OpenSendaiBench: A Benchmark Dataset of Building Exposure and Vulnerability Dynamics for EO-based Auditing of Global Disaster Risk

This Zenodo repository is the official global dataset for the research poster "Global Mapping of Exposure and Physical Vulnerability Dynamics in Least Developed Countries using Remote Sensing and Machine Learning” at 2nd Machine Learning for Re…, 12th International Conferen… in Vienna, Austria, on 11th of May 2024. The GitHub repository of Python codes can be accessed here: github.com/riskaudit/OpenSe…. The following technical info is from the four-page paper of this research poster. If you have any inquiries or would like to access any related materials, please feel free to visit my website (joshuadimasaka.com) or our project website (riskaudit.github.io), follow our project's GitHub repository (github.com/riskaudit), or send an email to jtd33@cam.ac.uk.

1. National Census-Derived Exposure Data

We rasterized every country-wide point dataset of building counts from the METEOR project with a defined physical vulnerability type at a spatial resolution of 15 arcseconds or approximately 500 meters at the equator (Huyck et al., 2019). We then implemented a rigorous probability-based approach in extracting 100 square tiles for each country. In sampling these 100 square tiles per country, we considered the number of physical vulnerability types that are present in every pixel to ensure that every label including those unlabeled pixels is represented. 

2. Time-Series Satellite Imagery

With the previously extracted geographical extents, we obtained the following pre-processed time-series satellite imagery via Google Earth Engine (Gorelick et al., 2017).

2.1. Sentinel-1 SAR GRD

At 10-m spatial resolution, we used the annual mean of the Ground Range Detected (GRD) scenes that are acquired from the dual-polarization C-band Synthetic Aperture Radar (SAR) instrument at 5.405GHz of Sentinel-1 satellite (Copernicus Sentinel data,…. As a result, covering the years from 2015 to 2023, we extracted nine annual mean of the two bands:

• VV (vertical transmit, vertical receive) and
• VH (vertical transmit, horizontal receive) signals.

To avoid data incompleteness across large areal extent, we disregarded filtering by orbital number and satellite direction. We also note that there are countries such as Angola, Comoros, Ethiopia, Kiribati, and Tuvalu with either partially or fully complete VV and VH signals because the orbit of Sentinel-1 satellite does not cover these areas for some time or only a single VV signal is available.

2.2. Sentinel-2 Harmonized MSI

With similar spatial resolution at 10 meters, we also extracted the annual median of the atmospherically corrected surface reflectance signals represented by the red, green, and blue (RGB) bands that are acquired from the MultiSpetral Instrument (MSI) of Sentinel-2 satellite (Copernicus Sentinel data,…. The aggregation by year also allows us to filter out and minimize the unnecessary cloudy or shadowy signals using the available and corresponding Sentinel-2 cloud probability dataset (Copernicus Sentinel data,…. Unlike Sentinel-1 SAR GRD, the resulting six annual median maps from 2018 to 2023 are all available for 47 countries (Note: Bhutan and Vanuatu already graduated from LDC status).

3. File and Folder Structure

Each <countryCode>.zip file has the following file and folder structure.

├───extent
│   └───<countryCode>_<nth>_of_<totalTiles>_<index>.geojson
├───groundtruth
│   └───<countryCode>_nbldg_<vulnerabilityCode>_<nth>_of_<totalTiles>_<index>.tif
└───obsvariables
    ├───SENTINEL1-DUAL_POL_GRD_HIGH_RES
    │   └───<countryCode>_<nth>_of_<totalTiles>_<index>
    │       ├───<year>_VV.tif
    │       └───<year>_VH.tif
    └───SENTINEL-2-MSI_LVL2A
        └───<countryCode>_<nth>_of_<totalTiles>_<index>
            └───<year>_RGB.tif

4. Custom Download Individual Country

The public may download the entire data repository or individual countries under the "Files" tab. We also provided the following customized download hyperlinks and expanded description of every country. 

1. AFG: Afghanistan
2. AGO: Angola
3. BDI: Burundi
4. BEN: Benin
5. BFA: Burkina Faso
6. BGD: Bangladesh
7. BTN: Bhutan (graduated from LDC status in December 2023)
8. CAF: The Central African Republic
9. COD: The Democratic Republic of the Congo
10. COM: The Comoros
11. DJI: Djibouti
12. ERI: Eritrea
13. ETH: Ethiopia
14. GIN: Guinea
15. GMB: The Gambia
16. GNB: Guinea-Bissau
17. HTI: Haiti
18. KHM: Cambodia
19. KIR: Kiribati
20. LAO: The Lao People's Democratic Republic
21. LBR: Liberia
22. LSO: Lesotho
23. MDG: Madagascar
24. MLI: Mali
25. MMR: Myanmar
26. MOZ: Mozambique
27. MRT: Mauritania
28. MWI: Malawi
29. NER: The Niger
30. NPL: Nepal
31. RWA: Rwanda
32. SDN: The Sudan
33. SEN: Senegal
34. SLB: Solomon Islands
35. SLE: Sierra Leone
36. SOM: Somalia
37. SSD: South Sudan
38. STP: Sao Tome and Principe
39. TCD: Chad
40. TGO: Togo
41. TLS: Timor-Leste
42. TUV: Tuvalu
43. TZA: United Republic of Tanzania
44. UGA: Uganda
45. VUT: Vanuatu (graduated from LDC status in December 2020)
46. YEM: Yemen
47. ZMB: Zambia

5. References

• Copernicus Sentinel data. Sentinel-1 SAR GRD: C-band Synthetic Aperture Radar Ground Range Detected, log scaling. developers.google.com, 2024a. Accessed: 2024-02-01.
• Copernicus Sentinel data. Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-2A. developers.google.com, 2024b. Accessed: 2024-02-01.
• Copernicus Sentinel data. Sentinel-2: Cloud Probability. developers.google.com, 2024c. Accessed: 2024-02-01.
• Joshua Dimasaka, Christian Geiß, and Emily So. Global mapping of exposure and physical vulnerability dynamics in least developed countries using remote sensing and machine learning [Poster], 2nd ML for Remote Sensing Workshop, 12th ICLR. Vienna, Austria. 11 May. 2024.
• Noel Gorelick, Matt Hancher, Mike Dixon, Simon Ilyushchenko, David Thau, and Rebecca Moore. Google earth engine: Planetary-scale geospatial analysis for everyone. Remote Sensing of Environment, 2017. doi: 10.1016/j.rse.2017.06.031. URL doi.org.
• C Huyck, Z Hu, P Amyx, G Esquivias, M Huyck, and M Eguchi. METEOR: Exposure data classification, metadata population and confidence assessment. report m3. 2/p. 2019.

Not all countries have the same set of physical vulnerability types and satellite imagery signals. Instead of downloading and manually checking each file, we recommend reviewing this spreadsheet first to understand the available features and groundtruth labels of each country. 

• Some countries do not have both VV and VH, but single VV only. This means that some areas are not covered by the satellite orbits.
• We do not suggest using the 2018 median map for the Sentinel-2 Harmonized MSI because most do not fully cover the defined extent or mask. This may be because of the limited spatiotemporal coverage of the satellite.