Landslide Susceptibility Modeling Using Logistic Regression and Artificial Neural networks in GIS: a case study in Northern Showa area, Ethiopia
No Thumbnail Available
Date
2007-02
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Addis Ababa Universty
Abstract
Landslide susceptibility mapping has been undertaken using topographic parameters and
lithologic information in north central Ethhiopia, Debresina area. The area has been one well
known for landslide occurrences throughout its history.
Before conducting the GIS analysis, extraction of some of the remote sensing data has been
undertaken involving rectification and enhancement of Landsat ETM+ image of May 2000
for lithologic units interpretations. Image classification of 5/7, 5/4, ¾ band combination
resulting in 70% accuracy and another classification using the first principal componets
which accounted for 98 % of the whole bands data was done resulting in 69% accuracy.
Further visual interpretation of the lithologic units has been undertaken to produce the final
lithologic map interpretation. In addition the digital elevation model (DEM) of the area was
obtained at 20m resolution by vectorizing contours from the topographic map of the area to
extract the topographic parameters.
In this study two susceptibility mapping methods have been employed: Logistic regression
and Artificial Neural Network (ANN). Both have been used to generate the weights to
represent the degree of contribution of selected seven parameters: Lithology, Slope, Aspect,
Plan curvature, Profile curvature and flow accumulation in different platforms than
geographic information system (GIS). Preceding to these the class weights of the various
parameters were obtained by BSA method. Finally raster calculation of the seven layers of
the parameters was conducted and two susceptibility maps were produced. The weights
generated by ANN signified the contribution of Planar curvature, Aspect and Slope type
towards landslide occurrence while that of the Logistic regression method signified
Lithology and flow accumulation scoring higher in contribution towards landslide occurrence
than the rest of the parameters. Finally, the outputs have been evaluated using the inventory
of slope failure from the same period as those used for training the models and one from the
recent massive landslide that occurred in the area.
Description
Keywords
Neural networks in GIS