Acoustic and visual cetacean surveys reveal year-round spatial and temporal distributions for multiple species in northern British Columbia, Canada

Cetaceans spend most of their time below the surface of the sea, highlighting the importance of passive acoustic monitoring as a tool to facilitate understanding and mapping their year-round spatial and temporal distributions. To increase our limited knowledge of cetacean acoustic detection patterns for the east and west coasts of Gwaii Haanas, a remote protected area on Haida Gwaii, BC, Canada, acoustic datasets recorded off SG̱ang Gwaay (Sep 2009–May 2011), Gowgaia Slope (Jul 2017–Jul 2019), and Ramsay Island (Aug 2018–Aug 2019) were analyzed. Comparing overlapping periods of visual surveys and acoustic monitoring confirmed presence of 12 cetacean species/species groups within the study region. Seasonal patterns were identified for blue, fin, humpback, grey and sperm whale acoustic signals. Killer whale and delphinid acoustic signals occurred year-round on both coasts of Haida Gwaii and showed strong diel variation. Cuvier’s, Baird’s, beaked whale and porpoise clicks, were identified in high-frequency recordings on the west coast. Correlations between environmental factors, chlorophyll-a and sea surface temperature, and cetacean acoustic occurrence off Gwaii Haanas were also examined. This study is the first to acoustically monitor Gwaii Haanas waters for an extended continuous period and therefore serves as a baseline from which to monitor future changes.


.1 Cetacean call detections and validation
Information regarding cetacean species that frequent the waters of Gwaii Haanas, their status under the Species at Risk Act (SARA) and the Committee on the Status of Endangered Wildlife in Canada (COSEWIC), and their sounds targeted by the manual and automated analysis are provided in Table S1. Table S1. Cetacean species off Gwaii Haanas waters, their status under the Species at Risk Act (SARA) and the Committee on the Status of Endangered Wildlife in Canada (COSEWIC), and their sounds targeted by the manual and automated analysis.

