The development of automated detection techniques for passive acoustic monitoring as a tool for studying beaked whale distribution and habitat preferences in the California current ecosystem

Includes bibliographical references (p. 159-166).In recent years, passive acoustic survey methods have become increasingly important in marine mammal population and ecosystem studies. Passive acoustic monitoring has been progressively combined with traditional visual survey techniques during line transect surveys for marine mammals. The objectives of this research were to test available automated detection methods for passive acoustic monitoring and integrate the best available method into standard marine mammal monitoring protocols for ship based surveys. Specifically, there were three overarching goals: 1) develop, test, and compare algorithms for automated classification of beaked whale signals; 2) employ and test techniques for beaked whale detection at sea; and, 3) use information from automated beaked whale detections to create the first acoustic based habitat models for beaked whale species and compare these with visual-based models for the same region. The goal of the first chapter was to evaluate the performance and utility of PAMGUARD 1.0 Core software for use in automated detection of marine mammal acoustic signals during towed array surveys. Three different detector configurations of PAMGUARD were compared. These automated detection algorithms were evaluated by comparing them to the results of manual detections made by an experienced bio-acoustician (author TMY). This study provides the first detailed comparisons of PAMGUARD automated detection algorithms to manual detection methods. The results of these comparisons clearly illustrate the utility of automated detection methods for odontocete species. Results of this work showed that the majority of whistles and click events can be reliably detected using PAMGUARD software. The second chapter moves beyond automated detection to examine and test automated classification algorithms for beaked whale species. Beaked whales are notoriously elusive and difficult to study, especially using visual survey methods. However, due to recent advances in passive acoustic monitoring techniques, beaked whales are now more effectively detected acoustically than visually during vessel-based (e.g. line-transect) surveys. Beaked whales signals can be discriminated from those of other cetaceans by the unique characteristics of their echolocation clicks (e.g. duration >175 ??s, center frequencies between 30-40 kHz, inter-click intervals between 0.2-0.4 sec and frequency upsweeps). Furthermore, these same characteristics make these signals ideal candidates for testing automated detection and classification algorithms. There are several different beaked whale automated detectors currently available for use. However, no comparative analysis of detectors exists. Therefore, comparison between studies and datasets is difficult. The purpose the second chapter was to test, validate, and compare algorithms for detection of beaked whales in acoustic line-transect survey data. Six different detection algorithms (XBAT, Ishmael, PAMGUARD, ERMA, GMM and FMCD) were evaluated and compared. Detection trials were run on three sample days of towed-hydrophone array recordings collected by NOAA Southwest Fisheries Science Center (SWFSC) during which were confirmed visual sightings of beaked whales (Ziphius cavirostris and Mesoplodon peruvianus). Detections also were compared to human verified acoustic detections for a subset of these data. In order to measure the probabilities of false detection, each detector was also run on three sample recordings containing clicks from another species: Risso's dolphin (Grampus griseus). Qualitative and quantitative comparisons and the detection performance of the different algorithms are discussed. Using data collected at sea from the PAMGUARD classifier developed in Chapter 2 it was possible to measure the clicks from visually verified Baird's beaked whale encounters and use this data to develop classifiers that could discriminate Baird's beaked whales from other beaked whale species in future work. Echolocation clicks from Baird's beaked whales, Berardius bairdii, were recorded during combined visual and acoustic shipboard surveys of cetacean populations in the California Current Ecosystem (CCE) and with autonomous, long-term recorders at four different sites in the Southern California Bight (SCB). The preliminary measurement of the visually validated Baird's beaked whale echolocation signals recorded from the ship-based towed array were used as a basis for identifying Baird's signals in the seafloor-mounted autonomous recorder data. Echolocation signals were segregated into four subsets based on a Gaussian Mixture Model with five mixtures over the peak frequency distribution to describe variability in the signal measurements. The median