P. Craneguy

ABSTRACT A study of sequential data assimilation of satellite derived sea surface temperature (SST) in the free surface primitive equation model MARS-3D [Lazure and Dumas, 2007] using Ensemble Kalman Filter [Evensen, 2003] is presented with application to the Bay of Biscay. Skill assessment of the data assimilation system is analysed over April-July 2006, a period for which independent temperature and salinity profiles are available over the Continental shelf. The spatial and temporal structure of forecast errors is investigated using an ensemble modelling approach (Monte-Carlo). Multivariate ensemble forecast statistics associated by distinct model error sources (wind forcing, model parameters) are shown to be neither homogeneous over the Continental shelf nor stationary. In this large space dynamical system, localization and filtering of small-sized ensemble correlations is needed to provide a consistent result for EnKF analysis. The localization used is inversely proportional to the bottom depth. Statistical analysis of the ensemble forecast reliability also reveals that SST forecast errors over the Continental Shelf of the Bay of Biscay are season-dependant: in spring they are mainly governed by the fraction of light loss due to scattering and absorption (extinction coefficient) which occurs over the Loire and Gironde plumes although they are dominated by wind stress and ocean mixing errors in summer. The potential of sequential data assimilation of SST to improve T-S model predictions over the shelf is investigated using independent in-situ temperature and salinity profiles over spring and summer test periods. The data assimilation system provides significant error reduction compared to the non assimilative one for temperature and salinity over the shelf. The efficiency of combined parameter and state estimation to reduce the SST model forecast biases over the shelf is shown over April-May, a period for which the forecast error is mainly governed by the extinction coefficient.

A study of sequential data assimilation of satellite derived sea surface temperature (SST) in the free surface hydrodynamic model MARS-3D [Lazure and Dumas, 2008] using Ensemble Kalman Filter [Evensen, 2003] is presented with application to the English Channel and the shelf of the Bay of Biscay. We focused our efforts on summer 2006, when observations are numerous and variability is high. We first identified uncorrelated key parameters of the model using a generalized sensitivity study. We found that the forecast ensemble generated by perturbations of those key parameters (extinction and turbulence closure coefficients, bottom roughness) is statistically consistent with model errors, but provide an ensemble with underestimated spread. Introducing errors in the initial conditions and in the atmospheric forcing (thanks to the ECMWF ensembles) significantly increases the ensemble variance. However, remote ensemble correlations highly suggest the needs of covariance localization spatial...

ABSTRACT A study of sequential data assimilation of satellite derived sea surface temperature (SST) in the free surface primitive equation model MARS-3D [Lazure and Dumas, 2007] using Ensemble Kalman Filter [Evensen, 2003] is presented with application to the Bay of Biscay. Skill assessment of the data assimilation system is analysed over April-July 2006, a period for which independent temperature and salinity profiles are available over the Continental shelf. The spatial and temporal structure of forecast errors is investigated using an ensemble modelling approach (Monte-Carlo). Multivariate ensemble forecast statistics associated by distinct model error sources (wind forcing, model parameters) are shown to be neither homogeneous over the Continental shelf nor stationary. In this large space dynamical system, localization and filtering of small-sized ensemble correlations is needed to provide a consistent result for EnKF analysis. The localization used is inversely proportional to the bottom depth. Statistical analysis of the ensemble forecast reliability also reveals that SST forecast errors over the Continental Shelf of the Bay of Biscay are season-dependant: in spring they are mainly governed by the fraction of light loss due to scattering and absorption (extinction coefficient) which occurs over the Loire and Gironde plumes although they are dominated by wind stress and ocean mixing errors in summer. The potential of sequential data assimilation of SST to improve T-S model predictions over the shelf is investigated using independent in-situ temperature and salinity profiles over spring and summer test periods. The data assimilation system provides significant error reduction compared to the non assimilative one for temperature and salinity over the shelf. The efficiency of combined parameter and state estimation to reduce the SST model forecast biases over the shelf is shown over April-May, a period for which the forecast error is mainly governed by the extinction coefficient.

ABSTRACT This study deals with the development of time-evolving multivariate data assimilation of satellite derived sea surface temperature (SST) and T-S profiles over the continental shelf. This work is being conducted in the framework of the PREVIMER project (www.previmer.org), whose primary objective is the development of an operational forecasting system for the coastal environment along the French coastlines. This presentation discloses a general overview of the project over the period 2008-2012, but it will focus on the results obtained during the initial phase of the project with respect to sequential data assimilation of satellite derived sea surface temperature (SST). This SST data assimilation in the free surface primitive equation model MARS-3D uses Ensemble Kalman Filter (EnKF): it is tested over the Bay of Biscay and the Gulf of Lion. Skill assessment of the data assimilation system is analysed over April-July 2006, a period for which independent temperature and salinity vertical profiles are available over the Biscayan continental shelf. Preliminary results of a similar data assimilation experiment for the Gulf of Lion are also discussed over April-July 2005. The spatial and temporal structure of forecast errors is investigated using an ensemble modelling approach (Monte-Carlo). Multivariate ensemble forecast statistics associated with distinct model error sources (wind forcing, model parameters) are shown to be neither homogeneous over the continental shelf nor stationary. In this large space dynamical system, localization and filtering of small-sized ensemble correlations is needed to provide consistent results through EnKF analysis. The localization used is proportional to the bottom depth. Statistical analysis of the ensemble forecast reliability also reveals that SST forecast errors over the Biscayan continental shelf are season-dependant: during spring, they are mainly governed by the fraction of light lost by scattering and absorption (extinction coefficient) which is due to the Loire and Gironde rivers plumes; during summer, they are dominated by the uncertainties over wind stress and ocean mixing. The potential of sequential data assimilation of SST to improve T-S model predictions over the shelf is investigated, using independent in-situ temperature and salinity profiles over the spring and summer test periods. The data assimilation system provides significant error reduction compared to the non assimilative one, for temperature and salinity over the shelf Finally, the efficiency of combined parameter and state estimation to reduce the SST model forecast biases over the shelf is shown over April-May, a period for which the forecast error is mainly governed by the extinction coefficient.