Noise-induced hearing loss in marine mammals: A review of temporary threshold shift studies from 1996 to 2015

Over the last few decades the environmental effects of intense, human-generated noise have become an increasing concern. It is now clear that underwater sound from construction, shipping, oil and gas exploration and production, military testing and training, and other human activities may reach sufficient amplitudes and possess sufficient durations to adversely affect the health and/or behavior of many marine animals. The effects of underwater sound on the hearing abilities of marine mammals are of particular concern, primarily because of the protected nature of these animals in many countries and the large extent to which they rely upon underwater sound for communication, orientation, and foraging (e.g., see Tyack and Clark, 2000).

Animals exposed to sufficiently intense sound may exhibit an increased hearing threshold, called a noise-induced threshold shift (NITS). If the hearing threshold eventually returns to normal, the NITS is called a temporary threshold shift (TTS). The magnitude of a TTS is a function of the recovery time—the amount of time that has elapsed since the cessation of the noise exposure. To indicate the recovery time associated with a specific TTS measurement, numeric subscripts are normally used: TTS₄ indicates a TTS measured 4 min after the exposure. If the hearing threshold does not return to normal, but leaves some residual NITS, the remaining NITS is called a permanent threshold shift (PTS).

Extensive studies of NITS in humans and small terrestrial mammals have been conducted to identify the fundamental noise parameters that affect hearing loss and to develop safe exposure guidelines for people working in noisy environments (e.g., see Kryter, 1970; Henderson et al., 1976; Melnick, 1991). These studies have generally consisted of long-term correlational analyses between workplace noise exposure and the occurrence of PTS, clinical examinations of TTS in humans, and laboratory studies of TTS and PTS in small terrestrial mammals. The similarities between the auditory systems of humans and small terrestrial mammals (e.g., chinchilla) have enabled a large body of data to be collected that is directly applicable to developing noise exposure criteria for people; however, differences between the auditory systems of marine and terrestrial mammals, as well as differences in sound propagation in-air and underwater, prohibit direct application of terrestrial damage risk criteria to marine mammals. For these reasons, acoustic exposure guidelines for marine mammals have primarily relied on measurements of NITS in representative marine mammal species, rather than extrapolation from terrestrial species.

The earliest marine mammal NITS experiments were sponsored by the U.S. Navy Program Executive Office for Undersea Warfare, to learn about potential auditory effects in marine mammals incidentally exposed to sonars (Ridgway et al., 1997). This was soon followed by a larger program initiated by the U.S. Office of Naval Research to investigate marine mammal TTS at three facilities in the Unites States: (1) the Hawai'i Institute of Marine Biology (HIMB), (2) Long Marine Lab (LML) at the University of California, Santa Cruz, and (3) the U.S. Navy Marine Mammal Program (MMP) at the Space and Naval Warfare Systems Center (SSC) Pacific. Since then, a number of TTS experiments have also been conducted by investigators at: (4) the Sea Mammal Research Company (SEAMARCO) in the Netherlands, (5) the Russian Academy of Sciences (RAS), and (6) Fjord & Baelt (FAB) in Denmark. Tests at HIMB have featured a bottlenose dolphin (Tursiops truncatus) exposed to relatively long duration, octave-band noise and short-duration tones (Nachtigall et al., 2003; Nachtigall et al., 2004; Mooney et al., 2009b; Mooney et al., 2009a). Work at LML has focused on relatively long duration, broadband noise, both in-air and underwater, with a harbor seal (Phoca vitulina), Northern elephant seal (Mirounga angustirostris), and California sea lions (Zalophus californianus) (Kastak and Schusterman, 1996; Kastak et al., 1999; Kastak et al., 2005b; Kastak et al., 2007). Experiments at the MMP have featured bottlenose dolphins, belugas (Delphinapterus leucas), and California sea lions exposed to relatively short-duration tones and impulsive sounds underwater (Ridgway et al., 1997; Finneran et al., 2000; Schlundt et al., 2000; Finneran et al., 2002; Finneran et al., 2003; Finneran et al., 2005a; Finneran et al., 2007b; Finneran et al., 2010a,b; Finneran and Schlundt, 2010; Finneran and Schlundt, 2013; Finneran et al., 2015). SEAMARCO tests have utilized harbor seals and a harbor porpoise (Phocoena phocoena) exposed to octave-band noise, tonal signals, and simulated impulses (Kastelein et al., 2012a; Kastelein et al., 2012b; Kastelein et al., 2013a; Kastelein et al., 2013b; Kastelein et al., 2014a; Kastelein et al., 2014b; Kastelein et al., 2015a; Kastelein et al., 2015b). The RAS studies featured belugas and Yangtze finless porpoises (Neophocaena phocaenoides asiaeorientalis) exposed to half-octave band noise (Popov et al., 2011b; Popov et al., 2011a; Popov et al., 2013; Popov et al., 2014; Popov et al., 2015). Finally, tests at FAB exposed a harbor porpoise to single underwater impulses (Lucke et al., 2009).

Although the experimental paradigms have been similar, the various groups have used different species, exposure parameters, and testing methods, highlighting the need for periodic review and synthesis of the existing marine mammal TTS data. This paper reviews the methods employed by the groups conducting marine mammal TTS experiments and the relationships between the experimental conditions, the noise exposure parameters, and the observed TTS. This paper is not intended to be a chronological history of efforts in this area, but rather an overview of the experimental methodologies and results. When possible, data from the different experiments are pooled to provide a more complete picture of the state of knowledge at this time for the effects of noise on pinniped and odontocete hearing.

The basic experimental approach for TTS measurements in marine mammals is analogous to that used to measure TTS in terrestrial mammals. Tests begin with a pre-exposure hearing threshold measurement at one or more frequencies. This is followed by the fatiguing sound exposure—the sound that may cause TTS. Finally, post-exposure hearing thresholds are measured at one or more frequencies. The TTS at each frequency is typically defined as the difference (in decibels) between the post-exposure and pre-exposure thresholds at that frequency, though some studies (e.g., Mooney et al., 2009b; Mooney et al., 2009a) have used an average “baseline” threshold instead of the pre-exposure threshold. To assess the recovery of hearing after a TTS, and to verify that the shift was in fact temporary, post-exposure thresholds are typically measured multiple times, over a period that may extend for several days.

In the absence of any existing data, the early experiments were designed to determine exposure parameters necessary to cause TTS, starting at low-level exposures and gradually increasing exposure level and/or duration until a measurable threshold shift was observed. This paradigm has also been repeated for novel stimuli and/or inexperienced subjects.

Table I lists the species, age, and sex of the subjects in the various studies. Pinniped species include the harbor seal (Pv; one male, two females), California sea lion (Zc; two males, two females), and Northern elephant seal (Ma; one female); odontocete species tested include the bottlenose dolphin (Tt; five females, four males), beluga (Dl; four males, four females), harbor porpoise (Pp; two males), and Yangtze finless porpoise (Np; one male, one female). The limited access to marine mammals has resulted in a large proportion of the data arising from a relatively small number of individuals. Many subjects have also been conditioned to tolerate the intense noise exposure; as a result, behavioral observations of the animals' reactions to the noise exposures may have questionable relevance to other individuals in different environments or with different prior experience.

