Abstract

The efficacy of communication depends on the detection of species-specific signals in background noise that includes other species’ signals. To avoid confusion with each others’ signals, species should partition communication space. I investigated this possibility for the dawn chorus of birds in an Amazonian rain forest. Acoustic censuses at a location in Matto Grosso, Brazil, detected 82 sedentary species of birds that sang frequently during dawn choruses. Eleven features of these species’ songs were analyzed to characterize the acoustic space of this community. The Euclidean distances between species’ songs in this acoustic space were then used to investigate spatial, temporal, and phylogenetic influences on the divergence of songs. Songs of species in the same stratum of the forest and during the same 30-min interval had the most dispersed songs. Songs of congeners and family members were more dispersed than songs of random species. These results indicate that in this complex acoustic environment, species singing at the same place and time partition signal space. These species either choose times and places for singing to minimize acoustic interference from other species or they have evolved different songs to reduce this interference.

Species-specific signals convey important information to conspecifics that enable them to recognize each other, to make appropriate mate choice decisions and to settle territorial disputes ( Bradbury and Vehrencamp 1998 ). Acoustic interference from background noise should decrease the efficacy of intraspecific communication by affecting the detectability and discriminability of conspecific signals ( Endler 1992 ; Wiley 1994 ). Background noise from both biotic and abiotic sources is ubiquitous in natural environments. In addition, many animals communicate in aggregations, such as frog choruses and avian dawn choruses, that make it especially difficult to discriminate conspecific from similar heterospecific signals ( Bremond 1978 ; Wiley 1994 ; Pfennig 2000 ; Gerhardt and Huber 2002 ; Wollerman and Wiley 2002b ; Brumm and Slabbekoorn 2005 ). To increase opportunities for correct detection of conspecific signals, signalers should evolve signals that contrast with the background noise of their environment ( Miller 1982 ; Endler 1993 ; Wiley 1994 ; Brumm and Slabbekoorn 2005 ; Wiley 2006 ).

Heterospecific signals are a common source of background noise ( Schwartz and Wells 1983 ; Wollerman and Wiley 2002a ). Signals with similar features have the greatest chance of interfering with each other and producing errors by receivers. Such errors include responses to signals from different species, which could lead individuals to respond to inappropriate rivals or mates, or a lack of responses to appropriate signals, which could result in additional time and risks in finding a mate or confronting a rival ( Wiley 1994 ). To minimize these errors and thus to avoid acoustic interference, species should partition acoustic space, a multidimensional space with axes defined by acoustic features, such as dominant frequency, duration, number of notes, and other attributes that characterize the structure of a signal ( Marler 1960 ; Miller 1982 ).

Because closely related species tend to be physically and behaviorally more similar than distantly related species, they might also have more similar songs and singing behavior. As a result, acoustic interference might occur more often among closely related species. Closely related species might also suffer greater consequences from responding to each other's signals because these responses would more likely lead to hybridization ( de Kort et al. 2002 ). In fact, recent studies have found that song divergence among sympatric congeners is greater than among allopatric congeners ( Doutrelant et al. 2000 ; Haavie et al. 2004 ; Seddon 2005 ).

Similarity of species’ signals could also result from the density of species in acoustic space. Nelson and Marler (1990) found that the central area of a community's acoustic space is more crowded than the periphery. Species in crowded areas of acoustic space have signals that are similar to each other, with increased chances of interference, whereas those in less crowded areas of acoustic space should incur less acoustic interference ( Nelson and Marler 1990 ).

Dawn choruses of birds in the tropics provide an example of communication in the presence of high levels of heterospecific background noise. The combination of high species diversity and a narrow window of time in which the majority of species sing should increase possibilities for acoustic interference and limit possibilities for song divergence. Beyond the basic species specificity of their songs, we know little about how these songs are distributed in acoustic space and perceived in noisy acoustic environments.

In this study, I examined an avian community in a southern Amazonian rain forest to investigate the ways in which species might partition acoustic space to improve communication with conspecific receivers. I tested the following hypotheses: 1) species’ songs are overdispersed in acoustic space, 2) closely related species’ songs are more dissimilar than those of other species in the same community, 3) species that sing at the same location and time have songs that are more dispersed than those of other species, and 4) songs of species in the center of occupied acoustic space are more crowed than those of species on the periphery. Overdispersed signals would indicate that selection for unambiguous species recognition has promoted the evolution of song features to improve intraspecific communication, or that individual singing behavior is plastic, and individuals choose to sing in times and places that reduce acoustic interference.

METHODS

Study location and acoustic censuses

This study was based on acoustic censuses conducted during both the wet season (February and March) and the dry season (June and July) in 2004 at the Rio Cristalino Private Natural Heritage Reserve, 40 km northeast of the town of Alta Floresta, Mato Grosso, Brazil (9°41′S, 55°54′W). The reserve consists of uncut lowland tropical moist forest (see Zimmer et al. 1997 ). The censuses consisted of recordings replicated in both space and time in terra firma habitat. Censuses were conducted at 3 sites, separated by 500 m to 1 km. Each census consisted of simultaneous recordings at 2 points 100 m apart at one of these sites. These simultaneous recordings were conducted at each of the 3 sites on 3 mornings during the wet season and 4 mornings during the dry season of 2004. Each census began 30 min before sunrise and continued until 30 min after sunrise. Recording resumed at 0700 and again at 0800 for 30 min (total time recorded during each census = 2 h). The recordings were made with Sony TC D5 Pro II and Marantz PMD222 tape recorders and Shure 33-1070D omnidirectional microphones placed 2 ± 0.1 m above ground. Because the microphones were omnidirectional, they simply hung downward from a small branch. In this sampling design, recordings were replicated at both small (100 m) and medium (1 km) spatial scales and at short-term (1 h), medium-term (2 day), and long-term (seasonal) temporal scales. They thus capture both spatial and temporal variation in vocal activity.

