LIDAR-BASED CREVASSE MAPPING, HAIG GLACIER, CANADIAN ROCKIES

Crevasses appear as open rifts or closed traces, at scales of millimeters to meters in width. Despite their uniform appearance, crevasses have different depth, width, and shape, with nearly vertical walls, and usually undergo change, such as widening and deepening, over the course of a melt season or in association changing glacier flow (stress) regimes (Benn & Evans, 2014). The dramatic increase in the use of laser altimetry and other finely resolved geodetic data (e.g. UAV, submeter optical satellite imagery) provide an opportunity to assess the prevalence, extent and changes of crevasses for mountain glaciers. Crevasses can also bias geodetic measurements of elevation and mass change unless they are properly accounted for.

We developed a framework for automatic mapping and geometry extraction of crevasses on glaciers, and masking them for analyses that are interfered with by crevasse topography. We apply an Artificial Neural Network (ANN) algorithm known as Self Organizing Maps (SOM) (Kohonen, 2001), which is strong in outlining linear trending features. Application of SOM with a suitable setting in network and input layers can recognize and map the entire morphology of crevasses.

Our study uses airborne LiDAR DEM from Haig Glacier in the Canadian Rocky Mountains, which is experiencing a transient response to the crevassed regions of the glacier as the glacier thins (Adhikari & Marshall, 2013). Crevasses on Haig Glacier are well captured in the LiDAR data when they are not snow-bridged at the end of the summer melt season (Aug/Sept). High-resolution data in this study makes it possible to develop 1st, 2nd, and 3rd derivatives of local glacier surface parameters to extract morphometric elements of the crevasses.

The SOM method, in addition to longitudinal and cross-sectional curvature layers, successfully extracted and characterized about 1000 crevasses on Haig Glacier in Sept 2015, with an overall accuracy of 98.4%. In addition, applying unsphericity can add a new dimension to crevasse evolution and their rim change studies. The mapping and filtering algorithm to remove crevasses from the glacier surface DEM results in an elevation difference map with 6.5 m greater accuracy. The resulting map also indicates diverse crevasse patterns on Haig Glacier, which provide insight into stress and flow conditions.