Recordings were conducted between March 2018 and April 2019 on six islands in French Polynesia (Fig. 2). Three of the islands were atolls from the Tuamotu Archipelago, Rangiroa, Raroia and Tikehau (sampled between March and November 2018), while the other three were high islands, Bora Bora and Moorea (both in the Society Archipelago, sampled in September 2018), and Mangareva (Gambier Archipelago, sampled in April 2019). The sampling activities were carried out as part of the Under The Pole III expedition (Concarneau, France; https://underthepole.org/). Considering the known faunal shift occurring between 60 and 90 m (Pinheiro et al. 2016, Pyle et al. 2016, Baldwin et al. 2018), two different depths were selected for sampling: one above and one below this depth range. Due to logistical constraints, the chosen depths were 60 and 120 m (except for Mangareva, where data are only available for 60 m depth).

Recordings were conducted using SNAP recorders (Loggerhead Instruments, Sarasota, FL, USA) connected to HTI-96 hydrophones (sensitivity ranging from −170.5 to −169.7 dB re 1 V, flat frequency response from 2 Hz to 30 kHz, sampling rate of 44.1 kHz, gain of 2.05, 16-bit resolution). The recording schedule was set to capture 1 min recordings every 10 min for a total duration of 62 h (2 days and 3 nights). The diel cycle was divided into four periods: the sunset period (between 5 and 6:59 p.m., n=3), night (7 p.m. to 4:59 a.m., n=3), sunrise period (5 to 6:59 a.m., n=3), and daytime (7 a.m. to 4:59 p.m., n=2).

The analysis focused on frequencies below 2 kHz (Raick et al. 2021a, 2023a, 2023b), because most fish sounds are commonly found within this frequency range (Parmentier et al. 2017, 2019, Raick et al. 2020), including CS1 and US1 (Bertucci et al. 2020, Raick et al. 2023a). The files were subsampled at 4 kHz using a MatLab® R2014b routine (MathWorks, Natick, MA, USA). Spectrograms were visualized using Raven Lite 2.0.3 (Cornell Lab of Ornithology, Ithaca, NY, USA; FFT=256). All files were audited and visually inspected to identify US1 and CS1 sounds. The number of sounds per file was then calculated.

Temporal series of the total number of sounds were used to visualize diel cycles. Differences between depths and islands were visualized using boxplots. Given that the number of replicates was lower for the day than in other temporal periods (n=2 vs. n=3), boxplots were created summing data from the night, sunset and sunrise periods (i.e., between 5 p.m. and 7 a.m.), in addition to boxplots using raw data. The results were similar for both methods. Only the first method is presented in the results. Normality was assessed using Shapiro-Wilk tests, and homogeneity of variances was assessed using the F-test of equality of variances (n_groups=2, to compare the number of sounds at 60 and 120 m) or Bartlett tests (n_groups>2, to compare the number of sounds between islands). Because neither of the two conditions was met, non-parametric tests were employed. The number of sounds at 60 vs. 120 m was compared using Mann-Whitney tests. To compare the number of sounds between islands, Kruskal-Wallis tests followed by Dunn post hoc tests were conducted for each depth separately. P-values were corrected using the Benjamini-Hochberg method. All statistical analyses were performed using R 4.0.5 (R Core Team, 2021) with α=0.05.

A total of 4318 sounds were detected in the audio files: 2318 CS1 and 2000 US1. The majority of US1 (91.65%) exhibited a classical pattern (Fig. 3), while others displayed variations of the classic pattern: 0.1% had a higher frequency distribution, and 8.25% were followed by one or several pulses (Fig. 3). These pulse series consisted, on average, of 4.5±0.85 pulses (mean±SD, measured in ten sounds) with a pulse period of 127±35 ms. The peak frequency of these pulses varied, but it always corresponded to the fundamental frequency (94±22 Hz) or the first harmonic frequency (215±7 Hz) when measured at the end of the classical pattern. The occasional presence of pulses after the main part of the sound was reported by Raick et al. (2023a).

