Skip to main content
eScholarship
Open Access Publications from the University of California

UC Berkeley

UC Berkeley Electronic Theses and Dissertations bannerUC Berkeley

What Controls Shallow Landslide Size Across Landscapes?

Abstract

Abstract

What Controls Shallow Landslide Size Across Landscapes?

by Dino Bellugi

Doctor of Philosophy in Earth and Planetary Science

and the Designated Emphasis in Computational Science and Engineering

University of California, Berkeley

Professor William E. Dietrich, Chair

Shallow landslides that usually involve only the colluvial soil mantle, are a widespread phenomenon in the United States and the world. Often triggered by extreme precipitation events, they can be the primary sources of debris flows, and are generally a threatening source of hazards, causing loss of life, destruction of property, and affecting communities all across the nation. Shallow landslides also play an important role in landscape evolution, dominating erosion in steeper landscape, unleashing debris flows that carve valley networks, and delivering sediment to rivers. The two primary aspects affecting the impact of shallow landslides, both in terms of downstream hazard and their geomorphic significance, are their location and size.

Theoretical and observational research has provided some insight on the controls on the size of shallow landslides. It has been observed that landslide exhibit a smaller size in grasslands than in forested areas and that landslides were smaller in areas where root strength decreased as a result land use change. The parameters that are most relevant for the occurrence of rainfall-triggered shallow landslides are slope, pore pressure, root and soil strength, and soil depth. Theoretical analyses have suggested that a decline in root strength results in failures having lower minimum lengths and widths, while low gradients, low pore water pressures, or high soil friction result in failures having higher lengths and widths. However, few if any studies examine the controls on both landslide location and size across a landscape. I hypothesize that the co-organization of landscape properties, such as slope, soil depth, pore pressure, and root reinforcement, controls the size and location of shallow landslides.

We currently lack mechanistic models for specifically predicting shallow landslide size across landscapes, thus reducing the effectiveness of landslide hazard delineation, and inhibiting our ability to formulate and apply mechanistic models for landslide flux and surface erosion. One reason for this is the one-dimensional representation of slope stability, generally applied in existing regional scale applications. Such a representation cannot produce discrete landslides and thus cannot make predictions on landslide size. Furthermore, one-dimensional approaches cannot include lateral effects which are known to be important in defining instability. These limitations can be addressed by a three-dimensional slope stability model, but its application to a landscape is challenging. Whereas the one-dimensional slope stability at a location can be determined independently of its dimensions and surroundings, multi-dimensional analyses require the treatment of discrete shapes. As these shapes are not known a priori, a search algorithm is required. This is a non-trivial problem, whose naïve solution (i.e. an exhaustive search) is of exponential complexity, rendering the problem effectively intractable at any relevant scale. Any new procedure must be sufficiently general to evolve with current understanding, but with a parsimonious parameterization in order to be compatible with available data. The procedure must be computationally efficient to be applicable at scales large enough to be relevant for geomorphological and hazard related questions, yet at sufficiently fine resolution to capture the fundamental mechanics of slope failure.

In this dissertation I develop a procedure which couples a novel slope stability model that captures the basic physics of shallow landsliding, with a new and efficient search algorithm based on spectral graph theory that can predict discrete shallow landslides. In order to apply this procedure at the regional scale, I define sub-models to produce the required data, when they are not available at the necessary resolution. These sub-models extract topographic attributes, compute the spatial distributions of soils, and estimate the root reinforcement and pore pressure fields. I define formal framework to evaluate the performance of the procedure, based on information retrieval theory. This procedure should advance our understanding and prediction capability, enabling me test the hypothesis that the co-organization of landscape properties, such as slope, soil depth, pore pressure, and root reinforcement, controls the size and location of shallow landslides. As these properties are mostly dictated by topography, I hypothesize that topography exerts a first order control on both location and size.

In chapter two, I present a multi-dimensional stability model framework that can be applied to landscapes at the regional scale. This slope stability model is mechanistic but not so mechanistic that its application becomes impracticable. It is fully three-dimensional in the treatment of the forces acting on a discretized slope element and it is statically determinate. The model considers the effects of root cohesion and pore pressures, and includes the effects of earth pressure in a manner that is compatible with natural slopes. Finally, this model is easily applicable to spatially gridded data, and requires only a modest parameterization facilitated by procedures defined to obtain spatially explicit parameter fields at the required resolution. The slope stability model allows for the characterization of the forces acting on all the boundaries resulting from the discretization of a landscape into slope element blocks (and thus the role of each block in the stability of the landscape). However, it requires a deterministic search procedure that is able to select discrete least-stable combinations of slope elements across a landscape to obtain meaningful shallow landslide predictions.

