A Global Approach to the Identification of Model Noise

Tuesday, 16 December 2014: 2:10 PM
Alexandre J Chorin, University of California Berkeley, Berkeley, CA, United States and Kevin Lin, University of Arizona, Mathematics, Tucson, AZ, United States
Data assimilation, e.g. in meteorology and geophysics, requires an estimate of the model noise, which is typically hard to find, in particular because the noise depends on the signal, i.e. on the values of the variable one is trying to estimate, and is difficult to separate from the signal. A global approach to the estimation of noise from data will be presented, in which the system of interest is imbedded in a stationary ensemble, and then the noise is separated from the signal via a reversible transformation of the ensemble. The global approach is particularly useful if the observation noise is significant. As an example, a global method is used to estimate the noise in a truncation of the Lorenz 96 model, and the results are compared with earlier work on stochastic parametrization and related constructions.