Effect of natural abiotic soil vibrations, rainfall and wind on anuran calling behavior: a test with captive-bred midwife toads (Alytes obstetricans)

Location and study species

The playback experiments were completed between in June and July 2020 in the Amphibian Recovery Center of the Sierra de Guadarrama National Park, where a captive population of midwife toads (Alytes obstetricans, Alytidae) is established for the reintroduction of the toads in their natural habitats (Martín-Beyer et al. 2011). This anuran species is widespread across Europe but has suffered substantial declines throughout its range and population size, especially due to chytridiomycosis and ranavirosis (e.g., Bosch et al. 2001, 2021). The genus is known for its remarkable reproductive behavior, with males carrying the eggs entwined around their hind legs on land for about a month, from fertilization to hatching. Males typically call in open spaces, while tending occurs mainly underground. During this period the eggs are kept under adequate temperature and moisture conditions, away from predation or infection by fungi, and finally, the mature egg masses are released at the shore of a body of water (Márquez and Verrell 1991; Márquez 1995; Márquez and Bosch 1995).

We tested a total of 15 male radio-tagged individuals that were temporarily housed in the outdoor enclosure of the Center. Animals are housed indoors in winter to avoid low temperatures and to allow them to breed year-round and are kept in outdoor enclosures from May to September. After each complete experiment, the tested animal was identified and returned to its indoor terrarium. This population was selected for testing, because travel between regions was forbidden in Spain during the 2020 field season due to the COVID-19 pandemic and the captive population was within the limits of the region of Madrid. We verified that every animal was healthy by observation when measuring its length and weight after the test. We ensured that every animal had the minimum stress possible, minimizing handling and transportation times. Every individual tested was left 24 h in the open enclosure before testing. Experiments started after sunset and the emission of stimuli started when the focal male seemed at the peak of its vocal activity.

Playback stimuli

For the experiments, we constructed a set of 9 playback crescendo stimuli trains: rainfall, wind in five different intensities (5 km/h, 6 km/h, 7 km/h, 8 km/h, 9 km/h), a pure tone of 200 Hz, a pure tone of 300 Hz, and a silence stimulus (control). All stimuli lasted 2 min and were modified by applying a linear ‘fade-in’ amplitude filter from 0% to 100%, using Audacity 2.0.2 (SourceForge, Carnegie Mellon University, Pittsburgh, PA, USA) Before applying the ‘fade-in’ amplitude filter, the stimuli were peak normalized to 100% with some exceptions: the amplitude of the 5 wind stimuli was scaled, so that the highest wind level (9 km/h) was peak normalized at 100% but the rest kept their relatively lower amplitudes. Therefore, the highest peaks of the 4 lower wind stimuli did not reach 100%. The advantage of using fade-in stimuli is that only the maximum intensity of each stimulus had to be adjusted to the conditions when re-cording in the field. This was adjusted using exactly the same equipment and recording level in the playbacks as was used in the recording of the wind playback and ensuring recording level did not saturate. As for rainfall the maximum level was adjusted to the maximum level of the higher wind speed.

Rainfall vibrations were recorded in Puente Ajolí, (5 m.a.s.l. El Rocío, Doñana, Huelva 37º08′01862 N 6º27′44.98″W) in 2012 with an Oyo ONE geophone vertically protected by a foam-covered structure to prevent the direct impact of the drops on the geophone. The stimulus was the same that was used in Márquez et al. (2016). To obtain the spectral components of the rainfall vibration, recordings, FFTs (1,024 points, sampling rate 48 kHz, 61.9 Hz bandwidth) were made using Audacity 2.0.2 and Raven Pro 2.5 (Lab of Ornithology, Cornell University, Ithaca, NY, USA). The recordings obtained had slightly irregular flat spectra toward the lower end of the spectrum. Based on this acoustic signature, a 2-min synthetic vibration stimulus was generated with Audacity 2.0.2 by low-pass filtering broadband noise (100% amplitude).

Wind vibrations were recorded in Puerto de Morcuera (1700 masl 40º50′32.76″N, 3º50′10.912 W, Madrid) in July 2020 in a mountain grassland habitat similar to the habitat of the extant natural population of midwife toads in Peñalara, and in the near vicinity. Vibrations were recorded with an Oyo ONE geophone protected (vertically and laterally) by a foam-covered structure to prevent the direct impact of the wind on the geophone, while the wind speed was monitored with a Bresser (model 7002510) portable weather station.

Both for rainfall and wind vibrations, the geophone was connected to a custom-made amplifier which in turn was connected to the input of a Sound Devices 744 T recorder.

The pure tones of 200 and 300 Hz were included to test if there was a detection bias in the range considered. 200 Hz was an approximate central frequency for the wind stimuli and 300 Hz was an approximate central frequency for the rainfall stimuli. Synthetic stimuli were synthesized with Audacity software, and natural stimuli were edited and played back with the same software too. An online randomizer (https://www.random.org/lists/) was used to generate the random sequence of the stimuli. A single rainfall track was used and a single wind track was used for each of the five wind intensity levels.

