In this work three versions of the Singular Evolutive Extended Kalman Filter (SEEK) filter are applied to a 1D implementation of a marine ecosystem dynamics model in two locations of the Northern Adriatic Sea to assimilate biogeochemical data. The scope is to gain insight in the benefit of the various levels of error covariance propagation: (1) the forgetting factor version, propagating analysis error correction directions and covariance matrix in reduced space, (2) a simplified version of the former, propagating the covariance matrix in reduced space only and (3) a new version that is proposed in this article abandoning the concept of forgetting factor for a more explicit approach in the approximation of the model noise covariance by statistical means. Twin experiments are presented comparing the various filters along with a free run and a non propagating scheme corresponding to an optimal interpolation to quantify the benefit of these sophisticated, but computationally heavier filters with respect to a simpler approach. The obtained results clearly show that the improvements achieved through the more advanced formulations of the propagation scheme are consistent with the level of sophistication in the design. The results for the filters with full propagation also overcome some unstable behaviour observed for the semi-propagating filter. The filter with statistical treatment of the dynamic noise further improved the results of the version with forgetting factor and full propagation showing a quicker convergence towards the ‘‘true solution’’ in the framework of the twin experiments.

A comparison of different versions of the SEEK filter for assimilation of biogeochemical data in numerical models of marine ecosystem dynamics. Ocean Modelling, 54-55

ZAVATARELLI, MARCO
2012

Abstract

In this work three versions of the Singular Evolutive Extended Kalman Filter (SEEK) filter are applied to a 1D implementation of a marine ecosystem dynamics model in two locations of the Northern Adriatic Sea to assimilate biogeochemical data. The scope is to gain insight in the benefit of the various levels of error covariance propagation: (1) the forgetting factor version, propagating analysis error correction directions and covariance matrix in reduced space, (2) a simplified version of the former, propagating the covariance matrix in reduced space only and (3) a new version that is proposed in this article abandoning the concept of forgetting factor for a more explicit approach in the approximation of the model noise covariance by statistical means. Twin experiments are presented comparing the various filters along with a free run and a non propagating scheme corresponding to an optimal interpolation to quantify the benefit of these sophisticated, but computationally heavier filters with respect to a simpler approach. The obtained results clearly show that the improvements achieved through the more advanced formulations of the propagation scheme are consistent with the level of sophistication in the design. The results for the filters with full propagation also overcome some unstable behaviour observed for the semi-propagating filter. The filter with statistical treatment of the dynamic noise further improved the results of the version with forgetting factor and full propagation showing a quicker convergence towards the ‘‘true solution’’ in the framework of the twin experiments.
Butenschoen M.; Zavatarelli M.
File in questo prodotto:
Eventuali allegati, non sono esposti

I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificativo per citare o creare un link a questo documento: https://hdl.handle.net/11585/126466
 Attenzione

Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo

Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 1
  • ???jsp.display-item.citation.isi??? 1
social impact