Wild zebra finches are attracted towards acoustic cues from conspecific social groups

Abstract

Social information gathered by observing others often supplements personal information collected from direct interactions with the physical environment during decision-making. Social information use may be particularly beneficial in harsh environments or if resources are distributed patchily, ephemeral, and unpredictable, and hence difficult to locate. We experimentally tested the use of acoustic cues in wild zebra finches (Taeniopygia guttata) as they flew around their arid habitat as a way of locating conspecifics on the ground, and potentially accessing useful social information. Joining a conspecific group may reduce the predation risk, and if they are foraging may also improve foraging efficiency, as the distribution of zebra finch food (grass seed) is scattered and unpredictable in their natural habitat, the Australian arid zone. We conducted playback experiments along vegetated creek lines radiating out from an artificial dam where all birds in the population were coming for drinking water. We broadcast recordings of vocalizations from foraging conspecific groups to birds using these creek lines to move to or from the water into the wider habitat. Zebra finches were more likely to land near the loudspeaker when conspecific vocalizations were broadcast compared to white noise. Birds flying low and close to the loudspeaker were most likely to land. Our results indicate that zebra finches use acoustic cues of conspecifics as a source of social information for grouping decisions. Use of such information may also enhance foraging efficiency in environments with unpredictable and scarce foraging locations, and reduce predation risk for calling and responding individuals.

INTRODUCTION

Individuals need to make many decisions on a daily basis. How effective these decisions are will depend on the information available to an individual, as this information facilitates decision-making among various alternatives. By collecting and using information during the decision-making process, an individual can optimize its behavior and may gain fitness advantages (Dall et al. 2005). Information can be obtained through own experience and direct interaction with the physical environment (personal information; Danchin et al. 2004; Dall et al. 2005) or through monitoring other individuals (social information; Valone and Templeton 2002). Information gathering entails costs for an individual (e.g. cost of time and energy), yet obtaining social information is normally less costly than obtaining personal information (Kendal et al. 2005). Social information may often be faster to obtain, but it can also be more prone to mistakes and can be irrelevant for an individual (Laland 2004; Kendal et al. 2005).

Socially acquired information can be produced either deliberately by signals or inadvertently by social cues and can comprise the mere presence and spatial location of an individual, its behavior, or its performance (Reed and Dobson 1993; Danchin et al. 2004; Dall et al. 2005). For instance, the presence of signaling resident birds attracts non-resident (Amrhein et al. 2004; Roth et al. 2009) and resident (Bircher et al. 2020) conspecific prospectors, and conspecifics to settle nearby (Alatalo et al. 1982; Ahlering et al. 2006); juvenile lizards preferentially establish territories near residents, leading to territorial aggregations (Stamps 1988); cliff swallows, Hirundo pyrrhonota, join conspecifics as soon as they see them foraging (Brown 1988); and pied flycatchers, Ficedula hypoleuca, prospect on conspecifics’ clutch sizes (Schuett et al. 2017) and use the clutch sizes of heterospecifics as cues to make breeding decisions (Forsman et al. 2012). House sparrows, Passer domesticus, produce calls to actively recruit conspecifics to foraging sites to reduce the risk of predation (Elgar 1986).

Social information is expected to be particularly important in unpredictable environments as information requires regular update, and social information provides real-time cues. A few laboratory studies have supported the prediction that social information use is important in unpredictable environments. For example, unpredictability in food quality led to increased social information use in nine-spined sticklebacks, Pungitius pungitius, in the laboratory (van Bergen et al. 2004). Similarly, starlings, Sturnus vulgaris, utilized social information under unpredictable but not predictable foraging conditions in the laboratory (Rafacz and Templeton 2003). Social information use in highly unpredictable environments, such as arid environments without a clear seasonality, has been rarely studied in the wild. However, in a breeding context, a recent study of the zebra finch, Taeniopygia guttata, found support for the use of social information with adults preferentially initiating breeding next to conspecifics at early breeding stages (Brandl et al. 2019b).

In a foraging context, social information may be particularly beneficial if food is patchily distributed, ephemeral and unpredictable (Rafacz and Templeton 2003; Egert-Berg et al. 2018) and hence difficult to locate by an individual. Individuals may then use social signals or social cues, for instance, those that reveal food locations from the presence of individuals with similar dietary requirements (social attraction) or from observing their (foraging) behavior from afar (local enhancement; Danchin et al. 2004). Joining foraging individuals or groups can have many advantages to an individual and these can outweigh potential disadvantages, such as competition, depending on the context, food availability at the foraging site, and an individual’s needs (Krause and Ruxton 2002). Advantages include reduced predation risk via dilution effects (Foster and Treherne 1981) or enhanced vigilance (Valone 1989; Beauchamp 2013) as well as increased discovery rates of food patches (Valone 1989), improved individual foraging efficiency, and decreased individual variance of food intake (Clark and Mangel 1984; Brown 1988; Giraldeau and Beauchamp 1999; Giraldeau and Dubois 2008).

