# Rapid flood and damage mapping using synthetic aperture radar in response to Typhoon Hagibis, Japan

## Abstract

During the aftermath of Typhoon Hagibis, we made flood and damage proxy maps, rapidly derived from synthetic aperture radar (SAR) data using change detection approaches. The maps have large spatial coverage over the Tokyo, Fukushima, Ibaraki, Iwate, and Nagano prefectures of Japan. The maps are also largely in agreement with various validation sources including aerial imagery, optical imagery and news sources. Apart from visual maps, we provide flood and damage extents in various formats compatible with geographic information system (GIS) applications. The data may potentially be used for applications such as typhoon risk modelling, investigating spatial correlations of typhoon impacts, and comparing alternative flood or damage mapping techniques.

 Measurement(s) damage • flood Technology Type(s) synthetic aperture radar Factor Type(s) geographic location Sample Characteristic - Environment cyclone Sample Characteristic - Location Japan

Machine-accessible metadata file describing the reported data: https://doi.org/10.6084/m9.figshare.11968443

## Background & Summary

Typhoon Habigis made landfall at the Izu Peninsula of Shizuoka Prefecture, Japan on 12 October 2019, UTC 08:30 as a Category 2-equivalent on the Saffir Simpson scale1. It was one of the strongest typhoons of the century to hit Japan, and had one of the fastest intensifications from a Category 1 to 5 status in less than a day. Reports reveal widespread impacts and at least 80 deaths, 135,000 people affected, 68,000 houses inundated and 10,000 buildings damaged1,2.

In response to the event, we mapped flood and damage extents across Japan over a week after the typhoon’s landfall to potentially support stakeholders in the region in assessing impacts and planning disaster relief efforts. The maps are derived from data acquired from SAR, a form of remote sensing. SAR is particularly helpful compared to other modes of remote sensing as its radar transmission in the microwave spectrum is independent of day-and-night visibility and weather conditions, such as persistent rain and cloud cover observed during typhoons. To identify flood and damage from SAR, we used change detection approaches where data acquired before and during an event are compared to detect land surface changes.

With the frequent six-day revisit of the Copernicus Sentinel-1 satellites operated by the European Space Agency, and ALOS-2 tasking efforts coordinated by the Japan Aerospace Exploration Agency and Sentinel Asia, timely SAR data were acquired following Typhoon Hagibis’ landfall. Rapid mapping with SAR data was facilitated by the Advanced Rapid Imaging and Analysis Singapore (ARIA-SG) system at the Earth Observatory of Singapore. This is an automated and scalable cloud-based SAR processing system for near real-time monitoring and rapid response for natural disasters. It is originally cloned from the ARIA system developed by the Jet Propulsion Laboratory and California Institute of Technology. Finally, four flood proxy maps (FPMs) and one damage proxy map (DPM) were disseminated through Sentinel Asia, a network comprising satellite data providers, analysis nodes and government agencies to support disaster management activity in the Asia-Pacific region3.

The paper is organized as follows: (1) principles of the SAR-based change detection techniques are explained in Methods; (2) maps are shown in Data Records; (3) and accuracies and uncertainties of the maps are discussed in Technical Validation for notes on usage of the maps. Due to the large spatial coverage of SAR data, the maps are most useful for providing relative spatial distribution information on the typhoon impacts. Potential usages of the data include typhoon risk modelling, investigation of spatial correlations of typhoon impacts, and comparisons with alternative flood or damage mapping techniques.

## Methods

### Flood proxy map

We obtained pairs of Sentinel-1 and ALOS-2 scenes (Table 1). Each pair consists of a scene acquired before, and another acquired after Typhoon Hagibis’ landfall. We selected each pre-event scene with acquisition time within a period of minimal precipitation, such that a pair can be used to detect floods brought by the typhoon using a change detection approach.

Next, we pre-processed Sentinel-1 single look complex (SLC) images in the ARIA-SG system using the InSAR Scientific Computing Environment (ISCE) processor4. Each co-event scene was co-registered to its corresponding pre-event scene. Co-registration used the National Aeronautics and Space Administration Version 3.0 Shuttle Radar Topography Mission (SRTM) Global 1 arc second digital elevation model (DEM)5 and dense subpixel offsets computed from the cross correlation of SAR amplitudes. All SLCs were then geocoded and resampled to the DEM’s spacing of approximately 30 m. For ALOS-2, we obtained L2.1 amplitude images from Sentinel Asia. These have been orthorectified and geocoded using a DEM.

