A data integration approach for earthquake damage assessment using VHR SAR and optical imagery

The new generation of spaceborne Synthetic Aperture Radars (SAR) and optical sensors provides metric or submetric resolution imagery, thus allowing, in principle, the detection of single building damage after an earthquake. However, the complexity of the image forming mechanisms within urban settlements, especially of radar images, makes the automatic detection of damaged buildings still a challenging task. The integration of different pieces of information, not only from remote sensing but also from geological and structural data sources, may help providing reliable damage assessment. Such an integration is foreseen in the APhoRISM (Advanced PRocedures for volcanic Seismic Monitoring) FP7 project which is the framework of the present study. In this work we will present a semiautomatic procedure exploiting Very High Resolution images acquired before and after an earthquake from both SAR and optical sensors for providing damage assessment products at single building scale. In order to test the proposed methodologies we use optical images from QuickBird satellite and Spotlight COSMOSkyMed SAR imagery of the seismic event that hit L'Aquila city (Italy) on April 6, 2009. For validation purposes two ground based damage maps are used. The first one refers to the survey performed by the Istituto Nazionale di Geofisica e Vulcanologia(INGV), while the second one is related to ground survey carried out by the Italian Civil Protection Department (DPC). Dealing with metric and submetric resolutions, objectbased change detection approaches are proposed. For segmenting optical images a GIS layer reporting building footprints is used. This allow the change analysis to be focused on the objects of interest, avoiding false alarms due for example to vegetation changes and temporary objects. As for SAR data, because of the complexity and peculiarity of building appearance in radar images, an adhoc segmentation technique of the preevent image has been developed. It is based on the use of morphological profiles to extract bright stripes and ridges caused by double bounce and/or layover mechanisms, the most distinctive features of the SAR building response. Looking at changes in these regions heavy damaged buildings can be identified. When a building collapses changes are also expected within the building footprint and in the shadow area. Typically an increase of the backscattering is observed in these regions due to the scattering contribution from debris and to the return coming from the ground previously occluded by the shadow. In order to single out such kind of changes a segmentation approach exploiting the KullbackLeibler Divergence (KLD) is proposed. A deep analysis of many change detection features, evaluated at object scale, is done by assessing their correlation with damage information provided by ground surveys. As for SAR data, the intensity ratio, the interferometric coherence, the intensity correlation and the KLD are analyzed. Regarding optical data, features describing texture and color changes are considered in addition to statistical similarity and correlation descriptors, such as the KLD and the Mutual Information. Exploiting these features, a non parametric classification approach based on the Bayesian Maximum A Posterior criterion is implemented for both SAR and optical data. The classification performances are not excellent when tested using the available ground truth, but a similar uncertainty has been observed comparing the INGV ground truth with that provided by the Italian DPC demonstrating the challenge of an accurate damage assessment even on ground (considering the difficulties encountered after an earthquake). SAR and optical data allow comparable performances in terms of damage sensitivity. Better performances in terms of