CGRG Bibliography of Canadian Geomorphology
Search Results
Author : Brown, O.H.; Utting, D.J.; Little, E.C.; Grunsky, E.C.; Harris, J.; and Peter, P.
Date : 2005.
Title : Remote predictive mapping of surficial geology in Nunavut using supervised classification techniques of Landsat and RADARSAT I data.
Publication : Joint Meeting of the Geological Association of Canada, the Mineralogical Association of Canada, the Canadian Society of Petroleum Geologists and the Canadian Society of Soil Sciences. May 15-18, 2005. Studley Campus of Dalhousie University, Halifax, Nova Scotia.
Issue :
Page(s) :
Abstract
To assist the planning and execution of fieldwork in 2005, predictive surficial geology maps for NTS 37E and 37H (North Baffin Island) were created using supervised classification techniques on Landsat and RADARSAT I data. If successful, the techniques applied to these areas may also be useful in other contiguous regions of northern Baffin Island where similar terrain and geology exists. In this study, three supervised classification techniques have been tested to determine which method produces the best predictive map: minimum-distance-to-mean, maximum likelihood and neural networks. The minimum-distance-to-mean classifier is the simplest algorithm which assigns a pixel to the class with the nearest mean value. The maximum likelihood classifier uses calculated probability to assign pixels to a class. The neural networks classifier identifies patterns in the data using statistics derived from the training sites. Supervised classifications were first conducted in NTS 37G, an area previously mapped by traditional means (i.e., air photos and ground truthing). The area was subdivided in two regions based on bedrock terranes: Bylot supergroup and other crystalline rock. The two regions were classified independently. This division was required because the sedimentary rock of the Bylot supergroup erodes to sandier material than surrounding rock. Within these regions, till, colluvium, alluvium and water were discernable. An “Other” class represents bedrock, glaciolacutrine, glaciofluvial and marine sediments — these materials were not easily differentiated because of their similar spectral signatures and limited number of training sites. The classifications were then compared to the surficial geology map derived through classical mapping techniques (e.g., aerial photograph analyses). The general distribution patterns of materials on both maps are similar. Differences between the maps occur because spectral characteristics of material classes may overlap due to similarities in material properties. Also, supervised classification techniques classify “pixel by pixel”, whereas traditional mapping groups areas of relatively homogeneous materials into polygons. The spectral signatures extracted from the known area were applied to the unmapped areas (NTS 37E, H) to produce three classified predictive maps. These maps will be used to assist with the planning of drift prospecting sampling, and identify potential sites requiring additional ground truthing.
Bibliography of Canadian Geomorphology