Predictive foundational AI for ecosystem resilience forecasting

Conserving ecosystems into the future requires a predictive ecology revolution. Foundational AI models are able to surpass the limitations of existing predictive models by encoding high complexity across global multi-sensor environmental data into a single digital representation, an ‘embedding’, which can be used as data in downstream analyses.

This project aims to produce the first embeddings explicitly trained on future climate projections across the land-sea interface up to 2100. The team will be focusing their approach on UK seabirds, a highly threatened yet functionally critical and mobile group spanning marine and terrestrial ecosystems, validated by tracking seabird’s movements at the UK’s warming edge on the Isles of Scilly. This provides a scalable testbed for extending predictive ecosystem engineering to other vulnerable taxa and regions.

The overall aim is to create a globally scalable toolkit for predicting future habitat suitability, identifying populations at risk of local extinction, and highlighting places where targeted interventions could improve resilience.

This project is funded by the Advanced Research + Invention Agency (ARIA) within their Engineering Ecosystem Resilience (EER) opportunity space, which is exploring if combining high-resolution measurement with targeted, resilience-boosting interventions could reverse biodiversity decline and prevent ecological collapse.