A Low-Cost Passive Acoustic Toolkit for Underwater Recordings

1. Introduction

Sound is a crucial component of the marine environment [1,2] and an effective means to study animal life in the ocean [3,4]. Many marine animals have evolved ways to use sound production and hearing in behavioural facets that are critical to their survival, such as cetaceans, marine invertebrates and numerous fishes [3,5,6]. Yet, in the oceans, anthropogenic low-frequency noise (<1 kHz) from shipping has steadily increased over the last 50 years [7,8,9] at a concerning average rate of 3 dB per decade [10]. Indeed, most human activities in the marine environment produce sound, and this pervasive input is increasingly altering the marine acoustic environment [11]. Depending on temporal patterns of traffic density, proximity to shipping routes, and acoustic propagation conditions [12] anthropogenic noise can have detrimental effects on marine biota and may disrupt entire biocommunities [13,14,15,16,17].

In the European Union (EU), low levels of underwater noise are a criterion for achieving ‘good environmental status’, as defined by the Marine Strategy Framework Directive (MSFD) in Descriptor 11: ‘Energy, including underwater noise’ [18]. Furthermore, each member state of the EU is required to research, monitor and report on the levels and spatiotemporal distribution of impulsive and continuous underwater noise [19]. In the open waters of the Mediterranean, the main sources of anthropogenic noise are the low-frequency sounds radiated from commercial ships and fishing vessels [20], while coastal waters are also subjected to small boat noise from recreational activities during the summer months [21,22]. Coastal marine protected areas (MPAs) implement various marine traffic restrictions and no-fishing policies [23,24], but their proximity, and often overlap, with tourist hotspots, coupled with the pervasive nature of underwater noise, limits their efficiency in safeguarding against this stressor [25,26,27].

The characterization of marine soundscapes and the study of underwater sound is therefore key for management bodies to develop informed noise mitigation policies [28,29,30]. To this end, passive acoustic monitoring (PAM) methods are particularly suited, either using fixed acoustic stations [31,32,33], moored oceanographic buoys [34], free-drifting calibrated hydrophones [35] or ad-hoc spot measurements [36]. Even though these methods can provide valuable and high quality measurements, their data collection efforts are often limited by financial and personnel availability constraints when aiming for large-scale coverage. In an effort to pursue low-cost alternatives to expensive underwater recording equipment, various initiatives have developed their custom versions of low-cost hydrophones, available either commercially with underwater housing, supporting software and duty-cycle options [37] or as homebuild do-it-yourself (DIY) designs based on off-the-shelf piezoelectric (PZ) elements for on-site recordings [38,39,40,41].

The aim of this study was to develop and test in-depth under different usage scenarios an easy to construct low-cost hydrophone, intended for large-scale acoustic sampling by citizen scientists and researchers alike. Its performance for demanding research applications such as marine traffic measurements and long-term soundscape monitoring was assessed across four field experiments at coastal sites, designed to capture both anthropogenic and biological sounds. The low-cost unit was compared with a scientific-grade hydrophone (the SNAP autonomous recorder from Loggerhead Instruments). The build proposed herein is part of an acoustic toolkit that consists of the hydrophone and a low-cost commercially available handheld recorder and optionally, it can be coupled with a cell phone device for easy-to-conduct field recordings. This hydrophone provides a cost-effective build (approximately 20 € per unit plus 30 € for a handheld recorder) and features a simple design that allows for straightforward sampling.

2. Materials and Methods

2.1. Hardware Components and Hydrophone Construction

Central to our low-cost hydrophone design is a thin piezoelectric disk 35 mm in diameter, identical to the PZ elements commonly used in gift cards, acoustic guitar pickups, and electronic buzzers. After soldering it to an off-the-shelf audio cable, the PZ disk is glued with cyanoacrylate (‘super’) glue onto a plexiglass disk of 4 mm thickness. The latter is then superglued to a PVC threaded nipple and acts both as the face of the hydrophone and as a lid for the enclosure (Figure 1a,b,c). At the back (recorder-facing) side of the hydrophone, the audio cable exits the enclosure through a brass pipe fitting that is screwed into the PVC thread and closely matches the cable’s diameter at its narrowest end. Finally, all connection points of the hydrophone are sealed to prevent water influx. Specifically, the PVC-to-plexiglass connection (Figure 1d-i) is sealed with a thick layer of adhesive sealant for marine applications (Sikaflex-291i), taking care not to cover the hydrophone’s face and the PZ element that sits behind it. The threaded part (Figure 1d-ii) is sealed with both Teflon tape and liquid Teflon, and the exit point of the audio cable (Figure 1d-iii) is protected with a layer of adhesive sealant and heat shrink tubing. The resulting unit offers adequate mechanical strength and enough weight to achieve slightly negative buoyancy. A full list of assembly components with indicative prices is provided in Appendix 1 (Table A1).

