The interest to analyse and forecast the sea surface velocity vector field in routine, or regarding peculiar situations is important regarding the objectives of the project. The complexity, the variability of the circulation of the water masses in the Gulf of Lion have already been pinpointed, this fact must be recalled for two reasons. The ship traffic in the area will become much more intensive in the future, as the two major economical poles (Barcelona and Marseille) will develop further their activities and their needs for exchanges with other harbours in the world. This will increase the pressure in terms of traffic security, and will certainly lead to the creation of specific traffic rails, and centres for survey. Such decisions will certainly need advanced information regarding the risks for the traffic, especially the mapping of the areas where current effects may be strong. This point is particularly important to ensure the transport of hazardous materials at sea. In peculiar situations, real time analysis and forecast may be crucial for the operator (ship routing, response to a pollution event), and the availability of synthetic maps of velocity vectors is precious.
Figure 5-5. AVHRR-4 images from 06/02/98 (left) and 17/01/98 (right). The 'blue hole' can be seen in the centre of the Gulf (dark area), the white band along the French coast is the Ligurian current. The two dates correspond to different flows: laminar (left), unstable with meanders (right).
USA authorities have asked since long to the NOAA to provide users in the marine domain with analysis maps based on AVHRR image processing. Such maps generally consist simply in a design of the general water masses (warm/cold currents limits, eddies, meanders, fronts), annotated with typical values (e.g. speeds for gyration and translation for an eddy). At the basis, two processes are being achieved. At first, SST maps are computed by pairs, separated by time intervals of 24 hours maximum. Each couple of images is assimilated into a numerical model, where the SST, i.e. the heat, is assumed to behalf as a passive scalar in the surface current. Such a heat advection model provides velocity maps which are annotated by expert analysts, as achieved in routine mode for meteorological analysis. The resulting maps are faxed to ships and harbours, using the same channels than meteorological bulletins.
Figure 5-6. ERS SAR image from 15/10/98. Natural slicks swirled by coastal eddies, near Barcelona.
This fact is not so well known by Europeans, as it has been well proven by the French skipper Florence Arthaud in 1992. When the sailor woman finished the race ‘La Route du Rhum’, she had to cross the Gulf Stream as fast as possible to the American coast, as she was just behind one other race ship. The winning route was then design using a NOAA SST analysis : to complete the race, the leading vessel had to cross a 150 km warm eddy by its centre, whilst the winner decided simply to ‘surf’ on the good edge of the same vortex. Each day, tankers crews avoid difficult positions using the same type of strategy along the western American coast.
Figure 5-7. Experimental test of the automatic displacement vectors extraction. Left: smoke plumes drifting in a slow air flow are enhanced by a laser tomography. Right: the plumes are detected and tracked, and the velocity vectors of their componentsare computed.
Figure 5-5 shows the interest to develop similar approach for the Gulf of Lion test site. Depending upon the deep water production rate in the Gulf, the flow rate of the Ligurian current (which flows from the Gulf of Genova to the coast of Provence, along the coast) may vary strongly. In the case of the 6 February 1998, the flow is regular, roughly parallel to the coast, suggesting a laminar regime. In the case of the 17 January 1998, the current shows many instabilities, with many meanders, and marginal eddies. It is clear that a numerical modelling of the first situation could lead to a proper result, as the theory of such flows is well achieved, but it is quite clear also that the numerical modelling of the second case — called also a turbulent case - should lead to less reliable results. This is the reason why forecasting centres need analysts in order to make corrections over the modelling results.
Figure 5-6 shows the state of the coastal circulation at a much finer scale, in the case of a turbulent flow. Seen as a zoom into the previous image — despite the fact that the imaged surface phenomena are quite different — the dynamical structure of the eddies at the 100km scale appears there clearly. This also indicates the reason why we may not expect to apply a complete numerical modelling in such situations: the structure, and thus the complexity of the flow is scale-invariant, which should imply the use of a huge — and noiseless — numerical model to achieve a direct simulation of the flow.
The approach we propose in this project is based both on the user needs, and on an empirical approach for the velocity vector retrieval : at first, one need to define properly what is a velocity vector. The end user is generally interested by the maximum velocity component that will be communicated to his ship — or the object he is tracking — by the surrounding flow. In other terms, in the most general case of a turbulent flow, we have to deal here with the component of the flow where the maximum of kinetic energy is available, i.e. the largest scale of the flow itself. It means that the velocity component of a turbulent current that may be of an interest for a user is the velocity vector of the largest coherent structure where the ship — or the oil spill or other pollution object — is floating (meander, eddy, etc.).
This is the meaning of the analysis bulletins available on broadcast by the NOAA. Our purpose is to reduce the time and the costs of analysis by the means of an automatic velocity vector retrieval scheme. Based at first on experimental works at Ecole des Mines (Nerot 1996), it has been demonstrated the possibility to use specific image processing techniques to characterise the coherent structures in non-stationary or turbulent flows (see Figure 5-7). Such algorithms, based on advanced mathematical tools such as the wavelet transform, have been extensively used in this study for the completion of the GIS of dynamical structures in Gulf of Lion. Associated with automatic recognition techniques, such databases allow to track an identified structure present in a couple of successive images (see Figure 5-7). The obtained velocity vector field is quite different from the one provided with numerical models, but gets closer to the type of information needed by the user: what are the important dynamical objects in my operating area, where are they, and what is their typical velocity?
Figure 5-8. Automatic detection for the Cataln thermal front (red polygon along the Spanish coast, centre left) on AVHRR-4 images acquired on 16 and 18 October 1995.
This approach has been used in a few cases, reported in previous reports, leading to original results for the measurement of velocity for peculiar waves (instabilities along thermal fronts, Figure 5-8). The orders of magnitudes for the retrieved velocities were far beyond the climatological current velocities, but revealed to be relevant with the experience of crews of civilian authorities who had to face those situations. This indicates that a more systematic application of the method should be developed, using in situ campaigns, in order to improve the quality and the reliability of its results.
The final aspect of this approach is that it provides an independent way to estimate the kinetic energy available at the largest scale of a turbulent oceanic flow : The method developed at UPC and described in section 5.3.3 to estimate the diffusion mechanisms and time scale at sea surface for pollutants may take some benefit from such information. This should result in a improved scheme for the estimate of turbulent diffusion of surface pollutants using satellite and airborne images.