Acoustic community

The acoustic censuses detected songs from 137 species, not including species that were recorded only while flying past the census points, such as parrots, hummingbirds, and nighthawks. The analyses included only species that sang during at least 1% of the total censused minutes (determined by vocal activity of a species during at least 52 of the 5276 min of total recording time) for a total of 82 species. The behavior and evolution of the other species might well have been affected by these more frequently singing species, but a sufficient number of clear recordings of their songs could not be obtained for the analysis. The excluded species presumably sang too infrequently to have much influence on the others. I analyzed one song pattern for each of the 82 species, with one exception. Male and female Buff-throated Woodcreepers, Xiphorhynchus guttatus , unlike other species in this study, had markedly different vocalizations and did not usually coordinate their songs. The analyses consequently included the complex vocalizations of both male and female X. guttatus as if they were 2 species for a total of 83 different song patterns. For species with multiple song patterns, such as Thryothorus genibarbis , I only used the pattern heard most frequently during the censuses. Because body size correlates with the frequency of vocalizations ( Ryan and Brenowitz 1985 ; Wiley 1991 ), my analyses included the mean mass for the male of each species as reported by Terborgh et al. (1990) and Dunning (1993) .

Acoustic analysis

To obtain clean examples of each species’ songs for analyses of their acoustic features, I used a Sennheiser ME66-K3U directional microphone and a Sony TC-D5 Pro II tape recorder or a Marantz PMD660 digital recorder. For species detected on the acoustic censuses, but not recorded with the directional microphone, I analyzed examples from the censuses. The tape recordings were digitized (16-bit accuracy, 22.05 kHz sampling rate, and WAV format) with WildSpectra2 (version 050415, http://www.unc.edu/∼rhwiley/wildspectra.html ).

One song each from 3 different individuals of the same species, identified by the locations of their territories, was analyzed with Wildspectra1 (version 051027; sampling rate of 22.05 kHz, frequency resolution 172 Hz, and temporal resolution 5.8 ms). When necessary, songs from individuals found at the same location but during different seasons or years were used for this analysis; although the censuses were only conducted in 2004, I visited the site and recorded bird songs over a 3-year period to maximize variation.

The SongSignature feature of Wildspectra1 was used to obtain measures of songs. This procedure involved manually deleting extraneous background sounds near but not overlapping with the notes of a song in a digitally generated spectrogram. When a box was dragged around the clean song, the program then computed the following measures from the selected song (time in milliseconds and frequency in Hz; Figure 1 ): 1) lowest dominant frequency, 2) highest dominant frequency, 3) overall dominant frequency, 4) song bandwidth (highest dominant frequency minus the lowest dominant frequency), 5) total number of notes, 6) song duration, 7) note rate (total number of notes divided by song duration), 8) change in note rate (ratios of the rates in each third of a song; Isler et al. 1998 ), 9) complexity of the first note (the bandwidth of the note divided by the duration of the note, in turn divided by the number of inflections in the note), 10) complexity of the last note, and 11) complexity of the average note (the mean of measurements of the complexity of 3 successive notes in the first, middle, and last third of each song; see Supplementary Material , Appendix A). The number of inflections in a note was determined from spectrograms by eye. The measure of note complexity was created to quantify the shape of the note, which might be important for species recognition. All subsequent analyses used the means of the measurements of each acoustic feature across the 3 individuals of each species.

Figure 1

Spectrograms of the songs of (a) Thamnophilus schistaceus , (b) Jacamerops aureus , (c) Schiffornis turdina , and (d) Crypturellus cinereus to illustrate measures of acoustic features. The x axis is time (ms) and the y axis is frequency (kHz). See the text for a description of the song features.

Figure 1

Spectrograms of the songs of (a) Thamnophilus schistaceus , (b) Jacamerops aureus , (c) Schiffornis turdina , and (d) Crypturellus cinereus to illustrate measures of acoustic features. The x axis is time (ms) and the y axis is frequency (kHz). See the text for a description of the song features.

Principal components analysis

Some of the acoustic features of songs were correlated with each other ( Table 1 ). To generate independent variables for the axes of acoustic space, I subjected the original features of songs to principal component analysis (PCA). Because this analysis requires variables with values for all individuals, when songs included only one note the measurements of the first note were also included as measurements for the average and last note. For songs with only 2 notes, I used the mean of the measurements of the notes. To normalize the total number of notes in a song, I used a square root transformation. PCA of the correlation matrix for the mean acoustic features of the 83 song patterns yielded 4 principal components (PCs) with eigenvalues greater than one, which together explained 82% of the variation ( Table 1 ). These 4 PCs were the 4 axes of acoustic space within which I located each of the 83 song patterns.

