FY 2007 PI Report - National Oceanographic Partnership Program

DECAF – Density Estimation for Cetaceans from passive Acoustic Fixed sensors
Len Thomas
CREEM, University of St Andrews, St Andrews, Fife, Scotland, UK. KY16 9RR
Phone: (+44) 1334-461801 FAX: (+44) 1334-461800 E-mail: [email protected]
Tiago Marques
CREEM, University of St Andrews, St Andrews, Fife, Scotland, UK. KY16 9RR
Phone: (+44) 1334-461801 FAX: (+44) 1334-461800 E-mail: [email protected]
David Borchers
CREEM, University of St Andrews, St Andrews, Fife, Scotland, UK. KY16 9RR
Phone: (+44) 1334-461843 FAX: (+44) 1334-461800 E-mail: [email protected]
Catriona Stephenson
CREEM, University of St Andrews, St Andrews, Fife, Scotland, UK. KY16 9RR
Phone: (+44) 1334-461831 FAX: (+44) 1334-461800 E-mail: [email protected]
David Moretti
NUWC, Code 74, Bldg. 1351, 1176 Howell St., Newport, R.I. 02841-1708
Phone: (401) 832-5749 FAX: (401) 832-4441
E-mail: [email protected]
Ronald Morrissey
NUWC, Code 74, Bldg. 1351, 1176 Howell St., Newport, R.I. 02841-1708
Phone: (401) 832-5600 FAX: (401) 832-4441 E-mail: [email protected]
Nancy DiMarzio
NUWC, Code 74, Bldg. 1351, 1176 Howell St., Newport, R.I. 02841-1708
Phone: (401) 832-3872 FAX : (401) 832-4441 E-mail: [email protected]
Jessica Ward
NUWC, Code 74, Bldg. 1351, 1176 Howell St., Newport, R.I. 02841-1708
Phone: (401) 832-5317 FAX: (401) 832-4441 E-mail: [email protected]
David Mellinger
CIMRS, 2030 SE Marine Science Drive, Newport, OR 97365
Phone: (541) 867-0372 FAX: (541) 867-3907 E-mail: [email protected]
Steve Martin
SPAWAR Systems Center San Diego, Code 2374, 53560 Hull Street, San Diego, CA 92152-5001
Phone: (619) 553-9882 FAX: (619) 553-2726 E-mail: [email protected]
Peter Tyack
Biology Department, Woods Hole Oceanographic Institution, Woods Hole, Massachusetts 02543
Phone: (508) 457-2041 FAX: (508) 289-2818 E-mail: [email protected]
Grant Number: NFWF 2007-0145-002
Determining the density and distribution of cetacean species is fundamental to understanding their
basic biology, and also to monitoring and mitigating the effect of man-made impacts on their
populations. However, this task is difficult because most cetacean species occur at low density and
over enormous areas, and because they spend relatively little time at the surface where they can be
seen using standard, visual surveys. Our primary long-term goal is to develop and test methods for
estimating cetacean density based on detecting the sounds cetaceans make underwater, using fixed
hydrophones. There are many potential configurations of such devices, so if it does prove possible to
estimate density reliably using passive acoustics, an important second goal (not addressed in this work)
is to determine which configurations is best for a each of a common suite of monitoring scenarios.
1. Develop statistical methods for estimating the density of cetacean species from fixed passive
acoustic devices. Methods should be applicable to a wide range of scenarios, including dense and
sparse arrays of permanent, bottom-mounted sensors and single bottom-mounted or floating sensors.
2. Demonstrate the utility and generality of the methods by implementing them in a set of key test case
studies. These test cases will also focus the methodological development to ensure their relevance to
real-world applications. We aim as far as possible to leverage data that have already been collected,
and classification and localization methods that have already been developed.
3. Promote adoption of the new methods in the marine mammal research community by (a) publishing
results in the peer-reviewed literature, (b) archiving data and results in publicly available electronic
storehouses (e.g., the Ocean Biogeographic Information System, OBIS), (c) holding one or more
workshops open to all interested researchers (participants at these will be self-funded).