Zebra finches are likely to use social information in a foraging context, given that they live in fission-fusion societies in ecologically harsh, unpredictable and spatially and temporally variable arid environments in Australia (Zann 1996; Griffith and Buchanan 2010; Morton et al. 2011; Funghi et al. 2020). Within the desert, zebra finches occur in areas of higher local plant productivity, which generate the grass seeds on which they forage (Stafford Smith and Morton 1990). Grass seeds are often patchily distributed, such that zebra finches sometimes search for feeding sites over many kilometers (Zann 1996). The high unpredictability of rainfall, and hence grass seed productivity in their environment may often leave zebra finches with outdated information about foraging sites. Combined with a potentially limited energy budget in harsh conditions, the use of social information to locate groups of zebra finches and potentially feeding sites should be particularly beneficial for them. Previous work on captive zebra finches has demonstrated that they are sensitive to, and able to use, information about the location of food provided by conspecifics (e.g. Beauchamp, 2006; Farine et al. 2015). Additionally, their food source (grass seed) is non-defendable but easily divisible, reducing the cost of social foraging (Giraldeau et al. 1990). Thus, it may even be possible that foraging conspecifics produce signals (e.g. calls) addressed at other individuals passing close by in addition or alternatively to undeliberate cues during foraging. The benefit of promoting social foraging to individuals already on a patch is that feeding on the ground in a relatively open environment with little shelter from vegetation makes zebra finches highly vulnerable to predation, especially to birds of prey (Zann 1996). Foraging in a group may therefore provide a variety of anti-predator benefits. Receivers would benefit from such signals/cues, as they can use them to locate foraging sites, next to also receiving anti-predator benefits.

Previous work has found that zebra finches rarely forage on their own and typically pairs or small groups of up to around 15 individuals forage together (McCowan et al. 2015; Brandl et al. 2021). Zann (1996) observed small feeding groups being joined by other small groups that were flying over foraging groups. The underlying mechanism promoting this social behavior has yet to be studied. Possible cues are the calls of conspecifics during foraging, or whilst moving around or resting between bouts of foraging, or the visual cue of seeing them on the nearly bare ground, or a combination of both. Social information use in foraging contexts has frequently been shown for visual cues like the location or behavior of observed animals foraging in various taxa (reviewed in Galef and Giraldeau 2001). Acoustic cues are also used in foraging contexts (e.g. Clay et al. 2012; Hillemann et al. 2019) and may often be easier to discern than visual cues, especially in densely vegetated areas, over longer distances or when resources are patchily distributed (Snijders and Naguib 2017; Martinez et al. 2018). In the case of vocalizing birds feeding at unpredictable and patchy locations, it may be more effective to use auditory over visual cues. In some species, individuals give specific food calls once they have located a food source (Bugnyar et al. 2001; Janik 2000; Clay et al. 2012). However, in social foragers such as zebra finches calls are of relatively low amplitude, and do not travel far when their hearing sensitivity is considered (Loning et al. 2022). Therefore, zebra finch short-range calls may predominantly be within group signals and only within a limited range function as food calls specifically given to attract and recruit conspecifics.

The aim of this study was to determine whether zebra finches use acoustic cues produced by conspecifics to home in, and join, or investigate those conspecifics. This aggregative behavior would provide opportunity for social benefits and the capture of social information, such as on locations of foraging sites. In a field experiment, we tested whether playback of vocalizations from zebra finches attract flying conspecifics to land nearby. The experiment was conducted close to pre-identified frequently used flight highways—vegetated creek lines radiating from a water source to the broader environment. We broadcast either calls recorded from naturally foraging zebra finches (treatment) or white noise (control). We predicted that if zebra finches use conspecific acoustic cues as social information, zebra finches passing over a playback of conspecific calls would land near the loudspeaker but would not land in response to the broadcast of control white noise. We further predicted that zebra finches are more likely to respond to the playback on their way out from a water source to the broader environment than on the inbound trip, as we expected drinking to be of higher priority on inbound trips in this dry and hot environment. Finally, we predicted zebra finches to be more likely to land early in the morning than during later hours, as foraging activity is high during the early hours of the day and then decreases (Zann 1996; Funghi et al. 2019).