Ocean environments were utilized at HIMB, MMP, and FAB. Several studies at the MMP were conducted in floating, netted enclosures located in San Diego Bay. Because of the variable ambient noise conditions (ship and boat traffic as well as other biological sources), early tests included intentionally generated broadband masking noise to provide a floor effect and keep thresholds consistent from session to session (Ridgway et al., 1997; Finneran et al., 2000; Schlundt et al., 2000; Finneran et al., 2002; Finneran et al., 2003). Later tests in San Diego Bay did not use masking noise (Finneran et al., 2015). Tests at HIMB were also performed in an open bay environment (Kaneohe Bay) with relatively high background noise levels, but no additional masking noise (Nachtigall et al., 2003; Nachtigall et al., 2004; Mooney et al., 2009b; Mooney et al., 2009a). Testing at FAB was conducted in a “semi-natural” outdoor pool with a natural bottom, two concrete walls, and two netted walls with free flow of water from the Baltic Sea (Lucke et al., 2009).

Tests at LML, SEAMARCO, and MMP were also conducted in pools with low ambient noise levels and no artificial masking noise (Kastak et al., 1999; Finneran et al., 2005a; Kastak et al., 2005b; Finneran et al., 2007b; Finneran et al., 2010a,b; Finneran and Schlundt, 2010; Kastelein et al., 2012a; Kastelein et al., 2012b; Finneran and Schlundt, 2013; Kastelein et al., 2013a; Kastelein et al., 2013b; Kastelein et al., 2014a; Kastelein et al., 2014b; Kastelein et al., 2015a; Kastelein et al., 2015b). TTS measurements by the RAS group utilized small water-filled tanks, with the animals suspended in slings during testing (Popov et al., 2011b; Popov et al., 2011a; Popov et al., 2013; Popov et al., 2014; Popov et al., 2015).

For in-air measurements, Kastak and Schusterman (1996) measured hearing thresholds in the presence of ambient background noise, while Kastak et al. (2007) performed tests in a hemi-anechoic chamber with very low ambient noise levels.

Each of the underwater test environments offers some advantages and disadvantages. Very low ambient noise levels can be achieved in pools or tanks, but the relatively small volumes of water can result in complex sound fields and highly variable stimuli, especially for pure tones (Finneran and Schlundt, 2007). Ocean environments typically feature much greater depth compared to pools or tanks, but also higher and more variable ambient noise conditions, which can elevate pre-exposure thresholds. Humes (1980) found that broadband, Gaussian masking noise sufficient to elevate pre-exposure thresholds by 20 dB lowered measured TTS₂ by 5 dB, thus it is possible that larger amounts of TTS would be obtained in quieter environments (e.g., see Parker et al., 1976; Humes, 1980). However, natural sources of ambient noise are often comodulated rather than Gaussian (Branstetter and Finneran, 2008). Results of previous marine mammal TTS experiments with varying levels of Gaussian masking noise and comparisons between TTS measured in quiet pools and ocean environments have also been somewhat ambiguous (e.g., Schlundt et al., 2000; Finneran et al., 2005a). To date, there have been no systematic investigations of the effects of masking noise on TTS in the same individual marine mammals.

Hearing threshold measurements have been carried out using both psychophysical (behavioral) and electrophysiological techniques. The behavioral approaches are based on the use of operant conditioning techniques to train subjects to reliably report the presence of a hearing test tone by providing a specific response and to withhold that response in the absence of a tone (a “go/no-go” paradigm). Signal levels typically begin at supra-threshold levels and are adjusted from trial to trial using a modified up/down staircase procedure in which the following tone sound pressure level (SPL) is reduced after a detection and increased after a non-detection. Thresholds are based on the collection of reversals—the transitions from detections to non-detections or vice versa. The major differences in approach involve the specific type of response and the trial presentation schedule.

Behavioral tests at HIMB, LML, and SEAMARCO were conducted in the context of a single-interval experiment, with each interval defined to the subject and possessing either a single tone stimulus or serving as a stimulus-absent trial. Subjects manipulated physical devices (e.g., pressed a paddle) to indicate the presence of a tone and remained at the test station to indicate the absence of a tone. At the end of each trial interval subjects were rewarded if the response was correct. The single-interval experiment is simple and the resulting data are easy to analyze and lend themselves to signal detection analyses; however, the method is relatively slow, since each trial interval only yields a single yes/no response and the trainer interacts with the subject before and after every trial.

At the MMP subjects were trained to produce an acoustic response (e.g., whistle) to indicate the presence of a tone and to stay quiet otherwise. The experiments used a modified version of the Method of Free Response (MFR)—multiple trials were presented during a single, relatively long observation period. This is in contrast to the traditional, single-interval experiment where each trial has a clearly defined start and stop and is followed by a reinforcement period. The primary advantage of the MFR with an acoustic response is speed, since multiple trials can be presented without requiring the subject to move or interact with the trainer. Using this method, thresholds are typically obtained within 2 to 4 min. The disadvantages of the MFR are the ambiguity in reinforcement and the difficulty of properly characterizing the subject's response bias. Since multiple trials (containing multiple responses) are conducted before the reinforcement period, there is some ambiguity regarding what behavior is being reinforced. In the original implementation of the MFR (e.g., Schlundt et al., 2000), trial intervals were not clearly defined, thus signal absent trials did not exist and the calculation of a false alarm ratio was complex. To overcome these limitations, the implementation of the MFR has been modified over the years to include a false alarm metric derived for use with the MFR (Schlundt et al., 2000), methods to scale reward to a subject's performance during the trial interval (Finneran et al., 2005a), and a light to clearly define each trial within the long trial block (Finneran et al., 2010b).

The psychophysical approach has traditionally been considered the “gold standard” for hearing assessment, since the data provide a direct indication as to whether the subject can hear the test tone. However, application of this method requires time to condition subjects to reliably report the presence of a test tone at near-threshold SPLs. There are limited numbers of marine mammals available for this type of training, and as a result there have been relatively few individuals/species tested.

Electrophysiological measurements are based on the measurement of auditory evoked potentials (AEPs)—small voltages reflecting the neural activity within the auditory pathway. AEPs in small odontocetes (i.e., dolphins and porpoises) are measured using small, passive electrodes, similar to those used for electroencephalogram (EEG) measurements in humans, embedded in suction cups and placed on the head. AEP amplitudes are relatively small, often less than ∼1 μV for dolphins, therefore hundreds to thousands of short-duration stimuli are typically presented, and the resulting AEPs synchronously averaged, to increase the signal-to-noise ratio. Although AEPs can be elicited from a variety of sound stimuli, marine mammal TTS experiments have relied on the auditory steady-state response (ASSR), also called the envelope following response, generated by sinusoidal amplitude modulated (SAM) tones or repetitive, short-duration tones containing only a few cycles (“tone pips”). The ASSR is formed when stimuli are presented sufficiently fast, or modulated at a sufficient rate, that transient evoked potentials overlap and form a steady-state signal (Picton et al., 2003). Modulation rates used during AEP TTS testing have been relatively high (600–1000 Hz), meaning that the resulting ASSRs are formed from overlapping short-latency responses, primarily from the auditory nerve and brainstem. The fundamental frequency of the ASSR equals the SAM tone modulation rate or tone-pip repetition rate, so the signals may be analyzed in the frequency domain, rather than the time domain. This yields advantages in signal detection, since the modulation rates used with marine mammals (and hence the ASSR fundamental frequencies) are typically well above the predominant frequencies of the background physiological noise.

