Workshop explores how advances in AI and Earth Observation could support restoration, monitoring and environmental policy.
Led by the CLR's Barbara Net-Bradley and Jessica Williams, this cutting-edge workshop brought together researchers, statutory agencies and restoration practitioners in Cambridge on 23–24 March 2026. It explored the latest technological advances emerging from collaboration between Cambridge’s Computer Science Department and the Centre for Landscape Regeneration.
The aim and outcome of the workshop was to bring people together who are at the forefront of this developing field, exploring how a new generation of geospatial tools can be applied to long-standing ecological challenges — including identifying and mapping habitats at scale, as well as monitoring biodiversity and assessing how healthy those habitats are.
The event was held at the David Attenborough Building and brought together participants from across government, research and practice, including Defra, Natural England, NatureScot, JNCC, UKCEH, the National Trust, Rewilding Europe, the Endangered Landscapes & Seascapes Programme (ELSP), PlantNet, and several universities and research organisations.
What is TESSERA and why it matters
The workshop focused on TESSERA, a geospatial foundation model developed at Cambridge and trained on large volumes of satellite imagery from the Sentinel-1 and Sentinel-2 missions, which capture radar and optical images of the Earth.
Unlike more traditional remote-sensing approaches, TESSERA builds a detailed picture of land conditions. This can then be used for a wide range of ecological tasks, making it faster and easier to create habitat and species models across much larger areas.
The challenges of environmental monitoring at scale
Across the UK and Europe, there is growing pressure to show whether investment in restoration and environmental land management is delivering results. However, current habitat monitoring systems are often expensive, fragmented and difficult to scale.
We asked participants to complete a questionnaire about the challenges they face. Several key issues emerged:
- limited access to ground-truth field data (real-world information collected directly at a site—rather than inferred remotely)
- weak standardisation between field and remote-sensing datasets (field data and satellite data)
- incompatible habitat classification systems across organisations (different organisations classify habitats in different ways, making it hard to compare data)
Robust monitoring underpins both ecological understanding and policy delivery. Agri-environment schemes, landscape recovery programmes and restoration initiatives all require reliable evidence of change over time.
Yet many current approaches still depend heavily on labour-intensive surveys or one-off habitat snapshots. A key message from the workshop was that technological barriers are shrinking quickly, while the main remaining challenges now lie in ecological definitions, data standards, monitoring methods and coordination between organisations.
Can geospatial foundation models support ecological monitoring?
The event showcased a series of applied examples demonstrating how geospatial foundation models can already support ecological monitoring in practice.
In the Cairngorms, TESSERA was used to predict the cover of species and habitat components — including heather, pine forest, cotton grass, deer grass and bare peatland — across the whole national park using a relatively small number of vegetation plots. It also showed potential for estimating woodland height in areas of heathland expansion at far lower cost than airborne LiDAR.
In Cumbria, the workshop highlighted the use of TESSERA for UKHab habitat classification and for detecting vegetation recovery following river restoration. This demonstrated how the model could support monitoring of landscape-scale recovery over time.
Additional case studies included tree species detection in Trentino, Italy, where TESSERA outperformed more conventional satellite composite methods, and wildfire disturbance mapping. In the latter case, a model trained in the United States was able to detect burned areas in Scotland with strong accuracy.
A further demonstration showed how the model could help identify potentially misclassified entries in existing government datasets, offering a rapid way to flag labelling errors and improve the quality of legacy monitoring data.
Taken together, these examples suggest that foundation models could make habitat monitoring faster, cheaper, more transferable and easier to update over time.
A growing ecosystem of tools
The workshop also placed TESSERA within a broader landscape of national and international monitoring initiatives.
Participants heard about Natural England’s Living England map, JNCC’s HabCon project, UKCEH’s long-running environmental monitoring programmes, and PlantNet’s work on citizen science-based, AI-driven species distribution mapping.
A strong theme throughout discussions was that the next step is not simply to build more tools in isolation, but to improve integration and interoperability across existing systems and datasets.
From proof of concept to practical use
The workshop identified several priority challenges for moving from promising demonstrations to operational use.
These include the need to:
- agree clearer definitions of habitat condition
- improve data access and metadata standards
- develop better methods for detecting change over time
- build more accessible interfaces for practitioners without specialist infrastructure
Participants also emphasised the importance of linking large-scale monitoring data to causal inference approaches, so that observed ecological changes can be more confidently attributed to specific interventions or policies.
For policy audiences, this is particularly relevant. A presentation on behalf of Defra highlighted the difficulty of evaluating agri-environment schemes, where ecological change can be slow, variable and hard to attribute.
The workshop’s overall assessment was that the tools needed for landscape-scale monitoring are now either available or close to being available. However, progress will depend on shared standards, improved data infrastructure, and sustained investment in the capacity needed to use these tools effectively.
What comes next
One of the clearest outcomes from the workshop was the need for a more coordinated roadmap bringing together Defra, statutory agencies, research institutions and delivery organisations.
Priority actions identified by participants included:
- developing EO-compatible (suitable for use with satellite data) habitat condition indicators
- standardising field survey protocols
- establishing a shared ground-truth data platform
- aligning classification systems to enable comparison across programmes
On the technical side, improving uncertainty reporting was also identified as a priority.
For the Centre for Landscape Regeneration, the workshop highlighted how this work connects directly to wider research across Cambridge and beyond. From species mapping in the Cairngorms to habitat mapping and recovery monitoring in Cumbria, discussions showed how open-source geospatial AI tools can help bridge the gap between research innovation and real-world landscape regeneration.
Acknowledgements
This meeting was primarily supported by a grant from the Endangered Landscapes & Seascapes Programme, with additional support from the NERC-funded Centre for Landscape Regeneration.
