Distribution and Behavior of the Bowhead Whale, Balaena mysticetus

Transcription

1 Clark, C. W., Charif R., Mitchell, S., and Colby, J Distribution and Behavior of the Bowhead Whale, Balaena mysticetus, Based on Analysis of Acoustic Data Collected During the 1993 Spring Migration off Point Barrow, Alaska. Rep. int. Whal. Commn. 46: Copyright International Whaling Commission, All rights reserved. Distribution and Behavior of the Bowhead Whale, Balaena mysticetus, Based on Analysis of Acoustic Data Collected During the 1993 Spring Migration off Point Barrow, Alaska Christopher W. Clark, Russ Charif, Steven Mitchell, and Jennifer Colby Cornell University, Cornell Laboratory of Ornithology, Bioacoustics Research Program, 159 Sapsucker Woods Rd., Ithaca, NY, 14850, USA ABSTRACT A successful acoustic study was conducted off Point Barrow, Alaska in spring 1993 as part of the continuing effort to census the Bering-Chukchi-Beaufort Sea bowhead whale population. New data collection and analyses methods were developed resulting in close to 1000h of hydrophone array recordings and the completion of 732h of acoustic location analysis within a year after the field season. A comparison of the old and new analyses methods indicates that the new method results in locations with less variability and slightly less bias than the old method. The offshore distribution of acoustically located whales indicates that throughout the 1993 migration period 86% of the vocalizing animals were within 4km of the nearshore ice edge. INTRODUCTION In 1993, from mid-april through May, an acoustic study was conducted off Point Barrow, Alaska as part of a major effort to census the Bering-Chukchi-Beaufort Sea population of the bowhead whale, Balaena Mysticetus. Previous acoustic studies were undertaken in 1984, 1985, 1986, and 1988, and population estimates have now been derived from the combined visual and acoustic data for 1985, 1986, 1988, and 1993 (Clark et al., 1986a, 1986b, 1986c, 1988; Clark and Ellison, 1988, 1989; Givens et al., 1995; Raftery et al., 1988, 1990; Raftery and Zeh, 1991, 1994; Zeh et al., 1988, 1990, 1991, 1995). The 1993 census effort was the most successful to date in terms of the numbers of whales seen, the amount of time with concurrent visual and acoustic data collection, and the stability of the weather and ice conditions. Spring 1993 off Point Barrow, Alaska was also unusual in that during the majority of the field season, there was little to no ice obstructing the visual census effort, and for most of the time there was open water along the nearshore edge of the shorefast ice. Partly as a result of the predominantly good conditions, the 1993 field effort resulted in the highest visual counts and the greatest number of acoustic array recordings of any census to date (George et al., 1995). In the past, the acoustic location analysis of the multi-channel array tapes has required a great deal of time, and the lack of acoustic locations for large portions of the season contributed to the uncertainty in the population estimate (Raftery and Zeh, 1991). In order to better accomplish the task of analyzing the acoustic array recordings anticipated from the 1993 census effort, a new acoustic location analysis system was developed. This new system was designed

2 to greatly improved the speed of the analysis process, as well as provide more precise and more accurate acoustic locations. Here results are presented on the locations and distributions of whales based on the completed analysis of the 1993 acoustic array data. Also, the new methods used in the analysis of the 1993 acoustic array data are described, and results from acoustic analyses using the pre (old) and 1993 (new) acoustic location methods are compared. METHODS Acoustic field study The primary objective of the acoustic field effort is to collect recordings of bowhead sounds, using arrays of hydrophones, during as much of the spring migration period as possible. The basic field methods for this task have been described in previous reports and will not be detailed here (see Clark et al., 1986a, 1986b; Clark and Ellison, 1988, 1989; Würsig and Clark, 1993). As in previous acoustic efforts, the census was conducted in April and May, during the bowhead spring migration off Point Barrow, Alaska. For the 1993 field effort, several improvements were made in the way acoustic data were collected. The first improvement involved expanding the size of the arrays. In the past, arrays of 3-4 hydrophones were installed and maintained with inter-hydrophone spacing of > 700m. In 1993, every attempt was made to install and maintain arrays of at least 4-5 hydrophones with inter-hydrophone spacing of > 800m. There were several reasons for expanding the size of the arrays. First, a larger array aperture can reduce the range and bearing errors associated with an acoustic position, especially for vocalizing whales that are >8-10km from the array. The net benefit to the census is a reduction in the uncertainty in the offshore distribution. Second, a longer array with more hydrophones increases the chances of detecting vocalizing whales that are migrating within several kilometers of the nearshore ice edge. With a short array (e.g., approximately 2km), nearshore whales are within the 120 array aperture for less than 0.5h, and there is less chance of a whale being acoustically located than with a longer array. Data recording Improvements were made in the data recording method by using a multi-channel digital audio tape (DAT) recorder (TEAC model RD 130T) rather than an analog recorder, as in the past, to record sounds. This calibrated DAT system has a flat frequency response (+ 0.5dB) in the 4-channel mode from DC to 10kHz, and a flat frequency response (+ 0.5dB) in the 8- channel mode from DC to 5kHz. Tape duration is 2 hours rather than 45 minutes, and each DAT tape contains a separate sub-code band so that the date and time are automatically recorded onto the tape. At the beginning of each tape recording, the acoustic personnel on duty noted the ambient noise levels for each of the recording channels. During the recording period for the tape, the person also qualitatively judged the acoustic recording condition. Quality assessment took into account such influences as the ambient level, amount of local interference at each hydrophone (e.g., ice knocking), and radio frequency interference. Ambient levels were noted as a value between 0 and -30dB relative to the saturation level of the tape. By this notation, the greater the negative value of the ambient level, the lower the level of noise in the 0-6kHz band, and the better the signal to noise ratio (SNR) for the whale sounds on the tape. During field operations, several tapes each week were routinely analyzed to verify the geometry of the hydrophone array and the proper operation of the data collection system. Description of old and new acoustic location analysis methods Acoustic locations are computed based on a) the measured positions of several (3-5) widely separated hydrophones (ca m apart), b) the known speed of sound in water page 2