2.2. Experimental Recordings

Two custom-made uncalibrated hydrophones were experimentally tested at four coastal sites of the Aegean and Ionian Seas (Greece): Marathonisi Islet at the National Marine Park of Zakynthos (NMPZ), the port of Mytilene at Lesvos Island, and the sites of Agrilia and Villa at Southeast Lesvos (Table 1, Figure 2).

During these test recordings, one custom-made hydrophone (hereafter referred to as ‘Nemo-1’) was always coupled with a Philips DVT 1120 handheld recorder that is part of the acoustic toolkit. The recorder of the second custom-made hydrophone (‘Nemo-2’) varied by site, and was either a Tascam DR-05X or an M-Audio MicroTrack II. The SNAP autonomous underwater recorder from Loggerhead Instruments, equipped with the HTI-96-Min acoustic sensor (sensitivity -170 dB re 1V/μPa), was used in all sessions as a reference, factory-calibrated scientific hydrophone. For ease of deployment, all acoustic devices were mounted on the same custom-built stainless steel base, which weighed approximately 8 kg in air (Figure 2f). A summary of recording settings per test session is provided in Table 1, while the recording sites are described in the sections that follow.

2.2.1. Recording Site 1: Marathonisi Islet at NMPZ

The National Marine Park of Zakynthos covers 83.3 km² of marine protected area along the south shores of Zakynthos Island (Ionian Sea, Greece). It fully encompasses Laganas Bay, a shallow water (<50 m) embayment with soft substrate dominated by unvegetated sandy beds and Posidonia oceanica meadows [42]. This MPA hosts one of the most important rookeries of the loggerhead sea turtle Caretta caretta in the Mediterranean [43,44], but is also subjected to severe anthropogenic pressures due to the recreational and economic importance of the area. The sandy beaches that stretch along inner Laganas Bay are heavily visited during summer, while nearshore infrastructure and coastal sites of high aesthetic value support a thriving tourism industry that peaks from June to August [45]. Leisure boat traffic is substantial during this season, and includes rigid inflatables, speedboat rentals, and hard-hulled eco-tourism boats that mostly launch from the ports of Limni Keriou and Agios Sostis (Figure 2b). The marine part of NMPZ is divided into three major management zones (A, B, C) and a peripheral (buffer) zone characterized by varying levels of protection and restrictions of human activities [23]. Zone A, at the eastern part of Laganas Bay, is a no-access area from May to October, while summer boat traffic (without anchoring) is permitted in zone B under a maximum speed limit of 6 knots. Zone C shares the rules of zone B, but boat anchoring is permitted year-round.

This zoning scheme keeps most speedboats away from the key nesting beaches of C. caretta at East Laganas Bay, but channels them near Marathonisi Islet and the rocky shores of Southwest Zakynthos. To investigate the performance of our custom-made hydrophones in monitoring speedboat traffic in high-use coastal waters, a test recording session was conducted near Marathonisi Islet on 26-06-2025 (Figure 2b, inset). Using an anchored small boat, the acoustic station was positioned on the seafloor at 3 m depth and recorded from 11:20 to 14:20 local time (Table 1). Concurrent visual observations were also conducted by the researcher on site, including boat distance between the acoustic station and the speedboat at their nearest point, boat direction, and perceived speed. Leisure marine traffic was dense throughout the recording period, and the distance between the acoustic station and passing speedboats ranged from a few meters to over 300 m.

2.2.2. Recording Site 2: Agrilia

A boat-based recording was carried out on 08-08-2025 in the vicinity of an aquaculture facility at Agrilia, Southeast Lesvos, in an effort to record the sounds produced by bottlenose dolphins (Tursiops truncatus) that frequent these waters [46]. The survey boat was anchored 200 m east of the aquaculture net pens (Figure 2d), and the acoustic station was lowered off the stationary boat to a depth of 7 m (Table 1). The recording started at 07:20 local time and was stopped after one hour due to worsening weather conditions. Although dolphins were neither sighted nor recorded, this fieldwork session captured the passage of a 195 m long ferry boat, northbound to Mytilene port. At its nearest point to the research boat, the distance between the ferry and the acoustic station was approximately 2 km.

