Learning to cope: vocal adjustment to urban noise is correlated with prior experience in black-capped chickadees

Abstract

Urban noise can interfere with avian communication through masking, but birds can reduce this interference by altering their vocalizations. Although several experimental studies indicate that birds can rapidly change their vocalizations in response to sudden increases in ambient noise, none have investigated whether this is a learned response that depends on previous exposure. Black-capped chickadees (Poecile atricapillus) change the frequency of their songs in response to both fluctuating traffic noise and experimental noise. We investigated whether these responses to fluctuating noise depend on familiarity with noise. We confirmed that males in noisy areas sang higher-frequency songs than those in quiet areas, but found that only males in already-noisy territories shifted songs upwards in immediate response to experimental noise. Unexpectedly, males in more quiet territories shifted songs downwards in response to experimental noise. These results suggest that chickadees may require prior experience with fluctuating noise to adjust vocalizations in such a way as to minimize masking. Thus, learning to cope may be an important part of adjusting to acoustic life in the city.

1. Introduction

Urban noise pollution may affect the diversity, density and breeding success of local avian communities [1–4]. Detrimental effects may be due to physiological stress and deterrence, but may also be due to acoustic interference when traffic noise overlaps in time and frequency with songs and calls of birds [5–7]. Such vocalizations play an important role in mate attraction and territory defence; masking noise can therefore undermine settlement and breeding success. As anthropogenic noise is often relatively low in spectral frequency, species with naturally high-frequency vocalizations appear to be generally less affected by urban noise than those with low-frequency vocalizations [1,8–11] (but see [12]). However, song plasticity is common among passerines, suggesting that researchers should consider not only what is biologically possible to sing, but also plasticity in what is actually sung in response to urban noise.

There are many bird species that can shift from singing relatively low- to high-frequency notes, and thus noise-dependent frequency use (spectral plasticity) may be an adaptive coping strategy [13–18]. Some species have been shown to shift individual notes or syllables within their songs to higher frequencies in noisy conditions [15,19], while others may selectively shift to higher-frequency song types from their repertoire [13]. A comparative study revealed that the link between geographical patterns of noise distribution and song frequency are especially strong for species that learn their songs through social experience [20]. However, although historical changes to population repertoires suggest that exposure to urban noise can influence loss or retention of different song types [21,22], to date no empirical data have shown that the degree to which songs are masked affects repertoire retention within individuals [23,24], nor is there any evidence that birds need experience to learn which song frequencies work best under different circumstances.

Even among species, which may be predisposed to avoid masking through innate spectral plasticity, learning or experience with anthropogenic noise may still be required to adjust songs in a way that results in masking release. Some species have the innate ability to dynamically shift or switch their songs (through song plasticity or selective use of repertoire; [25]). However, shifting to avoid anthropogenic noise is directional—only upward shifts will alleviate masking—whereas switching in other contexts (e.g. to avoid song matching) may allow bi-directional responses. Thus, even among species that can dynamically shift or switch songs, choosing which song variants in the repertoire effectively avoid masking may require prior experience with noisy conditions. Whether or not learning plays a role can be tested by comparing responses to experimental noise from birds in noisy territories near traffic versus those in quiet territories farther away, as the birds will differ in their experience with vocalizing under fluctuating noise levels.

Black-capped chickadees (Poecile atricapillus) lend themselves to testing whether there is a relationship between territorial noise levels and individual spectral plasticity. Unlike great tits (Parus major), black-capped chickadees do not have repertoires, but males can ‘pitch shift’ their single song type upwards or downwards through two to three distinct frequency groups during male–male territorial interactions [26,27]. Pitch shifting probably evolved as a mechanism to allow males to dynamically pitch-match (or avoid being pitch-matched by) rivals [28], but it may pre-adapt them with an efficient coping strategy to avoid signal masking from noise. Transmission experiments show that urban noise overlaps black-capped chickadee songs and has the potential to significantly reduce signal-to-noise ratios [29]. Further, observational and experimental exposure studies show that black-capped chickadees exhibit spectral plasticity in response to anthropogenic noise [30,31] and spectrally shift their songs away from narrow bands of overlapping noise [32]. However, we do not know whether this ability to avoid masking needs to be learned or whether the tendency to shift upwards in response to low-frequency noise is dependent upon prior experience with signalling under noisy conditions.

