Evaluation of Landslide Mapping Techniques and LiDAR-based Conditioning Factors

Abstract:

Landslides are a major geohazard, which result in significant human, infrastructure, and economic losses. Landslide susceptibility mapping can help communities to plan and prepare for these damaging events. Mapping landslide susceptible locations using GIS and remote sensing techniques is gaining popularity in the past three decades. These efforts use a wide variety of procedures and consider a wide range of factors. Unfortunately, each study is often completed differently and independently of others. Further, the quality of the datasets used varies in terms of source, data collection, and generation, which can propagate errors or inconsistencies into the resulting output maps. Light detection and ranging (LiDAR) has proved to have higher accuracy in representing the continuous topographic surface, which can help minimize this uncertainty. The primary objectives of this paper are to investigate the applicability and performance of terrain factors in landslide hazard mapping, determine if LiDAR-derived datasets (slope, slope roughness, terrain roughness, stream power index and compound topographic index) can be used for predictive mapping without data representing other common landslide conditioning factors, and evaluate the differences in landslide susceptibility mapping using widely-used statistical approaches. The aforementioned factors were used to produce landslide susceptibility maps for a 140 km² study area in northwest Oregon using six representative techniques: frequency ratio, weights of evidence, logistic regression, discriminant analysis, artificial neural network, and support vector machine. Most notably, the research showed an advantage in selecting fewer critical conditioning factors. The most reliable factors all could be derived from a single LiDAR DEM, reducing the need for laborious and costly data gathering. Most of the six techniques showed similar statistical results; however, ANN showed less accuracy for predictive mapping.

Keywords : LiDAR, Landslides, Oregon, Inventory, Hazard