2.2.3. Recording Site 3: Mytilene Port

A test recording was conducted at the port of Mytilene (Figure 2e) on 10-06-2025, focused on inboard vessels. From 17:40 to 18:10 local time, the acoustic station was deployed off the pier at 4 m depth (resting on the seafloor) and recorded the scheduled departure of a passenger boat (40 m in length) that travels daily between Mytilene and Ayvalik (Turkey). Recording settings for this session are listed in Table 1.

2.2.4. Recording Site 4: Villa

The coastal underwater soundscape of the Villa site (Figure 2c) was recorded for 34.3 consecutive hours between 14 and 15 August 2025, starting at 13:10 local time. As reported by the local diving center that frequently visits this site, the rocky reefs and Posidonia oceanica patches of Villa host several soniferous fish, such as the brown meagre, Sciaena umbra [47], the dusky grouper, Epinephelus marginatus [48], and various bottom-dwelling Scorpaenidae [49]. The acoustic station was fixed to the rocky substrate at a depth of 6 m (Figure 2f), while the handheld recorders were placed in a weatherproof case secured outside the seawater. Recording settings for this session are provided in Table 1.

2.3. Acoustic Data Processing

All acoustic recorders used in this study stored data in 16-bit WAV files, albeit with different file naming schemes and file storage lengths. For each test site, the initial preprocessing step consisted of manually synchronizing the concurrent recordings across all hydrophones, using the Raven Pro v1.6.5 sound analysis software [50]. The waveforms were visualized as separate channels in Raven Pro and were manually aligned in time using distinct audio marks (waveform peaks) that were added to the recording by the researcher on site. These marks were created by repeatedly hitting the steel station base with a metallic object in a rhythmic manner (both before deployment and after retrieval). The acoustic data files were cropped properly and renamed according to the SNAP file naming scheme (yyyymmdd_hhmmss), resulting in three audio data sets per recording site of identical timestamps and total duration. No filtering, resampling, or gain adjustments were applied to the raw data.

The synchronized acoustic series were then transformed by a fast Fourier transform (FFT) in Raven Pro, and the resulting spectrograms were visually and aurally inspected to obtain an understanding of the dataset contents. Sounds of interest were manually annotated, and audio segments needing further processing were accordingly marked. For the Marathonisi recording, a detailed analysis was performed in Raven Pro to manually mark all unique speedboat instances identified in the spectrograms, informed by the on-site visual observation log that recorded their vessel type and passage timestamp. Notwithstanding overlaps due to dense traffic, for each speedboat instance identified on the spectrogram’s time-frequency grid, a rectangular selection box was drawn that encompassed the sound of interest. The lower and upper frequency limits of the selection box were fixed at 30 Hz and 8 kHz, respectively, whereas the start and end time of the box defined the respective limits on the time axis. The following parameters were then extracted for each selection: 90% duration, dt_90% = t_95% - t_5% (s), where t_5% and t_95% are the points in time that define the first and last 5% of the sound’s energy, respectively; peak frequency, f_peak (Hz); center frequency, f_50% (Hz), defined as the frequency that splits the annotation box into two parts of equal energy; 90% bandwidth, Bw_90% = f_95% - f_5% (Hz), and interquartile bandwidth, Bw_50% = f_75% - f_25% (Hz), where f_5%, f_25%, f_75%, and f_95% are the frequencies that contribute to the first 5%, 25%, 75% and 95% of the sound’s energy, respectively. All aforementioned descriptors are standard outputs of the Raven Pro software. Subsequently, all speedboat selections were exported as separate WAVs, and the PAMGuide software [51] was used to compute their power spectral density (PSD).

To visually summarize the important contributors to the underwater soundscape over the 34.3-hour recording at Villa, the PAMGuide software and custom MATLAB scripts were used to produce long-term spectral average (LTSA) plots with 30 s averaging window. Root-mean-square (RMS) noise levels at 1/3-octave bands (TOLs) were also computed in PAMGuide for the Mytilene port and Agrilia sites. All spectrograms, PSD computations, TOLs, and LTSAs were produced in relative dB units (uncalibrated) for the custom-made Nemo hydrophones, while calibration data for the SNAP device were: 2 dB gain, clip level (peak-to-peak) 1.58 V, and hydrophone sensitivity -170 dB re 1V/μPa.

