Ground-truthing daily and lunar patterns of coral reef fish call rates on a US Virgin Island reef

: Coral reefs comprise some of the most biodiverse habitats on the planet. These ecosystems face a range of stressors, making quantifying community assemblages and potential changes vital to effective management. To understand short- and long-term changes in biodiversity and detect early warning signals of decline, new methods for quantifying biodiversity at scale are necessary. Acoustic monitoring techniques have proven useful in observing species activities and biodiversity on coral reefs through aggregate approaches (i.e. energy as a proxy). However, few studies have ground-truthed these acoustic analyses with human-based observations. In this study, we sought to expand these passive acoustic methods by investigating biological sounds and fish call rates on a healthy reef, providing a unique set of human-confirmed, labeled acoustic observations. We analyzed acoustic data from Tektite Reef, St. John, US Virgin Islands, over a 2 mo period. A subset of acoustic files was manually inspected to identify recurring biotic sounds and quantify reef activity throughout the day. We found a high variety of acoustic signals in this soundscape. General patterns of call rates across time conformed to expectations, with dusk and dawn showing important and significantly elevated peaks in soniferous fish activity. The data reflected high variability in call rates across days and lunar phases. Call rates did not correspond to sound pressure levels, suggesting that certain call types may drive crepuscular trends in sound levels while lower-level critical calls, likely key for estimating biodiversity and behavior, may be missed by gross sound level analyses.


INTRODUCTION
As global biodiversity continues to decline (Butchart et al. 2010), cost-effective and scalable ecosystem monitoring techniques are of increasing importance (Cardinale et al. 2012). Passive acoustic monitoring (PAM) is a non-invasive and highly scalable monitoring method which has been used to examine the health and biodiversity of marine and terres-trial ecosystems by extracting significant biological patterns from acoustic recordings over days to several months (Buxton et al. 2018, Gibb et al. 2019, Mooney et al. 2020. Animals produce and use sound for a wide variety of behaviors, including intentional acoustic signals for mate attraction, echolocation, group cohesion, and territorial defense as well as unintentional sounds from movement and feeding (Brad bury & Vehrencamp 1998). By detecting and moni toring changes in biological sounds of interest, it is possible to gain insight into the presence and activity of species and perhaps detect ecologically important behaviors such as foraging or spawning events (Myrberg 1986, Mann & Lobel 1995, Nordeide & Kjellsby 1999, Lobel 2002, Jones et al. 2004. Coral reefs comprise some of the most biodiverse habitats on the planet (Odum & Odum 1955, Connell 1978, Birkeland 1997. PAM on coral reefs has been implemented in multiple studies (e.g. Kennedy et al. 2010, Staaterman et al. 2017, Dimoff et al. 2021) but is just beginning to gain traction as a reef-habitat monitoring and management tool. Yet PAM on reefs is rapidly expanding due to its potential in tracking environmental changes happening in these critical habitats. Many reefs are threatened by a range of local and global stressors, from overfishing (Jackson et al. 2001, Pandolfi et al. 2003) and pollution (McCulloch et al. 2003) to climate change (Gardner et al. 2003. Reef ecosystems not only foster abundant biodiversity in our oceans but also provide countless ecosystem services, including fisheries (Smith 1978), coastal protection (van Zanten et al. 2014), recreational activities (Brander et al. 2007), and cultural importance (Johannes 2002). It is evident that changes in reef complexity and biodiversity have significant im pacts on these ecosystem services (Pratchett et al. 2014).
New methods for quantifying biodiversity at scale are therefore necessary to detect short-and longterm changes in reef complexity and health as global and local threats increase (Beijbom et al. 2012, Mooney et al. 2020, Kennedy et al. 2021. The most common method of reef population and community surveys is diver-based underwater visual censuses (Brock 1954). These observations can be extremely time-consuming, cost-prohibitive in remote locations, can introduce observer bias, and may alter fish behavior, likely confounding biodiversity metrics (Brock 1982, Sale & Sharp 1983, Thompson & Mapstone 1997, Edgar et al. 2004, Harvey et al. 2004. To reduce these biases, there has been increasing interest in non-invasive methods that monitor species abundance and biodiversity. Reefs lend themselves particularly well to PAM because of the abundance of intentional and unintentional biotic sounds from soniferous fish and invertebrates (Rice et al. 2022), and so by listening for reef sounds associated with soniferous fish and invertebrates, researchers can monitor reefs passively. Sound is an important component of many reef fish behaviors such as mating, feeding, and communication (Myrberg 1981, 1986, Mann & Lobel 1997, Tricas & Boyle 2021. Reef soundscapes also contain many ancillary sound cues which cause changes in fish and invertebrate activity (Radford et al. 2008).
