This study measured the degree of behavioral responses in blue whales (Balaenoptera musculus) to controlled noise exposure off the southern California coast. High-resolution movement and passive acoustic data were obtained from non-invasive archival tags (n=42) whereas surface positions were obtained with visual focal follows. Controlled exposure experiments (CEEs) were used to obtain direct behavioral measurements before, during and after simulated and operational military mid-frequency active sonar (MFAS), pseudorandom noise (PRN) and controls (no noise exposure). For a subset of deep-feeding animals (n=21), active acoustic measurements of prey were obtained and used as contextual covariates in response analyses. To investigate potential behavioral changes within individuals as a function of controlled noise exposure conditions, two parallel analyses of time-series data for selected behavioral parameters (e.g. diving, horizontal movement and feeding) were conducted. This included expert scoring of responses according to a specified behavioral severity rating paradigm and quantitative change-point analyses using Mahalanobis distance statistics. Both methods identified clear changes in some conditions. More than 50% of blue whales in deep-feeding states responded during CEEs, whereas no changes in behavior were identified in shallow-feeding blue whales. Overall, responses were generally brief, of low to moderate severity, and highly dependent on exposure context such as behavioral state, source-to-whale horizontal range and prey availability. Response probability did not follow a simple exposure–response model based on received exposure level. These results, in combination with additional analytical methods to investigate different aspects of potential responses within and among individuals, provide a comprehensive evaluation of how free-ranging blue whales responded to mid-frequency military sonar.
Sound production and reception are centrally important in the life history of all marine mammals, and their responses to natural signals as well as human noise can have both positive and negative fitness implications. However, we lack a comprehensive understanding of how most marine mammals respond to sound in their natural environment. Given the substantial scientific and regulatory interest in quantifying the effects of anthropogenic noise on marine mammals in recent decades (National Research Council, 1994, 2005; Southall et al., 2007, 2009, 2016; Hatch et al., 2016; National Academies of Sciences, Engineering and Medicine, 2017; Southall, 2017), there is a pressing need for detailed measurements of responses to acoustic disturbance in known and/or controlled exposure conditions. Regulatory requirements include quantifying marine mammal behavioral responses to noise with sufficient resolution to understand key aspects of behavior (e.g. foraging) that, if negatively affected, may have fitness consequences at both the individual and population level (King et al., 2015; McHuron et al., 2018; Pirotta et al., 2018).
The effects of military sonars on marine mammals have received particular attention. Specifically, focus has been placed on lethal mass strandings involving beaked whales associated with tactical mid-frequency (nominally 1–10 kHz) active sonar (MFAS) (see Filadelfo et al., 2009). However, both observational and experimental studies have documented sub-lethal behavioral responses to various kinds of sonar systems in an increasingly wide range of marine mammal taxa (e.g. Fristrup et al., 2003; Tyack et al., 2011; Miller et al., 2012, 2014; Moretti et al., 2014; Henderson et al., 2014; Sivle et al., 2015, 2016; Isojunno et al., 2016; Southall et al., 2016; Falcone et al., 2017). Responses range from brief and/or minor changes in social, vocal, foraging and diving behaviors to more severe modifications, including sustained avoidance of important habitat areas in some conditions (see Southall et al., 2016; Southall, 2017). Although sub-lethal, such responses may negatively influence vital rates in ways that, depending on their duration and severity, and the proportion of the population that is affected, may be consequential for protected or endangered marine mammal species. Direct, empirical measures of sub-lethal behavioral responses of marine mammals are thus needed in contexts where sonar exposure is known and can be compared within and across individuals (Southall et al., 2016). Specifically, given the regular exposure of various species to MFAS in and around military training areas, and the threatened or endangered status of most baleen whale species, understanding the frequency of occurrence and severity of how sonar affects behavior in these species has both scientific and regulatory importance.
Observational studies using passive acoustic monitoring have documented behavioral responses in several baleen whales to various types of operational military sonar systems (Miller et al., 2000; Fristrup et al., 2003; Martin et al., 2015). Controlled exposure experiments (CEEs) that use high-resolution animal-borne tags with movement and acoustic sensors provide detail on individual behavioral responses as well as the characteristics of received sound at the position of the animal (see Southall et al., 2016). Such approaches can increase the ability to empirically relate and quantify known sonar exposure with fine-scale aspects of behavioral responses (e.g. foraging) that are more difficult to measure with coarser observational methods. For instance, Nowacek et al. (2004) demonstrated responses of some North Atlantic right whales (Eubalaena glacialis) to controlled alarm stimuli. Sivle et al. (2016) identified behavioral changes of individual humpback (Megaptera novaengliae) and minke (Balaenoptera acutorostrata) whales exposed to towed operational military sonars.
Blue whales [Balaenoptera musculus (Linnaeus 1758)] are classified as endangered under the IUCN red list (Cooke, 2018). They are also considered endangered under the US Endangered Species Act of 1973 (16U.S.C. § 1531 et seq.), which, along with the US Marine Mammal Protection Act of 1972 (16U.S.C. § 1361 et seq.), affords them federal protections within the USA. Blue whales are the largest animals on the planet, yet they feed almost exclusively on small invertebrates (krill) in near-surface to deep (∼300–400 m) layers. They often occur in coastal waters, including along the California coast during summer and autumn. However, they also forage in pelagic areas, including in areas where military sonar is regularly used. Although, like for all baleen whales, there are no direct measurements of hearing in blue whales, they primarily produce and are presumably more sensitive to low-frequency sound. However, recent evidence (e.g. Goldbogen et al., 2013; DeRuiter et al., 2017) suggests they may be behaviorally sensitive in some conditions to mid-frequency sounds (1–10 kHz).
Behavioral responses of blue whales to MFAS and other mid-frequency sounds have been quantified using CEEs in a series of studies off the southern California coast (Southall et al., 2012; Goldbogen et al., 2013; Friedlaender et al., 2016; DeRuiter et al., 2017). These experimental studies have notably involved MFAS designed to simulate US Navy SQS-53C systems that were used in previous stranding events. The results of this previous work, which involved subsets of the data used here, demonstrate significant behavioral responses of many individual blue whales to MFAS and pseudorandom noise (PRN) of similar frequency and exposure level. Further, they illustrate several context-dependencies in behavioral responses, as noted by Ellison et al. (2012), including strong influences of individual behavioral state at the time of exposure, as well as prey distribution and density. DeRuiter et al. (2017) used hidden Markov models to evaluate behavioral state-switching, demonstrating greater probabilities for blue whales to either cease deep-feeding or fail to initiate deep-feeding behavior during sonar exposure. Collectively, these studies show generally that blue whales may respond to controlled noise exposures in different ways, and that a suite of contextual factors influenced response probability. However, results from these kinds of studies are more challenging to apply directly within regulatory applications, where more explicit individual information on response probability and severity are often required.
