Songbird community structure changes with noise in an urban reserve

Urban noise has the potential to negatively affect bird ﬁtness as it may interfere with communication and for instance, decrease predator detection or breeding activity. Regular exposure to urban noise may cause changes in bird communities and inﬂuence local distribution patterns. While some bird species may tolerate urban noise, others may not, which mechanisms underlie noise tolerance remain unknown. Whereas a large number of studies have focused on the responses of speciﬁc bird species or groups to urban noise, few have attempted to assess noise effects on species richness. Here, we assessed the association between (i) anthropogenic noise and bird richness, (ii) noise level and song modiﬁcation and (iii) species noise tolerance and detection frequency, by recording these variables at seven plots within an urban reserve in Mexico City. We found that species richness is negatively affected by increasing anthropogenic noise, but also by the increase of shrub cover. Nevertheless, over 90% of our species did not display signiﬁcant song modiﬁcations. This lack of response may be a result of our restricted sample size, although we cannot exclude a possible change in non-spectral components (e.g. change song timing) or that given a noise gradient, birds occupy areas where song modiﬁcation is not necessary. Our results suggest that urban avian species richness is signiﬁcantly affected by noise and vegetation cover. We expect that by limiting anthropogenic noise and providing diversity of vegetation cover, cities could be made attractive to bird species with different degrees of noise tolerance, promoting urban bird diversity. and parent–offspring


Introduction
The chief means of communication in birds consist of songs and calls (Catchpole and Slater 1995). To which degree these vocalisations are effectively transmitted between individuals may influence fitness, especially in the context of territorial defence, alarm calls and parent-offspring communication (Boncoraglio and Saino 2007;Brumm and Naguib 2009;Read, Jones, and Radford 2014). As such, selection has favoured species-specific bird songs characterised by sounds less prone to degradation (Morton 1986), attenuation (Forrest 1994) and/or the ability to overcome being masked by biotic or abiotic noise (Brumm and Slabbekoorn 2005). However, to what extent rapid changes in the acoustic landscape influence bird richness and entire songbird communities, especially those affected by anthropogenic noise, remains surprisingly unclear (Francis, Ortega, and Cruz 2009;Brumm and Zollinger 2013).
Anthropogenic noise has augmented dramatically during recent decades (Shannon et al. 2016), and is increasingly suggested to exert a strong selective pressure on birds (Dorado-Correa, Rodríguez-Rocha, and Brumm 2016;Kleist et al. 2016). Specifically, anthropogenic noise has been hypothesised to reduce transmission capacity during songbird communication by affecting both detection of conspecifics, and the distance along which birds can transmit their songs (Nemeth and Brumm 2010).
As a response, various bird species have been shown to respond by increasing the frequency (Halfwerk and Slabbekoorn 2009;Ríos-Chelén et al. 2013), bandwidth (Lohr, Wright, and Dooling 2003) and/or amplitude of their songs (Brumm 2004;Brumm and Zollinger 2011). Again, other species have been shown to shift the timing of their songs (Gil et al. 2015). Which particular strategy a bird applies seems to be species-and context-dependent. Nevertheless, various studies focusing on the effect of anthropogenic noise have shown that some birds may fail to respond altogether (Hanna et al. 2011;Dowling, Luther, and Marra 2012).
Several bird species have been found to express a limited capability to adjust their songs to increasing anthropogenic noise (Ortega 2012;Ríos-Chelén et al. 2013). Given their failure to respond, it is likely that these species may experience shifts in distribution range in the future (Slabbekoorn and Peet 2003). In order to measure potential changes in occurrence, a restricted amount studies that investigated effects of noise on birds has recently focused on the community level, rather than on individual species (Bayne, Habib, and Boutin 2008;Francis, Ortega, and Cruz 2009). Bayne, Habib, and Boutin (2008) and Proppe, Sturdy, and St Clair (2013) found that passerine abundance was negatively associated with anthropogenic noise. Similarly, Francis, Ortega, andCruz (2009), Pató n et al. (2012), González-Oreja (2017) and Perrillo et al. (2017), all working on entire bird communities (i.e. not limited to passerines, which was often the case), found a negative effect of noise on species richness. Nevertheless, the effect of noise on individual species varied, and was in some cases positive. These findings are consistent with the idea that there are two types of songbirds; those which are tolerant, and those which are non-tolerant to noise. As noise prevalence is expected to increase in the future, along with the growth of urban settlements (Marzluff 2001;Moore, Gould, and Keary 2003), the effects of noise are of great importance in urban areas as they can lead to a global homogenization of bird communities, particularly if certain species fail to occupy noisy areas (McKinney 2006;Slabbekoorn and den Boer-Visser 2006).
One motivation to study the effect of noise on urban birds is to promote urban avian diversity, which requires understanding how songbird communities respond to anthropogenic noise. In turn, this knowledge should be used to prioritise conservation areas. Here, we assessed the occurrence of species along noise gradients in an urban nature reserve (Reserva del Pedregal de San Á ngel-REPSA), located in Mexico City, México. Despite being surrounded by the extensive Mexico City metropolitan area, this protected environment has high conservation and diversity value, containing 30.6% of the ca. 840 species of land bird species found in Mexico (Banks et al. 2008). We conducted weekly censuses to characterise (i) songbird vocalizations (frequency and bandwidth) (ii) species richness and (iii) species detection frequency through several gradients of anthropogenic noise. We hypothesised that (i) within species, vocalisations should vary across noise levels, and (ii) species richness should be negatively associated with noise. We looked for evidence that species either adjust their songs between noise levels, or that some would be absent in noisier areas because of their incapacity to modify their songs. Furthermore, we expect non-tolerant species to have lower detection frequencies than tolerant species and that their distribution across noise levels would be more limited.

