Discussing effective conservation with all the UK Chief Scientists

Key findings from this study

• The meeting established that geospatial foundation models could enable interpolation of ecological data between sampled locations to assess intervention impacts at ecosystem scales.
• The authors presented evidence that traceable, open-source AI pipelines offer reproducible decision support while proprietary chatbots create black-box dependencies incompatible with governmental sovereignty.
• The gathering demonstrated that chief scientists from all UK statutory nature conservation bodies are actively exploring machine learning applications for accelerating nature recovery and preservation programs.

Overview

Conservation Evidence hosted a two-day meeting at Pembroke College in January to examine applications of artificial intelligence and machine learning for biodiversity conservation in the United Kingdom. The first day convened chief scientists from all five UK statutory nature conservation bodies plus the Joint Nature Conservation Committee. The second day featured a broader conference on evidence-based conservation practices. The event facilitated direct dialogue between academic researchers developing geospatial foundation models and conservation decision-makers responsible for implementing nature recovery interventions across UK jurisdictions. Presentations addressed the TESSERA project, which applies machine learning to ecological monitoring data, and the Conservation Evidence TAP system. The gathering emphasized the need for traceable, open-source AI pipelines rather than black-box commercial tools to maintain sovereign decision-making capability in conservation policy.

Methods and approach

The first day comprised a closed meeting where researchers presented the TESSERA geospatial foundation model to chief scientists from England, Scotland, Wales, Northern Ireland, and the Joint Nature Conservation Committee. Presenters demonstrated how machine learning embeddings could be combined with existing ground-based ecological data to interpolate between sampled locations. The proposed approach trains downstream models using the embeddings to enable larger-scale questions about ecosystem change and intervention impacts across outcomes currently difficult to assess. The second day featured public presentations in the Pembroke Auditorium covering evidence-based conservation methodologies, causal inference in biodiversity studies, and AI applications. Presentations were recorded and made available through both mainstream and Fediverse platforms. Discussions centered on choosing appropriate study designs for growing biodiversity datasets and avoiding uninterpretable causal relationships in post-hoc analyses.

Results

The chief scientists engaged in substantive discussion about how geospatial foundation models may influence future work of statutory nature conservation bodies. The meeting established next steps involving training and validating downstream models that combine ground-based data with machine learning embeddings. This approach aims to enable interpolation of datasets between sampled locations and support ecosystem-scale questions about intervention impacts. Presentations highlighted contrasts between off-the-shelf chatbots offering convenience but lacking reproducibility and provenance tracing, versus engineered open pipelines providing transparent processing. Example applications included mapping tree coverage in Northern Ireland and analyzing bird nest box design effectiveness. The conference format enabled direct connections between evidence synthesis researchers and territorial conservation agencies responsible for implementing biodiversity interventions across UK jurisdictions.

Implications

The gathering established a framework for ongoing collaboration between academic AI researchers and statutory conservation bodies on operationalizing machine learning for nature recovery. The emphasis on open, traceable processing pipelines rather than proprietary AI tools reflects broader concerns about maintaining governmental decision-making sovereignty in conservation policy. The proposed interpolation approach could expand the spatial scale and outcome diversity of ecosystem change assessments beyond current ground-based sampling limitations. This capability addresses a fundamental constraint in evaluating conservation interventions across heterogeneous landscapes. The meeting demonstrated growing normalization of evidence-based approaches in conservation practice, though significant design and causal inference challenges remain in utilizing expanding biodiversity datasets effectively. The connection between foundation model development and territorial agency needs creates pathways for research outputs to influence operational conservation decisions across England, Scotland, Wales, Northern Ireland, and cross-border coordination mechanisms.

Scope and limitations

This summary is based on the study abstract and available metadata. It does not include a full analysis of the complete paper, supplementary materials, or underlying datasets unless explicitly stated. Findings should be interpreted in the context of the original publication.