Abstract
Southwest Fisheries Science Center (SWFSC) has been using combined visual and acoustic
techniques to monitor marine mammal populations for the past eight years. Passive acoustic
monitoring was added to visual surveys in an effort to improve the accuracy of cetacean
population size estimates and increase the understanding of cetacean vocal behavior (Rankin et
al. 2008a-b). Acoustic detection methods are beneficial because they are not limited by most
weather conditions and are not restricted to daylight operations (Thomas et al., 1986). The
addition of passive acoustic monitoring techniques to ship-based surveys can increase both the
rate and distance of marine mammal detections (Clark and Fritrup, 1997; Gordon et al., 2000;
Barlow and Taylor, 2005). Passive acoustic methods are now an integral part of SWFSC’s
marine mammal monitoring protocol.
There are two main components to passive acoustic monitoring: detection and classification.
Detection refers to the ability to recognize marine mammal signals, whereas classification refers
to species-specific acoustic identification of those signals. Marine mammal detection requires
knowledge of marine mammal vocal behavior. Delphinid vocalizations are typically classified
into three categories: whistles, echolocation clicks, and burst pulse signals. Whistles are
continuous, narrow band, frequency-modulated signals. They can be pure tone or contain
harmonics of the fundamental frequency. Whistles are believed to function as social signals
(Janik and Slater 1998, Herzing 2000, Lammers et al. 2003) and range in duration from fractions
of a second to several seconds. They typically range in fundamental frequency from 2 to 30
kHz, depending on the species (Lammers et al., 2003; Oswald et al., 2004). Echolocation clicks
are impulsive, broadband signals that typically vary in peak frequency between 10 and over 100
kHz (Norris and Evans 1966; Au, 1980). These signals are used primarily for navigation and in
object discrimination (Au, 1993). Burst pulse signals are composed of short-interval broadband
click trains, resulting in a signal that may appear tonal due to the high repetition rate of the clicks
(Watkins, 1967; Herzing, 2000). Burst pulse sounds may be used as social signals as well as for
echolocation tasks (Dawson, 1991). These three categories of call types are not mutually
exclusive, as transitions from increasing click rates to click bursts to purely tonal signals can
occur during acoustic encounters (Murray et al., 1998).
Currently, SWFSC passive acoustic surveys of cetaceans require specially trained personnel to
continually monitor the hydrophone array signals in real-time in order to detect cetacean
vocalizations and plot bearings to the source. While effective, this method is time consuming
and costly. Automated detection of cetacean vocalizations would be a valuable tool during
marine mammal surveys, allowing for detection when experienced technicians are unavailable.
This technique is advantageous not only because it significantly reduces human effort, but also
because it removes sources of human error and bias in detection ability. Results from a recent
SWFSC study, show that acoustic detection capability varies by group size, species, and acoustic
behavior (Rankin et al., 2008b). These findings emphasize the need for comprehensive study of
species-specific vocal behavior. Reliable automated detectors could provide valuable
information about vocal behavior, species specific acoustic detectability, and vocalization rates
for several cetacean species. This is an important step in the effort to utilize acoustic line-
transect data to estimate population sizes for cetacean species.
The goal of this study was to evaluate the performance and utility of PAMGUARD 1.0 Core software for use in automated detection of marine mammal acoustic signals. Three different
detector configurations of PAMGUARD are compared. These automated detection algorithms
are evaluated by comparing them to the results of manual detections made by an experienced
bio-acoustician (author TMY). Ultimately, it is our goal to integrate automated detection and
localization methods into SWFSC’s acoustic marine mammal monitoring protocol and this work
is an important step in doing so.