Species Designation Vocalization types Reference
Blue whale Endangered (COSEWIC, SARA) A and B vocalizations 1 Fin whale Threatened (COSEWIC, SARA) 20 Hz and down-sweeps (40 Hz) Blue whales also produce D calls that are shorter than the A and B calls (<5 s), have sharp sweep rates between 95 Hz and 45 Hz, and are produced by both males and females 25,26 . Fin whale vocalizations most often consist of long, patterned sequences of 20 Hz notes 27 , but also of less frequent downsweeps in the higher frequency range (90 to 40 Hz) 28 . Humpback whales produce famous "songs" 29 as well as other vocalizations and percussive sounds, referred to as "social sounds" 30 . In BC, humpback whale tonal bubble feeding calls have also been documented 31 . Gray whales produce a variety of calls, including pulses and bonging signals, low-frequency moans, grunts, and subsurface exhalations, concentrated below 1500 Hz 32 . While most studies on gray whale sounds have been conducted in calving/breeding lagoons, recent studies showed that gray whales are also highly soniferous off BC while migrating (e.g., 8 ). In the Pacific Ocean, minke whales have been recorded making "boing" sounds 4 . Both north Pacific right whales (NPRW) and sei whales produce highly stereotyped and unique calls, up-calls 7 and downsweeps 33 , respectively, that are used for acoustic detection. NPRW also produce impulsive gunshot calls that, given their basic structure and non-uniqueness, are harder to use for PAM studies 34 .
Killer whales produce three types of sounds: clicks, whistles, and pulsed calls. Clicks are short pulses that usually occur in a series, also called click trains. Click durations range from 0.1 to 25 ms 10 . Whistles are single narrow-band tones in the 1.5 to 18.0 kHz frequency band. They have little harmonic structure and their duration ranges from 50 ms to 12 s 10 . Pulsed calls are harmonically structured with frequency ranges between 80 Hz and 12 kHz 35 . Echolocation signals from beaked whales are mostly frequency-modulated upsweep pulses, which appear to be species-specific and distinguishable by their spectral and temporal features 18 . Cuvier's beaked whales produce clicks with a centroid frequency of ~40 kHz that often had a characteristic C-shaped contour (Figure 3; 20 and 18 ). Clicks believed to be produce by Baird's beaked whales were sloped (characteristic of beaked whale clicks, Figure 3) with a centroid frequency of 20-28 kHz and an inter-click interval of 0. Automated analysis consisted of running three different types of detectors to identify cetacean calls based on known acoustic characteristics of calls by species or group of species. Blue, fin, and minke whales were detected using a spectrogram correlation approach based on 36 using time-frequency masks corresponding to the calls found in the North Pacific. The detector was configured to detect blue whale A, B and D calls, fin whale 20Hz pulses and 40 Hz downsweeps, and minke whale boing sounds. Humpback and killer whales were detected using the random forest-based classification approach described in 37 and 38 . Whistles and clicks from other odontocetes were detected using the general purpose contour detector described in 39 .
Manual analysis was performed in two ways. The first approach consisted of performing a 2.5% systematic manual analysis for all deployments and each channel (Supplementary Table 2). The systematic review resulted in 36 min per day (2.5% of mins in a day) being reviewed for each data set. The analysts visually and aurally examined each selected recording snippet using spectrograms created in the PAMlab software (JASCO Applied Sciences). Annotations were created to indicate marine mammal presence in each snippet. A lack of annotations was indicative of species absence within the detection range or species not vocalizing. The second manual analysis approach, described in 40 , was a manual review focused on the 1% of acoustic data files that had automatic marine mammal detections. All files selected for manual validation were reviewed aurally and visually using spectrograms in PAMlab. This second approach was specifically used to optimize the parameters of the automatic detectors (i.e., confidence thresholds and minimum number of detections accepted per file), and to calculate their performance for each location and time period. Although the detectors classify each specific signal, we only validated the presence or absence of species at the file level (one file corresponding to acoustic data recorded from one channel and one duty cycle, see Table 1). Presence of a species was defined by the presence of one or more detections for that species in a file. Acoustic signals were only assigned to a species if an analyst was confident enough in their assessment. In addition to the 1% of files selected for validation, every automated detection of minke, killer, and right whale was manually reviewed. All automated detections of blue and fin whale were also manually verified between 1 Mar and 1 Aug to confirm their daily presence. Performance of the detectors was quantified using the precision (P) and recall (R) metrics defined by: where TP is the number of true positives, FP the number of false positives, and FN the number of false negatives, calculated based on the presence/absence of species for each file reviewed with the second manual analysis approach. We calculated P and R for each species and each deployment.
Where the number of validated files was too low, and/or the overlap between manual and automated detections was too limited to calculate P and R, and/or when P < 0.75, automated detections were ignored and only validated results were used to describe the acoustic occurrence of a species (Table S3).

Source Level Estimation
Each whale species produces vocalizations in different frequency bands and at different source levels. Table S4 lists the vocalization types, source level values, and frequency bands for each species of interest for this study based on a review of the scientific literature. The source levels of each species were represented in the Monte Carlo simulation by sampling respective Gaussian distributions with the means and standard deviations defined in Table S4 (see also Figure S1).   Table S4.

Vocalization Depth
Propagation loss is depth dependent, so the depth at which an animal vocalizes can greatly influence the maximum distance at which its vocalization can be detected. The Monte Carlo simulation randomly selected the vocalizing animal depth from a distribution covering the typical depth range for the whale's vocalization of interest. Depth models were defined by reviewing vocalization depth data from the scientific literature for each species of interest and representing these data using log-logistic, Gaussian, or Uniform distributions. When raw depth data were available, the parameters of the distributions were fitted using Maximum Likelihood. Otherwise, the distribution parameters were approximated using the information available from the literature. When the information was available, the model represented the depths at which the vocalizations are produced. If such information was unavailable, the model represented the depths at which the whales were distributed (but not necessarily vocalizing). For killer whales, the depth model was based on unpublished tag data collected by Brianna Wright from DFO and was used in a previous study for DFO by 47 . We did not find information in the literature about the depth distribution of white-sided dolphins, so we used the same model as killer whales for that species. Sperm whales were considered uniformly distributed from 10 to 500 metres ( Figure S2).