Study area
The study was conducted at the Reserve del Pedregal de San Á ngel (REPSA), located in the main campus of Universidad Nacional Autó noma de Mé xico (UNAM) in Mexico City (core area ¼ 171 ha, buffer area ¼ 66 ha, total reserve area ¼ 237 ha). This reserve consists of a protected shrub land that grows on extensive lava floods from a nearby volcano which erupted at several periods, most recently about 1700 years ago (Siebe and Macías 2006). The area is interspersed with university buildings and surrounded by major avenues (Rojo and Rodriguez 2002).
We sampled in the reproductive and migratory seasons (January-April 2012). Recordings were obtained in seven circular, 120 m radius non-overlapping plots that were located within the core protected area of the REPSA (Fig. 1). Plots were located at least 120 m from the nearest reserve edge (acting as a buffer to avoid edge effects). In two of the three core areas of REPSA, three plots were placed along an approximately straight line, separated by similar distances from each other (exact distances differed due to variation in topography and core area size and shape). Each plot was visited six times, during which the following protocol was acted upon.

Recording sessions and species richness
Recordings in each plot were obtained at four sampling points along a straight line starting at the centre (0 m), then at 40 m, 80 m, and finally at the edge at 120 m. The taken direction (either N, NW, W, SW, S, NE, E or NE) from the centre of the plot was chosen randomly before starting the recording session. Species calls and songs were obtained during two recording sessions per sampling point, each lasting 150 min. The recordings (in .wav files) were obtained on weekdays without rain, nor strong wind, at dawn (starting 10 min before sunrise) and dusk (stopping 10 min after sunset). Hence, 56 recording sessions were performed, during which we collected species calls and songs at different distances from the centre of the plot to reduce the likelihood of recording the same individuals twice. During each recording session, visual bird surveys were conducted to reduce the possibility that we missed birds that were present, but not singing.
We recorded bird songs and calls using a directional microphone (Rode TM NTG-2) connected to a digital recorder (Marantz PMD221, sampling rate 44 kHz, 16-bit accuracy) during the reserve closing hours to avoid disturbance originating from human activities. The microphone was placed at breast height, parallel to the ground; directionality (N, W, S and E) was changed every minute. Noise measurements (Krauss et al. 2010) were made with a digital sound meter (SEW(R) 2310 SL, ANSI S1.4, Class II, used with 'A' frequency weighting, and slow response, factory calibrated with a 1 kHz tone at 94 dB SPL re 20 mPa). All dB values reported are SPL to re 20 mPa from here on referred to as 'dB'). The main source of noise for this area is a major highway that during weekdays contains a large amount of traffic and thus, noise (range of 37.4-67.4 dB). The following protocol was used at each sampling point (later used as recording session): (i) initial noise measurement (the average of five recordings with the sound meter pointing upward); (ii) 5 min of spontaneous song/call recording; (iii) 1 min of songbird playbacks. These playbacks consisted of five short (ca. 10 s) recordings randomly selected from a list of 28 (Supplementary Table  