PSY1v2 system:

The first level comes naturally from the multivariate nature of the PSY1v2 system: model forecast is sytematically compared to the observations, to form the innovation (the misfit) at the rigth plce at the rigth time. Thi innovation is then used by the OI to update the guess, but the direct analysis of the innovation is already a good validation. Bias and rms, and more generally all forecast scores, are computed (on the vertical with in situ data) for several regions and the basin, and their evolution with cycles can be checked. This helps to point out some poor performances. Innovation maps can then be used to locate the anomalies (bad forecast/doubtful data). In particular, geographical data inconsistency can be revealed at this stage. Finally, individual profiles can be compared for the forecast, the data and the climatology. A doubtful data can even be put on a black list (discarded). This first level validation is done for every cycle.

The second level of validation concerns analysis of time series that are significantly longer than the assimilation cycle. For instance, yearly departure from climatology can be compared to the data. Time series and vertical sections derived from in situ data help to identify the process (eddy, wave, upwelling ...) that cause the system faults (shift in location or intensity). System error bars, time correlation and spatial coherenceare sytstem statistics that help to detect some regional problems. Some special data such as tide gauge and driting buoys provide additional information on the system performances. This second level validation is done for every season of assimilation, and summerized for each year.

The third level of validation concerns checks that are indirect estimations of the system performances. For instance, consistency of the system can be diagnosed through a comparison with hypothesis. In particular, the level of error can be compared to the observational error. A posteriori parameters can be computed and compared to a priori values.. Finaly, transports, overturning and propagations can be compared to the current knowledge. This third level validation is done for every year of assimilation.