Long-term acoustic data collection
Passive acoustic monitoring was conducted from November 19, 2018 to October 3, 2020, with 683 days of recording effort overall (Supplementary Table 2), using a High-frequency Acoustic Recording Package (HARP)37. The HARP was deployed in Bahía Norte, Guadalupe Island, located approximately 150 miles offshore of México’s Baja California Peninsula (Fig. 1). The HARP was bottom-mounted and deployed to a depth of approximately 1100 m, with a calibrated hydrophone suspended ~30 m above the seafloor. The same hydrophone was used for both deployments to facilitate data comparison. The omnidirectional hydrophone sensor (ITC-1042, International Transducer Corporation, Santa Barbara, CA) had an approximately flat (±3 dB) hydrophone sensitivity from 10 Hz to 100 kHz of −200 dB re V/μPa. The sensor was connected to a custom-built preamplifier board and bandpass filter. The calibrated system response was corrected for during analysis. Data were sampled continuously at a 200 kHz sampling rate with 16-bit quantization, effectively monitoring a frequency range of 10 Hz–100 kHz.
Automatic detection and manual classification of beaked whale echolocation clicks
Beaked whales can be acoustically identified by their echolocation clicks38. These signals are frequency-modulated (FM) upswept pulses, which appear to be species-specific and are distinguishable by their spectral and temporal features. Cuvier’s beaked whale echolocation signals are well differentiated from the acoustic signals of other beaked whale species. They are polycyclic with a characteristic FM pulse upsweep, peak frequency around 40 kHz, and uniform inter-pulse interval of about 0.4–0.5 s39,40. Additionally, Cuvier’s beaked whale FM pulses have characteristic spectral peaks at approximately 17 and 23 kHz.
Beaked whale FM pulses were detected in the HARP data with an automated method using the MATLAB-based (Mathworks, Natick, MA) custom software program Triton (https://github.com/MarineBioAcousticsRC/Triton) and other MATLAB custom routines. After all potential echolocation signals were identified with a Teager–Kaiser energy detector41,42, an expert system discriminated between delphinid clicks and beaked whale FM pulses. A decision about presence or absence of beaked whale signals was based on detections within a 75 s segment. Only segments with more than seven detections were used in further analysis. All echolocation signals with a peak and center frequency below 32 and 25 kHz, respectively, a duration less than 355 μs, and a sweep rate of <23 kHz/ms were deleted. If more than 13% of all initially detected echolocation signals remained after applying these criteria, the segment was classified to have beaked whale FM pulses. This threshold was chosen to obtain the best balance between missed and false detections. A third classification step, based on computer-assisted manual decisions by a trained analyst, labeled the automatically detected segments to signal type and rejected false detections38. The rate of missed segments was approximately 5%, varying slightly between deployments. The start and end time of each segment containing beaked whale signals was logged, and their durations were summed to calculate cumulative hourly and weekly presence.
Manual detection of anthropogenic signals
Anthropogenic sounds, including UA devices, shipboard echosounders, and vessel engine noise, were detected by manually screening the HARP recordings with the MATLAB-based program Triton, using long-term spectral averages (LTSAs) displaying one hour of data at a time. The full bandwidth LTSAs (displaying data up to 100 kHz) used to detect UA devices and echosounders were created using a 5 s time average with 100 Hz frequency resolution. The mid-frequency LTSAs (displaying data up to 5 kHz) used to identify vessel noise events resulted from a decimation of the data by a factor of 20, and were created using a 5 s time average with 10 Hz frequency resolution. The start and end times of each anthropogenic signal encounter were logged, and their durations were summed to calculate cumulative hourly and weekly presence.
Computation of soundscape metrics
Soundscape metrics of the full HARP time series were computed using the MATLAB-based Triton “Remora” (a type of plug-in) Soundscape Metrics (https://github.com/MarineBioAcousticsRC/Triton/tree/master/Remoras/Soundscape-Metrics). Periods of 15-s disk-write noise, occurring on a repeating cycle every 75 s, were omitted from analysis. The broadband (0–100 kHz) acoustic data were processed to compute power spectral density (PSD) using Welch’s method within MATLAB (FFT length = 200,000 points, Hann window length = 200,000, FFT overlap = 0%), resulting in mean-square pressure amplitude (µPa2) at 1-Hz, 1-s resolution. For every 1-Hz frequency band, PSD levels per 1-min were calculated as the median of mean-square pressure amplitude (µPa2) over no less than 30 s for that minute. PSD levels were then converted to decibels (dB re: 1 µPa2/Hz).
To facilitate comparison of soundscape metrics to anthropogenic signals, standard frequency bands (American National Standards Institute, ANSI 1.11-2004) were selected, and median one-third octave band sound pressure levels (TOLs) were calculated over the full time series. TOLs in units of µPa2 were calculated by integration of PSD estimates of the mean-square pressure (µPa2) with a 1-Hz, 1-s resolution over each of 39 one-third octave bands, with the nominal center frequencies ranging from 13 to 80,000 Hz (IEC 61260-1995). The resulting TOLs with 1-s resolution were then used to calculate TOLs per 1-min as a median over no less than 30 s for that minute. The TOLs per minute were converted to decibels (dB re: 1 µPa2). Band levels for a nominal frequency of 125 Hz were transformed to reported TOL values for comparability by adjusting for frequency band bin width (bw) by subtracting 10 × log10(bw) from band levels, resulting in dB re: 1 µPa2/Hz. TOL calculations centered on 125 Hz are often reported as an indicator of vessel noise, as the energy concentrated in this frequency range is representative of the continuous low-frequency sound pressure levels generated by boat engines43.
