Data assimilation and integration

Many projects at the eScience Center revolve around continuously improving the model we have of a certain natural phenomena, such as a model of the weather or a model of crowd behavior. Models are abstractions and therefore by definition imperfect. To reduce the error propagation in a model, a method we commonly use is data assimilation, which periodically tunes a model to actual data to keep it from diverging from reality. This is usually applied to weather, ocean or climate models, but is also used to accellerate the convergence of machine learning.