During ASSR threshold testing, stimuli are presented at various SPLs and the resulting spectral amplitude of the ASSR is measured at the modulation frequency or tone-pip repetition frequency. From these data, two techniques have been used to define the threshold. In the first, a linear regression is performed on the ASSR input/output (I/O) function (i.e., the ASSR amplitude versus stimulus SPL data). The stimulus SPL at the point where the extrapolated regression line crosses the abscissa is then defined as the threshold. In the second technique, an objective (and sometimes statistical) procedure is used to determine the presence/absence of the ASSR at each stimulus SPL. Threshold is then defined as the lowest SPL at which a response was detected, or the mean SPL over a collection of reversal points (a change from ASSR detection to non-detection, or vice-versa).

The original marine mammal TTS studies all utilized behavioral test approaches; however, AEP methods are becomingly increasingly common for hearing assessment. ASSR thresholds have been shown to correlate with psychophysical thresholds from the same individual (e.g., Yuen et al., 2005; Finneran et al., 2007a; Schlundt et al., 2007; Mulsow et al., 2011), but the ASSR thresholds are typically 5–15 dB higher than the behavioral thresholds. Direct comparison between AEP and behavioral measurements is confounded to some extent by differences between stimuli: AEP measurements utilize short-duration/relatively large-bandwidth stimuli, while relatively long-duration, narrowband stimuli are used for behavioral threshold measurements. The difference between thresholds also increases as the test frequency increases above or decreases below the region of best hearing sensitivity.

Despite the similarities between AEP and psychophysical audiograms, there are important differences between the two test approaches. Psychophysical methods are based on cognitive tasks that directly “ask” the animal if it can hear a test tone. The results are influenced by auditory signal processing throughout the entire pathway, from the outer ear to the auditory cortex, and directly relate to whether the subject can perceive the sound stimulus. In contrast, the AEP methods utilized in marine mammal TTS testing measure the level of activity along the ascending auditory pathway from the auditory nerve to the brainstem. The AEP threshold reveals the experimenter's ability to measure the neurological voltage resulting from the sound stimulus, not the animal's ability to perceive the sound—although the two are certainly related, they are not equivalent. While both AEP and psychophysical thresholds are heavily dependent on the acoustic noise, AEP thresholds also depend on the residual background noise in the EEG. Higher levels of noise resulting from greater muscular activity could result in a failure to detect the AEP despite no actual change in the AEP amplitude itself. It is important to keep in mind the differences between the two hearing assessment techniques: Psychophysical methods determine the level of sound at which a subject decides to respond a certain proportion of the time, while AEP methods determine the level of sound that produces a detectable evoked response, often at the level of the auditory brainstem. Behavioral thresholds may be influenced by not only the subject's sensitivity, but also by response bias and motivational state. Electrophysiological thresholds are affected by both the amplitude of the basilar membrane response and the degree of synchrony of the response, with more synchronous discharges of auditory nerves resulting in larger AEPs and lower thresholds.

Other factors to consider with AEP testing are the manner in which the sound stimulus is calibrated and the frequency content of the stimulus. Relatively long-duration tones used for psychophysical testing are almost always specified in terms of their rms sound pressure; however, no universally accepted standards exist for calibrating AEP stimuli. As a result, some investigators have used the peak-equivalent rms sound pressure for tone pips or pip trains, while others have calculated the rms pressure over the entire stimulus duration, even when the stimulus possesses low duty cycle. SAM tones possess energy at the fundamental (carrier) frequency ± the modulation frequency. For odontocetes, modulation frequencies are typically 600 to 1000 Hz. At frequencies near and above regions of best sensitivity, SAM stimuli may be considered narrowband (e.g., 50 ± 1 kHz represents a 4% bandwidth); however, at lower frequencies the bandwidth may become very large (e.g., 4 ± 1 kHz = 50% bandwidth). In addition, tone pip and tone-pip trains have broader frequency content and care should be taken when interpreting frequency-dependent effects, especially with SPLs well above threshold. Also, stimulus frequency content should be based on the projected acoustic signals, not the electrical driving voltage, since the dynamic response of the transducer may significantly affect the projected acoustic signal. For example, short-duration tone pips may produce ringing in the transducer, resulting in more cycles in the acoustic signal compared to the electric signal, and non-flat amplitude response of transducers may result in an enhancement of certain frequencies in the acoustic signal. When combined with the high-frequency bias present in short-latency AEPs, the “effective” frequency range may differ from the center of the electrical signal spectrum.

Exposure conditions (Table I) used in marine mammal TTS studies can be grouped into three broad categories: Relatively long-duration octave-band noise (OBN) or one-half octave-band noise (HOBN), short-duration tones, and very short-duration impulsive sounds. OBN with various center frequencies with durations from 20 to 55 min has been used with all three pinniped species, a dolphin, and a harbor porpoise. HOBN has been used with belugas and finless porpoises. Both noise types enable comparison with a large amount of human and terrestrial mammal data, where relatively long-duration noise is used to represent steady-state noise conditions in industrial settings. TTS studies have been performed with dolphins and belugas, and a harbor porpoise exposed to tones with durations ranging from 1 s to 1 h. Most of these studies have employed continuous exposures, though four studies (Mooney et al., 2009a; Finneran et al., 2010b; Kastelein et al., 2014a; Kastelein et al., 2015b) have used intermittent tones. Tonal signals may be used to represent the effects of military sonars, fish finders, depth sounders, and other sources emitting steady-state, narrowband signals. Dolphins, a beluga, a harbor porpoise, and sea lions have also been exposed to impulsive sounds—transient sounds with relatively high peak pressure, short duration, and broad bandwidth. Impulsive sounds have been generated using both coherent (i.e., non-impulsive) sources, such as piezoelectric transducers (Finneran et al., 2000; Kastelein et al., 2015a), and impulsive sources, such as a seismic water gun, seismic air gun and an electrical arc-gap device (Finneran et al., 2002; Finneran et al., 2003; Lucke et al., 2009; Finneran et al., 2015). Signals from these devices have been used to represent impulsive sounds to which marine mammals may be exposed from underwater explosions, marine seismic surveys utilizing air guns or water guns, and impact pile driving.