We therefore investigated the relationship between frequency use in male black-capped chickadees and ambient noise levels through the use of observational and experimental techniques over a relatively large geographical scale. Specifically, we addressed three main questions: (i) Does song frequency use correlate with local ambient noise levels across different populations? (ii) Do males show immediate spectral plasticity in response to experimental noise exposure? And most importantly, (iii) do immediate responses to experimental noise exposure correlate with local ambient noise levels?

2. Material and methods

(a). Field recordings and noise exposure

Fifty-three male black-capped chickadees were recorded in the regions inside of and around the cities of Prince George, Quesnel, Kelowna and Vancouver, in central and southern British Columbia, Canada. Of the 53 chickadees, 42 were used to determine how frequency use correlated with local ambient noise levels and 28 were used to determine how males respond to experimental noise (17 males were used in both analyses; electronic supplementary material, figure S1). Sites were chosen over a gradient of both urbanization and noise levels (i.e. noisy sites came from both urban and rural areas). Recordings were performed between 27 March and 15 May during the spring dawn-chorusing periods of 2011, 2012 and 2013. Once located, we recorded singing males for a minimum of 5 min prior to experimental noise exposure and continued to record them during the 5 min experimental noise treatment.

Experimental noise was a synthesized noise imitating the frequency spectrum of traffic. It was created with SoX v. 14.3.2 [33] by using the equalizer option to attenuate white noise by 3 dB every 500 Hz up to 10 kHz [13] (figure 1c). Experimental noise was broadcast from a Roland Mobile Cube amplifier (Roland Incorporation, USA) connected to a Philips GoGear Raga MP3 player (Philips, Canada). To avoid startling the chickadees, the volume of experimental noise automatically faded in and out over 20 s at the start and end of the broadcast. In calibration trials, amplitude of the noise playback was determined for different MP3 volume levels and distances. Throughout the 5 min trial, playback amplitude was therefore maintained at approximately 67 dB(Z) (65 dB(A)) at the location of the chickadee by adjusting the MP3 volume to match the distance to the bird. If the chickadee moved to a different distance, the speaker was moved and/or the MP3 volume adjusted to compensate. All recordings of focal males were made with MKH70 Sennheiser microphones (Sennheiser Inc., Canada) on to Marantz PMD671 Digital recorders (Marantz Canada, LLC; 22 bit and 44.1 kHz sampling frequency) at a distance of between 5 and 20 m. We measured local, non-experimental, ambient noise levels (dB(Z)) through one to three readings of approximately 30 s each (averaged) before, during and/or after the recording of each male using either a Pulsar 30 (Pulsar Instruments, UK) or a Gold Line SPL120 L (Gold Line, USA) sound pressure level meter. Local ambient noise ranged from 48 to 78 dB(Z).

Figure 1.

Examples of (a) a spectrogram of black-capped chickadee song, (b) chickadee frequency use during a chorus sequence and (c) a spectrogram (left) and frequency spectrum (right) of the synthesized experimental noise. In (b), the bee-note frequencies of 219 successive songs sung during 10 min of dawn singing by an individual male before (white circles) and during (black circles) experimental noise exposure are shown. This individual showed clear pitch shifting through three distinct frequency clusters. The grey bands represent the upper and lower 25% frequency bandwidths used to calculate proportions of song sung.

(b). Sound analysis

Songs were extracted from recordings using SoX [33] and Avisoft-SASLab Pro v. 5.2.02 [34]. All sound analysis was performed with the R bioacoustics analysis package, seewave v. 1.7.2 [35] in R v. 3.1.0 [36]. We used a Hanning window length of 1024 for all frequency measurements and a length of 256 for all temporal measurements. The start and stop of each note was manually marked on a seewave spectrogram, and a bandpass filter from 1.25 kHz below the lowest note to 1.25 kHz above the highest note was applied. As the birds maintain consistent frequency ratios within pitch-shifted songs [37], a single reference frequency can be used to classify the frequency of the whole song. The bee-note is a typical metric for representing frequency of shifting songs in black-capped chickadees as its frequency is more consistent (flat compared with the decreasing fee) [26,27,38]; we therefore measured the dominant frequency of the second half of the bee-note (the -eee; figure 1a) and used this to classify song frequency throughout this study. As song frequency is perceived on a log-scale, we log10-transformed all individual song frequencies prior to averaging [39]. Poor-quality songs, defined as either overlapped by loud discrete noises (other, nearby birds) or too distant to be reliably analysed, were omitted.

