The study was conducted at the União Biological Reserve (União), located in the state of Rio de Janeiro, southeastern Brazil (22°25′30″S, 42°02′30″W). The reserve was created in 1998 to protect lowland Atlantic Forest remnants and the endangered Golden Lion Tamarin (Leontopithecus rosalia L.). União covers an area of 2548 ha, and the relief is comprised of alluvial plains, small rounded hills and mountain ranges with a maximum height of 376 m a.s.l. The climate is tropical wet. Mean annual precipitation is 1690 mm, with a moderate dry season between April and September (Oliveira 2002), and the mean annual temperature is 22 °C. The vegetation is comprised of lowland Atlantic rain forest (sensu Oliveira-Filho & Fontes 2000). During the last century, before the creation of the reserve, valuable native timbers were harvested from the forests of the formerly called União Farm, and some areas were covered with plantations of Corymbia citriodora (Hook) Hill & Johnson (formerly Eucalyptus citriodora). However, Rodrigues (2004) found that some forest stretches of the reserve have a high basal area and a very high richness of late-successional tree species, both of which are indicative of old-growth forests (Clark 1996).
The reserve is crossed by a gas pipeline and an electrical power line that pass through two distinct linear canopy openings within the forest (Fig. 1). The gas pipeline opening was created in the early 1980s, thus being ca. 25 yrs old by the time sampling was conducted (see 'Introduction' below), and is ca. 25-m wide. This opening is covered by grassy, relatively homogenous vegetation, which rarely exceeds 1 m in height. Vegetation management through manual clear-cutting is done along this canopy opening at least once every year, to prevent the establishment of deeply rooting woody species that would damage the buried gas pipeline. The power line opening was created in the early 1960s, thus being ca. 45 yrs old when sampled. This opening is ca. 100-m wide, and is covered by dense secondary vegetation composed of shrubs and scattered pioneer trees, mainly Cecropia sp. The weedy fern Pteridium arachnoideum (Kaulf.) Maxon forms dense, fairly homogenous, ca. 1.5-m high stands in many parts of the opening. This opening was subjected to annual clear-cutting until the year 2000, to avoid damage to the electricity towers. Furthermore, fire events were reported, as supported by the occurrence of dense stands of P. arachnoideum, an indicator species of fire-disturbed habitats (Martini et al. 2007b). Since 2000, vegetation was allowed to regenerate in some stretches of this canopy opening because the electricity towers there are high enough (e.g. on hilltops) to prevent forest regrowth reaching the wires. This caused the managed area within the canopy opening to decrease by some 40%, from 44.2 to 26.5 ha. Thus, linear canopy openings of the gas pipeline and power line differ in terms of age (25 vs 45 yrs), width (25 vs 100 m) and vegetation cover (grassy vs shrubby).
Figure 1. Map of União Biological Reserve. Stars, power line forest edge plots; triangles, gas pipeline forest edge plots; and squares, forest interior plots.
Download figure to PowerPoint
We sampled sapling, shrub and treelet communities in forest edges adjacent to the two linear canopy openings and in forest interiors far from any edge (Fig. 1). We selected 12 sites of mid-slope forest with no evidence of natural or human disturbance. Four of these sites were immediately adjacent to the gas pipeline canopy opening; four close to the power line; and the remaining four were located in forest interior areas at least 400 m distant from any edge (Fig. 1). In each of these sites, a 20 m × 50 m permanent sample plot was established. Gas pipeline and power line plots were placed completely within the forest, with one side close (1–5 m) to an imaginary line between closed-canopy and open areas. While allocating sampling units, we tried as much as possible to standardize the topographic position and forest physiognomy of the sampled areas. The sampling criteria constrained available sites for placing edge plots, in such a way that most plots belonging to the same edge type tended to be clustered in the landscape. Nevertheless, the minimum distance between any two plots was 100 m. All sampled forest edges had a similar, southeast-facing orientation.
We conducted the sampling of understorey communities in 2006. We sampled all treelets with DBH ≥ 5 cm and <10 cm within plots. Saplings and shrubs with DBH ≥ 1 cm and <5 cm were sampled in ten 5 m × 5 m subplots distributed within each plot. For this purpose, each plot was divided into 10 columns of four 5 m × 5 m subplots. In each column, one subplot was randomly assigned. Thus, the total sampled area for treelets (DBH 5–10 cm) was 1.2 ha, and for saplings and shrubs (DBH 1–5 cm) was 0.3 ha. Each individual had its DBH measured and was identified to species level whenever possible.
