Google Releases Earth AI Dataset to Map Hidden Nature Assets Across UK Farmland

Google’s new vectorized dataset maps fine-scale ecological features across England, including hedgerows, stone walls, copses, and linear woodlands.
The tool can support carbon accounting, biodiversity planning, and nature restoration without removing productive farmland from use.
The dataset uses Google Earth AI, Google Earth Engine, satellite imagery, LiDAR, and deep learning to identify features often missed by standard satellite systems.

Google Turns Countryside Data Into a Restoration Tool

Google has released a new Earth AI dataset designed to make overlooked natural features across UK farmland visible, measurable, and usable for climate and biodiversity planning.

The vectorized dataset converts high-resolution ecological maps into an actionable inventory of hedgerows, stone walls, copses, and other fine-scale landscape features. These features often sit between fields, beside boundaries, or across working farms. Yet many remain absent from national forest inventories because they are too small for standard satellite detection.

The release builds on Farmscapes 2020, the first large-scale, high-resolution map of overlooked woody features across England. Google developed that project with the Leverhulme Centre for Nature Recovery at the University of Oxford.

The latest dataset moves the work beyond pixels. It gives landowners, conservation groups, policymakers, and researchers more precise vector data. That matters for project design, carbon accounting, biodiversity monitoring, and nature restoration.

Why Small Landscape Features Matter

Forests play a central role in climate and nature systems. They store carbon, filter water, regulate ecosystems, and support biodiversity. But expanding large forests can compete with farmland, especially as food demand rises.

That trade-off has become one of the central challenges in land-based climate action. Governments and companies need to increase carbon storage and restore nature. They also need to protect food production and avoid leakage, where conservation gains in one place drive environmental damage elsewhere.

Fine-scale woody features offer a practical pathway. Hedgerows, shelterbelts, small copses, and linear woodlands can increase carbon storage and improve habitats without displacing crops. They also help connect fragmented ecosystems. This can support wildlife movement across agricultural landscapes.

Google said the new dataset aims to help make these small features count in planning and reporting.

“We developed a high-resolution deep learning framework to reveal fine-scale ecological features, like hedgerows and copses, that are typically invisible to standard satellite detection. This precise vector data offers a new pathway to address the climate and biodiversity crises on working lands without compromising food security.”

Inside the AI Mapping System

Google’s team had to solve three technical challenges: landscape complexity, ecological meaning, and computational scale.

Agricultural landscapes rarely follow simple shapes. A hedgerow can sit next to a field boundary or above a stone wall. Standard mapping systems struggle with overlapping features. Large maps also need to be broken into tiles, which can slice features at the borders.

Google used a dual-layer labeling system that combines submeter imagery and 1-meter LiDAR data. This allowed the model to detect both ground-level boundaries and above-ground features. It then used an algorithm to merge geometries across tiles. That helped produce complete shapes rather than fragmented map sections.

The second challenge was meaning. A woody pixel does not show whether a feature is a forest core, a wildlife corridor, or an isolated tree group. Google applied the Polsby–Popper compactness score to classify shapes by their physical footprint.

The model defined woodlands as substantial canopies with at least a 30-meter diameter. It classified small copses and individual trees as woody patches. It identified hedgerows and elongated corridors through stretched footprints with a compactness score below 0.5.

The third challenge was scale. England covers more than 130,000 km². Processing millions of woody features across that area would overwhelm many standard systems.

Google used Google Earth Engine to process thousands of S2-cell tiles in parallel. It also used Remote Sensing Foundations’ Vision-Transformer Backbone, trained on more than 300 million global satellite images. Google then fine-tuned the model on about 247 km² of annotated British countryside data.

What Investors and Policymakers Should Take Away

For executives, the dataset points to a broader shift in nature-related data. Companies face growing pressure to measure land, biodiversity, and climate impacts with greater precision. Investors also need more credible data to assess nature-based solutions and land-use risk.

For policymakers, the tool can support restoration planning across working landscapes. It may help identify where hedgerows, copses, and shelterbelts can expand carbon storage and habitat value without reducing agricultural output.

Google said it is also exploring how high-precision detection could support silvopasture and agrisilviculture systems. These approaches combine trees with livestock or crops. The technology may also help detect leakage around project boundaries.

The wider significance is clear. Nature restoration will not rely only on major forest expansion. It will also depend on the smaller features that shape rural ecosystems. By making them visible, Google is giving governments, farmers, and markets a stronger data foundation for climate and biodiversity action.