3 off Barrow, Alaska (1437m/s), and c) the time delays between the occurrences of the same whale call at the different hydrophones. The exact characteristics of the whale signals are not known, but can be described as amplitude-modulated (AM) and frequency-modulated (FM) sweeps (sometimes complex, sometimes simple), mostly in the range of Hz, sometimes accompanied by harmonics. The SNR varies considerably as there can be strong interference signals from seals, ice, snowmobiles, motorboats, wave action, or hydrophone strum. The transmission medium is very shallow (< 100m), isothermal (-1 to +1 C) arctic water. Sometimes the surface of the water is covered by a uniform sheet of young ice, sometimes by a jumbled mass of ice, sometimes there are floes of multi-year ice, and sometimes there is an avenue of open water, referred to as an open lead." Even if the lead is open, there may be large masses of floating ice (sometimes deeply keeled, multi-year ice) between the whales and the array. The hydrophones are deployed to a depth of 10-20m either over the ice edge along a pressure ridge or through a hole drilled through the ice. The hydrophones are free to move with the current, while their cables are anchored on the surface of the ice. Inaccuracies in the time delay estimation are primarily due to multipath arrivals of the signals at the hydrophones, and the presence of interference or noise. Inaccuracies in an acoustic location, given good time delays, can arise due to refraction or unexpected sound-ray geometry paths (probably due to ice, or the permafrost layer beneath the ocean floor). The displacement of the hydrophones due to local currents and the error in the survey method (< + 5m) introduce relatively minor errors into the estimate of a vocalizing whale s location. In the old method, the time delay between the occurrence of the same whale sound on two hydrophones was estimated by cross-correlation of the sound s spectrographic images (Clark et al., 1986a), using the duration and bandwidth of the signal as constraints on the cross-correlation process. The time offset associated with the maximum value of the crosscorrelation function was taken as the time delay between the occurrence of the sound at the two hydrophones. The position of the whale was calculated directly from the graphical intersection of only three cross-bearings, and the range error was calculated from the size of the triangle transcribed by the intersecting cross-bearings (see Clark et al., 1986a, 1986b for details of the old location process). In cases when four hydrophones were operating, and therefore six time delays were computed, only the three hydrophones that had the best SNR (as judged by the operator at the beginning of a tape) were used to determine a location. For the new method an acoustic location workstation was developed and implemented on an Apple Macintosh computer augmented with customized digital signal processing hardware and an advanced version of the Canary Software program (Charif et al., 1995). Using this system the field tapes are post-processed for acoustic locations of bowhead whale calls, a process that requires some human decisions and data interpretations. The operator identifies bowhead calls by listening to the original tape while simultaneously observing a realtime, multi-channel spectrogram display of all the hydrophone channels. The analyst recognizes when a bowhead call occurs, and selects a time-frequency box that bounds the call in each of the different channels. Time delay values are estimated by bandpass filtering the received signals, as delimited by the high and low values of the selected time-frequency box, and computing the cross-correlations of the waveforms (not their spectrographic images) for all pairs of hydrophones, where the waveform data are delimited by the start and stop times of the selected time-frequency box. The time offset associated with the maximum value of the complex demodulated cross-correlation function is taken as the time delay between the occurrence of the sound at a pair of hydrophones. The acoustic location is determined from the consistency of redundant information provided by the time delay estimates for all pairs of hydrophones. In the absence of errors, and assuming that the water depth is shallow relative to the range to the vocalizing whale, each time delay defines a hyperbolic cross-bearing intersecting the vocalizing whale's true position For example, a five hydrophone array yields ten time delays, and requires the simultaneous intersection of ten hyperbolic cross-bearings. In the presence of errors, there is typically no one source location that would yield the measured time delays, and location estimation is computed page 3

