Semi-Automated Object-Based Model for Producing Landslide Inventory Map (Case Study: Vazroud Watershed)

The first step in the study of landslide susceptibility and hazard is to produce a landslide inventory map.The usual methods of landslide mapping include field investigations, combined with visual interpretation of aerial photographs. Modern and newfangled remote sensing techniques, using satellite images, causes in facilitating of landslide mapping and reduce the required time. In this study, a semi automatic model based on object oriented analysis was developed in eCognition software to identify landslides at the central part of the Albourz Mountains in Vazroud watershed. A total of 105 landslide events were detected within the study area using documentation from previous research, interpretation of aerial photographs and visual inspection of the Google Earth images. 70% (74 points) of detected landslides were randomly selected for training, and the remaining 30% (31 points) cases were used for test. The model was developed based on spatial, spectral and geometry characteristics of training landslides as well as thematic maps of the study area. Results showed the model accuracy of 74.19% in the detection of test landslides. Also the accuracy of 41.28% was achieved in detection of the extent of landslides which in the condition of using high-resolution images, would be improved.