Sounding Out Our Environment: Operationalising machine learning analysis of environmental soundscapes

You only need to sit in a woodland on a sunny day to appreciate that sound contains a wealth of environmental data - the challenge is how to extract this information and make it operationally useful.

[Link to full project description]

In this project you will work with terrestrial and underwater soundscapes to extract environmental data using machine learning classification and acoustic metrics, combined with other sensor data. We have developed a novel off grid Raspberry Pi device called SOPRANO that allows us to perform these classifications in the field and telemeter the results back to a central data hub. Currently, this is applying bird and bat classifiers, but we want to extend this to work on classifying a range of novel systems including sediment transport in rivers, underwater biodiversity and weather.

You will work closely with a range of end-users, such as Forest Research and Natural England, to develop operationally useful solutions that we want to scale to many users beyond the project. Depending upon your interests, potential scenarios could include: monitoring of biodiversity change during rewilding using a space for time substitution, monitoring the impacts of land management interventions on biodiversity and / or Natural Capital, exploring the relationship between ecoacoustic data and measures of ecosystem condition or function, analysis of sediment transport during floods or associated with scour, or alerts for invasive species detection. 

This is now a realisable ambition because we have a convergence of low cost technology, machine learning tools and a growing awareness of how to analyse environmental soundscapes. Aside from the technical challenges, for the technology to reach its potential we need to develop evidence which can help support changes in practice for organisations with regulatory and statutory obligations around environmental monitoring and environmental reporting. 

Research questions

• How can we design and deploy soundscape classifiers to address environmental challenges?
• How can we best design field experiments that leverage the benefits of traditional campaign monitoring, continuous soundscape analysis and other environmental sensor data?
• How can soundscape classifier approaches be used to monitoring the effectiveness of policy interventions for biodiversity? (e.g. Agri-environment schemes, Protected Sites, Biodiversity Net Gain)
• How can we best combine soundscape analysis with remote sensing data to derive novel approaches for monitoring biodiversity and ecosystem function?
• What are the barriers to general update of soundscape analysis in environmental science and how can these be addressed?