Scaling and multivariate analysis of medium to large landslide events: Haida Gwaii, British Columbia

Gimbarzevsky (<CitationRef CitationID="CR12">1988</CitationRef>) collected an exceptional landsliding inventory for Haida Gwaii, British Columbia that included over 8,000 landsliding vectors covering an area of approximately 10,000 km<Superscript>2</Superscript>. This database was never published in the referred literature, despite its regional significance. It was collected prior to widespread application of GIS technologies in landsliding studies, limiting the analyses undertaken at the time. Gimbarzevsky identified landslides using 1:50,000 aerial photographs, and transferred the information to NTS map sheets. In our study, we digitized the landslide vectors from these original map sheets and connected each landslide to a digital elevation model. Lengths of landslide vectors are compared to the landsliding inventory for Haida Gwaii analyzed in Rood (<CitationRef CitationID="CR32">1984</CitationRef>), Martin Y et al. BC Can J Earth Sci 39:289–305 (<CitationRef CitationID="CR28">2002</CitationRef>); the latter inventory is based on larger-scale aerial photographs (~1:12,000). Rood’s database contains a more complete record of smaller landslides, while the inventory of Gimbarzevsky provides improved statistical representation of less frequent, medium to large landslides. It is suggested that combined landslide delineation at different scales could provide a more complete landslide record. Discriminant analysis was undertaken to assess which of nine predictor variables, chosen on the basis of mechanical theory, best predict failed versus unfailed locations. Seven of the nine variables were found to be statistically significant in discriminating amongst failed and unfailed locations. Results show that 81.7% of original grouped cases were correctly classified. Copyright Springer Science+Business Media B.V. 2012