Our research was carried out at the Shezaf Nature Reserve, a 40 km² area in the Arava region of Israel’s Negev Desert, characterized as a hyper-arid desert with an average annual rainfall of around 30mm (Anava et al., 2001; Keynan and Yosef, 2010). The Arabian babbler study population that this research is based on was established by Professors Amotz & Avishag Zahavi in 1971, and the population has been monitored continuously since then. The population currently contains approx. 120 individuals from 21 groups, and group size varies between 3–18 individuals. Each individual in the population is ringed with a unique combination of color rings for individual identification. The study population is habituated to the presence of human observers, allowing behavioral observations and data collection from close range (within 2–3m) without causing any perceivable behavioral change or stress (Zahavi and Zahavi, 1997). Typically, a group of Arabian babblers features one dominant breeding pair. The breeding season extends from February to July, with a single nest incubated at a time. All adult group members contribute to feeding the young and defending the nest against intruders and predators (Ridley, 2007). The incubation period lasts 14–15 days, and nestlings fledge after 14–16 days. However, the young remain reliant on adult feeding for an additional six weeks. By 12 months, juveniles reach adulthood, distinguishable by sexual dimorphism in beak shape and eye color dichromatism (Ostreiher, 1999; Anava et al., 2001; Ridley, 2007).

From January to September 2023, we studied the snake mobbing behavior of ten Arabian babbler groups, with each group comprising of five to ten individuals, totaling 59 individuals. During this period, some individuals dispersed or established new groups, particularly between February and April. Six of these groups successfully bred, producing 16 fledglings. Each group was recorded several times, averaging 6.2 ± 2.7 sessions. The free-roaming nature of the birds posed challenges in locating specific groups on certain days. Upon entering a group territory, we ensured that babblers were in visual or audible proximity before setting up a dummy snake model, either an Egyptian horned viper (Cerastes cerastes) or a painted saw-scaled viper (Echis coloratus) partially embedded in the sand 20 meters from the group. We randomly switched between the two viper models per group to avoid habituation to the snake model. In addition to the acoustic camera (see below), we positioned a high-resolution RGB (red, green, and blue wavelength) video camera approximately 3m from the snake model to enable clear identification of individual birds. Before attracting the babblers to the model with mealworms (one, up to three), we activated the acoustic camera ( Figure 1 ). The wide angle of the SoundCam 2.0 required the observer to maintain a 3-meter distance from the babblers, a distance measured prior to recording and typical for observing these birds (Ostreiher, 2003). Recordings were conducted in 1-minute segments with 20 second pauses for data storage, and triggering was manual. Intense mobbing typically occurred in the initial three to five minutes, likely diminishing as the stationary snake model failed to maintain the babblers’ interest. Recording sessions were concluded when the group either moved beyond 3m from the dummy or stopped vocalizing for one minute.

The Soundcam 2.0 acoustic camera was used to locate sounds of Arabian babblers during snake mobbing. The acoustic camera, comprising 64 microphones, each sampling a wide aperture angle 70° (FoV horizontal) at a rate of 200kHz, stored with a 24-bit sound resolution, allowing a recording range of 20 meters for sounds as weak as 33dB and with beamforming-methods from 800 Hz; below this threshold, the resolution is decreased (SoundCam 2.0, CAE Software and Systems GmbH, Guetersloh, Germany). Arabian babbler mobbing calls cover frequencies from 2000Hz up to 8000Hz. Acoustic beam size and frequencies visualized in the black and white video were manually adjusted in the field in consideration of the soundscape of the surrounding during the recording. Beam size on screen was manually changed by regulating the dynamic range. Therefore, localization of filtered frequencies in the video were visualized even when masking sounds from wind, or other acoustic disturbances were present in the surrounding. While the video setting was adjusted for optimal call visualization, the full acoustic data was saved in a TDMS file, ensuring that no information was lost due to filtering. For the acoustic camera to effectively address our questions, we had to ensure following conditions: (1) The mobbing was set in a predictable location to synchronize and focus both, the acoustic camera, and an additional high-resolution video camera. The latter was used to ease individual recognition. (2) The wide angle of the acoustic camera, combined with the close proximity of the individuals during mobbing, required maintaining a three-meter distance, to prevent compromised video resolution. We tested the spatial resolution of the system by placing two speakers playing back Arabian babbler mobbing sequences of one group, with calls naturally alternating and overlapping in time. We placed the speakers at different distances from each other along a horizontal line and recorded this scenario according to the same guideline used in the experiment. Measured distances between speakers were: 5cm, 10cm and 15cm and we analyzed 50 calls per distance category. Call loudness was not normalized to copy natural conditions. We then manually annotated the broadcasting speaker by observing the speaker (as done with the babblers) and we validated this using the known ground truth.