After pre-processing, we used the amplitude component of SLCs, A to compute the logarithmic amplitude ratio (LAR)6 between each pair of pre- and co-event SLCs (Eq. 1). ALOS-2 LARs were also passed through a median filter to reduce speckle noise7, and resampled to 25 m spacing.

$$LAR{=\log }_{10}\left(\frac{{A}_{co-event}}{{A}_{pre-event}}\right)$$
(1)

Typically, flood results in amplitude decrease or a negative LAR due to the specular surface of open waters as less signal is backscattered. Occasionally, amplitude increase over open water surfaces may be observed due to wind-driven capillary waves. Flood also results in amplitude increase or a positive LAR if the water body is adjacent to a semi-vertical feature, such as vegetation or buildings, due to a double bounce effect8. However, small amplitude changes may be attributed to noise as opposed to significant change due to actual floods.

Thus, we selected thresholds for both negative and positive LAR values to determine suitable extents of change which represent actual flooding. Thresholds were selected in consideration of the topography, and when available, known flooded places from aerial imagery from Geospatial Information Authority of Japan (GSI)9, optical imagery from Planet10 and news sources. For each FPM, all aerial imagery, optical imagery, and news sources used to guide the thresholding were acquired within 24 hours after, 5 hours after, and +/−12 hours of the SAR co-event scene acquisition respectively. All optical imagery used are 4-band PlanetScope Scenes provided in analytical mode, where the 4 bands are blue, red, green and near infrared (NIR), with values represented in digital numbers. The manual selection of thresholds using these various validation data is detailed below in Technical Validation.

Finally, we took the selected thresholds as global thresholds to apply to the whole SAR image extent. Then, all pixels with LAR values below the negative threshold or above the positive threshold were considered as flood, and flood pixels were smoothed to reduce noise. The result of thresholding is a binary image indicating flood or no flood, which was then converted into PNG, JPEG, GeoTIFF, KML and Shapefile formats for dissemination to users.

### Damage proxy map

To map damage with change detection, we obtained three Sentinel-1 SLCs (Table 1), with two scenes acquired before, and one scene acquired after the typhoon’s landfall. We used pre-event scenes which were acquired closest to the co-event scene to minimize the detection of land surface changes which occurred prior to the event. The SLCs were then pre-processed in the ARIA-SG system using ISCE algorithms4 similar to the Sentinel-1 flood mapping, with a resulting spacing of 30 m. For co-registration, we used the pre-event scene closest in time to the co-event scene as the reference.

We used the complex pixel value, c of pre-processed SLCs for the change detection analysis, where damage is inferred from a loss of coherence or decorrelation between SAR images11. Firstly, we computed the pre- and co-event interferometric coherences, γ (Eq. 2) from a pair of SLCs before the event, and another pair spanning across the event respectively.

$$\gamma =\frac{\left|\left\langle {c}_{1}{c}_{2}^{\ast }\right\rangle \right|}{\sqrt{\left\langle {c}_{1}{c}_{1}^{\ast }\right\rangle \left\langle {c}_{2}{c}_{2}^{\ast }\right\rangle }},\,0\le \gamma \le 1$$
(2)

The pre-event coherence represents change unrelated to the event, and was assumed to be the background value. Next, we obtained a coherence difference (COD) by subtracting $${\gamma }_{co-event}$$ from $${\gamma }_{pre-event}$$. Typically, human-made structures such as roads and buildings maintain high interferometric coherence, while vegetation and water surfaces show low coherence. A negative COD (or coherence gain), usually indicates surface changes occurring between the pre-event scenes and is associated with changes not related to the event. A positive COD (or coherence loss) indicates surface changes occurring between the co-event scenes spanning the event. Hence, the loss of coherence is most effective for detecting damage in built up areas caused by an event, and the types of change may include collapsed or submerged buildings and roads, and debris created by strong winds and storm surges. However, the COD is generally less effective over vegetation where coherence change may be random. Since only significant surface changes in the satellite line-of-sight can be detected, flood-induced internal building damage and cracked walls are not likely detected. We also note that greater loss in coherence generally correlates with greater severity of change, such as a fully collapsed building causing more significant coherence loss than partial collapse.