The above analyses of blue whale responses all involved methods assessing results across multiple individuals. These results demonstrate that some blue whales, which primarily use low-frequency sound, may be sensitive to mid-frequency noise and that their responses appear to be influenced by various contextual factors. However, there is a further need to quantify individual responses (or lack of responses) of specified type and severity associated with known noise exposure conditions. Such data are directly useful in deriving exposure–response probabilistic functions for specific exposure variables commonly used in regulatory frameworks [e.g. received levels (RLs)], as has been shown for Phase-I clinical trials in medicine and has been applied within other cetacean behavioral response studies (see Miller et al., 2012; Southall et al., 2016). Individual case-by-case analyses also enable the evaluation of how other response covariates, such as the source–individual range evaluated here, may also influence response probability (as in Harris et al., 2015). Although the present study includes individuals evaluated in a number of the studies cited above, by quantifying individual responses of blue whales to MFAS and PRN stimuli using whale-borne tags and CEEs, we provide a completely novel analysis that is more explicitly applicable in predicting response probability in ways that are useful in regulatory decision-making. Further, comparing multiple methods that have been used in other studies provides an important evaluation across analytical methods for response analyses at the individual level to identify behavioral change-points for use in exposure–response functions.
MATERIALS AND METHODS
Study area and general field methods
This study was part of a long-term, multi-disciplinary research collaboration – the Southern California Behavioral Response Study (SOCAL-BRS). The CEEs presented here used several different experimental treatments with tagged blue whales during summer and autumn months (June–October) from 2010 to 2014 in coastal and offshore areas of the Southern California Bight. Within years, CEEs were conducted on different days (with two exceptions in 2010, where two CEEs were conducted within days at locations >10 nm apart) in different geographical locations or spaced in time to the extent possible to reduce the occurrence of multiple exposures over short periods in the same area.
Detail on the SOCAL-BRS field methodology is provided in Southall et al. (2012, 2016) and is summarized here. Small (∼6 m) rigid-hull inflatable boats (RHIBs) were used to locate, tag and obtain positional and behavioral observational data for focal whales. A central research platform (M/V Truth; Truth Aquatics, Santa Barbara, CA, USA) supported many research components, including the portable experimental sound source, passive acoustic listening systems, and visual observers on an elevated (7 m) observational platform directly above the ship's bridge. Visual observers supported RHIBs in locating focal whales and monitoring marine mammal exposures during CEEs to meet specified permit requirements. Individuals were identified visually and from photos in the field, and in post hoc analyses to the extent possible using long-term photo identification records.
All research activities for this study were authorized and conducted under US National Marine Fisheries Service permit 14534; Channel Islands National Marine Sanctuary permit 2010-004; US Department of Defense Bureau of Medicine and Surgery (BUMED) authorization; a federal consistency determination by the California Coastal Commission; and authorizations AUP-06 and AUP-08 from Cascadia Research Collective's animal care and use committee (IACUC).
Quantifying individual blue whale behavior
Individual blue whale behavior was measured during phases defined as before, during and after CEEs using a combination of high-resolution tag sensors and detailed focal follow procedures (see Southall et al., 2012; Goldbogen et al., 2013). Tagging effort was concentrated on sub-adult or adult animals; no young calves (estimated by experienced field researchers as being less than 6 months of age) or mothers with young calves were tagged. Several types of motion sensing and acoustic tags were used. For the large majority of whales, DTAGs (versions 2 and 3) (Johnson and Tyack, 2003) were used. These tags included broadband hydrophones (<0.1 Hz to >100 kHz sensitivity) sampled at rates of 48–240 kHz depending on the tag type and configuration. Two whales in the first year of this experiment were tagged with B-Probes, sampled at rates of 20 kHz (see Oleson et al., 2007). For each tag type, hydrophones were either calibrated directly or sensitivity was determined from calibrated tags of the same type. Acoustic records included environmental sounds, instances of calls produced by tagged and other whales (see Goldbogen et al., 2014), known exposures to experimental stimuli, and other incidental anthropogenic noise including vessel noise and (in several instances) non-experimental military sonar of multiple types outside CEE periods. Tag-measured received levels (RLs) were quantified for both tag types using the same approach. The maximum RMS sound pressure level for each exposure stimulus within any 200 ms analysis window over the one-third-octave band was centered at 3.7 kHz, which contained the predominant sound energy of all exposure stimulus types (as in Tyack et al., 2011; Southall et al., 2012; DeRuiter et al., 2013; Goldbogen et al., 2013). Additionally, cumulative sound exposure levels (cSEL; in dB re. 1 µPa2 s) were measured as integrated sound energy across all received exposure stimuli (as in DeRuiter et al., 2013).
Fine-scale, three-dimensional movement data from individual diving, foraging, and other behavioral and kinematic parameters were obtained from pressure transducers and inertial measurement units at sampling rates from 5 to 250 Hz for DTAGs (Johnson and Tyack, 2003) and 1 Hz for B-Probes (Goldbogen et al., 2006; Oleson et al., 2007). For the DTAGs with higher sample sensor resolution, the following tag-derived measurements were used for analyses: depth (m); absolute heading (deg); heading variance (unitless); minimum specific acceleration (MSA; m s−2); vertical and horizontal speed (m s−1); feeding rate (lunges dive–1); and feeding lunge rate (lunges h−1). Heading variance was derived as relative variability between instantaneous absolute heading and median heading within each minute of tag data. The MSA was derived from three-axis accelerometers as an integrated metric of overall acceleration (Simon et al., 2012). For the two B-Probe deployments with lower sensor sample resolution, slightly different parameters were measured and used in analyses described below, including depth, fluking acceleration (m s−2) and overall speed (m s−1). For both tag types, the instantaneous velocity was determined by regressing the measured flow noise from tags against the orientation-corrected changes in depth during stable ascending or descending portions of dives; this was calibrated for each individual tag deployment and tag orientation within the deployment (as in Cade et al., 2018). The instantaneous velocity was then multiplied by either the instantaneous pitch cosine (to obtain horizontal speed) or sine (for vertical speed) (Goldbogen et al., 2006). Feeding lunges were manually identified based on dive profiles, tri-axial body acceleration and flow noise (as in Goldbogen et al., 2013). Given differential sensor sampling rates across tag types and sampling periods, all variables other than lunge rates were decimated to 1-Hz resolution. The minimum sampling rate across all tags (1 Hz) was sufficient to describe the most important biological relevant behaviors (feeding and diving).
Once animals were tagged, focal individual tracking commenced to obtain accurate spatio-temporal surfacing positions. Focal animal surface positions at known times were determined from: known RHIB locations combined with range and bearing measurements to animals, measured from a precision laser range finder (Leica Vector, Viper II); known animal surface locations based on recent surface footprint locations; or, in cases where direct measurements were not possible, visually estimated range and bearing from known RHIB locations to focal whales. Error in surface positions was estimated to be <10 m from directly measured locations and tens to hundreds of meters for visual estimates of range and bearing, depending on conditions and range from visual observers to whales. Focal whale positions were used to generate time-series maps of animal movement and relative (over-ground) speed estimates used in expert evaluation of potential response severity.
