Landslide Hazard Evaluation Using Artificial Neural Networks and GIS

Abstract

Lantau Island within the territory of Hong Kong has natural and undeveloped terrain (Dai et al. 2001), along with intense and frequent rainfall. It is thus a suitable area for preparation of a landslide susceptibility map of the area. The methods of using Landslide Susceptibility Values (LSV) and artificial neural networks (ANN) were applied in the GIS environment of ArcGIS 9.3 to prepare two landslide susceptibility maps of Lantau Island. The application of LSV and GIS to produce a landslide susceptibility map included the determination of LSV values of causative factors and calculation of a cumulative Landslide Susceptibility Index (LSI) for each pixel which was used to decide zones susceptible to landslides. The application of ANN required initially the preparation of input vectors from causative factors and output vectors of landslide susceptible zones by taking the LSV-produced map as the reference. The neural networks were trained and tested using the Neural Network Toolbox in MATLAB. The best network was obtained and applied further to predict the landslide susceptible zones for the whole study area and a landslide susceptibility map was prepared. These maps were compared with each other and with the landslide susceptibility map produced by Dai et al. (2001) using a logistic regression model. The landslide susceptibility map produced by applying ANN predicted more landslide susceptible regions with high and moderate susceptibility in the study area compared to the map produced using LSV. However, for some regions of the study area the LSV method performed better than the ANN method. Nevertheless, both methods produced quality maps and the performance of ANN was satisfactory, even with a small training dataset.