METHODS

Study site

The playback experiments were conducted in November 2016 in Gap Hills (30°56’58.1”S 141°46’02.2”E), New South Wales, Australia. The study site is an area of about 1.5 x 2 km centered around an artificial dam that is the only drinking water resource for the birds in the vicinity. At the study site zebra finches have been monitored since 2004 (Griffith et al. 2008) and are provided with artificial nest boxes for breeding (e.g. McDiarmid et al. 2018; Brandl et al. 2019a). In the calendar year of the study, as part of ongoing population ecology, we banded over 700 adults and juvenile birds in the area in which the study was conducted (and over 500 nestlings as well), and there were many hundreds of un-banded birds around as well.

Recordings and preparation of playback files

The sound files for the experiment were recorded several days before the onset of the experiment and contained zebra finch calls of naturally foraging zebra finch groups. We used a digital audio recorder (H4n Handy Recorder, ZOOM, New York, USA; wav files; sampling rate 44.1 kHz) and a directional microphone (ME66/K6, Sennheiser, Wedemark, Germany with a Classic-Softie Windshield, Rycote, Stroud, England) in a mount (InVision Softie Lyre Mount with Pistol Grip, Rycote, Stroud, England). Recordings of foraging zebra finch groups (two to approximately 20 individuals) were made from a hideout at a close distance (approximately 10 m). Of the recordings collected, we chose a total of 16 high quality recordings taken at different times and locations (recordings with frequently calling birds and low background noise; see below): twelve recordings of zebra finches randomly found foraging and socializing in Gap Hills, as well as four recordings of zebra finches in an experimental foraging context as part of another experiment. The 16 recordings were then used to construct 16 playback files (one playback file per original recording).

The playback files were constructed using Avisoft-SAS Lab Pro (R. Specht, Berlin, version 5.2.10). Per recording, we identified a sequence with frequently calling zebra finches and with as little background noise as possible. Playback files thus contained a variety of different zebra finch calls given during a natural foraging context, but we only used segments without song or begging calls (song and begging calls were rare on these recordings anyway). We constructed one-minute sound files from those recordings ranging from 100 to 360 calls per minute. Depending on the duration of the natural sequences, sequences recorded at one location were edited together to create sequences lasting one minute. The number of calls on each sound file was counted using their sound spectrograms in Avisoft. In three cases with fewer calls, we removed short silence sequences to create a playback file with at least 100 calls per minute. To remove remaining disturbing noise, the call sequences were filtered with the FiR bandpass filter (high pass: 0.5 kHz, low pass 12 kHz) and normalized to 90%. In addition to these 16 sound files with zebra finch calls suitable for the experiment, we generated a one-minute white noise sound file for a control treatment, using Avisoft.

Playback experiment

In total, we conducted 460 playback trials on 16 days (between 10^th and 28^th November 2016) between 7:00 and 12:30 hours at eight different locations within Gap Hills. The distance between locations ranged from 102 to 1107 m (mean 521 m, SE 251 m). Prior to the overall experiment, we had identified flyways that were frequently used by zebra finches. To identify flyways we had collected data on flight routes during regular circular transects around the artificial dam (200 m, 400 m, 600 m, and 800 m distance) between September and November 2016. The eight fixed locations for the playback experiments were located in about 50 m distance from such flyways and at least 50 m from the closest active nest box, but otherwise randomly set on a map (distance from water dam ranged between ca. 200 m and 700 m).

Playback trials were organized in sessions at one location. During each session at a given location, ten playback trials (each lasting one minute), of which in five trials we broadcast zebra finch calls and in five trials we broadcast white noise, were always carried out consecutively in an alternating order (see below). The white noise served as procedural acoustic control. One to five sessions, each consisting of the ten trials, were done each day but at different locations to avoid testing the birds on the same flyway several times. As soon as one or several zebra finches were seen approaching, a trial was started (i.e. a playback broadcast from the loudspeaker) when the zebra finch(es) flew towards the location of the loudspeaker or approached within 20 m distance (distance estimated) and if there was no other zebra finch present already near the loudspeaker (e.g. perching or foraging within a radius of 20 m). Playbacks were discarded when zebra finches were already above the loudspeaker when the playback started. Furthermore, if no zebra finch landed, we waited at least one minute between two trials to avoid that a bird in the distance could have heard the previous trial in advance. If one or more birds had landed during a trial, no new trial was started until all birds had left the site again.

