DefinePK

Abstract

Customary co-seismic landslide detection methods are meticulous, time demanding and very difficult to be adopted for regional-scale assessment. To provide on time fast and updated slope failure information, over the years numerous remote sensing based semi-automatic landslide detection methods have been developed and applied. However, implementation of these techniques requires some vital considerations like accurate and precise geometric correction, resampling and false positives elimination. To address these issues a new change-detection based technique of COSI-Corr (Co-registration of Optically Sensed Images and Correlation) has been adopted for mapping co-seismic landslides in Muzaffarabad area. However, due to COSI-Corr high sensitivity for detecting changes on the sub-pixel level, it is prone to detecting changes caused by vegetation, erosion and variation in built-up area size as false positive values. In this study, the influence of various factors like vegetation, sedimentation, erosion and built-up areas on landslides automatic detection results accuracy has been investigated. After the implementation of COSI-Corr technique, stepwise masking is performed. The false positives are successively removed from the landslide class by eliminating the noises resulting from drainage, urban sprawl and vegetation phonology. The results accuracy was increased after the application of each mask. The number of false positives was greatly reduced by the application of the vegetation-based mask. The best threshold found was 0.1 for which error of omission and error of commission was less than 11%. The results also showed that satellite images with medium spatial resolution could be successfully employed for the automatic detection of co-seismic landslides.