Therefore, to map damage, we determined a suitable minimum positive COD value to identify areas with significant decorrelation. Similar to flood mapping, a global threshold was manually selected based on knowledge of known damaged places from various validation sources including aerial imagery from GSI9 and news sources gathered in a similar way. Finally, a color map was applied to the thresholded COD to indicate increasingly significant decorrelation identified, corresponding to increasing COD values, and converted into PNG, GeoTIFF and KML formats for further use.

## Data Records

The Sentinel-1 SAR data we used to create our products can be obtained through the Amazon Web Services Public Dataset Program12. ALOS-2 data were obtained from Sentinel Asia and are not publicly available. The FPMs (Fig. 1) and DPM (Fig. 2) may be found in the ARIA-SG products13 (http://ariasg-products.earthobservatory.sg/EOS_ARIA-SG_201910_Japan_Typhoon_Hagibis) and ARIA share14 (https://aria-share.jpl.nasa.gov/201910-Typhoon_Hagibis_Japan) repositories respectively, where any updated map versions will be publicly released. Maps presented in this study are also freely provided through a static repository15.

The maps may be viewed through the PNG or JPEG files, or through KML files using Google Earth or GIS software. We also provide GeoTIFFs and Shapefiles of the flood and damage extents used to create the maps. These are GIS compatible and may be used for further spatial analyses. Lastly, each map has an accompanying text file detailing its usage, caveats and the SAR data it is derived from.

## Technical Validation

During the thresholding of the SAR-based FPMs and DPM, we calibrated the thresholding with different types of data where available. Firstly, we inspected 4-band PlanetScope optical imagery10 of 3 m resolution (Figs. 3 and 4). Timely optical images with minimal cloud cover were obtainable, but not covering all areas of the Tokyo and Fukushima prefectures. To inspect the images, we converted digital numbers to reflectance values, and made false color composites (FCCs) of reflectance using the NIR, red and green bands, assigned to red, green and blue channels respectively16. In general, water has lower reflectance across all bands, separating it from the vegetation that has higher reflectance in the NIR region. Thus, the FCCs show vegetation as red, clear water as dark blue, turbid water as cyan, and built-up areas as shades of blue, yellow or grey. We also computed the normalized difference water index (NDWI) from the reflectance bands (Eq. 3)17.

$$NDWI=\frac{Green-NIR}{Green+NIR}$$
(3)

The NDWI is proportional to water content, where positive values indicate moisture and negative values indicate water stress17. As the NDWI is generally less reliable over turbid waters, it was inspected with the FCC of reflectance.

Secondly, we used SRTM DEMs5 which highlight flood prone areas (Figs. 5 and 6), on the basis that water bodies tend to aggregate in low-lying areas and are confined by topographic features. Topographic data were available for all FPMs and covered whole mapping extents.

Thirdly, we used high resolution aerial imagery from GSI9 for a visual confirmation of flooded, non-flooded, damaged and non-damaged areas (Figs. 3, 4, 6 and 7). These were available for concentrated areas in the Tokyo, Fukushima and Nagano prefectures.

Overall, fitting thresholds could be determined such that the FPMs show consistency with the validation sources over open and less built-up areas, and sparse vegetation such as crop fields. The amplitude-based mapping is less reliable over dense urbanized areas due to a large mix of backscattering components within each relatively large resolution cell, and layover and shadow effects18. The FPMs are also only representative of flood extents at the time of SAR acquisition and do not necessarily reflect the peak of flooding. Apart from the differential responses of the drainage systems, the peak of flooding likely correlated with reported rainfall distributions and the typhoon track as well, which both migrated from south to northeast Japan in a little less than 24 hours after the typhoon’s landfall19. Hence, the FPMs covering Ibaraki and Nagano are more likely past the peak of flooding.