Methods to calibrate fatiguing sound exposures have varied across studies. Short-duration stimuli have typically been calibrated at the subject's listening position, under the assumption that only limited movement could occur during the exposure (e.g., Schlundt et al., 2000). Noise levels from longer-duration exposures have been estimated by comparing spatial variations in SPL to a subject's movement pattern during the exposure (e.g., Kastelein et al., 2012b) or by measuring the exposure using one or more hydrophones carried by the subject (e.g., Finneran et al., 2007b). In addition to spatial fluctuation in sound pressure caused by subject movements, received sound levels will also vary when subjects surface to breathe, potentially creating an intermittent exposure scenario.

Finally, in addition to noise exposures, most, but not all, marine mammal studies include control sessions where hearing thresholds are measured before and after simulated noise exposures to reveal variance in the TTS data in the absence of a fatiguing sound. A lack of true controls makes it difficult to distinguish “real” shifts in threshold from normal threshold differences due to inherent uncertainties in the threshold measurement process.

An important, yet often overlooked, parameter that affects the amount of TTS observed after a noise exposure is the methodology used to determine the pre- and post-exposure hearing thresholds. Just as AEP and behavioral methods may not result in identical thresholds, threshold differences determined using the two methods may also not agree.

Only two marine mammal studies have utilized both evoked potential and behavioral hearing assessments after the same intense noise exposure. Finneran et al. (2015) measured AEP and behavioral thresholds before and after dolphins were exposed to multiple impulses from a seismic air gun. Small amounts of TTS were obtained in the AEP thresholds, along with suppression of the AEP input-output functions, but no significant differences were seen in the behavioral thresholds. Using pure-tone exposures, Finneran et al. (2007b) reported large differences between TTS amounts determined using AEP and behavioral thresholds (Fig. 1). At the frequencies most affected by the noise exposure, the amount of TTS determined using AEP thresholds was always larger than that determined behaviorally, AEP threshold shifts required more time to recover, and AEP TTS of up to ∼10 dB were measured with no corresponding elevation in behavioral thresholds. The extent to which the observed differences depend on the exposure characteristics, amount of TTS, specific hearing test stimuli, or methodology is unknown. Terrestrial mammal data have shown varying degrees of agreement between AEP and behavioral TTS data (e.g., Benitez et al., 1972; Henderson et al., 1983); however, comparisons are complicated by use of the whole-nerve action potential in some terrestrial mammal studies and the brainstem AEP in others, differences in hearing test stimuli (e.g., clicks or tone pips), and differences in exposure durations and SPLs (Schmiedt, 1984; Abbas, 1988). Stimulus and response measurement parameters likely play a large role; e.g., Heffner et al. (2008) found better agreement between behavioral and AEP TTS and PTS when using noise burst stimuli rather than tone pips.

Despite the limited nature of the data in Fig. 1, the large potential differences observed between AEP and behavioral data challenge the validity of equating TTS data obtained with ASSR and behavioral methods. These data also suggest that the AEP measurements may provide a more sensitive indicator of the effects of noise—the AEP measurements can provide not only estimates of hearing threshold but also supra-threshold information in the form of input-output functions that may reveal changes in evoked response amplitude or latency even if thresholds are unaffected. These findings highlight the need for additional studies where both AEP and behavioral measurements are made after comparable noise exposures, using a variety of AEP stimuli. As more TTS studies are conducted utilizing AEP methods, the need to understand the relationship between the two data sets, and the proper manner to utilize both in developing acoustic safety criteria, will become increasingly important.

Auditory effects of intense noise exposure are not confined to the exposure frequency, but tend to spread to adjacent frequencies, especially higher frequencies as the exposure SPL and amount of TTS increase. The particular choice of hearing test frequency therefore affects the amount of TTS that is observed. Human and terrestrial mammal data show that for moderate and larger shifts, the maximum TTS occurs one-half to one octave above the exposure frequency (Cody and Johnstone, 1981; McFadden, 1986). This same trend has often been observed after marine mammal tonal and broadband noise exposures (Fig. 2).

For dolphins and belugas exposed to short-duration tones, Schlundt et al. (2000) reported most occurrences of TTS near one-half octave above the exposure frequency, though some TTS also occurred at the exposure frequency and an octave above the exposure frequency. More detailed investigations of the spread of TTS in dolphins, belugas, and finless porpoises exposed to tones and noise also reported maximum TTS close to one-half octave above the center frequencies of tones and noise bands, with progressively less TTS one octave above and at the exposure center frequency (Fig. 2, Nachtigall et al., 2004; Finneran et al., 2007b; Mooney et al., 2009b; Popov et al., 2011a; Popov et al., 2013). In contrast to these data, the maximum TTS in a harbor porpoise, sea lion, harbor seals exposed to OBN occurred at the exposure center frequency rather than above the exposure frequency (Kastak et al., 2005b; Kastelein et al., 2012a; Kastelein et al., 2012b).

Kastelein et al. (2014b) measured the frequency spread of TTS in a harbor porpoise as a function of the exposure SPL [Fig. 2(d)]. These data show a clear pattern of increasing upward spread of TTS as the exposure SPL increases, presumably as a result of the spread of the basilar membrane excitation pattern as the level of the fatiguing sound increased. This finding likely explains the discrepancies previously reported across investigations: Studies showing the maximum TTS at the exposure frequency typically involved relatively small amounts of TTS, often produced from relatively long-duration exposures. Longer-duration exposures may induce TTS at relatively low SPLs with limited upward spread and thus the major effects are concentrated near the exposure frequency. Shorter exposures require higher SPLs to induce TTS and thus tend to cause more upward frequency spread of the excitation pattern of the fatiguing sound, resulting in maximum TTS above the exposure frequency.

Regardless of the frequency of maximum TTS, measurable TTS can occur over a large frequency range, even from (relatively) narrowband exposures. For example, Finneran et al. (2007b) reported TTS in a dolphin over a frequency range of 20 to 50 kHz after exposure to a narrowband 20-kHz tone and Popov et al. (2013) reported TTS in belugas from 8 to 32 kHz after half-octave noise at centered at 11.2 kHz.

A systematic exploration of the frequency spread of TTS after impulsive noise exposures has not been performed in marine mammals; however the limited data for cetaceans shows TTS occurring at frequencies above the major frequency content of the impulsive noise (Finneran et al., 2002; Lucke et al., 2009; Kastelein et al., 2015a). It is likely that broadband exposures produce broadband TTS with an upward frequency spread in a manner similar to that seen with tones and noise bands.