Experimental design

Experiments were carried out for 15 days during male calling activity, starting from 10 p.m., from the 8th of June to the 20th of July 2020. Alytes obstetricans males were calling in an open-air enclosure area with a small accumulation of bricks on a corner, which the toads used as a refuge and from which they called. Toads were individually marked with radio tags. Each toad was only tested once and was returned to its indoor terrarium after reading the tag after the test. One experiment was completed per night. One to three toads were housed in the enclosure but tests were started when only one individual was calling. Occasionally a second non-focal toad would start calling during the test. This was noted and considered in the statistical analyses. Playback vibrations were generated with a portable computer connected to a Kenwood KAC-5205 amplifier (with all filters switched off) and a “Clark Synthesis TST429 Platinum tactile-sound transducer” buried 10 cm below ground 1.5 m away from the bricks. To monitor the intensities of vibrations reaching the toads we placed an Oyo ONE geophone about 10 cm from the refuge that housed the animals during the experiment. The geophone was connected to a custom-made amplifier with a fixed gain of 30db. To record the intensities and number of calls of the toads we used a microphone (Audio Technica AT3032) placed approximately 2 m away from the calling male. The microphone and geophone were connected to a Sound Devices 744 T recorder with 4 channels. Playback was not equalized. The spectral components of the recorded signals by the geophone were only visually inspected at the time of playback and were compared to the emitted natural stimuli to verify that they were no major spectral degradation effects. The emitted stimuli were well within the ranges of the emission equipment (transducer and amplifier). The experimenters were located 4.5 m away from the enclosure, remaining motionless during the playback (Fig. 1).

Fig. 1figure 1

Experimental setup in the field. The enclosure was 4.8 × 2.5 m wide. The microphone was placed 2 m away from the tested individual and the geophone was placed 10 cm away from it. The transducer was placed at a variable distance, but always more than 1.5 m away from the focal calling individual. Observers were located 4,5 m away from the enclosure

Test Protocol

Our test was based on the observation of a single calling toad before and during the emission of a crescendo stimulus train, a random sequence of stimuli separated by silence with each stimulus being linearly increasing in amplitude (“crescendo”, or Fade-in 0–100), similar to the technique used in Caorsi et al. (2019), Fig. 2. We emitted a playback train with 9 different stimuli to every individual (a total of 38 min). Every experiment started at 10 pm (sunset time, approximately). Each stimulus was 2 min long, with 2 min of silence before (pre-stimulus) and after (post-stimulus). The stimuli were presented in random order being different for each toad. Before each trial, we measured the temperature and the relative humidity.

Fig. 2figure 2

Playback scheme (crescendo stimulus train) showing the 9 playback stimuli (total 38 min) presented to each animal in random order. Triangles indicate the increase of the amplitude of the vibration emission within the 2-min treatment

Data extraction

For every stimulus, pre-stimulus and post-stimulus we measured the following acoustic characteristics and measurements: total number of calls, mean dominant frequency of the last 5 calls, and mean fundamental frequency of the last 5 calls. In addition, we determined for each experiment: the total number of specimens in the enclosure, the total number of specimens calling in the enclosure, background noise, time before hearing the first specimen singing, duration of a call of a specimen before turning silent during the playback of a stimulus, duration of the 1° call group defined as a regular calling, duration of the 2° call group defined as a regular calling, duration of the 3° call group defined as a regular calling and mean duration of all regular call groups. We define regular calling an interrupted call group using Koehler et al. (2017) terminology. These variables represent the duration of the calling bouts of the calls during playback tests. They were recorded to spot toads with low calling activity in case they should not be included in the analyses but we did not exclude any test based on these parameters. The analysis was carried out with the software Raven Pro on an Acer Nitro 5 notebook. The FFT window size used was 2048, passband filter 0.5–10 kHz.

Statistical analysis

To test the effect of vibration stimuli on the calling activity of the focal individuals, we used a general linear mixed-effects model (LMM, Baayen 2008), as the experiment followed a repeated measured design within individuals. Thus, we included the type of playback stimulus (9 levels) as fixed factor, and individual (15 levels) as random factor. Since environmental temperature and the presence of other calling males may influence the calling activity of the focal males, we included two additional factors as covariates to account for these effects, namely, air temperature (measured every time at the beginning of any experiment) and the maximum number of non-focal males exhibiting calling activity during the playback test. The air temperature was previously z-transformed to be centred and scaled. The response variable of the LMM was estimated as the difference in call rate (number of calls per minute) between the 2-min experimental period (stimulus) and an average of five 2-min silence periods (pre-stimulus and post-stimulus of a given treatment and the entire silence stimulus of a given individual). This variable represents the behavioral response of the focal males to the vibration stimuli in comparison with their behavior during periods without stimuli. The response variable follows a symmetric distribution and hence we fitted a LMM with Gaussian error structure and identity link using the function lmer of the R package lme4 (Bates et al. 2014). The interaction terms and random slopes were not included to reduce model complexity. To test the effect of the stimulus order regardless the kind of vibration, we fitted a similar LMM but replacing type of stimulus by the position of every stimulus in each experimental sequence as fixed effect. The model structure and fitting procedure remained identical. Visual diagnostics (Q–Q plots, residuals plotted against fitted values, etc.) revealed no obvious deviations from the canonical assumptions of linearity, normally distributed and homogenous residuals, and the absence of influential observations in both LMMs. Variance inflation factor (Field 2005) was applied to a standard linear model (excluding the random effect) using the function vif of the R package car (Fox and Weisberg 2011) and indicated no collinearity (VIF < 1.29). The LMMs were fitted using the Maximum Likelihood (rather than Restricted Maximum Likelihood; Bolker et al. 2008) and model inference established by full-null model comparisons (null model comprising without the fixed effect) with a likelihood ratio test using the R function anova (test = “Chisq”; Dobson 2002; Forstmeier and Schielzeth 2011). The effect of covariates was based on additional likelihood ratio tests, comparing the full with respective reduced models using the R function drop1 (Barr 2013). Confidence intervals for model estimates and fitted values were derived using the function bootMer of the R package lme4, using 1000 parametric bootstraps and bootstrapping over the random effects. These intervals were used for post-hoc pairwise comparisons. Finally, we calculated conditional and marginal coefficients of determination with the function r.squaredGLMM of the R package MuMIn (Barton 2022). Significance level was set at 5%. All statistical analysis and figures were performed using R v 3.6.3 (R Core Team 2020).

留言 (0)

沒有登入
gif