Comparison of different incremental analysis update schemes in a realistic assimilation system with Ensemble Kalman Filter

• Published in: Ocean Modelling
• Main Authors: Yan, Yajing, Barth, Alexander, Beckers, Jean-Marie, Brankart, J. M., Brasseur, P., Candille, G.
• Other Authors: GHER
• Format: Article in Journal/Newspaper
• Language: English
• Published: Elsevier Ltd 2017
• Subjects: Comparison; Ensemble Kalman Filter; Incremental analysis update; Probabilistic metrics; Bandpass filters; Integration; Continuous ranked probability scores; Ensemble forecast systems; Equilibrium modeling; Incremental analysis; North Atlantic Ocean; Kalman filters; Kalman filter; Atlantic Ocean; Atlantic Ocean (North); Physical; chemical; mathematical & earth Sciences; Earth sciences & physical geography; Physique; chimie; mathématiques & sciences de la terre; Sciences de la terre & géographie physique; North Atlantic
• Online Access: https://orbi.uliege.be/handle/2268/232437; https://orbi.uliege.be/bitstream/2268/232437/1/1-s2.0-S1463500317300653-main.pdf; https://doi.org/10.1016/j.ocemod.2017.05.002

peer reviewed In this paper, three incremental analysis update schemes (IAU 0, IAU 50 and IAU 100) are compared in the same assimilation experiments with a realistic eddy permitting primitive equation model of the North Atlantic Ocean using the Ensemble Kalman Filter. The difference between the three IAU schemes lies on the position of the increment update window. The relevance of each IAU scheme is evaluated through analyses on both thermohaline and dynamical variables. The validation of the assimilation results is performed according to both deterministic and probabilistic metrics against different sources of observations. For deterministic validation, the ensemble mean and the ensemble spread are compared to the observations. For probabilistic validation, the continuous ranked probability score (CRPS) is used to evaluate the ensemble forecast system according to reliability and resolution. The reliability is further decomposed into bias and dispersion by the reduced centred random variable (RCRV) score. The obtained results show that 1) the IAU 50 scheme has the same performance as the IAU 100 scheme 2) the IAU 50/100 schemes outperform the IAU 0 scheme in error covariance propagation for thermohaline variables in relatively stable region, while the IAU 0 scheme outperforms the IAU 50/100 schemes in dynamical variables estimation in dynamically active region 3) in case with sufficient number of observations and good error specification, the impact of IAU schemes is negligible. The differences between the IAU 0 scheme and the IAU 50/100 schemes are mainly due to different model integration time and different instability (density inversion, large vertical velocity, etc.) induced by the increment update. The longer model integration time with the IAU 50/100 schemes, especially the free model integration, on one hand, allows for better re-establishment of the equilibrium model state, on the other hand, smooths the strong gradients in dynamically active region. © 2017 Elsevier Ltd