We synchronized the acoustic camera (25fps) and video camera recordings (25fps) using adobe premium pro (Version 23.6.4), which also enabled us to place individual markers for each vocalizing babbler. The ring colors were distinguishable in the video’s RGB recordings, while the acoustic camera provided information on which individual was vocalizing and the timing of their calls (±40ms per frame) which were identified by the experimenter ( Figure 2 , upper). Due to the high resolution of RGB recordings, beak opening during vocalizations was occasionally used to reassure caller identity. The wide-angle setting of the acoustic camera also allowed us to capture vocalizations from individuals perched in nearby trees who later joined the mobbing circle ( Figure 2 , lower). We used Raven Pro 1.6.5 (K. Lisa Yang Center for Conservation Bioacoustics, 2024), to visualize snake mobbing vocalizations of 1–3-month-old juveniles joining the mobbing for the first time.

To explore the vocal dynamics of snake mobbing in Arabian babblers and to examine if social–vocal network parameters differ across age categories, we analyzed each snake mobbing event by creating a directional network, yielding a total of 62 networks from the 10 groups (6.2 ± 2.7 recordings per group), excluding defective recordings or trials with only one or two individuals mobbing. To reconstruct the networks each edge i–j in the network was strengthened when individual i vocalized after individual j. We set a minimum threshold of 120ms for the interval between the calls of i–j because shorter intervals are unlikely to be a response due to the minimum reaction time. Therefore, if a respondent i called more than 120ms after the individual j, the preceding individual was designated as the source. This 120ms threshold was selected based on the auditory reaction time found for other species and the focal one (Thorpe, 1963). Thus, if individual i called only 40ms after j, it is unlikely that it was responding to j. Pause duration between calls of different individuals was analyzed by ignoring the minimal threshold of 120ms, but setting a maximal threshold at 1600ms to portrait behavior during mobbing and not when individuals start to join. 85.9% of the data lay between the pause durations of 0–160ms.

Networks were generated using the “igraph” package in R-studio 2022.12.0 (Csardi and Nepusz, 2006; Posit team, 2022; R Core Team, 2022). We focused on network density to assess the participation of group members during vocal turn taking for the whole mobbing sequence. The network density measures the ratio between the actual number of connections in a network and the maximum possible number of connections. It is therefore a measurement of the proportions of pairs that call after each other. Density was additionally examined as a function of group size.

To test differences of individual and group vocal turn-taking based on age category during snake mobbing, we used generalized linear mixed models (GLMMs) and a linear mixed model (LMM). We chose the three response variables: “Inter-individual pause duration[ms]”, “Graph density as function of youngest juvenile in the group”, and “Number of calls as function of youngest juvenile in the group”. For each model, “group identity” was set as random factor and for the model with group density as response variable, number of individuals was added as a fixed factor. The generalized linear mixed models (GLMMs) and the linear mixed model (LMM) were calculated using the R-packages “lme4” and “arm” in R-4.2.2 (Bates et al., 2015; Gelman and Su, 2020). For the LMM, we assumed a Gaussian-, and for all GLMMs a negative binomial error distribution. The model fit was assessed through visual inspection of the residuals and by applying the R-package “DHARMa” to assure model fit (Hartig, 2022).

We estimated correct positions of calls when speakers were 5cm apart with 76% accuracy, for speakers being 10cm apart with 72%accuracy and for speakers being 15cm apart with 80% accuracy. All but two errors occurred in cases when the calls played by the two speakers overlapped in time, especially when one call was louder than the other ( Figure 3 ). We thus examined our accuracy specifically in overlapping calls and found that at 5cm we accurately identified the caller 44%, while it was 50% for 10cm and 40% for 15cm ( Figure 3 ). In reality, mobbing babblers typically stand at least 10cm apart and rarely call simultaneously. Moreover, we know based on the spectrogram when calls are overlapping and could pay special attention to these cases. Therefore, these results demonstrate the validity of the acoustic camera for studying the phenomenon.

