2014 GSA Annual Meeting in Vancouver, British Columbia (19–22 October 2014)

Paper No. 211-2
Presentation Time: 9:15 AM


BROSCOE, David, Algonquin College, Ottawa, ON K2G1V8, Canada and RUSSELL, Hazen A.J., Geological Survey of Canada, 601 Booth Street, Ottawa, ON K1A 0E8, Canada

Eskers have commonly been mapped and symbolized manually from aerial photographic interpretation as either lines (ridges) and polygons (sand and gravel). To-date no method has been deployed that could automatically extract esker extents and quantify the esker volume. A methodology is presented for the quantification of eskers that uses Canadian Digital Elevation Data (CDED), spectral remotely sensed imagery (e.g. LandSat, Spot), and legacy esker line work from Geological Survey of Canada publications. Using ArcGIS and an esker detection module (EDM) coded in Python, the CDED data are smoothed using user defined filter windows. A difference surface is produced that emphasizes ridge areas and is used to create polygons. The legacy esker line work is used as a training dataset to extract ridge areas within a user defined buffer. EDM results have been tested against the input training data and a local data set generated manually from aerial photographic interpretation. Depending upon terrain characteristics the success of the data extraction ranges from 65 to 81 % for the esker line work and 35 to 72 % for the more limited aerial photographic interpretation. The variable success reflects esker size related to both relief and width in the CDED data. Ongoing development of this methodology focuses on enhanced delineation of low-relief areas of the esker not captured by the DEM analysis through incorporation of spectral imagery. A multiclass (80-90) iso-cluster unsupervised classification of SPOT MSS data was completed to characterize the landscape. The isocluster classification is then overlain on the esker polygons. The most dominant classes in terms of area are identified and the user can specify the number of classes to be chosen. The originally topographically defined polygons are then merged with the selected intersecting spectral classification.