Acoustic data collection was conducted in the North-Western coast of Sardinia (Mediterranean Sea; Figure 1 ) at two sites: Tavolara Island and Santa Teresa Gallura (hereafter TA and ST). TA is located within the Marine Protected Area (MPA) of Tavolara Punta Coda Cavallo, established in 1997 and covering a total area of 15,280 hectares. The acoustic recorder was deployed at 35 m of depth, in the B zone (partial protection), where artisanal fishing is permitted while recreational fishing is prohibited. ST is within the Capo Testa Punta Falcone MPA, established in 2018 and spanning 5,000 hectares. Here, the acoustic recorder was placed within the C zone of the MPA, on a coralligenous cliff at a depth of 34 m, where both professional and recreational fishing are permitted. The site is also within the access corridor to the port, where navigation by ferry boats is allowed.

At both sites, the coralligenous habitat is primarily characterised by coralline algae and to a lesser extent by green algae (e.g., Halimeda tuna and Flabellia petiolata) and brown algae (e.g., Dictyotales spp.). The habitat also includes encrusting and prostrate sponges, as well as Alcyonacea such as the red gorgonian Paramuricea clavata (Pinna et al., 2021).

In each site, an underwater autonomous acoustic recorder (RASP-URec384k-Nauta RCS) was deployed via SCUBA divers. Each RASP was featured with a Sensor Technology SQ26-05 pre-amplified hydrophone (Sensitivity -164 dB re: V/µPa, flat frequency response between 100 Hz and 30 kHz), a DODOTRONIC programmable recorder (gain 0 dB), battery set, and 400 Gb-SD memory card. The recorders were set to record 15 min every hour in the period between mid-June and mid-August, obtaining 6 hours of recordings per day at a sampling rate of 192 kHz (16-bit resolution). To our knowledge, no studies in the Mediterranean Sea have evaluated the effectiveness of different duty cycles for monitoring fish acoustic communities. Accordingly, the choice of the duty cycles was based only on maximizing recording duration within the limits of battery consumption allowed by the recording systems.

Each recording was visually and aurally inspected by Raven Pro 1.6 (Cornell University), displaying the spectrogram 180 s at a time with frequencies between 0 and 8.5 kHz (FFT length = 8,192; Hamming window, 50% overlap), since most fish make sounds below 5 kHz (Amorim, 2006; Popper et al., 2019). Each clearly distinguishable fish sound, identifiable in both the oscillogram and spectrogram, was selected using Raven software. Then, the following acoustic characteristics were measured: sound duration (in sec), number of pulses, pulse rate (the ratio between sound duration and number of pulses), peak-frequencies (the frequency at which the highest sound power occurs, in Hz), centre frequency (the frequency that divides the sound into two frequency intervals of equal energy, in Hz) and bandwidth (the difference between the 5% and 95% frequencies, in Hz). The number of pulses and the pulse rate were manually measured only for sounds consisting of pulse series, while peak frequencies, centre frequency, and bandwidth were automatically measured by Raven for all selected sounds. Fish sounds were classified based on known fish sounds from the Mediterranean Sea (Parmentier et al., 2010; Picciulin et al., 2012; Kéver et al., 2012, 2016; Bertucci et al., 2015; Parmentier et al., 2018; Di Iorio et al., 2018; Bolgan et al., 2019) and a methodological framework initially proposed by Desiderà et al. (2019), which was based on the acoustic characteristics of the sound types and later modified ( Table 1 ). Five different sound types were classified as frequency modulated (sounds characterised by a changing frequency), namely “FM” and “SFM” (series of FM, emitted by unknown species), “DS” (down sweep frequency modulated sounds, emitted by unknown species; Desiderà et al., 2019), “DSS” (series of down sweep frequency modulated sounds, emitted by unknown species; Desiderà et al., 2019), “LDS” (slight-down sweep frequency modulated sound with peak frequency < 200 Hz, associated with Epinephelus marginatus; Bertucci et al., 2015). Seven sound types were composed of irregular pulse series, namely “LPS” (low frequency pulse series, associated with Epinephelus marginatus; Bertucci et al., 2015) and pulse series which were classified based on their acoustic characteristics applying a k-means clustering (MacQueen, 1967). This clustering method is a widely used unsupervised machine learning algorithm for dividing a dataset into k groups, or clusters, where k is a number specified by the analyst: the algorithm assigns objects to different clusters in a way that maximizes similarity within the same cluster (high intra-class similarity) and minimizes similarity between different clusters (low inter-class similarity), based on the Euclidean distance. Each cluster in k-means clustering is represented by its centre, or centroid, which is the mean of the points assigned to that cluster. The optimal number of clusters was estimated by gap statistic (Tibshirani et al., 2001) using the clusGap function from in R (Maechler et al., 2023). Based on the results (see Supplementary Materials , Supplementary Figure S1 ) six different pulse series were classified: “PS1” (long series of more than 20 pulses with peak frequency around 600 Hz, pulse rate of about 1 sec, emitted by unknown species), “PS2” (series of 5-7 pulses, with peak frequency around 300 Hz and duration shorter than 1 s, emitted by unknown species), “PS3” (series of 3 pulses, with peak frequency around 600-700 Hz and pulse rate of around 0.1 sec, emitted by unknown species), “PS4” (series of 6 pulses, with peak frequency between 600 and 800 Hz, and pulse rate around 1 sec, emitted by unknown species), “PS5” (series of 3-4 pulses with peak frequency around 250 Hz and duration of 0.4 sec, emitted by unknown species), and “PS6” (series of 3-4 pulses, with peak frequency around 600 Hz, longer than 2 sec, emitted by unknown species). Finally, three sound types were composed of regular pulses and have been associated with known species, namely RPS (regular pulse series associated with Sciaena umbra, Bonacito et al., 2001, 2002; Codarin et al., 2012; Picciulin et al., 2012, 2013, 2018, 2020, 2021), APPPS (accelerating pulse series with alternating pulse period associated with Ophidion rochei, Parmentier et al., 2010; Kéver et al., 2012, 2015, 2016; Picciulin et al., 2018), and UFPS (ultra-fast pulses series, associated with Scorpaena spp., Di Iorio et al., 2018; Bolgan et al., 2019). Two sound types were composed of a single pulse, namely “SB” (single boom, associated with Epinephelus marginatus; Bertucci et al., 2015) and “HSB” (a single boom with peak frequency higher than 135 Hz, emitted by unknown species).

