Investigations of ecoacoustic biodiversity characterized with remote sensing data using machine and deep learning approaches

Quinn, Colin Aidan (2023) Investigations of ecoacoustic biodiversity characterized with remote sensing data using machine and deep learning approaches. Doctoral thesis, Northern Arizona University.

Text Quinn_2023_investigations_ecoacoustic_biodiversity_characterized_with_.pdf - Published Version Download (6MB)

Abstract

Animal biodiversity assessment provides key measurements on the health and dynamics of animals in relation to habitat setting, human presence, and disturbances. As these factors continue to change due to climate change, introducing more intense and frequent wildfires, for example, assessing animal dynamics allows for proper protection and scientific understanding of the indicators and influences leading to animal community change. Ecoacoustic recording of the landscape is a rapidly evolving new approach in capturing the dynamics of animals and human impact. Recording passive soundscapes, the collection of noises emanating from the landscape, amounts to large quantities of acoustic data that require informed methods for summarizing underlying patterns so methods are generalizable and reliable. Many of these efforts over the previous decade and a half focused on a suite of metrics called acoustic indices that summarize dimensions of the acoustic signals such as frequency and amplitude. However, acoustic indices lack generalizability when compared among study regions particularly in relation to biotic noise sources (Biophony) and anthropogenic noise sources (Anthropophony). These gaps in understanding and lack of confidence in applying acoustic indices to our dataset of over 700,000 minutes of recordings from Sonoma County, California and the Soundscapes to Landscapes project led to the creation of the first research project in this dissertation where I classified soundscapes into informative types of sound called soundscape components using convolutional neural networks (Chapter 2). The second project then analyzed the patterns in classified soundscapes with a set of fifteen acoustic indices to investigate in what sonic conditions indices may be more reliable and how Biophony along with indices can model bird species richness (Chapter 3). From these two chapters I provided a classification approach to distill sounds types in 700,000 minutes of acoustic data and related these sounds to acoustic indices to assess their ability to reflect signals of biodiversity and robustness in the presence of confounding sound sources. The final project combined the above work to analyze spatio-temporal patterns of acoustic indices and soundscape components across Sonoma County (Chapter 4). This included modeling the ecoacoustic metrics with a suite of remote sensing variables in a species distribution modeling framework. Products from these models were used to understand the most important metrics, structural and climate metrics, in predicting ecoacoustic metrics. Maps were used to investigate increased Biophony levels following wildlife activity across increasing burn severities and patterns related to human impact. Combined, this work demonstrates the ability to leverage machine learning approaches to understand the complex nature of soundscape dynamics using large datasets in a regional-scale analysis.

Item Type:	Thesis (Doctoral)
Publisher’s Statement:	© Copyright is held by the author. Digital access to this material is made possible by the Cline Library, Northern Arizona University. Further transmission, reproduction or presentation of protected items is prohibited except with permission of the author.
Keywords:	Biophony; Distribution modeling; Ecoacoustics; Lidar; Machine Learning; Soundscapes; Human-Animal Ineraction
Subjects:	Q Science > QL Zoology
NAU Depositing Author Academic Status:	Student
Department/Unit:	Graduate College > Theses and Dissertations College of Engineering, Informatics, and Applied Sciences > School of Informatics, Computing, and Cyber Systems
Date Deposited:	23 Jun 2025 17:14
Last Modified:	23 Jun 2025 17:14
URI:	https://openknowledge.nau.edu/id/eprint/6196

Actions (login required)

IR Staff Record View

Downloads