Title: Model-reduced 4D-Var data assimilation in application to 1D ecosystem model

Authors: Joanna S. Pelc (Delft University of Technology, Delft, The Netherlands; Deltares, Delft, The Netherlands)
Ehouarn Simon (Nansen Environmental and Remote Sensing Center, Bergen, Norway)
Laurent Bertino (Nansen Environmental and Remote Sensing Center, Bergen, Norway)
Ghada El Serafy (Deltares, Delft, The Netherlands)
Arnold W. Heemink (Delft University of Technology, Delft, The Netherlands)

Nowadays, ecosystem models become more and more sophisticated. The number of their biological components, as well as their parameters is increasing. Even simple ecosystem models have strong nonlinear behavior. For such challenging environment data assimilation can be an extremely difficult task. Especially when one considers the adjoint-based techniques, obtaining the adjoints of such models is nontrivial. Also the model resolutions are much finer than in the past, which results in large sizes of the model states. This introduces a limitation for using the finite difference gradient approximations, since these ones are not suitable for large problems. The model-reduced 4D-Var (Vermeulen and Heemink, MWR, 2006) is a method proposed to deal with this problem. Based on a number of simulations of the original model, proper orthogonal decomposition is used to obtain a reduced model. The model-reduced 4D-Var is performed in the reduced space, therefore, the implementation of the adjoint of the tangent linear approximation of the original model is not required. Instead, it is approximated by the adjoint of the tangent linear approximation of the reduced model. The method is easily extended to the initial condition estimation, hence the parameter calibration is coupled together with the initial condition estimation. This method was effectively used in several applications. An advantage was shown especially for models characterized by periodical behavior. Since for these type of models, the number of required model simulations is relatively small. The ecosystem models also have periodic characteristics, therefore the method is a potential tool in ecological applications. Twin experiments have been conducted in a 1D ecological model. The study highlights the ability of the method to tackle the problem of combined initial condition and parameter estimation in such nonlinear framework.

GMAO Head: Michele Rienecker Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: May 27 2011