4 using a minimum sum-of-squares error technique based on all (not just three) time delay values. Appendix I provides further details of the new acoustic location method. Since errors are present in the acoustic location process, each acoustic location has a confidence region associated with it, where the confidence region represents an area of geographic locations (around the estimated location) whose time delay errors are less than a minimum factor of 80%. The boundary extremes of this 80% confidence region are used to compute the minimum range, maximum range, and bearing error statistics for each location. Appendix II provides details of the algorithm for computing the 80% confidence region. The acoustic location analysis requires some human interaction. The analyst listens to the tape while watching a real-time, multi-channel spectrogram display of all the hydrophone channels. The analyst recognizes when a bowhead call occurs, and selects a time-frequency box that bounds the call in each of the different channels. Once the call has been identified and bounded in time and frequency, the program automatically computes time delays for all pairs of hydrophones, computes the x, y position and 80% confidence region of the vocalizing whale, and plots the location and confidence region on a map. Figure 1 is an example of an acoustic location where its 80% confidence region is shown as a gray, almost elliptical area surrounding the acoustic location point. Fig. 1. Estimation of location and 80% confidence region for a calling whale using all 6 time-delay cross-bearings from a four-channel hydrophone array. The large confidence region around the whale s location is due principally to the cross-bearing from the time-delay between hydrophones 1 and 4, which intersects the other cross-bearings further from the estimated position than the other five cross-bearings. The 80% confidence region is considered the error region and defines the range and bearing errors associated with the location of the vocalizing whale. In some cases, it was obvious from the location plot that one or more of the computed time delay values was erroneous. To help resolve this problem, a screening algorithm was developed, tested and programmed into the final location analysis process. This algorithm is referred to as the "sigma minimizer" program since it minimizes the sum-of-squares error (sigma) in the location process. This program analyzes the set of all time delays for each location and discards zero, one, or two time delays to obtain the minimum sum of squares error (see Appendix I) between the time delay values of the estimated location and the time delay values as measured by the cross-correlation process. This procedure eliminates outlying time delays page 4

5 resulting from cross-correlations that are corrupted by multipath sound propagation, excessive ambient noise, or surface reverberation. An example of the improvement in acoustic location error as a results of this sigma minimizer method is illustrated in Figure 2. Fig. 2. The effect of eliminating an erroneous cross-bearing on a whale s acoustic location and associated error region (80% confidence region) for the same whale call as shown in Fig. 1. This estimate excludes the cross-bearing from the time delay between hydrophones 1 and 4. It illustrates the dramatic reduction in the range an bearing errors associated with the whale s location that results when the erroneous cross-bearing is eliminated. Once the sigma minimizer was tested and fully integrated into the location analysis software, all 1993 acoustic locations were re-computed using the modified program. As a final check on the output of the sigma minimizer each location was carefully examined, with the timedelay hyperbolas plotted on a map as in Fig. 2. Each location was then quality rated as either A, B, C, or D (where A is the best rating; D is the worst). Rating was done subjectively by visual examination of the intersecting time-delay cross-bearings. The quality ranking was based on how close the cross-bearings come to intersecting at a single unambiguous point. Locations rated A or B were considered acceptable, while locations rated C or D were considered unreliable and were not accepted. Although there is a certain amount of subjectivity in this process, locations qualified by different technicians yielded consistently similar ratings, and most locations were unambiguously acceptable or not acceptable sample periods selected for acoustic location analysis Twenty-one representative sample periods were selected for acoustic analysis from throughout the 1993 survey period. A listing of these sample periods is given in Table 1. These samples represent approximately 732 hours of tape recording effort when an array was operational, and contain at least 38,391 calls as tallied by acoustic personnel monitoring the array during the field season. During post-processing after the field season, all bowhead calls detected on the array tapes from the 21 periods were analyzed by the new acoustic location method. The results of that acoustic location effort are the basis for describing offshore distributions as presented here. page 5

6 Table 1. Listing of 21 sample periods for 1993 selected for acoustic location analysis. Start and stop times for the periods were modified slightly once the actual analysis was performed. Date Time Start and Date Time Stop represent the date and time for the beginning and end of the sample period, respectively. For example, in , 5 represents the fifth month, 15 represents the fifteenth day, 23 represents the 23 rd hour (11PM), and 00 represents zero minutes after the hour. Field Count refers to the numbers of bowhead sounds heard by acoustic personnel in the field but adjusted as possible so as not to include song notes. Tape Hours indicates the approximate total number of hours of array recording as determined by the number of tapes multiplied by two hours/tape. Offshore distributions Offshore distance distributions of vocalizing whales relative to the visual observation perch (origin) and the axis of the array (x-axis), were computed using the acoustic locations after processing by the sigma minimizer and quality rating procedures. Distance distributions were computed from a) the acoustic locations alone and b) acoustic tracks computed by an algorithm (Sonntag et al., 1986). For the distribution based on acoustic tracks, algorithm parameters used were the same as those selected by Zeh et al. (1995) for estimating 1993 population size based on the 1993 census study results. The offshore distance for a single acoustic location is the distance between the location and its projection onto the axis of the hydrophone array, while the offshore distance for an acoustic track is the distance between the location and its point of projection onto the axis of the array. By this procedure the offshore distance is the closest point of approach (CPA), and the distribution is referred to as the CPA distribution. For purposes of comparing distributions between different blocks of time in the field season, the 21 sample periods were lumped into six blocks. Each block covers a period of approximately one week, and the six blocks are referred to by capital letters A through F. The resultant CPA distributions for the acoustic tracks serve as the basis for comparing the distributions of vocalizing whales between different periods of the season and between different years when acoustic surveys were conducted. page 6

