Automated identification and quantification of rapid mass movement processes in a mountainous region of British Columbia

This research addresses the use of optical satellite sensor data and digital elevation model derivatives in the automated detection and quantification of rapid mass movements in the humid mountains of southwestern British Columbia. Whereas previous attempts to classify landslides using per pixel spectral response patterns have provided classification accuracies that are less than 60%, it is demonstrated that a combination of high resolution optical imagery, image segmentation and ancillary data derived from a digital elevation model can discriminate some types of mass wasting processes with accuracies of 80% or higher. The spatial resolution of the imagery is shown to be critical to the successful classification of landslides both in terms of information derived from textural analysis and in the ability to successfully segment landslide features. Furthermore, the data generated in this manner can be used for geomorphic research in terms of characterizing the occurrence of mass wasting within the bounds of the image scene and estimating event magnitude-frequency relations and modelling erosion on a regional scale. The data can also be produced for other geomorphic applications such as the calibration of landscape evolution models in order to predict landscape change due to specific processes over basin scales. The method has distinct advantages over the manual interpretation of analog data sources in the objectivity of the classification within a given image scene as well as a more rapid and accurate data throughput from the mapping phase to data analysis. This yields more confidence in the consistency of results and therefore in the comparison between datasets.