Google Earth AI Dataset Maps Hidden Nature Assets Across UK Farmland

Google has released a new Earth AI dataset that aims to make small natural features across agricultural landscapes easier to identify, measure, and include in climate and biodiversity planning.

The dataset focuses on fine-scale landscape features such as hedgerows, stone walls, copses, shelterbelts, and linear woodland. These features are common across farmland but are often too small, fragmented, or irregularly shaped to be captured properly by conventional satellite mapping systems. As a result, they can be underrepresented in land-use data, forest inventories, carbon assessments, and nature restoration plans.

The new resource builds on Farmscapes 2020, a high-resolution mapping project developed by Google in collaboration with the Leverhulme Centre for Nature Recovery at the University of Oxford. Farmscapes 2020 mapped semi-natural features across England’s agricultural landscapes using aerial imagery and machine learning. Google has now released a vectorized version of the dataset, turning pixel-based maps into more usable geospatial features that can support planning, analysis, and monitoring.

For farmers, land managers, conservation bodies, and public authorities, the shift from raster data to vector data is important. A raster map identifies features through pixels, while vector data represents them as defined shapes, lines, and polygons. This makes the information more practical for measuring length, area, location, connectivity, and overlap with other land-use layers.

Why Farmland Nature Data Matters

Agricultural landscapes are increasingly central to climate and nature policy. Governments and companies need to reduce greenhouse gas emissions, increase carbon storage, restore biodiversity, and improve resilience to floods, droughts, and soil degradation. At the same time, agricultural land must continue to support food production.

Large-scale afforestation and land conversion can deliver carbon and biodiversity benefits, but they may also compete with food production if poorly planned. Fine-scale nature features offer a more integrated option. Hedgerows, small woodland patches, field trees, stone walls, and shelterbelts can support wildlife corridors, provide habitat, store carbon, protect soils, reduce wind exposure, improve water retention, and connect fragmented ecosystems while remaining part of working farmland.

This is particularly relevant for the UK, where hedgerows and field boundaries have long shaped rural landscapes but have not always been mapped consistently in national datasets. Better mapping can help identify where restoration, expansion, or improved management could deliver the greatest environmental benefit.

How the Dataset Works

Google says the dataset uses Earth AI, Google Earth Engine, satellite imagery, LiDAR, aerial imagery, and deep learning techniques to detect features that are often missed by standard mapping tools. The underlying Farmscapes 2020 dataset provides 25 cm resolution probability maps for hedgerows, woodland, and stone walls across agricultural landscapes in England.

The model was developed using a vision transformer approach, a type of AI architecture used to analyse image data. According to Google’s Earth Engine documentation, the original dataset was generated by applying the model to high-resolution aerial imagery and producing separate probability layers for hedgerows, stone walls, and woodland or trees.

The newly released vectorized dataset is designed to make this information more directly usable. Instead of simply showing where a pixel is likely to represent a hedgerow or woodland feature, it provides mapped geometries that can be analysed in geographic information systems. This allows users to calculate metrics such as feature length, patch size, density, and connectivity across the landscape.

Google has said this approach can help users move from detection to planning. For example, a landowner could identify gaps in hedgerow networks, a conservation group could assess ecological corridors, and a public agency could compare restoration priorities across regions. For businesses and financial institutions, the data could also support nature-related risk assessment, carbon project screening, and supply chain sustainability analysis.

Potential Uses for Net-Zero and Nature Strategies

The release comes as companies and investors face growing expectations to measure both climate and nature impacts. Carbon accounting remains a priority, but biodiversity, land use, and ecosystem resilience are becoming more important in sustainability reporting and transition planning.

For the agricultural sector, better geospatial data can help connect farm-level decisions with wider climate and nature goals. Mapping hedgerows and woodland features can support baseline assessments, restoration planning, biodiversity net gain projects, natural capital accounting, and monitoring of nature-based solutions. It can also help verify whether interventions are adding habitat connectivity or increasing woody biomass over time.

The dataset could be useful for local authorities and national policymakers as they design land management schemes, conservation incentives, and climate adaptation strategies. It may also support companies with agricultural supply chains that need more detailed information about landscape-level environmental conditions.

However, the dataset is not a substitute for field verification. Google’s Earth Engine documentation notes several limitations in the Farmscapes 2020 dataset. The source imagery was captured between 2018 and 2020, meaning it does not reflect landscape changes after that period. Performance may also be reduced in dense urban or mountainous areas, and the stone wall class has lower accuracy than hedgerows and woodland because of class imbalance in the training data.

These caveats matter for practical use. The dataset can provide a strong baseline, but decisions involving payments, regulatory compliance, restoration finance, or carbon claims will still require appropriate validation, local expertise, and transparent methodology.

A Step Toward More Measurable Nature Restoration

The broader significance of Google’s release is that it shows how AI-based Earth observation can help close data gaps in nature and land-use planning. Many of the landscape features that matter for biodiversity and climate resilience are small, local, and difficult to capture at scale. By making them visible in a standardized geospatial format, datasets like this can help bring nature assets into decision-making processes that have historically focused on larger forests, protected areas, or land parcels.

For net-zero strategies, this is particularly relevant because land-based mitigation must be credible, measurable, and compatible with food systems. Fine-scale restoration will not replace deep emissions reductions, but it can support carbon storage, biodiversity recovery, and climate adaptation where it is well planned and monitored.

Google’s dataset provides one more tool for that planning process. Its value will depend on how it is used by farmers, researchers, policymakers, and companies, and whether it is combined with field data, ecological expertise, and transparent reporting. Used carefully, it could help turn overlooked farmland features into measurable assets for nature recovery and

Source: esgnews.com