Effects of noise levels and call types on the source

Effects of noise levels and call types on the source levels of killer whale calls Marla M. Holt,a) Dawn P. Noren, and Candice K. Emmons Marine Mammal Ecology Team, National Oceanic and Atmospheric Administration (NOAA), National Marine Fisheries Service (NMFS), Northwest Fisheries Science Center, 2725 Montlake Blvd. East, Seattle, Washington 98112

(Received 18 March 2011; revised 10 August 2011; accepted 29 August 2011) Accurate parameter estimates relevant to the vocal behavior of marine mammals are needed to assess potential effects of anthropogenic sound exposure including how masking noise reduces the active space of sounds used for communication. Information about how these animals modify their vocal behavior in response to noise exposure is also needed for such assessment. Prior studies have reported variations in the source levels of killer whale sounds, and a more recent study reported that killer whales compensate for vessel masking noise by increasing their call amplitude. The objectives of the current study were to investigate the source levels of a variety of call types in southern resident killer whales while also considering background noise level as a likely factor related to call source level variability. The source levels of 763 discrete calls along with corresponding background noise were measured over three summer field seasons in the waters surrounding the San Juan Islands, WA. Both noise level and call type were significant factors on call source levels (1–40 kHz band, range of 135.0–175.7 dBrms re 1 mPa at 1 m). These factors should be considered in models that predict how anthropogenic masking noise reduces vocal communication space in marine mammals. C 2011 Acoustical Society of America. [DOI: 10.1121/1.3641446] V PACS number(s): 43.80.Ka, 43.80.Nd [WWA]

I. INTRODUCTION

Human activities introduce many sounds in the ocean that have the potential to negatively impact marine mammals. As a result, several efforts aim to address impacts of anthropogenic sound on these animals including work on noise-induced hearing loss, behavioral responses to sound and acoustic masking (for review, see NRC, 2003; Nowacek et al., 2007; Southall et al. 2007). For example, some recent studies have modeled how masking noise from shipping and other vessel traffic reduces the active space of communicative signals in baleen whales and delphinids (Clark et al., 2009; Jensen et al., 2009). Accurate estimates of a variety of variables including those related to the sounds produced by signalers (source), sound propagation in the local environment (path), and hearing capabilities of potential listeners (receiver), in addition to masking noise attributes are needed to make informed decisions about anthropogenic sound impacts. These include the typical source levels, frequency ranges, and temporal characteristics of animal sounds. Knowledge about if and how animals vocally modify their behavior in response to changes in masking noise levels [e.g. the Lombard effect (Lombard, 1911)] are also needed for accurate assessments of such impacts. Several marine mammal species inhabit urbanized coastal waters where anthropogenic sounds are common. For example, southern resident killer whales (Orcinus orca) frequent the inland waters around Washington State and British a)

Author to whom correspondence should be addressed. Electronic mail: [email protected]

3100

J. Acoust. Soc. Am. 130 (5), November 2011

Pages: 3100–3106

Columbia where vessel traffic is pervasive. Forty years of work on their population dynamics has shown that their social structure is based on groups or pods of matrilines with no natal dispersal of either sex (Ford et al., 2000). This strictly fish-eating population, composed of three (J, K, and L) pods, is currently listed as endangered in both the U.S. and Canada. Risk factors to their recovery include vessel and noise interactions given a whale-watching industry that heavily relies on viewing this population during the summer months. Killer whales produce a variety of sounds including echolocation clicks for foraging and navigation and pulsed calls and whistles during social interactions. Call production is believed to serve important roles in the social dynamics of groups that travel and forage together (Ford, 1989). Some calls are stereotypic being stable in structure, repeated often, and easily classified aurally and spectrographically into discrete call types (Ford, 1987; 1989). Call type repertoire is population and even group specific (Ford, 1991; Miller and Bain, 2000). For example, the predominant call types in southern resident killer whales are “S1” in J pod, “S16” in K pod, and “S2iii” and “S19” in L pod during directional travel (Foote et al., 2008). Some call types include an independently modulated (harmonically unrelated) and sometimes overlapping higher frequency component (HFC) in addition to the relatively lower frequency part of the pulsed call (Miller and Bain, 2000). The HFC is typically more tonal compared to the lower frequency component (for example, Fig. 1). Anatomical and observational evidence indicates that odontocete (toothed whale) sounds are produced by the phonic lips in the nasal complex of the forehead and, with the exception of sperm whales, there are two sets with one