An increasing number of studies leverage these sounds and use a higher-level aggregate approach and proxies (i.e. energy, diel patterns, various indices) to estimate community assemblages, complexity, and reef fish diversity due to difficulty in assigning call type to species (Kennedy et al. 2010, Bertucci et al. 2016, Lin et al. 2017, Staaterman et al. 2017. However, these methods can be biased by certain factors, such as the choruses of acoustically dominant species of fish and invertebrates (Mooney et al. 2020), and often fail to verify the actual diversity of call types and their respective call rates. Consequently, bulk analyses can potentially induce inconsistencies when they are used as proxies for biodiversity and biotic activities on a reef.
Here, we sought to expand acoustic methods and data sets by investigating patterns and changes in fish call rates throughout the day using a manual survey method (e.g. Kaplan et al. 2014, Tricas & Boyle 2014, Silva et al. 2016 to identify call presence and discrimination. We examined and analyzed recurring biotic sounds to determine fish vocalization trends and variability at Tektite Reef in St. John, US Virgin Islands (USVI). We selected Tektite Reef based on its relatively high health condition compared to nearby reefs  and its long history of ecological and bioacoustic assessments (Edmunds 2013, which provide an important regional baseline and sufficient data to observe overall daily patterns in fish call rates. We then compared trends with existing bulk analytical methods applied to our data.

MATERIALS AND METHODS
Data were collected using a single-channel Soundtrap 300ST recorder (Ocean Instruments). The recorder was placed at Tektite Reef (18°18' 26.8" N, 64°43' 14.7" W; Fig. 1), 0.5 m above the benthos, attached to a rebar stake. From prior acoustic monitoring studies, we knew Tektite was acoustically rich relative to surrounding reefs and that this richness corresponded with observed high fish diversity . The SoundTrap was deployed from 19 March to 12 July 2017 and set to record on a duty cycle with 3 s of self-calibration and 60 s of recording every 10 min. For data analysis, 8 d were sampled by 2 independent, trained analysts in April and May 2017 during the full (11 April, 12 May), third-quarter (19 April, 18 May), new (26 April, 25 May), and first-quarter (2 May, 1 June) moons. Because such manual analysis was extremely timeintensive, the goal here was not to characterize all fish sounds found on Tektite Reef but to investigate prominent call types and quantify call rates during early summer, an acoustically active season , as a first look into call patterns and finescale acoustic diversity. This analysis would then provide a key comparison to batch analyses, for example using bandpass-filtered sound pressure levels (SPLs) as a proxy for fish activity (Staaterman et al. 2014, Bertucci et al. 2015. Furthermore, our analyses include fine-scale labels which will help validate subsequent automated acoustic event detection and classification efforts. To label individual acoustic events, we used an interactive graphical user interface created in MAT-LAB R2020a (MathWorks) to visualize and analyze 1 min sound files every 30 s for each recording day. Recordings were first down-sampled from their original 48 kHz sampling rate to 4000 Hz, allowing for analysis of sounds in the frequency band from 0− 2000 Hz containing nearly all fish vocalizations (Tricas & Boyle 2014, . Spectrograms (128 point, 32 ms Hamming window, 75% overlap, 1024 point Fast Fourier Transform size) were visualized in 10 s time frames to better observe calls. Individual calls were manually labeled with time− frequency bounding boxes within those 10 s spectrograms (see Fig. 2A). Time and frequency limits and corresponding labels were saved to a text file associated with each audio file. Because these reefs experience intermittent small boat traffic (Dinh et al. 2018), we labeled and removed periods with audible boat noise that could mask detection of fish calls, similar to . A final check of each file was made after initial auditing to parse out dense vocalization sequences.
