2 Why not pursuing the humpback whale case study 2. 3 Relevant information for humpback whale acoustic density estimation

Transcription

1 Estimating density of humpback whales using passive acoustic bottom-mounted detectors at PMRF - case study assessment and recommendations Tiago A. Marques 1, Len Thomas 1, Nancy DiMarzio 2, Ron Morrissey 2, Jessica Ward 2, Dave Moretti 2, David Mellinger 3, Steve Martin 4 and Peter Tyack 5 1 Centre for Research into Ecological and Environmental Modelling, The Observatory, University of St Andrews, St Andrews, KY16 9LZ, Scotland. 2 Naval Undersea Warfare Center Division, 1176 Howell St., Newport, Rhode Island, USA. 3 Cooperative Institute for Marine Resources Studies, Oregon State University, 2030 SE Marine Science Drive, Newport, Oregon 97365, USA. 4 Space and Naval Warfare Systems Center San Diego, San Diego, California USA. 5 Woods Hole Oceanographic Institute, Woods Hole, Massachusetts 02543, USA. Version s date: January 30, 2009 Contents 1 Introduction 2 2 Why not pursuing the humpback whale case study 2 3 Relevant information for humpback whale acoustic density estimation Humpback whale information Humpback whale acoustics Other relevant information PMRF PMRF and humpback whale data Possible approaches to estimating humpback whale density from passive acoustics Possible approaches Some unstructured ideas for future reference Rationale for changing to a minke whale case study 8 References 9 1

2 1 Introduction One of the tasks of DECAF project was Task 1. Case study: estimation of humpback whale density at the Pacific Missile Research Facility (PMRF), Hawai i. However, after DECAF began and a few startup meetings where held about this case study, it became apparent that humpback whales (HW) at PMRF were not the easy case study we had envisaged. This document has 4 objectives, each address in its own section following the introduction: 1. compile relevant current knowledge to HW density acoustic estimation 2. explain why we believed that this case study was not worth pursuing within DECAF 3. lay down some possible future research options regarding HW 4. present an alternative case study, namely minke whales (MW) at PMRF 2 Why not pursuing the humpback whale case study In the DECAF proposal, HW were considered to be the most straightforward case study, in effect just a simple application of conventional distance sampling methods (Buckland et al., 2001). It was stated in DECAF s proposal that [...] we focus on singing male humpback whales, for four reasons: (1) humpback songs (we will use song units, the lowest level of organization of the hierarchically structured song of humpbacks) are present in abundance in the acoustic data set available; (2) these vocalizations can often be detected on multiple hydrophones, making localization (at least in 2 dimensions) feasible; (3) the animals are shallow divers (Baird et al., 2000), relative to the depth of hydrophones, so the complication of 3-dimensional distance sampling can be ignored in the first instance, and to a reasonable approximation animals can be taken to be on the surface in calculating horizontal distances; (4) group size estimation is not a problem since singing males are widely spaced.. Of these reasons, we now know that (2) and (4) are likely incorrect, which lead to considerable implementation problems. For (2) we know that most vocalizations 2

3 Figure 1: A screen shot with humpback localizations, taken from the document that Steve Martin circulated around DECAF PMRF Notes - 29 May 2008). Blue stars are humpback locations, black dots are hydrophones, and green are the islands contours. occur outside the array (at least for the time periods we have looked at - cf. figure 1) so localization is difficult. This also means the standard assumption is false that density of true animal distances increases linearly with increasing distance from the point. For (4) we now see that animals may not be widely spaced, but some can be close to one another. Note however that in figure 1 these locations are from an extended period of time; it may still be true that at any instant the singing humpbacks are widely spaced. 3

4 3 Relevant information for humpback whale acoustic density estimation 3.1 Humpback whale information Humpback whale acoustics Both male and female humpback whales can produce sounds, however only the males produce the long, loud, complex songs for which the species is famous. Males are usually solitary. The purpose of the song is not yet clear, although it appears to have a role in mating. Au et al. (2006) report the maximum source level to vary between individual units in a song, with values between 151 and 173 db re 1 µpa. The classic paper about humpback songs is Payne and McVay (1971). The authors use the following terms: subunit < unit (shortest sound continuous to our ears in real time) < phrase < theme < song < song session (a series of songs for which there is no pause longer than a minute - these can take several hours). A single song can take over 30 mins (records of songs with 7 to over than 30 mins were made). They consider there is a general species specific song pattern. This differentiates two song types: A and B. Stafford et al. (2007) present a detection range modeling for 5 species of whales (including humpbacks) in the Gulf of Alaska, noting that modeled detection ranges varied greatly with input parameters and choice of ambient noise level. Regarding humpbacks, the authors stated that...humpback whale calls (350 Hz) would not be detected if the animals were within 2 km of the instrument due to shadowing, but would be detected with 100% probability between 5 and 12 km. There is evidence of convergence zone at the lowest ambient noise levels indicating that some humpback whale calls might be recorded up 45 km away, but not between 15 and 40 km.... It has been suggested that humpbacks can choose locations that are optimal regarding sound propagation over large distances. Somewhat anecdotal evidence has been collected supporting this, which is a problem under some settings for acoustic density estimation: it means that the detection function of the sounds is not simply a function of distance from the sound but also of some unknown optimal properties of the animals location. In Mercado and Frazer (1999) it is shown that humpbacks can greatly affect transmission range by adjusting their positions and sounds in response to environmental factors. Source depth, in particular, is shown to be a major determinant of which frequencies propagate the farthest. A preliminary analysis of range-dependent distortion suggests that spectral cues can potentially provide listening whales with information about how far a sound has 4

