Dissertations, Theses, and Capstone Projects

Date of Degree

9-2016

Document Type

Dissertation

Degree Name

Ph.D.

Program

Earth & Environmental Sciences

Advisor

Reza Khanbilvardi

Committee Members

Pedro Restrepo

Kyle McDonald

Nir Krakauer

Subject Categories

Geology | Geomorphology | Geotechnical Engineering | Risk Analysis

Keywords

Shallow Landslides, Rainfall-induced, Remote sensing, GIS, Logistic Regression

Abstract

Rainfall-induced landslides are one of the most frequent hazards on slanted terrains. They lead to considerable economic losses and fatalities worldwide. Intense storms with high-intensity and long-duration rainfall have high potential to trigger rapidly moving soil masses due to changes in pore water pressure and seepage forces. Nevertheless, regardless of the intensity-duration of the rainfall, shallow landslides are influenced by antecedent soil moisture conditions. To the present day, no system exists that dynamically interrelates these two factors.

This work establishes a relationship between antecedent soil moisture and rainfall expressed in the form of a Shallow Landslide Index (SLI) at 1km2 resolution for the United States. The proposed mathematical model is based on a logistic regression-learning algorithm that systematically adapts from previous landslide events listed in a comprehensive landslide inventory. Because landslides are considered to be the product of the interaction of static and dynamic factors, static factors are examined first. Also, because significant uncertainties are found when mapping factors in large spatial scales, buffer and threshold techniques are used to downscale areas and minimize uncertainties. Static parameters for 230 shallow rainfall-induced landslides in the continental United States are examined. ASTER GDEM is used as the basis for topographical characterization of slope and buffer analysis. Slope angle threshold assessment at the 50, 75, 95, 98, and 99 percentiles is tested locally. Further analysis of each threshold in relation to other parameters is investigated in a logistic regression model for the continental U.S. It is determined that lower than 95-percentile thresholds under-estimate slope angles and best regression fit can be achieved when utilizing the 99-threshold slope angle. This model predicts the highest number of cases correctly at 87.0% accuracy. A one-unit rise in the 99-threshold range increases landslide likelihood by 11.8%. The logistic regression model is carried over to ArcGIS where all static variables are processed based on their corresponding coefficients. A regional slope susceptibility map for the continental United States is developed and analyzed against the available landslide records and their spatial distributions.

Consequently, a mathematical algorithm is proposed to determine landslide probability as a function of static and dynamic factors employing accumulated water volume. As rainfall thresholds alone do not provide information about the soil wetness profile with depth, the Shallow Landslide Index (SLI) is intended to be an indicator of antecedent root soil moisture and rainfall accumulation over a 1km2 pixel area. Experimentally, root-soil moisture retrieved from AMSR-E and rainfall retrieved from TRMM are used as proxies to develop such index. Static and dynamic conditions leading to each landslide event are examined over 60-days, 30-days, 10-days and 7-days. The input dataset is randomly divided into training and verification sets where validation results indicate that the best-fit model predicts the highest number of cases correctly at 93.2% accuracy. The resulting equation is then incorporated in a python subroutine that calculates the SLI for each of the 900,000-pixel points. For each pixel, the algorithm incrementally tries values from 0 to the value that makes the event probability equal to 1.

Since AMSR-E and TRMM stopped working in October 2011 and April 2015 respectively, a solution that works for the future is presented. Root-soil moisture retrieved from SMAP and rainfall retrieved from GPM are used to develop models that calculate the SLI for the continental United States for 10-days, 7-days, and 3-days. The resulting models indicated a strong relationship (93.4%, 93.8%, and 93.7% respectively) between the predictors and the prediction value. Nevertheless, as of the writing of this work, the SMAP root soil moisture product has a mean latency of 7-days hence the SLI is functional for 10 or 7 days. It is expected that as SMAP’s latency is reduced, the SLI functionally can also be brought to a shorter period. The resulting SLI map can potentially be used as an indicator of the total amount of rainfall needed for a given duration of time to trigger a shallow landslide in a susceptible area.

Share

COinS