IMIS

Between April 2020 and October 2020 and in June 2021, we recorded underwater sound at different strategic points of the Belgian Part of the North Sea (BPNS). All the recordings were acquired from a drifting small boat with a hydrophone attached to a rope with weights. The length of the rope was chosen according to the depth, so that the hydrophone would be in average more or less between the 1/2 and 1/3 of the top water column. The exact depth was not possible to know real-time because the plotter was turned off, so the rope length was kept constant during each entire deployment. In this manuscript, we consider a deployment the data corresponding to the time when a hydrophone is in the water without changing any recording parameter. Three different boats were used (RIB Zeekat, Sailing boat Capoeira and working boat from RV Simon Stevin). Each of these recording consists of 30 to 60 minutes of continuous recording following the current by drifting with the engines and the plotter turned off. Drifting was chosen as an ecologically meaningful approach to measure coastal benthic habitats (Lillis et al., 2018) and spatial resolution was chosen over temporal resolution considering the available ship time and equipment. The locations were chosen to cover the 5 habitat types defined in (Derous et al., 2007) as well as some shipwreck areas to capture their specific soundscapes. The objective was to acquire short recordings above different shipwrecks which would give information about acoustic spatial distribution. The recordings were acquired while drifting to diminish the possible flow noise due to the current. The instruments used where a SoundTrap ST300HF (sensitivity -172.8 dB re 1 V/uPa, from now on, SoundTrap) and a Bruel & Kjaer Nexus 2690 and a hydrophone type 8104 (sensitivity -205 dB re 1 V/uPa, from now on, B&K) together with a DR-680 TASCAM recorder. The amplification in the Nexus was set to 10 mV, 3.16 mV or 1 mV, depending on the loudness of the recording location. The SoundTrap was set to sample at 560 kS/s and the B&K at 192 kS/s. To adjust for the different amplifications, at the beginning of each recording a calibration tone was performed. During each deployment, a GPS Garmin with a time resolution of 1s was synchronized with the instrument clock and stored the location during the entire deployment. A total of 54 different deployments were acquired. 14 of the sites were acquired simultaneously with the two recorders to confirm their interchangeability and assess their differences, so a total of 40 independent tracks were recorded. The metadata were stored in the Underwater Acoustics part of the European Tracking Network (ETN, https://www.lifewatch.be/etn/). Because acoustic changes in the soundscape were expected to be found in a small spatial (several meters) and temporal (several seconds) resolution, the data was processed in time windows of 2.5 seconds, overlapping 60%. The time window was chosen so the spatial resolution would be of 2 samples every 5 m, considering an average drifting speed of 1 m/s. All the acoustic processing was done using pypam (https://lifewatch-pypam.readthedocs.io/en/latest/). Each recorded file was converted to sound pressure using the calibration given by the manufacturer. For files with a calibration tone, the calibration value was computed from the tone and then the calibration signal was removed from the file. The rest of the data was processed according to the obtained calibration value. The sound pressure values obtained by the two instruments recording simultaneously where compared to make sure the calibrations where accurate. Per time window, first the Direct Current (DC) noise was subtracted by computing and subtracting the mean of the signal. Then a Butterworth band pass filter of order 4 was applied between 10 and 20,000 Hz, and the signal was down-sampled to twice the high limit of the filtered band. Once the signal of each time window was filtered to the desired bandwidth, the root mean squared value of the sound pressure of each one-third octave band (base 2) was computed per each 1 second-bin. This was done using a Butterworth band pass filter of order 2 per each one-third octave band, which resulted in 5 x 29 one-third octave bands per time window. Each acoustic sample was considered then to be these 5 consecutive one-third octave bands, with a total dimension of 5 x 29. A part from the acoustic features, each sample also stored a timestamp and the deployment metadata (instrument, end-to-end calibration, file path, hydrophone depth). The data are stored in one netCDF file per deployment, directly stored from pypam. Each netCDF file incorporated the deployment metadata together with the processed one-third octave bands. These files are stored in the deployments folder. The location of each expedition was stored in a gpx, which can be found under the gps/ folder. There is a data_summary_mda.csv file which summarizes the metadata for each deployment and links the deployment to the corresponding gpx file. Furthermore, some of the data were manually annotated for sound artifacts. These annotations are stored in the labels.csv file.