Information on grain sizes in gravel-bed rivers by automated image analysis

TLDR: In this paper, modified edge-detection algorithms combine edge seeding with an image porosity concept and partial watershed segmentation to reduce the time required in the field to characterize textural variation over gravel surfaces.

Abstract: The time required in the field to characterize textural variation over gravel surfaces can be reduced by taking vertical photographs for subsequent image analysis. We present modified edge-detection algorithms which combine edge seeding with an image porosity concept and partial watershed segmentation. The methods allow quick, reliable, and operator-independent size analysis from a wide range of vertical bed-surface images. They are tested using 24 naturally lit images of an exposed river bed with mixed lithologies and partial burial of gravel by sand. Grain-size percentiles derived by automated image analysis correlate closely with those from manual image analysis, with only small and consistent degrees of bias. They also correlate well with percentiles from field measurements with substantial bias, which, however, is consistent so that it can be corrected for, leaving a residual scatter of 0.25 (where = log2 mm = -) over a wide range of bed conditions. The bias depends somewhat on sand cover, and the biggest residual discrepancies are for tail percentiles.