A Comparison of Three Ensemble Kalman Filters Using a Thousand-Member AGCM Ensemble

Abstract:

We employ a state-of-the-art atmospheric general circulation model (the Community Atmospheric Model, version 6) to compare the performance of three ensemble Kalman filters using ensemble sizes an order of magnitude larger than typically used in operational settings. This setup allows us to address issues such as sampling errors and evaluate the relative importance of localization and inflation. While ensemble data assimilation approaches typically employ ensemble sizes of roughly one hundred members, this study employs a thousand-member ensemble to approximate the forecast error covariance matrix and uses it to implement a stochastic ensemble Kalman filter (EnKF). We compare the performance of this stochastic EnKF against two other filters: the ensemble adjustment Kalman filter and an EnKF with exact second-order observation perturbation sampling (EnKF-esops). The EnKF-esops aims to mitigate observation perturbation sampling errors in the stochastic EnKF.