Figure 2. Breeding sites sampled in the municipality of Ilhéus,
state of Bahia, northeastern Brazil. Sites A (-15.0844°S, -39.0080°W)
and B (-15.0844°S, -39.0133°W) were two kilometers far from highways
with car traffic (natural environments and no human noise), while sites
C (-14.9398°S, -39.0126 °W), D (-14.9289°S, -39.0177°W) and E
(-15.0866°S, -38.9987°W) were two a ten meters close to highways with
vehicle traffic (noise pollution), the distance from locations C and D
to E was 15 kilometers.
The sampling sites were defined as Natural (no traffic noise) or
Highways (with traffic noise). For each recorded individual, we obtained
the social situation in which the male was inserted, being: with chorus
(four or more conspecific individuals in vocalization activity within a
three-meter radius – CC) or without chorus (four or less conspecific
individuals vocalizing within a radius of three meters – SC).
Males of Phyllodytes luteolus in vocalization activity were found
through visual and auditory searches between 19:00 and 24:00 hrs. The
sampling points were quite representative due to their extensive areas
where they contained many individuals of bromeliads with the presence ofP. luteolus . To avoid re-recording the same individual in
vocalization activity, we started the search for P. luteolusmales on different sides of the sampling points, thus ensuring that
pseudoreplication biases did not occur, since the individuals were not
marked. After viewing the male, the advertisement calls were recorded
using a Sennheiser ME66 directional microphone coupled to a Marantz
PMD660 hand recorder at a distance of 0.5 meters from the vocalizing
male. As P. luteolus males have a long silent interval between
calls, each male was recorded for ten minutes, in order to obtain as
many calls as possible. The recordings were obtained in WAV format, 44
kHz sampling rate and 16-bit resolution. The bioacoustic analyses were
performed with the Raven Pro 1.6 software, using the following
configuration for the spectrogram: type = hann, FFT = 512, overlap =
50%. Three to five calls were analyzed for each individual. The
acoustic parameters extracted were: call duration (s), number of notes,
number of pulses per call, pulse duration (s), rise time (s), interval
between calls (s), repetition rate (calls/min.) and dominant frequency
(Hz). The acoustic terminologies followed Köhler et al. (2017). The
sound pressure level (SPL; intensity) was obtained for three to five
advertisement calls from each individual using a Minipa MSL-1301 decibel
meter (range: 30-130dB; 125ms, fast) positioned directly in front of and
0.5 meters away from the male in vocalization activity.
Shortly after the recordings were finished, the males were captured for
measurement of the snout-vent length (SVL) with a Western caliper
(accuracy of 0.1 mm). Air temperature and humidity were recorded using a
thermo-hygrometer (TechnoLine WS 9440, 0ºC precision). The variables
were obtained for all recorded individuals. The recordings are deposited
in the Fonoteca Neotropical Jacques Vielliard (catalog number; 0058292
to 0058297).
The noise level present in the environment was measured from the calling
site of each male recorded using a Minipa MSL-1301 decibel meter (range:
30-130dB; 125ms, fast, precision 1.5 dB), positioned 1.5 m above the
ground. The decibel meter was positioned in the four cardinal directions
(north, south, east, west), and for each direction three values of sound
pressure level (SPL; intensity) were obtained, every twenty seconds we
recorded the highest value, so for each one minute we obtained records
of the three highest values, so on for each direction. Subsequently, an
average of all recorded values was extracted to represent the, represent
the environmental noise present during the vocalization activity of
males. To describe the frequency noise felt by the calling males, we
also obtained frequency values (peak and bandwidth 90% frequency - Hz)
from the recordings of individuals near the road. In the recordings of
males from natural environments, the frequency values of the recordings
were not measured because they were close to zero due to the absence of
noise.
To verify if the environments close to the highway had higher
environmental noise than natural environments, we performed a t-test
using the average SPL obtained from the calling site of each male. To
test whether anthropic (hypothesis 1) and conspecifics (hypothesis 2)
noise alter the acoustic parameters (response variables) of the
advertisement calls of Phyllodytes luteolus , we performed
multiple regressions using the type of environment (with and without
traffic noise) and situation (with and without chorus) as explanatory
variables. Since body size and air temperature can influence anurans
calls (Gambale & Bastos, 2014; Morais et al., 2012; Wells, 2007), these
two factors were used as covariates in the models. We tested the
multicollinearity of the model’s independent variables through a
variance inflation factor (VIF) test. We considered variables with VIF
values greater than 3 to be collinear. The VIF did not indicate
collinearity between the variables (environment=2,680, situation=1,063,
SVL=2,520, and temperature=1,088). We evaluated the assumptions of
normal distribution of residuals and homogeneity of variances by graphs
of quantiles and residuals as a function of adjusted values (Zuur et
al., 2009), respectively. Acoustic parameters that did not meet the
assumptions (call duration, number of pulses, and interval between
calls) were logarithmized.
To test significant differences in the set of acoustic parameters
between the environments (with and without traffic noise), a
multivariate analysis of permutational variance (PERMANOVA, Anderson,
2001) based on a Euclidean distance matrix was used. Before this
analysis, we performed a Pearson correlation between the acoustic
parameters and removed the highly correlated ones (number of notes,
number of pulses, rise time and interval between songs). To verify the
homogeneity of the variances we used the PERMDISP (Homogeneity of
Multivariate Dispersion) (Anderson, 2004). For all tests we assigned a
significance level lower than 5%. The analyses were performed using the
”vegan”, ”lme4” and ”car” packages in the R software version 4.1.1 (R
Core Team, 2021).