EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Duneminds – Towards an Accessible Deep Learning Framework to Map Linear Dune Crests

Maike Nowatzki, David S. G. Thomas, Richard M. Bailey

Earth Surface Processes and Landforms · 2026-11 · 51 (11)

DOI: 10.1002/esp.70450 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Earth Surface Processes and Landforms, online erschienen am 15.11.2026. Verfasst von Maike Nowatzki und 2 weiteren Autor:innen. Die Arbeit nennt 86 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Clarendon Fund, Hertford College, University of Oxford, Royal Geographical Society.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Dune maps are fundamental for the research on dune systems and a long‐standing objective in aeolian geomorphology. However, they are often generated manually and thus lack reproducibility and feasibility for extensive study areas. This study introduces Duneminds , a deep learning framework built around a U‐Net convolutional neural network for the automated detection and mapping of dune crests. The model was applied to the south‐western Kalahari linear dune field and tested with a range of input configurations, including different data sets, levels of data augmentation and resolutions. Data sets included Sentinel‐2 optical imagery, the Copernicus Digital Elevation Model and derivatives of the latter. Among these, models trained on optical imagery outperformed those based on elevation data, with the best‐performing one achieving an IoU score of 0.77. We applied the framework to generate a continuous crest map of the entire south‐western Kalahari dune field, reproducing dune crest pathways and patterns of the study area.

Förderung

  • Clarendon Fund
  • Hertford College, University of Oxford
  • Royal Geographical Society

Zitieren

APA

Nowatzki, M., Thomas, D. S. G., & Bailey, R. M. (2026). Duneminds – Towards an Accessible Deep Learning Framework to Map Linear Dune Crests. Earth Surface Processes and Landforms, 51(11). https://doi.org/10.1002/esp.70450

Vancouver

Nowatzki M, Thomas DSG, Bailey RM. Duneminds – Towards an Accessible Deep Learning Framework to Map Linear Dune Crests. Earth Surface Processes and Landforms. 2026;51(11). doi:10.1002/esp.70450

BibTeX
@article{Nowatzki2026,
  author = {Nowatzki, Maike and Thomas, David S. G. and Bailey, Richard M.},
  title = {Duneminds – Towards an Accessible Deep Learning Framework to Map Linear Dune Crests},
  journal = {Earth Surface Processes and Landforms},
  year = {2026},
  volume = {51},
  number = {11},
  doi = {10.1002/esp.70450}
}
Kostenlos starten →