(c). Measurements

To determine the observed relationship between frequency use and local ambient noise conditions, we quantified three measures of frequency use for a 10 min period of recording prior to experimental noise or, if the male was not exposed to experimental noise, for a 10 min period of recording taken from the middle of the recording period. For each male, we calculated three different average song frequencies from all songs in the entire 10 min period based on the average of songs (i) sung overall (overall frequency), (ii) sung in the top 25% frequency range (highest-pitched frequency) and (iii) sung in the bottom 25% frequency range (lowest-pitched frequency). This resulted in a single averaged value per frequency measure per male. Frequency ranges were defined separately for each individual as the top or bottom 25% of their frequency range before exposure to experimental noise and omitting outlier songs (see below).

To examine the effect of experimental noise, we quantified five measures of frequency use from all songs in two 5 min periods of recording: one just before and one during experimental noise exposure. As above, for each male we calculated three different averaged song frequencies ((i) overall frequency, (ii) highest-pitched frequency and (iii) lowest-pitched frequency), but this time for each of the two 5 min periods. The frequency ranges were defined separately for each individual and each period as the top or bottom 25% of the frequency range calculated from songs sung in that particular 5 min period.

Black-capped chickadees do not gradually shift between frequencies, but sing songs in discrete frequency groupings (figure 1b); in response to noise they could shift songs upwards, but they could also sing fewer low-frequency songs and/or more high-frequency songs. This change may be detected as a small shift in average overall frequency, but would not be detected by changes in either the average of the highest- or lowest-pitched frequencies. Further, by only looking at changes in frequency, the strategy of shifting all songs up is indistinguishable from the strategy of singing fewer low-frequency songs. Thus, to determine whether chickadees use more or fewer songs from different discrete frequency groupings, we also quantified the proportion of songs sung from (iv) the upper bandwidth and (v) the lower bandwidth. Bandwidths were defined as the upper or lower 25% of the frequency range across all songs sung by that individual (i.e. across both 5 min periods) after omitting outlier songs (see below) (figure 1b).

In all cases, we omitted outlier songs before calculating frequency ranges or overall bandwidths. Outliers were defined as groups of up to 4 songs with frequencies above or below the 1.5× inter-quartile frequency range. We also omitted individual chickadees with fewer than 20 songs per 10 or 5 min period or when we were unable to calculate song frequencies for more than 15% of songs due to poor recording quality.

(d). Statistical analysis

We used the R package lme4 v. 1.1-11 [40] to create linear mixed models for all following analyses with region as a random intercept to account for any potential population effects between city regions; other studies have found minor regional differences in black-capped chickadee songs [41], but there were no regional differences found in this study. For mixed models, the R package lmerTest v. 2.0-30 [42] was used to calculate degrees of freedom using the Satterthwaite approximation and p-values.

To determine how local ambient noise correlated with frequency use in chickadees, we performed two analyses. First, we examined the relationship between overall frequency and ambient noise. Second, to determine whether chickadees adjusted highest- and lowest-pitched frequencies at different rates, the relationship between highest- and lowest-pitched songs and ambient noise were analysed together using frequency type (high or low) as a categorical predictor and including an interaction between ambient noise and frequency type. To account for the two observations per individual, we also included male ID as a nested random factor within region.

To analyse how local ambient noise affects the response to experimental noise exposure, we first calculated the difference in each measurement from before exposure to during experimental noise for each male. We then performed four analyses examining the relationship between these changes and local ambient noise. Similar to above, we analysed the change in overall frequency, as well as the combined analysis of the change in highest- and lowest-pitched song frequencies. We also conducted two additional analyses looking at the change in proportion of songs sung from the upper or lower bandwidths, respectively. Ambient noise was centred by subtracting the mean. Therefore, significant intercept estimates reflect significant overall changes in the response to experimental noise at an average level of ambient noise. Significant slope estimates of local ambient noise reflect an effect of local ambient noise on the change in frequency use during experimental noise exposure.

