As vital ecosystems across the globe enter unchartered pressure from climate change industrial land use, understanding the processes driving ecosystem viability has never been more critical. Nuanced ecosystem understanding comes from well-collected field data and a wealth of associated interpretations. In recent years the most popular methods of ecosystem monitoring have revolutionised from often damaging and labour-intensive manual data collection to automated methods of data collection and analysis. Sound ecology describes the school of research that uses information transmitted through sound to infer properties about an area's species, biodiversity, and health. In this thesis, we explore and develop state-of-the-art automated monitoring with sound, specifically relating to data storage practice and spatial acoustic recording and data analysis.
In the first chapter, we explore the necessity and methods of ecosystem monitoring, focusing on acoustic monitoring, later exploring how and why sound is recorded and the current state-of-the-art in acoustic monitoring. Chapter one concludes with us setting out the aims and overall content of the following chapters. We begin the second chapter by exploring methods used to mitigate data storage expense, a widespread issue as automated methods quickly amass vast amounts of data which can be expensive and impractical to manage. Importantly I explain how these data management practices are often used without known consequence, something I then address. Specifically, I present evidence that the most used data reduction methods (namely compression and temporal subsetting) have a surprisingly small impact on the information content of recorded sound compared to the method of analysis. This work also adds to the increasing evidence that deep learning-based methods of environmental sound quantification are more powerful and robust to experimental variation than more traditional acoustic indices.
In the latter chapters, I focus on using multichannel acoustic recording for sound-source localisation. Knowing where a sound originated has a range of ecological uses, including counting individuals, locating threats, and monitoring habitat use. While an exciting application of acoustic technology, spatial acoustics has had minimal uptake owing to the expense, impracticality and inaccessibility of equipment. In my third chapter, I introduce MAARU (Multichannel Acoustic Autonomous Recording Unit), a low-cost, easy-to-use and accessible solution to this problem. I explain the software and hardware necessary for spatial recording and show how MAARU can be used to localise the direction of a sound to within ±10˚ accurately. In the fourth chapter, I explore how MAARU devices deployed in the field can be used for enhanced ecosystem monitoring by spatially clustering individuals by calling directions for more accurate abundance approximations and crude species-specific habitat usage monitoring. Most literature on spatial acoustics cites the need for many accurately synced recording devices over an area. This chapter provides the first evidence of advances made with just one recorder.
Finally, I conclude this thesis by restating my aims and discussing my success in achieving them. Specifically, in the thesis’ conclusion, I reiterate the contributions made to the field as a direct result of this work and outline some possible development avenues.