Call rate for each file was calculated as the number of calls per minute (of all call types) of audio after excluding periods with vessel noise (e.g. . Average call rate per minute was determined throughout the 8 d of data for each 1 h interval. To observe patterns in dawn and dusk chorusing peaks, call rates were aggregated and averaged among night, dawn, day, and dusk (defined as 20:01−04:34, 04:35−05:53, 05:54−18:42, and 18:43−20:00 h, respectively) following data on www.timeanddate.com (Table 1).
As a comparison to call rate estimates, we quantified the received SPL within the fish band, a 50− 1200 Hz low-frequency band that includes most fish acoustic activity, as a proxy for large-scale biotic activity, following previous studies (Kaplan et al. 2018, Au & Banks 1998, Tricas & Boyle 2014. Files containing vessel noise were removed from the analysis. We filtered each file with a 50−1200 Hz 4-pole Butterworth bandpass filter, then calculated the rootmean-square (RMS) SPL for each 1 s period, transformed it to decibels, and corrected for the acoustic A linear mixed-effects model was used to observe how time of day and lunar phase affected call rates, using the 'lme4' toolbox (Bates et al. 2015) in R v.4.1.2 (www.r-project.org/). We modeled time of day and lunar phase as fixed effects to examine their importance to call rate and mean sound level. To account for variation between individual days, we ad ded day as a random effect on the intercept. We generated 4 nested models: a null model with only day as a random effect, 2 extending models that were the same as previous but sequentially added time of day and lunar phase, and finally a full model that included the interaction between time of day and lunar phase. A stepwise likelihood ratio test was performed from most to least complicated model using the 'anova' function from the 'car' package in R (Fox & Weisberg 2019) to evaluate the importance of each fixed effect and determine which model was best fitted to the data. A pairwise comparison using the 'lsmean' toolbox (Lenth 2016) in R was then used to determine whether average call rates at night, dawn, day, and dusk were significantly different from each other.
To understand the impact of individual calls on total energy content of the soundscape, we filtered all manually labeled fish sounds with a 50−1200 Hz 4pole bandpass filter. We then defined the call duration as the period containing 95% of the total acoustic energy (following Madsen et al. 2006) and calculated the received level (RL) as the RMS SPL over the 95% energy duration. Finally, we estimated the sound exposure level (SEL) for each call by multiplying the average intensity by the total duration (in decibels, by adding 10 × log 10 [d] to the RMS SPL in dB).

RESULTS
We addressed fish call rates on a coral reef across 8 d of recordings spread over 2 mo. Like many coral reefs, Tektite Reef provides habitat to a community of fish and invertebrates across trophic levels , Mooney et al. 2017); consequently, a diversity of call types were noted across these days and months. Given the previous efforts to address sound energy on reefs, this particular effort sought to address the temporal patterns of these reef fish calls with respect to overall sound levels. Call classification was not addressed here specifically but is a focus of ongoing work. To provide some context, Fig. 2 shows a series of example spectrograms with 4 call types that were seen most often. The low-frequency soundscape was primarily dominated by a variety of pulsed calls (Fig. 2B−D), though occasional tonal calls were also present, and at least one tonal call type (Fig. 2E) was seen intermittently throughout the 8 d.
Hourly rates of individual biological calls showed general peaks around dawn (05:30 h) and dusk (19:30 h) (Fig. 3). However, there appears to be a wide variety in call rate throughout each day. The highest call rate was seen at 19:30 h on 19 April (Fig. 3C), with 149.5 calls min −1 . The lowest call rate was 8.5 calls min −1 at 15:30 h on 2 May (Fig. 3E).
The linear mixed-effects model showed that time of day had a significant effect on call rate and that the effect of time of day depended on lunar phase (Table 2, Fig. 4A). However, phase alone did not have an impact on call rate. Akaike's information criterion (AIC) suggests the full model is best (AIC = 3145.2), while Bayesian information criterion (BIC) suggests Model 1 is the best (BIC = 3170.4). This discrepancy indicates that the effect of the interaction between time of day and lunar phase is tentative.