Physical changes to habitats that accompany urbanization often covary with anthropogenic noise; thus, to avoid this potential confound we selected sites across a broad range of each of urbanization and ambient noise (including both urban quiet and rural noisy). Further, we initially included an index of habitat urbanization (derived from PCA analysis of habitat characteristics assessed from Google Earth maps; methodology described in detail in [43]) as a covariate, and confirmed that there was no significant relationship between habitat structure and any of our response variables (with or without including local ambient noise levels, all p > 0.05). We therefore omitted this covariate and concentrated solely on the effects of local ambient noise and experimental noise.

For all analyses, assumptions of normality and constant variance were satisfied. Figures were created with R package ggplot2 v. 2.1.0 [44] and show model predictions, which demonstrate the relationship between response and explanatory variables while controlling for random effects. To improve interpretation of figures, frequency response variables were back-transformed (power-transformed) to Hz and frequency change variables were back-transformed to ratio changes prior to plotting. Proportions were left as proportions. All statistics were performed using R statistical software v. 3.2.5 [36]. Relevant statistical results are presented in the text and complete model results are found in electronic supplementary material, tables S1 and S2.

3. Results

(a). Correlations between frequency use and local ambient noise

During natural singing, there were significant positive correlations between all measures of frequency use and local ambient noise levels. Across individuals, average overall song frequency ranged from 3022 to 3517 Hz and increased with ambient noise (figure 2a; t₄₀ = 2.42; p = 0.020). Average frequency of highest-pitched songs ranged from 3108 to 3635 Hz and frequency of lowest-pitched songs ranged from 2934 to 3379 Hz. Overall, highest- and lowest-pitched songs also increased with ambient noise (figure 2b; t₇₂ = 2.32; p = 0.023). As defined, highest-pitched songs were significantly higher than lowest-pitched songs (t₄₀ = −2.44; p = 0.019). The interaction between frequency type (high versus low) and ambient noise was not significant (t₄₀ = 0.87; p = 0.389); thus the slopes of highest- and lowest-pitched songs did not differ.

Figure 2.

In black-capped chickadees song frequency increased with local ambient noise levels in (a) all songs, and in (b) both highest- and lowest-pitched songs. However, while slopes did not differ between highest- and lowest-pitched songs, intercepts did (b). Frequencies have been back-transformed from logs to Hz. Points are raw data values from individual territories. Lines represent model relationships (also back transformed). n = 42 individual chickadees in each panel.

(b). Experimental noise exposure and immediate spectral plasticity

Exposure to experimental noise did not result in any overall changes in absolute frequency before versus during exposure (all intercepts p > 0.676). However, changes in frequency use during experimental noise exposure did correlate with local ambient noise (figure 3). There was a non-significant trend for the change in average overall song frequency, which ranged from −400 to 286 Hz, to increase with ambient noise (figure 3a; t₂₀ = 2.02; p = 0.058). The change in average frequency of highest-pitched songs during versus before noise exposure ranged from −341 to 305 Hz, and the change in lowest-pitched songs ranged from −297 to 345 Hz. Overall, highest- and lowest-pitched frequencies increased with local ambient noise levels (figure 3b; t₄₃ = 2.21; p = 0.032). There was no significant interaction between frequency type and ambient noise (no difference between the slopes; t₂₆ = 0.42; p = 0.678), nor was there a difference in the magnitude of change between frequency types (t₂₆ = 0.41; p = 0.687).

Figure 3.

During experimental noise exposure, the change in frequencies sung ((a) all songs; (b) both highest- and lowest-pitched songs) by black-capped chickadees depended on local ambient noise levels. In all cases frequency increased with local ambient noise. However, neither slopes nor intercepts differed between highest- and lowest-pitched songs (b). Frequency changes were back-transformed from logs to ratios. Ratios represent a factor of increase (e.g. 1.10 times or 0.95 times the initial value indicate an increase or decrease, respectively). Points are raw data values from individual territories. Solid/dotted lines represent model relationships. The horizontal dashed lines represent point of no change from before experimental noise: points below the line reflect a downward change in frequency, points above the line reflect an upward change in frequency. n = 28 individual chickadees in each panel.