peak frequency for each of the four subsets measured from towed array and [long-term seafloor data] was at approximately 9 kHz [9 kHz], 19 kHz [16 kHz], 24 kHz [25 kHz], and 35 kHz [43 kHz]. Two distinct signal types were found, one being a beaked whale-like frequency modulated (FM) pulse, the other being a dolphin-like broadband click. Median center frequency ranged over all subsets and both recording situations from 17 to 36 kHz, -10 dB bandwidth from 6 to 13 kHz, and Teager-energy duration from 260 to 570 ??s. The median inter-pulse interval was 0.23 seconds. The description of Baird???s echolocation signals provided here will allow for studies of their distribution and abundance using towed array data without associated visual sightings and from autonomous seafloor hydrophones. The passive acoustic detection algorithms for beaked whales developed using data from Chapters 2 and 3 were field tested during a three year period to test the reliability of acoustic beaked whale monitoring techniques and to use these methods to describe beaked whale habitat in the SCB. In 2009 and 2010, PAM methods using towed hydrophone arrays were tested. These methods proved highly effective for real-time detection of beaked whales in the SCB and were subsequently implemented in 2011 to successfully detect and track beaked whales during the ongoing Southern California Behavioral Response Study (SOCAL-BRS). The three year field effort has resulted in (1) the successful classification and tracking of Cuvier's (Ziphius cavirostris), Baird's (Berardius bairdii) and unidentified Mesoplodon beaked whale species and (2) the identification of areas of previously unknown beaked whale habitat use. Thus, providing a better understanding of the relationship between beaked whale occurrence and preferred habitat on a relatively small spatial scale. These findings will provide information for more effective management and conservation of beaked whales. The final step in this research was to utilize the passive acoustic detection techniques developed herin to predictively model beaked whale habitat use and preferences in the CCE. To date beaked whale habitat models have been limited in utility due primarily to the small samples of visual observations available to inform the models. This chapter uses a multifaceted approach to model beaked whale encounter rates in the CCE. Beaked whale acoustic encounters are utilized to inform Generalized Additive Models (GAMs) of encounter rate for beaked whales in the CCE and compare these to visual based models. Acoustic and visual based models were independently developed for a small beaked whale group and Baird's beaked whales. Species distributions were modeled using a combination of fixed spatial features (depth, slope, aspect, and distance to the 2000m isobaths) and variable oceanographic variables (i.e., SST, SSS, logC, and MLD). Two models were evaluated for visual and acoustic encounters, one that also included Beaufort sea state as a predictor variable in addition to those listed and one that did not include Beaufort sea state. The visual and acoustic models differed markedly for both small beaked whales and Baird's beaked whale in the predictor variables retained in the best fit models and the regions of high encounter rate prediction. The visual models that included Beaufort sea state as a predictor variable retained this variable in the best fit resulting models. Acoustic models for the small beaked whales retained fixed spatial features of depth, slope, aspect and distance to the 2000 m isobaths as predictors in the best fit model, whereas only mixed layer depth was retained as a predictor in the best fit Baird's beaked whale acoustic model. The visual best fit model retained aspect and SST as predictor variables for small beaked whales and retained all predictor variables in the best fit Baird's beaked whale model. Differences in all models for Baird's beaked whale compared to the small beaked whales indicate that Baird's beaked whale habitat preferences may be distinctive from other beaked whale species. This work promotes current understanding of beaked whale distribution and habitat that can be used to inform beaked whale management and conservation efforts. This study has demonstrated the feasibility of using acoustic data to inform habitat models. Future work will benefit from utilizing acoustic data to inform habitat models for beaked whales and likely for other cryptic species as well. The culmination of this research has advanced techniques used in passive acoustic monitoring during towed array marine mammal surveys. The ability to efficiently detect and classify beaked whales using a towed hydrophone array represents a significant contribution to the field of marine mammal science. This work promotes current understanding of beaked whale distribution and habitat preferences and the highlights the role of behavioral and physiological processes in habitat selectio