S4) dawn-chorus recordings made at the same campus some years earlier (from May to April). Bird songs were subsequently played back following the protocol of Gunn et al. (2000); (iv) 5 min recording songs elicited in response to the playback; (v) final noise measure (as in point i). Visual detection/identification of birds was conducted during periods (ii) and (iv) using 10 Â 42 Nikon TM Monarch binoculars. We obtained 960 recordings and the same number of noise measures. Species were either visually identified in the field or diagnosed afterward from the recordings visualised using Avisoft SASLab TM and with the aid of Xeno-canto online database. We defined species richness as the amount of species present in a given area.

Vegetation characterization
Because vegetation density may attenuate noise (Aylor 1972), we used of Google Earth Pro V R to draw polygons to measure the relative presence of different vegetation types in each plot. These vegetation types were easily distinguishable on satellite images, given that only open ('shrub') or densely vegetated ('tree') communities occur in the area. Subsequently we calculated the surface area (m 2 ) of each of these polygons per plot, and converted these to percentages.

Song analyses
Following Pató n et al. (2012), we divided our noise measurements into three categories (referred herein as noise level): low (37.4-47.4 dB), medium (47.4-57.4 dB) and high (57.4-67.4 dB). Each species that was recorded at more than one noise level was included in the subsequent analysis of song attributes using three recordings per noise level. In total, we analysed between 3 and 9 recordings with good quality per species (28). The recordings included the same type of song or call (mainly the latter). Species were analysed in either call or song (similar sonograms can be found in Bermú dez-Cuamatzin et al., 2009). As the main focus of this study was the effect of noise on species richness, we did not attempt to maximise the sample to test for noise effects on song attributes. Thus, we did not conduct an in-depth analysis of the latter question and our results in this respect are to be taken as tentative.
Recordings were analysed with Avisoft-SASLab Pro TM . We divided the selected 5 min recordings into new sound files containing only the species of interest and a single song or call type (FFT length, 256; frequency resolution, 86 Hz). Afterwards, a high-and low-pass frequency filter with a cut-off was applied to these species-specific recordings. The filters were performed manually based on visual determination, thus, specific values of the applied filter varied between recordings. The latter allowed us to separate the signal from background noise, and improved noise-signal ratio. Total bandwidth, maximum-and minimum-peak frequencies were obtained through a power spectrum-based procedure using the automatic parameter measurements function (a single threshold À10 dB; min. freq. and max. freq. À17dB, with default parameters, see Supplementary Table S5). Using the results of these analyses, we performed four ANOVAs per species, each with one song attribute as dependent variable and noise level as an independent factor. We then applied a Bonferroni correction to account for multiple contrasts for each species. For significant variables, post hoc analyses were performed between noise levels (Turkey or Kruskal-Wallis tests, depending on the distribution of the response variable).

