| Attribute name | Label | Description |
|---|---|---|
| pubdate | ||
| label | ||
| score | ||
| county | ||
| map | ||
| id | ||
| geom |
AI Predicted Historic Woodlands by Woodland Type, Wales, (OS County Series – 1843-1893)
Welsh Government
This is an experimental dataset comprising of areas of different types of woodland, as depicted on the Ordnance Survey (OS) 1:25 000 Historic Landmark Map (Epoch 1).
This dataset is created in a similar way to AI Predicted Historic Woodlands, Wales (OS County Series – 1843-1893), however in this version, each predicted polygon in the dataset is labelled as one of the following types of woodland:
- Broadleaf
- Conifer
- Mixed Wood
- Bush
- Brushwood
- Orchard
- Osier
The labels have been determined using the original key from the Historic Landmark Map, which is made available through the National Library of Scotland.
The estimates are produced using an AI computer vision model called mask-rcnn. This model is trained on a small sample of manually labelled data which contains different areas of woodland as described above. The model learns the key features associated with each type of woodland, such as the distribution and pattern of pixels. The model is then applied to predict areas of woodland by identifying the same features on unseen data across the Wales.
As this technique involves predictive modelling, there are uncertainties with the dataset. For example, the model might incorrectly predict multiple types of woodland for a single area. This can happen in large, wooded areas with several different tree symbols marked near each other on the map.
To mitigate against this, we have applied a series of post-prediction processing steps to select a single woodland label for each predicted area. However, even with these validation steps, there is still a possibility of the misidentification of woodland areas.
For example, the model might not predict areas of woodland that exist, or the processing might result in different woodland labels assigned to polygons next to each other with the same tree symbols on the map.
There is also not complete coverage of this map across Wales. No estimates are made on areas which are missing from the map. Due to these reasons, the dataset should not be used to calculate the area, size or coverage of woodland.
Instead, we recommend this dataset should be used to indicate the likelihood of certain types of woodland existing in different areas in Wales during this period. As well as the boundaries of the wooded areas, the model also provides a probability score which represents the model’s confidence in that area containing that type of woodland.
This dataset is created using a different experimental methodology to other historic woodland outputs. For example, the Ancient Woodland Inventory uses other sources to identify areas of woodland. For this reason, the AI Woodland detections should be used independently to other data sources due to the risk in comparing areas of woodland from different sources that are not evaluated against the same criteria.
If you have any feedback on this experimental dataset and how it might be improved, please contact datascienceunit@gov.wales.
Use this data in another application
Please refer to the licence to understand the conditions of using this data.
Maps using this data (0)
Groups using this data (0)
Attributes (7)
Or display data on an existing map
- Type:
- Spatial data
- Publication date:
- 11 August 2025
- Licence:
- Open Government Licence for Public Sector Information (OGL)
-
Copyright:
- Point of contact:
- data@gov.wales
- Language
- English