To evaluate whether the sites showed differences in species richness irrespective of differences in plant density (Gotelli & Colwell 2001), we plotted a rarefaction curve of the observed number of species in each treatment (i.e. either gas pipeline, power line or forest interior) as a function of the number of individuals sampled using the software EstimateS (Colwell 2006). Curves and respective 95% confidence intervals were plotted for each size class separately. We checked for significant differences in observed species richness between treatments by comparing the 95% confidence intervals.
Partial redundancy analysis (pRDA; Legendre & Legendre 1998) was used to decompose the variation associated with treelet and sapling/shrub abundance data into two and three sets of explanatory variables, respectively. Sapling/shrub and treelet abundance data were normalized prior to analysis (see Legendre & Gallagher 2001). When saplings/shrubs were the response data set, the variation was partitioned into three sources: treelets (T), space (S) and edge effect (E). When treelets were the response data set, space and edge effect were used as explanatory data sets.
Edge effect, a qualitative vector with three states (gas pipeline, power line and forest interior), was coded as a dummy variable (Legendre & Legendre 1998), and the resulting matrix was used as an explanatory data set (see Borcard et al. 2011). The spatial coordinates x and y were expanded to comprise nine terms (x, y, x2, xy, y2, x3, x2y, xy2, y3) in order to capture more complex structures (Borcard et al. 1992). Only species that occurred in at least eight sampling units for the sapling/shrub (only response) data set and five sampling units for the treelet (explanatory or response) data set were used in the analyses in order to ensure more robust estimation of regression parameters. This difference was set because the number of variables in the response data set must be lower than N–1 (in our case, 12–1 = 11 sampling units), and lower than the number of variables in the explanatory data sets (see Legendre & Legendre 1998). We applied a forward model selection with the ordistep function in R to select explanatory variables maximally related to each response data set (R Foundation for Statistical Computing, Vienna, AT). The general algorithm proposed by Økland (2003) was used to decompose sapling/shrub or treelet abundance data into three and two sources, respectively.
For saplings/shrubs as response data set, there were 7 = 23−1 components. The analysis began by calculating the total variation explained (TVE) by all sets of explanatory variables, i.e. V(S∪E∪T). Then, we calculated the partial terms. First, we calculated the first-order partial terms: V((T)|S∪E), sapling/shrub variation unique to treelets, i.e. constrained ordination between normalized sapling/shrub abundance data and normalized treelet abundance controlling for the combined effect of space and edge; V((S)|T∪E), variation unique to space, controlling for the combined effect of treelets and edge effect; V((E)|S∪T) variation unique to edge effect, controlling for the combined effect of space and treelets. Then, we calculated the second-order partial unions for each combination of two data sets: V((T∪S)|E), V((E∪S)|T), V((T∪E)|S). The shared variation was assessed by subtracting the combined variation from the sum of the unique terms. For the shared variation between treelets and space, the calculation was V((T∩S)|E) = V((T∪S)|E)−V((T)|S∪E) + V((S)|T∪E); for shared variation between edge effect and space was V((E∪S)|T)−V((E)|S∪T) + V((S)|T∪E); and for the shared variation between treelets and edge effect was V((T∪E)|S)−V((T)|S∪E) + V((E)|T∪S). The unexplained variation (UV) was calculated as 1−TVE. Finally, the shared variation by all sets of explanatory variables was calculated by subtracting from TVE the sum of first- and second-order partial intersections.
For treelets as response data set, there were 3 = 22−1 components. TVE was calculated as V(S∪E). First-order partial terms were calculated as the variation unique to edge effect V(E|S), and the variation unique to space V(S|E). In order to calculate the shared variation, the sum of these unique terms variation was subtracted from TVE. Unexplained variation was calculated as 1−TVE.
We classified all sampled species into two groups according to their conservation importance. Species were classified either as (1) disturbance-tolerant, i.e. native pioneer or generalist species and exotic species; or (2) disturbance-sensitive, which includes native late-successional and habitat specialist species. This classification was based on data from published ecological, floristic and phytosociological studies; on information from herbarium records; and on our previous field experience within the Atlantic Forest (see Sansevero et al. 2011). Then simple and partial Mantel tests were used to relate abundance of disturbance-tolerant, disturbance-sensitive and Euterpe edulis palm to edge effect (after dummy variable coding) and space (x and y coordinates). Significance of correlations was tested by means of bootstrapping with 999 permutations.
Constrained ordinations and Mantel tests were run in the R environment with packages VEGAN and STATS. For Mantel tests, Euclidean distance was used as distance measure for all data, except for edge effect, for which distances were calculated using the binary method in function dist from the package Stats. We adopted a P-value of 0.1 in all analysis to avoid type II errors, as proposed in Mesquita et al. (1999).