Estimation of stochastic parameters with a nested ensemble Kalman filter

JUAN RUIZ; MANUEL PULIDO; GUILLERMO SCHEFFLER

Workshop; 7th WMO data assimilation symposium; 2017

WMO

A hierachical data assimilation approach based on the ensemble Kalman filter isproposed for inferring stochastic parameters -parameters of the model errorcovariance matrix- in a highly non-linear low order model. The proposed techniqueis based on the Rao-Blackwellization of the estimation problem. The premise of thetechnique consists of using an ensemble of ensembles, where each of the ensembles isassociated to an ensemble Kalman filter with the same characteristic as the dataassimilation system whose parameters are being estimated. We demonstrate theability of the technique to infer parameters related to the variance and spatialcovariances of stochastic representations of model error in the Lorenz-96 dynamicalsystem. The evaluation is conducted for stochastic twin experiments and onimperfect model scenarios with unresolved physics in the forecast model. Theproposed technique is successfully evaluated among different possible model errorcovariance structures. The technique is not thought to be applied operationally butoffline, as part of an a priori parameter optimization of the data assimilationscheme, which can be further extended to the estimation of data assimilationhyperparameters.