In chapter three, I define a procedure that can for the first time predict discrete landslides. Its foundation is a search algorithm based on spectral graph theory that can efficiently provide a good approximation for an otherwise intractable problem. This procedure relies on a slope stability model, as well as sub-models and data for, among others, topography, soil depth, and root strength, discussed in the previous chapter. However, the procedure is general, and is not confined by their choice: as better models and data emerge, the procedure can be easily modified to take advantage of such improvements. A formal framework is defined to evaluate the performance of the procedure, based on information retrieval theory. Applying the procedure to a synthetic landscape illustrates how landslide size and location are affected by the hetereogeneity of parameters such as root strength and pore pressure.

In chapter four, the procedure is applied to an instrumented catchment in the Oregon Coast Range using field-measured physical parameters, successfully predicting the size and location of the shallow landslide which destroyed the site during a storm in November, 1996. The procedure was then applied to a larger study area using modeled physical parameters, under a suite of diverse hydrological scenarios. The application of the procedure results in, and is able to reproduce the distribution of sizes and locations observed during the ten years of research at the site. Performance is quantified using a set of information retrieval measures, performing significantly better than a random classifier, demonstrating the applicability of the procedure.

In chapter five, a sensitivity analysis is performed to explore the controls on shallow landslide size and location. Rainfall, vegetation, soil, and topographic characteristics are systematically varied, resulting in probability density functions of predicted landslide size and location. I find that increasing precipitation or soil depth results in an increased number of predicted landslides. In contrast, increasing soil strength through root reinforcement or friction angle results in a decrease in the number of predicted landslides. Increasing soil depth results in predicted landslides being preferentially located in locations with steep slopes, while increasing soil strength results in predicted landslides being preferentially located in locations with high drainage area. Precipitation affects characteristic landslide location differently: if lateral re-distribution of water is dominant, landslides are predominantly found in locations with high drainage area; in contrast, when vertical infiltration dominates they are predominantly found in areas with steep slopes. Predicted characteristic size increases with increased precipitation and with increased root strength. However, it decreases when the increased strength results from an increase of the soil friction angle. Under uniform soil thickness, characteristic size decreases with increasing soil depth. When soil thickness distributions are instead controlled by topography, increasing soil depth causes the predicted characteristic landslide size to first increase and then to decrease, after a critical value, reflecting the stabilization effect of very thick soils.

In chapter six the effects of the fine scale variability of root strength on slope stability are examined, using a method which could be extended to represent the impact of spatial variability of landslide-relevant parameters on landslide size, location, and abundance. A simple dynamic hydrological model is used in combination with a ten-minute rainfall intensity record for a landslide-triggering storm. When comparing with a map of debris flows which occurred during the storm, the procedure predicts landslides in the observed debris flow source areas. Although over-prediction is greatly reduced, there remain a considerable number of predicted landslides in areas which did not fail during the storm event. Regardless, this is a promising result, as it suggests that this procedure is capable of capturing the timing of landslides (as well as their size and location), given a sufficiently resolved characterization of the hydrology.

I find that the spatial structure of soil depth, pore pressure, and root strength determines the areas favorable to landsliding that can be exploited by rain storms, resulting in the characteristic size and location distributions of rainfall-triggered landslides. Varying these controlling properties, even uniformly, changes the spatial distribution of these areas in the landscape. This results in new characteristic distributions of landslide size and location, as landslides sample different parts of the landscape. This reveals the first-order control exerted by topography on shallow landslides. Furthermore, the general spatial pattern of landsliding did not fundamentally change with the introduction of stochastic variability in root strength or with variations in the mechanism of pore pressure generation. This highlights the fundamental role played by the topographically-controlled distribution of soil thickness in defining landslide location.

Understanding hazards posed by rainfall-triggered shallow landslides requires predicting where landslides will occur, when they will occur, how big will they be, how fast they will mobilize, and how far will they go. This research constitutes a significant step in this direction by providing some of the first coupled predictions of where and how big landslides are, and demonstrating that capturing their timing is well within reach. By coupling this procedure with climate and vegetation models we can now explore the impact of climate and land use change on the landsliding regime. By integrating the procedure into a landscape evolution model we can then explore how, over longer time scales, landslides shape a landscape.

Main Content
For improved accessibility of PDF content, download the file to your device.
Current View