5 traveled Other relevant information The humpback whale (Megaptera novaeangliae) is a baleen whale. One of the larger rorqual species, adults range in length from 12 to 16 meters and weigh approximately kilograms. Found in oceans and seas around the world, humpback whales typically migrate up to 25,000 kilometers each year. Humpbacks feed only in summer, in polar waters, and migrate to tropical or sub-tropical waters to breed and give birth in the winter. During the winter, humpbacks fast and live off their fat reserves. Johnston et al. (2007) present a spatial analysis of the preferred wintering grounds of humpbacks in Northwestern Hawaiian Islands. A small number of DTag s (but no sound records?) seem to have been applied to humpbacks (Ware et al., 2006). Reisewitz (2002) contains a (potentially useful) annotated bibliography about Hawaiian Islands Humpback Whale National Marine Sanctuary. 3.2 PMRF The Pacific Missile Range Facility (PMRF), located on the northwestern coast of Kauai, Hawaii, is the world s largest instrumented test range capable of supporting surface, subsurface, air and space operations simultaneously. There are over 1,100 square miles of instrumented underwater range. The instrumented underwater range utilizes active pingers on objects of interest to detect, localize and track the objects. Many of PMRF s organic assets of bottom mounted underwater hydrophones can also receive vocalizations of several species of marine mammals. 3.3 PMRF and humpback whale data Twenty-four broadband hydrophones (100 Hz-19 khz and 100 Hz-45 khz) were recorded, primarily during the winter months (Feb-March) each year from 2002 through 2006; the resulting data comprise 40 days in total. Initial screening of the acoustic data has indicated that many acoustic detections are available for male humpback whales. Ten minute periods have been selected from separate days and were manually scanned by a human acoustic intelligence specialist, who has extracted recognizable cetacean vocalizations and estimated their location. The uncertainty in this location is unknown, and its reliability for some cases is dubious (e.g. some locations several km inland - see figure 1). Also it is hard to distinguish 5

6 between several animals singing close together, the same animal moving around or the same animal not moving with measurement error in the localization. The humpbacks have long and continuous songs, detected at large distances, create a chorusing which makes detection and localization of individuals more difficult. This actually accounts for a fair proportion of the background noise in these areas, as noted by Au et al. (2000), which creates additional problems to working with these sound records. 4 Possible approaches to estimating humpback whale density from passive acoustics 4.1 Possible approaches Here is a quick list and review of possible methods. Note that DTag, or other data about detectability, are assumed not known here. Adaptation of standard point transect methods. If we could localize animals 1 within a short time window, and if the distribution of true animal distances with respect to the hydrophones was known (assumed triangular in standard methods), then we could apply a snapshot version of standard point transect methods. Measurement error in localization could be an issue, but could be dealt with if something were known about the error distribution. This would give us the instantaneous density of singing males would need a multiplier (proportion singing) to get to density of animals (this is common to all the methods). There are a couple of problems with this approach. We may not be able to characterize measurement error, but this should be a minor problem. The unknown distribution of animals with respect to distance is unknown, but certainly not triangular. This is a major problem as it precludes using conventional distance sampling methods. If we knew the detection probability and range for each animal detected, we could estimate the unknown distribution of animals with respect to distance. This means, in turn, we could estimate effective detection radius for the hydrophone and hence effective area surveyed, and hence density. We could estimate detection probability of a sound received at a single hydrophone through propagation modelling but to do this we need to know the location of the animal, not just the range. We also need to know the 1 This is an animal-based method. 6