Table 1

Loadings for the first 4 PCs derived from measurements of the acoustic properties of avian songs

 PC1 PC2 PC3 PC4 
Eigenvalue 4.84 2.24 1.63 1.15 
Percentage of variance explained 40.34 18.71 13.62 9.58 
Lowest dominant frequency (Hz) 0.36 −0.28 0.06 0.26 
Highest dominant frequency (Hz) 0.39 −0.29 0.19 0.13 
Song bandwidth (Hz) 0.31 −0.22 0.3 −0.07 
Number of notes 0.18 0.45 0.08 0.43 
Total duration (ms) 0.01 0.23 0.57 −0.11 
Rate of song (notes/ms) 0.15 0.29 −0.37 0.59 
Change in rate first and second portion of the song 0.19 0.36 0.34 −0.12 
Change in rate second and third portion of the song 0.12 0.42 0.23 −0.04 
Dominant frequency of song (Hz) 0.38 −0.29 0.11 0.19 
First note complexity (bandwidth/duration/slopes) 0.34 0.17 −0.28 −0.34 
Last note complexity (bandwidth/duration/slopes) 0.35 0.11 −0.27 −0.22 
Average note complexity (bandwidth/duration/slopes) 0.36 0.12 −0.27 −0.38 
 PC1 PC2 PC3 PC4 
Eigenvalue 4.84 2.24 1.63 1.15 
Percentage of variance explained 40.34 18.71 13.62 9.58 
Lowest dominant frequency (Hz) 0.36 −0.28 0.06 0.26 
Highest dominant frequency (Hz) 0.39 −0.29 0.19 0.13 
Song bandwidth (Hz) 0.31 −0.22 0.3 −0.07 
Number of notes 0.18 0.45 0.08 0.43 
Total duration (ms) 0.01 0.23 0.57 −0.11 
Rate of song (notes/ms) 0.15 0.29 −0.37 0.59 
Change in rate first and second portion of the song 0.19 0.36 0.34 −0.12 
Change in rate second and third portion of the song 0.12 0.42 0.23 −0.04 
Dominant frequency of song (Hz) 0.38 −0.29 0.11 0.19 
First note complexity (bandwidth/duration/slopes) 0.34 0.17 −0.28 −0.34 
Last note complexity (bandwidth/duration/slopes) 0.35 0.11 −0.27 −0.22 
Average note complexity (bandwidth/duration/slopes) 0.36 0.12 −0.27 −0.38 

Loadings in bold face make important contributions to the component (loading >0.3, see McGarigal et al. 2000 ).

Quantifying acoustic space and nearest neighbor distance

To measure the separation of different species’ songs in acoustic space, I calculated the Euclidean distance between species’ songs in the 4-dimensional acoustic space just described. The nearest neighbor distance (NND) for each species was the distance to the closest neighbor in this acoustic space ( Figure 2 ). Because PCA normalizes the resultant PCs, it eliminates differences in scale that result from different units of measurement. The normalized PCs might not reflect the emphases that the different species of birds place on acoustic features during perception of sounds, but in the absence of any information about how the various species might weight these features, there was no biological justification for a different measure of distance.

Figure 2

Songs of 83 species in 2-dimensional acoustic space defined by the first 2 PCs. See Supplementary Material , Appendix A, for a list of species included in the plot.

Figure 2

Songs of 83 species in 2-dimensional acoustic space defined by the first 2 PCs. See Supplementary Material , Appendix A, for a list of species included in the plot.

To determine whether the acoustic community was clustered, random, or overdispersed in acoustic space, I used the R of Clark and Evans (1954) as a measure of dispersion in K dimensions. The test compares observed NNDs in a population, Ra , to that in a randomly distributed population, Re , of the same density, thus R = Ra/Re . If R = 1.0, the distribution is random. Scores approaching 0 indicate increasingly clumped distributions, and those above 1.0 indicate increasingly uniform distributions. I followed Clark and Evans (1979) to calculate the expected NND in 4 dimensions, re = 0.60813/ρ 1/4 , and the standard error of the mean distance to the nearest neighbor, r σ = 0.55326/ρ 1/4 , in a randomly distributed population of density ρ. To calculate the observed density, I calculated the volume of the total acoustic space (π 2 /2 × r4 for a 4-dimensional hypersphere). The radius was the Euclidean distance from the 4-dimensional centroid of the acoustic community to the location of the species farthest from the centroid. Outliers were not removed, so the hypersphere in effect incorporated a buffer strip around the occupied volume, as recommended by Donnelly (1978) .

Preliminary examination of the distribution of the song patterns in acoustic space revealed that species near the center were more clustered than those near the periphery. This pattern resulted in NND scores that were extremely clustered overall. Consequently, I calculated the centroid of the acoustic community and then selected the inner quartile of species in the acoustic community for the final NND analyses. Species in the inner quartile of the acoustic community are presumably most likely to create acoustic interference for each other ( Nelson and Marler 1990 ).

Acoustic community at multiple temporal scales

To investigate acoustic partitioning at different scales, NNDs were calculated for 1) all species from all census points across all days and seasons, 2) species at one point during one morning, and 3) species within a 30-min period at one point. To see if there were differences in NND dispersion between species that sang early and late in the morning, I investigated species that began to sing approximately at sunrise (0600) and those that began approximately at 0800. These two 30-min periods were the least similar to each other in species composition in my recordings, based on Jaccard's index of similarity (0.24 as opposed to >0.27). A 1-way analysis of variance (ANOVA) compared the vocal activity of each species during each of the four 30-min periods of the acoustic censuses.

The data were further subdivided to compare species that sang in the same or different strata of the forest and species that were closely or distantly related. I focused on the midlevel stratum of the forest, which had more species than the other strata (see Supplementary Material , Appendix A). Likewise, I focused on the suborder Tyranni (suboscines), which had more species than the other orders or suborders (see Results). For each spatiotemporal scale, except all species across all days and seasons, the mean of all nearest neighbor values was used as an overall measure of dispersion ( R ).

