Predicting shallow landslide size and location across a natural landscape: Application of a spectral clustering search algorithm

Abstract

Use policyThe full-text may be used and/or reproduced, and given to third parties in any format or medium, without prior permission or charge, for personal research or study, educational, or not-for-prot purposes provided that:• a full bibliographic reference is made to the original source • a link is made to the metadata record in DRO • the full-text is not changed in any way The full-text must not be sold in any format or medium without the formal permission of the copyright holders.Please consult the full DRO policy for further details. Abstract Predicting shallow landslide size and location across landscapes is important for understanding landscape form and evolution and for hazard identification. We test a recently developed model that couples a search algorithm with 3-D slope stability analysis that predicts these two key attributes in an intensively studied landscape with a 10 year landslide inventory. We use process-based submodels to estimate soil depth, root strength, and pore pressure for a sequence of landslide-triggering rainstorms. We parameterize submodels with field measurements independently of the slope stability model, without calibrating predictions to observations. The model generally reproduces observed landslide size and location distributions, overlaps 65% of observed landslides, and of these predicts size to within factors of 2 and 1.5 in 55% and 28% of cases, respectively. Five percent of the landscape is predicted unstable, compared to 2% recorded landslide area. Missed landslides are not due to the search algorithm but to the formulation and parameterization of the slope stability model and inaccuracy of observed landslide maps. Our model does not improve location prediction relative to infinite-slope methods but predicts landslide size, improves process representation, and reduces reliance on effective parameters. Increasing rainfall intensity or root cohesion generally increases landslide size and shifts locations down hollow axes, while increasing cohesion restricts unstable locations to areas with deepest soils. Our findings suggest that shallow landslide abundance, location, and size are ultimately controlled by covarying topographic, material, and hydrologic properties. Estimating the spatiotemporal patterns of root strength, pore pressure, and soil depth across a landscape may be the greatest remaining challenge.

Tree-root control of shallow landslides

“…In the large majority of cases, slope stability models add apparent cohesion to the soil to simulate root reinforcement (e.g., Milledge et al, 2014;Bellugi et al, 2015;Hwang et al, 2015). Few models include the effects of root distribution heterogeneity (Stokes et al, 2014), and none consider the stress-strain behavior of root reinforcement and the strength of roots in compression.…”

Hydro‐geomorphic behavior of contrasting tropical landscapes and critical zone response to changing climate

“…Also, even though a more detailed quantitative comparison between observed and simulated landslide scars could be informative (i.e. in addition to the landslide area comparison of Table ) (Bellugi et al, ), limitations in the landslide data set, the estimation observed landslide area at each landslide location highly uncertain (Larsen, ).…”

“…The magnitude of larger landslides was computed as the total area of unstable cells forming each cluster. This approach is simpler than the spectral clustering search algorithm of Bellugi et al (2015a), which is based on a landscape-scale hydrologic-stability model (Milledge et al, 2014;Bellugi et al, 2015b). The total simulated landslide area is greater for the B1 scenario in both watersheds (Table II).…”

A hydroclimatological approach to predicting regional landslide probability using Landlab

“…Despite its fine grid resolution, the SSURGO database (DOA-NRCS, 2016) only broadly captures topographic controls on soil depth and reflects existing conditions in the field based on soil surveys. In an attempt to improve the representation of spatial granularity and local uncertainties of soil depth, a soil evolution model is used (Dietrich et al, 1995;Simoni et al, 2008;Pelletier and Rasmussen, 2009;Tesfa et al, 2009;Bellugi et al, 2015). The model is run to develop time series of soil depth from which triangular distribution parameters for soil depth (mode, minimum and maximum) can be obtained and used in the Landlab LandslideProbability component.…”

“…We statistically evaluated our model using receiver operating characteristics (ROC) (Fawcett, 2006) and success rate (SR) curves (Bellugi et al, 2015).…”