Dynamical effects of inflation in ensemble‐based data assimilation under the presence of model error

Covariance inflation is one of the necessary tools enabling success ensemble Kalman filters (EnKFs) in high dimensional spaces and presence model error. Inflation maintains variance to a sufficiently large value, counteracting damping at analysis times its underestimate arising from sampling errors. In this work we investigate effect on dynamics EnKF ensemble. When focus recursive full cycle forecast-analysis-forecast, an apparently counter-intuitive multiplicative appears span EnK. particular, demostrate that changes alignment anomalies onto weakly-stable backward Lyapunov vectors. Whereas expected collapse subspace corresponding unstable portions spectrum, use contributes retain beyond subspace. Given error implies no longer fully confined local subspace, feature paramount importance for optimal filtering. We propose hybrid schemes whereby additive perturbations complement by suitably increasing dimension spanned The improves RMSE Lorenz 96 compared alone, emphasizing role when designing schemes.