To investigate whether smaller temporal scales had greater song dispersion than larger temporal scales, I compared R for species detected during the same 30-min period at one point to species detected at a point throughout the morning. To ensure independent samples, I randomly divided the different days of the acoustic censuses into 2 groups, each consisting of 20 point-days. The first group was used to analyze the species detected during the same 30-min period at a point and the second to analyze all the species detected at a point throughout a morning. A 1-way ANOVA compared R at these 2 temporal scales. Because R is a ratio of observed/expected NND for a specified number of species, comparisons of R at different scales is not confounded by the differences in numbers of species at these scales.

Phylogenetic distance, singing strata, and acoustic space

To compare the similarity between songs of closely and more distantly related species, I used ANOVA to compare species’ NNDs, Euclidean distances between congeners, and Euclidian distances between 1 of 2 congeners and a randomly selected species from the community. I used genera and families to group-related species because the quantitative phylogenetic relationships between many of the species in Amazonia are still unknown. This analysis was thus conservative. For genera that included more than 2 species, I randomly selected 2 for this analysis (see Supplementary Material , Appendix B). In a 1-way ANOVA, the categories of species (nearest-neighbor, congener, and random species from the community) were the predictor variables, whereas Euclidean distance was the response variable. Two tests were conducted. The first tested the hypothesis that congeners are nearest neighbors in acoustic space by comparing the distance between congeners and their nearest neighbor in acoustic space. The second tested the hypothesis that congener songs are more similar than songs of randomly selected species in the same community. It compared the Euclidean distance between congeners and a randomly selected species from the community. These methods were also used to compare the distance between songs of family members to the distances between nearest-neighbors, family members, and random species not in the same family. In this analysis, pairs of species in the same family always excluded congeners. In the analysis by genera, an assessment of the residuals showed 1 genus ( Xiphorhynchus ) as an outlier (greater than 2 standard deviations [SDs] from the mean). This genus was removed before the final analysis.

I also compared the distance between songs of species in the same stratum of the forest to the distance between songs of species in different strata. Based on published information and corroborated by personal observations at Rio Cristalino ( Ridgely and Tudor 1994 ; del Hoyo 2002 ; Remsen 2003 ; Zimmer and Isler 2003 ), I categorized each species as singing primarily on the ground (within 0.1 m of the ground), in the understory (0.1–4 m above ground), midlevel (4–15 m above ground), in the subcanopy (15–30 m above ground), or in the canopy (30 m above ground–top of trees). For each species, I calculated the Euclidean distance to the nearest neighbor in the same stratum of the forest and the nearest neighbor in a different stratum of the forest. A 1-way ANOVA included stratum of the forest (same and different) as the predictor variable and NND as the response variable. In an assessment of the residuals, 1 species ( Glyphorynchus spirurus ) was an outlier (greater than 2 SDs from the mean). This species was removed before the final analysis, although once again the removal of this outlier did not affect the statistical significance of the results.

Comparison of the center and periphery of the acoustic community

To measure changes in NND with distance from the center of the acoustic space, species were separated into the inner quartile and the outer quartile based on their Euclidean distance from the centroid of the acoustic community. A 1-way ANOVA compared NNDs of species on the periphery and near the center of the acoustic space. In an assessment of the residuals, one point in the outer quartile was an outlier and was removed before the final analysis, although the removal of this outlier did not affect the statistical significance of the results.

To investigate whether species near the center of acoustic space sang more frequently than species in the periphery of acoustic space, I used linear regression to examine each species’ mean amount of singing during the dawn chorus in relation to its Euclidean distance from the centroid of the acoustic community. This analysis included the 50% of the species closest to the centroid. The average amount of singing per morning for each species was calculated as the total number of minutes in which a species sang divided by the number of point-days on which it was recorded. In general, only 1 individual per species was recorded at each census point, but occasionally, a recording included 2 individuals of the same species, presumably because the microphone was near the boundary between their territories.

RESULTS

Acoustic community at multiple spatiotemporal scales

The 82 species that sang during more than 1% of the total census time (see Supplementary Material , Appendix A) included 51 suboscines (order Passeriformes, suborder Tyranni), 7 oscines (order Passeriformes, suborder Passeres), and 24 non-passerines (orders Tinamiformes, Galliformes, Columbiformes, Strigiformes, Caprimulgiformes, Trogoniformes, Coraciiformes, and Piciformes). Similar numbers of species were detected during the wet and dry seasons, 124 and 113, respectively. The mean number of species detected at each spatiotemporal scale revealed a nested structure with larger scales having more species and smaller scales having fewer species. For example, the mean number of species detected at any one site (comprising 2 points 100 m apart) across all sampling days was 114 ± 4.9, whereas during 1 season it was 82 ± 4.5, during 1 week 67 ± 4.5, and during 1 day 53 ± 5.8. In contrast, the mean number of species detected at one point per season, per week, and per day were 68 ± 6, 53 ± 5.6, and 39 ± 5.9, respectively. The mean number of species detected at any one point during the 30 min starting at sunrise was 19 ± 3.2 and during the 30 min starting at 0800 was 16 ± 2.4. ANOVAs revealed that 45 out of 82 species preferentially sang in a subset of the 30-min periods of the dawn chorus (see Supplementary Material , Appendix A). After a Bonferroni correction for multiple tests (adjusted P = 0.00061), only 20 species sang significantly more in some 30-min periods than others.

