Background Information
Japan is an island nation surrounded by oceans. In order to protect inland area from coastal disasters caused by sea breezes, sand blow high tides, and tsunamis, coastal forests are developed in large area. In Tohoku Region, a large area of coastal forests are suffered the damage by 2011 Tohoku Tsunami. According to a report by the Ministry of Agriculture, Forestry and Fisheries, Japan, the total area of coastal forest in Tohoku region suffered the inundation is 2800ha and 40% (1050ha) of forest area are considered to be damaged over 75% of area. According to Kamthonkiat et al., NDVI fluctuation is effective for monitoring changes in mangroves during the pre- and post- tsunami periods.Questions to be answered
How to detect the distribution of the damage on coastal forest and where and what type of vegetation suffered most by tsunami? (area/vegetation type/topography)Study Area
The study area is coastal area along Sanirku coast from Rikuzentakata-city, Iwate prefecture to Onagawa-cho, Miyagi prefecture in Tohoku region, Japan. The total length of the coastline is about 338km. There are some areas developed as artificial coastal forests for preventing coastal disasters including salt damage, high tides, and tsunamis. (Appendix Figure 1)Data and data collection
- Pre-Tsunami ASTER VNIR 3A01 (5/5/2010) (spatial resolution: 15m)
- Post-Tsunami ASTER VNIR 3A01 (4/13/2011) (spatial resolution: 15m)
- DEM (Spatial Resolution: 10m) made by the Geospatial Information Authority of Japan (http://fgd.gsi.go.jp/download/) (in Japanese)
- GIS data obtained from National Surveys on the Natural Environment (surveys on vegetation, rivers, lakes/marshes, coastlines, seaweed beds/tidal flats/coral reefs, etc.) by the Biodiversity Center of Japan (http://www.biodic.go.jp/trialSystem/top_en.html)
Both ASTER VNIR products are orthorectified. However, since both products are not atmospherically corrected, only the DNs are available for analysis in this study.
Remote sensing methods
1. Image normalization using Radiometric Control Set (RCS) MethodSince the remote sensing images were taken in different month, it is necessary to normalize the reflectance of images. In this study, the images of two dates are normalized using RCS method, which is same to Multiple Date Image Normalization Using Regression.
2. Calculate NDVI and the Simple Ratio Red/NIR ratio
In order to grasp the distribution and degree of damage, multiple different indexes are calculated and discussed.
3. Calculate the change
The change ratio of NDVI and Simple Ratio are calculated by the following equation.
(NDVI change ratio) = Log2((NDVI of Pre-Tsunami) / (NDVI of Post-Tsunami))
(SR change ratio) = Log2((SR of Pre-Tsunami) / (SR of Post-Tsunami))
4. Categorize the ratio by setting thresholds of the damage by sampling pixel value in area in which degrees of damage were known by field survey
1. Damage ratio > 75% (almost snapped and washed away)
2. Damage ratio > 25% (partially snapped and washed away)
Based on field surveys conducted in 2011 and manual detection of aerial photos, two areas for each class are selected and the mean values of change ratios are used as thresholds of each class. For the class “Damage ratio > 75%”, pixel values in forests of Koizumi and Rikuzentakata are sampled and the mean value is used as the threshold of the class. For the class “Damage ratio > 25%”, pixel values in forests of Oya coast and Oshima are sampled and the mean value is used as the threshold of the class.[appendix] In the next step, each image of change ratio are reclassified to the classes. In addition, based on the GIS data of vegetation survey, the pixels in “forests” located within 200m from coastline are extracted, visualized and the total areas of classes are calculated. (Appendix Figure 3, 4, 5)
5. Supervised classification of the images and change detection
In order to discuss the distribution of the damage in types of vegetation, supervised classification is conducted and compared with pre and post tsunami. The classes are consists of 1: Urban/Barren, 2: Forests, 3: Agriculture/Rangelands, 4: Water. For the coastal area that located within 200m from coastline, the confusion matrix is created.
Results
1. NIR/Red Simple Ratio and the Change Ratio
Simple Ratio
| Class | Pixel Count | Area(ha) | % |
|---|---|---|---|
| less than 25% | 42275 | 951.1875 | 86.07% |
| 25%~75% | 6338 | 142.605 | 12.90% |
| and over 75% | 503 | 11.3175 | 1.02% |
| sum | 49116 | 1105.11 |
In the 1105 ha of total area (49116 pixels) extracted from the procedure, 11 ha (503 pixels) of area is classified as “Damage ratio > 75%” and 142 ha of area is classified as “Damage ratio > 25%”.