0001-4966/2011/130(5)/3100/7/$30.00

C 2011 Acoustical Society of America V

Downloaded 16 Nov 2011 to 161.55.228.138. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp

(SLs) of the calls of northern resident killer whales, a population of killer whales that shares life histories and prey preferences but do not breed with southern resident killer whales despite some geographic sympatry (Ford et al., 2000). It was assumed that these estimates were based on off-axis recordings, given that data were collected from a hydrophone array towed in a parallel direction to the whales (Miller, 2006). Nonetheless, different call types had significantly different apparent SLs, and call types with a HFC had significantly higher SLs and about twice the predicted active space than those call types without a HFC (Miller, 2006). More recently, Holt et al. (2009) reported that call SLs in southern resident killer whales increased as background noise from nearby vessels increased. Thus, noise level seems to be an important variable that can affect estimates of killer whale call SLs, but sufficient data were only available for one call type (Holt et al., 2009). Although data were collected in low sea states and with no other vessel within 1 km, no detailed noise level measurements were reported in Miller (2006). Thus, noise level as an explanatory variable for the observed variability of killer whale call SLs have not been taken into account in previous investigations. The objectives of the current study were to investigate the SLs of a variety of call types in southern resident killer whales while considering background noise level as an important factor related to call SL variability. II. METHODS A. Location, equipment, and procedure

FIG. 1. Spectrograms of killer whale call types with high frequency components as indicated by the arrows (right graphs) and without high frequency components (left graphs).

on each side of the midline (Dormer, 1979; Cranford, 2000). There is the possibility that these two sets can act as separate sound generators, in addition to other potential structures (Cranford, 2000), and thus the overlapping HFC is thought to be produced independently from the other call component (e.g. Brown, 2008). For this reason, some investigators have labeled these sounds two-voiced calls or biphonations (Hoelzel and Osborne, 1986; Foote et al., 2008). Southern resident killer whales produce approximately 25 discrete call types and about half are HFC calls (Ford, 1987). Miller (2002) hypothesized that HFC calls communicate important orientation information for group coordination because these components are more directional. A number of studies have provided quantitative descriptions of the various sounds produced by free-ranging killer whales (Ford, 1989; Au et al., 2004; Miller, 2006). Miller (2006) reported variations in the apparent source levels J. Acoust. Soc. Am., Vol. 130, No. 5, November 2011

Data collection occurred in the transboundary waters surrounding the San Juan Islands, WA, and southeastern Vancouver Island, British Columbia, Canada (study area approximate range: 48 220 12“ N to 49 00 48” N, 122 430 18“ W to 123 190 46” W). All data were collected off of an 8-m research vessel during daylight hours (0900 to 1930) in the summers of 2007–2009 with sea states ranging from [1/2] to 2. When groups of killer whales were sighted, the research vessel was positioned approximately 1 km ahead and in the path of the group as they were swimming toward the vessel to maximize the number of on-axis recordings of killer whale sounds (with whales facing 0 relative to the recording hydrophone). The vessel motor was then shut off, and the acoustic equipment was deployed. Pod identification as well as individual identification based on photo-ID records were recorded along with the distance (estimated with a laser range finder within 1000 m, Yardage Pro, Bushnell) and direction (visually estimated in 30 increments) of surfacing individuals relative to the research vessel when possible. These data were taken to periodically verify that the group continued in the same direction toward the research vessel while recordings were made. Ranges used for source level calculations were obtained from hydrophone array localization as described in detail in the following text. After all individual whales passed the vessel, the equipment was recovered and water salinity and temperature were determined at each location as described in Holt et al. (2009). The acoustic recording equipment was similar to that described in Holt et al. (2009). The receiving transducer used Holt et al.: Source levels of killer whale calls in noise

Downloaded 16 Nov 2011 to 161.55.228.138. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp

3101

for all SL and noise level (NL) measurements was a factory calibrated ominidirectional hydrophone (Reson TC-4033, nominal sensitivity of 203 dB re 1 V/mPa) connected to a low-noise preamplifier (Reson VP1000 with a high pass setting of 1 kHz or VP2000 band pass setting of 1–100 kHz, both with equivalent gain). The receiving frequency response curve of the hydrophone connected to the preamplifier was calibrated on a yearly basis at the Naval Undersea Warfare Center Acoustic Test Facility in Keyport, WA, during the study period. The response curve varied little with the factory curve provided (62 dB in 1–40 kHz band, 63 dB in 0.02–94 kHz band). A four-element hydrophone array (LabCore Systems) with a 20-m aperture (for hydrophone spacing, see Holt et al., 2009) was deployed vertically with the first hydrophone at 5 m depth with a buoy and 10 kg weight. The measurement hydrophone was placed at depth of 7 m and at a range of 3.2 m relative to the shallowest (5 m depth) element of the array. Signals were digitized at a sampling rate of 192 kHz using a MOTU Traveler, recorded using a customized version of ISHMAEL 1.0 (Mellinger, 2001), and stored as time-stamped fivechannel wave files on a PC laptop for further analysis. The sampling rate was used to record the full frequency range of both echolocation clicks and calls of killer whales (Au et al., 2004; Miller, 2006). The 100 kHz low pass filter setting was chosen as the most appropriate fixed option available when the VP2000 preamplifier was used because the next lower option was 50 kHz. Although the Nyquist frequency was 96 kHz in this case, anti-aliasing filtering occurred with A/D conversion (MOTU, Inc.). Only calls were analyzed in the current study and the frequency bandwidth of analysis was further restricted as described in the following text. B. Analysis

The range of each call relative to the measurement hydrophone was estimated in ISHMAEL using time-of-arrival differences (hyperbolic localization, see Spiesberger and Fristrup, 1990) based on sound speeds calculated from average temperature and salinity profiles of that day (MacKenzie, 1981). Sound speeds varied little with depth indicating wellmixed water except near the Fraser river mouth; thus these data were not included. Localization error was assessed using previously recorded killer whales calls projected from a Lubell LL 9816 transducer (depth 9 m) at known source levels and horizontal distances. Received levels from known source distances during localization error assessment indicated that spherical spreading loss was an accurate assumption (as in Holt et al., 2009). The largest range errors occurred when the predicted range was either less than 40 m or more than 400 m. This was likely because hyperbolic localization curves intersect most accurately when the range is neither too close to the axis of nor too far away from the linear array. When these short and far simulated calls were excluded from the data set, the resulting average calculated source level error was 2.6 dB. Thus, only killer whale calls that were localized within an estimated range of 40-400 m were included in the analysis (as in Holt et al. 2009). The entire recording system (hydrophone, preamplifier, filters, A/D device input gain, and sound card input gain) 3102

J. Acoust. Soc. Am., Vol. 130, No. 5, November 2011

was calibrated using a pistonphone with a known sound pressure level (G.R.A.S. Sound and Vibration, model type 42AA with RA0078 coupler). Background NLs and call SLs were measured based on root-mean-square pressure (in dBrms re 1 lPa) over a standardized duration (250 ms) in the 1–40 kHz band by applying a low-pass digital filter at 40 kHz and then calculating calibrated measurements via the calibration utility in the time domain in SPECTRAPLUS version 5.0 (Pioneer Hill). The duration was chosen based on the shortest calls analyzed and the variability in NLs in the area, particularly from vessel traffic (Holt et al. 2009). This frequency range of analysis was chosen given the hearing range of captive killer whales (Szymanski et al., 1999) and the observed upper frequency limit of southern resident killer whale calls when recorded on-axis. Background NLs were measured within 9 s of a call. If the call was longer than 250 ms, the highest amplitude over a 250 ms window of the call was chosen as the received level (RL). The corresponding NL was subtracted from the RL before the source level was calculated (for details, see Holt et al., 2009). This was necessary because received levels of calls were not always well above the corresponding NLs. These steps were taken to ensure that any correlations of call levels and noise levels were not artificially created by only selecting calls with high signal to noise ratios. Call source level was then calculated in dBrms re 1 mPa at 1 m using the following equation: SL ¼ RL þ 20log10 R;

(1)

in which R was the estimated range from Ishmael. Regression analysis was run in SIGMAPLOT 11.0 (SYSTAT) to determine the relationship between SL and NL. Assumptions of the linear regression model were fulfilled including testing assumptions of linearity via residual plot inspections, and normality and homogeneity of variance via Shapiro-Wilk and Constant Variance tests, respectively, before linear regression analysis was performed. R-squared values were calculated from linear regression results as the regression sum of squares divided by the total sum of squares. A oneway analysis of variance (ANOVA) was first run to determine differences in SL among discrete call types (single effect) before an analysis of covariance (ANCOVA) was run to determine differences in SL among discrete call types (main effect) in which NL was a covariate (in SYSTAT 9). Differences in source levels among call types were then compared between the ANOVA and ANCOVA to explore the influence of NL on SL. III. RESULTS