The Clark and Evans (1954) test for overdispersion revealed that species were randomly distributed in acoustic space at all spatiotemporal scales. For the largest spatiotemporal scale (all species across all days), R was close to 1 ( R = 1.01, z = 0.01, P = 0.99). At single points during one morning, R varied from 0.83 to 1.22 with mean = 1.04 ( z = 0.75, P = 0.45). Species that sang together in the same 30-min period, starting at sunrise, had greater dispersion ( R ) than larger temporal scales, such as the whole morning, but were still not more uniformly distributed than expected by chance ( R = 1.19, z = 0.21, P = 0.84). There was no significant difference between the 30-min period starting at sunrise and the 30-min period starting at 0800 ( R = 1.20, z = 0.21, P = 0.84). Species that sang in the same 30-min period as well as in the same stratum of the forest showed even more dispersion ( R = 1.28, z = 0.3, P = 0.76). Closely related species singing in the same 30-min period showed the greatest dispersion ( R = 1.32, z = 0.35, P = 0.73). Species singing at any one point in the 30-min period beginning at sunrise had significantly greater dispersion of songs than did all the species singing at the same point ( Figure 3 ; F1,38 = 4.61, P = 0.038).

Figure 3

Dispersion in acoustic space ( R ) for birds that sang together at the same point during the one 30-min period (beginning at dawn) and birds that sang together at the same point but in different 30-min periods of a morning. The central line represents the mean and the lower and upper bars show the standard error. Species that sing in the same 30-min period have songs that are more dispersed than species that sing throughout the morning.

Figure 3

Dispersion in acoustic space ( R ) for birds that sang together at the same point during the one 30-min period (beginning at dawn) and birds that sang together at the same point but in different 30-min periods of a morning. The central line represents the mean and the lower and upper bars show the standard error. Species that sing in the same 30-min period have songs that are more dispersed than species that sing throughout the morning.

Phylogenetic distance, singing strata, and acoustic space

The acoustic censuses detected 11 genera with more than 1 species. In 9 of these genera, congeners were not nearest neighbors in acoustic space. The 2 exceptions were “Columba” and “Trogon.” The mean Euclidean distance between songs of congeners was 2.31, whereas the mean NND between congeners and their nearest acoustic neighbors was 0.81 ( F1,20 = 16.57, P = 0.0006). In contrast, there was no significant difference between within-genus Euclidean distances and Euclidean distances to species randomly chosen from the community.

The acoustic censuses detected 13 families that included more than 1 species. Euclidean distances to the nearest family member were much greater than the distance to their nearest acoustic neighbors, 2.89 and 0.87, respectively ( F1,24 = 28.78, P < 0.0001). In contrast, there was no significant difference between the random within-family Euclidean distances and the Euclidean distances to species randomly chosen from the community.

Eighty-two species were grouped into the 5 strata (14 ground, 11 understory, 37 midlevel, 10 subcanopy, and 10 canopy species). Species had mean NND within the same strata of 1.41 and mean NND between different strata of 1.03 ( F1,164 = 13.18, P < 0.0004).

Comparison of the center and periphery of the acoustic community

The quartile closest to the centroid consisted of 20 species (12 suboscines, 5 oscines, and 3 non-passerines) with an average mass of 71.4 g. The species in the quartile farthest from the centroid consisted of 21 species (14 suboscines, 1 oscine, and 6 non-passerines) with an average mass of 60.8 g. The mean NND of species near the center and near the periphery of acoustic space was 0.67 and 1.44, respectively. ANOVA revealed that species closer to the centroid of the acoustic community had smaller NND than species near the periphery ( F1,39 = 20.29, P < 0.0001). There were 40 species in the 50% closest to the centroid. There was a significant relationship between distance from the centroid of acoustic space and a species’ mean amount of singing during the dawn chorus ( Figure 4 ; R2 = 0.12, m = −3.73, P = 0.027).

Figure 4

Relationship between the mean number songs that a species sings each morning and the distance of the species’ song to the centroid of community acoustic space. This figure indicates that species with songs closer to the acoustic space centroid sing more frequently than species with songs further away from the acoustic space centroid.

Figure 4

Relationship between the mean number songs that a species sings each morning and the distance of the species’ song to the centroid of community acoustic space. This figure indicates that species with songs closer to the acoustic space centroid sing more frequently than species with songs further away from the acoustic space centroid.

DISCUSSION

In this study, I examined the effects of acoustic interference on song structure and singing in a community of birds in Amazonia. The results indicate that species singing in the same stratum of the forest and in the same 30-min interval of the morning were more dispersed in acoustic space than species singing in different strata and at different times in the morning. As expected, acoustic interference occurs between species with similar songs at similar places and times rather than across the entire avian community. Within sets of species that share the same time and place for communication, the challenge of recognizing conspecific signals against background noise, composed of similar heterospecific signals, has resulted in the overdispersion of signals in acoustic space.

Species in the center of the community acoustic space have closer nearest neighbors than species on the periphery of the community acoustic space. Because the center is more crowded than the periphery, central species have a greater chance of acoustic interference from heterospecific signals. Perhaps to compensate for their crowding in acoustic space, centrally located species sang more frequently than species further from the center.

Acoustic community at multiple spatiotemporal scales

The present study shows that at small spatiotemporal scales birds’ songs are overdispersed in acoustic space. Despite predictions that species in the same community should divide the acoustic space to improve signal detection and discrimination, previous studies have not been able to confirm signal partitioning. One reason might be that acoustic space has multiple axes, including song features ( Littlejohn 1959 ; Hodl 1977 ; Chek et al. 2003 ), the timing of signaling ( Sueur 2002 ), and the spatial location of signalers ( Drewry and Rand 1983 ; Duellman and Pyles 1983 ; Chek et al. 2003 ), which together create a large combination of parameters among which acoustic space could be divided by cooccurring species. For instance, if 2 species have similar dominant frequencies, they can divide acoustic space by selecting different times to sing, with 1 species singing early in the morning and the other late in the morning ( Luther 2008 ).