Synoptic environmental data
The overall vessel configuration and experimental paradigm were described in detail by Southall et al. (2012). However, subsequent to the original experimental design described therein was the inclusion of additional parameters related to the environmental contexts in which CEEs occurred.
Calibrated measurements of noise associated with SOCAL-BRS vessel operations were made under controlled, standardized conditions that were representative of typical field configurations. Remotely deployed drifting acoustic buoys supported passive acoustic recorders using both a primary surface float and an isolated smaller secondary float. Shock-reducing bungee cords were suspended from the secondary float, to which recorders were attached. Loggerhead DSG recorders (Loggerhead Instruments, Sarasota, FL, USA) were suspended to depths of ∼30 m depending on the angle of the suspension line (small sea anchors were used to maintain a vertical orientation) and tension in the bungee. The DSG recording units were affixed with HTI-96 hydrophones (High Tech Inc., Long Beach, MS, USA) with a nominal sensitivity of −180 dB re. 1 V µPa−1 and had a nominal 20-dB pre-amplifier gain; the recording unit had a resulting flat sensitivity of −160 dB re. 1 V µPa−1 (±3 dB) between 16 Hz and 30 kHz. Recording buoys were deployed on three occasions in offshore locations (200–500 m water depths) in areas near to where CEEs were conducted. Recordings were obtained over 3 days in sea state 2–4 conditions; data presented here were obtained from the lowest possible sea state condition. Both RHIBs (Ziphid and Physalus) were instructed to pass by the surface float suspending recorders at a range of ∼100 m at speeds of 5 and 10 kn. This range was commonly the distance at which focal follows before, during and after CEEs were conducted. The RHIBs traveled at variable speeds during focal follows, depending on the behavior of the individual being followed, with 5 kn being a typical speed and 10 kn likely closer to a maximum speed. The central research vessel (M/V Truth) was also instructed to pass recorders at ∼100 m range and speeds of 5–10 kn, which represented more of a worst-case scenario during CEEs (because the vessel was stationary and usually much further away), but was more realistic in context of environmental prey mapping. Additionally, the M/V Truth was instructed to position ∼1 km from recorders and maneuver as if suspending the simulated MFAS sound source. These measurements provided received sound levels associated with the operation of the sound source vessel at typical distances (range) from whales during CEEs, in isolation from the experimental signals used in CEEs. For vessel passes, 1-min acoustic recordings centered on the time of the closest point approach were selected for analysis. For each 1-min sample, one-third-octave band RMS levels (dB re. 1 µPa) were then computed for each 1-s interval. Median values of all 60 samples were then calculated and are presented as representative noise levels that would be received by a whale at a relatively shallow depth (∼30 m) and in typical proximity during approaches from each vessel. For the stationary M/V Truth maneuvering at ∼1 km range from recorders, 2-min acoustic recordings during the confirmed time of maneuvering were used. Similarly, for each sample, one-third-octave band RMS levels (dB re. 1 µPa) were computed for 1-s intervals. Median values of 120 samples were then calculated and are presented as representative noise levels that would be received by a whale at a relatively shallow depth (∼30 m) and in typical proximity during maneuvering of the M/V Truth for sound source deployments during CEE approaches. These values were then compared with comparable measurements of ambient noise made using the same methods during the same day and under similar conditions, with no experimental or other vessels operating within at least 3 km of recording buoys.
For some feeding whales during 2011–2014 CEEs, active acoustic methods were used to measure krill distribution and density in the proximity of feeding whales immediately before and after CEE sequences. The general approach in obtaining these measurements is described here; detailed methods for the collection and analyses of prey data are provided by Friedlaender et al. (2014, 2016) and Hazen et al. (2015). Once a tag was deployed on a focal whale and as conditions allowed, a pre-exposure prey mapping survey was conducted at or near (typically within ∼100 m) recent, known tagged whale surfacing positions. Across whales, this period lasted for 30–75 min prior to the onset of each full CEE sequence. This complete CEE sequence included three sequential 30 min phases (pre-exposure baseline, exposure and post-exposure; see below), each of which occurred in the absence of active acoustic sampling (i.e. echosounders were not active during CEE sequences). Following the CEE sequence, a second 30–75 min active acoustic prey mapping survey was conducted. Given the clear importance of prey distribution in the behavior of feeding whales and in their responses during CEEs demonstrated by Friedlaender et al. (2016), we sought to evaluate the available prey distribution data in the context of potential responses even though contextual prey data were not available for all CEEs. Thus, we use prey data when available to provide additional context to the derived response likelihood that was conducted uniformly for all whales.
The experimental methods and specifications for the experimental sound source used in CEEs for this study are described in greater detail by Southall et al. (2012) and summarized within the context of other recent studies using CEEs to study behavioral responses of marine mammals to sonar by Southall et al. (2016). Essentially, a standard before–during–after (A–B–A) experimental design (with 30 min phases for up to a total of a 90 min full experimental sequence) was used to quantify potential changes in individual movement, diving, feeding and other aspects of behavior where individual noise exposure was controlled and known.
Provided that numerous specific criteria were met regarding visibility, sea state, proximity to shore or other vessels, absence of very young calves, and other factors, the M/V Truth was positioned at a range (typically 1000 m) estimated to result in maximum received RMS sound pressure level at the focal whale of 160 dB re. 1 µPa. In instances where multiple tagged whales were being monitored but were not in the same social group, a focal individual was selected in terms of positioning the sound source while a second tagged individual was followed by a second RHIB, but at some (typically greater) range that was less explicitly controlled. The experimental sound source was then deployed to a depth of 25 m and transmitted one of two signal types (MFAS: max. 210 dB re. 1 µPa @ 1 m; or PRN: max. 206 dB re. 1 µPa @ 1 m) at 25 s intervals during CEEs (see Southall et al., 2012). Signals were ramped up from an initial source level of 160 dB re. 1 µPa @ 1 m in 3 dB increments to the maximum source level for each respective signal type within the first ∼7 min of exposure and were maintained at that level for the remainder of the CEE. Total exposure duration was a maximum of 30 min, but some exposure intervals were terminated early as a result of mitigation requirements (e.g. other animals swimming within 200 m of the active sound source) or because of equipment failure.
Following the completion of controlled noise exposure sequences, monitoring from archival tags and visual focal follow methods was maintained for at least 30 min. Early in this period, the experimental sound source was recovered, and the M/V Truth was directed to maintain a comparable range (∼1000 m) and speed relative to the focal whale (as done during the pre-exposure sequence). The RHIB maintained a comparable range and approach in the post-exposure as was done during the pre-exposure and exposure sequences. Complete CEE sequences thus consisted of constant monitoring using tags and visual follows of individuals from RHIBs during the consecutive 30 min pre-exposure, exposure and post-exposure sequences. During these periods, the sound source vessel was mobile at a deliberately comparable range and relative orientation for the pre- and post-exposure but stationary (drifting) during the exposure period.