Technical approach
In developing the statistical methods, we will build upon the existing substantial body of work on
distance sampling survey methods (e.g., Buckland et al. 2001, 2004). Fixed passive devices are
conceptually most similar to a type of distance sampling called point transect sampling. However,
there are several important issues that prevent a straight application of existing methods, as described
in the project proposal and in Thomas and Martin (2006). We propose to develop methods to address
these issues through a series of case studies, formulated as a set of linked tasks (see below). We will
start with situations where we believe it is feasible to produce robust estimates of absolute density, and
proceed incrementally to situations where our inferences may be less reliable. Our basic approach
throughout will be to develop the new statistical methods required for the situation and to apply the
methods using acoustic data that have already been collected (but in most cases will require
processing). In doing this, we will leverage the very significant efforts that have already been
expended in data collection, and also in the development of methods for data processing and analysis.
We stress that although we will be analyzing specific datasets, our methods are designed to be general.
The case studies are as follows:
• estimation of humpback whale density at the Pacific Missile Range Facility (PMRF), Hawai’i.
• estimation of sperm whale density at the Atlantic Undersea Test and Evaluation Center
(AUTEC) range, Bahamas
• estimation of beaked whale density at the AUTEC range, Bahamas.
• estimation of sperm whale density at AUTEC using single hydrophone data.
Project investigators and roles
The research is being undertaken by an internationally-leading, multi-disciplinary team of statisticians,
acousticians, cetacean survey specialists and biologists, drawn from academia and the US military. In
summary, our major roles are as follows:
• University of St. Andrews (UStA), St. Andrews, UK. Dr. Len Thomas, is project PI, and is
collaborating with Dr. Tiago Marques and Dr. David Borchers on development of the new
statistical methods and testing by simulation. Overall project management and coordination
across all institutions is performed by Dr. Catriona Stephenson.
• Space and Naval Warfare Systems Center (SSD), San Diego, CA. Mr. Steve Martin is
overseeing the first test case (Humpbacks at Hawai’i). Martin was PI on the ONR-funded
project to collect these data.
• Oregon State University (OSU), Newport, Oregon. Dr. David Mellinger is developing an
automatic classifier for humpback whales; he is also taking the lead on developing methods for
estimating density from single fixed sensors.
• Naval Undersea Warfare Center (NUWC), Newport, RI. Mr. David Moretti is leading a team
of engineers and acousticians, including Ms. Jessica Ward, Dr. Ron Morrissey and Ms. Nancy
DiMarzio. They will use new detection algorithms developed under this project, together with
the large collection of algorithms and hardware developed previously at NUWC (under the
Marine Mammal Monitoring on Navy Ranges (M3R) program, funded by N45 and ONR) to
extract data required for the case studies, as well as participating in developing and applying
the new density estimation methods.
• Woods Hole Oceanographic Institution (WHOI), Woods Hole, MA. Dr. Peter Tyack will, in
collaboration with others at WHOI (including Dr. Mark Johnson and Ms. Amanda Hansen),
provide estimates of vocalization behavior and movement data for sperm whales and beaked
whale species required to convert estimates of click density to estimates of animal density.
In addition to the core team of investigators, there is a project steering group of acknowledged experts
in the above fields, who provide annual feedback on progress and advice on future directions. The
steering group is composed of Dr. Jay Barlow (NOAA Southwest Fisheries Science Centre), Prof.
Stephen Buckland (University of St. Andrews) and Dr. Walter Zimmer (NATO Undersea Research
Work plan for coming year
At time of writing, we are still negotiating details of the funding contract, and this is beginning to
become a significant impediment to research progress. We hope that this will be finalized soon.
Assuming this is resolved, we intend to follow our work plan as laid out in the project proposal,
although some tasks have been put back to account for the late start. Specifically, we do not expect to
complete the work on humpback whales in the 2007-8 fiscal year, although we expect to make
significant progress on this task. We also expect to make significant progress on the other three case
studies, with processed acoustic data being available to project investigators from both PMRF and
AUTEC. In the coming year, we will complete development of the new humpback whale detector, and
develop methods for analysis of these data, together with distance-based multiple hydrophone data for
sperm and beaked whales. Project progress will continue to be monitored through monthly telephone
progress meetings, and we plan to hold an annual progress meeting in July 2008.
The project is still in its early stages, and some project partners have not been able to commence work
due to the delay in finalizing the funding agreement. Nevertheless, we held the project startup meeting
in July 2007, to coincide with a workshop in Boston that most co-investigators were already attending.
