Bottlenose dolphin (Tursiops truncatus) discrimination of harmonic stimuli with range-dependent signal degradation^a)

Anthropogenic noise sources are diverse and increasingly common in marine ecosystems, and include commercial, industrial, scientific, and military sources (Richardson et al., 1995). However, with regards to the behavioral reactions of marine mammals, no noise source has reached the same level of public and scientific interest as mid-frequency active military sonars (e.g., U.S. military AN/SQS-53C, abbreviated as 53C hereafter, D'amico and Pittenger, 2009). These sonars are frequency-modulated (FM) signals with multiple harmonics, characterized by high source levels (exceeding 230 dB re 1 μPa @ 1 m for 53C types) and have been implicated in marine mammal strandings, primarily those involving sensitive cetacean species {i.e., beaked whales [Simmonds and Lopez-Jurado, 1991; Frantzis, 1998; Balcomb and Claridge, 2001; Department of Commerce (DOC) and Department of the Navy (DON), 2001]}.

Recent controlled exposure experiments have aimed at determining how marine mammals react to sonar exposure (BRS and 3S, see Southall et al., 2016 for a review). Following a pre-exposure baseline period, marine mammals are exposed to either simulated or actual military sonars. Behavioral metrics during and following exposure to sonar are then compared to pre-exposure periods in order to determine the extent of any behavioral reactions. The primary independent variable in controlled exposure experiments has typically been the sound pressure level (SPL) received by the tagged individual, aligning the results with regulatory efforts focusing on received SPL (National Marine Fisheries Service, 2009). Data from a tagged Cuvier's beaked whale (Ziphius cavirostris) demonstrated an incidental non-experimental exposure to distant 53C sonar without a notable response (Deruiter et al., 2013). This is similar to the findings of Southall et al. (2014) in which there was no detectable change in the behavior of a Cuvier's beaked whale exposed to 53C sonar at a distance of approximately 70 km. These reports suggest that behavioral responses to distant sources differ relative to those elicited by closer sources, even at comparable received SPLs.

The perception of source range based on acoustic cues has not been experimentally examined in marine mammals. Studies with humans and other animals (especially birds) provide some hints as to the acoustic cues that allow for determination of source distance (Naguib and Wiley, 2001; Kolarik et al., 2016), and may therefore mediate behavioral reactions. These cues are inevitable outcomes of sound transmission through the environment, the most important and well-studied being (1) overall decrease in received level from spreading loss and absorption, (2) a relative decrease in the amplitude of high frequencies due to absorption, and (3) the presence of reverberation resulting from boundary interactions. These cues will likely be available to marine mammals judging the range to military sonar sources, which often project signals containing multiple harmonics in reverberant environments. Studies with the bottlenose dolphin and false killer whale (Pseudorca crassidens) have demonstrated auditory discrimination based on SPL and/or harmonic content (Yuen et al., 2007; Finneran and Schlundt, 2011; Branstetter et al., 2013); however, the perception of reverberation has never been examined.

The current study was designed to provide experimental evidence that marine mammals can differentially classify FM acoustic stimuli with similar received SPLs, but differing transmission-related acoustic cues of reverberation and frequency-dependent absorption. Bottlenose dolphins (Tursiops truncatus) were used as subjects. This species is representative of the mid-frequency odontocete cetacean functional hearing group, which includes beaked whales and many dolphin species (National Marine Fisheries Service, 2016). The methods used in this study were inspired by those used by Phillmore et al. (1998) with black-capped chickadees (Poecile atricapillus) and zebra finches (Taeniopygia guttata).

Dolphins were trained to listen to FM tones with multiple harmonics and provide conditioned phonic responses upon hearing tones with degrees of high-frequency attenuation (HFA) and reverberation (REV) that simulated distant sources. Novel probe stimuli simulating ranges that were intermediate to those used during training were then introduced to determine if these acoustic cues could result in perceiving acoustic sources on a gradient from near to far. Finally, HFA and REV were decoupled in stimulus generation to examine the relative importance of each cue in the dolphins' decisions.

