Investigating Satellite Rainfall Based Flood Modeling in Anticipation of GPM: Understanding the Worth of Spatial Downscaling and Satellite Rainfall Uncertainty

Realistic flood modeling in medium-large river basins requires rainfall data at hydrologically relevant scales ranging from 1--5 km. However, satellite rainfall data has historically been available at spatial resolutions that can be considered somewhat coarse for predicting the dynamic flood phenomenon (~ 25--100km). As a natural response to this limitation that has persisted for over a decade, hydrologists have devised numerous statistical spatial downscaling schemes till now. With the proposed Global Precipitation Measurement (GPM) mission, satellite rainfall data will gradually become more available at smaller scales (~ 10 km) in the next decade, prompting us to re-evaluate the worth of spatial downscaling for flood modeling. In this study, we therefore seek an answer to the question--- Which is the better option for satellite rainfall based flood simulation when rainfall data is available at coarse scale---a) an error propagation based ensemble streamflow scheme at the coarse (native) resolution or a probabilistically downscaled based ensemble streamflow scheme? The study is performed on the 970 km2 basin of the Upper Cumberland River in southeastern Kentucky bordering with Virginia and Tennessee. NASA satellite rainfall data products from the TRMM Multi-satellite Precipitation Analysis (TMPA) are used for the investigation. A statistical downscaling scheme of Perica and Foufoula-Georgiou (1996) and a satellite rainfall error modeling scheme of Hossain and Anagnostou (2006) are used for resolving the posed science question. Findings indicate that spatial downscaling does not unconditionally guarantee more accurate flood simulations as scale becomes smaller. The narrow range of uncertainty in flood simulation due to downscaling often misses the observed peak flow. On the other hand, error propagation at the native scale based on satellite rainfall uncertainty information tends to capture the natural variability during peak flows with much greater confidence.