Wireless sensor network-based acoustic localization for studying animal communication in terrestrial environments

This dissertation addresses problems in the development and application of passive acoustic localization systems in the context of terrestrial field biology, especially behavioral biology. A complete system with the ability to accurately determine the position of a variety of animals from the sound of their vocalizations has been developed. This has required the design and implementation of novel hardware, software, and methods of analysis.;The system developed to capture acoustic data in the field is a wireless sensor network, dubbed VoxNet. Each node is a self-contained, battery powered, four-channel microphone array. A collection of nodes forms a distributed, time-synchronized, array via wireless networking, which can be controlled from a distance to minimizes invasiveness. The array can also perform an acoustic self-survey with accuracy comparable to, or better than, GPS. A VoxNet array requires significantly less time and effort to deploy than the wired arrays of microphones typically used for acoustic monitoring and localization of animals. The design specifications of this system are provided in chapter 3.;The correlation envelope sum localization method has been designed to meet the requirements of field biologists, who have to operate in environments which pose severe challenges to most localization methods. This method, described in chapter 2, is efficient and tolerant of non-stationary noise, surveying errors, obstructions, and reverberations. It is also simple, allowing easy modification to improve performance for specific conditions, and may be used with data from any acoustic array, including wired microphone arrays.;The VoxNet system and correlation envelope sum localization method have been successfully used for observing animal behavior in the field. Chapter 4 demonstrates the ability of the system localize antbirds in a rainforest in Chiapas, Mexico. The spatial dynamics of yellow-bellied marmot alarm calls are analyzed in chapter 5.;A second method of computing localizations based on a consistency function was also developed. This method has good scaling properties and is suited to large-scale deployments and near real-time localization using a resource constrained sensor network. This is coupled with a marmot alarm call detector and applied to localizing marmot alarm calls in chapter 6.