The primary research objective was to assess the potential responses of blue whales to military sonar. Consequently, and given the novelty of the study, a disproportionate number of CEEs were conducted with MFAS stimuli. Following the first five exposure sequences during 2010 with MFAS, a 2:1 ratio of MFAS to PRN stimuli was used and tested in randomized order. While the primary experimental control was within the pre-exposure–exposure–post-exposure experimental design, a smaller number of complete ‘control’ sequences were conducted in which the full sequence was replicated and the sound source deployed but no noise stimuli were presented during the ‘exposure’ phase (Table 1).
In a single instance, a tagged blue whale was monitored while a CEE was conducted in coordination with an operational Navy ship (USS Dewey-DDG 105) using full-scale MFAS (SQS-53C). Given the higher source level (235 dB re. 1 µPa @ 1 m), in situ noise propagation modeling was conducted to position the vessel much further away from the individual in order to obtain the same desired maximum received level (∼160 dB re. 1 µPa). A relative orientation was selected such that the ship was generally approaching the whale but was not directed precisely toward it, and no course adjustments were made during transmissions. The ship transited a direct course at 8 kn and, given the inability to gradually increase the source level as was done with the experimental sonar, a slightly longer exposure period (60 min) with corresponding 60 min duration of pre-exposure and exposure phases was implemented.
Provided that tagged whales were being monitored according to specified criteria and conditions, CEEs were conducted irrespective of the animal's behavioral state at the time of exposure. To categorize each individual's behavioral state at the beginning of each CEE, the following post hoc criteria were used based on tag sensor data to define deep-feeding, shallow-feeding and non-feeding: the presence of a single foraging lunge during the baseline period was used to indicate a feeding state for the CEE; and any dive depth exceeding 50 m was used to distinguish deep from shallow diving.
Some CEEs were not fully completed, either because of tag failure or detachment, loss of visual contact with individuals for long periods, or premature termination of noise exposure resulting from required termination protocols or equipment failure. Because of the difficulty in obtaining large sample sizes for such experiments under field conditions, incomplete sequences were retained within partial analyses when possible. Where individuals were successfully monitored with tags and visual observations through the pre-exposure and at least half (15 min) of the exposure period, the CEE was included. Behavioral response analyses were conducted, although without the ability to evaluate potential recovery from any responses during post-exposure periods. This is an additional benefit of individual-based time-series analyses over a synthetic analytical approach.
Behavioral response analyses
Individual blue whale behavior and potential responses during noise exposure periods were evaluated in parallel using two different analytical approaches: a structured expert evaluation and a quantitative statistical analysis. Methods for each are discussed below and results are presented within each analytical method by individual and evaluated together based on CEE stimulus type and animal behavioral state at the start of CEEs.
Expert scoring analyses
A structured evaluation of selected, standardized data streams using methods derived by Miller et al. (2012) based on the Southall et al. (2007) response severity scaling developed by was conducted by two independent groups of subject matter experts, each containing three of the co-authors (group 1: A.F., A.K.S., J.A.G.; group 2: J.C., A.N.A., G.S.). Each group was provided synoptic time-series behavioral information in the form of annotated maps of individual spatial movement (from RHIB-based focal follows) and selected kinematic and behavioral parameters in time-series plots (extracted or derived from tag records). For DTAGs (40 of 42 individuals), these included: depth (m), feeding rate (lunges dive−1), MSA (m s−2), absolute heading (deg) and horizontal speed (m s−1). For the two B-Probe deployments, these included: depth (m), fluking acceleration (m s−2) and overall speed (m s−1). As in Miller et al. (2012), many of the scorers were involved in the original fieldwork and thus may have had some recollection of events during CEEs (although some occurred over 4 years prior to expert scoring). In order to minimize any biases resulting from experience, scorers in this study were blind to the individual whale ID, date and location of CEEs, exposure treatment, or precise timing of RLs of exposure signals, and CEEs were presented to groups in randomized order in terms of the date that the experiment was conducted. Experimental phases (pre-exposure, exposure and post-exposure) for each CEE were identified in all data plots provided to each scoring group. This allowed scorers to evaluate behavior in pre-exposure baseline conditions, identify potential behavioral changes during exposure at specified times, and assess whether any identified behavioral changes persisted throughout and/or after noise exposure. The three members of each group collectively evaluated these data plots and annotated maps and time-series data plots for each CEE. Maps showed the position of the experimental sound source at the start and end of the CEE, every surface location collected by RHIBs during individual focal follows identified in each CEE phase (with times shown for the first position in each phase), and a 1000 m radius around the source at the onset of exposure for scale.
Scorers were instructed to evaluate the annotated maps and data plots for each CEE and to identify any behavioral changes to the nearest minute that occurred based on the descriptions specified in the severity scale. Criteria for temporal descriptors were as follows: brief or minor changes were identified as those returning to baseline conditions during exposure; moderate duration changes were identified as those not returning to baseline conditions until into the post-exposure period; and extended duration changes were those not observed to return to baseline within the post-exposure period. If multiple changes were identified, all were reported based on visual inspection of plots. The two groups independently evaluated each CEE collectively and came to a consensus agreement about any identified behavioral changes, the time at which they occurred, and a confidence score (low, moderate or high) as to the overall severity score(s) for each CEE. Where no behavioral responses were identified, a severity score of 0 was assigned. Where multiple responses were identified, all were reported, but the most severe (highest score) was used as the resulting overall score for that CEE. Neither Southall et al. (2007) nor Miller et al. (2012) identified an increase in feeding as an adverse behavioral change. Because this was not included within the severity scale, when it occurred it was not systematically reported and scored by expert scoring groups here. It was noted on multiple occasions as a change but was not scored as an adverse reaction.
After each group independently completed their evaluation of all CEEs, both groups met to compare results. An adjudicator (B.L.S.) was selected to mediate the combined group discussion and served to break any irreconcilable disagreements that occurred about severity scores between groups. A consensus behavioral response severity score (0 for no response; 9 for most severe response), a confidence score (low, moderate or high) and specified exposure times for any changes were identified for all MFAS, PRN and control (no noise) sequences. If a behavioral response was identified, the time of the response was used to derive exposure RLs (maximum RMS and cSEL to that point within the CEE).
Exposure–response probability functions were then generated using recurrent event survival analysis to assess time-to-event changes using marginal stratified Cox proportional hazards models fitted to the severity score data (see Harris et al., 2015 for full details of model application to severity score data). These models combine the results from individual CEEs to estimate the likelihood of response as a function of exposure RL (in cSEL) and behavioral or contextual covariates. Models were fitted to broad categories of response severity levels (i.e. low, moderate, high) to ensure sufficient data to support the exposure–response functions. The resulting hazard models provide a relationship between exposure level and the probability of response at different severity levels, while accounting for selected contextual variables. Similar analyses have been conducted for pilot whales, killer whales and sperm whales (Miller et al., 2012; Harris et al., 2015), as well as humpback whales (Sivle et al., 2015).