The proportions of songs sung in upper or lower bandwidths also showed no overall changes during versus before experimental noise (all intercepts p > 0.22). Across individuals, changes in the proportion of songs sung from the upper bandwidth ranged from −90.0 to 87.8%, but there was no correlation between this change in proportion and local ambient noise levels (figure 4a; t₂₀ = 1.71; p = 0.102). By contrast, changes in the proportion of songs sung from the lower bandwidth during versus before noise exposure ranged from −92.5 to 99.0%, and significantly decreased by 3.2% for every dB(Z) increase in local ambient noise levels (figure 4b; t₂₆ = −2.61; p = 0.015).

Figure 4.

During experimental noise exposure, the change in the proportion of songs sung from (a) the upper bandwidth differed from (b) the lower bandwidth. The change in proportion of songs sung from the upper bandwidth (a) showed no relationship with local ambient noise. However, the change in the proportion of songs sung from the lower bandwidth (b) decreased as local ambient noise increased. Points are raw data values from individual territories. The dotted line represents a significant model relationship. The horizontal dashed lines represents point of no change from before experimental noise: points below the line reflect a downward change in the proportion of songs sung, points above the line reflect an upward change. n.s. = non-significant. n = 28 individual chickadees in each panel.

4. Discussion

We measured acoustic responses of black-capped chickadees to experimental noise exposure across a large geographical area and over a wide range of ambient noise levels. Using this study design, we found that male black-capped chickadees sang higher-frequency songs as local ambient noise increased [31] and that they responded with immediate spectral plasticity to broadcasts of experimental noise [30,32]. However, our design also revealed that the magnitude and direction of pitch-shifted responses varied with the levels of local ambient noise in the subject’s territory. When presented with experimental noise, males in noisy territories (presumably familiar with noisy conditions) quickly shifted their song frequencies upwards (observed as an increase in overall mean frequency as well as in parallel increases in frequency in highest- and lowest-pitched songs). They also switched to singing fewer songs from their lower bandwidth, but showed no change in the proportion of songs from their upper bandwidth. Males occupying quiet territories (presumably less familiar with noisy conditions) also adjusted their vocalizations, but decreased their song frequencies and sang more songs from their lower bandwidth. Birds in territories with intermediate noise levels were also intermediate in their vocal response. Thus, only black-capped chickadees with prior experience with noise showed immediate spectral plasticity that could potentially improve communication via masking release.

(a). Learning to be flexible

Our results suggest that, in the case of anthropogenic noise, the ability to achieve masking release through pitch shifting may depend on previous experience. Males in our study were not necessarily geographically distant from each other, yet they differed in their responses depending on local ambient noise levels. Males in noisy territories pitch shifted to higher frequencies, whereas those in quiet territories shifted to lower frequencies. Data suggesting that learning may be involved in the flexibility of spectral responses to anthropogenic noise have not been reported before. Previous studies addressing immediate spectral plasticity in response to noise have either been conducted at a single location, where all territories were similarly noisy, or local noise levels were not taken into account [13,14,30,45].

Pitch shifting probably reflects a strategic response to fluctuating noise conditions, enabled by a pre-existing pitch-shifting ability developed for social situations or for response to natural ambient noise fluctuations. This contrasts with reflexive responses such as the Lombard effect, in which elevated noise levels inherently induce a rise in amplitude, which in turn can be associated with a small increase in frequency [46]. Pitch shifting in the context of anthropogenic noise is more likely to be related to this species's pre-existing ability to pitch shift during dynamic vocal interactions [26–28]. However, natural ambient noise levels can also fluctuate over time (due to wind, rain, insects or other birds) or space (depending on proximity to flowing river or rustling leaves) [47–50]. Thus, even under natural conditions, pitch shifting may be used in non-social contexts to reduce masking by ambient noise.

Pitch shifting to reduce masking by ambient noise may also explain the unexpected result of males in quieter areas shifting downwards in response to experimental noise. If they simply had not learned how to respond to noise, we would have expected them to shift randomly. However, the cumulative volume of bird song across species during the dawn chorus can be extreme, particularly in the higher frequencies, and can greatly interfere with signal perception [51]. Thus, if noise in quiet areas is generally sung from other bird species (or even insect noise), it might be appropriate to shift down to avoid high-frequency interspecific masking. An added bonus of this strategy would be that lower-frequency songs attenuate less quickly than higher-frequency songs and will thus propagate farther, particularly in forested environments [52]. Further, it may also be physiologically easier for birds to produce lower-frequency songs than higher-frequency songs [53], and there is evidence that fee-notes are relatively louder in low-frequency songs compared with high-frequency songs [38].