Thresholds determined for the DPM were also fitting with validation sources over urban areas. However, the DPM may indicate inundated low-rise buildings in flooded areas and false damage pixels over vegetated areas where coherence is generally poor. The DPM should also be used as a proxy for damage extent, and may not be equivalent to conventional Post-Disaster Needs Assessments which commonly define damage as the loss of assets of monetary value. For example, cracks in building walls, water-induced damage to electrical wiring and damaged crops are unlikely detected with this SAR-based technique.

Uncertainties in the flood and damage extents in areas lacking validation may be present due to the global thresholding of LAR and COD values across the mapping area. This is because we assumed that flood- or damage-induced backscatter and coherence changes within a SAR footprint are more or less the same when calibrating thresholds to fit independent validation information of limited spatial coverage. Hence, higher noise levels in the mapped pixels where validation data is lacking is possible (Fig. 8). Additionally, artefacts from layover effects may be present due to a combination of the side-looking SAR satellite look angle, earth curvature, topography and buildings. Further work is ongoing to improve the reliability over different land covers and spatial variation of the maps.

## Code availability

The codes for SAR processing using the ISCE algorithms and for generating proxy maps are available via the following GitHub repositories: https://github.com/isce-framework/isce2 and https://github.com/earthobservatory/slcp2pm.

## References

1. 1.

Cabinet Office, Government of Japan. Damage Status Related to Typhoon No. 19 of Reiwa 1. Report No. r1typhoon19_23 (Cabinet Office, Government of Japan, 2019).

2. 2.

Asian Disaster Reduction Center. Japan: Typhoon: 2019/10/12, https://www.adrc.asia/view_disaster_en.php?NationCode=392&Lang=en&Key=2357 (2019).

3. 3.

Kaku, K. & Held, A. Sentinel Asia: A space-based disaster management support system in the Asia-Pacific region. Int. J. Disast. Risk Re 6, 1–17 (2013).

4. 4.

Rosen, P. A., Gurrola, E., Sacco, G. F. & Zebker, H. The InSAR scientific computing environment. EUSAR 2012 9th European Conference on Synthetic Aperture Radar. 730–733 (2012).

5. 5.

National Aeronautics and Space Administration, Jet Propulsion Laboratory. NASA Shuttle Radar Topography Mission Global 1 arc second. NASA EOSDIS Land Processes DAAC, https://doi.org/10.5067/MEaSUREs/SRTM/SRTMGL1.003 (2013).

6. 6.

Dekker, R. J. Speckle filtering in satellite SAR change detection imagery. Int. J. Remote Sens. 19, 1133–1146 (1998).

7. 7.

Lee, J. S., Jurkevich, L., Dewaele, P., Wambacq, P. & Oosterlinck, A. Speckle filtering of synthetic aperture radar images: a review. Remote Sensing Reviews 8, 313–340 (1994).

8. 8.

Mason, D. C., Giustarini, L., Garcia-Pintado, J. & Cloke, H. L. Detection of flooded urban areas in high resolution synthetic aperture radar images using double scattering. Int. J. Appl. Earth Obs 28, 150–159 (2014).

9. 9.

Geospatial Information Authority of Japan. Geographical Institute Map, Geographical Survey Institute, https://maps.gsi.go.jp (2019).

10. 10.

Planet Team. Planet Application Program Interface: In Space for Life on Earth, San Francisco, CA, https://api.planet.com (2019).

11. 11.

Yun, S.-H. et al. Rapid damage mapping for the 2015 Mw 7.8 Gorkha earthquake using synthetic aperture radar data from COSMO–SkyMed and ALOS-2 satellites. Seismol. Res. Lett. 86, 1549–1556 (2015).

12. 12.

Registry of Open Data on AWS. Sentinel-1 SLC dataset for South and Southeast Asia, Taiwan, and Japan, https://registry.opendata.aws/sentinel1-slc-seasia-pds (2019).

13. 13.

Earth Observatory of Singapore. Advanced Rapid Imaging and Analysis Singapore (ARIA-SG) Products Repository, http://ariasg-products.earthobservatory.sg (2019).

14. 14.

National Aeronautics and Space Administration, Jet Propulsion Laboratory & California Institute of Technology. ARIA: Advanced Rapid Imaging and Analysis Share Repository, https://aria-share.jpl.nasa.gov (2019).

15. 15.

