**2. General satellite image analysis for tsunami-affected areas**

#### **2.1 NDVI analysis using optical high-resolution satellite imagery**

#### *Tsunami inundation limit*

Recent advances in remote sensing technologies have expanded the capabilities of detecting the spatial extent of tsunami-affected areas and damage to structures. The highest spatial resolution of optical imageries from commercial satellites is up to 60–70 centimetres (QuickBird owned by DigitalGlobe, Inc.) or 1 metre (IKONOS operated by GeoEye). Since the 2004 Sumatra-Andaman earthquake tsunami, these satellites have captured images of tsunami-affected areas, and the images have been used for disaster management activities, including emergency response and recovery. To detect the extent of a tsunami inundation zone, NDVI (Normalised Difference Vegetation Index) is the most common index obtained from the post-event imagery, focusing on the vegetation change due to the tsunami penetration on land. The NDVI is calculated from these individual measurements as follows:

$$NDVI = \frac{NIR - R}{NIR + R} \tag{1}$$

where *R* and *NIR* stand for the spectral reflectance or radiance in the visible (red) and nearinfrared bands, respectively. Focusing on the existence of tsunami debris, 100 points were sampled to identify the NDVI threshold to classify the tsunami inundation zone. As shown in Fig. 1, the NDVI values are calculated within a range 0.34 ± 0.05. As a result, the extent of the tsunami inundation zone is determined by the supervised classification based on the NDVI threshold. As shown in Figure 2 (a), the QuickBird imagery clearly detects the vegetation change between pre- and post-tsunami. Tsunami debris can be seen along the edge of the tsunami inundation zone. Figure 2 (b) shows the result of the detection of the tsunami inundation zone by applying the threshold value of NDVI, and the result is consistent with the field survey.

Fig. 1. Threshold value of NDVI within the tsunami inundation zone obtained from the analysis of the post-tsunami satellite imagery.

Recent advances in remote sensing technologies have expanded the capabilities of detecting the spatial extent of tsunami-affected areas and damage to structures. The highest spatial resolution of optical imageries from commercial satellites is up to 60–70 centimetres (QuickBird owned by DigitalGlobe, Inc.) or 1 metre (IKONOS operated by GeoEye). Since the 2004 Sumatra-Andaman earthquake tsunami, these satellites have captured images of tsunami-affected areas, and the images have been used for disaster management activities, including emergency response and recovery. To detect the extent of a tsunami inundation zone, NDVI (Normalised Difference Vegetation Index) is the most common index obtained from the post-event imagery, focusing on the vegetation change due to the tsunami penetration on land. The NDVI is calculated from these individual measurements as

ൌ ܫܸܦܰ

ܴ െ ܴܫܰ ܴ ܴܫܰ

where *R* and *NIR* stand for the spectral reflectance or radiance in the visible (red) and nearinfrared bands, respectively. Focusing on the existence of tsunami debris, 100 points were sampled to identify the NDVI threshold to classify the tsunami inundation zone. As shown in Fig. 1, the NDVI values are calculated within a range 0.34 ± 0.05. As a result, the extent of the tsunami inundation zone is determined by the supervised classification based on the NDVI threshold. As shown in Figure 2 (a), the QuickBird imagery clearly detects the vegetation change between pre- and post-tsunami. Tsunami debris can be seen along the edge of the tsunami inundation zone. Figure 2 (b) shows the result of the detection of the tsunami inundation zone by applying the threshold value of NDVI, and the result is

Fig. 1. Threshold value of NDVI within the tsunami inundation zone obtained from the

(1)

**2. General satellite image analysis for tsunami-affected areas** 

**2.1 NDVI analysis using optical high-resolution satellite imagery** 

*Tsunami inundation limit* 

consistent with the field survey.

analysis of the post-tsunami satellite imagery.

follows:

Fig. 2. (a) Vegetation change found from pre- and (b) post-event imageries and estimated extent of tsunami inundation zone by the supervised classification of NDVI.

