Improving Flash Flood Monitoring and Forecasting Capabilities in West Africa with Satellite Observations and Precipitation Forecasts

Abstract

Flash floods are one of the deadliest types of floods due to their rapid onset and space-time characteristics. Early detection and warning of these threats require frequent estimates of the high precipitation rates that cause them. In this contribution, we will discuss the development work of a flash flood forecasting system based on satellite observations and short-to-medium range quantitative precipitation forecasts for the West Africa region of the SERVIR network. The currently available hydrologic forecasting system in the region does not have the capacity to predict hydrologic variables at the flash flood scale. Implementing such systems in emerging regions without suitable in-situ observational infrastructure (e.g., radars) presents additional challenges to the inherent limited predictability of these hazards. The newest generation of quantitative precipitation estimates (QPEs) in NASA’s Global Precipitation Measurement (GPM) mission’s Integrated Multi-satellitE Retrievals for GPM (IMERG) product offer a major opportunity to achieve the goal of improving short-fused and high-resolution precipitation information over the globe. IMERG provides precipitation information at (nominally) 10-km pixel resolution and 30-min intervals, which holds the potential to resolve rainfall intensities at the scale of flash floods. However, the minimum 4-hour data latency of the near real-time IMERG Early Run product lessens the utility of the precipitation estimates for the detection and forecasting of flash flood threats. We address this shortcoming with the development of an innovative machine-learning based nowcasting algorithm that blends IMERG estimates with geostationary satellite observations from EUMETSAT that have a much lower latency (approx. 15 minutes). We evaluate the utility of these nowcasts as input forcing to a distributed hydrologic model specifically designed to predict flash flooding conditions. The hydrologic model is configured at high spatial resolutions (100-m – 1-km) to describe the surface processes that buffer or exacerbate the effects of intense rainfall. Furthermore, we assess the lead times of flash floods forecasts forced with IMERG, the nowcasts and medium-range precipitation forecasts from a regional WRF model developed for the West Africa region.

Georgios C. Anagnostopoulos

Associate Professor of Electrical & Computer Engineering

I lead the Machine Learning Research Group at FIT.

Akshay Aravamudan

Doctoral Student of Computer Engineering

Xi Zhang

Senior Doctoral Student of Electrical Engineering

My research interests include point process analysis, modeling and optimization.