EGU23-8774
https://doi.org/10.5194/egusphere-egu23-8774
EGU General Assembly 2023
© Author(s) 2023. This work is distributed under
the Creative Commons Attribution 4.0 License.

A likelihood analysis of the Global Flood Monitoring ensemble product

Christian Krullikowski1, Candace Chow1, Marc Wieland1, Sandro Martinis1, Marco Chinni2, Patrick Matgen2, Bernhard Bauer-Marschallinger3, Florian Roth3, Wolfgang Wagner3, Tobias Stachl4, Christoph Reimer4, Christian Briese4, and Peter Salamon5
Christian Krullikowski et al.
  • 1German Aerospace Center (DLR), German Remote Sensing Data Center, Weßling, Germany (christian.krullikowski@dlr.de)
  • 2Luxembourg Institute of Science and Technology
  • 3TU Wien
  • 4EODC Earth Observation Data Centre For Water Resources Monitoring Gmbh
  • 5Joint Research Centre of the European Commission (JRC)

Flooding is a natural disaster that can have devastating impacts on communities and individuals, causing significant damage to infrastructure, loss of life, and economic disruption. The Global Flood Monitoring (GFM) system of the Copernicus Emergency Management Service (CEMS) addresses these challenges and provides global, near-real time flood extent masks for each newly acquired Sentinel-1 Interferometric Wide Swath Synthetic Aperture Radar (SAR) image, as well as archive data from 2015 on, and therefore supports decision makers and disaster relief actions. The GFM flood extent is an ensemble product based on a combination of three independently developed flood mapping algorithms that individually derive the flood information from Sentinel-1 data. Each flood algorithm also provides classification uncertainty information as flood classification likelihood that is aggregated in the same ensemble process. All three algorithms utilize different methods both for flood detection and the derivation of uncertainty information.
The first algorithm applies a threshold-based flood detection approach and provides uncertainty information through fuzzy memberships. The second algorithm applies a change detection approach where the classification uncertainty is expressed through classification probabilities. The third algorithm applies the Bayes decision theorem and derives uncertainty information through the posterior probability of the less probable class. The final GFM ensemble likelihood layer is computed with the mean likelihood on pixel level. As the flood detection algorithms derive uncertainty information with different methods, the value range of the three input likelihoods must be harmonized to a range from low [0] to high [100] flood likelihood.
The ensemble likelihood is evaluated on two test sites in Myanmar and Somalia showcasing the performance during an actual flood event and an area with challenging conditions for SAR-based flood detection. The findings further elaborate on the statistical robustness when aggregating multiple likelihood layers.
The final GFM ensemble likelihood layer serves as a simplified appraisal of trust in the ensemble flood extent detection approach. As an ensemble likelihood, it provides more robust and reliable uncertainty information for the flood detection compared to the usage of a single algorithm only. It can therefore help interpreting the satellite data and consequently to mitigate the effects of flooding and accompanied damages on communities and individuals.

How to cite: Krullikowski, C., Chow, C., Wieland, M., Martinis, S., Chinni, M., Matgen, P., Bauer-Marschallinger, B., Roth, F., Wagner, W., Stachl, T., Reimer, C., Briese, C., and Salamon, P.: A likelihood analysis of the Global Flood Monitoring ensemble product, EGU General Assembly 2023, Vienna, Austria, 24–28 Apr 2023, EGU23-8774, https://doi.org/10.5194/egusphere-egu23-8774, 2023.