Although this study investigated acoustic partitioning along multiple acoustic axes, it did not reveal an overall pattern of overdispersed signals in acoustic space. In fact, the distribution of species in acoustic space was clustered, with many species in the center and fewer species in the periphery (see discussion below). This distribution could result from similarities in the mechanisms of sound production across the majority of the species in the community. Most species were more or less similar in mass and bill shape, both important influences on the songs that a bird can most efficiently produce ( Ryan and Brenowitz 1985 ; Podos 1996 ). In addition, bird songs are adapted to optimize their transmission distance in their usual habitats ( Morton 1975 ; Marten and Marler 1977 ; Wiley and Richards 1982 ; Naguib and Wiley 2001). Because all the species in this study live in the same habitat, there should be convergence in songs for optimal transmission. Similar morphology and habitat could explain why the overall distribution of species in acoustic space was clustered rather than overdispersed.

Despite these restrictions on the evolution of acoustic partitioning, species that sang in the same 30-min intervals and in the same stratum of the forest had significantly greater signal dispersion than species in the same community that did not sing at the same place and time. The observed small-scale acoustic partitioning could have 2 explanations. First, species that habitually sing at the same time and place might evolve songs to avoid heterospecific interference. Alternatively, species might evolve behavioral plasticity to choose times and places for singing in order to avoid interference from other species with similar signals. In addition, both song features and singing behavior might evolve together.

Phylogenetic distance, singing strata, and acoustic space

In this study, the songs of syntopic species in the same genus or family were less similar than the songs of many less closely related species in the same community. The mean distance between songs of species in the same genus was almost equal to the distance to songs of species randomly chosen from the community. Although it is often thought that divergence among closely related syntopic species has evolved to reduce potential mating errors ( Hobel and Gerhardt 2003 ; Seddon 2005 ), this study provided no evidence that congeners had any more influence on the evolution of signals than did distantly related species. Possibly, congeners also use other cues for discrimination, such as the time of day (see Luther 2008 ), the stratum of the forest, or visual cues.

Although in general songs of congeners and family members were not nearest neighbors in acoustic space, in the genera Trogon and Columba congeners were acoustic nearest neighbors. In both cases, congeners used the same stratum of the forest, so the similarity of their signals might result from convergence of songs as a result of interspecific territoriality ( Cody 1969 ; de Kort et al. 2002 ). In such cases, species might also use other cues for recognizing conspecifics, such as visual cues, to avoid mating errors.

Songs of species in the same stratum of a forest might converge on similar characteristics as a result of adaptations to their signaling environment ( Marten and Marler 1977 ; Wiley and Richards 1982 ; Wiley 1991 ; Naguib and Wiley 2001; Nemeth et al. 2001 ; Seddon 2005 ). In this study, however, species in the same stratum of the forest were not acoustic nearest neighbors, presumably in order to avoid acoustic interference. Spatial separation, such as singing from different strata of a forest, should provide receivers with an additional cue for correct recognition of species ( Klump 1996 ).

Comparison of the center and periphery of the acoustic community

Species with songs near the center of the community's acoustic space might have greater difficulty identifying conspecific signals than species with peripheral songs because of the close proximity of many acoustic neighbors. For example, Nelson and Marler (1990) studied a community of 13 species of grassland birds in New York and found that species’ songs were more often in close proximity in the center of the community's acoustic space than on the periphery. Species near the center also required more information for accurate song recognition than species on the periphery of the acoustic space ( Nelson and Marler 1990 ). The present study showed that species centrally located in acoustic space in tropical forests also have closer neighbors than do peripheral species, just as Nelson and Marler (1990) found in the simpler communities of temperate fields in New York. Whether they also require more parameters for discrimination of conspecific song remains to be investigated.

Signal detection theory suggests that rare species would be at a greater disadvantage in the center of acoustic space than would common species ( Wiley 2006 ). Because the consequences of missed detections and false alarms are likely to be greater for rare species, they might evolve more distinctive signals that lie farther from the center of acoustic space ( Wollerman and Wiley 2002b ). However, birds are not the only source of biotic noise in a forest. Even though most frogs and insects produce sounds at different frequencies than most species of birds ( Ryan and Brenowitz 1985 ), all taxa that produce sounds at the same location share a common acoustic space. Birds on the periphery of the avian acoustic space could be near the center of the frog or insect acoustic space. Future studies should consider multiple taxa for a more complete picture of acoustic communities.

Individuals can compensate for interference from background noise by adjusting their signaling behavior. These adjustments could include an increase in signal amplitude, contrast with background noise, or the rate of signal repetition ( Brumm and Slabbekoorn 2005 ; Wiley 2006 ). In fact, signal detection theory predicts that increased redundancy can increase information transfer in the presence of noise ( Wiley 1994 , 2006 ). At least in some cases, birds increase the redundancy of their signals in the presence of background noise, either physical or biotic ( Lengagne et al. 1999 ; Brumm and Slater 2006 ). The increased frequency of singing by birds near the center of acoustic space in the present study might provide another example of increased redundancy in the face of background noise. Alternatively, species near the center of acoustic space might have denser populations than species further from the center of the acoustic community, so the increased frequency of songs might not apply to each individual.

In conclusion, this study has shown that songs of birds in Amazonian forests are overdispersed in acoustic space but that this overdispersion appears only between species that interact acoustically at nearly the same time and place. This study also confirmed that the center of the acoustic community is more crowded than the periphery.

FUNDING

Mellon Foundation; Explorer's Club; International and Latin American Studies at University of North Carolina; the University of North Carolina Graduate School.

