Assimilation of Sea Surface Temperature predicted by a satellite-based forecasting system in a doubly nested primitive equation model of the Ligurian Sea

Data assimilation is traditionally used to combine model dynamics and observations in a statistical optimal way. Assimilation of observations improves therefore hindcasts and nowcasts of the ocean state than otherwise obtained by the model alone. The observational constraints are necessary to reduce uncertainties and imperfections of the ocean model. Due to the obvious lack of future observations, the model forecast cannot be controlled by observations and the predictive skill degrades as the forecast time lag increases. The error grow is not only caused by the chaotic nature of the system but also by the biases and drifts of the model. The later part can be reduced by considering different models with different imperfections. Data assimilation provides the statistical frame for merging the different model results. A primitive equation model of the Mediterranean Sea (1/4° resolution) has been implemented with two successive grid refinements of the Liguro-Provençal Basin (1/20°) and the Ligurian Sea (1/60°) respectively (Barth et al, 2003). The dependence of the ``parent'' model and the embedded ``child'' model is bi-directional; it involves the exchange of boundary conditions and feedback between the models. Alvarez el al. (2004) developed a statistical predictor for forecasting the SST of the Ligurian Sea with a time lag of 7 days based on the previous remote sensed SST. The degrees of freedom of the SST are reduced by an Empirical Orthogonal Function (EOF) analysis. A genetic algorithm trained by the historical SST evolution in the Ligurian Sea is used to predict the EOF amplitudes. Observed and forecasted SST are assimilated in the hydrodynamic model and the results of this two experiments are compared to the model run without assimilation. The assimilation of the forecasted SST reduces the error of the model by an amount comparable to the assimilation of real SST, showing the potential of skill improvement of combining statistical and hydrodynamic models.