Validation of an Earthquake Damage Map from Vhr Optical Images Using a Ground Survey

In this work we investigated the effects to man made structures and buildings caused by the December 26th, 2003, Bam(Iran) earthquake. The epicentre of the seism was nearby the ancient urban area of Bam and caused strong damage. Thedataset was composed by preand post-earthquake QuickBird panchromatic images, with 60 cm geometric resolution.The very high resolution images acquired by QuickBird have been used to show the capability of this data to mapdamage at building scale by means of a segmentation approach. QuickBird took clear images of Bam on January 3,2004, eight days after the event and on September 30, 2003, about three months before.The use of very high resolution images from one side allows to detect very small details, while on the other handbuildings appear as rather complex structures difficult to be interpreted. Furthermore, they may be surrounded byscattering objects making less evident the contrast between the roofs and the ground, thus increasing the difficulties inthe segmentation process. False alarm signals affecting the change detection process can be due to the shadows andtheir variations along the year. This implies that a single band is not enough to classify such complex environment.Moreover, when using panchromatic very high resolution (VHR) images, some objects/classes may have similarradiance, thus preventing their correct recognition using a single channel. Therefore, the extraction of other informationsuch as shape of the objects or other geometrical information can be very useful when dealing with panchromaticimages. To this aim, morphological operators have been applied to our image dataset. In particular, the Open and Closeoperators have been adopted. Using these operators with different windows size (spanning from 3x3 up to 125x125pixels), a morphological profile for each object/class in the scene has been extracted from the original panchromaticimages acquired before and after the earthquake. Hence, all the buildings within the urban area have been identified bymeans of a supervised classification approach, both in the pre-seismic and post-seismic image. After this buildingidentification procedure at the resolution of the original Quickbird image, the damage level was computed within anarea of the urban settlement (e.g., a district) by comparing pixel radiance belonging to the same building before andafter the earthquake. The procedure has already been validate against a map identifying regions within the Bam citywith comparable degree of damage.In this work, we focus on the capability of the procedure to detect damage at single building scale, either total or partial.For this purpose, a validation process has been performed in order to compare the final high resolution damage mapextracted from QuickBird images with a ground survey mapping the buildings totally or partially collapsed during theearthquake. The latter, reported by Hisada et al. (2004), is a local survey of seven areas related to seven Strong MotionStations in Bam. A damage grade for each building within the areas has been assigned during the survey, according tothe five level of the European Macroseismic Scale 1998 (EMS98). The comparison of the ground truth with our highresolution damage map was fairly satisfactory and allowed us to better tune the change detection algorithm devote todamage estimation from VHR images. REFERENCESHisada, Y. and Shibaya, A., “Building Damage and Seismic Intensity in Bam City from the 2003 Iran, Bam,Earthquake”, Bulletin of Earthquake Research Institute, University of Tokyo, Vol.79, pp.81-93, 2004.