Dataset Name: PeatNet - Machine learning based segmentation of peatland drainage channels and erosion features in Scotland
Description:

This dataset contains various GIS layers of modelled peatland degradation features derived from 25 cm resolution Aerial Photography for Great Britain (APGB) data. 
Summary rasters of both drainage and erosion features at multiple spatial resolutions used in derived mapping, and polygon features at the site-scale are provided.
Further details on the processing and creation of these data layers can be found in the following publication:

    Macfarlane, F., Robb, C., Coull, M., McKeen, M., Wardell-Johnson, D., Miller, 
    D., Parker, T. C., Artz, R. R. E., Matthews, K., & Aitkenhead, M. J. (2024). 
    A deep learning approach for high-resolution mapping of Scottish peatland degradation. 
    European Journal of Soil Science, 75(4), e13538. https://doi.org/10.1111/ejss.13538

###

1. Usage Notes
This dataset is provided "as is," without any guarantees or warranties of any kind, either expressed or implied. Users should be aware of the following:

	a) Data Accuracy and Completeness:
		While this dataset has been prepared with care, it may contain errors, omissions, comissions, or outdated information. The dataset may also exhibit inherent limitations based on the quality and resolution of the input data. Users are advised to validate the data and ML/AI model outputs before relying on them for critical applications.

	b) Imaging Date and Temporal Relevance:
		The aerial imagery used to derive this dataset was captured on specific dates. Users should verify the imaging date(s) from the metadata to assess the temporal relevance of the data. This is particularly important for time-sensitive or dynamic environmental analyses such as restoration activities.

	c) Cloud and Snow Cover:
		Imagery in the dataset may be impacted by cloud cover or snow cover, which can obscure surface features. The presence of such features is not captured in the input imagery and as a result cannot be translated to this output. 
	
	d) Inherent Error and Bias in ML/AI Models:
		This dataset has been produced using a computer vision machine learning model which may introduce inherent error and bias. These biases can arise from various factors such as:

			Training Data: While input data was selected from a range of spatial and temporal locations, biases may propagate in the model's outputs.
			
			Model Limitations: No model is perfect, and users should expect a degree of uncertainty in predictions, classifications, or other automated outputs. Accuracy may vary depending on the complexity of the geographic area or specific environmental conditions (e.g., cloud or snow cover, vegetation etc.).
			
			Generalisation: Models may struggle to generalise across different geographies, seasons, or environmental conditions if they have not been adequately trained on diverse datasets.
		
		Users should carefully evaluate ML/AI outputs and not rely solely on automated results without further validation and verification, especially for critical decision-making. This dataset is not intended to replace desk and field-based surveys.

	e) No Warranty or Liability:
		The provider of this dataset makes no warranty, expressed or implied, as to the accuracy, completeness, or reliability of the data. The dataset is provided without any form of guarantee, and no claims are made regarding its suitability for any specific purpose. By using this dataset, the user agrees that the provider bears no liability for any consequences resulting from its use.

	f) Responsibility for Use:
		Users assume all responsibility for any conclusions drawn or actions taken based on the data. It is the user's responsibility to evaluate the data's suitability for their intended purpose and to conduct any additional validation, cleaning, or transformation that may be required.

	g) Updates and Maintenance:
		This dataset may not be regularly updated or maintained. Users should be aware that the data could become outdated and may no longer reflect the current state of the subject matter.

	h) Attribution:
		Where applicable, users should attribute the dataset according to the specified licensing or attribution requirements. This includes proper citation or acknowledgment in any published materials or derivative works.

	i) License:
		Users should review and comply with the terms of the dataset's license, including any restrictions on redistribution, modification, or commercial use.

	j) Get in Touch to Help Improve the Dataset:
		We welcome contributions to improve the quality and utility of this dataset. If you have suggestions, corrections, or additional data that can enhance the dataset’s accuracy or functionality, we encourage you to get in touch. Whether it's providing feedback on specific areas, contributing additional data, or suggesting improvements to the associated ML/AI models, your input can help refine and expand the dataset for the benefit of all users.
		
###

2. Data Summary
Data Format: Shapefiles (.shp), GeoTIFFs (.tif).
Coordinate Reference System (CRS): EPSG:27700 (OSGB36 / British National Grid)

###

3. Layer Descriptions
peatnet_drainage_density_XXm_2023.zip - zipfile containing a GeoTIFF (.tif) at XX metre resolution describing the modelled unit area density of drained peat as a result of artificial drainage channels. This is derived using the 30 metre buffer zone surrounding a drainage channel considered "drained" peat using the UKGHGI definition.

peatnet_drainage_drain_centrelines_2023.zip - zipfile containing a Shapefile of polyline features of the modelled drainage channel centrelines. This has been simplified to a tolerance of 1 metre using the Douglas-Peucker algorithm.

peatnet_drainage_drain_segmentation_2023.zip - zipfile containing a Shapefile of polygon features of the modelled drainage channel segmentations. This is the polygonised output from the 25cm image segmentation model, simplified to a tolerance of 1 metre using the Douglas-Peucker algorithm. 

peatnet_drainage_drained_area_2023.zip - zipfile containing a Shapefile of polygon features of the modelled drained area of peat.

peatnet_erosion_2023.zip - zipfile containing a Shapefile of polygon features corresponding to the modelled erosion features. This is the polygonised output from the 25cm image segmentation model, simplified to a tolerance of 1 metre using the Douglas-Peucker algorithm. Additionally, features with an area of less than five square metres have been removed.

peatnet_erosion_density_XXm_2023.zip - zipfile containing a GeoTIFF (.tif) at XX metre resolution describing the modelled unit area density of eroded and bare peat.

###

4. Data Sources
Aerial Photography for Great Britain (APGB)
URL: https://www.blueskymapshop.com/
Description: 25 cm resolution aerial photograghy

###

5. Citation and Licensing
Citation: If you use this dataset, please cite it as follows:

    Macfarlane, F., Robb, C., Coull, M., McKeen, M., Wardell-Johnson, D., Miller, 
    D., Parker, T. C., Artz, R. R. E., Matthews, K., & Aitkenhead, M. J. (2024). 
    A deep learning approach for high-resolution mapping of Scottish peatland degradation. 
    European Journal of Soil Science, 75(4), e13538. https://doi.org/10.1111/ejss.13538

License: Use is permitted under the terms of the James Hutton Institute Open Data Licence.

###

6. Contact Information
Author(s): Fraser Macfarlane
Institution: The James Hutton Institute
Email: fraser.macfarlane@hutton.ac.uk
Website: https://www.hutton.ac.uk/people/fraser-macfarlane/