Real-time Global Flood Monitoring and Forecasting using an Enhanced Land Surface Model with Satellite and NWP model based Precipitation

A community land surface model (LSM), Variable Infiltration Capacity (VIC) model, was enhanced by coupling with a hierarchical dominant river tracing-based runoff-routing model. The Dominant river tracing-Routing Integrated with VIC Environment (DRIVE) model formed the new core of an existing real-time global flood monitoring system (GFMS). It is the first time to use satellite-based real-time precipitation (with other data) to drive a state-of-the-art LSM for real-time flood monitoring for global domain at relatively high spatial (~12km) and temporal (3-hourly) resolution. In order to evaluate the new GFMS accuracy in flood event detection and flood magnitude estimation, we ran the DRIVE model for retrospective ~15 years (1998~) using both NASA TMPA research and real-time precipitation products, with the model simulations referred to as DRIVE-V7 and DRIVE-RT respectively. The DRIVE-RT and DRIVE-V7 derived very close probability of detection (0.90 vs. 0.93) and false alarm ratio (0.88 vs. 0.84) against archived flood events with duration greater than one day, which are much better than the old GFMS using a simpler hydrologic model driven by TMPA 3B42V6 research product. The DRIVE-V7 derived positive daily and monthly Nash-Sutcliffe coefficient (NSC) for 362 (32.3%) and 675 (60.2%) gauges, out of 1,121 in total from global rivers with observed daily streamflow data, with a mean of 0.39 and 0.212 respectively. The model performance generally decreases from tropics toward higher latitudes at annual, seasonal and daily scales, with DRIVE-V7 generally better than DRIVE-RT. However, their performances at daily scale had no significant difference for almost all regions except the northern mid-latitudes where TMPA V7 research product has much better quality than real-time data because of gauge data based corrections. A real-time evaluation on recent flood cases for the new operational GFMS (http://flood.umd.edu) demonstrated that the new GFMS had a fairly good performance in flood occurrence detection, flood evolution and magnitude calculation according to river gauge data. The DRIVE model has also been further developed for better flood detection, streamflow and inundation estimation at a much higher resolution (as fine as 1 km) through a novel hybrid-resolution coupling between the VIC model and a physically based routing model with high spatial-temporal dynamics delineation for floodplain or routed runoff re-distribution. We also extended the GFMS for 3~5 days flood forecasting using precipitation estimations from NWP models (i.e., NASA GEOS-5 and NOAA GFS).