NewsChronicles

Evolutions on the medium-resolution Atlantic prototype (1/3°, PSY1V2)

By Mounir Benkiran, oceanographer at Mercator Océan

This technical chronicle presents the evolutions made on the Atlantic medium-resolution (1/3°) prototype (PSY1V2), assimilating altimetry and in situ data. This new version is operational since February, 16th, 2005.

Contents:

• The problem and the implemented solution
• Benefits
• Operational production
• Useful links

December 2004 : the forecasters team pointed out an anomaly on the PSY1V2 bulletins: the system forecasted unusual cold water bubbles in the North Atlantic. These bubbles corresponded to cold water upwelling, affecting the surface, but also temperatures and salinities in-depth, particularly in the area where temperature strongly decreases (thermocline). These anomalies are visible on the 08/12/2004 temperature map, and, deeper, on the 08/12/2004 salinity map at 96m depth.

PSY1V2 Temperature , at 6m, on 08/12/2004

PSY1V2 Salinity, at 96m, on 08/12/2004

These errors were due to the parameterization of the multi-variate assimilation system. This one was requiring a high computer coast and was implemented sparingly. Computer optimizations on the assimilation platform were led which made possible significant improvements in the parameterization:

• By taking into account 20 empirical orthogonal functions instead of 10. This allows to better project on the vertical, the differences between model and data, which will correct the surface analysis (where there is a lot of data) towards the bottom.
• By using all the in situ profiles temperature and salinity data on the vertical, which was not the case before (we took only one data on two on the vertical).
• By assimilating SST Reynolds with a spatial resolution of 1°x1°. Before we took into account these SST with a 2°x2° resolution.

With this new parameterization, a new simulation has been led starting in 2003. The analysed situation on 8/12/2004 shows significant improvements (see figures below). First, the cold and less salted bubbles don't appear anymore. Secondly, noisy little structures have also disappeared, particularly in the tropics. General circulation doesn't change, but the great structures are not represented in the same way. The front cross the Gulf Stream is getting thinner and so is it along the Europ and North American coasts. In the tropical area, water masses distribution are more regular and make better assessment of the zonal extension of the great currents.

New PSY1V2 temperature, at 6m, on 08/12/2004

New PSY1V2 salinity, at 100m, on 08/12/2004

This new simulation has been validated. Figures below show the lesser necessity to correct the model every week. The root mean square (RMS) between model and in situ data has been plotted all over 2003 (before assimilation). In the old version (left), differences up to 2°C can be seen in the first 150 meters. They reach max 1°C in the new version. In the same way, differences greater than 1°C were observed down to 1,000 meters depth. Such discrepancies do not exist anymore in the new version: errors are reduced in surface and in depth. Temperature and salinity fields are then closer to the measured reality.

Root mean square (RMS) of differences between model and measurements according to time (horizontal axis) and to depth (vertical axis)

Idem with the new version

This new PSY1V2 version has replaced the old version in the operational chain since February, 16th, 2005. From this date, on-line bulletins correspond to this version. Earlier bulletins (until February, 9th, 2005) correspond to the previous version.

• Multivariate and multidata assimilation in Mercator, Mounir Benkiran, Mercator Newsletter n°13
• In situ data assimilation, Mercator chronicle
• On-line PSY1V2 bulletin