Of the CS1 sounds, 99.6% exhibited a classical pattern, 0.09% were followed by one or several pulse(s), 0.09% by one or several downsweeps (i.e. sounds with a frequency that decreases over time) and 0.04% by a fast pulse train. All the elements following the classical CS1 patterns had a harmonic structure with the same harmonic interval as the one measured at the end of the classical pattern. Additionally, the peak power (dB re one dimensionless sample unit) was equivalent for those elements. Furthermore, 0.13% of CS1 sounds had only one inflexion point and a longer first part of the sound (Fig. 3).

The number of sounds exhibited a pronounced diel pattern (Table 1, Figs 4 and 5). Both CS1 and US1 were predominantly present at sunset and during the night but were less common (sometimes even absent) during the day and sunrise periods (Table 1). At sunset, the number of sounds increased by a factor of 7 to 17 compared with daytime. CS1 sounds were two to three times more abundant in the first part of the night (7 to 11:59 p.m.) than in the second part of the night (12 to 4:59 a.m.) regardless of the depth. An increase was also observed for US1, but only at 120 m. In fact, the recorded sound counts were not homogenous across depths. Considering the entire diel cycle, US1 sounds were more abundant by a factor of 28 at 120 m than at 60 m. This difference was less significant for CS1: the number of sounds was higher by a factor of 1.3 at 120 m than at 60 m.

Although general diel patterns existed, there was a variability in the number of sounds recorded on the different islands (Fig. 6). At 120 m, for the entire diel cycle, the number of US1 varied between 44.8±13.1 per hour (at Tikehau) and 134.5±23.7 per hour (at Raroia) (Fig. 6A). At 60 m, for the entire diel cycle, Raroia was also the island with the highest number of recorded US1 (8.9±7.6 per h). There were significantly more US1 at Raroia than at all the other islands, both at 60 m (Kruskal-Wallis: χ²=60.35, df=4, P<0.0001; Dunn: Z=[−6.38, −5.17, −6.38, −2.94, and 4.92]; P=[< 0.0001,<0.0001,<0.0001, 0.007, and<0.0001]) and at 120 m (Kruskal-Wallis: χ²=44.50, df=4, P<0.0001; Dunn: Z=[−2.57, −3.89, −4.63, and 6.28]; P=[0.02, 0.00033,<0.0001, and<0.0001]). No US1 were recorded at 60 m for Bora Bora and Moorea.

The number of CS1 also varied between islands, with Rangiroa and Tikehau showing the highest number of CS1 at 120 m (Fig. 6B) (Kruskal-Wallis: χ²=38.41, df=4, P<0.0001; Dunn’s multiple comparisons test: Z=−3.9, −4.3, and 2.8 for Rangiroa, Z=−4.3, −4.7, and −3.2 for Tikehau; all P ≤ 0.009). At 60 m, there were statistically significant differences in the number of CS1 between islands (Kruskal-Wallis: χ²=281.9, df=5, P<0.0001), ranging from 0 (at Mangareva) to over 90 per hour. Moorea and Tikehau had more sounds than the other four islands (Dunn’s multiple comparisons tests: Z=−13.2, −10.1, and 7.5 and 11.4 for Moorea; Z=−8.1, −11.2, and −5.5 and −9.4 for Tikehau, all P<0.0001). Moorea was the only island with a higher number of CS1 at 60 m (93.3±12.3 per hour) than at 120 m (27.1±7.0 per h) (Mann-Whitney U test: W=54256, P<0.0001), while all other islands showed the opposite pattern (W=74234, 77450, 78328 and 68256, P=0.0024,<0.0001,<0.0001 and 0.69, with a non-significative result obtained for Tikehau).

