Ocean satellite data assimilation using the implicit equal-weights variational particle smoother

Abstract The implicit equal-weights variational particle smoother (IEWVPS) is a combination of the particle filter (PF) and weak-constraint 4-dimensional variational (4D-Var) method, that inherits the merits of both. The IEWVPS avoids the filter degeneracy of particle filters through an implicit equal-weights scheme and reduces the root mean square deviations (RMSDs) by introducing the 4D-Var method. This method has been tested using the Lorenz 96 model in a previous study, and we now implement it in the Regional Ocean Model System (ROMS), which is a realistic and complex ocean model. Two key problems, the representation of the analysis error covariance and the choice of the parameter α , were solved during this implementation. With an eddy-permitting model, satellite-based sea surface height (SSH) and sea surface temperature (SST) observations were assimilated with a set of 40 particles IEWVPS scheme. Compared with the ensemble 4D-Var method, the IEWVPS can reduce the bias introduced by perturbed atmospheric forcing, effectively improving temperature simulations in the upper 50 m while maintaining the RMSD of SSH at the same level. Therefore, the cooling effect caused by typhoons in the upper ocean is better characterized under the IEWVPS scheme than with previously used method. The ratio of RMSD to the ensemble spread indicates that the ensemble quality of the IEWVPS is much better than that of the ensemble 4D-Var. In addition, the computational cost of the IEWVPS is only slightly larger than that of the ensemble 4D-Var. One additional tangent linear model integration, one additional nonlinear model integration, and perturbation fields inputs/outputs are still needed.