Comparison of pixel, sub-pixel and object-based image analysis techniques for co-seismic landslides detection in seismically active area in Lesser Himalaya, Pakistan

Decision makers require timely and exact delineation of landslide information for effective and quick response to a co-seismic or rainfall-triggered calamity. Remote sensing (RS) techniques provide prompt disaster responses on a local to regional scale, particularly in remote, isolated, and inaccessible places. These techniques, which are divided into pixel, sub-pixel, and object-based techniques, each have their own set of advantages and disadvantages. This study aims to investigate and evaluate the virtues and drawbacks of pixel, sub-pixel, and Object-Based Image Analysis (OBIA)-based approach for detecting co-seismic landslides in Muzaffarabad, Pakistan (Lesser Himalayas). Using SPOT and ASTER imagery, a comparison of classification techniques based on the MLC (Maximum Likelihood Classifier), the Co-Registration of Optically Sensed Images and Correlation (COSI-Corr), and the OBIA was performed. On SPOT-5 images, MLC and OBIA approaches were applied, while ASTER images were used for sub-pixel classification. Overall accuracy for pixel-based MLC is 80.8, 90.9% for the sub-pixel COSI-Corr method, and 91.4% for OBIA. As a result, it can be inferred that COSI-Corr and OBIA-based classification outperformed the pixel-based MLC classification technique. The OBIA result is more spatially consistent than the pixel-based outcomes with speckled pixel effects, and depending on visual interpretation. Due to varied geomorphic conditions, the OBIA's ruleset makes it difficult to apply in different terrains. The COSI-Corr technique uses low-cost satellite data with a medium resolution to produce a plausible output. For this technique to work, suitable pre- and post-seismic event images (optical and digital elevation model) must be available (free of shadows and clouds).