Playbacks were broadcast through a JBL loudspeaker (JBL Clip, JBL, Los Angeles, USA), connected with a 10 m cable to a portable media player (iPod Nano, Apple, Cupertino, USA). Prior to playback, the amplitudes of all sound files were measured using a sound pressure level (SPL) meter (A setting, fast response, 1m high, outdoors at low wind, GA206 Sound Level Meter, Castle Group Ltd., Scarborough, England). All sound files, including the white noise control, were broadcast at a mean of 74 dB (mean amplitudes of all used sound files; range 69-81 dB). The amplitude was based on a previous laboratory study on zebra finch song amplitude of single males (Brumm et al. 2009) and an estimate of calling amplitude of single birds by Mouterde et al. (2014), taking into account that the sound amplitude may be slightly elevated when multiple individuals are calling at the same time. The amplitudes in these studies turned out to be louder than our own subsequent measurements from calibrated recordings in our field population (Loning et al. 2022). The loudspeaker was hidden along zebra finch flyways on the ground in a small bush, similar in height to the level at which zebra finches would normally be foraging and calling. One observer (C.A.) sat quietly on a stool around 10 m from the loudspeaker and watched for approaching birds. As soon as one or several zebra finches were seen approaching, a playback was initiated by starting the playback of one of the zebra finch recordings (randomly selected of those available from other locations) or of white noise. The order of playback treatments was alternated within each session, i.e. playbacks alternated between zebra finch calls and noise. In each session, five different zebra finch recordings were used, and no file was used twice. The sound files with zebra finch calls broadcast at each location of the experiment had been recorded at a different location in Gap Hills, to prevent the possibility that zebra finches were exposed to their own or familiar calls. Thereby, there were different possible sound files for each location of the experiment, which were used similarly often (14 to 18 times per sound file, 1 to 3 times per sound file per location). In every other session at a single location (i.e. at different days as only one session occurred at a location per day), the direction of observation by the observer was changed, so that one session considered birds flying only towards the water source (artificial dam) and the next session at that location only considered birds flying away from the water source.

For each trial, we noted the date, time, location (waypoint 1 to 8), and the sound file used (white noise or zebra finch file 1 to 16). The number of approaching zebra finches was counted for each trial as well as the number of responding birds within the group. The response was initially classified into five categories: (i) no response, (ii) flying lower and slower, (iii) flying circles, (iv) turning towards the loudspeaker, or (v) landing (in vegetation or on the ground within approximately 20 m of the playback speaker). Birds not responding followed their original flight direction without noticeable changes. For subsequent analyses, the responses were grouped into two categories: whether or not at least one bird of the group landed during a trial (please note that a “group” could here also consist of a single bird if a single bird was approaching during a trial). This binary variable “landing” (yes or no) thus provided the strongest response to the playback within the group. For the flight direction, a general direction was recorded, stating if the group was flying to or away from the water source. The flight height prior to response was noted as either “high” or “low”. For both categories, the estimated height (“high”: higher than approximately 5 m, “low”: lower than approximately 5 m), and the style of flying (“high”: straight and fast, “low”: following vegetation structure, lots of undulations, slower) was crucial, according to the two zebra finch flight modes described by Zann (1996).

Statistical analyses

Data were analyzed with generalized linear mixed-effects models (GLMMs) with binomial or Poisson error structure (for binary and count response, respectively). The sound file with zebra finch calls used in a trial as well as session nested within location were set as random terms (random intercepts) in all models (unless stated otherwise) to account for the potential pseudoreplication caused by repeated tests during sessions and at the same locations.

Even though we could not individually identify birds during the experiment, the number of birds at the study site was extremely large during the study (see section “Study site”) and hence it is unlikely that our data is all based on the same small set of birds measured repeatedly. Furthermore, individual zebra finches at our site (S. Griffith et al., unpublished data) and at another location (Zann 1996) usually only drink about once per day from a water source (unless temperatures approach 40^oC, which they did not during this study). Since a session of trials on a day was short and only recorded birds either flying towards or away from the water source at a location and locations were set at different flyways, it seems unlikely that we measured birds repeatedly on the same day. Also, as the time window for trials span 5.5 h across the day (and we assume that birds are reasonably consistent in their drinking times), it seems unlikely that many birds were recorded repeatedly across the study. Nevertheless, we conservatively applied permutation analyses on substantially, randomly reduced sample sizes of subsets of our data (see Supplementary Figure S1). Results of the permutation analyses revealed that reported effects of our sound treatment (see Results) were extremely robust against even large reductions in sample sizes. This suggest that even if some extent of pseudoreplication was present in our data due to repeated measures of some individuals, this is unlikely to have changed the outcome of the study, even if those trials would be removed completely. Additionally, we can expect that most of such potential pseudoreplication is already captured in our random terms where sessions were nested in locations and hence repeated measures at locations are already accounted for. One may even speculate that reported effects could be even larger in reality, as presumably repeated sampling of the same individuals would reduce their likelihood to land if they hear conspecific sounds in our playback experiment (e.g. they could learn that a particular site is not a location at which zebra finches are present).

