Data collection
Between June 27 and August 25, 2016, 6600 km of simultaneous visual and passive acoustic line transect surveys were completed on the National Oceanic and Atmospheric Administration (NOAA) ship Henry B. Bigelow5. Survey effort was distributed along saw tooth track lines spanning the continental slope from Virginia (US) to the southern tip of Nova Scotia (Canada) (36–42 N) and on several larger track lines over the abyssal plain. Two teams of visual observers independently recorded sightings of marine mammals using high-powered Fujinon binoculars (25 × 150; Fujifilm, Valhalla, NY) as well as environmental conditions (e.g. sea state) every 30 min.
The speed of sound in water was collected three times each day (morning, noon, evening) by measuring conductivity, temperature, and depth (CTD) at specific intervals in the water column. The sound speed closest to the depth of the towed hydrophone array was extracted. On alternating survey days, Simrad EK60 single beam scientific echosounders operating at frequencies of 18, 38, 70, 120 and 200 kHz were used to collect active acoustic data.
When possible during daylight hours (06:00–18:00 ET), passive acoustic data were collected continuously using a custom-built linear array composed of eight hydrophone elements and a depth sensor (Keller America Inc. PA7FLE, Newport News, VA) within two oil-filled modular sections separated by 30 m of cable (Fig. 1). The array was towed 300 m behind the vessel at approximately 5–10 m depth while the vessel was in waters more than 100 m deep and underway at speeds of 16–20 km/h. For more details see DeAngelis et al.31, with the only change being that two APC hydrophones and one Reson hydrophone in the aft section were replaced with HTI-96-Min hydrophones (High Tech, Inc., Long Beach, MS). The HTI’s had a flat frequency response from 1 to 30 kHz (− 167 dB re V/uPa ± 1.5 dB). Recordings were made using the acoustical software PAMGuard (v.1.15.02)34. This analysis used the data recorded by the last two 192 kHz sampled hydrophones in the array (MF5 and MF6).
The linear towed array included eight hydrophone elements and a depth sensor within two oil-filled modular sections separated by 30 m of cable. Six hydrophones sampled at 192 kHz (MF1–MF6) and two sampled at 500 kHz. The hydrophones were connected to two National Instruments sound cards (NI-USB-6356). A high pass filter of 1 kHz was applied by the recording system to reduce the amount of vessel noise in the recordings. This analysis used the passive acoustic data from MF5 and MF6. The schematic is not to scale.
Click detection and 2D event localization
The passive acoustic data were filtered using a Butterworth band pass filter (4th order) between 2 and 20 kHz and decimated to 96 kHz to improve sperm whale click resolution. Clicks were automatically detected using the PAMGuard (v.2.01.03) general sperm whale click detector with a trigger threshold of 12 dB.
Using PAMGuard’s bearing time display, all detections were reviewed to classify click types and mark click trains as “events” based on consistent changes in bearing, audible sound, ICI and spectral characteristics. Each event was marked to an individual level, tracking a whale from the first to the last detected click15,35. All events containing usual clicks were localized with PAMGuard’s Target Motion Analysis (TMA) module’s 2D simplex optimization algorithm. For further analysis, events were truncated at a slant range of 6500 m (Supplementary Fig. S1).
Echosounder analysis
A regression analysis was run using the R package MASS36. To account for overdispersion, a negative binomial generalized linear model (GLM) with a log link function was applied to a dataset of the daily acoustic detections33. Echosounder state (active versus passive), month (June, July, August), and habitat type (slope or abyssal) were included as covariates, with the total number of daily detections as the response variable. The track line distance covered per day was used as an offset for effort. The best fitting model was selected based on backwards stepwise selection using Akaike’s information criterion (AIC) and the single-term deletion method using Chi-squared goodness-of-fit tests37.
3D localization
Extracting a .wav clip for each click and attributing metadata
An automated process was developed using the R package PAMpal38 (v. 0.14.0) to extract the time of each click in the marked events from PAMGuard databases, generate a .wav clip for each click, and attribute all metadata (e.g., event 2D localization, array depth, radial distance, sea state, and sound speed) necessary for estimating the click depth.