The subjects of the study were three bottlenose dolphins: APR (female, 33-year-old, 170 kg), BLU (female, 52-year-old, 190 kg), and OLY (male, 33-year-old, 185 kg). All of the dolphins had previously participated in behavioral psychophysical hearing tests, in which they were trained to produce a conditioned phonic response upon the detection of an underwater acoustic stimulus. Behavioral audiograms for the three dolphins, obtained with an up/down staircase procedure in the test environment, are shown in Fig. 1. Upper-cutoff frequencies, defined as the frequency where psychophysical thresholds reached SPLs of 100 dB re 1 μPa were 110 kHz for APR, 45 kHz for BLU, and 65 kHz for OLY.

Testing was conducted in a 9-m × 9-m floating, netted enclosure in San Diego Bay (Fig. 2). Underwater noise levels in the enclosure were largely dominated by snapping shrimp and small boat traffic in San Diego Bay (Fig. 1). Noise levels were near 70 dB re 1 μPa²/Hz at 10 kHz and decreased linearly with increasing logarithm of frequency to about 55 dB re 1 μPa²/Hz at 70 kHz. Trials were not run in the case of vessels passing by the experimental enclosure.

A trainer (unaware of experimental trial type to prevent possible cues) attended to the dolphins during each session from a walkway spanning the corner of the pen, and the experimenter controlled trials from a research shack containing signal production and acoustic recording equipment approximately 10 m from the dolphins' location. The experimenter and trainer communicated using a hardwired headset system and monitored the dolphins' phonic responses with a hydrophone suspended near the subject. During testing, the dolphins were positioned on a “biteplate” station located 1 m under water, and 1 m directly in front of an International Transducer Corporation (ITC) 1032 piezoceramic projector (International Transducer Corporation, Santa Barbara, CA) used to transmit acoustic stimuli. A second underwater sound projector (LL916, Lubell Labs, Columbus, OH) was placed 2.3 m to the right of the dolphins' station, and was used for playback of a trial outcome indicator at the end of a trial; either a “buzzer” sound for correct responses or “boing” sound for incorrect responses. On correct responses, the trainer delivered a fish reward to the dolphin. On incorrect responses, the dolphin was recalled to the trainer without reinforcement.

Stimuli were designed as FM tones with multiple harmonics to allow for a degree of analogy with 53 C sonar signals; however, the ambient noise levels below 10 kHz (Fig. 1) precluded the use of harmonic signals with the exact frequency content of 53C sonar. Therefore, undegraded stimuli (i.e., simulating a received waveform identical to that at the source and without HFA, REV, or other range-dependent effects) and degraded stimuli (i.e., containing various range-dependent degrees of HFA and REV) featured fundamental frequencies at 10 kHz and harmonics extending to 60 kHz. This resulted in stimuli that had frequencies higher than the dominant noise in San Diego Bay, and were in a region where the dolphins had more sensitive hearing.

Undegraded stimuli were digitally synthesized (500-kHz sample rate) using custom software and saved as individual WAV files. Characteristics of the FM sweeps were defined by the (fundamental) start frequency, stop frequency, and duration. Each of these parameters had a base value that was randomized by a controlled percentage to generate individual (unique) WAV files. The fundamental start and stop frequencies were randomized over the range 10 ± 1 kHz (i.e., upsweeps and downsweeps were possible), and duration was 500 ± 50 ms. A rise/fall time of 20 ms was applied to all stimuli to avoid onset and offset transients. In addition to a 10 kHz fundamental, five additional harmonics at integer multiples of 2× to 6× the fundamental frequency were added. The levels of these harmonics were set such that the second harmonic of the underwater stimulus (i.e., accounting for the frequency response of the transmitting hydrophone) was −20 dB relative to the fundamental, and increasing harmonics decreased approximately 3 dB per harmonic (Fig. 3).

Degraded stimuli were generated in the same manner as undegraded stimuli; however, they contained various degrees of HFA and REV. The effect of HFA on the fundamental and harmonic components was simulated using estimated absorption coefficients at each frequency (Ainslie and McColm, 1998), adjusted to account for water properties in San Diego Bay. Simulated REV for an acoustic source at a depth of 50 m (at distances of 1 to 30 km) was created by adding approximately 12 000 replicates of the undegraded or HFA stimuli at randomized delays between 300 and 900 ms. At a received distance of 1 m, the amplitude of these replicates was set to −70 dB relative to the direct path (at longer simulated ranges the REV to direct path ratio was much greater, Fig. 3). The REV tail decreased exponentially such that it extended up to 900 ms after the termination of the simulated direct path waveform.

