This Casey_2022_Data_readme.txt file was generated on 2022-07-17 by Ari S. Friedlaender GENERAL INFORMATION 1. Title of Dataset: Data from: Acoustic signaling and behavior of Antarctic minke whales (Balaenoptera bonaerensis) 2. Author Information A. Principal Investigator Name: Dr. Caroline Casey Institution: UC Santa Cruz Address: 115 McAllister Way, Santa Cruz, CA 95060 Email: cbcasey@ucsc.edu B. Corresponding investigator Contact Information Name: Dr. Ari Friedlaender Institution: UC Santa Cruz Address: 115 Mcallister Way, Santa Cruz, CA 95060 Email: 3. Date of data collection: Feb-Mar 2018 and 2019 4. Geographic location of data collection: Western Antarctic Peninsula 5. Funding sources that supported the collection of the data: National Science Foundation Office of Polar Programs, World Wildlife Fund 6. Recommended citation for this dataset: Casey and Friedlaender (2022), Data from: Data from: Acoustic signaling and behavior of Antarctic minke whales (Balaenoptera bonaerensis), Dryad, Dataset DATA & FILE OVERVIEW These data were collected to describe and determine the acoustic behavior and associated context of calls from free-ranging Antarctic minke whales from tag-mounted hydrophones. The data presented here contain all of the acoustic calls that were detected along with specific call parameters for the identified call types determined by our analysis. Finally, exemplars of all calls types are provided as .wav files. 1. File List: File 1 name: Casey_2022_Minke_Data.xls File 1 Description: Sheet1: all acoustic calls detected from acoustic biologging tags and their associated file name, call type, behavioral and environmental context. Sheet2: for each acoustic call type determined, this sheet contains individual call file names as well as the Acoustic properties of each call and the calculated averages within each call type. File 2 name: 190309-52_10_093240.wav File 2 Description: Example of a 'Boom' call File 3 name: 190226-56_19_190118.wav File 3 Description: Example of a 'Downsweep' call File 4 name: 190226-56_31_230614.wav File 4 Description: Example of a 'Growl' call File 5 name: 190306-52_29_213928.wav File 5 Description: Example of a 'Rumble' call METHODOLOGICAL INFORMATION Spectrograms (Hann window; FFT size 2046; overlap 95%) of all available audio data were generated using Raven Pro (v.1.5, Center for Conservation Bioacoustics, 2014). Spectrograms were visually and aurally inspected by two individuals with extensive experience evaluating baleen whale acoustic behaviour from tag data. Recordings were then qualitatively compared to published data for Antarctic minke whales (Dominello & Širović, 2016; Filun et al., 2020; Risch et al., 2014; Schevill & Watkins, 1972; Shabangu et al., 2020; Van Opzeeland & Hillebrand, 2020). Other soniferous species that may have been present in this study area include humpback whales and Antarctic fur seals. The latter species are primarily vocal in air (see Page et al., 2002; Aubin et al., 2015), and recorded signals were qualitatively compared to published descriptions of Antarctic humpback whale vocalizations from this region (Stimpert et al., 2012; Van Opzeeland et al. 2013). Our recordings were then categorized into several descriptive types based on perceptual visual and acoustic similarities (neither the classification structure nor the number of groups was pre-determined). Discrete calls were defined as units of sound that could be readily identified and counted and separated from other signals by more than 3 s of silence. Only the vocalizations belonging to descriptive call types that were present on a minimum of five recordings on at least two separate tag deployments were included in subsequent analyses. A total of 651 suspected minke whale vocalizations were identified and qualitatively assigned to descriptive call categories. We used signal-to-noise ratios (SNR) to attribute calls to the immediate tagged individual or associated group. A relative SNR of 10 dB was the minimum required to include identified calls in further analysis. Factors including the swimming speed of the animal and the local ambient noise conditions may have led to a conservative selection bias in the signals included. For example, if calls were produced during periods with high background noise and were not salient (e.g., rapid acceleration or surfacing), these would not have met our criteria and been included in further analysis. Given that our objectives were focused on evaluating the broader behavioral and diel contexts associated with sound production, we are confident that the calls used for analysis were either produced by the tagged individual or a close associate in the group. This assumption was also corroborated by limited field observations collected concurrently with tag data. From the manually audited data, a subsample of high-quality calls for each descriptive call type wereselected for additional analysis. Only signals where all parameters of the spectral contour could be identified were included. A total of five acoustic features were measured for each call in Raven Pro (Hann window; low-pass filter 2000 Hz, FFT size 4096; frequency resolution 12 Hz, overlap 95%): duration 90% (s), center frequency (Hz), 1st and 3rd quartile frequencies (Hz), and 90% bandwidth (Hz). The 90% call duration and