Given data limitations for shallow and non-feeding behavioral states, the Cox proportional hazards models were only fitted to data from animals that were deep feeding in the pre-exposure period. For these cases, the first occurrence of each response level (severity scores 1–3, 4–6, 7–9) was determined based on consensus expert scored results for each CEE for inclusion in the models. For CEEs with a severity score of 0 (no response), the cSEL for the entire exposure sequence was used and the data were labeled as right-censored, meaning that no response was detected up to this exposure level. We fitted models to data from all CEEs associated with deep-feeding animals and included source–animal range (m) at the start of the exposure phase and signal type (MFAS or PRN) as covariates. Observations were assumed to be correlated within individuals but independent between individuals. The standard errors of the model estimates were corrected for the correlations within individuals using a grouped jack-knife procedure (Therneau and Grambsch, 2000). All possible model combinations from the null model through to two-way interaction terms were fitted and Akaike's information criterion (AIC)-based model selection was used. For the selected model, the proportional hazards assumption was verified (Kleinbaum and Klein, 2010; Harris et al., 2015). Analyses were conducted in R version 3.0.2 (https://www.r-project.org/) and exposure–response functions were generated as survival curves from the fitted models using the survfit function package (https://CRAN.R-project.org/package=survival).
Mahalanobis distance statistical analyses
A Mahalanobis distance (MD) method (Mahalanobis, 1936; see DeRuiter et al., 2013) was also used to statistically test for change-points in whale behavior. This approach involves the calculation of an integrated statistical distance-based metric that summarizes synoptic dive parameters from tag data and quantifies how they differ over time from those present within a specified baseline period (e.g. the pre-exposure period). The MD metric is a scale-invariant integrated ‘difference’ from baseline behavioral parameters calculated in multi-dimensional space and accounting for correlations between dimensions. It is calculated within a sliding temporal window across all dive parameters to identify the specific time (if any) at which overall behavior changed. A window duration of 5 min (a conservative average dive duration for blue whales across all behavioral states) was selected with an MD value calculated every 25 s (corresponding to the interval between the onset of individual noise transmissions during CEEs). The MD calculations require a variance–covariance matrix to quantify statistical relationships among all variables. We calculated this matrix for each whale using the full dataset for the entire deployment, excluding an initial 15-min period estimated (based on nominal blue whale diving behavior) to account for any tagging effects (based on Miller et al., 2009). The inclusion of the full dataset, including and following CEE periods, was deemed necessary to provide sufficient samples to accurately estimate matrix parameter values. It was also considered a conservative choice, in that if behavioral changes during or following exposure were such that the variance–covariance structure was altered, the MD analyses would be less likely to detect it when using the full dataset than if only pre-exposure data had been used.
The following behavioral parameters (all quantified from individual animal-borne tags) were used as input variables in calculating MDs. For DTAGs, this included: circular variance of heading (25 s window), MSA (m s−2), vertical speed (m s−1), horizontal speed (m s−1) and feeding lunge rate (lunges h−1, 15 min window), all at 1 Hz resolution. For the two B-Probe deployments, this included: overall speed (m s−1) and feeding lunge rate (lunges h−1) at 1 Hz resolution. Dive data from the 30-min pre-exposure period (where other contextual factors including experimental vessel presence were similar to those during exposure) were used as comparison baseline data; this period also began at least 15 min post-tagging. When a tagged whale was near the surface, all data points that were collected shallower than 10 m were replaced with median parameter values from the baseline period to result in MD values near zero. This was to account for artifacts introduced by noise in some input data streams, most notably accelerometer-based metrics. This effectively pulls MD values toward 0 as the proportion of data points obtained at shallow depths in a time-window increases. The MD was then computed between (1) average behavioral data parameters for the baseline period and (2) average data values within the 5 min sliding comparison window.
Exposure and post-exposure periods were then evaluated to determine whether an individual behavioral change occurred, when it began and when it ended. MD values exceeding the maximum value observed during the pre-exposure period were identified as behavioral changes. For consistency with the expert scoring severity assessment, detected changes associated with the onset of or increase in foraging were not considered responses that would have any potential negative effects for individuals. Therefore, they were not included in the expert severity scoring options and were not reported as detected changes.
A total of 48 CEE sequences were conducted for individual whales involving MFAS, PRN or no-noise ‘control’ exposures in (primarily) coastal and offshore areas spanning the Southern California Bight (Fig. 1). Data from six sequences in which tags detached prematurely or CEE sequences were terminated before 15-min of exposure were not included in this analysis as they failed to meet specified experimental criteria; the remaining 42 sequences met these criteria and were analyzed. These occurred within 33 discrete CEEs, as nine of these sequences involved two concurrently tagged and followed animals. In seven of these instances, simultaneously tagged whales were separated from one another and were followed by separate boats. In two cases, simultaneously tagged individuals occurred within close proximity and were being tracked within the same focal follow, although one of these the animals was later determined to be in different behavioral states during exposure. Four individual whales were later revealed through photo identification to have been exposed in two separate CEEs within the same year. In each scenario, CEEs were spaced by several days or weeks. Furthermore, in each case, animals received different treatment types and were in different behavioral states for subsequent exposures. This likely reduced, but did not eliminate, the potential that behavioral responses during the second CEE may have been influenced to some degree by exposure to the initial CEE.
The 42 discrete, randomized CEE sequences evaluated here were conducted during 2010–2014 field efforts within different exposure treatments and behavioral state contexts. The resulting distribution of CEEs conducted for individuals within these three different behavioral states for each treatment type are summarized in Table 1. Representative examples of different types of behavioral response results for three individual whales are provided (Fig. 2).
The results of CEE 2011-01 on 29 July 2011 with individual bw11_210b are shown in time-annotated maps and MD data plots with received cSEL (in dB re. 1 µPa2 s) in Fig. 2A,B. This was a deep-feeding blue whale exposed to MFAS at a source–whale horizontal range (at the start of the exposure) of 1.2 km. Clear changes in behavior were detected with both MD and expert scoring methods (high confidence scores) at virtually the same time (15:28–15:29 h), corresponding to a received cSEL of 119 dB re. 1 µPa2 s. Changes identified by adjudicated expert scoring included horizontal avoidance of sound source (severity score 7) and moderate cessation of feeding (severity score 6) (see Table S1 for expert scoring details). The results of CEE 2011-06 on 6 August 2011 with individual bw11_218b are shown in Fig. 2C,D. This was a deep-feeding blue whale exposed to PRN at a source–whale range (at the start of the exposure) of 5.6 km. No changes in behavior were detected with either MD or expert scoring methods (high confidence scores), despite a relatively high received cSEL of 168 dB re. 1 µPa2 s (see Table S1 for expert scoring details). The results of CEE 2013-06 on 26 July 2013 with individual bw13_207a are shown in Fig. 2E,F. This was a shallow-feeding blue whale within a control sequence conducted at a source–whale range of 0.5 km. No changes in behavior were detected with expert scoring methods (moderate confidence scores), although the presence of increased feeding was noted (see Table S1 for expert scoring details). The increase in feeding rate resulted in a gradual increase in the MD metric relative to the pre-exposure baseline condition and was thus detected as a change. As in several other instances where whales initiated or increased feeding during CEEs, the MD-detected change was noted, but was not considered a conflicting result to the expert scoring evaluation because an increase in feeding was not defined as an adverse behavioral response (Southall et al., 2007; Miller et al., 2012).
Expert scoring and MD results are presented for each treatment type and behavioral state category for each individual blue whale (Table 2). Received exposure levels for each whale either at identified change points or (where none were detected) maximum values for CEE sequences are also provided (Table 2). For CEEs with identified responses cSEL values at identified change points ranged from 97 to 155 dB re. 1 µPa2 s. Maximum cSEL values for CEEs where no change was identified ranged from 134 to 171 dB re. 1 µPa2 s. Source–whale range varied from 0.4 to 7.7 km for the simulated MFAS and 19.5 km for the single operational vessel MFAS signal, with a median range of 1.2 km. There was no significant correlation within experimental sound types (MFAS, PRN) across CEEs between RL and source–whale range.
The largest number of individual CEE sequences analyzed (n=29) occurred for blue whales engaged in deep-feeding during pre-exposure periods. Whales were most likely to respond during MFAS CEE sequences, with a similar overall proportion of individuals identified as changing behavior during exposure by both expert scoring (8 of 13) and MD (9 of 13) methods. A lower proportion of deep-feeding whales responded when exposed to PRN (4 of 11 in expert scoring analysis; 5 of 11 for MD) and almost no responses were detected in deep-feeding control sequences (0 of 5 for expert scoring; 1 of 5 for MD).
For a subset of deep-feeding whales (n=21), prey distribution and density were measured before and after CEE sequences to provide an environmental context for interpreting responses in this behavioral state. Given the knowledge of the importance of this contextual relationship, we include three examples of whale behavior and contextual prey data to illustrate how these measurements provide additional insight into changes in whale behavior and the interpretation of potential response (Fig. 3).
For bw11_210b on 29 July 2011 (Fig. 3A; Fig. S3), prey patch depth and density remained similar both before and shortly following the CEE (2011-01) in the area where the whale was feeding. Both expert scoring groups identified very similar behavioral changes with high confidence scores at approximately the same time as one another and similar to the MD analysis (see Table S1 for expert scoring details), which identified a clear change relative to not only the pre-exposure condition, but the entire behavioral record for this individual (including pre-CEE prey sampling periods). Given the similarity in the prey environment before and at least immediately after the CEE, these identified changes (avoidance and cessation of feeding) are unlikely the result of changes in the prey environment (from the exposure or otherwise). However, subsequent changes in the overall prey environment (more schools identified at various depths) and/or changes in the local prey environment based on the whale's geographic location may have also influenced whale behavior, particularly well after the CEE.
For bw11_218b on 6 August 2011 (Fig. 3B; Fig. S4), prey patches after the CEE (2011-06) were shallower than those measured before the CEE sequence. This whale appeared to progressively decrease its feeding depth and continue to feed during the CEE as it moved into an area with shallower patches. This gradual decrease in whale diving depth was not identified by either expert-scoring group as a behavioral response during the CEE (Table S1). A behavioral change point was identified within the MD analysis (see Fig. S4, where the MD trace crosses the dashed line representing the pre-exposure baseline value used as the response threshold), although this was a small increase above the pre-exposure baseline period and it was of smaller magnitude than the MD spike in this metric identified just after the pre-CEE prey sampling period.
For bw13_207a on 26 July 2013 (Fig. 3C; Fig. S5), prey patches measured around the CEE (2013-06) in the area where the whale was feeding were deeper and less dense following the CEE sequence than before exposure. The animal maintained a similar feeding depth before and during the exposure sequence but increased its feeding rate and switched to deeper feeding after the CEE, which also continued during the post-exposure prey sampling period. Neither expert scoring group identified any behavioral change in this CEE, but there was a discernible change detected using the MD method, associated with an increase in foraging during the exposure phase relative to the defined baseline (pre-exposure) period (see Table S1 for expert scoring details). These MD values were of similar magnitude to those measured during both prey sampling periods (before and after the full CEE sequence).
Cox proportional hazards models were fitted separately to responses of severity scores between 4–6 and 7–9; responses with severity scores of 1–3 were insufficient to apply this process. The Cox proportional hazards model selected by AIC for severity scores 4–6 retained only source–whale range as a covariate (ΔAIC=1.34), although its effect was not significant (P=0.316). The selected model met the proportional hazards assumption (global P-value from Chi-square test=0.079). The model selected by AIC for severity scores 7–9 was the null model (ΔAIC=1.03), with the model including source–whale range being the second best model according to AIC. Given the interest in understanding the role of source–whale range in the probability of responding, model results from the selected model for severity scores between 4 and 6 and the second-best model for severity scores between 7 and 9 were used to produce predicted exposure–response probability functions in terms of received exposure level for the two different response severity levels (moderate severity: 4–6; high severity: 7–9). In order to illustrate the relationship with source–animal range, response probability functions were calculated for the ranges over which most CEEs were conducted (1–5 km) (Fig. 4). These prediction plots suggest that the probability of a moderate response (severity 4–6) as a function of RL decreases rapidly as range increases, but the wide confidence intervals indicate substantial uncertainty in this relationship. The relationship is much less pronounced for high severity responses (severity 7–9), hence the selection of the null model.
Shallow-feeding and non-feeding whales
The second largest number of individual CEE sequences analyzed (n=8) occurred for blue whales engaged in shallow feeding during pre-exposure periods. No whales (0 of 7) were determined to change behavior during MFAS exposure by either expert scoring or MD methods. No PRN sequences were conducted for shallow-feeding whales. No responses were detected by either analytical method during the single shallow-feeding control sequence.
The fewest number of individual CEE sequences analyzed (n=5) occurred for non-feeding blue whales, although most of these individuals were determined to have an adverse behavioral response during CEEs across both methods. For MFAS CEE sequences, expert scoring determined such a response in one of two whales, whereas MD analyses detected adverse responses for both individuals. For PRN CEEs, expert scoring determined an adverse behavioral response in one of three non-feeding whales, whereas all three individuals were identified to have such a response using MD methods. No control sequences were conducted for non-feeding whales.
