Advanced

Shallow landslide susceptibility : modelling and validation

Pimiento Chamorro, Edgar LU (2010) In LUMA-GIS Thesis GISM01 20091
Dept of Physical Geography and Ecosystem Science
Abstract
Rainfall frequently triggers shallow landslides in mountainous areas
worldwide. Landslide susceptibility maps express the probability of
occurrence of landslides based on terrain conditions; they are useful for
disaster prevention and land use planning. This report is about val-
idating a qualitative approach to map global landslide susceptibility,
based on the weighted linear combination (WLC) of slope gradient,
soil type, soil texture, elevation, land cover and drainage density. The
parameters are derived from digital global databases. The accuracy
assessment was based on a detailed landslide inventory of a 160-km2
area in Japan, using the receiver-operating characteristic (ROC) plot
area under the curve (AUC). The AUC... (More)
Rainfall frequently triggers shallow landslides in mountainous areas
worldwide. Landslide susceptibility maps express the probability of
occurrence of landslides based on terrain conditions; they are useful for
disaster prevention and land use planning. This report is about val-
idating a qualitative approach to map global landslide susceptibility,
based on the weighted linear combination (WLC) of slope gradient,
soil type, soil texture, elevation, land cover and drainage density. The
parameters are derived from digital global databases. The accuracy
assessment was based on a detailed landslide inventory of a 160-km2
area in Japan, using the receiver-operating characteristic (ROC) plot
area under the curve (AUC). The AUC permitted to compare analysis
approaches and dierent parameter combinations. The AUC for the
WLC model was 0.47, below a random classication. Two approaches
improved the model accuracy, using the weights of evidence (WOE)
approach raised the accuracy to 0.64, and using a higher resolution
DEM raised the accuracy to 0.66. On the other hand, a quantitat-
ive approach based on logistic regression (LR) and using the software
package Spatial Data Modeller (SDM) produced models with AUC
between 0.67 and 0.71. The highest accuracy for a model including
lithology, slope gradient, prole curvature, plan curvature and elev-
ation. The reason for the higher accuracy of the LR models is that
the occurrence of landslides depends on local conditions, expressed by
the quantitative relations, while the qualitative weights of the WLC
model were developed for a global model using different criteria. (Less)
Please use this url to cite or link to this publication:
author
Pimiento Chamorro, Edgar LU
supervisor
organization
course
GISM01 20091
year
type
H2 - Master's Degree (Two Years)
subject
keywords
landslide susceptibility, weights of evidence, logistic regression, ROC plot AUC, validation
publication/series
LUMA-GIS Thesis
report number
6
language
English
additional info
Hiromitsu Yamagishi, Ehime University, Japan.
id
3559066
date added to LUP
2013-02-28 11:31:59
date last changed
2013-02-28 14:45:17
@misc{3559066,
  abstract     = {Rainfall frequently triggers shallow landslides in mountainous areas
worldwide. Landslide susceptibility maps express the probability of
occurrence of landslides based on terrain conditions; they are useful for
disaster prevention and land use planning. This report is about val-
idating a qualitative approach to map global landslide susceptibility,
based on the weighted linear combination (WLC) of slope gradient,
soil type, soil texture, elevation, land cover and drainage density. The
parameters are derived from digital global databases. The accuracy
assessment was based on a detailed landslide inventory of a 160-km2
area in Japan, using the receiver-operating characteristic (ROC) plot
area under the curve (AUC). The AUC permitted to compare analysis
approaches and dierent parameter combinations. The AUC for the
WLC model was 0.47, below a random classication. Two approaches
improved the model accuracy, using the weights of evidence (WOE)
approach raised the accuracy to 0.64, and using a higher resolution
DEM raised the accuracy to 0.66. On the other hand, a quantitat-
ive approach based on logistic regression (LR) and using the software
package Spatial Data Modeller (SDM) produced models with AUC
between 0.67 and 0.71. The highest accuracy for a model including
lithology, slope gradient, prole curvature, plan curvature and elev-
ation. The reason for the higher accuracy of the LR models is that
the occurrence of landslides depends on local conditions, expressed by
the quantitative relations, while the qualitative weights of the WLC
model were developed for a global model using different criteria.},
  author       = {Pimiento Chamorro, Edgar},
  keyword      = {landslide susceptibility,weights of evidence,logistic regression,ROC plot AUC,validation},
  language     = {eng},
  note         = {Student Paper},
  series       = {LUMA-GIS Thesis},
  title        = {Shallow landslide susceptibility : modelling and validation},
  year         = {2010},
}