As in terrestrial mammals, TTS generally increases with the noise SPL [Figs. 3(a), 3(d), and 3(g)]; however, the relationship is nonlinear and non-monotonic. Considering low SPL exposures, Ward et al. (1976) defined “effective quiet” as the highest SPL that would not produce a significant TTS or affect recovery from a TTS produced by a prior, higher-level exposure. To date, there have been no studies explicitly designed to measure effective quiet in a marine mammal; however, rough estimates can be made by noting the lowest SPLs that produced measurable TTS, regardless of exposure frequency or duration. This reveals that, at their most sensitive frequency, effective quiet would be less than 150–160 dB re 1 μPa for dolphins and belugas (Nachtigall et al., 2004; Mooney et al., 2009b; Popov et al., 2014), 140 dB re 1 μPa for Yangtze finless porpoises (Popov et al., 2011a), 124 dB re 1 μPa for harbor porpoises and harbor seals (Kastelein et al., 2012a; Kastelein et al., 2012b), and 174 dB re 1 μPa for California sea lions (Kastak et al., 2005b). As the exposure SPL increases beyond effective quiet, the relationship between TTS and SPL is exponential, so that TTS increases with SPL in an accelerating fashion. For higher exposure conditions, up to those producing 40 to 50 dB of TTS, TTS growth is generally linear with SPL.

TTS also generally increases with noise duration [Figs. 3(b), 3(e), and 3(h)] and increases linearly with the logarithm of time for moderate exposure durations and amounts of TTS [e.g., Fig. 3(h), Popov et al., 2014]. In terrestrial mammals, continued increases in exposure duration have been shown to ultimately fail to produce additional TTS, a condition referred to as asymptotic threshold shift (ATS). ATS has been hypothesized to represent the upper bound of PTS that could be produced by noise of a specific SPL, regardless of duration (Mills, 1976). Exposure durations sufficient to induce ATS in terrestrial mammals have generally been at least 4 to 12 h (Mills, 1976; Mills et al., 1979), much longer than the typical exposure durations used in marine mammal testing (Table I). As a result, no clear evidence of ATS has been seen in any marine mammal; however, it is likely that similar patterns of TTS growth would be found in terrestrial and marine mammals, including the occurrence of ATS. When ATS is taken into account, TTS growth with exposure duration is best described using exponential functions that feature a slow rate of growth at short durations, linear growth with the logarithm of time for short to moderate durations, and ultimately a plateau at long durations (Keeler, 1968, 1976; Mills et al., 1979; Patuzzi, 1998).

Sound exposure is defined as the time integral, over the duration of the exposure, of the instantaneous sound pressure-squared [American National Standards Institute (ANSI), 1994]; sound exposure level (SEL) refers to the sound exposure expressed in decibels, referenced to 1 μPa² s in water or (20 μPa)² s in air [American National Standards Institute (ANSI), 1989]. Because TTS depends on both the exposure SPL and duration, it has become convenient to use SEL as a single numeric value to characterize a noise exposure and to predict the amount of TTS. For interrupted or intermittent exposures, the cumulative SEL, defined as the total SEL calculated over the “on-time” of the noise exposure, is often used to characterize the exposure. SEL is linearly related to the SPL and logarithmically related to the exposure time, meaning that SEL will change on a 1:1 basis with SPL, and increase or decrease by 3 dB for each doubling or halving of exposure time, respectively. SEL has been shown to be reasonably effective at predicting the occurrence of TTS, especially over a limited range of exposure durations [Figs. 3(c), 3(f), and 3(i)]. Marine mammal studies have shown that the amount of TTS increases with SEL in an accelerating fashion: At low exposure SELs, the amount of TTS is small and the growth curves have shallow slopes. At higher SELs, the growth curves become steeper and approach linear relationships with the noise SEL. Marine mammal TTS growth data are fit reasonably well by the equation

where T is the amount of TTS measured a few minutes after exposure, L is the SEL, and a and b are fitting parameters [e.g., Figs. 3(c), 3(f), 3(i), Kastak et al., 2007; Kastelein et al., 2012b; Popov et al., 2014]. This particular function has an increasing slope when L < b and approaches a linear relationship for L > b (Maslen, 1981). The linear portion of the curve has a slope of a/10 and an x-intercept of b. After fitting Eq. (1) to the TTS growth data, interpolation can be used to estimate the SEL necessary to induce a specific amount of TTS or the exposure associated with the “onset of TTS”—the exposure level sufficient to produce a small (often defined as 6 dB) amount of TTS [e.g., Southall et al., 2007; Department of the Navy (DoN), 2008]. Figures 4 and 5 and Table II summarize the existing marine mammal TTS growth data for those studies with TTS ≥ 6 dB and sufficient data to enable a fit of Eq. (1).

The TTS growth rate—the rate at which TTS increases with exposure level—can be estimated from the slope of the linear portion of Eq. (1). Growth rates obtained in this fashion have been used to predict exposure levels necessary to induce 40 dB of TTS, as an indicator of the potential for injury (Southall et al., 2007). TTS growth rates with increasing SEL may vary widely depending on the exposure frequency, duty cycle, and the hearing test frequency (Finneran and Schlundt, 2013; Kastelein, 2014; Kastelein et al., 2014a). For example, in a dolphin exposed to 16-s tones and tested ∼1/2-octave above the exposure frequency, growth rates varied from ∼0.2 dB/dB at 3 kHz to 4.5 dB/dB at 28 kHz (Finneran and Schlundt, 2013). Growth rates in a harbor porpoise exposed to 1–2 kHz linear FM tones and tested at 1.5 kHz varied from 0.5 dB/dB at 10% duty cycle to 2.8 dB/dB at 100% duty cycle (Kastelein et al., 2014a). When the same porpoise was exposed to 6.5-kHz tones, the TTS growth rate at 6.5 kHz was ∼0.3 dB/dB, while at 9.5 kHz the rate was 1.3 dB/dB (Kastelein, 2014).

For plane progressive waves, sound exposure is proportional to sound energy flux density, so the use of SEL is often described as an “equal-energy” rule, where exposures of equal energy are assumed to produce equal amounts of NITS, regardless of how the energy is distributed over time. Since SEL changes by 3 dB for each doubling or halving of exposure duration, the use of SEL or an equal energy rule can also be described as a “3-dB exchange rate” for acoustic damage risk criteria. This means that the permissible noise exposure SPL will change by 3 dB with each doubling or halving of exposure time; e.g., an equal energy rule means that if the permissible exposure limit is 150 dB re 1 μPa for an 8-h exposure, the limit for a 4-h exposure would be 153 dB re 1 μPa.

Marine mammal studies, like terrestrial mammal studies, have shown that the equal energy rule has limitations and is most applicable to single, continuous exposures with similar durations (Kastak et al., 2005a; Mooney et al., 2009b; Finneran et al., 2010a). The temporal pattern of noise exposure is known to affect the resulting threshold shift. It is also well-known that the equal energy rule over-estimates the effects of intermittent noise, since the quiet periods between noise exposures will allow some recovery of hearing compared to noise that is continuously present with the same total SEL (Ward, 1997). For continuous exposures with the same SEL but different durations, the exposure with the longer duration will tend to produce more TTS. This can be seen in Figs. 3(c), 3(f), and 3(i) by comparing the TTS for various exposure durations with the same SEL: In most cases the exposure with the longer duration resulted in more TTS; however, some recent data also suggest that the equal energy rule may breakdown in the opposite direction. At moderate- to long-exposure durations, further increasing the exposure duration, and decreasing the SPL to maintain equal SEL, may eventually cause the TTS to decrease [Fig. 3(l), Popov et al., 2014]. For these reasons, some studies have modeled marine mammal TTS as a function of both exposure SPL and duration, representing TTS growth as a surface rather than a curve (Kastak et al., 2007; Mooney et al., 2009b; Finneran et al., 2010a; Popov et al., 2014).

