Hierarchical EnKFs for stochastic parameters and hyper-parameters inference

GUILLERMO SCHEFFLER; JUAN RUIZ; MANUEL PULIDO

Kobe

Simposio; International Symposium on Data Assimilation; 2019

Model errors in the ensemble Kalman filter are usually accountedthrough the use of multiphysics schemes, additive or multiplicativecovariance inflation, stochastic parameterizations or a combination ofthem. In these approaches, several parameters need to be optimized toproperly assess the model uncertainties. Parameters related to thecovariance structure of a stochastic additive perturbation can not beinferred using the state augmentation approach. We propose ahierarchical data assimilation scheme based on two nested ensembleKalman filter to infer offline optimal values for these type ofparameters. In a low-order system, it was possible to infer theamplitude and covariance structure of the additive stochastic forcingfor different types of covariance formulation. A simplified variant ofthe proposed scheme can also be applied to infer filterhyper-parameters like covariance localization scale and additiveinflation amplitude. These schemes can be used as an a priorioffline optimization of the data assimilation system.