IAP-25-085

Listening to Food Webs: Ecoacoustic insights into Bird Predator–Prey Networks in Changing Landscapes

Ecoacoustics — the study of environmental soundscapes — is increasingly being used to monitor land-use change because it offers a non-invasive, scalable, and continuous way to assess biodiversity and ecosystem health. Passive acoustic monitoring (PAM) using Autonomous Recording Units (ARUs) is especially valuable for studying birds, whose vocalisations dominate many soundscapes and provide rich ecological information [1-3]. Beyond measuring avian abundance and diversity, ecoacoustic approaches have the potential to reveal behavioural and ecological processes, such as identifying alarm calls to infer predator–prey interactions. When integrated with other biomonitoring technologies, such as drone surveys and environmental DNA (eDNA), artificial intelligence (AI) and deep learning methods can combine these complementary data sources to construct complex species-interaction networks [4]. Such approaches could transform our understanding of how biodiversity and ecosystem functioning respond to environmental change, yet empirical applications are rare.

This project will use the long-term Glen Finglas landscape-scale grazing experiment, established in 2002, to examine how variation in livestock (sheep and cattle) stocking densities influences upland bird and animal communities. After 15 years of grazing treatments, substantial shifts in vegetation and animal communities have been recorded [5,6], including higher bird species richness in ungrazed areas. These vegetation-driven changes are likely to cascade through other trophic levels, yet the impacts on whole food-web structure remain poorly understood. Focusing on maximising the information derived from ARUs, this project will detect, classify, and validate passerine calls across experimental treatments to infer predator–prey interactions. It will then develop integrated biomonitoring methods (including drone, field, and long-term datasets) aided by deep learning to construct highly resolved species-interaction networks encompassing plants, arthropods, mammals, and birds. Finally, the project will assess how grazing treatments influence the structure, complexity, and robustness of upland food-webs [7], providing policy-relevant evidence on how land management affects ecosystem resilience.

The project has three interlinked objectives:

OBJ. 1: Detect, classify, and validate bird alarm and other calls to infer inter- and intra-specific interactions.

OBJ. 2: Integrate biomonitoring technologies using data science and AI to construct upland food-webs consisting of plants, arthropods, mammals, and birds based on observed and inferred species interactions of these taxa collected over a 20-year period.

OBJ. 3: Compare the structure, complexity, and robustness of upland food-webs under different experimental grazing treatments.

Study System
The Glen Finglas grazing experiment in central Scotland was established in 2003 as a replicated, randomised-block design with six replicates of four grazing treatments, following baseline data collection in 2002. Each plot is approximately 3.3 ha and spans altitudes from 200–500 m above sea level. Treatments comprise:
(I) 9 ewes per plot (2.72 ewes ha⁻¹),
(II) 3 ewes per plot (0.91 ewes ha⁻¹),
(III) 2 ewes per plot (0.61 ewes ha⁻¹) plus autumn cattle grazing to match offtake in (II), and
(IV) ungrazed control plots.
Within each plot, vegetation, small mammals (field voles), and breeding birds are surveyed annually, with additional surveys of foliar arthropods and other taxa (e.g. moths, beetles, foxes) conducted every three years. Long-term data, including recent drone-based remote-sensing surveys [9], provide an exceptional basis for multi-trophic analysis.

Ecoacoustic Monitoring
An array of Autonomous Recording Units (ARUs) will be deployed across the 24 plots to obtain high-resolution spatio-temporal data under all four grazing treatments. Bird alarm and other calls will be detected, classified, and validated using both AI-based classifiers and direct field observations (e.g. nest focal watches and concurrent camera traps). These data will form the basis for constructing plot-level bird predator–prey interaction networks.

Data Integration and Analysis
The project will develop and test approaches for integrating multiple biomonitoring data streams—including ecoacoustic, drone, satellite, and eDNA datasets—using deep learning techniques. Specifically, large language model (LLM) approaches will be explored to maximise information from heterogeneous data sources to infer species interactions and evaluate changes in biodiversity and ecosystem functioning [4]. Network analyses, including Bayesian hierarchical models, will then be applied to assess how grazing intensity affects network structure, complexity, and robustness (i.e., the tolerance of networks to species loss [7]). The experimental design and replication make it well suited for hierarchical and Bayesian modelling frameworks.

Fieldwork and Collaboration
The student will conduct fieldwork at Glen Finglas with support from the James Hutton Institute (JHI), which curates the long-term datasets. A placement at JHI (Aberdeen) will provide access to data management and advanced analytical support. Collaboration with the British Trust for Ornithology (BTO) will offer hands-on training in the deployment, analysis, and interpretation of PAM data, as well as opportunities to extend validation studies across wider upland landscapes.

Year 1

Introduction to 20-year dataset and training in data science and AI, ecoacoustics, food-web construction methods, Deep Learning and ecological network modelling. Deployment of AudioMoths, camera traps and potential invertebrate biomonitoring technologies, and supplementary plant and animal sampling at Glen Finglas to fill identified ‘interaction gaps’ in the dataset (Season 1). Validation of bird calls and interactions using direct observations in the field.

Year 2

Deployment of PAM arrays, and supplementary plant and animal sampling at Glen Finglas (Season 2). Validation of interactions based on ‘predictive link’ models and field validation. Modelling the impacts of livestock treatments on the upland food-web, focussing on network structure, complexity and robustness (contrasting qualitative and quantitative structures and using Frequentist and Bayesian approaches). First publication.

Year 3

Training in adaptive network models, parameter testing and analysis using the long-term dataset. Forecast management outcomes using networks predictively. Second publication. Communicate results to stakeholders, the public and academic community.

Year 3.5

Thesis writing and final publications

The student will gain an exceptional combination of ecological, analytical, and data-science expertise by working across leading research organisations: Newcastle University, Durham University, the James Hutton Institute (JHI), and the British Trust for Ornithology (BTO). This interdisciplinary team provides expertise in applied ecology, ornithology, network science, and ecological modelling.

The student will have full access to Newcastle’s SAgE Faculty Researcher Development Framework (RDF), which offers training across the four Vitae domains: knowledge and intellectual abilities, personal effectiveness, research governance and organisation, and engagement, influence and impact.

Specifically, the project will provide advanced skills in:
1. Ecological census and biomonitoring methods – including deployment of ecoacoustic and drone-based technologies.
2. Network construction and data integration – applying Deep Learning and AI approaches, with support from Newcastle’s School of Computing and access to the Rocket HPC cluster.
3. Statistical and bioinformatic analysis – including Bayesian modelling, data visualisation, and the use of R and Python.
4. Science communication and stakeholder engagement – with opportunities to contribute to policy outreach and public communication (e.g., articles for The Conversation).

Further training in statistical methods will be available through JHI and Biomathematics and Statistics Scotland (BioSS). This integrated training programme will ensure the student develops the technical and transferable skills required for an interdisciplinary research career spanning ecology, AI, and environmental data science.