Detection frequency
To reduce the chance of measuring the same individual several times during the same recording session (i.e. leading to pseudoreplication) as much as possible, we used a measure of detection frequency per species, which was determined as follows.
Presence (1) or absence (0) of a species was recorded during the 10-min recording sessions at each sampling point, which, comprised the entire recording session before moving to the next sampling point. Subsequently, we summed the presences for each sampling point per species to obtain its detection frequency.

Statistical analyses
Analyses of species richness and noise were performed in R (R Development and Core Team 2008), using a negative binomial regression with species richness (¼number of species per sampling point) as dependent variable (y), and as explanatory variables, noise (as a continuous variable (Krauss et al. 2010)), shrub vegetation cover (%), recording period (dawn or dusk) and biologically meaningful interactions between these variables. Plot was introduced as random effect. Model selection was performed using the Akaike information criterion (AIC, Burnham and Anderson, 2002), which we used to rank models. Model comparison was performed by the sequential removal of one of the explanatory variables from the full model until all possible combinations were tested, after which only the best model, that is the model with the lowest AIC was selected. Furthermore, a correlation analysis was performed between noise range per species (maximum noise -minimum of noise were the species occurred) and detection frequency for illustrative purposes (Fig. 3). Fewer but species in noiser urban sites | 3

Results
We identified the songs or calls of 30 species in the 960 5-min recordings, in addition to eight calls that could not be identified. The negative binomial regression revealed that recording period and shrub vegetation cover negatively affect local species richness (Tables 1 and 2, Fig. 2). Overall, 50% of the decrease in species richness between recording periods can be attributed to differences in calling activity between dawn and dusk (P < 0.001; Table 2, Fig. 2). A further 13% decrease in richness was caused by noise increment per 1 dB (P < 0.001), while each 1% increase in shrub cover produced a decrease in bird species richness of 0.08% (P ¼ 0.001). The latter effect was dependent on the effect of noise on species richness (see Table 2). The formula to obtain the percentage values was: percentage change in the incident rate ¼ 1 -exp (variable coefficient) ( Table 2).
We detected a strong association between the detection frequency of each species and the noise range in which it was detected (P ¼ 0.001, SE ¼ 6.107, df ¼ 35). This outcome is not an inevitable consequence of the way we calculated detection frequency, since some species were only occasionally recorded, but were found in places with very different noise ranges (Fig. 3, see Supplementary Material for a list for species presence in Supplementary Table S2).
Using a limited (n ¼ 3) number of recordings per noise level per (288 recordings in total), we found evidence that only three of the species in our sample adjusted their song frequency according to noise level. Cynanthus latirostris (broad-billed hummingbird) produced higher frequencies when calling at intermediate noise levels (1163 Hz, higher, P ¼ 0.001), while Geothlypis nelsoni (hooded yellowthroat) and Pheucticus melanocephalus (black-headed grosbeak) called with higher frequencies at low noise levels (1115 Hz, P ¼ 0.04. and 683 Hz, P ¼ 0.03, respectively,

