10C.1 Advancing Flash Flood Forecasting Capabilities in West Africa with Machine Learning and Satellite Observations

Wednesday, 31 January 2024: 10:45 AM
339 (The Baltimore Convention Center)
Efthymios Nikolopoulos, Rutgers University, Piscataway, NJ; and A. Ali, W. Amponsah, G. Anagnostopoulos, A. Aravamudan, J. J. Gourley, V. Maggioni, M. Nasibi, V. Robledo, H. Vergara, and X. Zhang

Climate change is modifying the occurrence frequency and intensity of hydroclimatic extremes around the globe. This has a tremendous impact on the lives and properties of populations, especially in the most vulnerable regions of the world, such as West Africa (WA). The WA region is particularly vulnerable to the impacts of climate change due to the low adaptive capacity and high sensitivity of the socio-economic systems. For example, the 2009 floods in WA after torrential rains, affected 600,000 people in sixteen WA countries. Heavy rainfall is the main driver of floods in the region with the vast majority (>60%) of heavy rainfall events associated with less than 6h duration. Short duration and high intensity rainfall events result in flash floods, one of the most devastating types of floods. Flash flood events, however, are particularly difficult to monitor because they develop at space and time scales that conventional measurement networks of rain are not able to monitor effectively. In regions like WA where in situ observations are sparse and a network of ground weather radars does not exist, the only viable alternative for monitoring rainfall is from space-based platforms. Satellite precipitation observations combined with weather forecasts from numerical weather prediction models offer a unique opportunity to establish effective monitoring/forecasting flash flood systems in this region. As part of the NASA SERVIR program, we have developed a flash flood forecasting system for WA region, that is based on i) the EF5 (Ensemble Framework For Flash Flood Forecasting) distributed hydrologic modeling system, ii) the near-real time satellite precipitation estimates from NASA IMERG (Integrated Multi-satellitE Retrievals for GPM) and iii) a novel machine learning-based nowcasting technique. The proposed forecasting system will provide critical information for the design and operation of relevant water infrastructure (e.g., levees, storm drainage, dams), flood insurance, and floodplain management and will ultimately assist climate adaptation and hazard mitigation strategies. In this work, each component of the developed framework is presented and results of flood predictions for selected flash flood events in the WA region are discussed. Preliminary results highlight the significant challenges to forecasting skill posed by factors such as data stream latency, satellite precipitation uncertainties and prediction lead time. Concurrently, they offer important insights for future improvements in such systems.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner