Evaluation of an Application of Adjoint Methods to Yellow Sea Modeling

In the application of adjoint methods, adjoint model equations are used to compute the directions in which unknown model parameters should be adjusted to achieve an optimal fit between modeled dynamics and observations. Many researchers have successfully applied an incremental adjoint modeling approach to infer open ocean boundary forcing that accounts for available observations. In this study, an adjoint model system, comprising a simple model and its inverse in conjuction with a more complex forward model, is applied in the Yellow Sea to predict tidal elevations and currents throughout the region using the incremental approach. The nonlinear finite element model ADCIRC-2DDI is used as the complex forward model with the linear finite difference adjoint model of Thompson in the inverse assimilative loop. Field measurements include sea surface elevations derived from altimeter data and tidal stations, and currents from three ADCP locations. Various implementations of the feedback between the linear inverse model and the nonlinear forward model are examined. Comparisons between large domain solutions without data assimilation and limited area domains forced with adjoint model predictions on the open boundary form the basis for evaluation of the approach. Incompatibilities in model discretization, resolution, and boundary condition formulation severely limit the advantages of the application of the incremental approach using the finite difference adjoint model of Thompson as implemented here. --Original contains color plates: All DTIC reproductions will be in black and white.