Propagation Loss Modelling
Propagation loss was modelled along four radials (Table S5) in different directions at Ramsay Island and Gowgaia Slope ( Figure S3) to sample the propagation loss characteristics as a function of range and azimuth. The Bellhop Ray Tracing Model was used to calculate propagation loss for frequency bands with components above 1.0 kHz 48 . The model accounts for absorption in seawater, which can be important at the high frequencies considered in this study. Propagation loss for low frequencies (<1.0 kHz) was modelled with the Range-dependent Acoustic Model (RAM; 49 ). Both models take as input the sound source location and depth, geoacoustic parameters of the ocean bottom, sound speed profiles for the water column, and a profile of the bathymetry along the modelled radials. Propagation losses were calculated at three frequencies within each 300 Hz band and then averaged to provide a propagation loss estimate for each frequency band of killer whales, humpback whales, sperm whales, and Pacific white-sided dolphins. In the frequency bands of fin, gray, and blue whales, propagation losses were calculated every 1 Hz.
All resulting propagation loss values were plotted as a function of range, and the data were fit with an equation of the form: for frequency (f) in Hertz, depth (z) in metres, and range (R) in metres. The α(f,z) term in Equation S1 accounts for volume absorption in decibels per kilometer and is only used for curve fitting at the higher frequencies of killer whales, sperm whales, pacific white-sided dolphins, and part of the humpback whales. The resulting PL values were used in the Monte Carlo simulation for the corresponding depths and frequency bands.

Bathymetry
Water depths throughout the modelled locations were obtained from the ETOPO2 database 50 with a 2minute resolution. Bathymetry values were extracted and re-gridded onto a Universal Transverse Mercator (UTM) Zone 9 coordinate projection with a regular grid spacing of 10 m.

Geoacoustics
The geoacoustic parameters of sediments are compressional or P-wave speed, shear or S-wave speed, density, P-wave attenuation, and S-wave attenuation. These parameters were derived for use in a previous noise mapping study 51 from data from two sources: • Geological Survey of Canada (GSC) surficial sediment point sampling data.
Geoacoustic models were derived from bottom classification based on grain size 52 .

Sound Speed Profile
For the Gowgaia Slope location, no conductivity, temperature and salinity (CTD) measurements were available, so summer water column sound speed profiles were computed from salinity and temperature profiles obtained from the Generalized Digital Environment Model (GDEM) version 3.0 in August (summer).
In summer, heating of the surface results in a surface thermocline leading to higher surface sound speeds and downward acoustic refraction. For the propagation loss modelling, each radial was characterized by eight sound speed profiles (one every 50 km).
At the Ramsay Island location, water column sound speed profiles for summer were computed from conductivity, temperature, and salinity (CTD) measurements from the study area in August 2018. Sound speed profiles were calculated using the Del Grosso equation reformulated by Wong and Zhu for the 1990 International Temperature Scale 53 . In summer, heating the surface results in a downward refracting sound profile.

Detection range results
The detection range probabilities at Gowgaia Slope for blue whale B-song notes, fin whale 20 Hz pulses, humpback whale songs, killer whale pulsed vocalizations, and sperm whale off-axis clicks are summarized in Table S6.
The detection range probabilities at Ramsay Island for gray whale moans, fin whale 20 Hz pulses, humpback whale songs, killer whale pulsed vocalizations, and for Pacific white-sided dolphin whistles are summarized in Table S7.  Table S7. Median detection range (in metres) along each radial of Ramsay Island for Probability of Detection, P, of 0.1, 0.5, and 0.9.

Species Radial Summer detection ranges (m)
3 Diel patterns   None observed Potentially part of delphinid clicks *Note the hydrophone stopped recording early July at the same time that the visual survey was completed. 16 Figure S5. Results of the GAMM analysis for the correlation between sea surface temperature (SST), chlorophyll a (chla) and acoustic recorder locations and blue whale acoustic presence. Figure S6. Results of the GAMM analysis for the correlation between sea surface temperature (SST), chlorophyll a (chla) and acoustic recorder locations and fin whale acoustic presence. Figure S7. Results of the GAMM analysis for the correlation between sea surface temperature (SST), chlorophyll a (chla) and acoustic recorder locations and humpback whale acoustic presence. Figure S8. Results of the GAMM analysis for the correlation between sea surface temperature (SST), chlorophyll a (chla) and acoustic recorder locations and sperm whale acoustic presence. Figure S9. Results of the GAMM analysis for the correlation between sea surface temperature (SST) and chlorophyll a (chla) and Baird's beaked whale acoustic presence.