Deep Planet and National Trust - Habitat Mapping for Conservation
Introduction
Deep Planet, a leader in AI-driven remote sensing, partnered with the National Trust, a major landholder protecting nearly 250,000 hectares of land and coastline, to enhance habitat mapping and conservation efforts. This case study focuses on the work carried out over the surveyed areas of Kipscombe (Devon), a site under National Trust stewardship.
The challenges centre around accurate ground-truthing and surveying. This pilot in remote sensing offers a tool to address these issues, but as an emerging technology, it requires further validation. Preliminary results (with final reports pending) have shown some discrepancies between satellite-derived insights and fieldwork (e.g. soil samples), which merit further investigation.
Project Objectives
The primary objectives of this initiative were to:
Create detailed, high-precision habitat maps for key sites. Each map tracks the distribution of habitat types important to National Trust site management.
Enable continuous, real-time monitoring of vegetation health to detect changes and signs of degradation. Remote sensing data enables retrospective analysis across the last five years.
Support weekly monitoring of carbon, soil moisture, and nutrients across sites to evaluate the effects of land management practices, using Deep Planet’s online portal.
Methodology
Deep Planet utilised satellite imagery combined with AI-driven analytics to map vegetation with high precision. The approach included:
Satellite Data Acquisition: Multi-spectral imagery was collected for National Trust properties, with resolution down to 0.5m to enable fine-scale classification of habitat types.
Machine Learning & AI Analysis: Models were trained to classify vegetation and habitat types and monitor temporal change.
Ground-Truth Validation: Field data and historic vegetation records were used to validate model predictions.
Key Findings & Impact
Data Visualisation & Reporting: Interactive maps and analytical reports were generated, with National Trust staff using the platform to understand baseline conditions and guide future management decisions.
Enhanced Habitat Classification: AI models classified 8 habitat types (Pasture, Grassland, Fern, Scrub, Woodland, Gorse & Heather, Bare Ground) across 200 hectares with >90% accuracy.
Sustainable Land Management: Mapping supported conservation planning and targeted restoration work.
Scalability: While the methodology has strong potential to scale across additional National Trust sites, further validation and cross-referencing with field data is needed.
Limitations & Risks
At this early stage in the project, continued fieldwork & ground-truthing remain vital for continuing to improve model outputs.
Protocols need to be developed to improve the efficiency of fieldwork surveys and reduce labour hours associated.
Conclusion & Prospects
This collaboration has demonstrated the potential of AI and remote sensing in conservation and sustainable land management. Next steps may include:
Cross-referencing remote sensing outputs with other monitoring and validation methods.
Expanding the project to new regions, enabling National Trust teams to access frequent insights without relying solely on in-field surveys.
Enhancing the AI models with additional field data to improve classification accuracy.
Developing real-time monitoring and alert systems for dynamic conversational responses.
The pilot delivered a baseline habitat map for 200 hectares—a task that would have taken significantly longer with traditional field methods.
According to Jack Ward, a representative from the National Trust, remote sensing has the potential to be a powerful tool in monitoring landscape-scale nature restoration.
This partnership indicates how emerging technology is supporting biodiversity protection and climate-resilient land management.