Research was conducted in lowland Amazonia within the Tapajós National Forest (TNF), a 560 000-ha area managed by the Instituto Brazileiro de Meio Ambiente e Recursos Naturais Renováveis. The TNF is in the Tapajós area of endemism, one of eight such recognized areas in Amazonia (da Silva, Rylands & Fonseca 2005). The Tapajós area of endemism comprises 648 862 km2, has lost only 9·3% of its forest cover, and is among the least threatened regions of the Amazon. However, this estimate of forest loss does not include selectively logged areas, and thus is an underestimate of overall impact. Over 28% of the Tapajós area of endemism enjoys some level of protection. However, only 0·7% of land is protected strictly, with the rest dedicated as indigenous reserves (24·1%) or subject to sustainable use (3·5%).
The TNF (3·36° S, 54·95° W) is located on the east bank of the lower Tapajós River in western Pará, Brazil (Fig. 1). The climate is tropical, with mean monthly temperatures between 24·3 and 25·8 °C (Silva 1989). Rainfall is substantial (≈1920 mm per year), with most precipitation occurring from December to May, and a modest dry season from August to October. Several distinct moist and wet forest types occur in the TNF, with terra firme forest constituting 33% of forested lands and encompassing the entire study area. Terra firme is characterized by gently rolling terrain on poor upland soils (dystrophic yellow latosol; Silva 1989). Canopy height ranges from 30 to 40 m, with emergent trees reaching 50 m.
A system of roads and trails at km 83 of the Santarém–Cuiabá Highway provided access to forest, including four 100-ha experimental blocks. These blocks were within a 5000-ha grid established as a demonstration for logging practices in terra firme forest. Two control blocks were undisturbed forest and two cut blocks were subjected to RIL. Control blocks were adjacent to each other and were 1–2·3 km from cut blocks. Cut blocks were separated from each other by 2·5 km of selectively logged forest.
Timber harvest in cut blocks was completed in December 1997, 20 months before initiation of the study. In addition to minimizing damage related to felling, skidding or log processing, RIL techniques that were applied to cut blocks harvested fewer trees and a reduced volume of wood (<19·0 m3 ha−1) compared with forest subjected to traditional selective logging (≈40 m3 ha−1). In RIL forest, an average of 3·94 (18·70 m3) and 3·79 (18·73 m3) trees per hectare were removed from cut blocks. All trees harvested were >45 cm dbh. Manilkara huberi, Manilkara paraensis, Protium pernevatum, Dinizia excelsa and Piptadenia suaveolens were the most commonly harvested species at TNF (Keller et al. 2004).
Two groups of understorey sites were selected based on habitat physiognomy: treefall gaps and closed canopy (henceforth gap and closed canopy, respectively). Sixteen gaps formed by natural treefalls and 16 closed-canopy sites were selected in each control block; similarly, eight gaps resulting from logging of individual trees and eight closed-canopy sites were selected in each cut block (Fig. 2). All sites were >30 m from the edge of the block. Each block was divided into four quadrats, and an equal number of gaps and closed-canopy sites were located in each quadrat to ensure dispersion of sites throughout each block. To enhance comparable sampling of variation within quadrats, each gap was associated with a closed-canopy site (Fig. 2). Location of an associated closed-canopy site with respect to a particular gap site was random with respect to direction, and at a random distance between 25 and 50 m. Associated sites were closer to each other than they were to any other site. The size of gaps from tree harvest (Ȳ = 219 m2 ± 55 SE) in cut blocks were indistinguishable from the size of gaps that were formed naturally (Ȳ = 286 m2 ± 37 SE) in undisturbed forest (Wunderle, Henriques & Willig 2006).
Figure 2. Arrangement of net sites in 100-ha blocks of terra firme forest in Tapajós National Forest, Pará, Brazil. Each block was divided into four quadrats (delimited by dashed lines) and an equal number of sites (circles) were placed in gaps (open) and closed-canopy (solid) sites. (a) In each block of undisturbed forest, four closed canopy and four gap sites were located in each quadrat. (b) In each block of logged forest, two closed-canopy and two gap sites were located in each quadrat.
Download figure to PowerPoint
Sampling was organized into four time periods: June–July 1999, August 1999, November–December 1999, and April 2000 (Saldanha 2000). During each sampling period, each site was surveyed for one night. Depending on gap size and shape, 24 m of net in various combinations and configurations of 6- and 12-m segments were used to sample bats from gaps. Matching net configurations were used in corresponding closed-canopy sites. All mist nets (four-shelf, 2·6 m tall, 35 mm mesh) were erected at ground level and checked every hour from 18.00 to 01.00 h. Depending on intersite distances, two or three pairs of gap and closed-canopy sites within the same block were surveyed each night. Order of site selection within a block was random, and sites within each block were sampled in a temporal fashion (all sites of one block were sampled before sampling in another block). In addition, block order was randomized each sampling period. Nets were closed during heavy rain. If heavy rain occurred for more than 2 h of netting on any night, data from that night were excluded from analyses, and another night of netting was scheduled to replace lost sampling effort. To minimize effects of lunar phobia (Crespo et al. 1972) on sampling efficacy, netting was not conducted within two nights of a full moon.