Despite these limitations, however, the equal energy rule continues to be a useful concept, since it includes the effects of both noise amplitude and duration when predicting auditory effects. SEL is a simple metric, allows the effects of multiple noise sources to be combined in a meaningful way, has physical significance, and is correlated with most TTS growth data reasonably well—in some cases even across relatively large ranges of exposure duration (e.g., Figs. 4 and 5). The use of cumulative SEL to predict the effects of intermittent exposures also errs on the side of caution, since it will always over-estimate the effects of intermittent or interrupted sources, which may be important when developing mitigation policies for protected species.

The frequency content of the noise exposure can have a dramatic effect on the induced TTS and the TTS growth rate. In particular, noise at low frequencies, well below an animal's most sensitive hearing range, will tend to produce less TTS compared to noise with the same SEL and duration but with higher frequency content. Figures 4 and 5 include examples of the manner in which the amount of TTS and the TTS growth rate may vary with exposure frequency in dolphins, belugas, and finless porpoises. In the dolphin, larger TTS and steeper growth rates were observed as the exposure frequency increased above from 3 to ∼10 kHz, with the steepest growth rate and largest TTS occurring after the 28 kHz exposures (Finneran and Schlundt, 2013). Belugas showed smaller growth rates at 90 kHz compared to 11–45 kHz (Popov et al., 2013). In finless porpoises, TTS amounts and growth rates also varied with exposure frequency, with less TTS observed after 22- and 90-kHz exposures compared to 32- and 45-kHz exposures (Popov et al., 2011a).

The onset of TTS—defined here as the exposure level necessary to produce 6 dB of TTS—also varies with exposure frequency. Onset TTS exposure levels can be estimated by first plotting measured TTS versus (cumulative) SEL (e.g., Figs. 4 and 5), then interpolating to find the exposure level required for 6 dB of TTS. Figure 6 shows the variation of onset TTS with exposure frequency for dolphins and belugas, for those datasets with TTS values bracketing 6 dB; i.e., no extrapolation was performed. For comparison, TTS measures from studies where the minimum amounts of TTS induced were larger than 6 dB are also shown. For these data, there is roughly a 20-dB difference between TTS onset from 3 to 14 kHz, highlighting the need to consider exposure frequency when assessing the potential effects of noise. For a harbor porpoise, a 30-dB difference in TTS onset was reported between exposures at ∼1.5 and 6.5 kHz (Kastelein et al., 2014a; Kastelein et al., 2014b).

Human noise damage-risk criteria have long used auditory weighting functions to capture the frequency-dependent nature of the effects of noise by emphasizing noise amplitude at frequencies where animals are more susceptible and de-emphasizing the amplitude at frequencies where animals are less susceptible (29 CFR 1910.95, 2008). The functions may be thought of as frequency-dependent filters that the noise is passed through before a single, weighted exposure level is calculated.

Human auditory weighting functions were derived from equal loudness contours—curves that show the combinations of SPL and frequency that result in a sensation of equal loudness magnitude in a human listener (Suzuki and Takeshima, 2004). Equal loudness contours are in turn created from data collected during loudness comparison tasks. Analogous tasks are difficult to perform with non-verbal animals; as a result, true equal loudness contours are available for only a single marine mammal (a dolphin) across a limited range of frequencies (2.5 to 113 kHz) (Finneran and Schlundt, 2011). In the absence of equal loudness contours for other species and at other frequencies, weighting functions to predict TTS onset have been estimated using estimated auditory bandwidth (Southall et al., 2007), measured auditory thresholds (Nedwell et al., 2007), and equal latency contours (Wensveen et al., 2014). Equal-loudness based functions agree closely with onset TTS data in dolphins (Finneran and Schlundt, 2013), but are limited in frequency extent; it is not known if the agreement would hold at lower frequencies where onset TTS would be expected to change more significantly with frequency. Threshold-based weighting functions capture the variation of TTS onset at the lower frequencies and are very intuitive; however, at higher frequencies, the exposure SELs required for onset TTS show less variation with frequency compared to audiograms; i.e., the iso-TTS contours exhibit some compression in this region and appear “flatter” than the audiogram. Data from dolphins and porpoises also suggest that there may be a frequency region of heightened susceptibility in odontocetes, from approximately 10 to 40 kHz, that does not mirror the audiogram. The available data therefore suggest that threshold-based methods are not appropriate for broadband weighting functions to predict TTS onset in odontocetes. Finally, equal latency-based approaches offer somewhat mixed results: At low SPLs the functions closely mimic threshold curves and thus offer little advantage beyond the thresholds themselves (Mulsow et al., 2015). At higher SPLs, the porpoise latency curves show more compression than exhibited in the (limited) TTS onset data available at lower frequencies (Kastelein et al., 2014a; Kastelein et al., 2014b; Wensveen et al., 2014). At the present time therefore, it appears that none of the available methods is completely satisfactory. TTS growth data at various exposure frequencies, for multiple individuals within a number representative species, thus represent a critical data gap.

The large variety of observed recovery rates, combined with differences in exposure parameters, hearing test methods, and subjects, has made the development of universal models for TTS recovery in marine mammals difficult. Early TTS studies, especially those employing novel stimuli, typically generated relatively small amounts of TTS, thus recovery information was limited. More recent studies have induced more TTS and provide more recovery information. At present, enough data are available, across multiple species, to begin to establish some governing relationships.

Figure 7 shows examples of recovery data from 11 studies utilizing dolphins, belugas, a harbor seal, and a harbor porpoise exposed to steady-state tones or narrowband noise (Finneran et al., 2010a,b; Popov et al., 2011b; Kastelein et al., 2012a; Kastelein et al., 2012b; Finneran and Schlundt, 2013; Kastelein et al., 2013b; Popov et al., 2013; Kastelein et al., 2014a; Kastelein et al., 2014b; Popov et al., 2014). As seen in terrestrial mammals, TTS measured in these studies typically decreases with increasing recovery time; however, the relationship is not linear but appears to be linear with the logarithm of recovery time. For dolphins, TTS has been reported to decline at a rate of 4 to 6 dB/decade when the initial shifts were ∼5 to 15 dB (Nachtigall et al., 2004; Finneran et al., 2007b; Mooney et al., 2009b), while similar initial TTSs have resulted in recovery rates from 4 to 13 dB/decade in a harbor porpoise and harbor seals (Kastelein et al., 2012a; Kastelein et al., 2012b; Kastelein et al., 2013b). Recovery rates of 15 to 23 dB/decade have been measured in a dolphin and belugas after 25 to 40 dB initial TTS (Finneran et al., 2007b; Popov et al., 2014). Across studies, recovery rates tend to increase with increasing initial TTS; i.e., larger initial TTSs tend to result in faster recovery rates, although total recovery times are still typically longer for larger initial shifts.

