VHR SAR and Optical sensors for earthquake damage mapping: automatic and semiautomatic techniques fo

1
VHR SAR and Optical sensors for earthquake damage
mapping automatic and semi-automatic techniques
for features extraction

• Marco Chini, Christian Bignami, Salvatore
  Stramondo
• Istituto Nazionale di Geofisica e Vulcanologia
  (INGV) Remote Sensing Laboratory, Rome, Italy

2
Outline

• Introduction
• Building map extraction
• by VHR SAR
• by VHR Optical
• Single Building Damage Level (SBDL)
• Test case Bam (2003)
• Conclusions

3
Introduction - 1

• Satellite Remote Sensing provides a synoptic view
  of surface effects of extreme events and has
  assumed an important role in supporting natural
  disaster management and mitigation.
• Concerning earthquakes EO data are used to
  generate a detailed map of damage to urban
  structures.
• The potential of SAR and Optical images can be
  exploited to support the decision chain of Civil
  Protection for crisis management and to provide a
  rapid mapping of damage to buildings and
  infrastructures.
• Combining SAR and Optical images overcome some
  limitation as those due to weather conditions and
  Sun illumination.

4
Introduction - 2

• In terms of Civil Protection, in order to manage
  the crisis phases, the prompt response, the fast
  detection and classification of damages grade
  affecting buildings and infrastructures can be
  useful to support the rescue activities.
• This is particularly true especially in remote
  areas or where the infrastructures are not well
  developed to ensure the necessary communication
  exchanges.
• Indeed, in the aftermath of these severe
  disastrous events one of the most urgent needs is
  to estimate with sufficient reliability and
  rapidity the amount of population and
  infrastructures affected for different degrees of
  damage.

5
Introduction - 3

• VHR (SAR and Optical) data are potentially able
  to provide very detailed mapping of the urban
  environment with a meter/sub-meter ground
  resolution.
• At the same time, the increase in volume of data
  creates additional problems in terms of automatic
  classification.
• An urban area appears as a complex structure
  containing buildings, transportation systems,
  utilities, infrastructures, commercial and
  recreational areas. At different scale, a simple
  building may appear as a complex structure also,
  with a lot of architectural details surrounded by
  gardens, trees, buildings, roads, shadows, social
  and technical infrastructure and many temporary
  objects, such as cars, buses, etc

6
Introduction - 4

• If on one hand the increase of geometrical
  resolution leads to a fine representation of the
  scene, on the other hand it results in a decrease
  of resolution in the spectral domain.
• It is evident that this problem is intrinsically
  related to the sensor resolution and it cannot be
  solved by increasing the number of spectral
  channels.
• Thus, a multiscale contextual analysis for
  exploiting contextual relations must be included
  in the analysis of VHR data in order to model the
  spatial information of the pixels.

7
Building map extraction

• Classification of VHR (SAR or Optical) image
  before the earthquake, in order to extract the
  map of the buildings to define a damage level at
  building scale, useful for the Civil Protection
  activities.
• The classification process must be fast, thus the
  selection of the most important input features
  and a selection of a faster classifier (i.e.
  unsupervised) is mandatory.
• Change detection analysis has to be applied only
  to the extracted building map, using pre- and
  post-seismic optical or SAR data.
• Definition of change detection features for
  characterizing the changes occurred in terms of
  damage levels is necessary.

8
Building map extraction Rome test case

• Input Panchromatic image second order
  textural parameters
• The land use map (9 classes) has been obtained
  using a Neural Network and pruning technique for
  reducing the input space and optimizing the net.
  
• The pruning step reduce the input space from 241
  to 140.

QuickBird image July 19, 2004
Land use Map of Rome
F. Pacifici, M. Chini, W. J. Emery, A neural
network approach using multi-scale textural
metrics from very high resolution panchromatic
imagery for urban land-use classification.
Remote Sensing of Environment, Vol. 113, No. 6,
pp. 1276 1292, June 15th, 2009.
9
Building map extraction Indianapolis text case

• Input Intensity SAR image Open and Close
  Morphological profiles with an anisotropic and
  asymmetric structure element.
• The land use map (7 classes) has been obtained
  using a Neural Network and pruning technique for
  reducing the input space and optimizing the net.
  