Species identity, sex, reproductive condition, age (juvenile, subadult or adult), mass and standard morphometric measurements were recorded for each bat captured. To facilitate identification of recaptures within the same sampling period, bats were marked with a small notch in the border of one of the pinnae; notch location was unique for each time period. Recaptures from previous sampling periods were not discriminated from new captures in subsequent sampling periods. This ensured that the spatiotemporal focus of the study, the average number of individuals (new captures) per species per sampling period, was not overestimated by counting the same individual on multiple occasions within a sampling period. Accurate field identification of bats was facilitated by collecting a series of voucher specimens from the area prior to the study, but not at the sites of actual field work (Saldanha 2000). This synoptic series is deposited in the Museu Paraense Emílio Goeldi (Belém, Brazil). Nomenclature followed Simmons (2005) except in recognizing Artibeus planirostris instead of A. jamaicensis (Lim et al. 2004) as occurring in Amazonia. We classified bats into broad foraging ensembles (taxonomic subsets of a guild) based on published recommendations (Gardner 1977) that reflect the primary components of each species’ diet. Phyllostomids captured at TNF represented four foraging ensembles (Table 1), including sanguinivores, nectarivores, frugivores and gleaning animalivores (a composite of foliage-gleaning carnivores and foliage-gleaning insectivores).
Table 1. Ensemble association, familial classification, and number of captured individuals in each combination of management and physiognomy for each of 39 bat species in Tapajós National Forest, Brazil. Species richness, total number of captures, and a rarity threshold are presented for each combination of management and physiognomy. Rarity thresholds are specific to each combination of management and physiognomy, and equal the abundance of phyllostomids in that combination divided by phyllostomid richness in that combination. Species with abundances greater than or equal to the rarity threshold for a particular combination of management and physiognomy were considered common and are indicated by bold numbers. Conservation status: LR:nt, lower risk, near threatened; VU A2c, vulnerable, population reduction of at least 20% projected or suspected within 10 years in areas of occupancy, extent of occurrence, or quality of habitat
|Subfamily/species||Feeding ensemble||Conservation status||Control||Cut||Total|
|Desmodus rotundus||Sanguinivore|| ||2||2||0||0||4|
|Choeroniscus godmani||Nectarivore|| ||0||1||0||0||1|
|Choeroniscus minor||Nectarivore|| ||0||3||0||0||3|
|Glossophaga soricina||Nectarivore|| ||4||4||4||11||23|
|Lichonycteris obscura||Nectarivore|| ||0||1||0||1||2|
|Lonchophylla thomasi||Nectarivore|| ||5||30||5||16||56|
|Chrotopterus auritus||Gleaning animalivore|| ||0||3||0||0||3|
|Glyphonycteris sylvestris||Gleaning animalivore||LR:nt||0||1||0||1||2|
|Lampronycteris brachyotis||Gleaning animalivore|| ||1||0||0||0||1|
|Lophostoma carrikeri||Gleaning animalivore||VU A2c||1||0||0||0||1|
|Lophostoma silvicolum||Gleaning animalivore|| ||4||14||1||11||30|
|Micronycteris hirsute||Gleaning animalivore|| ||1||1||0||0||2|
|Micronycteris megalotis||Gleaning animalivore|| ||2||2||0||2||6|
|Mimon crenulatum||Gleaning animalivore|| ||0||6||0||0||6|
|Phylloderma stenops||Gleaning animalivore|| ||1||1||0||2||4|
|Phyllostomus discolor||Gleaning animalivore|| ||0||1||2||5||8|
|Phyllostomus elongatus||Gleaning animalivore|| ||6||8||0||2||16|
|Tonatia saurophila||Gleaning animalivore|| ||8||18||1||10||37|
|Trachops cirrhosus||Gleaning animalivore|| ||2||2||0||0||4|
|Trinycteris nicefori||Gleaning animalivore|| ||3||1||1||0||5|
|Carollia brevicauda||Frugivore|| ||11||7||15||12||45|
|Carollia perspicillata||Frugivore|| ||152||172||71||138||533|
|Rhinophylla pumilio||Frugivore|| ||8||22||3||2||35|
|Ametrida centurio||Frugivore|| ||2||0||0||0||2|
|Artibeus gnomus||Frugivore|| ||7||6||9||3||25|
|Artibeus litratus||Frugivore|| ||57||98||13||72||240|
|Artibeus planirostris||Frugivore|| ||3||10||0||3||16|
|Chiroderma trinitatum||Frugivore|| ||0||4||0||3||7|
|Chiroderma villosum||Frugivore|| ||0||1||0||1||2|
|Mesophylla macconnelli||Frugivore|| ||3||1||0||0||4|
|Platyrrhinus helleri||Frugivore|| ||0||2||3||7||12|
|Sturnira lilium||Frugivore|| ||1||0||0||1||2|
|Sturnira tildae||Frugivore|| ||1||3||0||0||4|
|Uroderma bilobatum||Frugivore|| ||4||10||6||5||25|
|Vampyressa thyone||Frugivore|| ||0||2||0||0||2|
|Phyllostomid richness|| || ||28||35||16||25||39|
|Total phyllostomid abundance|| || ||327||543||153||356||1379|
|Rarity threshold|| || ||11.7||15.5||9.6||14.2||35.4|
Because all mist nets were placed at ground level, the presence or abundance of some taxa, particularly species in the Emballonuridae, Natalidae, Vespertilionidae and Molossidae, may be underestimated in multistrata tropical rainforests. This problem is not universal, as many species of emballonurids and vespertilionids may be captured at ground level more frequently than in elevated nets (Peters, Malcolm & Zimmerman 2006). To minimize such complications, we restricted analyses to populations of phyllostomid bats. All sampling methods, including those for volant mammals, involve some degree of species-specific bias (capture probabilities are not the same for all taxa, especially for those with different sensing or locomotor modalities). This is particularly problematic when estimating community-level characteristics such as species diversity or evenness, as differences in abundance or catchability among species are reflected in metrics of biodiversity that weight species presence by relative abundance. Such concerns are less critical for comparisons of intraspecific metrics such as abundance, because interspecific differences in sensing or locomotor modalities are not germane. The important assumption when comparing abundances is that the biases associated with a sampling method are equivalent in different levels of a treatment factor (e.g. cut vs. control forest). We use the term ‘abundance’ to refer to the number of captures for each species because of its ease of exposition, with the understanding that variation in the number of captures at a site can be a consequence of the catchability, habitat use or density.