The recovery data in Fig. 7 were fit (nonlinear regression, OriginLab, 2010) with the equation

where t is the recovery time (in min), TTS(t) is the TTS (in dB) at time t, TTS₄ is the TTS (in dB) at t = 4 min, and m is the recovery rate (dB/decade). The fits were generally good, with the mean adjusted R² = 0.891 [standard deviation (SD) = 0.107]. Recovery rates varied over a wide range, from 2.9 to 26 dB/decade, and were linearly related to TTS₄ [Fig. 8(a)]:

with the best fit values c₁ = 0.52 ± 0.041, c₂ = 2.7 ± 0.69, and the adjusted r² = 0.629. Equations (2) and (3) can be used as a first approximation to TTS recovery in marine mammals exposed to steady-state noise and, if the uncertainties of the fitting parameters are taken into account, the bounds of the recovery function can also be estimated within some desired level of confidence [Fig. 8(b)]. When fit using all the data, Eq. (3) accounts for ∼ 63% of the variability in the recovery rate data; the remaining unaccounted variability is presumably related to differences in subjects, exposure parameters, and hearing test methodology. Slightly better fits of Eq. (3) to the recovery rate data were obtained by considering only the data for dolphins and a harbor porpoise exposed to noise or tones with frequency content from 1 to 4 kHz [Fig. 8(c)]. In this case, best fit values were c₁ = 0.46 ± 0.051 and c₂ = 3.7 ± 0.74, and the adjusted r² = 0.689.

Logarithmic functions such as Eq. (2) generally provide good fits to TTS recovery data from steady-state exposures (Luz and Hodge, 1971), however, they have limited applicability for small and large time values (Keeler, 1968; Botsford, 1971; Keeler, 1976). Equations (2) and (3) assume that recovery depends only on TTS₄ and is therefore independent of the exposure conditions that produced TTS—an assumption that has been shown to be invalid for human data (Melnick, 1991). Simple linear-log equations also cannot adequately fit more complex recovery patterns that have sometimes been observed (Fig. 9). These include recovery data containing regions with variable slopes (e.g., Finneran et al., 2010a; Popov et al., 2011b; Kastelein et al., 2012b; Kastelein et al., 2014a) and those exhibiting delayed TTS growth and/or recovery (e.g., Finneran et al., 2007b; Popov et al., 2011a; Finneran and Schlundt, 2013; Popov et al., 2013).

Complex TTS recovery patterns can be fit using modified multi-exponential functions such as

where n₁, n₂, T₁, and T₂ are fitting parameters (Patuzzi, 1998). Equation (4) was fit to the TTS recovery data in Fig. 7 using nonlinear regression (OriginLab, 2010), with T₁ and T₂ shared across all datasets and n₁ and n₂ allowed to vary for each dataset. Fits were generally good, with the mean R² = 0.933 (SD = 0.105). Best-fit values for the time constants T₁ and T₂ were 10 ± 0.22 min and 580 ± 34 min, respectively. Similar to the recovery rate from Eq. (2), the parameters n₁ and n₂ were linearly related to the amount of TTS₄ [Fig. 10(a)], and the dispersion in the n₁, n₂ data tends to become smaller when the range of subjects, exposure conditions, and test parameters narrow [Figs. 10(b) and 10(c)].

Both the linear-log and multi-exponential models provide good fits to some subsets of the existing recovery data; however, a deep understanding between the model parameters and the subject, exposure, and hearing test parameters remains elusive. Therefore, although descriptive models can be developed based on existing data, it is generally not known how well these models can be extrapolated to other species and exposure scenarios. Development of more robust models for recovery will first require a systematic effort to build datasets of TTS recovery curves measured from various combinations of exposure frequency, SPL, and duration. There is at present also very little understanding of the conditions that produce delayed growth or recovery. This may represent an especially critical data gap, since human data suggest that recovery functions featuring delayed TTS growth or recovery indicates more hazardous exposure conditions (Ward, 1976).

The manner in which noise energy is distributed in time has a critical impact on the resulting amount of TTS, since partial recovery of hearing may occur during the quieter intervals within the exposure (Ward, 1997). The addition of temporal pattern as an independent variable substantially complicates efforts to understand and predict TTS resulting from intermittent exposures. For marine mammals, this is compounded by the relatively few studies that have addressed this issue: Only four studies have examined the role of intermittency and/or exposure duty cycle on TTS in marine mammals (Mooney et al., 2009a; Finneran et al., 2010b; Kastelein et al., 2014a; Kastelein et al., 2015b). The results of these studies mirror those found with terrestrial mammals, primarily that TTS can accumulate across multiple exposures, but the resulting TTS will be less than the TTS from a single, continuous exposure with the same total SEL [Figs. 5(b) and 11].

A variety of approaches have been proposed to predict TTS arising from intermittent noise exposures. The approaches can be broadly separated into two categories, referred to here as “exposure summation” models and “TTS summation” models. Exposure summation models focus on combining the individual noise exposures, so that an intermittent noise exposure can be modeled as a single, continuous exposure with an “effective” SPL and duration. In contrast, TTS summation models focus on combining the TTS resulting from an individual exposure with the TTS remaining from any prior exposures.

The various exposure summation models differ in the manner in which the effective SPL and effective duration are derived. The most common technique is a cumulative or equal energy approach, where the intermittent exposure is treated as a single, continuous exposure with the same cumulative SEL of all of the individual exposures [Southall et al., 2007; Department of the Navy (DoN), 2008]. This technique does not account for recovery of hearing in quiet intervals and will therefore increasingly over-estimate TTS as the duty cycle decreases (i.e., as the relative amount of quiet increases). An alternative exposure summation model, termed the “single-ping equivalent” (SPE) method [Department of the Navy (DoN), 2001], treats an exposure to multiple sounds with equal durations as a single exposure whose SPL is related to the sum of the squared sound intensities (i.e., sound pressures raised to the fourth power), rather than the sum of the squared sound pressures as in the cumulative energy model. The SPE approach was designed to account for recovery between individual exposures and thus will predict less TTS than the cumulative energy model; however, the SPE has not been directly validated with TTS data (see Finneran et al., 2010b). For relatively rapid fluctuations in noise level, TTS can be estimated by averaging the amount that the exposure SPL exceeds the effective quiet SPL over the course of the exposure (Ward, 1997). Finally, Ward et al. (1959) treated any residual TTS existing at the beginning of a subsequent exposure as additional exposure time, a type of exposure summation model applicable to noise with constant SPL.