Slant delay
Using the methods established by DeAngelis et al.31 and custom Matlab R2021a (MathWorks Inc., Natick, NA) scripts, the multipath arrival of clicks and surface reflected echoes were used to mathematically convert the linear array into a 2D planar array and estimate 3D localizations. Using the .wav clips exported from PAMPal, the time delay between the click and the corresponding surface reflected echo, known as the slant delay, was measured via autocorrelation. Within the autocorrelation solution’s envelope of correlation values, the optimal slant delay was measured using the peak with the highest correlation value above a threshold of 0.02 and within an expected time window after the direct click of 0.0005–0.015 s. Although theoretically a surface reflected echo could have arrived less than a millisecond (< 0.00006 s) after a sperm whale’s click, a minimum time of 0.0005 s was applied because otherwise the autocorrelation solution consistently selected a peak within the direct click. In addition, as a result of the multi-pulse structure of sperm whale clicks11,15, the autocorrelation solution can select a peak from a second or third pulse of the direct click if it has a larger amplitude than the peak of the surface reflected echo. The inter-pulse interval (IPI) of a direct click can be consistent over short periods of time39, while the time between the direct click and surface echo can vary as the whale moves up or down in the water column. These patterns can occasionally be visualized and distinguished using an echogram (Fig. 2). Therefore, for each event a plot of the measured slant delay over event time was manually reviewed for expected patterns, and probable false detections were removed.
(a) Echogram illustrating the energy recorded in the 0.015 s after each of 2425 clicks in an event. The vertical axis includes 144 samples at a decimated frequency of 9600 Hz (0.015 s). The P1 and P2 pulses of the direct click are visible as are the surface reflected echoes (SRE) of these pulses. (b) Plot of the measured slant delay (s) over the click time (HH:MM). Slant delay increases to ~ 0.01 s and then decreases over a time period of approximately 34.5 min. Some of the false measurements of slant delay, not consistent with the pattern, are marked.
Average depth and standard deviation
For each click with a measured slant delay a click depth was estimated by combining the slant delay with the required metadata31. Click depth values were truncated at 4000 m as any greater measurements were likely based on false detections of the surface reflected echo. For each event, the click depths were used to calculate an average depth. The 3D localization was rejected if the average depth exceeded the seafloor depth40 where the event was recorded. A standard deviation was calculated based on the combined uncertainties from the slant delay, 2D localization, and depth measurement while adjusting for the number of clicks in an event.
Dive depth patterns
For each event with a 3D localization, patterns in the click depths over time were manually reviewed and categorized. If a descent, bottom phase and ascent were present in the click depths (Supplementary Fig. S2), longer events (> 5 min) were categorized as U shaped, and as shallow (< 800 m), medium (800–1600 m), or deep (> 1600 m) based on the maximum click depth (Fig. 3).
Example of click depths (m) over time (min) for events categorized as (a) U shaped and shallow (< 800 m), (b) U shaped and medium (800–1600 m), (c) U shaped and deep (> 1600 m).
Click depths were then binned at 400 m intervals to account for an animal’s unknown horizontal movement over time as well as uncertainty in the estimated click depths, and the total time an animal spent within each depth bin was calculated. For each event with a U shaped click depth pattern, the depth bin in which the bottom phase occurred6 was determined. Finally, to assess if a whale was diving in the water column or close to the seafloor, the depth bin in which the 90th percentile of the click depths was recorded was compared to the bin including the seafloor depth. If the whale was more than 400 m above the seafloor, it was determined to be diving in the water column.
Distance sampling
Depth-corrected average horizontal perpendicular distances
For each event, a depth-corrected average horizontal perpendicular distance was calculated using the TMA derived perpendicular slant range and the average depth or an assumed depth in the Pythagorean theorem19,31. The weighted mean, first quartile, and third quartile of the average depths were tested as assumed depths for events excluded from 3D localization. If depth was greater than or equal to the slant range the perpendicular distance was coerced to 0, indicating the whale was diving directly below the track line. The resulting distribution of depth-corrected perpendicular distances that aligned most with distance sampling theory was used in the final distance analysis.
Acoustic density and abundance estimation
The R package Distance41,42 (v.1.0.4) was used to estimate two separate detection functions based on the uncorrected slant ranges and the depth-corrected perpendicular distances. Half-normal, uniform, and hazard rate key functions were tested with cosine, simple polynomial, and Hermite polynomial adjustment terms. The best fitting models were selected based on the AIC, the Kolmogorov–Smirnov (K–S) test, the Cramer-von Mises (CvM) test, quantile–quantile plots, and visual review of the fitted models43. The probability of detection, abundance, and effective strip half width (ESW) were then estimated for foraging sperm whales.
Permitting authority
Data used in this manuscript were collected during surveys that were completed under U.S. Marine Mammal Protection Act permit numbers 17355 and 21371 issued to the Northeast Fisheries Science Center.
Source: Ecology - nature.com