The percentage of occurrence of each fish sound type was calculated as the abundance of the sound type over the total sound types counted daily. Acoustic richness was calculated as the number of different sound type per day of recording, acoustic abundance was calculated as the total number of fish sounds per day of recording, while acoustic composition was expressed by the acoustic abundance of each fish sound type per day of recording.

Acoustic richness and abundance were analysed through a permutational analysis of variance (PERMANOVA) based on Euclidean distance, using Site (two levels: TA and ST) and Period (two levels: early-summer and mid-summer, June 14–July 15 and July 16–August 16, respectively) in a crossed design. The sampling period was divided into early and mid-summer to account for the differing temperature conditions observed during these two periods (La Manna, pers. obs.). A non-metric multidimensional ordination (nMDS) was used to visually represent the similarity of fish acoustic composition between sites across periods. Values were fourth root transformed before calculating the Bray-Curtis similarity. Spatial and temporal differences in the structure of fish acoustic composition were analysed by a PERMANOVA based on Bray-Curtis resemblance matrix of fourth root transformed data, using the same model of univariate analysis. The Pair-wise test was used to discriminate between levels of significant factors.

To capture hourly trend in fish acoustic abundance, the percentage of occurrence of each sound type per hour was calculated and represented in several radar plots, distinguished per site.

To verify if the number of recording days was sufficient to represent the whole acoustic richness of each site, a species accumulation curve was calculated with the “speccacc” function of the Vegan package in R. To investigate the relationship between species richness and recording duration, we compared acoustic richness across different duty cycles by selecting various schemes from the 15 minutes of recording per hour: one minute per hour, five minutes per hour, fifteen minutes per hour, and fifteen minutes per hour exclusively during nighttime (from dusk to dawn). Thus, a Generalized Linear Model (GLM) with a Poisson distribution was applied to evaluate the influence of recording duration (a categorical variable with four levels: 1 min/hour, 5 min/hour, 15/hour min, and 15 min/hour at night) and its interaction with site (a categorical variable with two levels: TA and ST) on acoustic richness. For this analysis only the recordings of August, when the highest sound abundance was found, were used (see Results).

All statistical analyses were carried out in R (R Core Team, 2019) using the packages ‘vegan’ (Oksanen, 2017) and ‘MASS’ (Venables and Ripley, 2002).