The number of calls in a sound file and the day of trial had no influence on whether at least one bird in a group landed during a trial (GLMM with binomial error structure; number of calls: χ ² = 2.12, DF = 1, P = 0.145, day: χ ² = 2.44, DF = 1, P = 0.119, N = 460 trials). Therefore, neither the number of calls nor the day was considered in any further analysis.

In the first model, we tested whether the response (binary variable: “landing”, i.e. at least one bird landed) depended on the treatment (zebra finch calls or white noise), the time of day (in hours; data mean-centered), the group size (number of birds approaching in a trial; mean-centered), whether the group was flying towards or from the direction of the water source, the flight height (high or low) and the passing distance to the loudspeaker (estimated distance between speaker and group at trial start, in meters, range 0-20 m; mean-centered). In the second model, we assessed whether the group size of zebra finches passing by, the treatment or the time of day (fixed terms) influenced how many zebra finches landed. This model only included trials in which at least one zebra finch responded (i.e. landed). We also ran one hurdle model as alternative to above two models (same explanatory variables) which gave qualitatively the same results (see Supplementary Table S1). A hurdle model contains a mixture of binomial and count model, where the count model does not include any zeros (Zuur et al. 2009; in our case: the birds either landed or not, but it was not possible that zero birds landed if landing occurred). We decided to present the model outcome of the above two individual models in the main text, as the main question of our study is whether birds are more likely to land when hearing the zebra finch calls compared to the control, white noise playback (first model).

In the third model, we aimed to validate the white noise as a control treatment and to verify that the noise did not deter the zebra finches. We used a data subset of white noise trials only and tested the effect of time, number of zebra finches passing, flight direction, and height as well as the distance between zebra finches and speaker on the birds’ response (i.e. whether at least one bird landed) in a trial. If zebra finches were deterred by the sound of white noise, we would expect that birds flying more closely to the playback would be less likely to land than those flying past at a further distance to the speaker. This model included the session nested within location as random term.

Maximal models were simplified by stepwise deletion of the least significant terms, as long as the removal did not significantly reduce the explanatory power of the model (likelihood ratio tests between subsequent models; Crawley 2007). We provide associated test statistics in Table 2. In addition, we provide the test statistics for the full models as well as likelihood ratio tests comparing each maximal model and its respective null model (without any fixed but random terms as specified above; Forstmeier and Schielzeth 2011) in the Supplementary Table S2. The data were analyzed using R (version 3.1.2 R Development Core Team 2020). All GLMMs were conducted using the function “glmer” or “lmer” of the R package “lme4” (Bates et al. 2014). Model assumptions were visually assessed using diagnostic plots and we tested for overdispersion (all models were underdispersed).

Table 2

Summary of test statistics from GLMMs of the playback experiment. Model 1 used the whole data set (treatment and control trials) to test which variables influenced the likelihood that at least one bird in a passing group landed (binary response variable “landing”); model 2 only used data from trials in which at least one bird in a passing group landed to assess which variables influenced how many birds landed (response); model 3 used only data from white noise (control) trials and had the binary variable “landing” as response. Significant P-values highlighted in bold.

Model	Response	Random term	Variance	Fixed term	Coef.	Χ2	df	P	N
Model 1	Landing	Sound file	0.123	(Intercept)	–3.615				460
(All data)	yes/no	Session: Waypoint	0.215	Sound treatment [ZF]	1.593	4.934	1	0.026
		Waypoint	0.204	Time	(0.093)	0.514	1	0.473
				Number of birds	(–0.139)	2.518	1	0.113
				Direction [to]	(–0.638)	3.613	1	0.057
				Flight height [low]	0.875	6.964	1	0.008
				Distance to speaker	–0.063	6.965	1	0.008
Model 2	No. birds	Sound file	<0.0001	(Intercept)	0.678				58
(Birds	responding	Session: Waypoint	<0.0001	Sound treatment [ZF]	(–0.179)	0.795	1	0.373
responding		Waypoint	<0.0001	Time	(–0.003)	0.001	1	0.971
data only)				Number of birds	0.259	39.23	1	<0.001
Model 3	Landing	Session: Waypoint	<0.0001	(Intercept)	–3.048				230
(White noise	yes/no	Waypoint	<0.0001	Time	0.289	1.378	1	0.241
trials only)				Number of birds	0.095	0.627	1	0.428
				Direction [to]	(–0.529)	0.735	1	0.391
				Flight height [low]	(0.393)	0.409	1	0.522
				Distance to speaker	–0.079	3.407	1	0.065

Χ²: Χ² value of likelihood ratio test; coef.: coefficient; df: degrees of freedom; P: significance after likelihood ratio test; N: number of trials; ZF: zebra finch calls. Coefficients in brackets: coefficients of non-significant terms just before dropping the terms; other coefficients: from minimal adequate model (coefficients in brackets cannot be compared to coefficients from the minimal adequate models, since the simplification alters coefficients). Coefficients for a factor level (specified in square brackets) give the difference to the reference level. Bold P-value denotes significant effect. The time, number of birds and distance to speaker were mean-centered.

