Automated landslide detection model to delineate the extent of existing landslides

Abstract

Landslides are one of the most common natural hazards and cause major socioeconomic impacts worldwide. Identifying the locations of the active or inactive landslides before development may play a major role in identifying areas of high risk. Traditional methods for inventorying landslides involve field surveying and interpretation of photogrammetric data. The advent of recent remote sensing technologies has expedited this process, and as a result, several computer-based algorithms used to identify the locations of past landslides have been proposed. Computer-based analyses provide significant advantages over traditional methods; however, a majority of these computer-based analyses require the user to define the properties of the landslide prior to the search and require supervision and quality assurance. The purpose of this study is to present a simple, new methodology that can be implemented with readily available tools and datasets without the need to supervise the analysis after the parameters regarding landslide morphology are defined for that region. This methodology is referred to as automated landslide detection model (ALDM). Three areas with LiDAR bare earth digital elevation models (DEMs) have been used to test the ALDM, each consisting of a varying range of mapped landslide features. The ALDM results were compared against data obtained from the Pennsylvania Department of Conservation and Natural Resources and landslides that were determined visually from the hillshade map of the study area. The results demonstrate that the ALDM method was able to accurately capture both the landslides and non-landslides in all of the areas evaluated with accuracies of 70% and 92%, respectively. Additionally, the study showed that the proposed ALDM method could be implemented in different regions where landslides of different shapes and sizes could be detected.