Received SPL values (dB rms re 1 μPa) between 50 and 1200 Hz showed similar trends, with general peaks at dawn and dusk (Fig. 5). There was variation in energy during each of the 8 d; however, the highest variability was seen during the day (~05:54− 18:42 h), with large peaks such as that at 09:30 h on 10 May (Fig. 5B) and 12:30 h on 26 April (Fig. 5D). The average received SPL depended only on time of day and not on lunar phase (Table 3).
While call rates and SPLs appeared to follow a generally similar diel trend, fish calls were generally short (median 95% energy duration: 0.13 s; Fig. 6A) and low amplitude (median RL: 98.5 dB re 1 μPa; Fig. 6B), resulting in very little energy contributed to the overall soundscape (Fig. 6C). As a consequence, we did not find a significant correlation between re ceived SPLs and call rate (Fig. 6D).   Table 3. Linear mixed-effects models implemented to examine drivers of received sound levels (RL) on the reef. See Table 2 for further details. Results show that time of day significantly improved the model (* and bolded p-values for p < 0.05) first-quarter moon on 2 May (red) and 1 June (blue). Note that per-file sound pressure level was measured at 30 min intervals for the same files analyzed in Fig. 3

DISCUSSION
New cost-effective ways to determine coral reef health and biodiversity are needed to understand changes in reef health due to many current stressors. Marine soundscapes have been of increasing interest because they facilitate scalable longer-term observations, are cost-effective, and can support near realtime detection of reef animal activity. However, to our knowledge, most acoustic methods for observing reef soundscapes take a higher-level aggregate approach with limited, albeit time-intensive, assessments of the in situ acoustic activity of reef fish and invertebrates. The goal of this study was to determine how call rates within a coral reef soundscape vary throughout the day using a detailed manual method that is often applied to the signals and calls of other vocalizing taxa but rarely applied to fish call rates and characterizations. This work highlights the temporal patterns of these call rates while comparing them to the overall acoustic sound levels on the reefs (Fig. 6).
Using PAM data from St. John (USVI) and manual analysis methods, we observed daily call rates at Tektite, a well-studied reef (Edmunds 2002(Edmunds , 2010. Call rates varied widely, ranging from 10 to 150 calls min −1 . We found an increase in call rates around dusk and dawn, similar to those seen in previous studies on this reef  and elsewhere (Staaterman et al. 2014, Kaplan et al. 2018). These methods tend to quantify overall SPLs and thus could be dominated by repeated, high-intensity calls from certain fish or invertebrate choruses (Radford et al. 2008, Parsons et al. 2016, Kahl et al. 2021, Bolgan et al. 2022 and miss more subtle lower-amplitude calls. Interestingly, we did not see a clear association be tween call SPL and call rate, despite generally similar diel trends (Fig. 6). Most calls were short, low-amplitude (and low signal-to-noise ratio) calls and thus did not contribute much acoustic energy to the larger soundscape (Fig. 6C). Given a median SEL of 90.3 dB re 1 μPa 2 ·s and a median SEL of 113.4 dB re 1 μPa 2 ·s per 60 s file, it would take an average of 204 calls to increase the total acoustic energy by 3 dB. There could be many reasons that many reef fish calls are relatively low amplitude, but in part, this may be an ecologically important strategy; to communicate sufficiently with your neighbor conspecifics but not necessarily acoustically advertise one's position on a predator-rich reef.
We also noted an approximate halving of call rates during the day (compared to the crepuscular time periods), suggesting a lower level of activity within the fish community during those periods. Further, these lower daytime call rates underscore the need for caution when applying brief 'snapshot' recordings, especially if they are made during the daytime (Kennedy et al. 2010), as they may not fully capture the acoustic diversity of the reef. Presumably, the fish community did not change substantially during the 8 d recording period, but these data show the call rates did change significantly.
Our current SPL data (Fig. 5) show an increase during the day similar to previous studies , which suggests an increase in biological activity. However, we suspect that these differences may be a result of bulk analyses picking up some component of the background soundscape. This may be simply wind-driven wave sounds from warmer daytime temperatures or an abundance of a few harderto-discriminate call types (such as choruses and signals near background), which would elevate overall sound levels but be hard to discern and label in our manual analyses. Our analysis methods are not driven by the energy of calls, but each individual call is weighed equally. By taking a more individualized approach, we eliminate background noise sources, such as wind and vessels, that may dominate the soundscape and largely influence energy analyses. Of course, dominant, repetitive individual calls could in fluence this method, but at least anecdotally (Fig. 2), this was not the case. Further research on call-type characterization would help address this uncertainty.
