Statistical Signature of Deep-seated Landslides

Thursday, 18 December 2014
Chandana Gangodagamage, Los Alamos National Laboratory, Los Alamos, NM, United States, Efi Foufoula-Georgiou, Univ Minnesota, Minneapolis, MN, United States, Patrick Belmont, Utah State University, Department of Watershed Sciences, Logan, UT, United States, Benjamin H Mackey, University of Canterbury, Christchurch, New Zealand and Theodore K Fuller, Simon Fraser University, Burnaby, BC, Canada
We investigate the statistical signature of deep-seated landslides using basin wide topographic data and flowpath arrangement and explore the extent to which these globally derived signatures can be used to locally map landslides. We used directed distance from the divide, which accounts for the distance traveled along flowpaths starting from significant ridgelines, as a scale parameter and demonstrate that local slope vs. directed distance and curvature vs. local slope offer powerful means for identifying the presence of landslides in a landscape. By exploring a threshold on the probability distribution of local slopes conditional on directed distance we show that mapping of landslide features is possible. We apply the methodology to three 0.5 to 2.5 km2 watersheds in northern California and document three regions of distinct geomorphic signatures [Gangodagamage et al., 2011, http://dx.doi.org/10.1029/2010WR009252]. In region A, hillslope gradient increases with distance from the divide and flowpaths are divergent or parallel. Region B corresponds to the zone with highly convergent flowpaths and exhibits the strongest signal of landslide related features. Region C is a moderately convergent zone that transitions into the fluvial channel network. Next, we use specific quantiles of the probability density function of local slopes conditioned on directed distance from the divide to map individual landslide features. This analysis allows us to explore the 3D morphometry of the landslide affected basins and to develop a supervised set of ensemble templates for landslides as a function of local slope vs. directed distance (DD) relationship. Then we use this template and demonstrate that the landslide affected basins can be identified by iterative matching the landslide signature template with the basin wide signatures of the tributary basins in the South Fork Eel River, CA. Finally, we perform a multiscale analysis of the contributing area parameterized by directed distance from the divide and demonstrate that the landslide-disturbed landscapes show reduced complexity and degree of multifractality, that is, a reduction in the degree of spatial heterogeneity in the distribution of extreme contributing areas for a given distance from the divide.