Real-Time Analysis of the 2019 Mozambique Flood Using Satellite Rainfall and the Global Flood Monitoring System (GFMS)

Satellite observations and models are being used to monitor and forecast floods in real-time on a global scale. Results from the March 2019 flood in Mozambique associated with Cyclone Idai is used as a main example of such activity utilizing output from the Global Flood Monitoring System (GFMS) (http://flood.umd.edu) and other products. The GFMS has been developed and used in recent years to provide real-time flood detection, streamflow estimates and inundation calculations for most of the globe. The GFMS is driven by satellite-based precipitation, with the accuracy of the flood estimates being primarily dependent on the accuracy of the precipitation analyses and the land surface and routing models used. The GFMS utilizes precipitation estimates from GPM’s IMERG rainfall product (.1 deg., 30 min. resolutions). The routing calculations are done at both 12 km and 1 km resolution. IMERG validation in terms of daily precipitation will be described as well as comparisons between GFMS inundation calculations and satellite-based inundation observations from optical, Synthetic Aperture Radar (SAR) and passive microwave data.

Prior to landfall NWP estimates of rainfall were used in GFMS to forecast location and intensity of flooding. The forecast NWP estimates from the NASA GEOS model were larger than eventually estimated from the satellite IMERG estimates (peak values 1000 mm vs. 600 mm). Direct validation of the IMERG results for this case was difficult because of the lack of local conventional ground estimates because the storm’s destruction, but surrounding and prior validation seem to confirm an underestimate of storm rainfall, especially the peak values. The Idai validation results are also compared to those done over the U.S. with Hurricanes Florence and Barry. Underestimation also seems to be present in those cases.

The 1 km resolution inundation maps calculated by GFMS are compared to SAR estimates. In terms of latency GFMS was generating inundation estimates as soon as it started flooding about 15 March, but the first SAR estimates were produced by 20 March. The accuracy of the GFMS estimates were evaluated both for simultaneous instantaneous estimates on 19-20 March and also with a time-integrated (15-19 March) GFMS inundation map. The results show significant underestimation of the SAR-based flood area, although this is ameliorated by the use of time integration. These effects seem to be related to a combination of the underestimation of the rainfall amounts and a hydrologic phase error. The GFMS inundation maps are also compared to other products, including an estimate based on passive microwave data and terrain information.

An approach for possible integration of these types of estimates to provide a consistent method for users to access the best data quickly are described.