Data assimilation combines observational data with numerical simulation models. The methodology allos to improve the initialization of model predictions, determining model deficiencies, but also to enhance data sets by augmenting the data with dynamical information from numerical models simulating e.g. ocean physics or biogeochemistry. This combination can fill data gaps by an interpolation which accounts for the dynamical information provided by the numerical model. Further the observed information can be used to improve unobserved variables, and even fluxes. This is accomplished through the use of dynamically estimated cross-covariances between the observed and unobserved variables. The assimilation can result in data sets which, at the resolution of the model, exhibit smaller errors than using the observations or the model alone. I will discuss the method of ensemble-based data assimilation on the example of ocean-biogoechemical modeling with the assimilation of satellite ocean color data.