Vessel noise characterization
Median values of vessel noise were calculated for the closest point of approach for all vessels during each condition. These values were compared for each condition for the RHIBs Ziphid and Physalus with comparable measurements of ambient noise made using the same recorders and methods during the same day and similar conditions, with these vessels operating at much further ranges from recording buoys (Fig. S1). Ambient noise measurements were also compared for each passage condition for the M/V Truth with comparable measurements of ambient noise made using the same recorders and methods during the same day and similar conditions, with this vessel operating at much further ranges from recording buoys (Fig. S2A,B). For the stationary M/V Truth maneuvering at ∼1 km range from recorders, median noise values were calculated relative to ambient noise during the same day and similar conditions (Fig. S2C). Both RHIBs and the M/V Truth were clearly detectable over ambient noise for both speeds at these close ranges, with different relative spectral distribution of noise energy at different speeds for each vessel. Based on the associated noise levels and frequencies and typical ambient noise during non-vessel periods, their operation is likely audible to subjects over ranges typical during CEEs, particularly the RHIBs at their typical operating speeds and ranges from animals. However, as a part of the experimental design during the pre-exposure (baseline), exposure and post-exposure sequences, these represent relatively continuous levels of additional noise exposure. During sound source deployment, the M/V Truth conducted small maneuvers to remain stationary. The measurements of ambient noise during this period demonstrated that these maneuvers and the presence of the vessel were not discriminable over noise measured using the same recording system in the absence of the M/V Truth. That is, although vessels were likely audible during their operation, particularly during pre- and post-exposure periods, when the M/V Truth was following focal animals, during exposure periods, noise from the sound source vessel received by experimental subjects was predominately or exclusively the result of experimental exposures.
This study generated the largest sample size (n=42) for any experimental behavioral response study involving sonar conducted to date for any marine mammal species (Southall et al., 2016). Although the number of individual CEEs conducted in some behavioral states and treatments were limited, dozens of controlled individual experiments were conducted using high-resolution movement and acoustic sensors for individuals in well-defined exposure contexts. These results provide direct and robust means of evaluating how an endangered species responds to noise exposure, including simulated and actual military MFAS signals that have been associated with lethal responses in other species. The analytical approach provides a direct means of quantifying individual behavior and behavioral responses within known noise exposure conditions in such a way that probabilistic response functions may be generated in light of important contextual variables. Such data provide an empirical basis for modeling efforts to evaluate potential consequences of disturbance at broader population scales (King et al., 2015; McHuron et al., 2018; Pirotta et al., 2018).
Blue whales responded to noise in some but not all CEE sequences (19 of 37 for MD analysis; 14 of 37 for expert scoring) and in almost no control (no-noise) sequences (1 of 6 for MD analysis; 0 of 6 for expert scoring). Treatment types had variable sample sizes, but responses were generally equally likely to occur for MFAS and PRN exposures. Other than a single instance detected only with the MD method, none occurred during control (no noise) sequences. Nine CEEs involved exposure of multiple individuals, although nearly all of these included animals in separate groups. A small number of CEEs involved paired individuals or subsequent exposures to the same individuals and in two instances in the first year of the study animals could have been remotely exposed to an earlier CEE prior to being the focal animal in a subsequent CEE later in the day. Although these could call into question the treatment of all individuals as independent samples, they were treated as such here (rather than excluding individuals) given the small number of instances relative to the overall sample size. Further, we took into consideration the fact that in all but one instance these CEEs all involved differences in individual behavioral state and/or treatment type.
Responses generally included short-term changes in diving behavior, small-scale (a few kilometers) horizontal avoidance of sound source location and/or cessation of feeding activity. Recovery to typical pre-exposure behavior in most CEEs typically occurred within the post-exposure phase. However, the short-term and relatively rapid nature of recovery should be considered within the context of acknowledged differences between the MFAS from an experimental source and operational MFAS. The experimental MFAS is stationary, includes a ramp-up escalation of the source level, and the overall duration is relatively short (tens of minutes). Operational MFAS training involves much louder and constant levels and can occur over many hours or even days in the case of multi-ship operations (see Moretti et al., 2014). It can also occur at any hour of the day and throughout the year, whereas CEEs here were only conducted during daylight hours in the summer and autumn.
Two different analytical approaches were applied to evaluate behavioral changes from baseline conditions within individuals using high-resolution, time-series kinematic and acoustic data. This approach included both quantitative statistical change-point methods and structured expert scoring assessment of deviations from baseline conditions during exposure by subject matter experts. The MD method is inherently objective in that it simply identifies changes in a suite of variables from baseline (pre-exposure) conditions and is thus equally likely to detect a behavioral change associated with a presumably positive outcome (e.g. an increase in foraging behavior) as a presumably negative outcome (cessation of feeding). Further, the selection of a response ‘threshold’ for MD strongly affects the probability of statistically detecting a behavioral response. Here, a fairly low MD value was selected as a change-point threshold, namely, an MD value within the exposure period exceeding that measured during the pre-exposure period. This results in a higher likelihood of identifying a behavioral response than if an alternate threshold were selected (e.g. two standard deviations exceeding the pre-exposure maximum) or if MD values during exposure exceeded the pre-exposure maximum value across the entire tag record. However, the intent here was to identify a discernible change in behavior during an exposure period with a similar context as pre-exposure conditions (e.g. local environmental variables, proximity of vessels) rather than aiming to identify a change that was more unusual than any other change measured for that or any other blue whale. Not surprisingly, the MD method was more likely to detect a change than expert scoring, both in controls and exposures. However, once detected changes associated with the onset of feeding (presumably not an adverse behavioral change) were discounted, results were quite similar across individuals. Some differences were still observed, but for 32 of 42 CEEs (76%), the methods agreed as to whether an adverse behavioral change occurred (where changes associated with the onset of feeding were excluded). Further, detected changes tended to occur at similar exposure times and associated RLs. Expert scoring methods were consequently consistent with the MD method in identifying behavioral changes, but this approach also has the advantage of being descriptive and identifying changes associated with various types of behavior (movement, feeding), including variability in response severity and the level of confidence in discerning response both within and between groups. Although both methods have advantages and limitations, the general agreement here was encouraging, and having used both methods provides more comprehensive insight into changes during experimental exposures. Future studies should consider integrating objective statistical change-point analyses (e.g. MD results) within expert evaluation of potential responses.
