Turning imagery into evidence
Our data scientists, statisticians and ecologists build AI that turns aerial, drone and satellite imagery into accurate, auditable ecological data, and then tell you what it means. Within our team we cover the end-to-end process including: survey design → image processing → AI detection and mapping → expert review → field validation → statistical analysis → GIS layers, reports and interactive dashboards.
Every model we deliver sits inside a documented workflow, from raw imagery to the densities, habitat maps and reports that regulators rely on.
What our AI does
• Detects and counts wildlife. YOLO models find and identify birds, marine mammals and other wildlife in survey imagery and video, down to targets of around 2 cm, and count densely packed colonies and roosts from drone imagery.
• Maps habitats. U-Net segmentation, combined with the Segment Anything Model (SAM) and ensemble approaches in hybrid models, produces fine-scale habitat maps aligned to UKHab and Fossitt, with repeat mapping to track condition and change for restoration, Biodiversity Net Gain and monitoring.
• Learns from experts. Training data is labelled by experienced ornithologists and ecologists and held in curated image libraries that can be audited, reused and shared safely.
From what to why
Many providers can find animals in an image. Few can tell you what the results mean. Our statisticians turn AI outputs into answers:
• How many are there? Site-wide abundance estimates with confidence intervals.
• Where are they, and why? Density surface and species distribution models linking animals to depth, sea temperature, colonies and habitat.
• How is it connected? Network analysis of habitat patches, colonies and foraging areas.
• Is it changing? Trend and BACI analyses that separate real effects from natural variation.
• What does it mean for the project? Robust inputs to collision risk, displacement and population viability models for EIA and HRA.
AI error rates are carried through to the final estimates, so uncertainty is reported honestly.
Quality you can trust
• Accuracy reported transparently by species, habitat and platform.
• Statistically designed audits, including checks for what the model missed.
• Version-controlled, reproducible models and data.
• Validated against ground truth from our own field ecologists.
• Proven on large, high-resolution survey datasets.
Built with Python, PyTorch, YOLO, U-Net, SAM, Label Studio, WebODM, R, Shiny and QGIS.
Talk to us about how AI and statistical modelling can make your surveys faster, more consistent and more cost-effective, and turn the results into answers you can act on.