Ethics statement

The project was approved by the Macquarie University Animal Ethics Committee (ARA#2015/017), NSW Parks and Wildlife Service, and the Australian Bird and Bat Banding Scheme. It followed the ASAB and ABS guidelines for the use of animals in research.

RESULTS

The likelihood that at least one zebra finch landed in a trial was significantly higher when conspecific calls were broadcast (in 20.0% of 230 zebra finch call trials) compared to when the white noise control was replayed (in 5.2% of 230 white noise trials; see Table 1, Model 1 in Table 2, Figure 1a, Supplementary Figure S2, Supplementary Tables S1 and S2). Birds that passed the loudspeaker closer, both in height (Figure 1b) and in distance, were significantly more likely to land (Model 1 in Table 2, Supplementary Tables S1 and S2): In trials in which at least one zebra finch landed, the median distance at which birds flew past the loudspeaker was 5 m (maximum: 20 m); whereas zebra finch groups that did not land flew past at a median distance of 10 m. At the same time, at least one zebra finch landed in 17.1% of trials in which zebra finches were flying low, but in only 9.5% of trials in which zebra finches were flying high. Birds that flew back from the water source to the broader area away from the dam tended to be more likely to land than birds on the inbound flight to water (Model 1 in Table 2, Supplementary Tables S1 and S2): at least one zebra finch landed in 10.4% of the trials during inbound trips, whereas at least one bird landed in 14.8% of the trials when moving away from the water source. There was no effect of the time of day or the number of birds in a flock on zebra finch tendency to land in a trial (Model 1 in Table 2, Supplementary Tables S1 and S2). All 16 zebra finch sound files attracted zebra finches to land at some point (the different sound files attracted birds to land in 5–43% of trials in which they were used). The landing frequency in the eight locations ranged between 5 and 26% of the trials.

Table 1

Frequency of trials in which no bird or one to eight birds landed in relation to the sound treatment (zebra finch calls or white noise control trials). These data are visualized in Supplementary Figure S2.

No. birds landing	White noise	Zebra finch calls
0	218	184
1	3	22
2	2	15
3	2	3
4	3	3
5	1	1
6	0	1
7	0	0
8	1	1
total	230	230

Figure 1

Probability of at least one zebra finch landing in trials (a) with either zebra finch calls or white noise as a playback (N = 460 trials: N of zebra finch call trials = 230, N of white noise trials = 230); (b) in relation to the observed flight height of zebra finches (N = 460 trials: N of high flight = 273, N of low flight = 187). Boxes range from the first to the third quartiles; horizontal lines in the boxes show medians; diamonds show means; whiskers show 1.5x the interquartile ranges or the extremes (whichever are smaller). Box plots are based on model output for 1000 simulations for each observation with “treatment”, “flight height” and “distance to speaker” (mean-centered) as fixed terms, “session nested within location” and “sound file” as random terms (minimal model, Model 1 as in Table 2).

In those trials in which at least one zebra finch landed, the number of birds observed in the approaching group and the number of birds landing in this trial were highly positively correlated, whereas the type of sound file replayed (zebra finch calls or white noise control), as well as the time of day, had no effect on the number of birds in a group that landed (Model 2 in Table 2, Supplementary Tables S1 and S2). Either the whole group, a single individual, or a pair of birds landed (Figure 2). In 88.9% of trials, in which at least two individuals passed and landing occurred (N = 36), the group landed as a whole. The maximum group size responding observed was eight, whereas the maximum observed group size of passing zebra finches was 20.

Figure 2

Number of zebra finches responding (i.e. landing) in relation to the observed groups sizes of passing zebra finches in a trial. The sizes of the dots are proportional to the frequencies of observations (largest dot size represents N = 22 trials). Data are shown for groups in which at least one zebra finch landed Model 2 in Table 2).