Recordings were overall collected for 64 days at TA (from June 14 to August 16, 2023) and 62 days at ST (from June 16 to August 16, 2023), detecting a total of 18,886 and 9,889 fish sounds, respectively. The analysis of recordings has identified 17 distinct sound types (17 at TA and 12 at ST, Figure 2 ) with a frequency distribution very different between sites and temporal periods. The most frequently recorded sound type was the RPS (the pulse series associated with Sciaena umbra; occurrence percentage of 81% at St and 38% at TA), which was almost continuously present over the entire study period both in TA and ST ( Supplementary Figure S2A ). The second most common sound, the UFPS (the “/kwa/” sound associated with Scorpaena spp.), was recorded exclusively at ST, with an occurrence percentage of 15%, and an increasing abundance from the end of July up to the mid of August ( Supplementary Figure S2B ). The sound associated with Ophidion rochei (APPPS) was extremely rare ( Figure 2 ) and was recorded mainly in TA and only in few days of June and August ( Supplementary Figure S2C ). The sounds related to Epinephelus marginatus (namely, SB, LPS and LDS) were recorded almost exclusively in TA ( Figure 2 ) and during the whole period ( Supplementary Figure S2D ). The HSB, a single boom like sound, but with highest peak frequency than that described by Bertucci et al. (2015), showed the same daily presence of the SB emitted by Epinephelus marginatus ( Supplementary Figure S2D ). Following RPS and UFPS, the PS2, PS3, and PS5 sounds had the highest occurrence percentages ( Figure 2 ). PS2 and PS5 were both more abundant in TA and in June-July compared to August ( Supplementary Figure S2E ). PS1, PS4 and PS6 were quite rare ( Figure 2 ) and were recorded over few distinct days ( Supplementary Figure S2E ). The frequency modulated sounds (DS, DSS, FM and SFM) were recorded in less than 5% of the recordings at both sites ( Figure 2 ). While DS and DSS were continuously present over the entire study period, FM and SFM were recorded only in few days of August ( Supplementary Figure S2F ).

Acoustic richness was significantly higher at TA compared to ST but showed no significant variation across the periods ( Table 2 ). PERMANOVA and nMDS found significant variations in both fish acoustic abundance ( Figure 3 ) and composition ( Figure 4 ) based on site and period interaction ( Table 2 ). Specifically, post hoc pair wise comparisons on PERMANOVA identified a higher fish acoustic abundance in mid-summer compared to early summer, but only in ST, and that at TA the acoustic abundance was higher than in ST in both periods. Furthermore, pair wise comparisons also indicated that the fish acoustic composition at ST was much more heterogeneous between periods than at TA ( Figure 4 ).

Sound type occurrence was highly dependent on the hour of the day. Some sounds were exclusively recorded during specific nocturnal (e.g., RPS between 9 and 11 pm, PS2 between 10 and 11 pm, SB and HSB between 10 pm and 7 am) or diurnal hours (e.g., PS3 from 7 am to 8 pm) at both sites. A few sound types were recorded at different hours depending on the site (e.g. DS, Figures 5 , 6 ).

To assess the adequate number of days of recording for capturing acoustic richness, a species accumulation curve was calculated for each site. At TA, 94% of the total species were identified after 11 days of recording, while at ST, the curve did not plateau, indicating the need for additional recording days to detect less abundant species ( Figure 7 ).

The negative binomial GLM applied on the acoustic richness of August showed that the duration of recording influenced the ability to capture the species richness: namely longer duty cycle (15 min/hour and 15 min/hour at night) had the highest acoustic richness at both sites ( Table 3 , Figure 8 ).

Seventeen distinct sound types were identified, highlighting a diverse acoustic community within these coralligenous ecosystems. The significant difference in sound detection between the two sites suggests that TA had a more acoustically active or rich fish community during the study period compared to ST, which may be attributed to site-specific environmental factors, habitat structure, species composition or highest level of protection. The number of sound types recorded was lower than those observed in the coralligenous of Corsica and France (Di Iorio et al., 2021) and this discrepancy may be due to differences in regional fish assemblages and/or variations in the classification of fish sounds. In fact, in our study, single-pulse sounds were excluded, and pulse series were statistically classified (based on spectral and temporal acoustic characteristics), potentially accounting for the variation in the number of identified sound types.

The most frequently recorded sound type was the RPS, associated with Sciaena umbra, which was quite more prevalent at ST likely suggesting a difference in the species abundance between the two sites. The UFPS sound, linked to Scorpaena spp., was the second most common sound, and it was exclusively recorded at one site (ST), where it accounted for 15% of the total sounds. In contrast, other sound types such as PS2, PS3, and PS5 were more commonly recorded at the other site (TA). All other sound types were detected with low occurrence (<5%) at both sites. These findings suggest that a few dominant species significantly shape the local soundscapes at ST, where other soniferous species may either be present at very low densities or rely less on acoustic signals for communication (Rice et al., 2017). A similar site-specific variation in acoustic activity has been observed in other Mediterranean coralligenous reefs. Di Iorio et al. (2021) suggested that higher levels of sound production in the coralligenous reefs are associated with more complex habitats, likely due to the enhanced shelter and breeding opportunities habitat heterogeneity provides for fish species. This would suggest that habitat complexity may play a crucial role in driving acoustic activity, as already found in other habitats and ecosystems (Lyon et al., 2019; La Manna et al., 2024b), a hypothesis that warrants further investigation.