Stimuli were converted to analog at a rate of 500 kHz using a PCI-6251 data acquisition card (National Instruments Corporation, Austin, TX), low-pass filtered (−3 dB at 100 kHz, Krohn-Hite Corporation, Brockton, MA), amplified (Pro 2400, Hafler Corporation, Tempe, AZ), and applied to the 1032 projector. Stimuli were calibrated without the dolphin on the biteplate before and after each session using a TC4013 hydrophone (Reson Corporation, Slangerup, Denmark) placed at the midpoint of the sound reception areas on the mandible. Signals were amplified and filtered (+32 dB, 100 Hz high pass, VP1000, Reson Corporation, Slangerup, Denmark), converted to digital using the PCI-6251, and stored to hard disk for analysis.

Equating the dolphins' perceived levels across stimuli posed a challenge, i.e., typical metrics of overall SPL or peak SPL may not exactly represent the perceived loudness of the FM tones with various HFA and REV conditions. To control for the use of signal level as a discrimination cue, stimuli were first calibrated using the maximum root-mean-square SPL in a 250-ms sliding window over the entire signal duration. This SPL was used to create a transmitting voltage response for each individual stimulus, and the stimuli were then equated at a level of 120 dB re 1 μPa. The stimuli were roved ±10 dB such that the received levels varied between 110 and 130 dB re 1 μPa (see Sec. II E below), thereby eliminating perceived level as a potential cue in making decisions.

The dolphins were initially trained to respond to the presentation of a single 30-km degraded stimulus. Once this behavior was reliable, repetitive undegraded background tones were played back at levels below the dolphins' hearing thresholds (and ambient noise) and increased incrementally in amplitude over many trials. This process continued until the dolphins reliably withheld phonic responses in the presence of audible undegraded background tones and responded to degraded target tones.

The next training step was to introduce stimuli with increasing variability in the start frequency, stop frequency, and duration. During the first parts of the training process, a 1% factor was used. Successive training sessions introduced stimuli with approximately 2 and 5% variability, resulting in an increasing FM upsweep and downsweep nature on a proportion of the stimuli. Eventually stimuli with 10% parameter variability were introduced, and this value was used for the generation of all stimuli for the rest of the study. Inclusion of the roving SPL was achieved in a manner similar to that used to introduce stimulus frequency and duration variability. The roving level was introduced at low levels (i.e., ±1 dB), and increased to ±10 dB once the dolphins had learned to withhold responses to changes in SPL.

A trial began when the experimenter initiated the playback of a set of eight background stimuli that consisted of either undegraded tones (Experiments 1 and 2) or tones with simulated 1-km HFA and REV (Experiments 3 and 4). These tones were repeated once every 2 s, and the background stimuli were presented randomly without replacement until all stimuli within the set had been presented once. When each background stimulus had been presented once, all were replaced, and the process was restarted. This was conducted to ensure that the dolphins were not over-trained on one particular background stimulus. As noted earlier, the amplitudes of each stimulus were randomly roved over a ±10 dB range on each presentation to remove perceived loudness as a cue in discriminating stimuli.

Once tones were being presented, the trainer instructed the dolphin to swim to the biteplate, and one of three types of trials was initiated. For target trials, the background FM tones were repeated for a pseudorandom interval from 2 to 12 s (1 to 6 background stimuli, Fig. 4). This period was then followed by the presentation of a single degraded target stimulus with HFA and REV simulating 30 km (i.e., greater than that of the background stimuli). If the dolphin produced the conditioned phonic response within 2 s following the onset of the target stimulus (hit), the buzzer was played indicating a correct response, and the trainer provided fish reinforcement. Control trials were identical to target trials except that the pseudorandom interval was followed by the presentation of a background stimulus. The buzzer was played and the dolphin was given fish reinforcement if the phonic response was not produced (correct rejection). One fish was given to the dolphin for correct responses on both target and control trials. If a dolphin did not respond on a target trial (miss) or responded on a control trial (false alarm), the boing sound was played, and it was recalled to the trainer and not provided fish reinforcement. Responses during repetitive background tones on either a target or control trial were also scored as false alarms, and the trial was terminated early with a boing sound.