The simplest TTS summation model would be that of adding the individual shifts together, so the TTS from an intermittent exposure is equal to the sum of the TTSs from the individual exposures. The modified power law (MPL) model (Humes et al., 1988; Humes and Jesteadt, 1989, 1991), does not add TTS values directly, but instead transforms TTS values into quantities proportional to power or energy that are then raised to some power and added. The resulting TTS is found by performing the reverse transform on the sum. The MPL has been shown to provide good agreement to experimental data for human and terrestrial mammal TTS after exposure to a variety of intermittent noise conditions (Humes and Jesteadt, 1991; Macrae, 1993, 1994). Finneran et al. (2010b) found that the modified power law model (Humes and Jesteadt, 1989) fit the growth of TTS in a dolphin exposed to multiple, short-duration tonal noise exposures better than cumulative energy, SPE, or TTS summation; however, the data were very limited and it is not known to what extent this method would fit data from other test conditions. Clearly, additional data are needed regarding TTS growth after exposures with different SPLs, durations, and duty cycles.

The term “impulse noise” is generally used to denote any short-duration, high-amplitude sound with relatively broad frequency content and relatively fast rise time. Terrestrial mammal studies of the auditory effects of impulse noise have revealed that impulse noise may be particularly hazardous to hearing, and that the variability associated with NITS measurements is higher when using impulsive fatiguing sources (Henderson and Hamernik, 1986). In addition to the SPL, duration, and frequency content, the rise time and number of impulses will also affect the resulting amount of NITS (Henderson and Hamernik, 1986).

Few TTS studies have been conducted with marine mammals exposed to impulsive noise sources, and many of the studies have produced no effect or only small effects on auditory thresholds. Finneran et al. (2000) exposed dolphins and a beluga to single impulses from an array of underwater sound projectors designed to produce pressure signatures resembling underwater explosions, but found no TTS after exposure to the highest level the device could produce (SEL = 179 dB re 1 μPa² s). Similarly, no TTS was found in two California sea lions exposed to single impulses from an arc-gap transducer with SELs of 161 to 163 dB re 1 μPa² s (Finneran et al., 2003).

Two recent studies found only small effects of multiple impulses on odontocete hearing. Kastelein et al. (2015a) reported behaviorally measured mean TTS of 4 dB at 8 kHz and 2 dB at 4 kHz after a harbor porpoise was exposed to a series of impulsive sounds produced by broadcasting underwater recordings of impact pile driving strikes through underwater sound projectors. The exposure contained 2760 individual impulses presented at an interval of 1.3 s (total exposure time was 1 h). The average single-strike, unweighted SEL was approximately 146 dB re 1 μPa² s and the cumulative (unweighted) SEL was approximately 180 dB re 1 μPa² s, with peak SPL ∼ 180 dB re 1 μPa. Finneran et al. (2015) found no behavioral TTS in three bottlenose dolphins exposed to a sequence of 10 impulses produced from a seismic air gun. A 9-dB shift in AEP thresholds occurred in one dolphin at 8 kHz, and the amplitudes of auditory steady-state responses were suppressed at 4 and/or 8 kHz in two dolphins. The air gun impulses were produced at a rate of 10 s/impulse, with maximum cumulative SELs for 10 impulses of 193 to 195 dB re 1 μPa² s.

The most salient effects of impulse noise were reported for a beluga and harbor porpoise exposed to single impulses generated by seismic sources. Finneran et al. (2002) reported behaviorally measured TTSs of 6 and 7 dB in a beluga exposed to single impulses from a seismic water gun (unweighted SEL = 186 dB re 1 μPa² s, peak SPL = 224 dB re 1 μPa) and Lucke et al. (2009) reported AEP-measured TTS of 7 to 20 dB in a harbor porpoise exposed to single impulses from a seismic air gun (at TTS onset, unweighted SEL = 162 dB re 1 μPa² s and peak SPL = 196 dB re 1 μPa).

Although limited, the impulse TTS data appear to follow the trend of increasing spread of TTS to higher frequencies. The TTS in the beluga and the harbor porpoises occurred at frequencies above the predominant energy in the exposures, suggesting an upwards shift in TTS as one would expect based on terrestrial mammal data (Finneran et al., 2002; Lucke et al., 2009). It is also likely that the failure of air gun impulses to produce TTS in a dolphin at cumulative SELs higher than those producing TTS in a beluga exposed to a single impulse was related to the primarily low-frequency content of the exposures (Finneran et al., 2015). Various methods have been proposed to reconcile the frequency content of impulses with the (generally higher) region of best sensitivity in marine mammals, typically utilizing an auditory weighting function derived from narrowband TTS data, auditory thresholds, equal loudness contours, or equal latency contours (e.g., Southall et al., 2007; Finneran and Jenkins, 2012). Development of robust models for TTS growth and recovery from impulsive exposures will require systematic efforts to measure the amount of TTS produced by impulsive sounds with various combinations of SPL, SEL, duration, frequency content, and temporal pattern.

Many marine mammal species are amphibious, therefore effects of noise must be evaluated both in air and underwater. To relate the effects of noise exposures in air and underwater, Kastak et al. (2007) found that noise exposures with equal durations could be equated on the basis of a sea lion's sensation level in each medium; however, the comparisons were limited and it is not known to what extent this relationship could be generalized. More data would be necessary, for multiple species across a wide range of frequencies, to conclusively demonstrate this relationship.

One of the chief limitations of the marine mammal TTS data is that most data have been obtained from a relatively small number of individuals; as a result, little is known about the inter-individual variation expected to occur in measured TTS. Only a few studies have featured multiple subjects: For example, Finneran et al. (2013) found similar TTS in two dolphins exposed to 16-s, 3-kHz tones; however, Kastelein et al. (2012a) and Popov et al. (2013) found significant differences in TTS measured in harbor seals and belugas, respectively, exposed to octave or half-octave noise bands (Fig. 12). It is clear from the available data that some amount of inter-individual variation should be expected, that potential differences in auditory effects would likely vary with exposure parameters, and that larger variances across subjects would likely occur at higher exposure levels, where initial NITSs are larger.

Over the past twenty years much progress has been made in understanding the potential adverse effects of noise on the hearing of marine mammals; however, many data gaps remain. The most critical gaps involve the manner in which exposure frequency affects the resulting patterns of TTS growth and recovery. TTS growth curves at various frequencies are needed for representative species so that effective weighting functions can be developed to predict the onset of TTS and establish upper safe limits to prevent PTS for various noise frequencies. The noise sources of greatest concern—such as military sonars and seismic air guns—involve acute exposures to high-intensity, intermittent sounds, but significant questions remain regarding the rate of TTS growth and recovery after exposure to intermittent noise and the effects of single and multiple impulses. Data at present are also insufficient to construct generalized models for recovery and determine the time necessary to treat subsequent exposures as independent events. More information is needed on the relationship between AEP and behavioral measures of TTS for various stimuli. Finally, data on noise-induced threshold shifts in marine mammals are available for only a few species, and few individuals within these species, and questions remain about the most appropriate methods for extrapolation to other species.

The author thanks Carolyn Schlundt for a critical review of the manuscript and Jason Mulsow and Dorian Houser for helpful discussions on the content. The author also thanks Sam Ridgway and Robert Gisiner for reviewing and providing insight on the origins of the marine mammal TTS research efforts. Financial support for this effort was provided by the U.S. Navy Living Marine Resources (LMR) Program and U.S. Fleet Forces Command.