| The problem of assimilating satellite remote sensing observations into the land surface modeling system to enhance the estimation of various hydrologic state/flux variables was investigated in this study. The assimilation problem in land surface hydrology was treated as a dynamic merge of information from the satellite into the modeling system. Mathematically, this question is posed as an optimal state estimation of a dynamic system, using the theories of state space systems, statistical modeling, Bayesian estimations, etc. Different from the assimilation in other fields, the assimilation here has to accommodate a number of particular behaviors of the underlying physical processes, the modeling system, and the remote sensing data from satellite sensors. Such particular behaviors include the highly nonlinear properties of the hydrologic dynamic system, the large and complicated (non-Gaussian) uncertainties (noises) in both satellite data and the model forcing inputs, the transfer processes between land surface variables and remotely sensed quantities and the noises so involved, and the physical constraints (e.g. the balance of water and energy) on the estimations. To meet these special needs of the problem, this study identified, developed and tested a number of traditional and new statistical techniques for the hydrologic data assimilation. These techniques include (1) the ensemble Kalman filter, (2) the particle filter, (3) the copula model, and (4) the method to enforce equality constraints in statistical estimations. A series of assimilation experiments were carried out to investigate the effectiveness, efficiency and drawbacks of these techniques, the strength of satellite data, and challenges as well. Experiments are performed in the form of both identical twin experiments using synthetically generated data and real assimilation experiments on real satellite or in-situ data, at different climate/vegetation regimes, and at both point and large river basin scales. Results confirm the great potential of satellite remote sensing in improving our ability to characterize the land surface hydrologic system, if the behaviors of the target dynamic system and data are carefully studied and the assimilation methods are carefully chosen. |