Whether at least one bird landed or not during white noise trials was not influenced by the distance between the loudspeaker and bird(s) (see Model 3 in Table 2, Supplementary Table S2). In addition, no other tested variable (time of day, number of birds, flight direction, and flight height) had an effect on the likelihood that at least one bird landed in the control trials.

DISCUSSION

Here we show that zebra finches on flyways between the wider environment and a water source were attracted to playbacks of vocalizations of conspecifics, and were more likely to land near the playback site broadcasting zebra finch vocalization than when control sounds were played. Moreover, zebra finches flying low and in closer proximity to the playback loudspeaker were more likely to land than zebra finches passing by further away and at greater height. Zebra finches tended to be more likely to land on flights away from the water source towards the wider area than during inbound flights and usually the entire group responded, instead of single individuals. Landing as response to the playback was not affected by either the time of day or the group size of the flying birds.

Our results support our main prediction that zebra finches are attracted to land when hearing calls of conspecifics. Regardless of the function of the calls produced by foraging zebra finches and irrespective of whether other zebra finch vocalizations also have an attracting function, we show that they provide a source of social information and resulted in conspecific attraction. This suggests that zebra finches do not require visual cues to locate groups of conspecifics.

Our results further indicate that zebra finches collect information on the presence of conspecifics, and potentially on the location of foraging conspecifics spontaneously while traveling through the area in which they live. The closer the birds pass by, the more likely it should be to hear the broadcasted calls of conspecifics (Loning et al. 2022) and the lower the costs of interrupting a current trip. Additionally, whilst we recorded only the height of flight, this is related to the mode of flight. As characterized by Zann (1996), zebra finches have two separate modes of flight that are likely to differ in their function. Whilst both modes of flight are for locomotion, low flight is usually slower and exhibits more undulations than high flight, and is typically less direct, whilst high flight is fast and straight and is typically used when birds are moving from one point to a clearly defined destination. These differences in flight may have hence influenced the zebra finches’ decisions to land. Low flight might be less targeted and used while searching for a grouping or foraging option or while spontaneously prospecting for social information from conspecifics. Zebra finches might also be able to use social cues and might be better able to do so, when conspecifics are in close proximity. It should be equally possible for them to perceive these acoustic cues while they are on a trip to the water source, back to the nesting sites, or searching for feeding sites and use this information right away. A general attraction to conspecific presence was previously found in this zebra finch population in relation to nesting site decisions (Mariette and Griffith 2012) and could also be used for gathering social information in foraging site decisions. Especially in habitats with high variability, the presence of conspecifics can be an important cue for making decisions, as was shown for breeding habitat decisions in Baird’s sparrows, Ammodramus bairdii (Ahlering et al. 2006).

Our findings raise the interesting question of whether the calls of zebra finches that we used in our playback have a primary function within the foraging group and are simply eavesdropped by passing zebra finches opportunistically. For example, in other species of birds and bats, individuals use inadvertent acoustic social information of foraging conspecifics (Chaves-Campos 2003; Martinez et al. 2018) or vocalizations of heterospecifics with similar foraging ecology (Gu et al. 2017; Hügel et al. 2017). Alternatively, zebra finches may produce directed, deliberate calls to recruit conspecifics and to increase group sizes in order to lower individual predation risk while being exposed on ground, for instance during foraging. If calls are an active recruitment signal, one would expect them to be loud and to be expressed at a higher rate when flying conspecifics are detected and in areas with more birds passing by, such as along the flyways where we conducted our experiment. All zebra finch vocalizations, even the loudest “distance call”, however, are soft and, in combination with their hearing thresholds, are not audible for them over a very long range (Loning et al. 2022). One would also expect higher calling rates when few birds are foraging as many of the anti-predator benefits of attracting others would decrease with group size. Deliberate recruitment calls in a foraging context have been reported amongst other socially foraging species, for example in the house sparrow, Passer domesticus, when the available food source is divisible (Elgar 1986). In line with this interpretation, zebra finches represent relatively non-aggressive flock foragers and grass seeds are an easily divisible food source, which cannot be individually monopolized. At this stage, both, the use of calls elicited as inadvertent cues or production of deliberate recruitment calls to attract those birds in hearing range seem possible explanations for our findings, and they are not mutually exclusive. The relatively low amplitude of most calls in the zebra finch repertoire compared to the louder contact calls (Zann 1996), however, suggests that the calls probably primarily function for within group communication (Loning et al. 2022), although the range of attraction may likely be increased somewhat if a group of birds is calling simultaneously due to the additive properties of sound (Embleton 1996). Regardless of whether the calls are used as cue or as signal, both parties are likely to benefit, for instance in a foraging situation when sufficient food is available. Future studies are needed to shed further light into the context at which zebra finches are more likely to call to better explain the underlying mechanisms of social attraction we found.

