Lost historic woodland 1891-1914 (AI predicted) dataset for England
Published 27 August 2026
Applies to England
© Crown copyright 2026
This publication is licensed under the terms of the Open Government Licence v3.0 except where otherwise stated. To view this licence, visit nationalarchives.gov.uk/doc/open-government-licence/version/3 or write to the Information Policy Team, The National Archives, Kew, London TW9 4DU, or email: psi@nationalarchives.gov.uk.
Where we have identified any third party copyright information you will need to obtain permission from the copyright holders concerned.
This publication is available at https://www.gov.uk/government/publications/historic-woodland-datasets-for-forestry-proposals/lost-historic-woodland-1891-1914-ai-predicted-dataset-for-england
Find out about areas in England that may be suitable for new woodland, based on lost historic woodland information.
About the lost historic woodland (AI predicted) dataset
A national dataset for England showing land where woodland was mapped between approximately 1891 and 1914 but is no longer present. The dataset has been generated using an artificial intelligence (AI) model applied to historic 2nd edition Ordnance Survey (OS) maps. It indicates the extent and type of historic woodland. This historic woodland dataset has been compared to the 2024 National Forest Inventory (NFI) to identify areas of historic woodland that no longer survive.
The dataset can support the development of forestry projects, analysis, interpretation and decision making. It can help inform strategy, planning and research.
The AI predictions were subject to post-processing to create a consistent, usable national dataset for England published under an Open Government Licence.
Summary of the dataset
The dataset uses an AI model that identifies woodland symbols from digitised historic OS maps. These areas of woodland have been compared to the 2024 NFI dataset to see where woodland shown on the historic maps is no longer present. It shows probable extent and composition of ‘lost’ historic woodland.
You can use the outputs to support research, planning and environmental analysis. In particular where understanding long-term woodland presence and change over time is important.
The Welsh Government Data Science Unit and the Forestry Commission developed this dataset. The National Library of Scotland provided access to the historic mapping used.
How to use the data
Lost woodland areas may retain characteristics that make them suitable for woodland creation. You can use the data to help inform decisions in your forestry proposal about future land management when used with other evidence.
You can combine the dataset with other environmental, ecological or land use datasets to provide further context. For example, when assessing habitat networks, landscape change or environmental opportunity areas.
You should interpret the outputs from this dataset alongside other available evidence. This may include modern datasets, local records and specialist knowledge. It may be also necessary to apply quality assurance, filtering or validation processes. You should not rely on the outputs in isolation.
The dataset is not suitable for precise measurement of lost woodland extent or for use in statutory or regulatory decisions without supporting evidence or acknowledging these limitations.
What the dataset contains
The dataset consists of polygons representing areas where woodland was mapped on historic OS maps but no longer exists according to the 2024 NFI. Each polygon includes confidence scores that indicate how strongly the model predicts the presence of different woodland types that previously existed within that area.
In some cases, more than one woodland type may be associated with a single polygon. This reflects uncertainty or mixed woodland characteristics in the underlying historic map.
Woodland types included in the dataset are:
- broadleaf
- brushwood
- bushes
- conifer
- mixed woodland
- orchard
- osiers
How the dataset was created
Historic OS map tiles were processed and analysed using an image segmentation model trained to recognise woodland symbols and patterns initially within Wales. The model was trained using manually labelled historic OS map data and designed to detect different types of woodland based on their cartographic representation.
The trained model was applied systematically across England to generate predictions for woodland extent. These predictions were converted into polygon features and georeferenced to align with modern mapping system coordinates.
The Welsh Government Data Science Unit and Forestry Commission applied a series of geospatial processing steps to improve the usability of the dataset. These included:
- combining overlapping predictions
- resolving inconsistencies between adjacent areas
- simplifying geometries
- and ensuring the dataset could be used effectively within Geographic Information Systems
Comparison was then made between these detections and the 2024 NFI. This identified areas of woodland that had been lost since 1891-1914.
Where to access the dataset
You can download the lost historic woodland 1891-1914 (AI predicted) dataset for England from the:
You can also view it using the Forestry Commission map browser.
The map is open data and is covered by the terms and conditions of the Open Government Licence.
Data quality and limitations
This dataset is derived from the predictions of an AI model and includes a degree of uncertainty. It does not represent a definitive or complete record of lost historic woodland and should not be interpreted as such.
Accuracy is influenced by the quality and consistency of the underlying historic OS maps and the ability of the AI model to correctly interpret different map symbols. Some areas of England have lower map quality or use alternative symbology, which can affect prediction accuracy.
Known issues include:
- misclassification of non-woodland features such as buildings, or errors where shading or linework on the historic maps resemble woodland symbols
- artificial boundaries or visible edges between areas where map tiles were processed separately
- small, fragmented polygons created during data processing
Differences between historic map sources, including variations in scale and symbol design, can lead to inconsistencies in how lost woodland is represented across the dataset. Some woodland types, particularly brushwood, are more prone to misclassification and may require additional filtering or validation before use.
Further information
For further information on the lost historic woodland 1891-1914 (AI predicted) dataset for England, email historic@forestrycommission.gov.uk.
For more advice on the historic environment in forestry, read our historic environment guidance for forestry in England.