Research / Methods
Study Areas and Bird Censuses
We studied canopy birds at two lowland rainforest sites, one in Pico Bonito National Park in northern Honduras (DLA), the other in central Amazonia near Manaus, Brazil (LNN), during two independent research projects. Detailed descriptions of the study areas and census methods are in Anderson (2009) and Naka (2004). Briefly, both sites lie at elevations below 350 m. Forest structure is similar, characterized by a closed canopy reaching to approximately 35 m, with abundant epiphytes and lianas. Annual rainfall averages 2900 and 2400 mm for Pico Bonito and Manaus, respectively, and is seasonal, with a pronounced dry season of approximately 3–5 months.
To make the data sets from the original studies more strictly comparable, we made minor changes to data summarization for the present study. We therefore provide a brief overview of census methods before discussing the standardization of data. The principal method for studying birds in Honduras and Brazil was censuses from canopy-based viewpoints, following the protocol of Loiselle (1988). In Brazil, canopy viewpoints consisted of three canopy towers separated by 10 to 45 km. Three censuses were made monthly from each tower over a complete annual cycle from November 1999 to November 2000. In Honduras censuses were made from 30 trees within a 100-ha plot from April 2006 to April 2007. Additional observations were obtained from 30 point-count stations along two ground transects. Canopy censuses began 30 min after sunrise and lasted 3 hr, during which we recorded all birds seen or heard in the forest canopy within 150 m of the observer (Loiselle 1988).
The data used in analyses are the maximum number of individuals and species observed per 3-hr canopy census. For point counts from the ground in Honduras, numbers of individuals and species for all points covered in a single walked transect are summed. For Honduras birds only, we categorized all detected, into one of four forest strata: (1) ground (soil, leaf litter, and fallen logs), (2) understory (the space from the ground to 2 m), (3) midstory (the space between the understory and canopy), and (4) canopy (the sum of all tree crowns exposed to the sky above; Figure 1C in Bongers 2001).
Our analyses exclude nocturnal species, aerial foragers (swifts, swallows), and scavengers (vultures) because these bird species were observed solely as flyovers. To facilitate comparisons at the assemblage level, we assigned all bird species to one of six major feeding guilds: (1) raptors, (2) nectarivores (exclusively hummingbirds), (3) frugivores (diet includes a substantial portion of fruit at least during some seasons, seeds not destroyed but presumably dispersed; Moermond and Denslow 1985), (4) granivores (seeds destroyed; parrots), (5) insectivores, and (6) omnivores (species that regularly feed on fruits, insects, nectar, and sometimes small vertebrates). We omit the guild insectivoreomnivore (Naka 2004) and include those species within the omnivore guild. Assignment to guild is based in part on Stiles and Skutch (1989), Terborgh et al. (1990), and Robinson et al. (2000), combined with our own personal observations.
One of our primary objectives was to distinguish the “core” members of the canopy assemblage, species that regularly breed in, winter in, or migrate through the forest canopy, from those species that are not characteristic of the forest canopy and that occur as visitors from lower levels of the forest, as visitors from nonforest habitats, or as vagrants (Remsen 1994). For Honduras, we used census data to quantify birds´ preference for the canopy stratum with the method of Neu et al. (1974), which compares the observed frequency of use of a given resource or habitat with an expected frequency derived from the available proportion of that resource or habitat. To maintain a 95% confidence level, we used Bonferonni’s adjustment to set confidence limits around the observed frequency of detection in the canopy stratum for species with ≥4 detections. A significant preference for the canopy was indicated by expected values below the 95% confidence limits for the observed values (Haney and Solow 1992, da Silva et al. 1996), and we refer to species that met this criterion as the core canopy species. Furthermore, we assigned numeric values (ground = 0, understory = 1, midstory = 2, canopy = 3) to the four strata defined in Honduras so that we could calculate a stratum average for each species. These procedures could not be applied to Brazil, where detections below the forest canopy were not recorded. Instead, we adopted the list by Cohn-Haft et al. (1997) of residents having the forest canopy as their preferred habitat. Because the methods for defining core canopy species in Honduras and Brazil differ, we attempt no quantitative comparisons of core canopy species (e.g., richness, abundance distributions) at the two sites.
