Rapid co-seismic landslide mapping using object-based change detection by bi-temporal GaoFen-2 in Tibet, China near Nepal

Although numerous methods have been applied for event-induced landslides based on imagery data in past decades, it is still challenging how to delineate landslides quickly and accurately at a regional scale. In this paper, we developed a change detection and rule-based object-oriented image analysis framework using bi-temporal optical data from the GF-2 satellite and topographic constraints for rapid landslide detection in Tibet near Nepal. First, available pre- and post-event high-resolution optical images were selected and preprocessed, and the hillshade as an additional layer was used to image the initial segmentation extracted from DEM data. Then, two popular segmentation scale optimization methods were combined to determine several optimal scale parameters for multi-scale classification and further analysis. Furthermore, we used a hierarchical knowledge-based classification approach to distinguish landslides from the background with the spectra, texture and terrain features series, which can reflect changes in the earth surface. Finally, the identification result of the proposed unsupervised framework was verified using a manual inventory and qualitatively by optical images of the unmanned aerial vehicle. Additionally, the effectiveness of this approach is further validated by comparisons with the regularly used random forest-based method. The experimental results revealed that our framework is more accurate and efficient for landslide practical applications.