Background & Summary

Underwater ambient sound levels are a critical component for the health of the marine ecosystems. Marine organisms are evolved to get relevant information by listening to the soundscape, whose acoustic signature reveals the occurrence of natural events and vocal species1,2. In this context, the input of underwater noise induced by human activities has been linked to detrimental effects on marine fauna3,4,5,6,7. As a result, the anthropogenic underwater noise has been recognised as a pollutant of international concern and has been addressed by international agreements8. The U.S. National Oceanic and Atmospheric Administration’s Ocean Noise Strategy (ONS), for example, focuses on the evaluation and management of the human-generated noise and its effect on marine species, supporting the goals of the U.S. National Ocean Policy9. Further, the European Union’s Marine Strategy Framework Directive (MSFD) requires the EU member states to monitor and mitigate noise pollution to reach a “Good Environmental Status” of the marine environment. Setting up monitoring cross-border programmes aiming to evaluate the underwater sound levels at sub-regional scale is recommended in the MSFD context (EU Directive 2008/56/EC).

Global efforts on monitoring underwater sound levels resulted in long-term projects dedicated to target areas10, including, among others, the US Outer Continental Shelf and the US coastal waters (ADEON, NOAA CetSound Project, respectively), the British Columbia, the Vancouver Port waters, the Canadian Atlantic coast waters and the Gulf of St. Lawrence (ECHO program11, ESRF and SeaWays projects, respectively). Underwater soundscapes have been investigated also in Australia12,13, Eastern, Southern and South East Asia14 and South Africa15 waters, as well as in Artic16 and Antartic17 waters.

Continuous sound monitoring EU projects have been established in the Northeast Atlantic (JOMOPANS and JONAS), in the Baltic Sea (BIAS) and in waters between Scotland and Ireland (COMPASS)18,19,20,21. In contrast, no extensive research on the underwater sound continuous levels has been developed in the Mediterranean Sea so far. Pilot monitoring studies have been run in the context of the EU QUIETMED project22 together with few other local studies23,24,25, including some done in the Adriatic Sea26,27,28,29. The EU Interreg Italy-Croatia project SOUNDSCAPE (Soundscapes in the North Adriatic Sea and their impact on marine biological resources) has been therefore established to implement a shared monitoring network for a coordinated transnational assessment of the underwater ambient sound in the North Adriatic Sea (NAS). The Adriatic has been recognized as one of the important sub-regions of the Mediterranean Sea by the MSFD; the NAS is its shallowest, northernmost part. Most of NAS is considered to be an Ecologically and Biologically Significant Area (EBSA, Convention on Biological Diversity), as well as hosting several marine and coastal Natura 2000 sites, and protected areas30. Whilst having a very vulnerable biodiversity31, NAS is highly impacted by increasing maritime traffic, tourism and resource exploitation32. As a result, NAS biota is currently under the combined pressure of the anthropogenic impact33 and climate change34,35.

The SOUNDSCAPE dataset presented in this paper contains 20 and 60 seconds averaged sound pressure levels (SPLs) collected at nine monitoring stations, from the Gulf of Trieste till about the Middle Adriatic Pit (Fig. 1; Table 1), from March 2020 to June 2021. This dataset is essential for establishing baselines that document acoustic conditions over time on the regional scale and represents the first dataset of this kind in the Mediterranean Sea. The collected data are crucial to assess the ecosystem health, to evaluate the consequences of new possible marine development and to promote a knowledge-based management of the marine resources. Additionally, the SOUNDSCAPE dataset includes the most restrictive COVID-19-induced lockdown phase (March–April 2020), providing a unique benchmark for spatial and temporal comparative analysis.

Fig. 1
figure 1

Soundscape recording stations in the Northern Adriatic Sea; vessel traffic (a) and bathymetry (b) are highlighted. Vessel traffic is represented as total number of vessel passages in 2020, obtained from EMODnet Human Activities, Vessel Density Map. (revision date 2022-03-21).

Table 1 Coordinates, bottom depth and sediment types of the recording locations.


The workflow shown in Fig. 2 summarizes the steps undertaken to obtain the SPL datasets from the underwater noise raw data collected in the field. The workflow entails two main blocks: (i) “Data Acquisition” describes the process of sound recording and wav files uploading on the SOUNDSCAPE-dedicated server to store the data; (ii) “Data Processing” shows the steps that lead to the processing of wav data to calculate Sound Pressure Level data.

Fig. 2
figure 2

Workflow of the acquisition and processing of underwater noise data to obtain SPL20,60 dataset.

The applied procedures are in accord to guidelines developed by other international projects or agreements10,36. Used terminology followed ISO 1840537, IEC 61260-1:201438 and JOMOPANS39 Terminology Standards and it is summarized in Table 2.

Table 2 Used acoustic terminology.

