MORPHOMETRIC ANALYSIS OF GLACIFLUVIAL EROSION FORMS IN BEDROCK, KANGERLUSSUAQ, GREENLAND

The planimetric shape and spatial distribution of >8500 bedforms eroded by glacial meltwater in a gneissic bedrock sill in Watson River at Kangerlusuaaq, West Greenland, are analyzed to test whether computationally based shape recognition techniques distinguish forms that occur at hydrodynamically distinct locations. The intent of these analyses is to develop an approach suitable for assessing whether small-scale meltwater forms, like those at Kangerlussuaq, differ from larger sculpted forms argued by some as evidence for subglacial erosion by large meltwater outburst floods. We build sub-centimeter resolution digital image maps of forms at upflow (220 m²), crest (350 m²), and downflow (500 m²) locations along the sill by assembling low-level (4 m) pole photography and rectifying these image maps with 2 m grid total station survey data. Form boundaries are traced, scanned and analyzed with shape analysis software to derive form “feature” data (length, width, perimeter, area, and other derived measures). Measures of form separation derived from covariation matrixes of feature data identify which variables may best discriminate hydrodynamically distinct forms.

Initial analyses of individual or “context-free” forms indicate substantial overlap of feature data, but that that forms at crest locations on the sill separate from those at upflow and downflow sites for some form measures. These groupings are likely explained by the prevalence of circular to elliptical pothole and scallop forms at upflow and downflow sites in contrast to en echelon transverse forms at crest sites. Dimensionless feature data related to circularity (Perimeter²Area^-1) and asymmetry produce the best form separation results, whereas area, perimeter and various scale-dependent measures are poor form separators. Large synthetic populations generated from optimal feature data are used to develop receiver-operator characteristic (ROC) curves which provide a means for evaluating the confidence with which new forms may be assigned to a group. These ROC curves together with ongoing context-sensitive analyses of form feature data provide a means for determining whether these small-scale forms differ fundamentally from other more controversial glacial erosion forms.