PSD levels per 1-min were calculated as the median of mean-square pressure amplitude (µPa2) over 200 Hz frequency bins and no less than 30 s for that minute. PSD levels were then converted to decibels (dB re: 1 µPa2/Hz). PSD levels for 19.3–19.5 kHz, 29.1–29.3 kHz, and 49.9–50.1 kHz were extracted to measure energy from specific anthropogenic sound sources, as these frequency ranges were representative of the UA signal, 28 kHz echosounders, and 50 kHz echosounders, respectively. These narrowband, 200 Hz wide frequency bins were chosen as proxies for these three anthropogenic sounds because more broadband measurements (e.g., TOLs) would also contain additional energy from other, non-target signals. Echosounders transmitting at 50 kHz produced pings with a peak frequency of 50 kHz and energy between approximately 49–51 kHz, and 28 kHz echosounder pings had a peak frequency of ~28.8 kHz and energy between approximately 27 and 31 kHz. PSD levels for 29.1–29.3 kHz were chosen as the proxy for 28 kHz echosounders to avoid any spectral overlap with an ultrasonic antifouling pulse containing energy from 28.4 to 29.0 kHz.
Cetacean visual surveys
Twelve field trips to Guadalupe Island were undertaken between May 2017 and October 2020 at various times throughout the year (Supplementary Table 3). During these expeditions, boat-based visual surveys for cetaceans were conducted, which entailed at least two observers scanning with the naked eye for beaked whales and other cetaceans during daylight hours. Information on all cetacean sightings was logged systematically, including details such as the species, group size, and GPS location. Cetacean research activities were conducted from large vessels (17–40 m boats) and/or pangas (~7 m skiffs). Some of these field trips coincided with the 2018 and 2019 shark cage diving tourism seasons, while others occurred when tourist boats were absent from the island, such as during the spring periods of 2017 and 2019, as well as during the fall of 2020 when the tourism season was canceled due to the COVID-19 pandemic. Visual survey tracklines and sighting locations of Cuvier’s beaked whales were plotted to create maps showing search effort and sighting distributions under various conditions related to the presence of tourist vessels (Fig. 1).
Opportunistic acoustic recordings
Opportunistic passive acoustic recordings were collected on several occasions near vessels assumed to be equipped with UA devices, using hand-deployed recorders at depths of approximately 2–7 m below the sea surface (Supplementary Table 5). In March 2021, acoustic data were collected approximately 500 m away from the Royal Princess cruise ship anchored in the Bay of La Paz, México. A Marantz PMD661 recorder (96 kHz sampling rate, 24-bit resolution) was used with an omnidirectional Reson TC4013.1 hydrophone (sensitivity −211 ± 3 dB re V/μPa, frequency response 1 Hz to 170 kHz, 50 kHz low-pass filter), and was programmed to record continuously at a 96 kHz sampling rate.
For all other opportunistic recordings, a SoundTrap® ST300HF recorder (Ocean Instruments, Auckland, New Zealand) was used to collect acoustic data at distances <300 m away from a variety of vessels. The ST300HF was programmed to record continuously at a 576 kHz sampling rate with a 400 Hz high-pass filter, for an effective frequency range of 400 Hz–276 kHz. With these settings, the instrument had an approximately flat frequency response (±3 dB) from 500 Hz to 150 kHz. In August 2021 at Guadalupe Island, acoustic data was collected with this recording system near the Nautilus Belle Amie, an anchored shark cage diving tourist vessel in Bahía Norte. This recording setup was also used to collect acoustic data on three separate occasions in October 2021 in San Diego Bay, California, USA near the berthed cruise ships Grand Princess, Majestic Princess, and Koningsdam, and was also used in the harbor of Ensenada, México to collect acoustic data near the berthed cruise ships Navigator of the Seas, Carnival Miracle, and Oceania Regatta in November 2021, December 2021, and January 2022, respectively. All opportunistic acoustic recordings were manually screened for the presence of UA signals by scanning 5 s spectrograms (1000-point FFT length, 0% overlap) of the full bandwidth data using the MATLAB-based program Triton.
Statistics and reproducibility
To evaluate the impact of the vessel-based anthropogenic sound sources on Cuvier’s beaked whale acoustic presence, statistical analyses were conducted in MATLAB. Hourly sums of Cuvier’s beaked whale detections (cumulative number of minutes per hour with echolocation clicks) were calculated over the complete HARP time series, and because these data were not normally distributed a non-parametric test was applied. To examine if the number of beaked whale detections per hour differed depending on the presence/absence of the various anthropogenic signals, a Kruskal–Wallis one-way analysis of variance test was computed for hourly sums of Cuvier’s beaked whale detections under the following noise conditions: (1) UA devices, echosounders, and vessel noise absent; (2) Vessel noise present; (3) Echosounders present; (4) Vessel noise and echosounders present; (5) UA devices present; (6) UA devices and vessel noise present; (7) UA devices and echosounders present; and (8) UA devices, echosounders, and vessel noise present. Post hoc multiple comparisons were then made using the “multcompare” function in MATLAB with Bonferroni correction to determine pairwise differences between the conditions. P values < 0.05 were considered statistically significant.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
