HIGH RESOLUTION SATELLITE IMAGERY NEW PERSPECTIVES FOR THE EARTHQUAKE RISK MANAGEMENT Lucian CHIROIU Denis Diderot University, Pôle Image Laboratory Paris, France Geosciences Consultants Paris, France WORKSHOP ON APPLICATION OF REMOTE SENSING TECHNOLOGIES FOR DISASTER RESPONSE University of California, Irvine. September 12th, 2003
Introduction Application of high resolution satellite imagery to Bhuj (India) earthquake of January 26 th, 2001 & Boumerdes (Algeria) earthquake of May 21 th, 2003. - 1m IKONOS imagery (source: spaceimaging.com & European Space Imaging) - 0.60m QUICKBIRD imagery (source: DigitalGlobe) principles of damage detection by photo-interpretation - mono temporal analysis - multi temporal analysis - typical damage examples (detection of a soft story damage) automatic damage detection - recent advancements damage mapping fast estimation of human casualties conclusions & perspectives
Principles of damage detection Mono temporal analysis Radiometric heterogeneity of the roofs Geometric irregularities of contours Absence of shadows 0 25 m 0 25 m The roofs are not continuous, indicating important damage. Example on Zemmouri, Algeria Absence of shadow, indicating collapse. Example on Zemmouri, Algeria
Principles of damage detection Examples of complete damage: easy detectable Complete damage in Bhuj, India, after January 26 th earthquake, 2001 Complete damage in Zemmouri, Algeria, after May 21 th earthquake, 2003
Principles of damage detection Mono temporal analysis difficulties for detecting damage for masonry or adobe traditional construction, in dense urban environment Traditional bhonga constructions, in Bhuj, India (source: world-housing.net) Urban zone in Bhuj, India Urban zone in Zemmouri, Algeria
Principles of damage detection Multi temporal analysis change detection Optical imagery: - changes in the structure, in the texture or in the contour - the absence or the decrease of shadows stereoscopic analysis before after Example of complete damage of a multi-story building in Boumerdes, Algeria, after May 21th earthquake, 2003
Principles of damage detection Multi temporal analysis high accuracy of interpretation detection of typical soft storey damage before after Example of a soft story damage in Boumerdes, Algeria, after May 21th earthquake, 2003
Automatic damage detection Recent advancement: damage detection by morphological analysis Example of a test zone in Bhuj Original image Filtered image
Principles of damage detection Building contours layer (in yellow) overlaid on the convex envelope layer (in red) Automatic detection of damaged buildings Accurate results for a few test zones Main difficulty: extraction of the building footprints
Damage mapping Damage detection Mapping of affected zones : a) Building level b) Zone level c) Region level Additional mapping: - Accessibility (roads & other important features) - Urban structure (building types) - Possible locations of relief camps Example of damage mapping at the level of a building, on Boumerdes
Damage mapping Example of damage mapping at the level of a zone, at Zemmouri
Damage mapping Example of damage mapping at the level of a region, in Boumerdes province
Damage mapping Data can be integrated into a GIS base Standard freeware allows viewing and basic data analysis (vector & raster format) Ex: ArcExplorer, ProViewer, TNT Atlas, SVG viewer, Freelook, Java utilities Raster data can be compressed by specific software Ex: ECW, MrSID (250 Mb 5Mb) Rapid handling of the data Transfer by internet to the rescue teams already deployed on the affected zone Transfer of low resolution maps ( *.tiff or *.jpeg format) by satellite networks (cell phones)
Fast estimation of casualties Assumptions (casualty ratios): - regarding the complete damage, given that the structure is completely destroyed, it was assumed that 80% of the occupants are dead, and 20% of the occupants are injured. - regarding the extensive damage, it was assumed that 5% of the occupants are dead, and 60% are injured.
Fast estimation of casualties Application to Zemmouri, Algeria: Damage Complete Extensive Density prs/m 2 0,027 0,027 Surface (m 2 ) 16270 132270 Affected persons 440 3571 Death Rate 20% 60% Field estimations: more than 400 dead persons 1 Casualty ratios Injury Rate 80% 5% Total : Injured Persons 88 2143 2231 Dead Persons 352 179 531 Application to Bhuj, India: Damage Density prs/m 2 Surface (m 2 ) Affected persons Casualty ratios Death Rate Injury Rate Injured Persons Dead Persons Complete 0,028 251250 7035 20% 80% 4221 352 Extensive 0,028 180900 5065 60% 5% 1013 4052 Total : 5234 4404 Field estimations: 5065 dead persons and 10925 injured persons 2 1 according to Le Monde journal of May 27 th 2 according to UNDMT reports
Conclusions Accurate damage detection by photo - interpretation (ex: recognition of soft storey damage) Automatic damage detection methods of remote sensing are still under development Constraints and difficulties related with : - dense urban environment - extraction of the building s contours - cloud coverage - timing of imagery acquiring Accurate mapping of affected zones Reliable fast estimation of casualties Preliminary reconnaissance of the urban environment Support to crisis management & disaster recovery Reduced delay of analysis allowed by a manual cartography (20 km 2 / 2 operators / 4 hours)
Perspectives Development of building inventories (easily recognition of various building types) Use or development of geomatic products: urban DEM, GIS demographic databases, etc - More precision for the existing loss estimation models - Interesting costs of production ( gain of time & money) Simulation of an urban DEM