Forecasting Skill Assessment of a Doubly, Two-Way Nested Model of the Ligurian Sea driven by Assimilation of Sea Surface Temperature

The GHER 3D primitive equation model is implemented with threedifferent resolutions: a low resolution model (1/4Ž°) coveringthe whole Mediterranean Sea, an intermediate resolution model (1/20Ž°)of the Liguro-Provençal basin and a high resolution model (1/60Ž°)simulating the fine mesoscale structures in the Ligurian Sea. Boundaryconditions and the averaged fields (feedback) are exchanged betweentwo successive nesting levels.
The model of the Ligurian Sea is also coupled with the assimilationpackage SESAM. It allows to assimilate satellite data and in situobservations using the local adaptative SEEK (Singular EvolutiveExtended Kalman) filter. Instead of evolving the error space by thenumerically expensive Lyapunov equation, a simplified algebraicequation depending on the misfit between observation and modelforecast is used.
Starting from the 1st January 1998 the low and intermediate resolutionmodels are spun up for 18 months. The initial conditions for theLigurian Sea are interpolated from the intermediate resolutionmodel. The three models are then integrated until August 1999. Duringthis period AVHRR Sea Surface Temperature of the Ligurian Sea isassimilated. The results are validated by using CTD and XBT profilesof the SIRENA cruise from the SACLANT Center.
In a second validation exercise,the AVHRR SST and model forecast %(not yet affected by the SST used for validation) are decomposed at different scalesusing a horizontal wavelet transform.The processes characteristic for each scale are isolated in the model and the SST images.The error statistics calculated on the wavelet amplitudes can thus be related to the ability of the model in forecasting the correspondingmarine processes.
In a second validation exercise, the AVHRR SST and model forecast areboth decomposed at different scales using a horizontal wavelet transform.The characteristic processes for each scale are isolated in the modeland the SST images. The error statistics calculated on the waveletamplitudes can thus be related to the ability of the model inforecasting the corresponding marine processes.