Discussion
We used a combination of methods to assess which bird species are found in areas of an urban reserve, the Reserva del Pedregal de San Á ngel in Mexico City, subject to varying degrees of anthropogenic noise. These allowed the detection of a significant portion of the bird diversity reported for the area. Using the categories in Chá vez and Gurrola (2009), we detected 42.1% of the common resident species, 46.7% of the abundant residents and 100% of the very abundant resident species. We believe that this is a fair representation of the bird fauna of the REPSA, especially since our methods were not designed to maximise detection of some bird groups such as shrikes or swifts. Some of the potentially confounding factors common to this type of studies were avoided by (i) recording at closing times thus avoiding the disturbance caused by human presence on our recordings; (ii) establishing a 120 m buffer area to avoid edge effects, visual and traffic pollution; and finally (iii) performing visual bird assessments (see Methodology).
We accordingly collected data on chiefly Passeriformes as this was the most abundant bird group encountered during our study. Passeriformes as most birds groups rely heavily on acoustic communication (Kroodsma and Byers 1991;Catchpole and Slater 1995) and are thus likely to be affected by anthropogenic noise (Yang and Cardoso 2010;Brumm and Zollinger 2011). Indeed, we recorded a significant decrease in species richness as noise increases. This is relevant since, as urban green  Perrillo et al. (2017); noise: 31-45 dB has also found a negative link between noise and bird species richness. However, these authors performed their studies in multiple urban parks across urban-rural gradients (8-28 parks of 0.28-90.7 ha). What follows is that the low noise levels recorded in those studies might be associated with areas other than those in which high levels of noise occur, as the high noise areas are not necessarily connected to the areas where low noise is present. Specifically, small parks may show high levels of urban noise, while larger parks provide a gradient between high noise level and low noise levels, thus giving a wider spectrum of noise for bird species to choose from. We suggest that in such circumstances, birds in small noisy parks are deprived of the option to access quieter areas in the same patch or its surroundings (unless there is a large park in the vicinities, which is often not the case in urban areas with high noise levels), and thus may (i) stay put and modify their song, (ii) stay put and not modify their song or (iii) move away from the area, for instance to a larger park outside their home rage. This response would both impoverish local avian diversity and bias it toward species, which can modify their song.
In addition to noise, vegetation was found to affect bird species richness. In this study we separated shrub cover (a xerophytic scrub vegetation community) from tree cover, which contains communities of Pinus, Abies, Alnus and Quercus tree species (Castillo-Arguero et al. 2004)-as such, we separated open (shrub) and close (tree) vegetation communities. An increase in shrub cover (and therefore, a decrease in forest cover) was subsequently found to negatively affect bird species richness. This difference may be explained by variation in the availability of food, protection or nesting resources characterising these vegetation communities (Miller, Brooks, andCroonquist 1997, Pino et al. 2000). Furthermore, because shrub cover is an open vegetation type, noise may be able to pass though more freely compared with more densely vegetated areas such as forest (Aylor 1972;Wiley and Richards 1982), and therefore affect relatively more bird species.
In spite of a substantial bias toward oscine passeriformes in our sample, we found that only three species (17%) changed some parameters of their song when noise level increased (see also Potash 1972;Lowry, Lill, and Wong 2012). These were the hummingbird Cynanthus latirostris, the warbler Geothlypis nelsoni and the grosbeak Pheucticus melanocephalus. Interestingly, the former increased the frequency of its calls at intermediate, but not at high noise levels. This result contrast with recent studies, which found that hummingbirds are associated with noisy areas (Francis, Ortega, and Cruz 2009;Ortega and Francis 2012), and that these species increase the peak frequency of their vocalisations when exposed to high noise levels (Pytte, Rusch, and Ficken 2003;Francis, Ortega, and Cruz 2011). We note, however, that what is characterised as high noise levels by Pytte, Rusch, and Ficken (2003) prompting the blue-throated hummingbird (Lampornis clemenciae) to increase the peak frequency of its calls (ca. 40 dB) is equivalent to our medium noise level (see above). Hence, we suggest that comparisons between studies pertaining the effects of anthropogenic noise on bird species richness should include actual measures (for instance, of dB) rather than broad categories.
