Products Interpreted bulletins	#01 | #02 | #03 | #04 | #05 | #06 #07 | #08 | #09 | #10 | #11 | #12 #13 | #14 | #15 | #16

The oceanography forecaster's tools

There are many complementary professions at Mercator Ocean. The oceanography forecaster intervenes at the end of the operational chain by validating oceanic bulletins as they "come out of the oven" and approving them for publication and distribution to users, while also helping users meet specific needs, for instance for scientific campaigns. In this bulletin, the forecasters at Mercator Ocean talk about themselves and their work.

By Isabelle Charon, forecaster at Mercator Ocean

Contents :

Oceanography forecasters monitor the state of the ocean and provide users with the most reliable and up-to-date information possible. For sailors and fisherman, but also scientists doing research at sea, information on the current and predicted state of the ocean is obviously vital. This is why an operational forecasting department was set up at Mercator, once the first, medium resolution, Atlantic prototype (alias PSY1) had been commissioned in January 2001. Three years later, Mercator Ocean has claimed a place for itself in the world of operational oceanography and scientists have made important breakthroughs in data modelling and assimilation. Mercator forecasters can now use 3 prototypes to produce analyses and 14-day forecasts every week:

• PSY1V2, descendent of the first PSY1 prototype, models the North and inter-tropical Atlantic (up to 20°S), with medium resolution (1/3 of a degree) and assimilates both altimetry and in situ data.
• PSY2V1, the high resolution prototype (1/15°) for the North Atlantic (up to 10°N) and the Mediterranean, which for the moment only assimilates altimetry data.
• PSY2G, the first global prototype, a precursor of the PSY3 prototype, which will model sea ice and the ocean with a resolution of 1/4 of a degree.

These three prototypes are the basic tools which forecasters use to 'monitor' the ocean in order to provide users with maps of currents as well as temperature sections . Previously, forecasters had to examine the results of models by comparing them and evaluating their respective quality. This is done by the validation team which is part of the project team. Mercator scientists strive to do this in what is known as a 'long loop' validation cycle and the team of forecasters in turn try to do their validation through a 'short loop' or real time process.

The purpose of this bulletin is thus to illustrate the validation activity performed by our team, by means of a few practical examples.

These data come from measurements made at sea using various devices (such as Argos floats, CTD, XBT , drifting buoys, anchored buoys, etc). These are temperature and salinity measurements and for drifting buoys and Argo floats, they are measurements of currents (to determine their velocity and direction). At Mercator Ocean, the Coriolis data centre, located at Ifremer, transmits this data in real time.

In addition to their usefulness for the PSY1V2 prototype, which assimilates them directly, these in situ data are very useful for validation purposes, since they are 'field truthing data' which can then be compared with the forecasts of our models (this is particularly true for the PSY2 prototype, which does not yet assimilate this data and where consequently they are used as a totally independent data set).

The differences found between the forecasts of our PSY2 prototype and the 'field truth data' are, moreover, submitted to the public for critical comment in our web validation bulletins, which have been published each week since 1 September 2004. When the differences are positive, this means that the model is 'too hot' or 'too salty'. Inversely, negative differences mean that the model is 'too cold' or 'not salty enough', depending on whether we are referring to temperature or salinity.

The following map shows an example of this with temperature anomalies of our PSY2 model for each observation point. These observations were collected during the week preceding the 6 October run (each week the model is run, in other words the analysis and forecasting programme is run). Even though they can still see data gaps, or 'deserts', the number of observations for the North Atlantic has been increasing over the last few years. An observation campaign is taking place at the moment in the Mediterranean, as can be seen from the 'line' between Greece and Nice. We can also see that during this week, our model overestimated the temperature in the Mediterranean (giving positive differences). These real time maps provide valuable information to forecasters (on condition that an observation has been made for the area under study!) and they eagerly look forward to these measurement campaigns which 'rake' a particular zone, thus enabling scientists to establish general ocean trends and to identify eddies and upwellings.

Map showing differences in temperature calculated by the Mercator model (prototype PSY2) and temperatures measured at sea on 6 October 2004 (click to zoom in)

Along the same lines, the Ovide oceanography campaign undertaken last June by Ifremer, was very rewarding for forecasters. This campaign, which is held every four years, performs measurements along a line between Portugal and Greenland. Throughout the campaign, Mercator Ocean provided maps showing currents, temperatures and salinity to the scientists on board the research vessel, who were able to validate the Mercator forecasts by comparing them with measurements taken from the ship in real time. In return, Mercator received all of the data measured, in order to assimilate them for validating prototypes. The following diagrams can be used to compare our prototypes with the data observed during the campaign. They are temperature sections of the ocean along the ship's route. The same diagram was plotted using Levitus climatology (which is what meteorologists call 'normal seasonal values'). It is easy to see how modelling helps to accurately describe mesoscale structures when compared to methods using climatology alone.