We also observed variability in call rate throughout each individual day. There are evident peaks in the mornings of 11 and 19 April. Upon further investigation, these peaks appear to be dominated by a single call type, which appears as a triple pulse with a frequency range between 300 and 1200 Hz. During this chorusing, rapid pulsing occurs for about 10 min at a time with up to 40 individual calls within that period. While it is unclear what animal is causing these sounds and for what reason, it is possible that it may be one individual or species. Determining which species is making sounds has been a long-standing question and need in bioacoustics (Fish & Mowbray 1970). New methods such as rebreather or passive acoustic arrays and camera systems are now being applied to address such questions (Tricas & Boyle 2014, Mouy et al. 2018. In many cases, we are still discerning how fish call characteristics are influenced by environmental conditions. Compared to mammals, fish call repertoires are somewhat fixed, but it appears that key events such as changes in temperature, upwelling, or habitat loss may influence call rates and signal parameters (Connaughton et al. 2002, Mann & Grothues 2009, Papes & Ladich 2011, Lamont et al. 2022). More work is needed here to better apply our bioacoustic assessment analysis tools.
Finally, this study underscores that diel patterns are a major driver of call rates, more so than lunar trends observed in the USVI and Hawaiian coral reefs (Kaplan et al. , 2018. Given that bulk analysis methods often compare reef health or biodiversity based on sound levels, a possible next step could be to apply these call rate observations to additional reefs and compare noted call rates and diversity across a range of reef conditions. However, one limitation is the time-intensive nature of such manual ana lyses (hence the application of diversity and sound-level metrics) (Mooney et al. 2020). Machine learning is also an approach recently applied to coral reef sounds which has been used to identify reef fish choruses and bioacoustic patterns (Lin et al. 2017, Lin & Tsao 2020. Manually labeled data sets provide an important ground-truth resource with which to validate automated machine-learning methods and ultimately allow for scaling up individual-level PAM. Quantifying the call rates of fish, the parameters of those calls, and how such sound patterns may be driven by environmental or human-induced (i.e. noise) changes is a rapidly emerging field (Munger et al. 2022). However, the optimal method of detecting potential changes has yet to be determined. The data here help lay the groundwork for applying emerging methods using traditional detectors or machine learning. SPLs on their own may not be the best metric, but actual call rates can certainly help support those methods as well as elucidate behaviors.
Open-source collections of fish sounds are still in their infancy but are rapidly developing (i.e. recently launched www.fishsounds.net; Looby et al. 2022). Manual identification and labeling of calls, such as this work, or manual labeling of calls and subsequent unsupervised identification of call types (Parra-Hernández et al. 2020, Sainburg et al. 2020) are critical for populating these data collections with a wealth of individual fish and invertebrate sounds.
While this was a quantitative analysis of call rates, a subsequent analysis is underway to examine call types, differentiation, and potential niche with over 100 call types defined (authors' unpubl. data). Discerning call types in coral reefs is challenging because of the high number of calls and the ability of individual species to make a variety of call types depending on behavior (Kasumyan 2008). Previous studies have attempted to determine call types in aquarium settings but have shown that it is impossi-ble to recreate specific behaviors and sounds in captivity (e.g. Hawkins & Amorim 2000, Širović & Demer 2009). Therefore, peer-reviewed efforts to discern call categorization are needed. These can eventually be used for training automated algorithms to simultaneously detect and parse out the rich fauna found in coral reefs, similar to how current deep neural nets are now capable of classifying hundreds of bird sounds with high accuracy (Kahl et al. 2021). Such an individual-level approach to the analysis of coral reef soundscapes may be promising for tracking local abundance of keystone species, invasive species, or sentinel species that respond early to ecosystem changes (Tricas & Boyle 2021) and thus have the potential to transform coral reef ecosystem monitoring.