Classifying Ditch and Stream Channels Mapped From High-Resolution Digital Elevation Models Using Machine Learning
https://doi.org/10.5878/r0x8-kx56
This data contains the digital elevation models with 0.5 m resolution and polyline shapefiles with the location of channels from the 12 study areas used in this study.
It also has the scripts to generate the datasets used to train the machine learning model to classify channels into ditches and streams, and calculate the hydrological indices. The code to train the model is also included, along with the models obtained.
For the ground truth data, the channels were mapped differently based on their type: ditches were manually digitized based on the visual analysis of some topographic indices and orthophotos obtained from the DEM. Streams were mapped by initially detecting all natural channel heads, then tracing the downstream channels, and finally manually editing them based on orthophotos.
We recommend using Docker to set the environment.
Download 10 files with open access (137.89 MiB)
Files with open access (10)
Files with open access (10)
Citation and access
Citation and access
Access level:
Creator/Principal investigator(s):
Research principal:
Principal's reference number:
- SLU.seksko.2024.4.4.IÄ-1
Data contains personal data:
Yes
Type of personal data:
Names of user accounts indicating who performed certain steps of the data processing.
Citation:
Language:
Method and outcome
Method and outcome
Data format/data structure:
Data collection - Datorbaserad observation
Data collection - Datorbaserad observation
Mode of collection:
Datorbaserad observation
Description of the mode of collection:
Experter från Skogsstyrelsen digitaliserade manuellt dikena inom de 12 studieområdena spridda över Sverige baserat på hillshade- och högpassmedianfiltret som erhållits från DEM. Dikena digitaliserades manuellt med historiska foton och aktuella ortofoton (upplösning från 0,17–0,5 m).
Vattendragen kartlades genom att initialt detektera alla naturliga kanaler, sedan spåra de nedströms kanalerna och slutligen manuellt redigera dem baserat på ortofoton.
Source of the data:
- Geografiskt område
Spatial resolution:
0.5 meter scale
Instrument
Instrument
Name:
GIS-programvara
Geographic coverage
Geographic coverage
Geographic description:
The data covers 12 study areas spread across Sweden, containing information related to channel type for small water channels. More information with the precise locations can be found at the README.pdf file.
Administrative information
Administrative information
Responsible department/unit:
Department of Forest Ecology and Management
Contributor(s):
Funding
Funding
Funding agency:
- Marianne och Marcus Wallenbergs stiftelse
Öppnar nytt fönster hos ror.org.
ROR
Award title:
WASP-HS
Funding information:
Topic and keywords
Topic and keywords
Swedish Standard Classification of Research Subjects 2025:
INSPIRE topic categories:
Relations
Relations
Publications
Publications
Citation:
Dos Santos Toledo Busarello, M. (2025). Mapping small water channels using machine learning. In Acta Universitatis Agriculturae Sueciae. Swedish University of Agricultural Sciences. https://doi.org/10.54612/a.1vuvm11qn6
ISBN:
Citation:
Busarello, M. D. S. T., Ågren, A., Westphal, F., & Lidberg, W. (2025). Automatic detection of ditches and natural streams from digital elevation models using deep learning. In Computers & Geosciences (No. 105875; Vol. 196). https://doi.org/10.1016/j.cageo.2025.105875
SwePub:
Metadata
Metadata
Version 1