Trials were generated using a pseudorandom sequence with an initial probability of 0.7 for target trials for APR and BLU and 0.6 for OLY (due to the latter subject's more liberal response bias). After three consecutive trials of a particular type, the conditional probability of that trial type was raised to the fourth power (i.e., reduced) for the subsequent trial. In the case that four trials of a particular type occurred in a row, the conditional probability was set to the initial probability raised to the fifth power. Each time the trial type changed from target to control or vice versa, the probability of a target was reset to its initial value. Using this procedure, sequences exceeding three consecutive trials of the same type were minimized (although they did still occur), and control trials comprised approximately 35%–40% of the total trials conducted in the study.

Probe trials constituted the third type of trial, where novel tones with degrees of HFA and/or REV intermediate to the background and target stimuli were presented after the 2- to 12-s pseudorandom interval. These trials were designed to determine how the dolphins perceived aspects of signal degradation relative to the background and target stimuli. There were no “correct” or “incorrect” responses to probe trials, and neither the buzzer nor the boing sound was played at the end of the trial. At the end of a probe trial, the dolphin was recalled to the trainer without fish reinforcement and then cued to begin the next trial. Consequently, the potential for outcome-based learning on probe trials was minimized. Sessions with probe trials contained six probe trials out of a total of 60 trials, with a single unique probe stimulus presented in each ten-trial block. This ratio was used in order to prevent frustration on the part of the dolphins for having multiple unreinforced trials for which there was no incorrect or correct answer.

Table I summarizes the background, target, and probe conditions that were used for each of the four experiments described below.

A complementary metric for quantifying performance on probe trials was the latency from the onset of a tone to the onset of a dolphin's response [response time (RT)]. Subject RT has been shown to correlate to the perceptual similarity of stimuli in discriminations, with longer latencies corresponding to higher degrees of perceived similarity and increasingly difficult discriminations (e.g., Stebbins and Miller, 1964; Pfingst et al., 1975; Dooling and Okanoya, 1995; Mulsow et al., 2015). In the case of the current study, RT also provided a means of determining which aspects of the stimuli the dolphins attended to in making discriminations. The time courses for HFA and REV on degraded stimuli differed, with HFA being present throughout the entire stimulus duration and REV beginning 300 ms after tone onset. Therefore, RTs less than 450–500 ms (based on the minimum RT of approximately 150–200 ms in bottlenose dolphins; Ridgway et al., 1991; Ridgway and Carder, 2000; Blackwood et al., 2003; Mulsow et al., 2015) reflect decisions solely based on HFA, while longer RTs allow for REV to be taken into consideration as well.

For each probe trial including a response, stimulus onset was aligned with waveforms and spectrograms of the underwater acoustic recordings using Raven Lite software (Cornell Lab of Ornithology, Ithaca, NY). The responses of each subject were identified visually and by listening to each recording. Each RT was calculated as the difference between probe stimulus onset and the onset of the dolphin's response. The median RT for each probe condition and subject was then calculated.

After the dolphins performed the discrimination task between undegraded background and 30-km target stimuli at a criterion level of 85% correct on three consecutive sessions (including hits and correct rejections), novel probe stimuli were introduced. The goal of Experiment 1 was to ensure that the dolphins responded as expected to novel probe stimuli that had not been used as background or target stimuli in the training phase.

The probe stimuli included undegraded and degraded stimuli that were generated in the same manner as the training stimuli, with up to ±10% variability in the parameters of start and stop frequency, and duration. Two novel undegraded and four novel 30-km degraded probe stimuli were presented in pseudorandom order across six 10-trial blocks. Three 60-trial sessions were completed per dolphin, with unique novel probe stimuli generated for each presentation. Following completion of three sessions, data within condition (i.e., undegraded versus degraded) were pooled for analysis of the dolphins' performance. Target and control trials were also analyzed with regards to percent correct and false alarm rate.

All three dolphins correctly identified changes from undegraded stimuli to the 30-km degraded (i.e., non-probe target trials) at rates greater than 97% (Table II). The false alarm rate for OLY was highest at 11%, while the other two dolphins rarely responded on control trials (2%). The dolphins performed as expected on the probe trials, responding to all presentations of the novel 30-km degraded stimuli with relatively short RTs (Table III), with one exception in the case of OLY. Also as expected, the dolphins withheld response to all of the novel undegraded probe stimuli. The RTs suggested that the dolphins were primarily basing their responses on HFA as opposed to REV. These results demonstrated that novel probe stimuli not used for training reliably elicited responses from the dolphins for 30-km degradation, and did not elicit responses for undegraded conditions.

