Natural Disaster Damage Evaluation Using Fully Polarimetric Techniques with Spaceborne SAR Data

Si-Wei Chen1, Yong-Zhen Li1, Xue-Song Wang1, Motoyuki Sato2

1 State Key Laboratory of Complex Electromagnetic Environment Effects on Electronics and Information System (CEMEE),

National University of Defense Technology, China. chenswnudt@163.com

2Tohoku University, Japan. sato@cneas.tohoku.ac.jp

Abstract

This paper focuses on earthquake/tsunami damage investigation over urban areas by exploring the multi-temporal

spaceborne ALOS/PALSAR PolSAR data. The polarimetric scattering mechanism changes before- and after-tsunami, at the city block scale, have been examined using model-based decomposition and PO angle techniques. These analyzes are used to establish the relationships between the polarimetric scattering mechanism changes and damage levels. The basic scattering structures such as ground-wall dihedral structures from the built-up areas were found to be stable even over a long temporal baseline. Two polarimetric indexes have been proposed for damage level indication. Experimental results validate the efficiency of these two indicators, since the built-up areas with different damage levels can be well discriminated. These results demonstrate the importance and efficiency of full polarimetric information for natural disaster assessment.

1. Introduction

The occurrence of the observed natural disasters, such as earthquakes and tsunamis, appears to have increased in recent

decades. Quick observation of the damage caused by an earthquake and tsunami is extremely important. High resolution optical images are usually used [1]. However, the use of optical sensors is limited by cloudy weather and nighttime darkness. In contrast, synthetic aperture radar (SAR) can work day and night. SAR data is particularly useful when weather conditions are not suitable for optical sensing. Several studies using SAR and a combination of SAR and optical data for damage evaluation have been presented [2-4].

However, limited work has been reported [5-6] using full polarimetric SAR (PolSAR) data. Generally, full polarimetric techniques can better assist the understanding of scattering mechanisms, and provide additional and more accurate information compared to partial polarization modes [7]. Complementary to previous studies, we will explore polarimetric techniques for damage investigation over urban areas. The study case is the great Eastern Japan earthquake and tsunami of March 11th, 2011. Full PolSAR datasets acquired by ALOS/PALSAR are used. This work aims to find relationships between the polarimetric parameters and the damage level. Firstly, polarimetric scattering mechanism changes are examined by recently developed model-based decomposition [8-10] and polarization orientation (PO) angle technique [11]. Two polarimetric indicators are then proposed for damage level indication.

2. Fully Polarimetric Techniques

2.1 Model-Based Decomposition

Subject to the reciprocity condition ( )HV VHS S= , the coherency matrix is 11 12 13

21 22 23

31 32 33

T T TT T T T

T T T

⎡ ⎤⎢ ⎥= ⎢ ⎥⎢ ⎥⎣ ⎦

. Polarimetric model-based

decomposition is an effective technique for understanding the scattering mechanisms of PolSAR data 7. The general principle is to decompose a polarimetric matrix into a summation of several basic scattering models. For incoherent decomposition, using the coherency matrix as an example, the form is

d dbl v vol s oddT f T f T f T= + + + (1) where dblT , volT and oddT are respectively the double-bounce, volume and odd-bounce scattering models; df , vf and sf are the corresponding decomposed coefficients.

978-1-4673-5225-3/14/$31.00 ©2014 IEEE

After the decomposition, decomposed powers for each scattering mechanism are available, represented by dP , vP and sP for the double-bounce, volume and odd-bounce scattering mechanisms, respectively. The dominant scattering mechanism for each resolution cell can be determined thereby.

2.2 Polarization Orientation Angle

In urban areas, the PO angle is highly correlated to the orientations of buildings. The scattering from an urban area is

relatively deterministic and so are the measured PO angles. For buildings parallel to the flight pass, the PO angle is zero, while oriented buildings will rotate the polarization basis and induce PO angle shifts from zero. For damaged built-up areas, the scattering from surviving buildings and remaining foundations could be also deterministic. However, the scattering from the randomly oriented debris becomes less or non deterministic. The induced PO angles from such debris will also distribute randomly and the fluctuation could be significantly increased. Consequently, exploring the difference of the PO angle distributions before and after the tsunami has the potential to reflect the damage condition. The expression of the PO angle is

( )231

22 33

2Re1 tan 0,14

Tn n

T Tθ π−⎛ ⎞

= ± =⎜ ⎟−⎝ ⎠ (2)

3. ALOS Data And Study Area

The great tsunami caused by the March 11th earthquake struck the coast of Northeast Japan and caused significant damage.