• The pruning step reduce the input space from 321
  to 50.

Land use Map of Indianapolis
TerraSAR-X image of Indianapolis, July 1, 2007
QuickBird image of Indianapolis, July 11, 2007
M. Chini, F. Pacifici, W. J. Emery,
Morphological operators applied to X - band SAR
for urban land use classification, Proc.
IEEE/IGARSS 2009, Cape Town (South Africa), 13
18 July, 2009.
10
Building map extraction Bam text case

• Input Panchromatic image Panchromatic Images
  filtered by Open and Close Operators with
  different window size (Morphological profile)
• The buildings have been classified using an
  unsupervised classifier

Buildings Map of Bam City
QuickBird image September 30, 2003
M. Chini, N. Pierdicca, W. J. Emery, Exploiting
SAR and VHR optical images to quantify damage
caused by the 2003 Bam earthquake. IEEE
Transactions on Geoscience and Remote Sensing,
Vol. 47, No. 1, pp. 145 - 152, January 2009.
11
Single Building Damage Level (SBDL)
SBDL a features which can be representative of
the damages occurred to each building. e.g.
Pearson correlation coefficient between pre and
post seismic pixels belonging to each building.
12
Test Case Bam (2003)

• On December 26, 2003 the southeastern region of
  Iran was hit by a 6.5 Moment Magnitude (Mw)
  earthquake whose epicenter was located very close
  to the historical city of Bam.

• DATASET
• Two QuickBird panchromatic images
• September 30, 2003
• Off-nadir angle 9.7
• January 4, 2004
• Off-nadir angle 23.8

13
Test Case Bam (2003)
Three different levels of damage Light (or No
Damage), Medium and Heavy.
14
Test Case Bam (2003)
Pan 09/30/2003
Pan 01/04/2004
15
Test Case Bam (2003)
Pan 09/30/2003
Pan 01/04/2004
16
Test Case Bam (2003)
Pan 01/04/2004
Pan 09/30/2003
17
Test Case Bam (2003)
Comparison between SBDL statistic and Ground
Survey Total Buildings SBDL12498 Ground
survey12063 Error3,6 Heavy damaged SBDL6150 G
round survey6615 Error7

• Using the threshold on the image difference (pre-
  and post-) has been possible only classifying the
  completely collapsed buildings. Medium and low
  damage not well classified.

18
Test Case Bam (2003)
Damaged buildings from ground survey European
Macroseismic Scale 1998 EMS98
Grade 1 Negligible to slight damage Grade 2
Moderate damage Grade 3 Substantial to heavy
damage Grade 4 Very heavy damage Grade 5
Destruction
Y. Hisada, A. Shibaya, M. R. Ghayamghamian,
(2004), Building Damage and Seismic Intensity in
Bam City from the 2003 Bam, Iran, Earthquake ,
Bull. Earthq. Res. Inst. Univ. Tokyo, Vol. 79
,pp. 81-93.
19
Test Case Bam (2003)
20
Test Case Bam (2003)
Optical Correlation coefficients vs EMS98 damage
EMS98

• 3 different damage levels recognizable
• G1G2G3 G4 G5
• Good detection of collapsed buildings

21
Conclusions

• The proposed approach can be applied to VHR
  optical or SAR images indifferently.
• The SBDL index has been defined and applied to
  the Bam test case.
• Results are satisfactory collapsed buildings can
  be well recognized.
• Improvements more accurate separation G4 G5
  by
• improving the classification and building mask
• reducing the parallax error (system requirements)
• We are applying the method to the LAquila
  earthquake using COSMO-SkyMed and QuickBird image
  using very detailed ground truth from ground
  survey by INGV experts.