Deep Learning: A new tool for mapping and analysis of dune fields (e.g. Rub'Al Khali sand sea) - Archive ouverte HAL Accéder directement au contenu
Poster De Conférence Année : 2023

Deep Learning: A new tool for mapping and analysis of dune fields (e.g. Rub'Al Khali sand sea)

Résumé

The surface of Earth and Mars present abundant periodic topographic forms at different scales (mm- km) and in different environments (e.g. Aeolian, Subglacial) called bedforms (e.g. dunes, drumlins). These bedforms develop at the interface between the moving fluid and a deformable and/or erodible material. Generally, these bedforms are grouped in field which presents a great scientific interest because their morphological characteristics (e.g. shape, size, orientation, complexity, organization) depend on the physical characteristics of i) the moving fluid (e.g. velocity, direction of movement, viscosity), ii) the available sedimentary stock (e.g. density, shape, grain size). Thus, local to regional mapping and producing a morphometric analysis of these bedforms can provide elements to understand their processes of formation and to reconstruct the (paleo)flow dynamics of the fluids involved. First, we developed an automated protocol in order to extract the characteristics of aeolian dunes (figure 1): Dune mapping and analysis in QGis, Python and eCognition are performed using a steps succession: i) Residual Relief extraction as a key feature in deep learning; ii) Deep learning approach (Convolutional Neural Network (CNN), used for the morphology segmentation process from DTM derivatives; iii) Volumetric Obscurance approach (DTM derivative) used for the crestlines extraction. Finally, we extract the morphometric indexes (sinuosity, elongation, circularity) of each dune. We applied this method to analyze the Quaternary aeolian dunes of Rub'Al Khali desert, in the Arabian Peninsula. We choose this analogue because of its wide surface (660,000 km²), ii) its dune diversity, and iii) the different observation scales within the available dataset (Copernicus DTM 30m: https://opentopography.org/). Our results show that these successive steps have allowed us to produce a detailed map of the dune morphologies of the Rub'Al Khali desert at two observation scales (kilometer and hectometer). In total, 78,000 dunes and crestlines were mapped in 6 hours of processing on 58,000 km² surface. The dune shape vectors were used to calculate and analyze the dune morphometry. Morphometric indices highlight three spatial domains that illustrate a spatial variability for circularity index (strong for star and dome dunes), sinuosity index (strong for mega-barchanoid ridges) and elongation index (strong for linear dunes) oriented to the median wind direction.
Fichier principal
Vignette du fichier
POSTER_MARID_Daynac_2023.pdf (29.61 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-04384098 , version 1 (10-01-2024)

Licence

Paternité

Identifiants

  • HAL Id : hal-04384098 , version 1

Citer

Jimmy Daynac, Paul Bessin, Stéphane Pochat, Régis Mourgues. Deep Learning: A new tool for mapping and analysis of dune fields (e.g. Rub'Al Khali sand sea). Marine and River Dune Dynamics – MARID VII, Apr 2023, Rennes, France. , 2023. ⟨hal-04384098⟩
21 Consultations
11 Téléchargements

Partager

Gmail Facebook X LinkedIn More