Four-Dimensional Variational Data Assimilation Experiments with a Multilevel Semi-Lagrangian Semi-Implicit General Circulation Model

Abstract

Four-dimensional variational data assimilation (VDA) experiments have been carded out using the adiabatic version of the NASA/Goddard Laboratory for Atmospheres semi-Lagrangian semi-implicit (SLSI) multilevel general circulation model. The limited-memory quasi-Newton minimization technique was used to find the minimum of the cost friction. With model-generated observations, different first-guess initial conditions were used to carry out the experiments. The experiments included randomly perturbed initial conditions, as well as different weight matrices in the cost function. The results show that 4D VDA works well with various initial conditions as control variables. Scaling the gradient of the cost function proves to be an effective method of improving the convergence rate of the VDA minimization process. The impacts of the length of the assimilation interval and the time density of the observations on the convergence rate of the minimization have also been investigated. An improved assimilation was obtained when observations were available in selected segments of the assimilation window. Moreover, our 4D VDA experiments with the SLSI model confirm the results obtained by Navon et al. and Li et al. concerning the impact of the length of the assimilation window. The choice of an adequate lime distribution of observations along with an appropriate length of assimilation interval is an important issue that will he further investigated.