bandwidth were selected because they minimize measurement error. Temporal parameters were always measured from the waveform, and spectral parameters were measured from the spectrum or spectrogram. Of all the calls identified, 230 vocalizations were deemed suitable for this detailed acoustic analysis (Table 1). We conducted a linear discriminant analysis (LDA) with cross-validation using the five variables measured for each call type to confirm the initial categorization of sounds by trained observers. LDA uses a linear combination of values from two or more independent discriminating variables that best group cases into their a priori assigned classes. In this analysis, we assigned call type as the group identifier and acoustic measures as the discriminant variables. Percent correct classification obtained from the classification matrix (generated by the LDA) provided a metric of how well the measured variables separated the calls into each subjective call category. This analysis provides a relative confidence score for the accuracy of the human-derived call classifications. Association of Call Types with Behavioral and Environmental Variables Diel Calling Rates—To evaluate whether specific calls were associated with either day or night, we first determined sunrise and sunset times for each tag deployment using the NOAA ESRL sunrise/sunset calculator (https://www.esrl.noaa.gov/gmd/grad/solcalc/sunrise.html). Diel period (day/night/twilight) was determined from the angular sun position at a given location and time using the MATLAB package “Sunrise Sunset” (https://www.mathworks.com/matlabcentral/fileexchange/55509-sunrise-sunset). Twilight was defined as the period when solar elevation was < 6° degrees below the horizon and was determined in order to exclude those transitional time periods from daylight and nighttime hours. All identified calling events were then categorized as being produced either during day or night. To determine diel patterns in calling rates for each call type, we divided the total number of daytime calls by the total number of daylight tag hours. This provided a measure of call rate. We then replicated this process for nighttime calls. This allowed us to determine the hourly calling rates for each call type during day and night, by tag deployment. We then compared the proportion of each identified call type’s cumulative occurrence during daytime and nighttime periods across tag deployments. Foraging Versus Non-Foraging – To evaluate whether specific calls were associated with either foraging or non-foraging behavior, we used previously developed methods for evaluating kinematic and motion data to determine the timing of lunge feeding events (Cade et al. 2021). All kinematic and motion tag data were decimated to 10 Hz, tag orientation on the animal was corrected for, and animal orientation was calculated using custom-written scripts in MATLAB 2014a (Johnson & Tyack 2003; Cade et al., 2021). Animal speed for all deployments was determined using the amplitude of tag vibrations (Cade et al., 2018), and the norm of the jerk signal (vector sum of the difference) was calculated using the ‘njerk’ tool at animaltags.org. Animal speed and jerk were the primary metrics used to characterize the kinematic signature of a rorqual feeding lunge, which includes an increase in speed and overall body acceleration followed by a rapid deceleration (Goldbogen et al., 2017). For each recorded dive, we used the presence of one or more lunges to label a dive as feeding. Additionally, we corroborated feeding lunges that occurred during periods when tag video was recorded. We generated a record of foraging and non-foraging dives for each tag deployment, which included both the dive depth and time of day. We compared the proportion of each identified call type occurring during foraging and non-foraging states. The resulting matrix was evaluated using Chi-square analysis to determine whether calls were produced more or less frequently among these categories compared to parity. DATA-SPECIFIC INFORMATION FOR: Casey_2022_Minke_Data Sheet1 1. Number of variables: 7 2. Number of cases/rows: 741 3. Variable List: Animal Deployment ID: tag ID number for each deployment with acoustic data WAV file name: specific file name for individual acoustic files in which a call was detected Comments: general information on call heard Nominal Call Type: best guess by acoustic observer of call type based on acoustic properties Acoustic Analysis (Y/N): whether the call was used in our analysis to define acoustic parameters of each call type Number of Lunges on dive: all values greater than 0 were used to categorize a dive and subsequent acoustic call as being associated with foraging or not Sun Status: based on the hour of the day to categorize whether a call was observed during day, night, or twilight Sheet2 1. Number of variables: 7 2. Number of cases/rows: 255 Variable List: XXXX Files: distinct type of Acoustic call Q1 Freq: first quartile acoustic frequency in Hz Q3 Freq: third quartile acoustic frequency in Hz BW90%: 90% percentile bandwidth in Hz Center Freq: center of call frequency in Hz Peak Freq: peak acoustic frequency in Hz Duration 90%: 90th percentile call duration in seconds 4. Missing data codes: none 5. Specialized formats or other abbreviations used: