NEXT-COAST develops and integrates advanced technological approaches to understand and mitigate the impact of terrestrial predators on coastal bird populations. The project combines drone-based surveys, high-resolution remote sensing, computer vision, bioacoustics, and machine learning within a unified framework for ecosystem modelling and conservationplanning.

The research is embedded within the EU-funded FARMBIRD project, comprising 11 international partners, and focuses on the Leybucht pilot region as a model system for coastal avifauna under increasing ecological pressure.

AI-based monitoring and individual-level ecology

A central component of the project is the automatic re-identification (re-ID) of individual predators, enabling robust estimation of population size, spatial dynamics, and social structure. Moving beyond species-level detection, this approach allows the identification of individuals and their potential specialisations and habitat use patterns.

We employ a video-based framework using camera and drone footage, supported by transformer architectures, integrating multiple feature domains such as appearance (coat colour, morphology, facial features) and movement (gait dynamics). This multi-modal approach enables reliable individual recognition under highly variable field conditions.

In parallel, passive acoustic monitoring is implemented to quantify predator activity and disturbance intensity in situ. By analysing vocalisations, we aim to characterise not only the presence of predators but also the underlying causes and intensity of disturbance events, as well as the monitoring of adult birds and chicks.

Integrative analysis of ecosystem dynamics

This integrative framework, combining predator re-identification, bioacoustics, and telemetry - including tagged redshanks - makes it possible to identify key drivers of predation and disturbance, as well as emergent dynamics between species and their environment. These insights form the empirical basis for the development of targeted habitat restructuring measures and adaptive management strategies, including selective removal where necessary.

To support decision-making, NEXT-COAST is developing a digital twin of the Leybucht ecosystem. This modelling framework combines correlative and process-based approaches within a Bayesian hierarchical structure, enabling the simulation of management interventions prior to field implementation.

A key methodological challenge lies in addressing limited data availability, incomplete understanding of ecological processes, and the inherent complexity of ecological systems. To overcome these constraints, the project integrates approaches such as Bayesian inference and physics-informed neural networks, allowing mechanistic knowledge and empirical data to be combined in a coherent modelling framework.