AUTOMATED EXTRACTION OF BARCHAN POSITIONS AND MOVEMENT VECTORS

Geomorphic studies of Aeolian features have traditionally focused on single or small collections of examples over limited regions due to the large scale of objects, their extensive spatial distribution, and difficulty in visiting and monitoring multiple research sites. Because of these constraints, comprehensive analysis of a statistically robust number of features across environments is not currently possible and comparison between features, capture of their time dependent characteristics and analysis of forcing mechanisms at different scales remains difficult.

Barchan morphometry is difficult and time consuming to delineate both in the field and from imagery. This has led to studies characterized by small sample sizes resulting in vastly different, and statistically unreliable, estimates of dune morphological parameters giving rise to a limited ability to compare between studies. This reduces our understanding of how these features move and evolve over time.

In this research we use a semi-automated approach to extract features and movement vectors along 40 barchan paths in the Laayoune dune field of Western Sahara. Barchan polygons were extracted from four annual images (2010-2013) using edge detection, clutter reduction, and object extraction algorithms. A predictor equation based on operator estimation of offset between a small number of barchan pairs was used to estimate positions between time slices and object matching algorithms were used to identify the "most likely" pair candidates between images and to predict where all objects would be located in subsequent images. For each object identified the trailing toe point and centroid were captured along the average downwind movement vector. An image of the barchan was also captured in each time slice.

Analysis indicates annual mean movement ranging from 72-103 m/yr (mean ~88 m/yr or ~0.24 m/day). Mean directional vectors showed only minor differences between 2010 and 2013 with an average direction of movement of 183.4 degrees, consistent with the prevailing wind direction measured at Laayoune. Using our approach, we repeated the analysis on a portion of the dune field used in earlier field-based barchan movement studies and confirmed that that our semi-automatic derived movement vectors were nearly identical to those published earlier.