The seriously damaged regions around the city of Ishinomaki, Miyagi prefecture are selected for study. Most of the houses were completely washed away. The building damage map is shown in Fig. 1. During the ALOS lifetime, it has acquired five PolSAR datasets covering the study area. The acquisition information is summarized in Table I. There are four before-tsunami and one after-tsunami full polarimetric acquisitions.

Fig. 1. Damage map. Copyright 2011 ZENRIN CO., LTD.

TABLE I Acquisition Information of the Multi-temporal

ALOS/PALSAR Full Polarimetric Image Acquisitions Over Ishinomaki City, Miyagi. The Corresponding Local Weather Information is Cited from Japan Meteorological Agency.

Image

No. Acquisition

date Incidence

angle (deg) Wind speed

(m/s) Temperature

(℃) D1 April 8th, 2011 23.832 2.8 10.5 D2 November 21st, 2010 23.796 2.1 14.0 D3 April 2nd, 2009 23.774 5.8 4.6 D4 May 13th, 2007 23.773 7.5 13.9 D5 March 28th, 2007 23.780 1.3 5.8

The incidence angle is at the imaging scene center.

4. Urban Damage Evaluation

The results of model-based decomposition and estimated PO angle are shown in Figs. 2-3. Comparing the before- and

after-tsunami results, the scattering mechanism changes due to the damage are obvious. The resolution of this PolSAR data is not fine enough to identify each building. Therefore, the scattering mechanism changes are carried out at the city block scale. Nine built-up patches with dense buildings inside but different damage levels are delineated. These patches are highlighted in Figs. 2-3. The damage level is defined as the percentage of the buildings which were flushed away in one built-up patch and are calculated from the damage map. The damage level is divided into four groups: 80-100% (patch 1), 50-80% (patches 2 and 3), 20-50% (patches 4-6) and 0%-20% (patches 7-9). Higher damage level relates to a more seriously damaged area.

N

9

6

1

2

5

4

3

8

7

(a) (b) (c)

Fig. 2. Decomposition results. (a) D5, (b) D2, and (c) D1.

N

9

6

1

2

5

4

3

8

7

(a) (b) (c)

Fig. 3. Estimated PO angle. (a) D5, (b) D2, and (c) D1.

The damage level is determined by the fraction of flushed away buildings in a local built-up patch. This reduction in the number of intact buildings may produce an equivalent reduction in the amount of ground-wall dihedral structures. Since the double-bounce scattering in urban areas directly relates to the ground-wall dihedral structures, the ratio of the double-bounce scattering-dominant contributions after- and before-tsunami can reflect the same decreasing trend. Thereby, the ratio of the double-bounce scattering-dominant contributions after- and before-tsunami is proposed as the first index for damage level assessment

( )

( )( )

( )( )

D ,D D ,

D ,

DominantRatio

Dominantd n i

n m id m i

P

P− = (3)

where i is the built-up patch number and 1,2, ,9i = . D Dn m− is one multi-temporal pair. Ratio values from after- and before-tsunami pairs are shown in Fig. 4(a). The temporal baselines vary from 138 days to

1472 days, while the spatial baselines vary from 1747 m to 4680 m. For all these configurations, the linear relationship between this ratio index and the damage level is clear: the ratio drops with increasing degree of damage.

PO angle links to the orientations of buildings. At block scale, buildings usually have similar orientations. However, after the tsunami, some buildings were flushed away, leaving a number of building foundations and much debris. Therefore, the homogeneity of the PO angle in a block is reduced with increasing damage level. The standard deviation of the differences of two sorted PO angle sequences is proposed as another damage indicator

( ) { }( ) { }( )( )D D , D , D ,Std p qn m i n i m i

std θ θ− = − (4)

where { }( )D ,p n iθ is the PO angle sequence sorted in descending order. ( )std ⋅ is used to obtain the standard deviation.

The index values from after- and before-tsunami pairs for each selected built-up patch are shown in Fig. 4(b). For all these various temporal and spatial baseline combinations, the trend is clear that with increasing damage level the ( )D D ,Std n m i− values

also increase accordingly. The standard deviation is around or over 10 for patch 1 and is within 4 - 8 for patches 2 and 3. For less damaged patches 4-6, the standard deviation is within 2 - 4 , while for patches 7-9, it is below 2 . Therefore, at the block scale, comparisons of the PO angle distributions provide valuable information for understanding damage conditions.