7 Call tracks Acoustic call tracks were identified after the acoustic location process was completed using the same methods as in previous years (Clark, 1989; Clark, Bower, and Ellison, 1990; Clark and Bower, 1991). The term call track refers to something different than the term acoustic track. An acoustic track is based on the output of a tracking algorithm and is strongly influenced by the parameters of swimming speed and migration heading. An acoustic track is not believed to represent the actual movement pattern of an individual whale, but instead represents statistically derived linkages between a series of acoustic positions. In contrast, a call track is a series of acoustic locations linked into a track based on the similarity of the calls vocal characteristics and an acceptable maximum swimming speed independent of swimming direction. In the past, call tracks were determined by careful inspection of the visual images of bowhead sounds and cross-correlation of the sounds in conjunction with a map of the acoustic locations for the sounds (Clark, 1989; Clark, Bower and Ellison, 1990; Clark and Bower, 1991). In the 1993 call track analysis, the same process was used but now aided by having real-time audio spectrograms and mapping displays that allowed one to quickly relisten to sounds while looking at their visual images and locations. The resultant call track is a set of acoustic locations that are believed to represent the same whale or the sequential positions of a group of whales traveling together so closely as to be indistinguishable acoustically. Figure 3 illustrates two call tracks for two different whales migrating northward past the visual perch during the same time period. Fig. 3. Two acoustic tracks constructed by linking together successive locations of similar distinctive calls. Beginning and end times are shown for each track. Re-analysis of 1988 array data using new methods To compare the old and new acoustic location methods on bowhead sounds recorded on an array, a limited set of 38 hours of acoustic array tapes from the 1988 field census study were re-analyzed using the new acoustic location method. Re-analysis of 1985 calibration data using new methods To compare the old and new acoustic location analysis methods on sounds played back from a known source location, the acoustic array data from the calibration test on 28 April, page 7

8 1985 were re-analyzed using the new method. In the 1985 field calibration test, approximately 60 frequency sweeps, simulating a bowhead call in the Hz band, were transmitted through a J11 loudspeaker at each of five different sites. The frequency response of this transducer was flat, + 1 db, in the Hz band. Later in 1985, 10 random sounds from each site were processed by the existing acoustic location method (Ellison et al., 1986). For the re-analysis, the exact same sounds in the playback sequence as analyzed by the old method in 1985 were re-analyzed using the new location method. RESULTS Field data collection Table 2 provides a summary of the acoustic effort to record bowheads during the 1993 spring migration season. The first hydrophone was installed 1600 h on 14 April. This single hydrophone was acoustically monitored for whale sounds but only one tape was recorded over the next 23 hours as the first array was being installed. The first array was working by 2030 on 15 April, and arrays were maintained for most of the season until acoustic monitoring ended at 1405 h on 30 May. Array monitoring and recording was essentially continuous during this time (see Table 2) except for a two-day period between 17 and 19 May when only a single hydrophone was operational. No recordings were made for this single hydrophone, but acoustic monitoring and field counts of whale calls continued. Two gaps in monitoring occurred: a 4.5 hour gap on 17 May and a 29.8 hour gap between May, both due to relocation of the ice camp. page 8

9 Table 2. Summary of 1993 bowhead whale acoustic recording effort. For 1993, a total of 49,301 bowhead calls were noted by acoustic personnel in the field during 952 h of acoustic monitoring, and 530 DAT tapes were recorded for the entire season. Figure 4 shows the daily call counts and daily monitoring effort for the entire field season. Acoustic field tallies as shown in Fig. 4, especially in the late April period, sometimes included bowhead song notes and are therefore slightly different from counts registered by acoustic personnel during the post-season analysis effort. Conditions were generally good throughout the season as indicated by the average acoustic conditions and average ambient recording levels noted by acoustic personnel. page 9

10 Fig. 4. Histogram of daily field counts of bowhead sounds tallied by acoustic personnel, and the line graph of daily acoustic monitoring effort for the 1993 field season. Altogether there were 25 arrays installed and maintained for a total of hours (Table 2), with an average array life span of hours. A total of 496, two-hour tapes were recorded when arrays were operational. One-hundred thirty-three hours of recordings were made from 3-hydrophone arrays, 316 hours from 4-hydrophone arrays, and 540 hours from 5- hydrophone arrays. An additional sixty-eight hours of effort were spent listening to 1-2 hydrophones, with no recordings made. The number of arrays was somewhat higher than in previous acoustic studies (Clark et al. 1986a, 1988), largely due to minor adjustments in hydrophone locations to minimize equipment loss and maximize array performance. Acoustic location analysis Acoustic location analysis was completed for 732 hours of array recordings that included all of the 21 sample periods. Table 3 lists the results of the acoustic analysis. The acoustic analysis, before correction using the sigma minimizer, resulted in 7494 acoustic locations. As a result of the re-analysis process using the sigma minimizer and quality rating procedures, the original set of 7494 acoustic locations was reduced to The tracking algorithm analysis yielded 1458 acoustic tracks, of which 701 contained only a single location and 757 contained multiple locations. For those tracks with multiple locations, the average number of locations per track was 3.4, the average length of an acoustic track was km, and the average duration of a track was minutes. The average bearing of a track was degrees relative to the axis of the array. page 10

11 Table. 3. Summary of the acoustic location analysis of the 1993 array tapes. Values listed under Acoustic tracks are the totals for each of the six blocks A-F. page 11