#### *Damage and recovery monitoring of mangrove*

Because mitigation and protection against the 2004 Indian Ocean Tsunami was one of the important services that mangrove ecosystems provided in the affected areas, a six-year program to conserve and rehabilitate mangrove forests in the tsunami-impacted areas was implemented by the Thai Government after the tsunami. However, information on mangrove restoration and reforestation is limited to field surveys. Monitoring proposals were applied for a damaged mangrove area. Kamthonkiat et al. (2011) used ASTER images acquired in 2003, 2005 (two months after the 2004 Indian Ocean tsunami), 2006 and 2010 and the analysis using NDVI to monitor the mangrove recovery in tsunami-impacted areas in the southern part of Thailand. Figure 3 depicts the area of mangroves in 2003 in red and the area impacted by the tsunami in 2005 in dark blue and white for the same location. After the mangrove trees were uniformly or homogeneously replanted in the same location in the last quarter of 2005 in Takuapa District, the areas marked in red increased in 2006 and increased still further in 2010, as shown in Fig. 3 (*Note*: red represents vegetation or mangroves, white represents bare soil/sand, and blue/dark blue represent water). The recovery process can be detected, as some parts in light blue became red in 2006, and most became red in 2010 meaning the mangroves recovered to nearly the normal condition before the tsunami attack. These results show the abilities of geoinformatic technologies, especially regarding the time series analysis.

Fig. 3. Damage and recovery process due to the 2004 tsunami in Takuapa, Thailand

Application of Remote Sensing for Tsunami Disaster 147

For the areas where some trees in the forest were washed away, the significant backscattering characteristics changed from volume scattering to surface scattering with significant roughness. These kinds of characteristics affecting the backscattering echo were identified in the tsunami-affected areas in the TerraSAR-X image. Following Nojima et al. (2006), the regression discriminant function for building damage was calculated from two characteristic values, the correlation coefficient and the difference in backscattering coefficient for pre- and post-event SAR images. First, following the accurate positioning of the two SAR images, a speckle noise filter with a 21×21 pixel window (Lee, 1980) was applied to each image. The difference value, *d*, is calculated by subtracting the average value of the backscattering coefficient within a 13×13 pixel window in the pre-event image from the post-event image (after – before). The correlation coefficient, *r*, is also calculated from the same 13×13 pixel window (Matsuoka & Yamazaki, 2004). The result of applying regression discriminant analysis, using the *d* and

 *ZR1* = -A·*d* – B·*r* (2) Here, *ZR1* represents the discriminant score from the SAR images where the values of parameter A and B are 1.21 and 4.36, respectively. The pixels whose *ZR1* value is positive (red) are interpreted as suffering severe damage (Fig. 5 left). Because both coefficients are negative, higher and negative *d* or smaller *r* produce larger *ZR1* values. A preliminary formula for the C-band dataset was used because that for the X-band was unavailable. For this reason, the backscattered echoes were stronger in the post-tsunami image. To detect such damaged areas using image analysis, cases where the reverse occurs need to be considered. Therefore, the following Equation (3) was also calculated based on a positive

 *ZR2* = A·*d* – B·*r* (3) Here, *ZR2* represents another discriminant score where the values of parameters A and B are 1.21 and 4.36, respectively. Using this formula, the pixels whose *ZR2* value is positive (red)

Fig. 5. Computed *ZR1* and *ZR2* from TerraSAR-X image to determine inundation area

*r*, is shown in Equation (2).

value for the difference in backscattering coefficient *d*.

might be assigned as damaged areas (Fig. 5 centre).