Tay, C. W. J. et al. Rapid flood and damage mapping using synthetic aperture radar in response to Typhoon Hagibis, Japan. figshare, https://doi.org/10.6084/m9.figshare.c.4726868 (2020).

16. 16.

Prost, G. L. Remote Sensing for Geoscientists: Image Analysis and Integration. (CRC Press, 2013).

17. 17.

McFeeters, S. K. The use of the Normalized Difference Water Index (NDWI) in the delineation of open water features. Int. J. Remote Sens. 17, 1425–1432 (1996).

18. 18.

Refice, A., D’Addabbo, A. & Capolongo, D. Flood Monitoring through Remote Sensing. (Springer, 2017).

19. 19.

Yotsumoto, J. The trail of destruction left by Typhoon Hagibis. NHK World-Japan, https://www3.nhk.or.jp/nhkworld/en/news/backstories/712/ (2019).

20. 20.

Geospatial Information Authority of Japan. Global Map Japan version 2.2 Vector data, https://www.gsi.go.jp/kankyochiri/gm_japan_e.html (2016).

## Acknowledgements

We sincerely thank Sentinel Asia and the Japan Aerospace Exploration Agency for their tasking efforts and providing ALOS-2 data for rapid mapping. We acknowledge the European Union Copernicus and European Space Agency for timely tasking and processing Sentinel-1 data, and Amazon Web Services for their provision of the Public Dataset Program to enable the efficient download of Sentinel-1 data. This research is supported by the Earth Observatory of Singapore (EOS), the National Research Foundation (NRF) of Singapore, and the Singapore Ministry of Education under the Research Centres of Excellence initiative, and by a Singapore NRF Investigatorship to E.M.H. (Proposal ID NRF2018NRF-NRFI001-21). Part of the work was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration. This work comprises EOS contribution number 284.

## Author information

Authors

### Contributions

C.W.J.T. carried out the flood mapping and analyzed validation data for the FPMs and DPM in this study. S.H.Y. carried out the damage mapping, with an advisory role in flood mapping, and is the lead developer for ARIA disaster response algorithms and systems. S.T.C. automated data acquisition and processing, and is the lead developer of the ARIA-SG system. A.B. provided the method for optical imagery analysis. J.J. worked together with S.H.Y. for the damage mapping. E.M.H. was heavily involved in all discussions and writing.

### Corresponding author

Correspondence to Cheryl W. J. Tay.

## Ethics declarations

### Competing interests

The authors declare no competing interests.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Rights and permissions

The Creative Commons Public Domain Dedication waiver http://creativecommons.org/publicdomain/zero/1.0/ applies to the metadata files associated with this article.

Reprints and Permissions

Tay, C.W.J., Yun, SH., Chin, S.T. et al. Rapid flood and damage mapping using synthetic aperture radar in response to Typhoon Hagibis, Japan. Sci Data 7, 100 (2020). https://doi.org/10.1038/s41597-020-0443-5

• Accepted:

• Published:

• ### Modelling typhoon-induced extreme river discharges: A case study of Typhoon Hagibis in Japan

• Joško Trošelj
•  & Han Soo Lee

Journal of Hydrology: Regional Studies (2021)

• ### Illogical dam policy in Japan

• Yoshiyasu Takefuji

Natural Hazards (2020)

• ### A Review of Factors Associated with Nutritional Problems and Improvement Initiatives after Natural Disasters

• Naoko Miyagawa
•  & Nobuo Nishi

The Japanese Journal of Nutrition and Dietetics (2020)

• ### WBF-2019 Core Research Cluster of Disaster Science Planning Session as Disaster Preparedness: Participation in a Training Program for Conductor-Type Disaster Healthcare Personnel

• Junko Okuyama
• , Hiroyuki Sasaki
• , Shuji Seto
• , Yu Fukuda
• , Toshiki Iwasaki
• , Toru Matsuzawa
• , Kiyoshi Ito
• , Takako Izumi
• , Hiroki Takakura
• , Fumihiko Imamura

Journal of Disaster Research (2020)

• ### Antipyretic Medication for a Feverish Planet

• Markus Stoffel
• , David B. Stephenson
•  & Jim M. Haywood

Earth Systems and Environment (2020)