Assimilation of SST data in the POSEIDON system using the SOSSTA statistical-dynamical observation operator

Abstract. In spite of their long-standing availability, the optimal assimilation of sea surface temperature (SST) observations retrieved from infrared and microwave space-borne sensors is still challenging in oceanographic forecast systems. One prominent problem stems from the fact that ocean general circulation models do not resolve the diurnal variability of SST data as measured by satellites. In order to improve SST data assimilation schemes and enhance the exploitation of swath SST data, an observation operator capable of representing the SST diurnal cycle is introduced and called SOSSTA. Firstly, a one-dimensional turbulence model is used to produce a data set of upper ocean temperature profiles with corresponding skin and subskin SSTs. A canonical correlation analysis is then used to extract the maximally correlated modes of variability between temperatures at depth and skin/subskin SST, conditioned to atmospheric state (insolation and wind speed). These canonical correlations form the novel observation operator, which is implemented in the POSEIDON model forecasting system (Aegean Sea) to test the assimilation of daytime SST retrievals from the SEVIRI infrared radiometer. Comparison of misfits (off-line assessment) suggests that the new operator outperforms the mere use of the first model level to calculate SST innovations. Real-world data assimilation experiments indicate that the use of the SOSSTA operator is beneficial to the skill scores and in particular improves the sea surface height analysis and forecast skill scores, whose improvement is maintained throughout a one year long experiment.