#### **2.2 TerraSAR-X image analysis**

Among the various sensors, SAR (Synthetic Aperture Radar) is remarkable for its ability to record the physical value of the Earth's surface (Henderson and Lewis, 1998). Unlike passive optical sensors, SAR enables the observation of surface conditions day or night, even through clouds. SAR interferometric analyses using phase information have successfully provided quantification of relative ground displacement levels due to natural disasters (Massonnet et al., 1993). More importantly, intensity information obtained from SAR represents a physical value (backscattering coefficient) that is strongly dependent on the roughness of the ground surface and the dielectric constant. Based on this idea, models for satellite C- and L-band SAR data were developed to detect building damage areas due to earthquakes by clarifying the relationship between the change in the backscattering coefficient from pre- and post-event SAR images (Matsuoka & Yamazaki, 2004; Matsuoka & Nojima, 2009) and then applying the models to tsunami-induced damage areas (Koshimura & Matsuoka, 2010). TerraSAR-X, which is the first German radar satellite with highresolution X-band, was successfully launched on June 15, 2007, and has been in operation for data acquisition since early 2008. The day after the event, TerraSAR-X observed the coastal area in the affected regions by the StripMap-mode, which captures the Earth's surface with an approximately 3-metre resolution. Typically, man-made structures show comparatively high reflection due to the cardinal effect of structures and the ground. Open spaces or damaged buildings have comparatively low reflectance because they scatter the microwaves in different directions. Buildings may be reduced to debris by earthquake ground motion, and in some cases, the debris of buildings may be removed, leaving the ground exposed. Thus, the backscattering coefficient determined after building collapse is likely to be lower than that obtained prior to the event (Matsuoka & Yamazaki, 2004; Nojima et al., 2006). Inundated areas also show a lower backscattering coefficient because of the smooth surface and the dielectric constant of water bodies (Fig. 4 centre). By examining the backscattering characteristics of tsunami damage in typical areas, however, the reverse case occurred in some damaged areas in farmlands and controlled forests. To explain these anomalies in the post-tsunami TerraSAR-X image, several factors need to be considered, such as changes of the Earth's surface and its materials. Scattered debris from collapsed buildings, visible in the farmlands and bare ground in the post-tsunami image, show brighter reflections than in the pre-tsunami image (Fig. 4 centre).

Fig. 4. Comparison between TerraSAR-X image and IKONOS (GeoEye) image

Among the various sensors, SAR (Synthetic Aperture Radar) is remarkable for its ability to record the physical value of the Earth's surface (Henderson and Lewis, 1998). Unlike passive optical sensors, SAR enables the observation of surface conditions day or night, even through clouds. SAR interferometric analyses using phase information have successfully provided quantification of relative ground displacement levels due to natural disasters (Massonnet et al., 1993). More importantly, intensity information obtained from SAR represents a physical value (backscattering coefficient) that is strongly dependent on the roughness of the ground surface and the dielectric constant. Based on this idea, models for satellite C- and L-band SAR data were developed to detect building damage areas due to earthquakes by clarifying the relationship between the change in the backscattering coefficient from pre- and post-event SAR images (Matsuoka & Yamazaki, 2004; Matsuoka & Nojima, 2009) and then applying the models to tsunami-induced damage areas (Koshimura & Matsuoka, 2010). TerraSAR-X, which is the first German radar satellite with highresolution X-band, was successfully launched on June 15, 2007, and has been in operation for data acquisition since early 2008. The day after the event, TerraSAR-X observed the coastal area in the affected regions by the StripMap-mode, which captures the Earth's surface with an approximately 3-metre resolution. Typically, man-made structures show comparatively high reflection due to the cardinal effect of structures and the ground. Open spaces or damaged buildings have comparatively low reflectance because they scatter the microwaves in different directions. Buildings may be reduced to debris by earthquake ground motion, and in some cases, the debris of buildings may be removed, leaving the ground exposed. Thus, the backscattering coefficient determined after building collapse is likely to be lower than that obtained prior to the event (Matsuoka & Yamazaki, 2004; Nojima et al., 2006). Inundated areas also show a lower backscattering coefficient because of the smooth surface and the dielectric constant of water bodies (Fig. 4 centre). By examining the backscattering characteristics of tsunami damage in typical areas, however, the reverse case occurred in some damaged areas in farmlands and controlled forests. To explain these anomalies in the post-tsunami TerraSAR-X image, several factors need to be considered, such as changes of the Earth's surface and its materials. Scattered debris from collapsed buildings, visible in the farmlands and bare ground in the post-tsunami image, show

brighter reflections than in the pre-tsunami image (Fig. 4 centre).