The second experiment was designed to investigate how the dolphins would classify probe stimuli with degrees of stimulus degradation intermediate to the undegraded and 30-km stimuli. Probe stimuli were created with simulated ranges of 1, 10, and 20 km. Two novel probes of each intermediate range condition were presented in the pseudorandom fashion described for Experiment I over five consecutive 60-trial sessions (i.e., six probe stimuli per session, ten probe trials per intermediate range). For analyses, the responses to probe trials for each intermediate range condition were pooled.

The dolphins' RTs and performance metrics are given in Tables III and IV, respectively, and probe results are shown in Fig. 5. Correct responses on target trials and false alarm (FA) rates were consistent with Experiment 1, with OLY again having the highest false alarm rate. The pooled percentage of responses for the 1-km degraded probes was lowest at 63%, and all probes at 10 - and 20-km ranges elicited responses, with the exception of one 10-km probe presented to BLU (resulting in overall response rate of 96% for that condition). Response patterns were similar for BLU and APR, while OLY responded to all ten probe trials at each condition. This was likely a result of the more liberal bias in reporting the presence of range-related degradation in this dolphin (i.e., higher false alarm rate). Increases in subject RT tracked response rates, with the longest RTs corresponding to low response rates.

As the background stimulus waveform used up to this point in the experiment was a tone without either HFA or REV, it makes it possible that the dolphins perceived all of the probe stimuli somewhat similar to the 30-km target based solely on the presence of any range-related signal degradation. A more powerful test would examine the responses for probes at ranges between background and target stimuli both with degrees of HFA and REV [i.e., with transmission-related degradation more similar to the stimuli of Phillmore et al. (1998), actually recorded in the field]. Experiment 3 therefore was conducted using such conditions.

To determine how the dolphins made similarity judgments for probes intermediate between two degraded stimuli, the three subjects were trained to discriminate between the standard 30-km target condition and a repetitive 1-km background (versus the undegraded background used for Experiments 1 and 2). The dolphins were trained for this procedure by introducing the new degraded background stimulus in the place of the previous undegraded background, and reinforced in a manner consistent with control and target trials.

Once the dolphins' performance had reached the criterion of 85% correct on control and target trials across three consecutive sessions, probe trials were included in the same fashion described above for Experiment 2. Probe stimuli simulating 2 -, 10 -, and 20-km ranges were included in Experiment 3. Two novel probes of each intermediate range condition were presented in pseudorandom fashion over five consecutive 60-trial sessions (i.e., six probe stimuli per session, 24 probes total per dolphin, eight probes total per intermediate range). Probe trials were pooled according to range for analyses.

The dolphins APR and BLU both transitioned to the new background stimulus and were able to discriminate the 30-km degraded target from the degraded 1-km condition immediately (the 85% criterion needed to introduce probe stimuli was reached after the first three sessions). The task proved more difficult for OLY, and his performance did not reach the criterion necessary for moving forward to the probe trial phase after nine sessions. All further results reported for Experiments 3 and 4 are therefore only from APR and BLU.

The RTs and performance metrics for the dolphins are given in Tables III and V, respectively, and probe response patterns are shown in Fig. 6. Correct responses on target trials and FA were consistent with APR's and BLU's patterns in Experiments 1 and 2. Unlike Experiment 2 where APR responded to presentations of 1-km probes against an undegraded background, APR did not respond to any of the presentations of the closest probe condition of 2 km against a 1-km background. Response rates for APR to 10 - and 20-km probe stimuli, however, tracked what was observed in Experiment 2 and were at 100%. For BLU, the results were qualitatively similar to those from Experiment 2, displaying a progressively increasing response rate with increasing probe range.

The results of Experiment 3 provide a demonstration that the dolphins can classify stimuli based on varying degrees of range-related degradation qualities, as opposed to just their presence. A final experiment was designed to determine the relative importance of HFA and REV in the dolphins' classifications and responses to the probe stimuli.