Video of Honduras Birds in the Pico Bonito Rainforest
We rarefied rates of species accumulation to compare species richness in canopy assemblages. Rarefaction curves are derived from repeated and random resampling of the pool of observations and plotting the average number of species represented by n individuals; they are therefore a statistical representation of species-accumulation curves (Gotelli and Colwell 2001, Magurran 2004). We calculated Chao 1 and Chao 2 nonparametric estimators (Magurran 2004) to estimate species richness from each study area. Chao 1 is an abundance-based estimator that relies on the number of species represented by a single individual to estimate species richness, whereas Chao 2 is an incidence-based estimator that uses the number of species detected in a single sample to estimate richness. For these calculations we used Estimates version 7.5 (Colwell 2005). We used the inverse of the Simpson index (1/d) to characterize the evenness of species in Honduras and Brazil (Smith and Wilson 1996, Magurran 2004).
First, we calculated this index for the data from Honduras and Brazil, respectively, as an approximation of evenness for the overall assemblage of canopy birds. We then used a randomization procedure to obtain confidence limits around the overall values. Specifically, we bootstrapped individual daily censuses until we obtained a sample that contained the same number of censuses as constituted the original empirical sample. We repeated this process 1000 times to obtain 95% confidence limits around the index of diversity. Bootstrapping and randomization were done in R (R Development Core Team 2008).
We used a new approach to determine the 20 dominant canopy species in Honduras and Brazil, a recurrent theme in descriptions of avian assemblages (Loiselle 1988, Karr et al. 1990, Robinson et al. 2000, Naka 2004). Ideally, dominance is described in terms of the relative density of individuals and biomass (Terborgh et al. 1990, Robinson et al. 2000, MacKenzie et al. 2006), although various proxies have been used in the absence of these data, including percentage of overall detections (Blake 2007), total number of detections (Loiselle 1988), mean number of individuals detected per census (Naka 2004), and frequency of detection (Naka 2004). To standardize comparisons of the two sites, we used a procedure that takes into account two such measures of relative abundance: frequency of observation and average number of individuals per observation. Specifically, we multiplied the mean number of individuals per survey and the proportion of surveys in which a species was detected and ranked species by the product. This measure more accurately accounts for the difference between species that are observed regularly in small numbers and species observed infrequently but in larger numbers, usually in single-species flocks. We compared the observed composition of dietary guilds, numbers of edge species, and numbers of migrant species in each group of core canopy species with null distributions drawn from each regional pool of species through a randomization procedure. At each site we conservatively defined edge species as those found in both continuous forests and forest edges, in gardens, or in semi-open and nonforest habitats.
We made these determinations primarily on the basis of personal experience and Stotz et al. (1996). To assemble each regional pool of species we considered all species of possible occurrence in the canopy of primary forest, excluding terrestrial, aquatic, and aerial species as well as regional species not known to frequent primary forests. For Honduras we considered those species found below 350 m in Pico Bonito National Park (Bonta and Anderson 2002), and for Brazil we considered species listed by Cohn-Haft el al. (1997) as occurring in the Biological Dynamics of Forest Fragments Project (BDFFP) north of Manaus. We used a bootstrapping procedure to randomly draw a number of species from a given regional pool equal to the number of species in the region’s group of core canopy species. We then tallied the number of edge species, migrants, and species in each dietary guild and repeated this procedure 1000 times to obtain confidence estimates around a randomly generated assemblage composition. We inferred a result to be significant when the observed values fell above or below 95% of the null values. For some analyses we desired a balanced comparison of equal survey effort in Brazil (117 canopy censuses) and Honduras (56 canopy censuses). For this purpose we narrowed the Brazil data set to 56 censuses by selecting those censuses whose Julian dates most closely matched those of the corresponding canopy censuses in Honduras. Analyses that used this restricted data set are noted below.
Previous studies have shown that the great variety of social systems of tropical birds necessitates that a variety of methods be used to estimate population densities, that correcting for observation biases in avian communities with such high species richness is complex and not possible for all species present, and that relationships between the true population density of a species and estimates derived from such methods remain unclear (Terborgh et al. 1990, Robinson et al. 2000). We acknowledge that our comparison of distinct avifaunas of distant sites will introduce bias in density estimation. We emphasize that an attempt at correcting density estimates for a limited number of species under these circumstances would not fully rectify the problem of detection biases in the assemblages under consideration, nor would it allow us to fully address the structure of whole avian assemblages as proposed. Finally, an additional focus of our study was a comparison with the results of the two remaining canopy-based studies of canopy bird assemblages from Panama (Greenberg 1981) and Costa Rica (Loiselle 1988) for which no corrections would be possible. We instead adhere to the use of detections as a proxy for population density (Greenberg 1981, Loiselle 1988, Karr et al. 1990, Robinson et al. 2000, Naka 2004, Blake 2007), and we restrict our comparisons of the data to broad analyses of general patterns that we believe reflect taxonomic and functional patterns of real assemblages and broad-scale biogeographic patterns that are the result of structuring mechanisms operating at the assemblage level.