12 Figure 5 shows a scatterplot of all 6042 acoustic locations from the 732 hours of acoustic analysis, while Figures 6A and 6B show the CPA distributions for all locations and all 1457 acoustic tracks derived form these acoustic locations by the tracking algorithm, respectively. Figure 7 shows histograms of the offshore distributions of the acoustic locations from each of the 6 blocks, A-F. Fig. 5. Scatterplot of all 6,042 acoustic locations for The origin represents the visual observation perch. page 12

13 Fig. 6. Top: CPA distribution for all 1993 acoustic locations (n=6,042). Bottom: CPA distribution for all 1993 acoustic tracks (n=1,437). page 13

14 Fig. 7. CPA distributions for each of the six blocks (ca one week periods) during the 1993 field season based on acoustic locations (n=6,042). Call track analysis Call track analysis was performed for 10 of the 21 sample periods (see Table 3), yielding a total of 115 call tracks. Only tracks with durations greater than five minutes are included, since tracks of shorter duration are considered unreliable for computing swimming speed and heading. page 14

15 Comparison of old and new acoustic location analysis methods; 1988 data In the comparison of the old and new methods based on the analysis of 38 hours of array data from 1988, although the same array tapes were re-analyzed, not all of the calls located by the new method were the same as those located by the old method. Some calls analyzed in 1988 were not located by the new method and some calls analyzed by the new method were not located by the old. In the comparison presented here only the same calls successfully located by both methods are considered. For the tapes from the periods selected for comparison, the new method located 236 of the same calls that had been located by the old method. Nineteen of the calls failed to yield acceptable locations by the new method: 15 were considered as unreliable because the estimated range was greater than five times the array aperture and 4 were given a quality rating of C by the analyst. Nine of these 19 calls were also within a few degrees of the edges of the 120 sector in which locations are considered acceptable. The location data for the remaining 217 calls were then used to compare the old and new methods. Offshore distributions Figure 8 shows the CPA distance distributions generated by the old and new methods for the 217 calls. The percentages of distances < 4 km for the two distributions are identical (76.5%), the median values of the CPA distributions are nearly identical (2580 m old, 2575 m new), and there is no significant difference between the two distributions (KS two-tailed test, for alpha = 0.05 level, D = , Da = ). Fig. 8. CPA distributions for 217 calls recorded in 1988, as determined by the old and new acoustic location methods. page 15

16 Range and bearing estimates Figure 9 compares the range and bearing estimates between the old and new method for each of the 217 locations. As one would expect, range estimates agree closely between the two methods for calls made by nearby whales, with increasing variability as range increases. Bearing estimates are very consistent, except for two calls, one of which is at the extreme edge of the 120 sector. Fig. 9. Comparison of range and bearing estimates generated by 1988 and 1993 methods for 217 bowhead calls recorded during the 1988 acoustic census. page 16

17 Figure 10 shows distributions of the differences between range and bearing errors for each pair of 217 location estimates as determined by the old and new methods. Average range error for the old method was m compared to the average range error of m for the new method. When considering range error as a percentage of the range, the averages are % for the old method and % for the new method. Bearing error was for the old method and for the new method. These results indicate that the range and bearing errors associated with an acoustic location as estimated by the new method are approximately three to four times lower than the errors as estimated by the old method. Fig. 10. Distributions of differences between new and old range errors and bearing errors. Negative differences indicate that the new method yielded a lower error value. page 17

18 Re-analysis of 1985 calibration data using 1993 acoustic location methods For the re-analysis of the 1985 calibration data, the same ten sound transmissions as originally analyzed for each of the five sites by the old method were re-processed using the new method. In this calibration exercise the location of the source was determined by a visual crossfix on the source position using two theodolites (azimuth resolution, 10"), and the visual crossfix position is considered the true location of the sound source. Results from this re-analysis are listed in Table 4 along with the old analysis results (Ellison et al., 1986). The average percentage difference between the visual and acoustic ranges using the old method is 3%, while the average difference is 2% for the new method. The coefficient of variation (CV) for the new method is 0.5 compared to 2.5 for the old method. Table 4. Locations of five 1985 acoustic calibration sites as determined by the old and new acoustic methods, compared with the visual position as determined by two theodolite crossfixing. Range ± SD (m) is the average and standard deviation for the distance (n=10) from the visual observation perch to the location of the acoustic source, while Bearing ± SD ( o mag.) is the average and standard deviation for the bearing (n=10) to the source from the perch relative to magnetic North. DISCUSSION The bowhead census effort in 1993 represents the most successful survey ever conducted for bowhead whales in terms of the number of visual sightings, numbers of array tapes obtained and the proportion of total hours monitored. This research effort gathered more visual sightings (>3300 new whales, George et al., 1995) and acoustic array recordings than any previous census. Acoustic array recordings included nearly 1000 hours of array data and over 69,000 calls. During the peaks of the migration, there were periods when calling rates were in excess of 1000 calls per hour. These are the highest rates ever noted during any spring census effort. The development of a new acoustic location system significantly decreased the time required for processing acoustic locations and allowed us to analyze significantly more hours of page 18