The probability of landing depended on the flight height of the groups with low flying birds being more likely to land. Similarly, the probability to land tended to depend on whether birds were on an inbound trip to a water source or on the way back. Together these findings indicate that decisions to respond are traded off against the purpose of a trip. Birds were more likely to land when flying low and undulating. In this flight mode, the zebra finches may have actively searched for grouping or foraging opportunities, for instance by seeking cues of foraging conspecifics, as active response to an immediate need to forage. Birds tended to be more likely to land and “join conspecifics” on their trip from the water source than on the trip towards the water source, supporting our prediction that drinking may be of higher priority on inbound trips. Birds on an inbound trip towards the water source may have had an immediate requirement to drink and thus tended to be less responsive to social information during the flight and less willing to join a group and/or forage on the way. On the way back from the water, birds might have been less constrained in their time, and thus be more responsive to grouping and/or foraging opportunities. Thus, our findings show that responses to conspecific calls are context-dependent, and perhaps reflective of individual needs and trade-offs (water is often a pressing need in this hot environment; Cooper et al. 2019; Funghi et al. 2019). These trade-offs can provide one mechanism for the fission-fusion societies in zebra finches where individuals join and leave social groupings flexibly. Since each pair has to deal with their own trade-offs to optimize the resource acquisition for themselves and their brood, such flexibility will likely be adaptive and appear as an emergent property of individual and pair decisions.

Our observed group sizes of approaching zebra finches were very similar to those reported in McCowan et al. (2015). The number of birds responding matched the number of birds in the approaching group in nearly every trial where zebra finches landed, meaning that a group tested in a trial normally landed as a whole, rather than splitting into parts. This is in line with the expectations of stable (foraging) flocks (see Zann 1996), and might reflect that birds with similar interests and trade-offs join each other and synchronize on flights to water and food sources. Such synchronized strategies would allow individuals and groups to find food patches faster by joining already foraging conspecifics while also reducing predation risk (Foster and Treherne 1981; Inman and Krebs 1987; Valone 1989; Beauchamp 2013).

The hypothesis that zebra finches would land more frequently during trials in the early morning was not supported by the results. Zebra finches’ diurnal pattern in crop filling usually peaks right after sunrise and again shortly before dusk (Zann and Straw 1984). The trials started between 07:00 and 12:30 hours, which might have been already too late to capture the first foraging trip of feeding flocks (see Funghi et al. 2019 for daily foraging patterns in the population). Alternatively, the birds may have had other motivation to join group of conspecifics than for foraging.

Taken together, our findings that zebra finches respond to playback from foraging conspecifics specifically on trips back from water to the broader habitat where foraging and nesting sites were found, provides new insights into the underlying mechanisms of structuring in fission-fusion societies. We have found that on around 20% of occasions, individuals that were passing by, were diverted to investigate the acoustic cues of a conspecific social group. This provides an opportunity to form a larger aggregation, and also a potential feeding opportunity. Whilst this may not seem a high level of recruitment, the birds in the population were actively breeding during the study and many birds would have been engaged in high levels of parental care. Passing birds may have been on the way to water, or had crops full of either food or water for their nestlings, or may simply have not been interested in joining conspecific groups or exploring new foraging opportunities at the time of passing, for various other competing reasons (for example the need to incubate or brood at the nest). The recruitment level demonstrated, whilst not huge, is likely to have been biologically highly significant for signaler and receivers as it allows birds recruiting conspecifics fairly quickly to bolster numbers, for instance at a foraging site, and would quickly enable single birds or small groups of birds to find and join a larger group. Group structure in terms of membership and group size is likely to be optimized on a fine scale, through the individual decisions that are made as birds receive cues or signals of social foraging, and decide whether or not to heed them and join a group. Selective responsiveness to social information allows animals to optimize the trade-offs in such things as the investment in different types of activity, and also in the size of the group that is joined, which in turn may reduce or increase the risk of predation (Dall and Griffith 2014). Future studies could now test whether additional visual cues increase the extent of social information use and whether specific recruitment calls are responsible for triggering the joining behavior.

We thank Hugo Loning for reading an earlier version of the manuscript.

FUNDING

This work was supported by a grant from the “Deutsche Forschungsgemeinschaft” (SCHU 2927/3-1 to WS and SCG). CA received a student Hamburglobal scholarship. MN was supported by a Nederlandse Organisatie voor Wetenschappelijk Onderzoek grant ALW.OP334 during writing the manuscript.

DATA AVAILABILITY

Analyses reported in this article can be reproduced using the data provided by Adrian et al. (2022).