Assimilating remote sensing observations across the terrestrial water cycle in a drought forecasting system

The application of data assimilation techniques in hydrologic studies has been gaining traction in the last 10-15 years. Most of these studies have focused on a single water cycle component, while few studies have examined methods of assimilating multiple observations from different sensors and of different hydrologic variables. The latter is challenging since any potential disparities in the observations could lead to suboptimal estimates after assimilation. The optimal estimates of hydrologic states, such as soil moisture, can be used as initial conditions for hydrologic forecasting systems. A multi-sensor and multi-variable data assimilation forecast system has been developed at JPL (RHEAS, Regional Hydrologic Extremes Assessment System) with an initial focus on forecasting drought characteristics. The core of the flexible RHEAS framework is the VIC hydrology model, while a number of satellite data products are used to drive or constrain RHEAS that correspond to different components of the water cycle. Precipitation is obtained from TRMM, CMORPH, NRLgeo, REFAME, and PERSIANN-CCS; ET from PT-JPL and PM-MOD16; snow covered area and dust radiative forcing in snow from MODSCAG and MODDRFS, respectively; soil moisture (where possible) from AMSR-E and SMOS; and groundwater change derived from GRACE. The initial study area includes two basins: the Sacramento-San Joaqin and the Upper Colorado River basins. The benefit of implementing RHEAS in these basins is the existence of in-situ measurements of rainfall, soil moisture and streamflow. The experiment starts with a baseline simulation using only in-situ measurements to drive the model, while a hindcast experiment is performed by predicting the hydrologic fluxes with a model driven by an ensemble of meteorological forcings. Additionally, initial results are presented from a hindcast experiment of the summer 2012 drought over the Midwest United States.