The acoustics of climate change - long-term observations in the arctic oceans

点击翻译按钮获取中文摘要

The Arctic environment is experiencing accelerated climate change; increasing temperatures in the air and in the ocean contribute to reducing the extent of sea ice, which is essential for regulating Earth's climate through ice-albedo feedback. With less ice coverage, more solar radiation is absorbed by the ocean waters leading to a causal sequence of decreased ice formation, rising sea levels and more destabilised weather events in the south. The reduction of sea ice in the Arctic is also opening the waters for new shipping routes and developments in maritime infrastructure, which further impacts the marine ecosystem and pollutes the subsea soundscape. Both loud impulsive sounds and increased ambient noise from anthropogenic activity are major stressors for many marine species, in particular whales who make use of the idealised conditions in the Arctic for long-distance communication. This project aims to monitor the effects of climate change in the Arctic environment using passive acoustic monitoring (PAM) techniques. The soundscape of the Arctic oceans can be measured continuously, capturing sounds from shipping, sea ice processes and animal vocalisations along with changes to ambient noise levels. Studying how these sources contribute to the soundscape of the region over time will be highly informative of the rate of climate change of the region.While acoustic signatures from many underwater sources are well documented, there are new opportunities with machine learning to automate identification of acoustic events. Machine learning techniques will be developed to isolate individual sound sources from recordings for more rigorous statistical analyses. The evolution of sounds across the seasons and along the years is of particular interest, as it reflects the effects of climate change. We will use PAMGuide as a basis for PAM data processing (PAMGuide is open source, developed at Bath and used around the world). This will be used to assess conventional acoustic metrics, such as power spectral density and third-octave band level. Machine learning packages currently considered for the project are TensorFlow and Ketos (itself based on TensorFlow and designed for acoustic data) for automating acoustic signature identification. Early stages of the project will likely focus on obtaining an understanding of machine learning, not only in how to utilise it but to be able to scrutinise performance of algorithms as well.Data is provided by Ocean Networks Canada (in a shallow bay in Arctic Canada), National Oceanic and Atmospheric Administration (deep-sea moorings north of Alaska) and the EU project HiAOOS (High Arctic Ocean Observation System) (across the Arctic). Data from ONC and NOAA has already been collected, whereas data from the HiAOOS project is expected to be received in 2025. Funding for the project is provided by the Engineering and Physical Sciences Research Council (EPSRC) Doctoral Training Partnerships (DTP). The data stored at Bath will be subject to a Data Management Plan, in line with University guidelines.