A land data assimilation system (LDAS) is designed to provide continental-scale estimates of surface moisture and temperature states-and water and energy flux exchanges with the atmosphere-by integrating micrometeorological forcing data (such as surface precipitation, incoming radiation, wind, air temperature, and humidity) with a land surface model. When measurements of land system states are available, data assimilation approaches can be used to optimally update model states based on the assumed magnitude of modeling and observational errors [e.g., Reichle and Koster, 2005].LDAS estimates have value for water, energy, and biogeochemical studies as well as for the initialization of land surface states, such as soil moisture and temperature, in numerical weather prediction or seasonal climate forecasts. To improve forecast performance, collaborative LDAS projects are currently under way for the North American (NLDAS; http://ldas. gsfc.nasa.gov), European (ELDAS; http://www. knmi.nl/samenw/eldas), and global (GLDAS; http://ldas.gsfc.nasa.gov) domains. Preliminary results demonstrate that the incorporation of LDAS approaches into subseasonal forecast initialization can lead to statistically significant improvements in precipitation predictability .However, the value of LDAS predictions is tempered by the impact of forcing data errors and structural limitations in land surface models. Reducing these errors and improving the value of LDAS estimates for forecast applications will likely require new global remote sensing measurements. Potential sources for providing such data include the Global Precipitation Measurement (GPM) satellite constellation under development by NASA [Flaming, 2005] and currently unfunded satellite mission concepts focusing on the use of L-band radar and microwave radiometry to retrieve surface soil moisture [e.g., Entekhabi et al., 2003]. Exploiting the complementary nature of multiple types of water cycle measurements significantly enhances the accuracy of LDAS system outputs.
LDAS Error SourcesLand surface models at the core of LDAS systems are used to make continuous estimates of land surface moisture and temperature states. Both parameterization (model structure) and forcing (model input) errors reduce the accuracy of these estimates. Correcting these errors requires the accurate continental-scale retrieval of both model state (e.g., soil moisture) and forcing (e.g., precipitation) variables.Structural uncertainty in land surface models manifests itself in highly variable estimates of soil moisture and evapotranspiration [Koster and Milly, 1997] among various land surface schemes, even when models are forced with the same precipitation and meteorological input data. Soil moisture measurements taken from space provide a valuable constraint for reducing this uncertainty. Examples include the use of remotely sensed soil moisture to calibrate model soil hydraulic parameters [Burke et al., 1997] and to compensate for a poor model parameterization by updating model states using data assimilati...