A Comparative Analysis of Flood Frequency Mapping Approaches for Climate-Resilience in South Sudan

Résumé fourni par la source

In 2022, South Sudan was ranked as the world’s most vulnerable country to climate change and the one most lacking in coping capacity. Furthermore, it is also one of the world’s most politically fragile nations. The country is facing challenges related to riverine flooding, including four consecutive years of floods (2019-2022) that have displaced hundreds of thousands of people and left many struggling to access food.Flood extent and frequency mapping based on remote sensing products is being explored by the European Space Agency's Global Development Assistance (GDA) programme's thematic area of Climate Resilience, as a collaboration between GMV and the World Bank in South Sudan.Floods are mapped with Synthetic Aperture Radar (SAR) imagery from Sentinel-1 (S-1), and the 5-day VIIRS flood fraction product. The former has a native pixel size of 10 m (GRD), whereas it is 375 m for the latter. This resolution disparity is bridged aggregating 9x9 S-1 pixels (which also reduces speckle “noise”) and downscaling the VIIRS product using the flood fraction and the 90 m Copernicus Digital Elevation Model to determine which pixels are more likely to be flooded.Sentinel-1 flood delineation detects significant deviations from the standard 'dry' stratus using by-track geo-median (sigma-nought) or terrain-flattened gamma-nought image classification. The latter method includes the closest VIIRS 8-day mosaics to prevent false positives in semi-arid regions. Both approaches aim to identify flooding, even beneath vegetation canopies.Due to the absence of in-situ data, it was not possible to validate the results but an intercomparison was conducted, including different S-1 methods. The downscaled VIIRS product yielded the largest flood extents and frequencies, likely due to its higher imaging frequency (14 h). Consequently, the deviation-based Sentinel-1 products exhibit similar spatial patterns but with lower frequencies and extents due to longer revisit times. These S-1 methods failed to detect flooding in some areas marked as high-frequency flooding by VIIRS, this is attributed to a mischaracterization when the reference image is already flooded. In contrast, the classification-based Sentinel-1 product captured actual flood frequency but was prone to omission and commission errors. Combining maximum flood frequency from both Sentinel-1 products, while masking false positives with VIIRS, reduces errors while preserving maximum spatial detail.The resulting Earth Observation (EO)-based maps provide key information on the extent, frequency, and persistence of recent flooding seasons (2017-2022). This detailed flood hazard information can raise awareness of flood risk among local institutions and communities. For such purpose, EO data is consolidating its role in helping reduce flood risk to citizens’ lives and livelihoods, as ground data is very sparse across many countries. By combining EO-based flood hazard maps with exposure datasets such as for population, building or crops, we provide additional country-wide information on the potential impacts of recent floods. The service covers the entire country of South Sudan and enables the creation of a flood hazard and exposure index, allowing the World Bank team to detect flooding hotspots and prioritize investment accordingly. These efforts will help the government develop detailed flood risk management plans.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.