19 array recordings (732 h) than had been analyzed in any of the three previous acoustic studies. With older analysis methods, several years were needed to complete a significant portion of the acoustic array tape analysis. Another advantage of the new location system is that all the data (including the original sound files) are now stored in a digital archive that allows for easy recomputation of locations and associated errors, should further improvements be made in the methods for estimating location and confidence regions. There are several conclusions from the comparison of results from the old and new methods based on the re-analysis of 38 hours of 1988 data. First, the two methods yield essentially identical location distributions for the same sample of calls (n=217), with absolute differences in range and bearing estimates increasing as the distance to the vocalizing whale increases. Differences were also effected by the bearing to the call (i.e., differences were larger for calls located near the edges of the 120 sector). These effects were first noticeable at ranges of > 3 km, and were more noticeable at ranges > 6 km (see Fig. 9). Second, the range error and bearing error associated with a location were four and three times greater with the old method than with the new method, respectively. This is not an unexpected result since the method of determining time differences between the occurrence of the same sound at different hydrophones was improved in the new method through better application of signal processing techniques. This reduces the uncertainly in the time delays, resulting in lower range and bearing errors (see Figs. 1 and 2). The primary conclusion from the comparison between results based on the re-analysis of the 1985 calibration data using the old and new methods is that the new location method results in locations with less variability but only slightly less bias than locations using the old method. Evidence for the conclusion that both methods have approximately the same bias is supported by the results showing that the average percentage difference between the visual and acoustic ranges to a known acoustic source using the old method was 3%, while the average difference was 2% for the new method. Evidence for the conclusion that the new method is less variable is supported by the results showing that the coefficient of variation (CV) for the new method is 0.5 compared to 2.5 for the old method. Thus, the new method is, on average, about five times more precise than the old method. Results based on the acoustic location analysis of the 1993 array data indicate that the majority of vocalizing whales were within several kilometers of the perch at their CPA (see Fig. 6). Specifically, in 1993, 86% of the whales were acoustically located within 4 km from the ice edge from which the array was deployed and visual observers were working, and 94% of the acoustic locations were within 6 km. This is in contrast to previous years when the minority of whales tracked were < 4 km from the perch on their closest point of approach. For example, in the 1985 season, 71% of the whales were < 4 km from the perch and 76% were within 6 km. In 1986, 50% of the whales were < 4 km from the perch and 60% were within 6 km. These percentages for 1985 and 1986 are corrected for the effects of array aperture (see Clark and Ellison, 1988, 1989). That is, for the 1985 and 1986 estimates of CPA distribution, a correction factor was used in order to compensate for the effects of the small array aperture (Clark and Ellison, 1988, 1989). In 1993 an array geometry correction factor was unnecessary because the array size was 4500 m (more than twice the length of previous arrays) and the majority of whales were passing within 2 km of the perch. Unlike 1985, 1986, and 1988, when there were sometimes dramatic shifts in the offshore distribution of vocalizing animals during the migration period, the offshore distribution in 1993 remained relatively consistent throughout the season, as seen by comparing the CPA distributions for the six seasonal blocks in Figure 7. The reason for this consistency may be related to our array locations being more southerly than in previous years in combination with fairly stable weather and ice conditions. Unlike previous seasons, we were able to remain in one location and operate a large array almost continuously for a month, and this period covered the major portion of the migration. The combined results for all four census years underscore the fact that within-year and between-year variability in the whales offshore distribution during page 19

20 the spring migration is often pronounced and must be properly taken into account in any population census efforts. The number of calls detected by field personnel per unit time was higher in 1993 (49 calls/h) than in 1985 (21 calls/h), but lower than in 1986 (60 calls/h). These same differences existed after lab analysis ( calls/h; calls/h; calls/h). The percentage of calls located was the lowest of any previous year ( %, %; %), although the number of locations per acoustic track was quite consistent between years ( locs/track; ; ). This might be an indication that when a whale is in the 120 sector within which reliable locations are obtained, there is a high probability that it will vocalize several times. However, this could also be an indication of an upper limit on the tracking algorithm's ability to link a whale for greater than 3-4 locations. The reason for the lower location efficiency in 1993 compared to previous years is not easy to determine, but probably is related to the fact that such a high percentage of whales were traveling within several kilometers of the nearshore ice edge. This would help explain both the higher call counts and the decreased ability to locate vocalizing whales. Whales that are traveling close to the ice edge spend a greater proportion of their time within acoustic detection range of the array but are more difficult to reliably locate. Therefore, if a large proportion of whales are traveling along the ice edge, as they did in 1993 compared to previous years, than a large proportion of calls will be heard on the array. One might expect under this circumstance that the whales would be located more easily. However, this is not the case since whales that are traveling close to the ice edge are also approaching the array along its axis. In this configuration the location of a vocalizing whale is difficult to determine since all the time delay bearings pointing toward the whale will be parallel to each other resulting in a location that is unreliable. This is the reason that we adopt the procedure of not accepting any location with an acoustic bearing that is within 30 of the axis of the array, and the reason why fewer whales will be located when traveling along the ice edge. Thus, a vocalizing whale that is traveling within a few kilometers of the ice edge spends relatively less time within an array s 120 acoustic location sector (see Clark and Ellison 1988, 1989) than a whale that is traveling 2-10 km offshore of the array, and has a lower chance of being located. This would explain the lower percentage of acoustic locations obtained per total calls in 1993 because, although many more calls were detected, many more of those whales were calling from either of the two 30 sectors that are outside of the region within which acoustic locations are considered unreliable. In 1985, although nearly half of the whales were within 4km of the ice edge, the location efficiency was the highest of all four years; nearly four times the efficiency as in 1993 (see Clark and Ellison, 1985). However, in 1985 only half as many whales were traveling within 2km of the ice edge when compared to 1993; 33% in 1985 vs. 60% in 1993, so that the ice edge effect was not as significant in 1985 as it was in In 1986, only 21% of the vocalizing whales were detected within 2km of the ice edge, but the efficiency was still lower than in This might be explained by the fact that in 1986 when the lead was open, there were many periods when the majority of whales in were >8 km offshore (Clark and Ellison, 1989) and during these periods many calls were detected but the locations were unreliable because the animals were to far offshore from the array. These results further emphasize the fact that the distribution of animals across the lead varies tremendously between years depending primarily on ice conditions. Exactly how representative the acoustic offshore distribution is of the true distribution remains somewhat of an open question. There is some evidence by Clark and Ellison (1988, 1989) supporting the conclusion that the acoustic distribution is not biased since both calling rate and call type are independent of a whale's offshore distance from the array out to a range of approximately 15 km. Also, aerial transect survey data, although limited, are in general agreement with acoustic distribution results on a year to year basis (Withrow and Goebel-Diaz, 1989); a result that supports the conclusion that the acoustic distribution is representative of the true distribution. In any case, we believe that a combination of visual and acoustic methods will provide a better page 20

