The study was conducted in and around the Nonggang National Nature Reserve (22°28'- 22°30'N, 106°56' -106°58'E; elevation 59-215 m a.s.l.), using the nests of the Red-whiskered Bulbul Pycnonotus jocosus. Nonggang is a tropical limestone site, where researchers have spent a decade investigating the breeding ecology of the avifauna . It is located in the southwest part of Guangxi Zhuang Autonomous Region, Southern China and has a tropical monsoonal climate . The Red-whiskered Bulbul is abundant in buffer areas of the reserve which includes degraded forest, as well as plantations of sugarcane and the Chinese Chestnut (Sterculia nobilis), and its nesting has previously been studied in this area . The species has also been studied in other tropical sites of southwest China, and been shown to attract a range of nest predators, including birds, mammals, and reptiles . Its breeding season lasts about five months, from early April to late August, and we worked on this study throughout the breeding season, during three years, 2018-2020.
Measurements at the nests and recordings
We recorded begging calls at six nests in the buffer zone during May of 2018. We checked nests every two days, to document when the eggs hatched and hence the age of the nestlings. When nestlings were five days old and parents were not present at the nest, we positioned the end of a CR:152B model Cirrus Research sound level meter (SLM) about 1 cm close to the beak of the nestlings. The nestlings responded to this stimulus by producing begging calls, as if in response to the parent [for similar behavior, see 24]. The SLM recorded 3 min of amplitude data, using its fast response setting and A weighting, with one data point collected every 1 s. The nestling calls’ fundamental frequencies were between ~ 4.5 and 6 kHz (Fig. 1).
At the same time as we recorded the amplitude of the vocalizations using the SLM, we also recorded those vocalizations using a K6 Sennheiser ME 62 omnidirectional microphone, placed 10 cm away from the nestlings, embedded in a Telinga parabola, and attached to a Marantz PMD 671 digital recorder. Recordings were made at 44,100 kHz sampling rate and in the .wav format. We then repeated this process when nestlings were 7 days old. We took care not to overly disturb the nest, and the whole recording process at a nest lasted only 10 minutes. Our manipulations did not appear to affect the success of the nests adversely, as four of six (66.7%) of the nests in which we worked fledged successfully, higher than the average success of nests in the area (nest success: 9/31 nests, 29.0%) .
In analyzing the SLM amplitude measurements, we found the difference between the amplitude when the chicks were begging (averaged across all such seconds), and the amplitude when the chicks were not begging (i.e. background noise), and calculated the amplitude of the begging . We found no significant differences in amplitudes between the 5- and 7-day recordings. Pooling these two types of recordings together, we had 12 measurements of the amplitude of begging, averaging 73.1 ± 5.9 dB at 1 cm (hereafter referred to as “mean amplitude”). The loudest exemplar was 83.9 dB at 1 cm (hereafter referred to as “maximum amplitude”).
We placed an FHD-480 model RICH video camera at three nests to document typical parental behavior. According to the video cameras, we found the frequency of parental feeding was approximately 5-6 trips/hour, with the earliest feeding about 6:00 AM, and the latest about 6:00 PM.
Noise selection and recording
We selected traffic noise as the type of noise used in this study because it is the perhaps the most pervasive kind of anthropogenic terrestrial noise. We recorded traffic noise at close range because at this distance we found the high-frequency components to be considerable, so that they provided masking towards the lower frequencies of the begging of the bulbuls (i.e., around 4.5 kHz, Fig. 1). If nestling birds perceive some interference from traffic noise and increase their begging volume, as the experiments of Leonard ML and Horn AG  suggest they might, the higher frequency components of their vocalizations (e.g., 6 kHz and above) would not be as masked as their lowest frequencies (e.g. 4.5 kHz), and this might make them more vulnerable to predators that have sensitive high-frequency hearing.
We recorded traffic noise 5 m from large highways near Nanning, the capital city of Guangxi Zhuang Autonomous Region, using the same recording equipment as above, and with the microphone pointed into the flow of traffic. We made 6 separate recordings 3 km apart. Although the amplitude of the recordings was made so that it did not change over time, there were some fluctuations in the mix of frequencies on the recordings: the amount of energy above 4.5 kHz could be as high as ~ 1/10 of the total energy as a car approached, and less than ~ 1/50 when the cars were departing or distant.
Design of playback tapes
There were two kinds of playback treatments to prepare: begging tapes and the same begging sounds mixed with traffic noise. To make the begging tapes, we took the 1 min sections of the 3 min recordings with the greatest number of begging calls. Using Raven Pro (Version 1.5, Cornell Laboratory of Ornithology) we filtered the tapes (below 0.7 kHz), and edited them to get rid of background noises, while keeping the original repetition rate of the begging calls. To make the mixed tapes, we used Audacity (version 2.4.1, Audacity Team) to mix with any begging recording an equal amplitude of traffic noise, so that at peak amplitudes (i.e. when the chick called) the total amplitude of the mixed tape consisted of half begging and half traffic noise (as judged using the dimensionless units of amplitude [kU] in Raven Pro’s waveform display, see Fig. 1). Begging was partially masked by the traffic noise, although the majority of the power of the traffic noise was at lower frequencies. The masking effect had some fluctuation according to the approach and departure of the cars, with an approximately average situation shown in Fig. 1.
