A NEW METHOD FOR THE SEMI-AUTOMATED MAPPING OF DRUMLINS AND MEGA-SCALE GLACIAL LINEATIONS

Relict drumlins and mega-scale glacial lineations (positive relief, Longitudinal subglacial Bedforms – LBfs) have been used in reconstructions of paleo ice-sheet configuration and dynamics though their genesis remains equivocal. Morphometric data is needed to test genetic hypotheses, but the present data set is mainly derived from manually-mapped footprints that are geographically restricted, and morphometric variables do not capture the full range of morphologic variation. More complete inventories must first overcome the subjectivity of manual mapping (MM). (Semi-)automated mapping (SAM) techniques are designed for objectivity and speed but published SAM methods for LBF mapping have not been highly successful. We propose a new method, Normalized Closed-Contour Method (NCCM), and test it in two terrains of different complexity (A is most complex, B is least complex). NCCM assumes that each LBf footprint can be represented by a lowermost, normalized local relief (NLR) closed contour. In a top-down approach, footprint objects are identified by querying a multitude of candidate objects (every NLR closed contour) against a morphometric ruleset encapsulating the morphometric range of LBfs in a study area. Rules are based on a manually-mapped sample. NLR is modeled from a digital elevation model and is used instead of elevation a.s.l. because the latter is not suitable for defining LBf footprints on sloped terrain. NCCM was applied with 4 different NLR models. Performance was assessed in relation to manually-mapped footprints using both cell-based and object-oriented statistics. A hydrology-based model was the best performing NLR model (detection rate= 92%; Kappa=0.49) in test area A. A moving window-based NLR model computed for each cell as z-z_min/z_max-z_min was the best performer (detection rate=86%; Kappa=0.51) in test area B. In both areas, NCCM-footprint datasets match reference footprints well in terms of the mean and standard deviation of both footprint orientation and footprint elongation. The similarity of NCCM performance in the two test areas argues for reproducibility. Additionally, NCCM is easy to implement and scalable. Future work should focus particularly on improving accuracy in terms of footprint size and on reducing overdetection in complex terrains.