For each of the 17 most common species of phyllostomid in TNF, a generalized linear mixed-effects model (GLMM; Venables & Ripley 2002) with the assumption of Poisson errors quantified the effects of management (cut vs. control forest), forest physiognomy (gap vs. closed canopy), and their interaction on abundance (average number of captures per time period at each site). In all GLMMs, management and physiognomy were model I factors, and block (a model II factor) was nested within levels of management. Because of the nesting of blocks within levels of management, factors quantifying this nesting, as well as the interaction of physiognomy and block within management, were included in each GLMM. All GLMMs were executed in the R programming environment (R Development Core Team 2005) and the MASS and nlme libraries. We were interested in population-level responses of each species to management and physiognomy, rather than overall multivariate evidence of effects on the phyllostomid assemblage. Therefore results were interpreted without application of Bonferroni sequential adjustments (BSA). Considerable controversy surrounds the use of such adjustments, and the exploratory nature of this research argues against the use of such highly conservative approaches (Moran 2003; Roback & Askins 2005). Because of conservation and forest management implications, we were more concerned with detecting responses to management and forest physiognomy than we were with the potential for type I errors. Therefore we report exact P values without application of BSA, and discuss all responses that were significant (P ≤ 0·05) or approached significance (0·05 < P ≤ 0·10).
Because species abundances reflect the vulnerability of a species to extinction, species rarity is a topic of increasing interest in ecology and conservation biology (Kunin & Gaston 1997; Rodrigues & Gaston 2002; Hartley & Kunin 2004). We used a statistical metric of rarity (Camargo 1992) that considers a species to be rare if its abundance is less than the average abundance (n̄) of all species in an assemblage (<N/S, where N = total number of individuals in an assemblage and S = species richness). Because differences in the density or type of vegetation, as well as flowering or fruiting phenology, can affect local or site-specific capture rates, and because more pairs of sites (16 vs. eight) were sampled in control forest than in cut forest, we calculated n̄ij separately for each combination of management (i) and physiognomy (j). Subsequently, n̄ij was used as the rarity threshold for each respective combination of management and physiognomy: species with abundances < n̄ij were considered to be rare and species with abundances ≥ n̄ij were considered to be common.
Qualitative comparisons (common vs. rare; present vs. absent) of species between treatments were confounded by unequal numbers of captures and unequal numbers of sites (Table 1). For example, if five species were rare in control forest but absent from cut forest, these differences could be explained by differences in quality of habitat (actual differences in species associations) as well as by differences in number of captured individuals or number of sites sampled in each management type (sampling bias). To address sampling issues, we employed rarefaction (Heck, van Belle & Simberloff 1975) to standardize the number of individuals for each comparison. Rarefaction was based on 1000 bootstrapped samples of the number of individuals in the smaller sample. Similarly, we used an incidence-based approach (Colwell, Mao & Chang 2004) to account for differences in the number of sites between treatments as well as to estimate parametric values of richness for each combination of management and physiognomy. To compare the number of rare species in different combinations of management and physiognomy, three sets of rarefaction analyses were used to compare control forest with cut forest, control gaps with cut gaps, and control closed-canopy sites with cut closed-canopy sites. Rarefaction was conducted in MATLAB ver. 6·1 (2001) (Math Works, Inc., Novi, MI, USA) and analyses of incidence were conducted in EstimateS ver. 7·5 for Windows (Colwell 2005). We used these techniques to evaluate the influence of sampling on the ability to detect species presence in each management type, not in an attempt to estimate parametric measures of diversity. This approach established levels of confidence for the absence of species in particular habitats.