Estimating landslide occurrence via small watershed method with relevance vector machine
Journal
EARTH SCIENCE INFORMATICS
Journal Volume
13
Journal Issue
2
Pages
249
Date Issued
2020
Author(s)
Abstract
The mechanism of landslide occurrence is complicated due to the dependency and nonlinear relationship of various physical factors. A promising prediction model, which can be used to locate the high-risk regions and a corresponding prevention strategy can be prepared to reduce the slide occurrence and its consequence, is therefore desired. To perform the landslide assessment for a large-scale slope, this study proposes to use the method of small watershed that is integrated with Relevance Vector Machine (RVM) to enhance the prediction accuracy. Effect of physiographic and hydrological factors such as slope steepness, dip slope ratio, landslide ratio, and cumulative rainfall are investigated. To estimate the occurrence of landslide, RVM first maps the aforementioned factors into a feature space using Gaussian radial basis function. A linear boundary, distinguishing landslide or not, is then obtained through the search of the optimal weights. To find these weights, a Bayesian theory-based optimization problem is formulated and solved by iteratively reweighted least squares algorithm and the Laplace approximation procedure. The proposed model is validated by the data collected from Kaoping River Basin. Results indicate that the proposed RVM-based small watershed approach possesses a prediction accuracy of 87.5%, which is better than those of using Support Vector Machine (SVM), Least-Square Support Vector Machine (LS-SVM), and logistic regression, providing authorities in their hazard alert system to minimize the life or property losses caused by the landslide.
Subjects
Small watershed; landslide; RVM; Physiographic factor; Hydrological factor
SDGs
Publisher
SPRINGER HEIDELBERG
Type
journal article
File(s)![Thumbnail Image]()
Loading...
Name
3-1 參考作1.pdf
Size
1.8 MB
Format
Adobe PDF
Checksum
(MD5):bbeaaad955f73c0f88a5132eeef37afa