2. NDVI and the Change Ratio
NDVI
| Class | Pixel Count | Area(ha) | % |
|---|---|---|---|
| less than 25% | 41856 | 941.76 | 85.22% |
| 25%~75% | 6881 | 154.8225 | 14.01% |
| and over 75% | 379 | 8.5275 | 0.77% |
| sum | 49116 | 1105.11 |
In the 1105 ha of total area extracted from the procedure, about 9 ha of area is classified as “Damage ratio > 75%” and 155 ha of area is classified as “Damage ratio > 25%”.
3. The distribution of the damage
| Height (m) | Damage ratio < 25% (ha) | Damage Ratio > 25% (ha) | Damage ratio > 75% (ha) |
|---|---|---|---|
| 0 - 2.5 | 61.8525 | 32.49 | 7.29 |
| 2.5 - 6.6 | 80.6625 | 22.455 | 1.1025 |
| 6.6 - 10.2 | 83.5875 | 21.5325 | 0 |
| 10.2 - 15.5 | 91.5525 | 15.8625 | 0 |
| 15.5 - 20.4 | 95.355 | 12.8475 | 0 |
| 20.4 - 24.8 | 98.9775 | 9.0675 | 0 |
| 24.8 - 30.3 | 99.3825 | 8.235 | 0 |
| 30.3 - 36.2 | 98.3025 | 9.225 | 0 |
| 36.2 - 45.8 | 94.8825 | 7.6725 | 0 |
| 45.8 - 120.2 | 94.0725 | 6.8175 | 0 |
Height is classified into 10 classes based on quantile classification of height values of each pixel in coastal forests within 200m from coastline (Appendix figure 6). Obviously the lowland areas suffered more severe damage than other area. In addition, coasts facing the epicenter (southwest from study area) suffered more severe damage.
4. Supervised classification of the images and change detection
| (Post)Urban/Barren (ha) | (Post)Forest (ha) | (Post)Agricultrue/Grass (ha) | (Post)Water (ha) | (Post)sum (ha) | |
|---|---|---|---|---|---|
| (Pre)Urban/Barren (ha) | 73.5075 | 96.39 | 28.53 | 23.8725 | 222.3 |
| (Pre)Forest (ha) | 30.3525 | 506.4525 | 17.01 | 16.155 | 569.97 |
| (Pre)Agriculture/Grass (ha) | 40.5675 | 114.1875 | 40.0275 | 29.7675 | 224.55 |
| (Pre)Water (ha) | 7.3575 | 17.2125 | 10.0575 | 26.865 | 61.4925 |
| sum (ha) | 151.785 | 734.2425 | 95.625 | 96.66 | 1078.3125 |
Within the area in the forest features assigned by vegetation survey, 58% area is actually classified as forest in the pre-tsunami image. Overall, comparing pre-tsunami and post-tsunami images, total vegetation area (forest + agriculture/grass) was increased by 35ha, from 794ha to 829ha. The total forest area was increased by 164ha, from 570ha to 734ha. The total water area was increased by 35ha. (Appendix Figure 9)
Discussions
1. Advantages and disadvantages of using NDVI and Simple RatioCompared to the simple ratio, NDVI compresses the range of values in vegetation and expands the range of values in urban and water area. Since this study compares the change ratio of each indices and focuses coastal forest, the result of the Simple Ratio has wider histogram (appendix) of pixel values and extracts more pixels with severer damage ratios. In addition, atmospheric corrections of images are required for future work in order to acquire more accurate indices values.
2. Supervised classification of the images and change detection
In the area classified as forest in the pre-tsunami image, 16ha of area is classified as water in the post-tsunami image, mostly supporting the extraction results from the Simple Ratio and NDVI change ratio methods. However, considering the classification results, because the FROM-TO information is confused, the result does not provide more information.
Conclusion
Based on the NDVI and SR change ratio, in the coastal forests in the study area, more than 8.5ha of area is considered to be suffered 75% of the forest. Similarly, based on the supervised classification method, about 16ha of the coastal forests in the study area were completely inundated.References:
Kamthonkiat, D., Rodfai, C., Saiwanrungkul, A., Koshimura, S., & Matsuoka, M. (2011). Geoinformatics in mangrove monitoring: damage and recovery after the 2004 Indian Ocean tsunami in Phang Nga, Thailand. Natural Hazards and Earth System Science, 11(7), 1851–1862. Retrieved from http://www.nat-hazards-earth-syst-sci.net/11/1851/2011/
Appendix
Figure 1. Study Area
Figure 3. Sampling area for Damage Ratio > 75% in Rikuzentakata and Koizumi
Figure 4. Sampling area for Damage Ratio > 25% in Oshima
Figure 5. Sampling area for Damage Ratio > 25% in Oya coast
Figure 6. Classification of height values based on quantiles in ArcGIS tool
Figure 7. Distribution of NDVI change ratio (full view of study area)
Figure 8. Distribution of Simple Ratio Change Ratio (full view of study area)
Figure 9. Comparison of Pre-Tsunami and Post-Tsunami Classification Results