Thus, in generally quiet areas, even if occasional bursts of noise are low-frequency (e.g. wind or intermittent anthropogenic noise), shifting song frequencies downwards could have the effect of maximizing song amplitude while avoiding conspecific (or heterospecific) masking. In areas where noise is consistent, loud and dominated by low-frequency anthropogenic sources, this strategy may not be enough to overcome noise interference. Over time birds may learn to shift songs upwards rather than downwards, which would have the effect of minimized masking from anthropogenic noise. Consequently, the use of pitch shifting in response to anthropogenic noise, as seen in our study, may be an extension of pitch shifting in either social or natural noise-related contexts, or both. However, in no context are we aware of any empirical data suggesting that adults require experience to pitch shift appropriately.

(b). Implications for urban success

The detrimental impact of masking by urban noise is frequency-dependent, as traffic noise has a strong bias to low frequencies. Consequently, negative effects on the distribution and density of breeding territories and reproductive activities are biased towards species that rely on communication through low-frequency vocalizations [1,2]. Singing at high frequencies or the ability to shift signals upwards in pitch may lead to masking release [54], although these have not been shown to actually result in fitness benefits [55]. Further, it may well be that a pitch shift increases audibility but decreases signal attractiveness and thereby confronts the singer with a trade-off in a frequency-dependent functional compromise [55–57].

For example, pitch shifting may enable black-capped chickadees to quickly and competently reduce masking by anthropogenic noise, but this use may also interfere with communication. Both relative note amplitude and consistency in frequency ratios between notes convey information about dominance in male black-capped chickadees [38,58]. Discrimination between high- and low-ranking males based on these song characteristics is best observed in low-frequency songs [38]. Thus, males avoiding low frequencies may inhibit the ability of conspecifics of either sex to assess their quality. Further, avoiding lower-frequency songs may limit a male's vocal range, and thus make it more difficult to perform well during dynamic pitch shifting interactions [27], which can result in males losing paternity in their nests [59]. Thus, pitch shifting to minimize the effects of masking could result in a functional compromise [55,56], such that masking may be reduced but at the expense of efficient communication with both conspecific females [4,60] and males [61,62].

Even if switching to higher-frequency songs is adaptive, with few or no functional compromises, the time scale over which learning occurs or the ability to learn may limit urban colonization and/or persistence. Our results show that ‘immediate’ plasticity probably requires learning or experience to develop, and thus time and ability. Longer-term studies following multiple noise exposures to the same bird may reveal how long it takes for this ability to develop, or whether or not it can even be learned by all birds. Birds in more quiet territories could learn to pitch shift to higher frequencies either through direct auditory experience/feedback, or through social feedback [13]. However, birds that have learned to cope with noise may be especially smart and innovative. Urban noise could therefore be contributing to selection for typical features of successful urban phenotypes [11,12,63–65]. Black-capped chickadees are fairly common in urban green spaces, and thus the ability to learn or the time taken to learn appropriate spectral plasticity may not be problematic. The aptitude for quickly learning to cope may even provide them with a benefit relative to other, less flexible species that do less well in and around noisy cities.

Ethics

This work was carried out with approval from University of Northern British Columbia Animal Care and Use Committee (protocol no. 2011-05).

Data accessibility

Data reflecting frequency measures and ambient noise levels are available at Dryad: http://dx.doi.org/10.5061/dryad.669qn.

Authors' contributions

S.E.L. participated in study conception, design and coordination, collected field data, curated data, carried out statistical analyses and drafted the manuscript. H.S. participated in study conception and design, and helped to review and edit the manuscript. K.A.O. participated in study conception, design and coordination, and helped to review and edit the manuscript. All authors gave final approval for publication.

Competing interests

We have no competing interests.

Funding

Funding was provided by the James L. Baillie Memorial Fund from the Society of Canadian Ornithologists (SCO-SOC) to S.E.L., by the Natural Sciences and Engineering Research Council of Canada (NSERC) through a personal PGS doctoral scholarship to S.E.L. and through a Discovery grant to K.A.O., and by the University of Northern British Columbia through a Graduate Entrance Research Award and a Research Project Award to S.E.L.