Desert dunes can be distinguished by a wide variety of morphologies. They may, for example, show different shapes, crest heights, wavelengths, or orientations creating morphological patterns between and even within dunefields. Patterns can be explained by differences in wind strength and orientation, sediment supply, or vegetation.
This project will use machine learning approaches to explore the relationship between morphological patterns and their potential drivers in linear dunefields.
Linear dunes of different morphology in the Kalahari (left) and the Namib (right).
Images provided by Copernicus, Sentinel-2
Parts of the Kalahari and Namib deserts in southern Africa were chosen as study area since they are home to extensive (linear) dunefields showing different morphological patterns and activity states.
Map: The red polygons show the intended study areas - linear dunefields of the Namib (A) and Kalahari (B) desert.
In order to meet the main aim of the project, the exploration of the relationship between dunefield morphology and drivers, two objectives are pursued.
(1) The first objective is the quantification and description of the dunefield morphology. The dunes in the study area will be mapped using an automated approach based on Sentinel-2 satellite imagery and a convolutional neural network (CNN) before morphometrics such as crest height, orientation, or wavelength are calculated.
(2) Furthermore, a model that scores the activity or responsiveness of dunes will be used to determine sub-areas that are currently most active and thus likely responsive to current environmental influences driving dune morphology.
In these sub-areas the relationship between potential morphological drivers (i.e. climate / environmental parameters) and the calculated morphometrics will be explored by applying a black box approach based on another CNN.