Are Icelandic harbor seals acoustically cryptic to avoid predation?

: Male harbor seals ( Phoca vitulina ) produce stereotypic underwater roars during the mating season. It remains unclear to what extent roar structures vary due to predation levels. Here, seal roars from waters with many (Iceland) and few (Denmark and Sweden) predators were compared. Most Icelandic roars included a long pulse train and a pause. Icelandic roars occurred less frequently, lasted longer (20.3 6 6.5s), and were recorded with lower received sound levels (98.3 6 8.9dB re 1 l Pa root mean square) than roars from Denmark and Sweden. Local extrinsic factors may shape sound production in harbor seals more than previously reported. V C 2021 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/) .


Introduction
Harbor seals (Phoca vitulina) are distributed in temperate and polar coastal waters along the eastern and western North Pacific and North Atlantic coasts (Berta, 2009;Bowen et al., 2009;Stanley et al., 1996). Harbor seals mate underwater, but details of their mating behavior remain unclear (Hayes et al., 2006). Some evidence indicates a hotspot lek system, consisting of underwater display territories where males compete while being noticed by females (Boness et al., 2006;Hayes et al., 2004a;Hayes et al., 2004b). Males also appear to patrol along the coastline in a form of mate-guarding or scrambled competition (Van Parijs et al., 1997;Van Parijs et al., 1999).
During the mating season, after lactation and before molt (Thompson, 1988), underwater calls are produced, presumably by males. The most frequently produced harbor seal call is a broad-band roar with a frequency content below 4 kHz and shorter than 20 s (Hanggi and Schusterman, 1994;Van Parijs et al., 2000b;Bjørgesaeter et al., 2004;Sabinsky et al., 2017). Their reported average source level is 144 dB root mean square (rms) re 1 lPa at 1 m recorded from wild harbor seals in Alaska (Matthews et al., 2017) and approximately 140 dB rms re 1 lPa at 1 m from one captive harbor seal in California (Casey et al., 2016). Acoustic communication is impacted by a variety of factors, such as the time of year or the type of environment. It is also conceivable that higher predation risk from, e.g., orcas (Orcinus orca) (Deecke et al., 2002) and gray seals (Halichoerus grypus) (Brownlow et al., 2016) affect harbor seal underwater communication during mating. Harbor seals exposed to predators may face an exacerbated trade-off between reproduction and self-maintenance and are thus expected to adapt signal parameters to maximize their fitness.
Icelandic harbor seals are a suitable population to investigate the evolutionary influence of predator-prey interaction on call production. They are genetically distinct from other North Atlantic seals with little or no exchange with other populations (Stanley et al., 1996;Goodman, 1998). Orcas (Samarra et al., 2018) and gray seals (Granquist and Hauksson, 2019a) are present in Icelandic waters, making it likely that roars may be shaped to reduce predation risk. In this study, we recorded Icelandic harbor seals to understand how their calls differ from areas with little predation pressure in Danish and Swedish waters. Some of the differences could potentially be attributed to the different predation pressures rather than to the large geographical distance between the different harbor seal populations.

Study sites
Data collection took place at two study sites (Heggstaðanes and Illugastaðir,N 65.6 ,W 20.9 ) separated by approximately 20 km, on the Northwest coast of Iceland (supplementary Fig. 3), 1 where 9% of the total Icelandic harbor seal population (approximately 9400 animals) occurs (Granquist and Hauksson, 2019b). At Heggstaðanes, seals haul out on six rocky sites [H1-H6, supplementary Fig. 3(c)]. 1 At Illugastaðir, Vatnsnes peninsula, two parallel rocky skerries are located approximately 100 m apart. There are tidal variations at both study sites with an average tidal amplitude of 50 cm (Meteo365.com Ltd., 2018). Seabed structure at both study sites is rough sand, shells, or rocky with brown algae.

Acoustic recordings: Acoustic data logger deployments
Two self-contained underwater sound recorders were used (SoundTrap 300 HF, OceanInstruments, Auckland, New Zealand). Deployments were made in 3-15 m water depth. One recorder was deployed between the skerries at Illugastaðir from July 3 to September 5, 2017 [supplementary Figs. 4(A) and 5(c)] 1 and recorded for 10 min every hour. The second recorder was deployed and recorded continuously 11 times at Heggstaðanes [supplementary Figs. 3,4(B), and 5(c)] 1 from July 4 until September 13. Both SoundTraps recorded WAV files with a sample rate of 48 kHz and 16-bit resolution.
The SoundTraps were calibrated before field work by the manufacturer and after field work by relative calibration, both in air and underwater. Details of system calibration can be found in the supplementary material. 1

