean proportions of species (compared to the pre-inventory data) that were initially lost from the plots (2001) and that were still absent (2009) were compared with the proportion that displayed a delayed local extinction in two linear mixed-effect models (in the same way as the Bray–Curtis dissimilarity index described earlier). In a similar manner, the contrasts were analysed between the proportions of species that recolonized vs. those that were still absent among the species lost immediately after logging. The proportion of new species in the plots was analysed in two ways: proportion of species with an early vs. a time-lagged increase (among the species that had colonized between 1998 and 2009) and the proportion of species that had colonized first but then became locally extinct vs. those that were still present in the second inventory among the newly colonized species recorded in the first re-inventory. All analyses were carried out as for the Bray–Curtis dissimilarity index described earlier.
The relationship between the change in occupancy and the change in local frequency (i.e. the mean number of segments [1–5] that a species was found in, including only the occupied plots) was analysed by Pearson's correlation tests for the two periods and for species increasing or decreasing in occupancy (≥2 plots) separately.
We summed the number of subplot records (number of 200 m2 records) of red-listed species for each plot and analysed both this value and the number of species per plot across treatments and years. Hylander et al. (2005) found that the number of records of red-listed species in 200-m2 subplots was similar to the number of easy delimited small patches (in general <0·01 m2) of these species in each plot. In the analyses, we omitted all red-list species found only in the references (mostly species on rock) and a somewhat doubtful species found only in the last inventory (Scapania cf. glaucocephala). Moreover, the Swedish red-list has been revised twice since the first inventory; however, we have only included those species named in the most recent red-list in this study (Gärdenfors 2010), which means that three red-listed species in Hylander et al. (2005) were omitted from the analyses. The analyses of the number of red-list species and the red-list subplot records consider the occurrence patterns of the following seven species: Anastrophylum hellerianum, Calypogeia suecica, Lophozia ascendens, Lophozia longiflora, Lophozia polaris, Scapania apiculata and Tayloria tenuis. The difference between the years and among the treatments in numbers of species and records was tested by Wilcoxon signed-rank test because of small numbers, but the results were displayed as percentage changes in the means for pedagogical reasons (change data could not be analysed because some cells would be divided by 0). However, we also analysed the percentage change from the prelogging situation omitting sites without records in the prelogging data (Buffer strips: N = 11, Clear-cuts: N = 8 and References: N = 7). With this approach, we could compare the change of species occupancy and species records among the treatment groups.