3. Results

3.1. Recordings of Recreational Boat Traffic at NMPZ

A total of 182 speedboat instances were identified during the 3-hour recording period at Marathonisi site, corresponding to a heavy traffic density of about 60 speedboat passages per hour. Figure 3 shows a typical 20-minute segment between 12:54 and 13:14 local time, concurrently recorded with the SNAP, Nemo-1, and Nemo-2 hydrophones. The synchronized spectrogram plots show that the low-cost hydrophones can adequately record speedboat vessel traffic in shallow waters, both in terms of boat counts and overall received sound pressure levels (Figure 4). When compared to the 182 instances identified in the SNAP dataset, only 8 passages could not be visually detected on the Nemo data, which corresponded to distant (>300 m) and/or slow-moving, low-frequency inboard vessels. As illustrated in the spectrogram plots and in the PSD of the recordings (Figure 3f,g), the low-cost hydrophones have poor reception abilities at frequencies below 200 Hz, are more effective in the mid-to-high frequency range (>1 kHz), and have resonance peaks around 3 kHz that saturate their spectrograms.

Received sound pressure level values at all three hydrophones were computed (Figure 4) for the same 20-minute segment presented in Figure 3. Apart from a few SPL peaks where the SNAP hydrophone picked up short duration broadband crackling sounds that were not registered by the Nemo devices (e.g. at 12:54:00, 13:03:30, 13:04:20, 13:09:40), the uncalibrated received pressure levels at both Nemo hydrophones were in good agreement with the reference SNAP. Overall, the comparative results show that even though the low-cost hydrophones have limited response at lower frequency bands, they are able to record events even at 250 Hz (Figure 4b), provided the source is a high energy sound such as a close-range passing speedboat.

Quantitative spectral descriptors (30-8000 Hz) were also calculated for the 182 speedboat instances recorded at Marathonisi (NMPZ) using SNAP and Nemo-1 hydrophones, and are displayed in Figure 5 as box plots. The Nemo-1 derived descriptors are biased towards higher frequencies, both due to the resonant frequency at 3 kHz that makes peak (f_peak) and 50% frequency (f_50%) metrics centered at this hotspot, and the low sensitivity at low frequency sounds that do not contribute energy to the Nemo-1 selections. As a result, the energy-weighted bandwidth metrics have a smaller range than those of the SNAP, while all other frequency metrics of speedboats are overestimated.

Summary descriptive statistics of selected temporal and frequency metrics are reported in Table 2 for the SNAP and Nemo-1 hydrophones. The mean energy-weighted duration (dt_90%) of Nemo-1 recorded speedboats is similar to that of SNAP, albeit underestimated by 2.1 s. Bandwidth (Bw_90%) and 95% frequency (f_95%) are the only descriptors that are comparable to the reference (SNAP) ones, while the f_peak, f_50%, and f_5% metrics are highly biased.

3.2. Recordings of Shipping Noise at Agrilia Site

The one-hour recording session at Agrilia captured the passage of a northbound ferry heading to Mytilene port, at an approximate distance of 2 km from the acoustic station. Excluding a small (5 m) outboard boat that was conducting regular maintenance work at the nearby aquaculture facility, no other vessel was visible in the vicinity. The concurrent recordings and corresponding noise measurements of this event are displayed in Figure 6, both for the reference, calibrated SNAP and the low-cost, uncalibrated Nemo-1 hydrophone.

The PSD spectrogram plots (Figure 6a,b) show that the event is clearly registered by the SNAP device across the full spectrum examined. The event is also discernible at the Nemo-1 spectrogram and appears overall similar to that of SNAP, although displayed more faintly (Figure 6b). The received sound pressure levels recorded by the SNAP and Nemo-1 indicate a significant difference in the frequency response of the two hydrophones. The reference SNAP device employs the HTI-96-Min hydrophone, which has a flat frequency response from 2 Hz to 30 kHz, as per the manufacturer. It therefore registers a monotonic increase in received SPL values for each progressively broader frequency band, and, when compared to the approximately 112 dB re 1 μPa baseline prior to the ferry passage (Figure 6c, 07:22–07:30), it registers a maximum 17 dB re 1 μPa increase in SPL at the closest point of the ferry. On the other hand, Nemo-1 is effectively unresponsive at low frequencies below 200 Hz, where most of the ferry’s radiated noise is, registering an overall SPL increase of only 6 dB (relative) throughout the event (Figure 6d). This is also reflected in the RMS noise measurements at 1/3 octave frequency bands (Figure 6e,f), where the Nemo-1 1/3-octave bands begin to contain acoustic energy above approximately 150 to 200 Hz.