Analysis of seal calls
Recordings were screened using Adobe Audition, version 1.5 (Adobe, Inc., San Jose, CA). Seal calls were manually selected from the recordings and time stamped. We only selected calls with a signal-to-noise ratio better than 6 dB, measured as the difference between the rms intensity of the call and the ambient noise within the bandwidth of the call. Calls were analyzed with a custom-made script (Sabinsky et al., 2017) in MATLAB, version R2016b (MathWorks Inc., Natick, MA).
Each roar recording was down-sampled to 5 kHz and high-pass filtered at 80 Hz (four-pole Butterworth filter). The calls were manually divided into seven different color-coded consecutive sections. The measured parameters for every section were duration (s), energy (lPa 2 s), rms (Pa), centroid frequency (Hz), peak frequency (Hz), rms bandwidth (Hz), À10 dB bandwidth (Hz), and kurtosis (Southall et al., 2008). Mean, maximum, minimum, and standard deviation (SD) of each parameter were plotted. In addition, the total duration (s) of each call (including pauses) was measured. Received sound level (rl, in lPa rms over a duration of a 95% energy window) was calculated as described in Madsen and Wahlberg (2007) and expressed as dB rms re 1 lPa rms.
All statistical tests were done in R for Windows, version 3.3.1 (R Foundation, Vienna, Austria) with R Studio (version 1.1.383). To find data-driven groups in the Icelandic data set, a k-means cluster analysis was applied on the roar burst parameters (see above), including the total call duration. Before this, the data were scaled, and k was determined with k ¼ 3 as best fit by the ratio of the sum of squares divided by the total sum of squares for each k-means model. All parameters of the roar burst were compared with a canonical linear discriminant analysis (CLDA) to harbor seal roar bursts from Blinderøn in Limfjord in 2010 and 2011 (Denmark), Juvre in the Wadden Sea (Denmark), and Kalmarsund (Sweden) (Sabinsky et al., 2017). Statistical differences of how the site (as descriptive variable) affected the factors of total call duration, energy, and roar burst duration were tested using three generalized linear models (GLMs).

Results
All detected calls were recorded at Heggstaðanes (Iceland) at the first haul-out site (H1) on July 4 from 9:40 a.m. until July 5 at 9:00 a.m. There was no apparent pattern in calling over the recording period, but calls could be heard at all times of the day and night.
In total, 76 calls fitting inclusion criteria were observed at H1. Average 6 SD received sound level of the roar burst was 98.3 6 8.9 dB re 1 lPa rms. Mean total duration was 20.3 6 6.5 s, and the fundamental frequency (f0) was 75-100 Hz.
The analyzed calls had seven components labelled A, B, C, D, E, F, and G ( Fig. 1) arranged in nine different sequences (Table 1). The highest number of calls had the combination AB---FG, with components C, D, and E variously missing in the middle positions (Table 1). In 96% of the cases, there was a "first pause" (B) after the "first pulse train" (A). Each component's duration was extracted from the oscillogram, which showed that the first pulse train (A) was the longest component (duration in seconds, Table 1).
Three different groupings could be identified within the calls. Clusters overlapped, but classification based on the k-means function could be accomplished with overall 51.6% of calls classified correctly (k-means, n ¼ 76, k ¼ 3).
For the comparison of the Icelandic roars with Danish and Swedish roars, roar burst parameters were extracted from all three sites (Table 2). ARTICLE asa.scitation.org/journal/jel For the cluster analysis of Icelandic, Danish, and Swedish roar bursts, the following roar burst parameters were included: duration, centroid frequency, rms-bandwidth, -10 dB bandwidth, and the percentage of energy in the roar burst compared to the total call energy. Furthermore, the measured total duration of the call was used for comparison between sites. Four significant linear discriminant functions were derived, of which the first two explained 91% of the total Fig. 1. Examples of oscillograms and spectrograms of a complete "ABCDEFG" call (top) and an "AB---FG" call (bottom). Background noise is shown in blue, "first pulse train" (A) is green, "first pause" (B) is black, "second pulse train" (C) is pink, "second pause" (D) is black, "start of growling" (E) is red, "roar burst" (F) is cyan, and "end groan" (G) is yellow. The top call had a signal-to-noise ratio of 12 dB, and the bottom call had a signal-to-noise ratio of 11 dB. Signals were resampled at 5 kHz, and the spectrogram had a fast Fourier transform size of 512 points, Hann window, and 25% overlap. Table 1. Call types with the encountered combinations of call components (A-G) and number of calls of that call type (A ¼ first pulse train, B ¼ first pause, C ¼ second pulse train, D ¼ second pause, E ¼ start of growling, F ¼ roar burst, and G ¼ end groan) and duration of each of the seven components (mean 6 SD). Roar burst component (F) was always present, as this defined which calls to analyze as roars.

