What is the key of Data Assimilation?

Data assimilation is a technology which improves the accuracy of model simulation by fusing information from multiple sources. There are a variety of data assimilation approaches, such as Kalman filter, Particle filter and other 3D-Var, 4D-Var algorithms. It hard to evaluate the effect of data assimilation in a operation way since the huge field work. Most of papers published in the past decade were talking about algorithms borrowed form other fields. Very little of them considered the error evaluation of information for data assimilation, including model simulations and satellite observations. It is not difficult to fuse information for data assimilation purpose if the uncertainties are known. But if we do not know these uncertainties, the benefit of data assimilation is hard to evaluate. Ensemble Kalman filter was developed for solving this problem. It provides a way to estimate simulation uncertainties by disturbing forcing for an ensemble of simulation. But How to perturb forcing is also a problem. The key to successful data assimilation is  estimate uncertainties of model simulation and satellite observation reasonably.