The stimuli in Experiments 2 and 3 had REV and HFA that were coupled to the sources simulated range. To investigate the relative extents to which the dolphins used the features of HFA and REV to classify the probe stimuli, Experiment 4 independently manipulated HFA and REV while using background and target conditions that were consistent with Experiment 3. OLY did not serve as a subject due to the previous inability to perform the task with 1-km background conditions.

Background and target stimuli were degraded tones simulating ranges of 1 and 30 km, respectively. Six probe stimulus classes were used. Three of the classes corresponded to simulated REV conditions at 2, 10, and 20 km (2 -, 10 -, and 20-km REV probe stimuli in Table I). Each of these stimuli had HFA characteristics set at a simulated range of 1 km, the same as the background stimulus. The other three probe stimulus classes comprised the reciprocal conditions: HFA conditions of 2, 10, and 20 km, and REV characteristics set at 1 km. Each probe condition was presented six times per dolphin.

The RTs and performance metrics for the dolphins are given in Tables III and VI, respectively, and probe response patterns are shown in Fig. 7. Correct responses on target trials and FA rates were consistent with APR's and BLU's patterns from all of the previous experiments. For probe trials where REV was held constant at 1 km, APR and BLU performed identically. Neither dolphin responded to HFA conditions simulating 2 km, but both responded to all probe trials with 10 - and 20-km HFA conditions. Performance was markedly different for probe trials in which HFA was held at 1 km but REV varied. Neither dolphin responded to any of the probes with 2-km REV conditions. At 10-km REV, both APR and BLU responded to a subset of the probes: BLU to 50% (3/6) and APR to 33% (2/6). Interestingly, increases of REV to 20 km did not result in an increase in the percentage of responses for BLU, and APR did not respond to any stimuli for this probe condition. The most notable pattern in the RT data was that the longest RTs occurred for REV only probe stimuli. This is of course expected, as the first REV cues do not arise until 300 ms into each stimulus. These longer latencies also likely arose in part from the lower response rates for these stimuli, which were below 50% for both dolphins.

These experiments demonstrate that dolphins can use range-related features of signal degradation to classify acoustic stimuli independent of the received SPL. Work on acoustic discrimination in cetaceans has previously demonstrated that this is likely the case. Studies by Yuen et al. (2007) and Branstetter et al. (2013) demonstrated that odontocetes can use the harmonic structure of stimuli to make discrimination decisions. In the case of the study by Yuen et al. (2007), the tones did not feature a roving SPL and the false killer whale may have used perceived level to discriminate the tones. Branstetter et al. (2013) used a roving SPL of ±3 dB, therefore reducing the potential for level cues. The current work expanded upon these studies through the use of multiple background and target conditions, and unreinforced probe trials. These additions made the dolphins' responses more indicative of a general principle of similarity to a relatively close or distant source, rather than any deviation from a more specific template (although there was clearly a good degree of acoustic similarity in the current FM tones, despite differences in start and stop frequencies, and durations).

The probe results of Experiments 2–4 suggest a decision-making process based on similarity to the background or target stimuli. The responses in Experiment 2 could have been based on the presence of reverberation or the attenuation of harmonics relative to the fundamental. The simulation of HFA at 1 km resulted in attenuation of about 3 dB (relative to 0 km) at 20 kHz, and roughly 18 dB at the sixth harmonic of 60 kHz. Despite these large (and likely easily discriminable) differences in harmonic profile, the two dolphins with the most consistent performance (APR and BLU) only responded to these probe trials 40%–50% of the time with median RTs of over 1 s. This suggests that these two dolphins perceived the 1-km probes as somewhat similar to the undegraded background tone. The 100% response rate for OLY reflects a decision bias that was much more liberal (although the longer RTs in the 1-km condition for OLY might suggest a greater degree of similarity to the background than the longer range probes).

The responses of APR and BLU during Experiment 3, when the background tones had 1-km degradation, provide further insight. Here, the degree of HFA relative to the 1-km background was the same as that for the closest-range probe stimuli in Experiment 2; ranging from 3 to 18 dB for the second through the sixth harmonic. The higher response rates for the 1-km probes in Experiment 2 suggest that the dolphins classified the probe stimuli in that experiment as being more different from an undegraded background. In both Experiments 2 and 3, the near 100% response rates on the 10 - and 20-km probes demonstrate that the degradation was sufficient for those stimuli to be classified as similar to the 30-km target, or at least different from a 1-km degraded background.