3.3. Recordings at Mytilene Port

For the recordings conducted at the port of Mytilene, PSD spectrograms of a single inboard vessel departure (40 m in length) were computed for the SNAP, Nemo-1, and Nemo-2 hydrophones, accompanied by RMS noise measurements for all 1/3-octave frequency bands from 32 Hz to 8 kHz (Figure 7). The results show similar patterns to those observed at Agrilia and Marathonisi, i.e. the Nemo low-cost hydrophones are able to amply register the high energy event at mid to high frequencies, but perform poorly below 200 Hz and systematically peak at the resonant frequency of about 3 kHz (Figure 7e, 1/3-octave band centered at 3162 Hz).

3.4. Long-Term Recordings of the Coastal Soundscape at Villa Site

Long-term spectral average plots were produced to summarize the continuous recordings (34.3 hours) at Villa site obtained with the SNAP and Nemo-1 hydrophones (Figure 8). The diurnal cycle of crustacean sound activity at high frequencies is prominent in both plots, as are the various small-scale fishing inboard vessels that frequent these coastal waters. The Nemo-1 output is highest at the resonant frequency of 3 kHz, effectively saturating the spectrogram during the night, when crustacean snapping sounds mostly contribute to the soundscape. At lower frequencies around 300 Hz to 1 kHz (Figure 8c,d), most high-energy instances were recorded by Nemo-1, albeit signals below 200 Hz are practically absent from its spectrogram. Although the entire acoustic dataset was scrutinized for fish sounds that were regularly present in the SNAP data, especially during dusk, only three such low-frequency instances were spotted in the Nemo-1 series, corresponding to sounds produced by fish that were apparently very close to the acoustic station (Figure 9).

Appendix A.1

Table A1. List of components used for the construction of the Nemo hydrophones; reusable components across multiple units are marked with an asterisk (*).

Component	Function	Price (€)
Piezoelectric disk (35 mm)	Acts as the acoustic element of the device	1.5 /unit
Audio cable (10 m)	Connects the PZ disk to the recording device	5 /unit
⌀60 mm plexiglass disk (4 mm)	Is attached to the PZ disk and acts as a lid to the hydrophone’s enclosure	3 /unit
PVC threaded nipple	Engulfs the acoustic element (housing)	0.85 /unit
Brass pipe fitting	Threaded on the PVC extension, it closes the recorder-side of the hydrophone and helps the audio cable exit the watertight enclosure. Additionally, its weight keeps the hydrophone submerged underwater	6 /unit
Heat shrink tubing *	Used for sealing and abrasion resistance at the point where the audio cable exits the enclosure	0.15 /m
Sikaflex-291i *	Used for sealing the front and back side of the hydrophone	12 /300 ml
Liquid teflon*	Used for sealing the connection between the brass fitting and the PVC thread	11 /100 ml
Teflon tape *	Used for sealing the connection between the brass fitting and the PVC thread	0.5 /10 m
Cyanoacrylate glue *	Used for attaching the PZ disk on the plexiglass disk, and for attaching the plexiglass disk on the PVC extension	2.5 /unit

Table A1. List of components used for the construction of the Nemo hydrophones; reusable components across multiple units are marked with an asterisk (*).

Component	Function	Price (€)
Piezoelectric disk (35 mm)	Acts as the acoustic element of the device	1.5 /unit
Audio cable (10 m)	Connects the PZ disk to the recording device	5 /unit
⌀60 mm plexiglass disk (4 mm)	Is attached to the PZ disk and acts as a lid to the hydrophone’s enclosure	3 /unit
PVC threaded nipple	Engulfs the acoustic element (housing)	0.85 /unit
Brass pipe fitting	Threaded on the PVC extension, it closes the recorder-side of the hydrophone and helps the audio cable exit the watertight enclosure. Additionally, its weight keeps the hydrophone submerged underwater	6 /unit
Heat shrink tubing *	Used for sealing and abrasion resistance at the point where the audio cable exits the enclosure	0.15 /m
Sikaflex-291i *	Used for sealing the front and back side of the hydrophone	12 /300 ml
Liquid teflon*	Used for sealing the connection between the brass fitting and the PVC thread	11 /100 ml
Teflon tape *	Used for sealing the connection between the brass fitting and the PVC thread	0.5 /10 m
Cyanoacrylate glue *	Used for attaching the PZ disk on the plexiglass disk, and for attaching the plexiglass disk on the PVC extension	2.5 /unit