Several bird species present in our study area have been previously described to avoid noise (for instance A. californica, P. melanocephalus and T. bewickii) (Francis, Ortega, and Cruz 2011;Francis et al. 2012), perhaps due to the lack of ability to adapt. Indeed, we did not detect song modification in any of these species during our study. Other species, which were previously reported to modify their song in response to noise (M. melodia, S. passerine, T. migratorius and C. mexicanus; Wood and Yezerinac 2006;Francis et al. 2012;Bermú dez-Cuamatzin et al. 2009) were not found to do so in the present study.
Several factors may explain why we found song adjustment in only a few bird species, and not in others. First, our sample Table 1 Model selection (negative binomial regression) to explain species richness ( ¼ number of species per sampling point) as a function of noise (continuous), shrub vegetation cover (%), recording period (dawn or dusk) and biologically meaningful interactions between these (denoted by a colon), with sampling plot declared as random effect. Using the Akaike information criterion (AIC) we sequentially removed explanatory variables and retained the model with the lowest AIC Independent variables AIC D AIC Noise þ recording period þ plot þ shrub cover þ noise: recording period a þ noise: shrub cover 740.6 0.6 Noise þ recording period þ plot þ shrub cover þ noise: shrub cover 740 Noise þ recording period þ plot þ shrub cover 751.1 11.1 Noise þ recording period þ plot 749.6 9.6 Recording period þ plot 757.9 17.9 Noise þ plot 828.7 88.7 a Non-significant interaction.
¼ standard error of the coefficient, P (z) ¼ probability associated to the z-value. Negative coefficients mean that the effect is negative (e.g. there are fewer bird species as noise increases, and there are fewer species at dusk than a dawn).
Fewer but species in noiser urban sites | 5 size is relatively small, which leads to a reduced power to detect effects, unless they are large. Accordingly, failure to detect adjustment may indicate that such an effect may have been relatively small. Another reason may be that the breadth of noise levels under which we recorded species was insufficient in many cases to prompt song adjustments. Further, it is possible that calls, which make the bulk of our records, are less plastic than songs, since they are made of one or very few elements that can be modified and often have very specific functions which may be compromised if changed (Kroodsma 2004;Lowry, Lill, and Wong 2012). Nevertheless, Potvin, Parris, and Mulder (2011) and Potvin, Mulder, and Parris (2014) showed ample evidence of changes in calls in response to noise. Finally, it may be that the cost of changing the frequency or bandwidth across noise levels is higher that the benefits that may result from such an adjustment (Read, Jones, and Radford 2014). Such strategy may be more effective for the noise levels that we encountered, and at the relatively large, complex environment of the REPSA urban reserve, where birds have the option of moving to quieter areas instead of shifting song attributes. Finally, we cannot discard the occurrence of Table 3 Association between noise (low, medium, high) and song frequency, in the three species in which there was an effect (see supplementary information for a full report)  other changes in song structure/performance that were not measured in this study, such as differences in singing time or call amplitude (McClure et al. 2013).
The data presented in this study support the hypothesis that the increment of noise leads to a decrease in bird species richness, and that noise may also affect abundance of species (if we take detection frequency as a proxy). For noise-tolerant species, noise may promote their abundance relative to other species, as those with higher values of detection frequency had also the wider noise tolerances. It can be argued that species with high detection frequency may have higher noise tolerance due to their larger distributions. However, we observed that abundance per se does not explain the noise tolerance differences (i.e. Atlepest pileatus, a 'common' species at REPSA, has equal range as 'very abundant' species like Carpodacus mexicanus; Chá vez and Gurrola 2009; see Supplementary Table S3).
Despite increasing evidence on the negative effect of noise on bird richness and abundance, empirical evidence on responses involving entire bird communities remains scant. Here, we provide an example on how species richness decreases primarily due to time changes (dusk and dawn, as species detection decreases), secondly, by an increase of noise, and finally by the increase in shrub cover. We suggest that increase in noise may lead to urban bird communities dominated by a limited number of noise-tolerant species. In an increasingly urbanised world, the composition of these communities may therefore depend, to a degree, on the species' capacities to adapt their vocalisations to prevailing noise (Brumm 2006;Luther and Derryberry 2012). Whether rare and endemic bird species will be able to do so, and compete with more tolerant species, remains unclear (Slabbekoorn and Ripmeester 2008). We expect that a future focus on the relative roles of song adaptation, tolerance and competition between species may be most fruitful in assessing effects of anthropogenic noise on bird communities (Slabbekoorn and Ripmeester 2008;Proppe, Sturdy, and St Clair 2013).

Supplementary data
Supplementary data are available at JUECOL online.