1 2 3 4 5 6 7 8 90

0.2

0.5

0.8

1

D1-D2 D1-D3 D1-D4 D1-D5

1 2 3 4 5 6 7 8 9

0

2

4

6

8

10

12

D1-D2 D1-D3 D1-D4 D1-D5

(a) (b)

Fig. 4. Damage evaluation. (a) Ratios of the double-bounce scattering-dominant contributions, (b) Standard deviations (in deg) of PO angle differences.

For further validation, these two index parameters from before-tsunami pairs are also compared. For before-tsunami pairs, the values of ( )D D ,Ratio n m i− are always around 1, while the values of ( )D D ,Std n m i− are kept below 2 These investigations confirm the efficiency and the stability of the these two parameters. More detailed results will be presented in the symposium.

5. Conclusion

This paper focuses on tsunami damage investigation over urban areas by exploring the multi-temporal PolSAR data. It is

aimed at providing an understanding of the natural disaster from the viewpoint of radar polarimetry. The polarimetric scattering mechanism changes before- and after-tsunami have been examined using model-based decomposition and PO angle techniques. Two polarimetric indexes have been proposed for damage level indication. Experimental results validate the efficiency of these two indicators, since the built-up areas with different damage levels could be well discriminated. These results demonstrate the importance of full polarimetric information for damage assessment, which could be complementary to other remote sensing techniques.

6. Acknowledgments

The authors would like to thank JAXA for providing the ALOS data. This work was supported in part by the National

Natural Science Foundation of China under Grant 41301490.

7. References

1. M. Pesaresi, A. Gerhardinger, and F. Haag, “Rapid damage assessment of built-up structures using VHR satellite data in tsunami-affected areas,” Int. J. Remote Sens., vol. 28, no. 13/14, pp. 3013–3036, Jun. 2007.

2. F. Bovolo and L. Bruzzone, “A split-based approach to unsupervised change detection in large-size multitemporal

images: Application to tsunami-damage assessment,” IEEE Trans. Geosci. Remote Sens., vol. 45, pp. 1658–1670, Jun. 2007.

3. D. Brunner, G. Lemoine, and L. Bruzzone, “Earthquake damage assessment of buildings using VHR optical and SAR

imagery,” IEEE Trans. Geosci. Remote Sens., vol. 48, pp. 2403–2420, May 2010.

4. P. Gamba, F. Dell'Acqua, and G. Trianni, “Rapid damage detection in the Bam area using multitemporal SAR and exploiting ancillary data,” IEEE Trans. Geosci. Remote Sens., vol. 45, pp. 1582–1589, Jun. 2007.

5. M. Sato, S. W. Chen and M. Satake, “Polarimetric SAR analysis of tsunami damage following the March 11, 2011 East

Japan earthquake,” Proc. IEEE, vol. 100, no. 10, pp. 2861–2875, Mar. 2012.

6. S. W. Chen and M. Sato, “Tsunami damage investigation of built-up areas using multi-temporal spaceborne full polarimetric SAR images,” IEEE Trans. Geosci. Remote Sens., vol. 51, no. 4, pp. 1985–1997, Apr. 2013.

7. J. S. Lee and E. Pottier, Polarimetric Radar Imaging: From Basics to Applications, Boca Raton, US: CRC Press, 2009.

8. S. W. Chen, X. S. Wang, Y. Z. Li and M. Sato, “Adaptive model-based polarimetric decomposition using PolInSAR coherence,” IEEE Trans. Geosci. Remote Sens., vol. 52, no. 3, pp. 1705–1718, Mar. 2014.

9. S. W. Chen, X. S. Wang, S. P. Xiao and M. Sato, “General polarimetric model-based decomposition for coherency matrix,” IEEE Trans. Geosci. Remote Sens., vol. 52, no. 3, pp. 1843–1855, Mar. 2014.

10. S. W. Chen, M. Ohki, M. Shimada, and M. Sato, “Deorientation effect investigation for model-based decomposition over oriented built-up areas,” IEEE Geosci. Remote Sens. Lett., vol. 10, no. 2, pp. 273–277, Mar. 2013.

11. J. S. Lee, D. L. Schuler, T. L. Ainsworth, E. Krogager, D. Kasilingam, and W. M. Boerner, “On the estimation of radar polarization orientation shifts induced by terrain slopes,” IEEE Trans. Geosci. Remote Sens., vol. 40, pp. 30–41, Jan. 2002.