An automated acoustic biodiversity assessment tool for rapid woodland bird health and diversity checks in eastern Australia

The use of bioacoustics has revolutionised wildlife monitoring on the back of readily available hardware and well-developed theory underpinning its application. However, the difficulty of automating analyses of recordings has limited its widespread effective use by non-specialists. We will leverage existing acoustic datasets from our past and ongoing projects (e.g., Australian Acoustic Observatory), to rapidly build, test and deploy machine-learning based algorithms (termed ‘acoustic recognisers’) for 130 key bird species from south-eastern Australian woodlands. A functional, cost-effective monitoring tool for these avian communities is in high demand, as this community readily detected using acoustic sensors, and is characterised by a high proportion of threatened taxa that are closely linked to floral and other biodiversity measures onsite. Recognisers help by automating the analysis of collected audio, outsourcing this task to a computer algorithm that provides the user with a certainty of a given species’ call being detected, with a quantified error rate based on traditional survey data we have already collected. This approach removes the need for onsite expertise, allowing non-experts to collect and analyse acoustic recordings then monitor and report on diversity and its changes with confidence. Our recognisers will be compatible with existing infrastructure for bioacoustics (e.g., Open Ecoacoustics), and most importantly developed and tested statistically to yield measures of their accuracy that users can incorporate into reporting outputs, providing a rapid, automated solution for users to assess woodland bird community condition.

Funder: Department of Climate Change, Energy, Environment and Water
Award Value: $419,475