Initial data provision activities have been completed. OSU has begun development of the humpback
whale song unit detector based on data from PMRF. UStA has begun development of methods based
on those of Moretti et al. (2006). This UstA work is on the case of a dense array of hydrophones
where accurate localizations cannot be made on detected vocalizations, but for which detections on one
or more hydrophones are certain for a group of animals diving together. The test data for this work is
beaked whales at AUTEC (Figure 1), although the methods are designed to be general.
The humpback detection results are too preliminary to report, so we focus on the count-based density
estimation methods. Assessment of previous methods and respective assumptions led us to propose
new approaches to the estimation procedures described in Moretti (2006). Specifically, a new “dive
counting” algorithm has been developed, where a spatial smoother is applied to a moving window
time-series of detected vocalizations at an array of hydrophones (Figure 2). This is used to determine
the number of group dive initiations per unit time. This can, in turn, be converted to an estimate of
animal group density, provided we have some information on the frequency at which groups start
dives. Such information is available from DTAG data. With knowledge of the mean group size, the
estimate of animal group density can be used to estimate animal density. On first investigation, this
approach appears promising, but it will take further development and validation before it can be
applied in more general situations. Another promising alternative is closer to standard cue counting
methods as described in Buckland et al. (2001), and we plan to process data into the required format
and investigate this in the coming year.
National Security
The US Navy is committed to marine mammal risk mitigation, both on testing ranges and exercises
outside of these areas. Methods developed under this project will contribute substantially to risk
mitigation capabilities, both in enabling more effective planning of testing and training for times and
places that minimize exposure of marine mammals to underwater sound, and also potentially in realtime monitoring of marine mammal presence.
Quality of Life
Cetaceans are an iconic part of the world’s biodiversity; the project will enable us to better monitor
their numbers and so conserve them for future generations.
None at present.
The N45 and ONR-funded Marine Mammal Monitoring on Navy Ranges (M3R) program has
developed tools capable of detecting and tracking marine mammals in real time on Navy ranges
(see proposal). Archival and new data from this program is being used to provide much of the
input data for the current project.
US Navy Pacific Fleet and Office of Naval Research have funded PMRF data collection and
analysis (see project proposal), including a manual analysis of acoustic snapshots of data that
will form the basis for a ‘quasi ground truth’ for evaluation of the Humpback whale song unit
The UK Defense Science Technology Laboratory (DSTL) is funding a PhD student based at
UStA from 2007-2010, co-supervised by Thomas, John Harwood (UStA) and Chris Clarke
(Cornell), to work on estimation of cetacean density from sparse arrays of hydrophones, such as
those of the IUSS SOSUS array. This work will proceed in close collaboration with DECAF.
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.
Buckland, S. T., Anderson, D. R., Burnham, K. P., Laake, J. L., Borchers, D., and Thomas, L. (2004).
Advanced Distance Sampling. Oxford University Press, Oxford.
Moretti, D., N. DiMarzio, R. Morrissey, J. Ward, and S. Jarvis. 2006. Estimating the density of
Blainville’s beaked whale (Mesoplodon densirostris) in the Tongue of the Ocean (TOTO) using
passive acoustics. Proceedings of the MTS/IEEE Oceans 2006 Conference and Exhibition.
Figure 1. The beaked whale test data set at AUTEC span sapproximately 6 days, from April 26th
2005 till May 2nd 2005. On the top panel each small dot represents the number of hydrophones with
non-zero counts per minute, as a function of minutes since the beginning of the recording. The
shaded areas represent night periods (18:00-06:00). The green vertical lines represent 24:00. The
bottom left plot represents the spatial layout of the AUTEC range, with each number representing a
hydrophone ID (smaller ID numbers represent inactive hydrophones during the recording period).
The bottom right plot represents a typical data set of counts of detected clicks (vocalizations) for a
given minute, illustrating the data from which we aim to estimate beaked whale density.
Figure 2. Illustration of the dive counting method, applied to beaked whale detections for a selected
29 minute period at the AUTEC range. Compare with the bottom panels of Figure 1 for spatial
reference and units. Each frame shows spatial smoothing (using a generalized additive model)
applied to click counts over a moving window of length 10 minutes, shifted by 1 minute in each
frame. A new dive was initiated in minute 31 (probably before this and then detected in this
minute), as indicated by the black circle. Similarly, the dive that started in minute 32 (black circle) is
likely outside the range boundary, and hence would be ignored in the analysis.