Fig. 4. Comparison between TerraSAR-X image and IKONOS (GeoEye) image

**2.2 TerraSAR-X image analysis** 

For the areas where some trees in the forest were washed away, the significant backscattering characteristics changed from volume scattering to surface scattering with significant roughness. These kinds of characteristics affecting the backscattering echo were identified in the tsunami-affected areas in the TerraSAR-X image. Following Nojima et al. (2006), the regression discriminant function for building damage was calculated from two characteristic values, the correlation coefficient and the difference in backscattering coefficient for pre- and post-event SAR images. First, following the accurate positioning of the two SAR images, a speckle noise filter with a 21×21 pixel window (Lee, 1980) was applied to each image. The difference value, *d*, is calculated by subtracting the average value of the backscattering coefficient within a 13×13 pixel window in the pre-event image from the post-event image (after – before). The correlation coefficient, *r*, is also calculated from the same 13×13 pixel window (Matsuoka & Yamazaki, 2004). The result of applying regression discriminant analysis, using the *d* and *r*, is shown in Equation (2).

$$\mathbf{Z}\_{R1} = \mathbf{-A} \cdot \mathbf{d} - \mathbf{B} \cdot \mathbf{r} \tag{2}$$

Here, *ZR1* represents the discriminant score from the SAR images where the values of parameter A and B are 1.21 and 4.36, respectively. The pixels whose *ZR1* value is positive (red) are interpreted as suffering severe damage (Fig. 5 left). Because both coefficients are negative, higher and negative *d* or smaller *r* produce larger *ZR1* values. A preliminary formula for the C-band dataset was used because that for the X-band was unavailable. For this reason, the backscattered echoes were stronger in the post-tsunami image. To detect such damaged areas using image analysis, cases where the reverse occurs need to be considered. Therefore, the following Equation (3) was also calculated based on a positive value for the difference in backscattering coefficient *d*.

$$Z\_{R2} = \mathbf{A} \cdot \mathbf{d} - \mathbf{B} \cdot \mathbf{r} \tag{3}$$

Here, *ZR2* represents another discriminant score where the values of parameters A and B are 1.21 and 4.36, respectively. Using this formula, the pixels whose *ZR2* value is positive (red) might be assigned as damaged areas (Fig. 5 centre).

Fig. 5. Computed *ZR1* and *ZR2* from TerraSAR-X image to determine inundation area

Application of Remote Sensing for Tsunami Disaster 149

Fig. 7. Example of building damage classification criteria for the 2009 Samoa tsunami

In 1993, a tsunami accompanied by a M7.8 earthquake off the south–west coast of Hokkaido, Japan, struck Okushiri Island, which is 30 kilometers west of Hokkaido, within 5 minutes after the quake, causing more than 200 casualties. In particular, the Aonae district in the southernmost area of Okushiri Island suffered devastating damage due to an approximately 11-m tsunami that struck from the west coast of the island as well as fire caused during and after the tsunami attack (Murosaki, 1994). Visual damage inspection was conducted using pre- and post–tsunami aerial photographs acquired on 29 October 1990 and 14 July 1993 (one day after the event occurred), as shown in Fig. 8. Because the Aonae district suffered from extensive fire during and after the tsunami attack, it is not possible to discriminate between tsunami and fire damage by the aerial photographs alone. Thus, focusing on the existence of house roofs, the structural damage was categorised into five classes according to the damage area, whether flooded or burned, reported by Shuto (2007). The number of inspected houses and structures was 769, and the result of the structural damage interpretation in Aonae district is shown in Table 1 (Koshimura et al., 2009a). The method to detect the damaged area using SAR image analysis was applied to the tsunami-affected area in Okushri Island. Using a set of pre- and post–tsunami SAR images acquired by JERS (Japanese Earth Resources Satellite), Matsuoka & Yamazaki (2002) calculated the correlation and difference in the backscattering coefficient to represent the changes in the tsunamiaffected area. To detect the impacted area, the discriminant score, Equation (2), was

**3.1 The 1993 Hokkaido Nansei-Oki tsunami** 

Two discriminant scores, *ZR1* and *ZR2*, were calculated for the TerraSAR-X image pair using the described procedure. The threshold values for Z*R1* and *ZR2* were determined to be 6 and 0, respectively. The extracted areas where the *ZR1* is larger than 6 or the *ZR2* is larger than 0 are shown in red in Fig. 6.