Playback tapes were then made that were 12 hours long, with a 1-min section of begging calls followed by 10 min of silence repeated many times to replicate the patterns that had been recorded by the videocameras. There were a total of five playback treatments: 1) begging played at mean volume (BEG), 2) begging played at maximum volume (BEGMAX), 3) mixed (begging calls + traffic noise) played at mean volume (the same volume as treatment 1, BEGNOISE), 4) mixed played at maximum volume (the same volume as treatment 2, BEGNOISEMAX), and the silent control (SILENT).
Small speakers (AM1 Plus Abramtek speaker, China, which have an approximately flat frequency response curve between 100 and 10000 Hz) were hung 3-10 cm from the nest. Volume was controlled by adjusting the volume of the speaker so that it matched the mean amplitude (begging and noise at ~77 dB, with noise at ~75 dB, producing an amplitude of begging at ~73 dB, all assessed at 1 cm from the speaker) or the maximum amplitude (begging and noise at ~87 dB, with noise at ~84 dB, producing an amplitude of begging at ~84 dB, again at 1 cm from speaker) of the measurements of begging. Playback was started at 6 AM and played until 6 PM for seven straight days, with 5-day old nestling recordings used the first four days, and 7-day old nestling recordings used the last three days. On each day a playback exemplar (one of six) was chosen randomly, as long as it had not been used at that nest before.
In the beginning of the project (the 2018 field season), we conducted the experiment at nests that had been used that year. After the nestlings fledged or were predated, we waited one week, and then reused the nest, provided they were not damaged. Wearing plastic gloves, we first cleaned the nests of any excrement or egg shells. We then placed eggs made of plasticine in the nests. These eggs were made to mimic Red-whiskered Bulbul eggs in their size (approximately 16 mm X 22 mm) and coloring (marks were painted on the plasticine using waterproof paints). In addition to the speaker, we also placed at the nest an infrared camera (SG-660v, Shenzhen Siyuan Digital Technology Co., China) to detect predation events. This camera was always positioned slightly higher than the nest so that the eggs were visible, and was kept 0.5-1.0 m away horizontally.
A nest was considered to be predated if at least one egg was damaged or disappeared. Each nest was assigned to one treatment randomly, as long as nests with the same treatment were not within 250 m of each other. In the 2018 year, we used 20 such nests. A problem with the natural nests was that they were frequently damaged by rainfall, limiting the number of nests in which we could put the artificial eggs. In order to increase the sample size, we used hand-made artificial nests, constructed out of dried twigs and vines to be of similar size and shape to the bulbuls’ nests and painted to resemble them. In 2018, we installed 40 of these artificial nests in locations where we had observed actual nests over the previous two years (in other projects), but where these nests had been damaged. The overall predation rate was similar to that of the natural nests (see results). We therefore used artificial nests in the last two years of the study (2019 and 2020, 100 nests), again placing them at places where real nests had been located that year. The birds did not reuse the same locations year-after-year.
Predators were identified if they: a) approached within 2 m of the nest on the day on which the nest was predated , and b) by bite marks on the plasticine eggs. In general, bite marks of small mammals showed small incisions from their teeth or a thin layer of the egg gnawed off (Fig. 2). In contrast, the beaks of the large predatory birds made wide, penetrating gouges into the artificial egg. If all the eggs disappeared at once without any pictures, we suspected a snake. If a nest was predated by the same species more than once, we only count one predation event. However, in the analyses of classes of predators, we do count multiple predation events if the species of predator were different (but such predation events by multiple predators occurred for only two nests).
To understand how treatment affected nest predation overall, we used the most detailed information available, the number of days the eggs in a nest were exposed and not predated. Because there was a maximum of seven days exposed, we expressed this number as a proportion of the maximum, and specifically, the response variable was a matrix, composed of two numbers for each nest: the days the eggs were exposed without predation (Dp) combined to (1-Dp). We then constructed a generalized linear mixed (GLM) model with a binomial distribution to understand the effect of playback treatment on this matrix, using the R statistical environment (R Core Team, 2021), base code. Results from this model were overdispersed, so we re-ran the model with a quasibinomial error structure, followed by Tukey HSD multiple comparisons. We then tested three planned contrasts corresponding to the three hypotheses: 1) whether predation was greater for the playback treatments, or the control, 2) whether predation was greater for maximum amplitude or mean amplitude, and 3) whether predation was greater for playback without traffic noise, or with traffic noise.
For understanding which predator classes (birds, mammals or reptiles) predated nests, we used a simpler, frequency table-based approach, as GLM models did not converge. For each of the three hypotheses we constructed a two-by-two table summarizing the number of nests predated or not predated by the predator class (with three classes being birds, mammals and reptiles), dependent on the characteristic of the playback (control vs treatment, amplitude, inclusion of traffic noise). We tested whether these tables showed evidence of differential response based on the characteristic of the playback with Fisher Exact Tests. We consider p-values < 0.05 significant.