Juveniles younger than a month did not engage in snake mobbing. At this stage, they primarily follow caretakers and beg for food from the adults involved in mobbing, while ignoring the presence of the snake. The youngest mobbing individual that we observed was a 1–2-month-old, un-ringed individual. Although unable to produce a proper mobbing call, this juvenile exhibited attempts to mob the snake-model, showing vigilance and emitting shortened begging calls ( Figure 4A ). The youngest of the nine precisely ageable juveniles to participate in mobbing and emit accurate mobbing calls, was 48 ± 3 days old.

Analyzing the mobbing sequences revealed that Arabian babblers take turns vocalizing with preferred pause durations between consecutive calls. The most common pause is around 175ms long, which is in-line with the reaction time discussed above, and with additional peaks appearing at approximately integer multiplications of 175 ± 25ms ( Figure 4B ).

In order to explore the importance of learning the temporal coordination of snake mobbing, we examined how babblers of different ages time their pauses ( Figure 5A ). When examining pauses between calls emitted by individuals of different age groups, it can be seen that very young juveniles aged 1.5–3 months vocalize with significant shorter pauses when responding to each other compared to when responding to older individual such as adults (β=0.42, SE=0.12, z(280)=3.89, p<0.001, GLMM, Table 1 , Figure 5B ), or juveniles aged 4–6-month (β=0.45, SE=0.15, z(280)=3.07, p<0.01, GLMM, Table 1 , Figure 5B ). These shorter pauses may diminish the opportunity for other individuals to join the mobbing vocally.

To investigate vocal turn-taking of mobbing babblers, we tested if the social–vocal network density and call rate differ across groups defined by the youngest participant. We found that groups containing very young juveniles aged 1.5–3 months exhibited reduced density compared to groups composed solely of adults (β=0.2, SE=0.06, z(62)=3.23, p<0.001, LMM, Figure 6A , Table 2 ). Groups with intermediate and older juveniles (4–6-month-old) exhibited intermediate network densities which did not differ statistically from each other, or from groups with 1.5–3-month-old juveniles (β<0.2, p>0.05, LMM, Figure 6A , Table 2 ). Therefore, the densities of networks follow an age gradient from low densities of groups with young babblers towards high densities in adult groups. Graph density was not correlated with group size (β=−0.02, SE=0.015, z(62)=−1.52, p>0.05, LMM, Table 2 ). The differences in network density could also not be explained by calling propensity. Groups composed of solely adults, actually called significantly less than groups with juveniles of any age category (β=−0.51, SE=0.24, z(62)=−2.17, p<0.001, GLMM, Figure 6B , Table 2 ). The mobbing events containing solely adults therefore show a higher vocal connectivity, even though the call rate of adult groups is lower than in groups with juveniles ( Figure 6C ).

Arabian babbler mobbing–vocal networks generally exhibit symmetry between in-degree and out-degree edges (n=62 networks) ( Figure 7B ). This symmetry arises from alternating calls between two or more individuals. To illustrate the temporal dynamics of these networks, we visualized the social vocal networks of the same group during two different snake mobbing events, each lasting three minutes ( Figure 7 ). The group consisted of 4 adults, one older juvenile (J1 ~8 months old) and two younger juveniles (J2–3 ~1.7 months old). These events were nine days apart and the differences were pronounced. In the first event only the older juvenile (J1) participated in mobbing, vocalizing after the subordinate female F2. However, in the subsequent event (9 days later) the younger siblings (J2–3) who previously only begged, now actively joined in, mostly calling in alternations with the dominant male (M1), the second ranked male (M2) and the second ranked female (F2) ( Figure 7 ). The density of networks decreases from 0.7 to 0.57, after juveniles join the mobbing.

The acoustic camera can be useful for studying various animal systems where the localization of multiple vocalizing animals is necessary. Our setup can be used for recording the role of vocalizations during take-off and group flight in birds like flocking Tristram’s starlings (Onychognathus tristramii) and monk parakeets (Myiopsitta monachus, Figures 8A, B ). The operational constraints of the acoustic camera, such as time and proximity, can be mitigated by connecting the SoundCam 2.0 to a laptop and mounting it on a tripod. When connected to a laptop, longer video and high-quality sound recordings can be performed, and the operator can maintain a greater distance from the camera to minimize disturbance. This setup is for example suitable to record animals in roosting, feeding, or courting sites such as our recordings of the Greater mascarene flying fox (Pteropus niger) in Mauritius ( Figure 8C ), or congregating Middle east tree frogs (Hyla savignyi) courting during the night ( Figure 8D ). The maximal distance to record animals is ~20 meters (for most species), which can limit the application of the camera especially in regard to flying wild birds.