Call SLs and background NLs were calculated from recordings collected over 25 days (4 days in 2007, 11 days in 2008, and 10 days in 2009). A breakdown of recording days by pod presence is shown in Table I. Source levels were calculated for 763 discrete calls and ranged from 135.0 to 175.7 dB re 1 mPa at 1 m with a mean of 155.1 dB re 1 mPa at 1 m (66.5 SD). Some call types occurred more often than others, and for the subsequent analysis, we only included those call types with a sufficient sample size (n > 35) as listed in Table II Holt et al.: Source levels of killer whale calls in noise

Downloaded 16 Nov 2011 to 161.55.228.138. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp

TABLE I. Pod presence by number of recording days. Pod ID

Number of recording days

Percentage of total

J K L J, K J, L K, L J, K, L Total

9 2 2 0 5 3 4 25

36% 8% 8% 0% 20% 12% 16%

based on Ford (1987). These include the predominant call types produced by each of the three southern resident killer whale pods (Foote et al., 2008). Call types with a HFC were S2iii, S19, and S44 (Ford, 1987; Fig. 1, Table II). Source level ranges (highest SL minus lowest SL) within each call type varied between 25 and 36 dB, depending on the call type. Results of the one-way ANOVA showed a significant effect of call type on source levels (ANOVA: F6,520 ¼ 5.22, P < 0.001). The unadjusted mean differences in source levels between call types based on pairwise comparisons (Tukey test) are listed in Table II. Although source levels appeared to differ by call type, there were also differences in the mean corresponding NLs (Fig. 2). For example, noise levels for S3 and S16 call types appeared to be lower, on average, compared to noise levels for other call types. Furthermore, SL significantly increased as corresponding NL increased for each call type, as shown in Table II and Fig. 3. The slope of the regression of SL vs NL for all 763 calls was 0.75 (P < 0.001). The slope of each linear regression for each call type ranged from 0.46 to 1.25 (Fig. 3). Source level vs NL for the S1 call type can be found in Holt et al. (2009). Note that the current study includes an increased sample size of the S1 call type (listed in Table II), but there were no significant differences between slopes or intercepts of the regression of S1 SL vs NL between data from Holt et al. (2009) and the data reported here (based on Student’s t tests). The assumption of homogeneity of regression was fulfilled for an ANCOVA because there was no difference in the slope of SL vs NL among call types (i.e., no significant interaction between call type and noise level, P ¼ 0.46, using

FIG. 2. Box plots showing median (dark line), 25th–75th percentile (box) and 5th-95th percentile (whiskers) of source levels (shaded, left side y axis) and paired noise levels (unshaded, right side y axis) for each call type.

GLM). Source levels were significantly different among call types when NL was incorporated into the model as a covariate factor (ANCOVA: F6,519 ¼ 7.04, P < 0.0001). However, Tukey pairwise comparisons of the adjusted means (with removal of the NL variance effects) revealed some disparate results compared to pairwise comparison of unadjusted means (i.e., without NL as a covariate factor), which are also listed in Table II. The differences in the adjusted means for these pairwise comparisons ranged between 2.4 and 4.5 dB as shown in Table II. IV. DISCUSSION

The mean source level of 763 discrete calls of southern resident killer whales reported here was within 1.5 dB of those reported for northern resident killer whales despite several differences in methods between studies including equipment, localization methods, analysis bandwidth and duration, and transmission loss assumptions (Miller, 2006). In the current study, a significant positive relationship was found between SL and NL for all call types with a sufficient sample size indicating that killer whales demonstrate a Lombard-like response to changes in noise levels. This assumes that the observed positive relationship did not result from our inability to detect and localize lower source level

TABLE II. Summary of southern resident killer whale call types reported in this study, including those with a high frequency component (HFC). Sample size (N) and linear regression results of source level by noise level for each call type and pairwise comparisons results of source levels between call types based on the Tukey HSD test; unadjusted mean differences are shown in parentheses and based on ANOVA (call type was a single factor), adjusted mean differences are based on ANCOVA with noise level as a covariate; significant differences (P < 0.05) are shown in bold, with the exception of S1 vs S19 in which P ¼ 0.05 (indicated by *), NS ¼ not significant.

Call Type

S1 S2iii S3 S10 S16 S19 S44

HFC?

No Yes No No No Yes Yes

N

164 105 61 42 37 65 53

P-value regression results

Holt et al., 2009 0.013