Temperature section: 'field truth' data measured on board (click to zoom in)	Temperature section: Mercator analysis performed by the 1/3° prototype with assimilation of in situ data (click to zoom in)
Temperature section: Mercator 1/15° prototype analysis with no assimilation of in situ data (PSY2V1) (click to zoom in)	Temperature section: Levitus climatology (normal seasonal values) (click to zoom in)

In addition to sea height data measured by altimetry satellites, which is then assimilated by prototypes, forecasters can later also use sea surface temperature data, based on satellite observations and in situ data. This sea surface data is available in the form of a grid, which can be exploited more easily than isolated observations:

• The Sea Surface Temperature (SST) analyses Safo (from Météo France) produce a daily field of sea surface temperature with a higher resolution (1/10°) than that of a 1/2° Reynolds analysis used for 'forcing' the models.
• The Armor analyses (from CLS) produce weekly 3D fields for temperature, salinity and currents down to a depth of -700 m.

These fields are systematically compared with analyses produced by our prototypes. For instance, the following graphs show the difference between the Mercator temperature and the Safo temperature, calculated for each run and averaged for the zone known as 'New Iceland' which is the precise area covered by the Ovide campaign (for which several analysis zones were defined). For the PSY2 high-resolution prototype, the data have been available since 2003 (dotted line) and for 2004 (solid line). For the medium-resolution prototype PSY1V2, data have only been available since 2004. We can see that for this zone, Mercator is too cold in winter and too hot in summer, however with less significant differences for PSY1, which is what had been observed during the Ovide campaign.

Differences between Mercator temperatures (PSY2 prototype) and Safo temperatures

Differences between Mercator temperatures (PSY1 prototype) and Safo temperatures

The monthly means are also monitored to see whether the models are drifting over time and to compare the behaviour of the different prototypes. For instance, an analysis of the month of September 2004 in the Bay of Biscay gave the following maps:

Map showing differences between Mercator temperature (PSY1 prototype) and Safo temperature in September 2004 (click to zoom in)

Map showing differences between Mercator temperature (PSY2 prototype) and Safo temperature in September 2004 (click to zoom in)

It may be observed that the temperatures were, on the whole, over estimated by Mercator, but that the assimilation of in situ data enabled PSY1 to limit some of its imperfections, for instance along the continental shelf in the Bay of Biscay.

More generally, the data is compared to climatological data to closely monitor the evolution of the ocean. This monitoring has, for example, revealed the effect of the 2003 heat wave on temperature in the Mediterranean (see the ocean July 2004 bulletin).

In addition to analyses, forecasters often have to provide forecasts, in other words predictions of the state of the ocean, since what sailors need is to be able to spot the current vein ahead of time, to enable them to 'win' a few extra miles and what research scientists need is to be able to head towards the 'right' eddy to take their measurements.

Each week, the PSY1 and PSY2 prototypes provide daily forecasts for currents, temperature and salinity, covering up to 14 days ahead.

A very simple way of evaluating the quality of these forecasts is to compare them with reanalyses which have been made: the model is run a posteriori, for the same week as for the forecast and with all of the available information (in situ and satellite data). These observations are generally more numerous 'after the event', since there is more time in which to receive them. More particularly, the atmospheric forcing used (surface wind, heat flux, evaporation/precipitation) is based on the actual meteorological situation and not on forecasts. For the moment therefore, reanalysis gives us the best possible knowledge of the state of the ocean.

In this way, a forecasting score, called RMS (root mean square) is defined by statisticians (it characterises the error variation). The score is evaluated for a given zone. The higher the score, the worse the forecast.

An example of variation in score is shown below for the "Prestige" zone (which was tested during the shipwreck of the oil tanker in the Bay of Biscay) for two surface parameters: current (current is a speed, or a vector. It is thus described by its east-west U component and its north-south V component) and the wind stress (in fact, the force created by the wind on the ocean surface, which is also determined according to its U and V components). The score is calculated for the dates D+0 (analysis), D+7 (7-day forecast) and D+14 (14-day forecast).

Examples of forecasting scores for surface current (to the left) and wind strength (to the right). These scores were calculated for the PSY2 prototype, in the Prestige zone, for the period from January 2004 to June 2004. (click to zoom in)

We can see that the quality of the forecast decreases fairly rapidly as we get further and further away in time. Indeed, ocean surface fields are affected very strongly by the meteorological situation and the quality of the forecast will thus depend enormously on the quality of the weather forecast used for forcing the model. Now, the fields used are those used for operational forecasting by the European Centre for Medium-Range Weather Forecasts (ECMWF), which only covers a ten-day period. From 10 to 14 days, the Mercator models thus run with very approximate forcing values. We can see that errors for these currents are closely related to those for the winds and that we cannot reasonably use these forecasts for this zone for periods of more than 7 days.

Conclusion

The team of Mercator oceanography forecasters, who are doing pioneering work in their field, have now acquired sufficient maturity for them to make an objective and reliable judgement of the quality of Mercator prototypes. The production chains are operational and the tools are now there to be used; in addition to real time validation, a quarterly validation of the PSY2 prototype is being published and will soon be complemented by a PSY1 validation. Mercator is becoming more and more reliable and the team as well. The general feeling is one of great enthusiasm and the desire to do even better work.