Identifying Building Damage Patterns in the 2016 Meinong, Taiwan Earthquake Using Post-Event Dual-Polarimetric ALOS-2/PALSAR-2 Imagery

The 2016 magnitude 6.4 Meinong earthquake caused catastrophic damage to peoples lives and properties in Taiwan. Synthetic Aperture Radar remote sensing is a useful tool to rapidly grasp the near real-time building damage to areas affected by the earthquake. Previous studies employed X-band single polarized high-resolution synthetic aperture radar imagery to identify building damage. However, suitable X-band single polarized high-resolution synthetic aperture radar imagery is not always accessible. Therefore, this research applied L-band dual-polarimetric ALOS-2/PALSAR-2 data to analyze the radar scattering characteristics of three types of affected buildings in the 2016 Meinong earthquake. The results show that collapsed buildings are characterized by a weak double-bounce scattering due to a reduced dihedral structure, while the characteristics of slightly damaged buildings are similar to those of undamaged buildings. Furthermore, the discrimination ability of a series of polarimetric, texture, and color features derived from the dual-polarimetric SAR data for three types of buildings affected by the earthquake are quantified based on a statistical analysis using the pixels in the combined areas of layover, shadow, and building footprint of each building. The results of the statistical analysis show that the spaceborne dual-polarimetric ALOS-2/PALSAR-2 images have good potential to distinguish between slightly damaged buildings and collapsed and tilted buildings. However, it is still difficult to distinguish between collapsed and tilted buildings. In addition, the results of the statistical analysis show that the mean value and variance value of the Gray-Level Co-Occurrence Matrix of the span image are two suitable features by which the categories of building damage can be distinguished. The polarimetric and color features demonstrated poorer performance in terms of distinguishing between damage categories than the texture features.