The response patterns and RTs from Experiment 4 reflect the strategies that the dolphins used to make decisions based on HFA and REV. That no responses were observed for either the 2-km HFA or REV is consistent with results for the fully degraded probes during Experiment 3. The 10 - and 20-km HFA degradation conditions clearly resulted in those probes being classified as more similar to the 30-km target. The patterns for the 10 - and 20-km REV are more difficult to interpret. The REV for the 10-km distance resulted in the dolphins perceiving those probes as more similar to the target, but at a lower rate than the equivalent level of HFA. The REV conditions for 20 km apparently did not make these probes markedly more similar to 30-km target stimuli for BLU, and resulted in an apparently decreased similarity to the target stimuli for APR. The reason for these patterns—especially the decrease in response rate for APR—is currently unknown. However, these results confirm that in the context of this experiment, REV was less important than HFA in responding to stimuli.

Of the cues examined in this study, REV would be the most expected to faithfully represent absolute source range without prior experience with the signal (Mershon and King, 1975), as very few natural sounds contain features that approximate reverberation (Naguib and Wiley, 2001). Both HFA and SPL vary to large degrees in natural signals independent of propagation, and while these features may generally imply a relative source distance, prior knowledge of the signal structure plays an important role in using these cues to accurately judge relative range in humans (Coleman, 1962; Kolarik et al., 2016). It is therefore interesting that both APR and BLU appeared to use HFA to a greater degree than REV when making decisions in this experiment, particularly given the different audiometric profiles between the two dolphins. This is perhaps a result of the dolphins having a general template for the harmonic structure of the FM sweeps in the study. The fact that HFA was present over the entire duration of a stimulus and was available prior to REV cues may have also resulted in the dolphins giving HFA precedence over REV in making decisions.

There are a number of caveats that must be considered when extending the results of this study to perceptual processes of dolphins in general. First, this study used few probe trials per condition, especially in the case of Experiment 4. A greater number of probe trials would likely have increased the degree of gradation between conditions and the stability of median RT values. The selected distances of the probe, background, and target stimuli certainly affected these observed response percentages as well. Changes in these parameters will result in different response patterns, although a similar overall pattern of increased response probability with increasing distance is likely (e.g., the differences between Experiments 2 and 3). The ambient noise in the testing enclosure is also an important consideration in interpreting the results. While the harmonic components of undegraded stimuli were all above ambient noise, higher degrees of HFA decreased the signal-to-noise ratios of the harmonic components above the fundamental, and the decreased stimulus amplitudes in the tails of stimuli with REV may have been more likely to become masked by noise. These issues may partially explain the lack of gradation in the responses to probe trials at the longest ranges, and the reduced influence of REV relative to HFA in response probability.

While received SPL may convey some information on source distance (Coleman, 1962; Kolarik et al., 2016), the observed use HFA and REV qualities to classify stimuli independent of received SPL support controlled exposure experiment observations of behavioral reactions being dependent on distance from an experimental acoustic source (Deruiter et al., 2013; Southall et al., 2014). The current study's response probability gradients do not, however, conclusively show that HFA and REV directly translate to the perception of distance extending along an axis away from the animal (as also noted by Phillmore et al., 1998). Field studies with birds have demonstrated this translation of acoustic information to distance through close-range playbacks of stimuli that simulate relatively longer ranges, eliciting responses that are appropriate for the simulated longer range as opposed to that of the source speaker (Naguib and Wiley, 2001). Comparable studies with wild marine mammals, or carefully designed laboratory studies, may be used to demonstrate this capability in marine mammals.

The authors thank J. Powell, R. Dear, M. Tormey, G. Goya, T. Wu, R. Echon, K. Christman and the animal care and training staff at the Navy Marine Mammal Program. The study followed a protocol approved by the Institutional Animal Care and Use Committee at the Biosciences Division, Space and Naval Warfare Systems Center (SSC) Pacific and the Navy Bureau of Medicine and Surgery, and followed all applicable U.S. Department of Defense guidelines. Financial support was provided by the U.S. Navy Fleet Forces Command. This is Contribution 193 of the National Marine Mammal Foundation.