Forecast assessment in the mediterranean sea : A structure oriented approach

The MFSTEP1 project is an international scientific collaboration program which aims
to create an operational forecasting system for the Mediterranean sea. The simulations
provided at the basin scale are 10 days forecasting fields in a 3-D ocean. The hydrodynamic
model primitive equations are combined with the data assimilation scheme
SOFA2. The data collection is done in a near real time process and the set of XBT and
SLA observations are used in one week assimilation cycle.
The forecast assessment is traditionally realised using classical statistic tools like
RMSE or the bias and the assimilation benefit is estimated by skill scores using as
reference the free model, persistence or also climatology. The process is essentially
based on the comparison of two fields at a fixed time, one corresponding to the simulations
and the other one to the observations. The interest of such statistical methods
comes in the quick and sensitive appreciation they provide about the quality, accuracy
and consistency of the simulation. However this kind of assessment procedure
brings in it self a conceptual contradiction: performances of a dynamical process are
measured using a snap shot view of the ocean state.
A system evolution assessment procedure is carried out within the framework of the
MFSTEP hindcast. The hindcast system is intrinsically analysed (without independent
informations) comparing the background forecast evolution with the abrupt variation
which occurs at the observations assimilation time steps. The system evolution between
two consecutive days is analysed using a decomposition method. The temperature
and salinity fields evolution in a sub-region of theWestern Mediterranean basin is
seen in a structural point of view and decomposed in three elements : a global spatial(2D) displacement which conserves the internal features, a global intensity variation
which expresses the system energy changes, and an internal pattern changes ensemble.
The index of evolution used is a mean squared difference between the two consecutive
simulations. The displacement contribution is estimated after the determination of the
shift (field translation) which minimises the local mean squared difference between
the translated field and the next simulation. The intensity variation contribution is calculated
as the difference of the squared mean fields. The remaining difference after
manipulations is considered as the internal pattern changes contribution to the system
evolution.