NASA MODIS Flood Mapping Product Assimilation in Operational Flash Flood Warning Systems

Regional operational systems that support forecasters for the real-time warning of flash flood events have been implemented worldwide in the last decade. These systems provide hydrological and meteorological agencies the tools to produce timely alerts and warnings for flash floods in small basins (area of order of 100 km²). The systems use precipitation forcing from hourly geo-stationary satellite precipitation products (Hydro Estimator, NESDIS), adjusted by raingauges and in several cases by satellite microwave observations, as input to a land surface model that updates the moisture state of the land and tracks the degree of stream bankfull exceedance. The output of the model consists of indices that estimate the amount of rain of a certain duration that is needed over a given small basin in order to cause minor flooding (bankfull flow) at its outlet. These indices are compared with local available data and nowcast products by forecasters and enable the generation of prompt flash flood warnings and alerts.

Soil moisture is the principal state variable in estimating the rainfall-runoff relationship in a given catchment. Antecedent soil moisture conditions directly impact the ability of additional precipitation inputs to infiltrate, rather than becoming surface runoff. In the operational systems the focus is on the water balance over the flash-flood prone small watersheds, and downstream inundation effects or large scale groundwater effects are not significant for the majority of the upland flash flood prone basins, and these effects are not considered. However, some lowland areas and flat terrain near large rivers experience standing water long after local precipitation has ceased. This may be the result of backwater effects from swollen rivers or raised groundwater levels from upland source areas. The NASA Office of Applied Science is producing an experimental product from the MODIS instrument on the Terra and Aqua satellites that detects standing water, beyond reference water, at a daily time interval and with a 250m resolution. This presentation discusses the potential utility of this product to adjust the soil water estimates of the operational systems for flash flood prone basins in low lying areas to improve local flash flood warnings.

Given that a portion of the catchment area is inundated, the total volume of the upper soil can be divided in to two separate volumes, inundated and not inundated. MODIS observations are then used to update the inundated portion of the total volume at the time of observation, thereby updating the total volume of water in the upper soil zone. The difference between modeled soil saturation fraction estimates and unity, representing full inundation, for those basin-days experiencing full inundation was found to be nearly normally distributed. Conditional sampling was used to generate model error estimates for a given range of modeled upper soil water. Those error estimates were used in the context of Monte Carlo ensemble forecasting of soil water and flash flood potential. Numerical experiments with six months of data (July 2012 – December 2012) showed that MODIS inundation data, when assimilated to correct soil moisture estimates, increased the likelihood that bankfull flow would occur, over non-assimilated modeling, at catchment outlets for approximately 44% of basin-days during the study time period. For these basins this is a significant reduction of the bias that leads forecasters to underestimate local flash flood threat.