Can data assimilation be successful in geophysics?
IPGP - Îlot Cuvier
Séminaires communs Géomagnétisme-Paléomagnétisme
Department of Mathematics, University of California, Berkeley
The task in data assimilation is to use data to update the prediction of a mathematical model. In theory, the solution of this problem is the model state conditioned on the available data, however difficulties can arise in the computation of this random variable. I will discuss the two main difficulties of numerical data assimilation in geophysical applications: the large dimension of the state and the nonlinearity of the underlying dynamics. It will become clear that numerical data assimilation can be successful for linear systems even if the state dimension is huge (provided the assumptions on the noises are reasonable). I will also point to methods which show promise when coping with nonlinearity.