Call type
Call components Number of calls Call duration was longer at Icelandic than at Danish and Swedish sites (GLM, p < 0.001). Roar bursts were as long in Heggstaðanes (Iceland) as in Blinderøn in Limfjord (Denmark) (GLM, p ¼ 1) but shorter than at Juvre in the Wadden Sea (Denmark) and longer than in Kalmarsund (Sweden) (GLM, p < 0.001). The carried energy in the roar burst was significantly lower in Heggstaðanes (Iceland) than at any other location (GLM, p < 0.001; Fig. 2).

Discussion
Underwater mating roars from Icelandic harbor seals differed in duration and received energy level from the roars produced by Danish and Swedish harbor seals. All Icelandic roars were recorded around one haul-out site, and roars were only heard early in the recorded season. In Denmark, calls are heard throughout the mating season and at many sites around the haul-out colonies (Sabinsky et al., 2017). Also, studies from Scotland (Van Parijs et al., 1997;Van Parijs et al., 1999) and east and west coasts of North America (Van Parijs and Kovacs, 2002;Hayes et al., 2004a;Boness et al., 2006;Nikolich, 2016;Matthews et al., 2017) indicate that harbor seal vocal displays are usually easy to detect and recognize during the breeding season. Icelandic harbor seals thus appear to be more silent during the mating period than other populations.
The timing of mating in Icelandic harbor seals is largely unknown. Our recordings indicate that mating was ongoing in early July, even though it may have started earlier. Gentry (1998) argued that higher latitudes, with harsh weather and strong but predictable seasonality, seem to favor short lactation periods in seals. This could indicate a possible earlier ending of the lactation period in Iceland than in Danish and Swedish populations. However, even though there were no calls recorded throughout most of July, mating-related displays were observed in Iceland in this time period, when calls are ubiquitous in Danish and Swedish waters. Therefore, mating in Iceland could have continued throughout July, but without vocalizations (Sullivan, 1981;Bishop, 1967) or with a change of location [such as suggested by Van Parijs et al. (2000a)]. It is also possible that the observed mating-related displays were from younger animals not involved in mating (Thompson, 1989;Perry, 1993;Nicholson, 2000).
The low received levels of the Icelandic calls could either be caused by a large propagation loss (i.e., seals being far away from the data logger) or by the seals emitting low intensity signals. It would be surprising if the transmission loss could affect recordings to the extent observed here, especially given the high number of seals observed close to the hydrophone during recordings. If the low received levels were due to seals calling at a distance with the same intensity as vocalizations recorded in Denmark and Sweden, these vocalizing individuals would have to be several kilometers from the haulout sites, which is highly unusual for harbor seals (Van Parijs et al., 2003;Sabinsky et al., 2017).
The most parsimonious explanation for the very low received levels in our recordings is therefore that seals themselves emit very low source levels. The size of the investigated Icelandic haul-out sites was 46-59 animals (supplementary Fig. 6), 1 which is much fewer individuals than many of the Danish and Swedish recording sites. Fewer animals, and therefore possibly fewer males, at each local site could have been a reason for reduced acoustic activity and low emitted source levels. It was unknown during this study if several individuals were vocalizing during the recording period. However, call structure variability suggests multiple callers (see Van Parijs et al., 2000b), which contradicts the idea of the source level being low because of few individuals calling.
Some of the differences in call parameters, such as the lower source level and longer duration of Icelandic calls, could also be explained by an adaptive strategy to avoid predation. Predators, such as gray seals or orcas, are much more prevalent in Icelandic waters (Deecke et al., 2002;van Neer et al., 2015) than, for example, in Danish waters. Orcas may indeed use passive listening to detect and locate marine mammal prey as a specialized hunting strategy, instead of relying on echolocation (Deecke et al., 2005). Orca hearing is most sensitive at 15-20 kHz [at 35 and 36 dB re 1 lPa, respectively;  ARTICLE asa.scitation.org/journal/jel While this strategy may reduce predation risk from eavesdropping orcas, it also decreases the likelihood of the call reaching its intended recipient. Harbor seals may use different strategies to resolve this dilemma. In Alaska, harbor seals use much higher source intensities than what we presume is the case in Iceland, despite being in an area frequented by orcas (Matthews et al., 2020). This may help the males to maintain their competitive edge during mating, with the risk, however, of being consumed by orcas. Thus, lowering the frequency emphasis rather than the source level may be the way out of this dilemma for Alaskan harbor seals (Matthews et al., 2017). The prolonged calls in Iceland, with less intensity, may constitute another way to keep the signals attractive and honest during mating activities, while at the same time decreasing predation risk.