SUPPLEMENTARY MATERIAL

Supplementary material can be found at http://www.beheco.oxfordjournals.org/

I would like to thank R. H. Wiley for help in all aspects of this project. I would like to thank Maria Alice dos Santos Silva and Mario Cohn-Haft who helped with Brazilian research visas and John Luther, Amy Upgren, Brad Davis, Vitoria da Riva Carvalho, and the staff of the Rio Cristalino RPPN for assistance in the field. I would like to thank Mario Cohn-Haft, Andrew Whitaker, Kevin Zimmer, Dan Lane, and Alex Lees for help with initial species identification of unknown vocalizations from my acoustic censuses.

References

Bradbury
JW
Vehrencamp
SL
Principles of animal communication
 , 
1998
Sunderland (MA)
Sinauer Associates
Bremond
JC
Acoustic competition between the song of the wren (Troglodytes troglodytes) and the songs of other species
Behaviour
 , 
1978
, vol. 
65
 (pg. 
89
-
98
)
Brumm
H
Slabbekoorn
H
Acoustic communication in noise
Adv Study Behav
 , 
2005
, vol. 
35
 (pg. 
151
-
209
)
Brumm
H
Slater
PJB
Ambient noise, motor fatigue, and serial redundancy in chaffinch song
Behav Ecol Sociobiol
 , 
2006
, vol. 
60
 (pg. 
475
-
481
)
Chek
AA
Bogart
JP
Lougheed
SC
Mating signal partitioning in multi-species assemblages: a null model test using frogs
Ecol Lett
 , 
2003
, vol. 
6
 (pg. 
235
-
247
)
Clark
PJ
Evans
CS
Distance to nearest neighbor as a measure of spatial relationships in populations
Ecology
 , 
1954
, vol. 
35
 (pg. 
445
-
453
)
Clark
PJ
Evans
FC
Generalization of a nearest neighbor measure of dispersion for use in K dimensions
Ecology
 , 
1979
, vol. 
60
 (pg. 
316
-
317
)
Cody
ML
Convergent characteristics in sympatric species: a possible relation to interspecific competition and aggression
Condor
 , 
1969
, vol. 
71
 (pg. 
222
-
239
)
de Kort
SR
den Hartog
PM
ten Cate
C
Diverge or merge? The effect of sympatric occurrence on the territorial vocalizations of the vinaceous dove Streptopelia vinacea and the ring-necked dove S. capicola
J Avian Biol
 , 
2002
, vol. 
33
 (pg. 
150
-
158
)
del Hoyo
J
Jacamars to Woodpeckers. Handbook of the birds of the world
 , 
2002
Barcelona (Spain)
Lynx Editions
Doutrelant
C
Leitao
A
Otter
K
Lambrechts
MM
Effect of blue tit song syntax on great tit territorial responsiveness-an experimental test of the character shift hypothesis
Behav Ecol Sociobiol
 , 
2000
, vol. 
48
 (pg. 
119
-
124
)
Donnelly
K
Hodder
I
Simulations to determine the variance and edge-effect of total nearest neighbor distance
Simulation methods in archeology
 , 
1978
London
Cambridge University Press
(pg. 
91
-
95
)
Drewry
GE
Rand
AS
Characteristics of an acoustic community: Puerto Rican frogs of the genus Eleutherodactylus
Copeia
 , 
1983
, vol. 
4
 (pg. 
941
-
953
)
Duellman
WE
Pyles
RA
Acoustic resource partitioning in anuran communities
Copeia
 , 
1983
, vol. 
3
 (pg. 
639
-
649
)
Dunning
JB
Jr
CRC handbook of avian body masses
 , 
1993
Boca Raton (FL)
CRC Press
Endler
JA
Signals, signal conditions, and the direction of evolution
Am Nat
 , 
1992
, vol. 
139
 (pg. 
S125
-
S153
)
Endler
JA
The color of light in forests and its implications
Ecol Monogr
 , 
1993
, vol. 
63
 (pg. 
1
-
27
)
Gerhardt
HC
Huber
F
Acoustic communication in insects and anurans: common problems and diverse solutions
 , 
2002
Chicago (IL)
University of Chicago Press
Haavie
J
Borge
T
Bures
S
Garamszeg
LZ
Lampe
HM
Moreno
J
Qvarnstrom
A
Torok
J
Saetre
GP
Flycatcher song in allopatry and sympatry—convergence, divergence and reinforcement
J Evol Biol
 , 
2004
, vol. 
17
 (pg. 
227
-
237
)
Hobel
G
Gerhardt
HC
Reproductive character displacement in the acoustic communication system of green tree frogs (Hyla cinerea)
Evolution
 , 
2003
, vol. 
57
 (pg. 
894
-
904
)
Hodl
W
Call differences and calling site segregation in anuran species from central Amazonian floating meadows
Oecologia
 , 
1977
, vol. 
28
 (pg. 
351
-
363
)
Isler
ML
Isler
PR
Whitney
BM
Use of vocalizations to establish species limits in antbirds (Passeriformes: Thamnophilidae)
Auk
 , 
1998
, vol. 
115
 (pg. 
557
-
590
)
Klump
GM
Kroodsma
DE
Miller
EH
Bird communication in the noisy world
Ecology and evolution of acoustic communication in birds
 , 
1996
Ithaca (NY)
Cornell University Press
(pg. 
321
-
338
)
Lengagne
T
Jouventin
P
Aubin
T
Finding one's mate in a king penguin colony: efficiency of acoustic communication
Behav Genet
 , 
1999
, vol. 
136
 (pg. 
833
-
846
)
Littlejohn
MJ
Call differentiation in a complex of seven species of Crinia (Anura, Leptodactylidae)
Evolution
 , 
1959
, vol. 
13
 (pg. 
452
-
468
)
Luther
DA
Signaler-receiver coordination and the timing of communication in Amazonian birds
Biol Lett
 , 
2008
, vol. 
4
 (pg. 
651
-
654
)
Marler
P
Lanyon
WE
Tavolga
WN
Bird songs and mate selection
Animal sounds and communication
 , 
1960
, vol. 
Vol. 7
 
