Data assimilation methods

Through the SEAMLESS project new and advanced data assimilation methods have been developed for all CMEMS Marine Forecasting Centres (MFCs).

Global and Iberian-Biscay-Irish seas MFC

The current Global (GLO) MFC uses a SEEK filter with prescribed covariances to assimilate ocean colour in reanalysis simulations, which are also nudged to climatological nutrient fields. No biogeochemical analysis/forecast is performed with the GLO system and no biogeochemical assimilation at all is performed in the IBI system. In SEAMLESS, the Université Grenoble Alpes enhanced a stochastic ensemble-based system for the NEMO-PISCES model used in both the GLO and IBI MFCs. The system is based on 4D analysis with the LETKF for the assimilation of both altimetry and ocean colour data.

Table 3.1. Comparison of the CMEMS DA methods before (dark grey) and after recent developments (light grey) undertaken in the 3D MFC configurations, error covariance schemes, updated biogeochemical variables (PHY= phytoplankton, Chl = chlorophyll, N = nitrate, P = phosphate), and uncertainty estimation approaches.

The current Arctic Sea (ARC) MFC only includes a direct insertion of ocean colour data in the real-time analysis system, and a Deterministic Ensemble Kalman filter and smoother (DEnKF/EnKs) for the reanalysis of the past ocean state. In SEAMLESS, NERSC advanced the ARC MFC by introducing the estimation of new biogeochemical parameters along with the biogeochemical in the DEnKF/EnKs system.

Baltic MFC

The current Baltic (BAL) MFC delivers reanalysis by means of the LSEIK and using prescribed covariance matrices. This system assimilates in situ biogeochemical profiles, by using PDAF with the NEMO3.6-SCOBI model. No assimilation is produced with the NEMO-ERGOM analysis and forecast system. In SEAMLESS, the Alfred Wegener Institute developed a new ensemble, multivariate assimilation system for NEMO-ERGOM, based on the LESKTF and the Hybrid Filter LKNETF implemented in PDAF, for the assimilation of both ocean colour and sea surface temperature.

North West Shelf-Seas MFC

The current North West Shelf-Seas (NWS) MFC assimilates ocean colour in real-time analysis and forecasts and reanalysis of the past ocean state, by using the 3D-variational assimilation of the NEMOVAR software. In SEAMLESS, Plymouth Marine Laboratory developed a new system with ensemble-3DVAR (hybrid) capacity for the biogeochemical component of the NWS system. The ensemble is generated by perturbing the six most sensitive biogeochemical model parameters identified in SEAMLESS, using an available ensemble of physical states.

Mediterranean Sea MFC

The current Mediterranean Sea (MED) MFC uses a 3D-variational assimilation system to assimilates ocean colour for reanalysis and both ocean colour and biogeochemical-Argo float data in real-time analysis/forecasts. In SEAMLESS, the The National Institute of Oceanography and Applied Geophysics (Italy) developed a new (pre)operational assimilative modelling system based on the ensemble approach of the SEIK filter.

System	DA method	Error covariances / Error subspace	Updated BGC variables	Uncertainty estimation
NWS CMEMS	NEMOVar	Prescribed stats (monthly climatol.)	Univariate DA (Chl) + balancing scheme	No
NWS SEAMLESS	Hybrid Ensemble/NEMOVar	3D ensemble-based	Univariate DA (Chl) + balancing scheme	Ensemble spread
IBI CMEMS	None
GLO CMEMS	SEEK Filter Fixed basis	Prescribed stats (seasonal climatol.)	Bi-variate (PHY + N) + adjustment scheme	No
N Atlantic SEAMLESS	Stochastic Ensemble Filter	4D ensemble-based (space + time)	Full state vector	Ensemble spread
MED CMEMS	3DVarBio	Prescribed stats	Multivariate (PHY+N+P)	No
MED SEAMLESS	SEIK	3D ensemble-based	Multivariate (PHY+N+P) or Full state vector	Ensemble spread
BAL CMEMS	None
BAL SEAMLESS	LESKTF & Hybrid Filter LKNETF	3D ensemble-based	Multivariate (Chl+ 3 phytoplankton variables) or Full state vector	Ensemble spread
ARC CMEMS	DEnKF/EnKS	3D ensemble-based	Full state vector	Ensemble spread
ARC SEAMLESS	DEnKF/EnKS updated	3D ensemble-based	Full state vector + BGC model parameters	Ensemble spread