Evolving topics in data assimilation
• How can we use ensembles to improve forecasts, skill vs. spread, etc.?
• How can we use the data assimilation cycle to identify and correct model bias?
• How should OSSEs be designed?
• Can we solve variational data assimilation with nonlinear obs operators by using conjugate gradient minimization?
• What can we do when a feature (a storm) is in the wrong place in the background?
• What can we do when only half of a feature is observed by a satellite?
• Would it be possible/better to use radiances in data assimilation instead of retrievals?
• Can we solve the 4d-VAR problem?
I will describe some advances in some of these areas and make some predictions of the future evolution of data assimilation. In principle, modern approaches solve these issues. In practice, approximations and short cuts are required and keeping these concerns in mind can help to understand results from modern systems, and point the way to improving them.