For these last thirty years, earth observation and numerical models improved greatly and provide now a huge amount of accurate, yet heterogeneous, information on geophysics fluids dynamics and structures. Optimization methods from the eighties called variational data assimilation are capable of merging information from different sources. They have been used to estimate the parameters of numerical models and better forecast oceanic and atmospheric flows. Unfortunately, these powerful methods have trouble making benefit of always more complex information, suffering from the lack of available powerful calculators. The approach developed here, focuses on the use of multigrid methods, that are commonly used in the context of differential equations systems, to solve high resolution data assimilation. Multigrid methods are iterative methods improved by the use of feedback corrections evaluated on coarse resolution. First in the case of linear assimilation, we study the robustness of multigrid approach and the efficiency of the coarse grid correction step. We then apply the multigrid algorithms on a non linear 1-D Burgers equation and on a 2-D Shallow-Water model. We study two types of algorithms, the Gauss Newton Multigrid, which lays on global linearization, and the Full Approximation Scheme. Their behavior is compared to more traditional approaches as incremental and multi-incremental ones.