Furthermore, while acoustic richness did not vary significantly across periods, acoustic abundance was highest in mid-summer at ST. The nMDS and PERMANOVA analyses confirmed significant differences in fish acoustic community composition between the two sites across the periods: the higher R-value attributed to site (0.051) suggests that spatial differences had a more substantial impact on community composition than variations between periods (R-value=0.031). This could indicate that while the diversity of vocalizing species remained relatively stable, the abundance of acoustic signals increased towards the end of summer, at least in ST, likely due to seasonal peaks in reproductive or territorial behaviours (Staaterman et al., 2014; Parsons et al., 2016; McWilliam et al., 2017; Kaplan et al., 2018) or changes in environmental conditions. These insights are particularly important in the context of climate change and other environmental stressors (tourism, boat traffic, underwater noise), which can alter the natural rhythms of marine ecosystems (Sueur et al., 2019). Therefore, by monitoring these changes, we can detect early signs of ecological disruption and assess the resilience of coralligenous habitats (La Manna et al., 2024a).

The diversity of sound types was also highly dependent on the time of day, with some sounds being specific to nocturnal or diurnal hours. For example, RPS (associated to Sciaena umbra) and APPPS (associated to Ophidion rochei) were mostly recorded between 9 and 11 pm and between 2 am and 6 am, respectively, accordingly with the hourly pattern found for these species in other Mediterranean sites (e.g., Kéver et al., 2016; Parmentier et al., 2018; Desiderà et al., 2019; La Manna et al., 2021; Bolgan et al., 2022). In contrast, few other sound types (e.g. PS3) were more common during daytime hours. This temporal partitioning of acoustic activity likely reflects the daily rhythms of different species, with some being more active or vocal at night and others during the day. Diel periodicity in fish acoustic activity has been documented in various regions (McCauley and Cato, 2000; Luczkovich et al., 2008; Rice et al., 2017; Panicker and Stafford, 2023) including the Mediterranean Sea (e.g., Picciulin et al., 2013; Buscaino et al., 2016; Parmentier et al., 2018; La Manna et al., 2021; Bolgan et al., 2022). In many of these locations, fish calling activity and chorus typically start in the early evening, intensify around sunset, peak a few hours after sunset, and conclude by sunrise. Some other fish instead are more vocal during dawn and dusk, known as crepuscular peaks, which coincide with increased feeding activity and predator-prey interactions (Elise et al., 2019; Bertucci et al., 2020; Carriço et al., 2020). By capturing these daily rhythms, acoustic monitoring can provide a more comprehensive understanding of the ecological dynamics within coralligenous communities. High-resolution temporal data can help identify critical windows of activity (Bolgan et al., 2022), essential for the survival and reproduction of key species, that can be used for designing effective monitoring strategies or temporal restrictions of human activities.

Species accumulation curves revealed a wide inconsistency between sites suggesting that at TA a relatively short monitoring period would have been sufficient to capture most of the acoustic diversity, while at ST additional recording days (over 60) would have been necessary to detect possible less abundant or more elusive species. This difference could be either due to lower species abundance or a less uniform distribution of acoustic activity at ST, making it harder to detect all species within a limited timeframe.

Furthermore, longer duty cycle (15 min/hour and 15 min/per hour at night) captured the highest acoustic richness, consistently at both sites, indicating that extended recording periods are more effective in detecting the full range of species acoustic activity. However, it is important to emphasize that these results pertain to the maximum duty cycle investigated (15 minutes) and the maximum number of sampling days permitted by our recording systems. Even if further research is needed to explore different duty cycles and recording durations, our results are consistent with other studies which obtained a significantly increased in the detection of species richness and acoustic events by longer recording periods (Parsons et al., 2016; Wood et al., 2021). While longer recordings offer a more comprehensive view of the acoustic environment, increasing the chances of detecting rare or sporadic biological sounds, they also introduce significant challenges: extended recording times can quickly consume battery power, particularly in remote or underwater settings where regular maintenance is difficult. Moreover, longer recordings produce large volumes of data, which can be challenging to store, manage, and be analysed. These findings highlight the crucial need to balance extended recording durations with practical limitations, such as battery life and data processing demands. This underscores the importance of carefully planning sampling strategies to optimize the effectiveness of acoustic monitoring without overconsuming logistical resources. Achieving this balance is key to ensuring an efficient and sustainable use of passive acoustic monitoring in ecological research (Risch et al., 2014; Gibb et al., 2019; Sugai et al., 2019).