Hardware components and calibration

The acoustic recordings were made by using autonomous passive underwater acoustic recorders (APUARs; Sono.Vault by Develogic Subsea Systems GmbH, Hamburg, Germany). Each recorder was featured with an omnidirectional broadband Neptune Sonar D60 Hydrophone characterized by a sensitivity around −192.7 dB re 1V/µPa (flat frequency response: 10 Hz – 20 kHz ± 3 dB). The processing chain includes a high-pass filter (cut-off frequency in the range of 3–10 Hz), a preamplifier and a 16 bit analogue to digital converter (ADC). The 16-Bit ADC has a high frequency reject filter with 500 kHz and it is otherwise limited by the input amplifier which has a bandwidth of approximately 100 kHz. The ADC is the last component in the processing chain. The data that is stored comes directly from the ADC.

The waterproof pressure resistant housing contained a programmable recorder with variable gain, a battery set consisting of lithium D-Cells and up to 1TB-SD memory cards.

Hydrophone calibration was achieved by the manufacturer using a calibrated reference projector; the reference projector was calibrated as well using a free-field three-transducer reciprocity calibration. Both procedures are compliant with the IEC 60565-1:2020 international standard40. The recorders were set to record continuously at a sampling rate of 48 kHz, providing a recording bandwidth of approximately 22 kHz with 16-bit resolution.

Additional information and hydrophone sensitivity curves in the pertinent frequency range are available in the SOUNDSCAPE Deliverables 3.2.141 and 3.6.342.

Acoustic data acquisition

A total of nine APUARs were deployed (Fig. 1). In most of the stations, the recorders were anchored to the bottom with a rig design consisting of an anchor, an acoustic releaser, the logger itself secured by polypropylene rope and extra flotations (e.g., sphere with the diameter of 25 cm mounted at minimum of 100 cm from the hydrophone), as illustrated in Fig. 3. The rig design above the anchor was positively buoyant. This ensured that the loggers were suspended about 3 m above the seabed throughout the deployment. In the stations MS1, MS5 and MS6, deployment and recovery operations were carried out by scuba divers, so no acoustic release was needed.

Fig. 3
figure 3

Sketch of the SOUNDSCAPE standard rig deployed on the seafloor, with hydrophone set at ~3 m above the seafloor (range from 2 to 6 m).

The system calibration was checked in situ just before the deployment and after the recovery by using an air-pistonphone Grass 42AC (Grass Instruments, West Warwick, RI, USA), that generates a known sound pressure level at 250 Hz. Additionally, profiles of water conductivity, temperature and pressure were recorded by using a CTD probe. Metadata were collected for each deployment and recovery, including name, geographic position and the depth of the measurement site, start and stop time for each recording, equipment ID number and set up data, calibration data and weather conditions. Additional details on the deployment and recovery protocols are available in the SOUNDSCAPE Deliverable 3.2.243. Typical measurement duration for the stations was 3 months, after which each device needed to be recovered to download data and to remove biological fouling. The measurement period covered about one full year and four months (from 1 March 2020 to 30 June 2021). Table 3 shows the data coverage for each monitoring station.

Table 3 Data temporal coverage (days per month) for each recording station.

Acoustic data storage and processing

The collected .wav files were stored on two servers at CNR ISMAR (Venice, Italy) and IOR (Split, Croatia). No data compression was applied to the original files. The whole data-set has been processed by the same processing executable tool, that was developed specially for the SOUNDSCAPE project by the University of Gdansk together with CNR-ISMAR.

The processing steps are briefly summarized:

  1. (i)

    each 1-sec segment is read from the wav file (i.e. 48000 values, being the Sample Rate equal to 48000 Hz) and a discrete Fourier Transform is applied;

  2. (ii)

    the power within one third-octave (base 10) band38, U(F), is calculated as

    $$U\left(F\right)=\frac{1}{{N}^{2}}{\sum }_{{b}_{1}}^{{b}_{2}}{\left|A\right|}^{2}$$

    where N is the number of samples, A are the coefficients in the discrete Fourier transform and b1 and b2 are the indices corresponding to the lower and upper frequencies of a given one-third octave band;

  3. (iii)

    the SPL (Lp) averaged over 1 second (hereafter SPLs1 dB re 1 μPa) is obtained as

    $${L}_{p}(F)=10{\cdot {\log }}_{10}(U(F))-{S}_{Dev}(F)$$

    where F refers to each one third-octave (base 10) frequency band38 and SDev is a factor related to the Develogic Sono.Vault hydrophone sensitivity, the recording gain and calibration process;

  4. (iv)

    20 and 60 seconds averaged SPLs (hereafter SPLs20 and SPLs60) are then calculated from 1 second averaged SPLs (SPLs1) by using the following Eq. (3):

    $$SP{L}_{n}=10\cdot \,lo{g}_{10}\left(\frac{1}{n}{\sum }_{i=1}^{n}1{0}^{\frac{SPL{s}_{1i}}{10dB}}\right)dB,$$

    for each SPLs1i with n = 20 or 60;

  5. (v)

    output data of SPLs1, SPLs20 and SPLs60 are produced.