7 distribution of source levels, as this has a great effect on probability of detection but that is not well known. One helpful option here would be to choose as our detector something that focusses on some part of a song unit that is known to have a relatively homogeneous source level, to make this part easier. This has the additional advantage that detecting all instances of a single song unit is probably easier than detection of all types of song units. One issue here is how to turn detection probability for a single hydrophone into probability of getting the range, given that the likely way we will get the range is by localizing the sound source which requires detection on 3 hydrophones. This could conceivably dealt with by extending conventional methods. Another issue is that this kind of estimator (Horvitz-Thompson) is biased when detection probability is estimated, and here it is indeed estimated. The bias is worse when detection probability is low, and here detection probability will be low for some sounds received. Lastly, note that we can not only localize animals, but also get a probability density function (PDF) for each observation, which allows us to incorporate location measurement error in our assessment of detection probability. Many details here need to be resolved. An alternative method would be to use spectrogram cross-correlation for pairs of hydrophones to get bearings to sound sources, without need for classification. Bearings from multiple pairs could be used to determine how many sound sources (i.e. individuals) there were, for some snapshot in time. To turn this into a density estimate, will need to know the proportion of times an individual is calling which we may be able to get some information on, and the effective detection radius (EDR). The latter comes from integrating the probability of detection given distance over the PDF of true animal distances. We can possibly get the probability of detection from propagation modelling, but the problem here is that the source is now ill defined, and will have a very wide (and unknown) distribution of source levels hence there will be considerable uncertainty in prob of detection. To get the PDF of true animal distances, one option would be to use external information about relative density of calling whales from prior studies, to give us this. (Possible examples are Johnston et al. (2007), Frankel et al. (1995), Adam Frankel s thesis, etc.). This method seems quite tenuous. 7

8 4.2 Some unstructured ideas for future reference A few ideas not properly thought through, but which we list here for future reference include: An option suggested by Steve Buckland was to set up trials by sending a boat to near singing whales to record when the sound is produced, and then seeing where it is detected. This would also be useful for getting source levels. This option was considered to be infeasible within the bounds of DECAF. Mercado and Frazer (1999) implies that it might be possible to estimate range from spectral characteristics of sound. In terms of localizing animals, it might be possible to get bearing from time-difference-of-arrival of sounds at different hydrophones (via spectrogram cross-correlation, or other ways), and perhaps to get range via multipath methods for single hydrophones. Perhaps there are ways to integrate these? D. Mellinger thought it would be possible to identify a fixed point in a song, by focussing on a specific unit within each song, and getting, say, the first instance of it (by identifying the other parts of the phrase). This will have to be optimized for each season, since songs change each season, and would have to be a unit used by all whales within the period for which density is being estimated (if that s possible!). Another suggestion raised was the use of data from the migration period where the issue of non uniformity of distances might not present itself. But it seems that this is unlikely to work, as migrating whales don t go through the middle of the range. 5 Rationale for changing to a minke whale case study It was suggested that we could focus on minke whales (MW), because their spatial distribution might be such that the hydrophones at PMRF can be considered a random sample of their locations. In particular, MW are also often detected within the range, rather than (as HW) predominantly off the range. This circumvents the major shortcoming of the HW case study. It brings however an additional problem, because little is known regarding MW sound production rate. So there would be 8

9 two options here, one which implements an animal based method, or one which implements a cue based method, but then the conversion of cue density to animal density requires some cue rate estimate for which there is, at least currently, very little information available. References Au, W. W. L., Mobley, J., Burgess, W. C., Lammers, M. O., and Nachtigall, P. E. (2000). Seasonal and diurnal trends of chorusing humpback whales wintering in waters off western Maui. Marine Mammal Science, 16(3), Au, W. W. L., Pack, A. A., Lammers, M. O., Herman, L. M., Deakos, M. H., and Andrews, K. (2006). Acoustic properties of humpback whale songs. The Journal of the Acoustical Society of America, 120, Baird, R., Ligon, A., and Hooker, S. (2000). Sub-surface and night-time behavior of humpback whales off Maui, Hawai i: A preliminary report. Technical Report Report prepared under Contract 40ABNC050729, Hawaiian Islands Humpback Whale National Marine Sanctuary, Kihei, HI. Buckland, S. T., Anderson, D. R., Burnham, K. P., Laake, J. L., Borchers, D. L., and Thomas, L. (2001). Introduction to distance sampling - Estimating abundance of biological populations. Oxford University Press, Oxford. Frankel, A. S., Clark, C. W., Herman, L. M., and Gabriele, C. M. (1995). Spatial distribution, habitat utilization, and social interactions of humpback whales, Megaptera noveangliae, off Hawaii, determined using acoustic and visual techniques. Canadian Journal of Zoology, 73, Johnston, D. W., Chapla, M. E., Williams, L. E., and Mattila, D. K. (2007). Identification of humpback whale Megaptera novaeangliae wintering habitat in the Northwestern Hawaiian Islands using spatial habitat modeling. Endangered Species Research, 3, Mercado, E. I. and Frazer, L. N. (1999). Environmental constraints on sound transmission by humpback whales. Acoustical Society of America Journal, 104, Payne, R. S. and McVay, S. (1971). Songs of humpback whales. Science, 173,

10 Reisewitz, A. (2002). Hawaiian islands humpback whale national marine sanctuary - annotated research bibliography. Technical report, University of Hawaii. Stafford, K., Mellinger, D., Moore, S., and Fox, C. (2007). Seasonal variability and detection range modeling of baleen whale calls in the Gulf of Alaska, The Journal of the Acoustical Society of America, 12, Ware, C., Arsenault, R., Plumlee, M., and Wiley, D. (2006). Visualizing the underwater behavior of humpback whales. IEEE Computer Graphics and Applications, pages