21 estimate of the distribution than only one method alone since the two methods provide independent measures of the same phenomena. Acoustic offshore distribution results could be improved by installing hydrophones across the lead rather than restricting the array to the lead edge as is presently done. This could be accomplished using a few autonomous hydrophone recording systems deployed on the bottom several kilometers offshore of the array. Visual distribution results could be significantly improved by increasing the number of transect surveys. Since manned surveys in the Arctic are expensive, we suggest that surveys using small remotely operated aircraft with multiple high resolution cameras might be a cost effective option to adequately sampling the offshore distribution of bowheads migrating off Point Barrow, Alaska. ACKNOWLEDGMENTS Funding for this study was provided by the State of Alaska and by the North Slope Borough, P.O. Box 69, Barrow, Alaska, USA. Thomas Albert was a constant source of advice and support throughout this research. He and John Craighead George were instrumental in the planning and successful completion of the field season. We are very grateful to Mayor George Ahmaogak and all the people of Barrow for their support and hospitality during this project. Thomas Calupca, Adam Frankel, Christine Gabriele, Christy Giessinger Hartendy, Mia Grifalconi, Matt Irinaga, Togulik Opie, Samuel Payne, and Susie Tigigilook provided dedicated assistance during the field study effort. Sean Cunningham did a fantastic job in the development of the software programming. Thomas Calupca did equally magnificent work in the design and fabrication of the customized signal processing hardware. Katherine Brese, Carol Carson, and Janet Doherty were extraordinarily diligent in the laboratory analysis and organization of the data sets. Judy Zeh provided invaluable help in the design of the sampling scheme, and completed the acoustic tracking algorithm analysis. REFERENCES Charif, R.A., Mitchell, S., Clark, C.W Canary 1.12 User s Manual. Cornell Laboratory of Ornithology, Ithaca, NY. 229 pp. Clark, C. W Call tracks of bowhead whales based on call characteristics as an independent means of determining tracking parameters. Rep. int. Whal. Commn. 39: Clark, C.W., and Ellison, W.T Numbers and distributions of bowhead whales, Balaena mysticetus, based on the 1985 acoustic study off Pt. Barrow, Alaska. Rep. int. Whal. Commn. 38: Clark, C.W., and Ellison, W.T Numbers and distributions of bowhead whales, Balaena mysticetus, based on the 1986 acoustic study off Pt. Barrow, Alaska. Rep. int. Whal. Commn. 39: Clark, C.W., Ellison, W.T. and Beeman, K. 1986a. An Acoustic Study of Bowhead Whales, Balaena mysticetus, off Point Barrow, Alaska During the 1984 Spring Migration. A Report to the North Slope Borough, Department of Wildlife Management. Marine Acoustics. 145 pp. Clark, C.W., Ellison, W.T., Beeman, K. 1986b. A preliminary account of the acoustic study conducted during the 1985 spring bowhead whale, Balaena mysticetus, migration of Point Barrow, Alaska, Rep. int. Whal. Commn. 36: Clark, C.W., Ellison, W.T. and Beeman, K. 1986c. Acoustic tracking of migrating bowhead whales. Oceans 86, IEEE Oceanic Eng. Soc., New York: Clark, C.W., Ellison, W.T. and Beeman, K An Acoustic Study of Bowhead Whales, Balaena mysticetus, during the 1985 Spring Migration. A Report to the North Slope Borough, Department of Wildlife Management, P.O. Box 69, Barrow, Alaska. 143 pp. page 21

