Semi-automated mapping of aeolian sand dunes: A case study from northern Finland

Aeolian dunes are formed by the accumulation of wind-blown sand and its movement over time. While their shape may reflect wind directions during the period of formation, stratigraphic features in aeolian dunes record past climate and environments. In northern Fennoscandia, abundant dune fields serve as detailed records of Holocene arctic climate change. Specifically, buried soil and charcoal layers observed in mostly parabolic dunes preserve a rich history of past changes in climate, environment, fire history and land use. Understanding the evolution of the dunes in the face of these changes is crucial to project how Arctic environments respond to future climate change. Such studies require knowledge on the location, size, and shape of the aeolian sand dunes, which involves mapping them as polygon features. However, manual mapping of these landforms is very time-consuming. Semi-automatization can be used to develop a transferable, reproducible and scalable mapping approach, overcoming this issue.

Here we utilize a semi-automatic method involving machine learning models to map aeolian sand dunes in two study areas in northern Finland. While we rely on manual mapping based on a 2 m digital elevation model (DEM), its hillshade and Google satellite imagery for establishing a training dataset, segmentation and classification of sand dune objects is carried out using DEM derivatives such as slope, convergence index, general curvature, and topographic position index through support vector machines and random forest algorithms. We validate the training dataset and mapping results during field campaigns. Our semi-automatic object-based mapping method enables a mostly correct identification of dune objects, including attribute information on their size, shape, and morphological characteristics, compared to our field investigation. False positives occur at locations of great similarity between parabolic dunes and other topographic features. Future steps include reducing the false positives and transferring the approach to additional areas to ultimately develop a method for automated, regional-scale mapping of aeolian sand dunes in Arctic Fennoscandia.