This study focused on investigating the diel cycle of two sound types, US1 and CS1, which were previously described in MCEs in French Polynesia. These sounds were known to occur between 5:00 and 6:59 p.m. on various coral reefs (Raick et al. 2023a). Additionally, in Moorea, CS1 sounds were documented to take place between 7 and 9 p.m. on the photic reef at a depth of 20 m (Bertucci et al. 2020). The findings of our study reveal that US1 and CS1 sounds are predominantly produced during the night and sunset hours, rather than during sunrise and daytime (with US1 occurring 26 times more and CS1 50 times more during these periods). This observation suggests that the species emitting these sounds are likely nocturnal. Given the complex nature of the sounds, it was expected to record them primarily at night. Diurnal species typically use sounds to complement visual signals, whereas nocturnal species generally lack visual signals (Ruppé et al. 2015). Both CS1 and US1 exhibit harmonic characteristics, a trait found in several families of teleosts (Mélotte et al. 2019, Raick et al. 2020, 2021b, 2023c), and are FM, similar to the sounds produced by Batrachoididae (Tower 1908, Bass et al. 1999, Rice and Bass 2009), a family not known to occur in French Polynesia (Siu et al. 2017). FM sounds are less commonly observed in fish than pulse series. Some are known to be associated with species from various families, such as Batrachoididae (Tower 1908), Gobiidae (Lugli et al. 1997) and Serranidae (Lobel 1992). However, these specific families are not known to produce sounds resembling CS1 and US1. Understanding the diel cycle of US1 and CS1 should aid in identifying these nocturnal sound-producing species. Based on the results, to identify the vocal species, investigations should primarily be conducted during sunset or shortly after at a depth of 120 m for US1 and during the early part of the night at both 60 m and 120 m for CS1.

This study confirmed the depth preference of the species emitting the studied complex and FM sounds, as the number of US1 and CS1 was higher at 120 m than at 60 m. The effect of depth was more pronounced for US1, with a 30-fold difference between 120 m and 60 m. This finding aligns with the study conducted by Raick et al. (2023a), which indicated that the US and CS categories (to which US1 and CS1 belong) were 10 and 2 times more abundant, respectively, at 120 m than at 20 m (specifically studied between 5 and 6:59 p.m.). Therefore, US1 and CS1 are characteristic of the deeper part of MCEs. MCEs have distinct acoustic sources that are unique to them, as previously hypothesized by Bertucci et al. (2017), and are not merely propagated from the photic reef.

The geographic distribution of recorded US1 and CS1 sounds was not uniform. There were extreme situations, such as the absence of US1 sounds in the recordings at 60 m in Bora Bora and Moorea. Moorea also stood out in terms of CS1 sounds: they were more abundant at 60 m than at 120 m. However, when comparing between islands, it is important to consider that the recordings were not made during the same time period for technical reasons, such as transporting a boat with rebreathers from one island to another, given the size of French Polynesia (5000000 km²) (Rougerie et al. 1997).

If we focus on the sites at 120 m, fewer CS1 sounds were recorded at Moorea and Bora Bora than at the other three islands (Rangiroa, Raroia and Tikehau; there was no data available at 120 m for Mangareva). Two factors could explain this difference: geographical variations (differences between archipelagos) or geomorphological differences (different types of islands: atolls vs. high islands). High islands are known to have a more variable bathymetric profile due to their younger age (Pichon 2019). However, it is not possible to distinguish between these two factors because Moorea and Bora Bora are the only studied islands belonging to the Society Archipelago, while all the other islands studied at 120 m belong to the Tuamotu Archipelago, and both Moorea and Bora Bora are high islands, whereas all the islands in the Tuamotu Archipelago are atolls.

The analysis of the boxplots also suggests that, for some islands, US1 may be more abundant where CS1 are less abundant (e.g. in Bora Bora) or vice versa (e.g. in Tikehau). However, this hypothesis requires further substantial observations to be confirmed in order to better understand the acoustic niche of each sound type.