22 Clark, C. W., Bower, J.L., and Ellison, W.T Acoustic tracks of migrating bowhead whales, Balaena mysticetus, off Point Barrow, Alaska based on vocal characteristics. Rep. int. Whal. Commn 40: Clark, C. W. and Bower, J.L Intercall intervals and acoustic tracks of bowhead whales off Point Barrow, Alaska, based on passive acoustics. Rep. int. Whal. Commn 41: 761. Ellison, W.T., Clark, C.W. and Beeman, K Acoustic location techniques and calibration methods used during the spring 1984 and 1985 bowhead whale, Balaena mysticetus, migrations. Rep. int. Whal. Commn. 36:502. George, J.C., Suydam, R.S., Philo, L.M., Albert, T.F., and Zeh, J.E Report of the spring 1993 census of bowhead whales, Balaena mysticetus, off Point Barrow, Alaska with observations on the 1993 subsistence hunt of bowhead whales by Alaska Eskimos. Rep. int. Whal. Commn. 45: Givens, G.H., Zeh, J.E., and Raftery, A.E Assessment of the Bering-Chukchi-Beaufort Seas stock of bowhead whales using the BALEEN II model in a Baysian synthesis framework. Rep. int. Whal. Commn. 45: Raftery, A. E., Turet, P., and Zeh, J. E A Bayes empirical Bayes approach to interval estimation of bowhead whale, Balaena mysticetus, population size estimation. Rep. int. Whal. Commn 38: Raftery, A. E., Zeh, J. E., Yang, Q. and Styer, P. E Bayes empirical Bayes interval estimation of bowhead whales, Balaena mysticetus, population size based upon the 1986 combined visual and acoustic census off Point Barrow, Alaska. Rep. int. Whal. Commn 40: Raftery, A. E. and Zeh, J. E Bayes empirical Bayes estimation of bowhead whale population size based on the visual and acoustic census near Barrow, Alaska in 1986 and Rep. int. Whal. Commn 41: 759. Raftery, A. E. and Zeh, J. E Bowhead whale, Balaena mysticetus, population size estimated from acoustic and visual census data collected near Barrow, Alaska in Rep. int. Whal. Commn 45: 448. Sonntag, R.M., Ellison, W.T., Clark, C.W., Corbit, D.R., and Krogman, B.D A description of a tracking algorithm and its application to bowhead whale location data during the spring migration near Point Barrow, Alaska Rep. int. Whal. Commn. 36: Withrow, D. and Goebel-Diaz, C Distribution of bowhead whales near Point Barrow, Alaska, Rep. int. Whal. Commn 39: Würsig, B. and Clark, C.W Behavior. In: J. Burns and J. Montague (eds.) The Bowhead Whale Book. Allen Press. Lawrence, Kansas. pp Zeh, J. E., Turet, P., Gentleman, R. and Raftery, A.E Population size estimation for the bowhead whale, Balaena mysticetus, based on 1985 visual and acoustic data. Rep. int. Whal. Commn. 38: Zeh, J.E., Raftery, A.E., and Yang, Q Assessment of tracking algorithm performance and its effect on population estimates using bowhead whale, Balaena mysticetus, identified visually and acoustically in 1986 off Point Barrow, Alaska. Rep. int. Whal. Commn, 40: Zeh, J.E., Raftery, A.E. and Schaffner, A. Revised Estimates of Bowhead Population Size and Rate of Increase. Paper SC/47/AS10 presented to the IWC Scientific Committee, May 1995 (unpublished). 26pp. page 22

23 Appendix 1: Location Estimation Algorithm Geometry and propagation model The algorithm assumes that the sound source is a point source in a two dimensional homogeneous isotropic non-dispersive linear medium, emanation circular waves at a constant speed of ca = m/s. The algorithm uses the surveyed locations of the hydrophones, given in x,y coordinates in meters from the visual perch, with the y -axis pointing true north. Contrary to this algorithm assumption, we have evidence (from some 1992 calibration results) that the sound ray propagation path is not always a straight line through the medium. Multipath signal arrivals have been clearly observed under conditions when large multi-year ice flows were between the source and hydrophone receivers. The hydrophone positions are subject to surveying error, and the hydrophones themselves undergo some motion as they dangle suspended by their cables from the surface. Measurements n The algorithm takes as inputs ( 2) measured time delays of sound arrival between the n hydrophones. In general, not all of the hydrophone pair time delays are used in computing the location estimate. The analyst marks one hydrophone as the best channel. If the measured time delay between channel k and the best channel is greater than max delay, the time required for sound to travel directly through the medium between the positions of the two hydrophones, then all time delay pairs involving channel k are unused. (This is the mechanism by which the analyst may disqualify a noisy hydrophone, since moving the time-frequency selection out of range will cause time delays using that channel to exceed max delay.) In addition, if the measured time delay between a channel pair i and j (neither of which is the best channel) is greater than the corresponding max delay, then the pair (i, j ) is not used. Statistical model Given the algorithmic assumptions, there is a deterministic function τ = τ ( ) M x (1) which gives a vector of time delays τ as a function of the source position x. (M refers to the propagation model, hydrophone positions, etc.) The location estimator is an inverse of this function. We can postulate a simplified statistical model, that the measured time delays obtained from a source located at position x are given by τ = m τ ( x)+ w M, (2) where w is a Gaussian random vector with covariance matrix T E[ ww ]= σ 2 I (3) page 23