Efficient data assimilation into a complex, 3-D physical-biogeochemical model using partially-local Kalman filters

Abstract. Advanced Kalman filtering techniques were used to assimilate pseudo ocean color and profile data into a complex, three-dimensional coupled physical (POM)-biogeochemical (ERSEM) model of the Cretan Sea ecosystem. The assimilation schemes, the Singular Evolutive Partially-Local Extended Kalman (SEPLEK) filter and its variant called SFPLEK, are based on the standard SEEK filter in which the Kalman correction is made along a set of "global" and "local" directions, determined via a so-called "global-local EOF analysis". The global functions are used to represent the ecosystem large-scale variability. They are allowed to evolve in time in the SEPLEK filter to follow changes in the model dynamics, while they remain invariant in the SFPLEK filter. The local functions always remain invariant and are determined in such a way as to independently represent the different spatial regimes of the ecological model. This helps to improve the estimation of fine-scale variations while requiring significantly less computational time compared to the SEEK filter.

Several assimilation experiments were performed to assess the relevance of the assimilation system and to study its sensitivity to different choices of global/local EOFs. The SFPLEK filter was used in all the sensitivity experiments in order to efficiently measure the representativeness of the different set of correction directions, as well as to save computational time. Assimilation results suggest that the use of global-local correction directions clearly enhances the filter's performance under different assimilation setups. The choice of the local directions should, however, be carefully considered, taking into account the model regional variability and the characteristics of the observational system.