JASA

https:/doi.org/10.1121/10.0043324

Statistical analysis using generalized additive models
‘GAMSs) and random forest (RF) models (Sec. IIE).

A. Measurement setup
Selecting an appropriate measurement location was crucial
 to maximize the number of recorded CTV passages. To
identify a suitable site, automatic identification system
(AIS) data from the JOMOPANS Project (2020) were analyzed,
 focusing on tracks of vessels classified as “high-speed
orafts.”
Helgoland serves as a key departure point for CTVs,
servicing four offshore wind farms to its north: Amrumbank
West, Kaskasi II, Nordsee Ost, and Meerwind Süd/Ost.
Siven its strategic position and high vessel traffic, a measurement
 site was chosen within the northern approach fairway
 to Helgoland.
To minimize the risk of equipment loss due to fishing
activity, additional AIS data from the JOMOPANS Project
were reviewed to avoid high-density fishing areas. The
selected measurement location is shown in Fig. 1. Two
hydrophones were deployed at this site, positioned 60m
apart to ensure data redundancy. Recordings were collected
:ontinuously from June 24 to October 08, 2024. The black
lines in Fig. 1 indicate the AIS-tracked CTV passages
recorded during this period.
The recorders were installed in custom-made cages that
were held vertically in the water column with floating
buoys. Both recorders were measuring for 55min every
hour, ensuring sufficient time for the recorder to properly
save the audio files in WAV format. The sensitivity of the
tecorders was tested prior to deployment using a custommade
 acoustic calibrator at 125 Hz (IEC, 2019). The sample
rate was set to 96 kHz, enabling an analysis of decidecade
bands up to 40kHz. Table I provides an overview of the

8'

° 54°N
3 15.00’
5

127

Longitude

8°E

FIG. 1. Tracks of crew transfer vessels around Helgoland (black) and other
marine traffic (orange) between June and October 2024.

J. Acoust. Soc. Am. 159 (4), April 2026

TABLE I. Overview of measurement location and used equipment.

Parameter Recorder 1 Recorder 2
Latitude 54.23°N 54.23°N
Longitude 7.82°E 7.82°E
Water depth 24.8 m 24.5 m
Recorder type Sound Trap ST600 Sound Trap ST600
Sampling rate 96 kHz 96 kHz
Duty cycle 55 min/1h 55 min/1h
Sensitivity —177.7dB re 1 V/uPa —177.9dB re 1 V/uPa

location of the measurement, the equipment used, and the
settings chosen.

B. Data collection and preprocessing
The raw data files were quality controlled following the
recommended practice from Ward et al. (2021). Then all
recordings were processed to compute 1-s sound pressure
levels (SPLs) for all decidecade frequency bands between
LOHz and 40kHz using the tool BSOUNDH, developed by
Fraunhofer IDMT (Oldenburg, Germany) during the BSH
SOUND Mapping Project (2021). The tool uses an integrated
 filter bank and was tested to conform with the common
 standards (IEC, 2014, 2016).
C. Identification of CTV passages
AIS data, sourced from the European Maritime Safety
Agency (EMSA), were analyzed to identify vessel passages
where the distance of the closest point of approach (DCPA)
to the hydrophone was less than 1 km (mean DCPA 429m,
standard deviation 229 m). Only passages without other vessels
 within 2 x DCPA were considered, resulting in the
identification of 3243 passages, of which 2526 were CTV
passages. To correlate these passages with the acoustic data,
peaks in the broadband SPL time series were identified
within +2 min of the time of the closest point of approach
(TCPA) for all passages. Only peaks with a minimum prominence
 of 6dB above their local acoustic background were
considered, where prominence is defined as the level difference
 between a peak and the lowest surrounding local minimum.
 All 810 preselected peaks were then manually
validated using a custom MATLAB interface (see Fig. 2). The
closest point of approach (CPA) for each passage was determined
 by examining the narrowband spectrogram up to
1000 Hz. Specifically, the minimum of the characteristic
U-shaped interference pattern—known as the Lloyd’s mirror
effect, which results from the interference between direct
and surface-reflected sound paths—was manually identified
(Urick, 1983). However, manual selection of these minima
was complicated by uncertainties that may arise from
frequency-dependent effects, the directional characteristics
of the sound source, or its relative position and motion.
The shift between manually validated TCPAs and
TCPAs from the AIS data reveals a constantly increasing
time drift between the recorders and the AIS data. Both
recorders exhibit a similar clock drift (Fig. 3). A linear

Basan et al. 3407