Specifically, the factor SDev in formula (2) is computed by the following formula (4) based on the information provided by the Develogic Sono.Vault manufacturer.


Where SH (F) (dB/V) is the sensitivity of the hydrophone for each one-third octave frequency band as extracted from the calibration sheet (Table 4), LUcal (dB/V) is introduced to take into account the recording gain of the APUAR, K is a constant value, being equal to 49.0309 related to the signal used by manufacturer Develogic during the calibration process (see SOUNDSCAPE report41 for details).

Table 4 Sensitivity (dB/V ref 1 μPa) of the hydrophone SH(F) for each one-third octave frequency F extracted from the calibration sheet of the manufacturer Develogic.

Data Records

The dataset of 20 and 60 seconds averaged Sound Pressure Levels (SPL) output files collected by SOUNDSCAPE and described in this paper is available on Zenodo44.

Data are archived using structured HDF5 files, each one containing metadata and SPL data according to the ICES (International Council for the Exploration of the Sea) continuous noise data portal specification, with time stamps relative to UTC time provided in compliance to ISO 860145.

Technical Validation

In order to ensure the data quality, a check on the collected data by times series visualisation was carried before the data analysis: it did not highlight spurious signals or transient artefacts due to deployment settings, nor systematic artefacts due to flow noise, which is consistent with the study areas being characterized by low tidal currents. Moreover, data recorded before and during the deployment, during and after the retrieval and while the deployment vessel was in close proximity of the recorder were removed.

The measured data may be contaminated by the system self-noise. Self-noise fluctuations in output of an acoustic receiver system are caused by the combination of acoustic self-noise and non-acoustic self-noise (electronic self-noise). The acoustic self-noise sound is usually caused by the deployment, operation, or recovery of a specified receiver and its associated with the deployment of the acoustic sensor and platform (e.g., noise from moorings and fixtures, flow noise, etc.) whereas the non-acoustic self-noise is related to fluctuations in the output of a receiver system in absence of any sound pressure input37. In the SOUNDSCAPE project, the introduction of unwanted acoustic self-noise in the recordings was prevented by deployment rig’s design and deployment procedure. Attention was given (i) in the mooring settings to minimize the self-noise (i.e., use of soft ropes and avoidance of the metal parts) and (ii) in the positioning of the deployments, by locating them at a distance from the coast that assured no interaction with external infrastructures that could generate unwanted sounds. Further, the monitoring stations were not sited in high tidal flow areas and hydrophones were placed close to the bottom. The SOUNDSCAPE non-acoustic self-noise due to the electrical noise is calculated to be better than 58 dB re 1μPa2/Hz at 63 Hz and better than 53 dB re 1 μPa2/Hz at 125 Hz, according to the manufacturer technical specifications.

Finally, a quality control has been applied to the software used for the analysis (Fig. 2). To validate the correct functioning of the SOUNDSCAPE processing app (ANP) applied to the wav data, the latter was tested again other already validated software (SpectraPLUS and independent MATLAB routines). A subset of data was processed with the SOUNDSCAPE ANP and the SPLs of each one-third octave (base 10) band were compared with the ones calculated by a validated tool: the SOUNDSCAPE ANP was able to reproduce almost the same results. Namely, the mean absolute difference between trusted tool and SOUNDSCAPE ANP results was equal to 0.08 dB, being less than 0.1 dB in most of the frequencies, with the exception of the lower frequencies (less than 25 Hz), where it was lower than 0.3 dB.

Usage Notes

To post process the SPLs20,60 data, CNR-ISMAR developed a Python script that was deployed as a Jupyter Notebook interactive document46, that is here released.

The workflow of SPL data post processing is simple. After reading SPLs20,60 files and (i) selecting a time window to define the investigated period, (ii) a recording station and (iii) a given one-third octave (base 10) band, it is possible to compute some metrics to create tables and to visualize efficiently the data (see Fig. 4 for examples). Statistics can be calculated for each one-third octave (base 10) band over various temporal analysis windows (based on UTC time) like for example one hour, one day, one month, one year10. Once the time window is selected, tables with descriptive statistics can be produced including percentile values (1th, 10th, 25th, 50th, 75th, 90th, 99thpercentiles) and the arithmetic mean. The Python script can also generate graphs such as time series plots, to visualize the temporal evolution of SPL data, and descriptive plots, to highlight the principal statistics of the data distribution over the time window. Data can be aggregated also to check their distribution between stations.

Fig. 4
figure 4

Examples of SPL20,60 data post processing outputs generated applying the Python post processing script to the released dataset.