Abstract:
Sophisticated land surface models can now run globally at high resolutions on inexpensive computing platforms. The accuracy of their output is limited by the quality of the input data used to parameterize and force the models, the model developers' understanding of the physics involved, and the simplifications necessary to depict the Earth system economically. Numerous streams of relevant satellite observations are now available, but they have their own problems, including data gaps, errors from multiple sources, and low resolutions. Furthermore, remote sensing is not yet able to provide a complete picture of all the processes and conditions we wish to assess. The advantages of both land surface modeling and remote sensing can be harnessed by data assimilation, which synthesizes discontinuous and imperfect observations with our knowledge of physical processes, as represented in the models. Multiple data assimilation algorithms are now being implemented in the Global Land Data Assimilation System (GLDAS) at NASA. These will be discussed, along with the potential pitfalls of multivariate data assimilation. Furthermore, we have begun to design and test approaches for constraining land surface models with terrestrial water storage information derived from the Gravity Recovery and Climate Experiment (GRACE). The potential value of GRACE for hydrological research and applications is huge, as it is the only remote sensor currently able to detect water variability below the upper few centimeters of soil. However, the unique spatial and temporal characteristics of GRACE products present a special challenge.