Toward background error covariance hybridization for climate prediction

The Norwegian Climate Prediction Model (NorCPM) combines the Norwegian Earth System Model (NorESM) with the Ensemble Kalman Filter (EnKF) and aims at providing seasonal to decadal climate predictions. On nowadays supercomputer, it is not computationally tractable to run more than 30 members (and 5 members with the high resolution version of NorCPM), which results in sampling issues when estimating the background error covariance matrix.

To overcome these issues, an hybridization method derived from previous work from (Hamill and Snyder, 2000) has been used and led to the implemeantion of 2 methods: climatological hybridization and dual resolution. These 2 methods allow for a reduction of sampling error when compared to standard EnKF.

The hybrid covariance method are tested with the quasi-geostrophic model within the DAPPER package. It is shown that the method outperforms the standard implementation of the EnKF in particular for small ensemble size.

Further work will assesses the performance of the two methods with NorCPM in the context of twin experiments.