These findings demonstrate the kinds of context-specific differences in behavioral response identified by Ellison et al. (2012). Along these lines, they also complement and expand upon the findings of Goldbogen et al. (2013) and DeRuiter et al. (2017) regarding the importance of behavioral state in terms of response probability for blue whales, specifically the increased likelihood of response in deep-feeding animals. This study provides a different perspective on this behavioral state dependency in evaluating individual response type and severity for known exposure conditions for a relatively large sample size. Given these observations, we note the contextual differences between the simulated MFAS and some kinds of operational MFAS sources such as the SQS-53C sonar used in one CEE here; there are greater contextual similarities between the experimental source and other common operational military MFAS sources such as helicopter-dipping sonars. The experimental MFAS has proven useful in demonstrating previously unknown aspects of behavior, response and context dependency in these species, but, as we have shown, differences in exposure parameters can influence response probability. Additional research, some of which has been conducted and some of which is underway, is needed to further evaluate the importance of contextual differences in sound source type (e.g. source level, movement, spectral features) and proximity. This approach with individual animals where exposure range was known allowed for a quantification of behavioral response probability as a function of proximity to the sound source (Fig. 4) for the ranges tested. Deep-feeding whales had a higher response probability when located closer to the sound source for comparable RLs, although there is considerable uncertainty within the relationships and insufficient data to test this relationship for other behavioral states. Given the available data at this point, a simple relationship between source range, RL and response probability across all whales does not appear to exist. Further evaluation of the potential range-dependence identified within this study using a dedicated experimental design to test and further resolve these seemingly important range–RL relationships is needed before firm conclusions can be drawn. Specifically, additional studies should explicitly evaluate different dimensions of the RL–range space, including potential changes during near but quieter exposure conditions.
Whale dive depth has been closely linked to changes in prey patch depth, thus prey can both mediate the response to sonar playbacks when prey are dense and confound potential responses when prey distributions are not known. Although a direct quantitative comparison is not possible for all individuals, given the absence of before and after prey data in some cases, our results were consistent with Friedlaender et al. (2016) in suggesting that the behavior of feeding blue whales is broadly influenced by features of the prey environment in ways that likely mediate responses to CEEs. Specifically, two of the three instances where the MD detected CEE responses were potentially a result of changes in prey while expert scoring classified 0 of the 3 as a CEE response (see Table S1 for additional details). This highlights a potential strength in expert scoring in identifying specific aspects of a response in the absence of known important contextual variables. Changes in prey patch depth have been shown to result in commensurate changes in whale dive depth, and for some individuals, the likelihood of a behavioral response to navy sonar during a playback is reduced with increased prey density while foraging.
Many regulatory efforts to evaluate the effects of noise on marine mammals have primarily or exclusively used received noise exposure level as a predictor of response probability and have sought to develop more robust predictive associations. As illustrated by Ellison et al. (2012), a host of contextual factors can influence behavioral responses to noise. Several key contextual influences were identified here (and see Goldbogen et al., 2013; Friedlaender et al., 2016; DeRuiter et al., 2017) that have strong effects on whether and how endangered blue whales respond when exposed to military MFAS signals or PRN of similar frequency and duration. Responses were mediated by a complex interaction of the animal's behavioral state at the time of exposure, features of the environment and the relative proximity of sound sources. Without identifying behavioral state using objective, quantitative metrics (e.g. dive depth, presence of foraging lunges) and considering this as a relevant contextual variable, it would have been much more difficult to unravel the complexity of these relationships across studies. Identifying this, within certain contexts, indicates that an increase in RLs is in fact associated with an increase in response probability. Although this complexity is not yet fully understood, relating response probability, exposure level and behavioral state dependency will enable a more insightful and informed understanding of exposure–response relationships. This does not mean that each behavioral state and/or prey contextual condition must be informed by distinct and empirical exposure–response risk functions for management applications. Rather, integrated risk functions within behavioral states (e.g. foraging, traveling) and a small subset of contextual covariates (e.g. range) might be informed by targeted experimental studies in some species where relatively large sample sizes may be obtained (see Southall et al., 2016; Southall, 2017).
These results provide further evidence and increased resolution on how baleen whales respond to noise exposure. They also provide much-needed direct measurements of behavioral responses in an endangered species commonly exposed to MFAS within important habitat areas off California. As has been noted in other studies (see Southall et al., 2016; Southall, 2017), results from locations where sonar exposure is common are likely much different from the behavioral responses of animals from areas where sonar exposure is uncommon or absent. Although blue whales are likely low-frequency specialists, they can and do respond to sounds presented to them with primary energy in the 3–4 kHz range associated with many MFAS systems found in commercial, naval and recreational platforms. Whales that do respond appear to recover to typical behavioral patterns relatively quickly based on the results from these CEEs, and their probability of response should be considered given the contextual dependencies described in this study. With increased energetic demands and needs for high-density prey, even the cessation of feeding for a short time could have consequences for the fitness of these large animals (see Goldbogen et al., 2013). If they are chronic, they could manifest as population-level effects. Future experimental studies and targeted monitoring informed by these results should focus on the energetic and, in turn, biological consequences of behavioral responses across different behavioral states.
This 5-year study represented a major portion of the overall SOCAL-BRS project and could not have been conducted without the tireless support of many dedicated field personnel. We sincerely appreciate the Truth Aquatics team for their sustained involvement and support. We would also like to specifically acknowledge the contributions of: Kristin Southall, Peter Tyack, Jay Barlow, Shannon Rankin, Fleur Visser, Annie Douglas, Erin Falcone, Katy Laveck, Jeff Foster, Todd Pusser, Glenn Gailey, Doug Nowacek, Louise Burt, John Hildebrand, Ron Morrissey, Greg Juselis, Jolie Harrison, Tami Adams, Sarah Wilkin, Ned Cyr, Frank Stone, Bob Gisiner and Mike Weise. We also thank two anonymous reviewers, who greatly improved the quality of the manuscript.
Conceptualization: B.L.S., A.F., J.A.G.; Methodology: B.L.S., S.L.D., A.F., A.K.S., J.A.G., E.H., C.M.H., D.M., S.G., J.C.; Software: J.A.G., E.H., C.C., S.F., D.E.C.; Formal analysis: B.L.S., S.L.D., A.F., A.K.S., J.A.G., E.H., C.C., D.E.C., A.N.A., C.M.H., G.S., S.G., J.C.; Investigation: B.L.S., S.L.D., A.F., A.K.S., J.A.G., A.N.A., G.S., D.M., S.G., J.C.; Resources: J.A.G., C.C.; Data curation: B.L.S., J.A.G.; Writing - original draft: B.L.S.; Writing - review & editing: B.L.S., S.L.D., A.F., A.K.S., J.A.G., E.H., C.C., S.F., D.E.C., A.N.A., C.M.H., D.M., S.G., J.C.; Visualization: S.L.D., A.F., J.A.G., C.C., S.F., D.E.C., C.M.H., J.C.; Supervision: B.L.S., J.C.; Project administration: B.L.S., D.M., J.C.; Funding acquisition: B.L.S., J.C.
Primary funding for the SOCAL-BRS project was initially provided by the U.S. Navy’s Chief of Naval Operations Environmental Readiness Division and subsequently by the U.S. Navy’s Living Marine Resources Program. Additional support for environmental sampling and logistics was also provided by the Office of Naval Research, Marine Mammal Program. Additional support to enable S. Guan's participation was provided by the U.S. National Oceanic and Atmospheric Administration.
The authors declare no competing or financial interests.