Washington (DC)
American Institute of Biological Sciences
(pg. 
348
-
367
)
Marten
K
Marler
P
Sound transmission and its significance for animal vocalization
Behav Ecol Sociobiol
 , 
1977
, vol. 
2
 (pg. 
271
-
290
)
McGarigal
K
Cushman
S
Stafford
S
Multivariate statistics for wildlife and ecology research
 , 
2000
New York (NY)
Springer-Verlag
Miller
EH
Kroodsma
DE
Miller
EH
Character and variance shift in acoustic signals of birds
Acoustic communication in birds
 , 
1982
, vol. 
Vol. 1
 
New York (NY)
Academic Press
(pg. 
253
-
295
)
Morton
ES
Ecological sources of selection on avian sounds
Am Nat
 , 
1975
, vol. 
109
 (pg. 
17
-
34
)
Naguib
M
Wiley
RH
Review: estimating the distance to a source of sound: mechanisms and adaptations for long-range communication
Animal Behav
 , 
2001
, vol. 
62
 (pg. 
825
-
837
)
Nelson
DA
Marler
P
Stebbins
WC
Berkley
MA
The perception of bird song and an ecological concept of signal space
Comparative perception, Vol. 2, complex signals
 , 
1990
New York (NY)
Wiley
(pg. 
443
-
478
)
Nemeth
E
Winkler
H
Dabelsteen
T
Differential degradation of antbird songs in a neotropical rainforest: adaptation to perch height?
J Acoust Soc Am
 , 
2001
, vol. 
110
 (pg. 
3263
-
3274
)
Pfennig
KS
Female spadefoot toads compromise on mate quality to ensure conspecific matings
Behav Ecol
 , 
2000
, vol. 
11
 (pg. 
220
-
227
)
Podos
J
Motor constraints on vocal development in a songbird
Anim Behav
 , 
1996
, vol. 
51
 (pg. 
1061
-
1070
)
Remsen
JV
del Hoyo
J
Elliott
A
Christie
D
Furnariidae (ovenbirds)
Handbook of the birds of the world
 , 
2003
, vol. 
Vol. 8
 
Barcelona (Spain)
Lynx Editions
(pg. 
162
-
357
)
Ridgely
RS
Tudor
G
The birds of South America
 , 
1994
Oxford
Oxford University Press
Ryan
MJ
Brenowitz
EA
The role of body size, phylogeny, and ambient noise in the evolution of bird song
Am Nat
 , 
1985
, vol. 
126
 (pg. 
87
-
100
)
Schwartz
JJ
Wells
KD
An experimental study of acoustic interference between two species of neotropical tree frogs
Anim Behav
 , 
1983
, vol. 
31
 (pg. 
181
-
190
)
Seddon
N
Ecological adaptation and species recognition drives vocal evolution in neotropical suboscine birds
Evolution
 , 
2005
, vol. 
59
 (pg. 
200
-
215
)
Sueur
J
Cicada acoustic communication: potential sound partitioning in a multispecies community from Mexico (Hemiptera: Cicadomorpha: Cicadidae)
Biol J Linn Soc
 , 
2002
, vol. 
75
 (pg. 
379
-
394
)
Terborgh
J
Robinson
SK
Parker
TAI
Structure and organization of an Amazonian forest bird community
Ecol Monogr
 , 
1990
, vol. 
60
 (pg. 
213
-
238
)
Wiley
RH
Associations of song properties with habitats for territorial oscine birds of eastern North-America
Am Nat
 , 
1991
, vol. 
138
 (pg. 
973
-
993
)
Wiley
RH
Real
LA
Errors, exaggeration, and deception in animal communication
Behavioral mechanisms in evolutionary
 , 
1994
Chicago (IL)
University of Chicago Press
(pg. 
157
-
189
)
Wiley
RH
Signal detection and animal communication
Adv Study Behav
 , 
2006
, vol. 
36
 (pg. 
217
-
247
)
Wiley
RH
Richards
DG
Kroodsma
DE
Miller
EH
Adaptations for acoustic communication in birds: sound transmission and signal detection
Acoustic communication in birds
 , 
1982
New York (NY)
Academic Press
(pg. 
131
-
181
)
Wollerman
L
Wiley
RH
Background noise from a natural chorus alters female discrimination of male calls in a neotropical frog
Anim Behav
 , 
2002a
, vol. 
63
 (pg. 
15
-
22
)
Wollerman
L
Wiley
RH
Possibilities for error during communication by neotropical frogs in a complex acoustic environment
Behav Ecol Sociobiol
 , 
2002b
, vol. 
52
 (pg. 
465
-
473
)
Zimmer
KJ
Isler
ML
Thamnophilidae (typical antbirds). Handbook of the birds of the world
 , 
2003
, vol. 
Vol. 8
 
Barcelona (Spain)
Lynx Editions
Zimmer
KJ
Parker
TAI
Isler
ML
Isler
PR
Survey of a southern Amazonian avifauna: the Alta Floresta region, Mato Grosso, Brazil
Orn Monogr
 , 
1997
, vol. 
48
 (pg. 
887
-
918
)