Fig. 6. Threshold values for *ZR1* and *ZR2* in the case of the 2011 Tohoku tsunami in Miyagi prefecture. The tsunami inundation area was extracted when the *ZR1* was larger than 6 or the *ZR2* was larger than 0.

#### **3. Tsunami damage detection and classification by remote sensing**

This section mainly focuses on how remote sensing is used for further research on the detailed classification of tsunami damage areas using structural damage as an example. By taking advantage of satellite remote sensing, the spatial distribution of structural damage by a tsunami can be identified. SAR images are widely used to determine tsunami-affected or inundated areas using the reflection property or backscattering coefficient as mentioned in the previous section. However, through inspecting a set of pre- and post-tsunami satellite images visually or manually, the presence of building roofs can be interpreted. The highest spatial resolution of commercial optical satellite imaging is up to 60-70 cm (QuickBird) or 1 m (IKONOS). The advantage of using high-resolution optical satellite images for damage interpretation is the capability of understanding structural damage visually. These images also enable us to comprehend the spatial extent of damage at the regional scale, where posttsunami surveys hardly penetrate because of limited of survey time and resources. However, note that no structural types were identified by the interpretation of the satellite images. Additionally, the damage feature that can be identified from the satellite images is only structural destruction or major structural failure, which reveals the change of a roof's shape, namely "collapsed" and "major or severe damage." Accordingly, the interpretation "Destroyed" means "Collapsed" or "Major or severe damage," and "Survived" is classified as "Moderate," "Minor," "Slight" and "No" damage. An example of building damage classification is shown in Fig. 7.

Two discriminant scores, *ZR1* and *ZR2*, were calculated for the TerraSAR-X image pair using the described procedure. The threshold values for Z*R1* and *ZR2* were determined to be 6 and 0, respectively. The extracted areas where the *ZR1* is larger than 6 or the *ZR2* is larger than 0

Fig. 6. Threshold values for *ZR1* and *ZR2* in the case of the 2011 Tohoku tsunami in Miyagi prefecture. The tsunami inundation area was extracted when the *ZR1* was larger than 6 or the

This section mainly focuses on how remote sensing is used for further research on the detailed classification of tsunami damage areas using structural damage as an example. By taking advantage of satellite remote sensing, the spatial distribution of structural damage by a tsunami can be identified. SAR images are widely used to determine tsunami-affected or inundated areas using the reflection property or backscattering coefficient as mentioned in the previous section. However, through inspecting a set of pre- and post-tsunami satellite images visually or manually, the presence of building roofs can be interpreted. The highest spatial resolution of commercial optical satellite imaging is up to 60-70 cm (QuickBird) or 1 m (IKONOS). The advantage of using high-resolution optical satellite images for damage interpretation is the capability of understanding structural damage visually. These images also enable us to comprehend the spatial extent of damage at the regional scale, where posttsunami surveys hardly penetrate because of limited of survey time and resources. However, note that no structural types were identified by the interpretation of the satellite images. Additionally, the damage feature that can be identified from the satellite images is only structural destruction or major structural failure, which reveals the change of a roof's shape, namely "collapsed" and "major or severe damage." Accordingly, the interpretation "Destroyed" means "Collapsed" or "Major or severe damage," and "Survived" is classified as "Moderate," "Minor," "Slight" and "No" damage. An example of building damage

**3. Tsunami damage detection and classification by remote sensing** 

are shown in red in Fig. 6.

*ZR2* was larger than 0.

classification is shown in Fig. 7.


Fig. 7. Example of building damage classification criteria for the 2009 Samoa tsunami
