5.3 Climatologies

Satellite have two significant strengths over most conventional sampling techniques, firstly, a single sensor sampling a large area will tend to provide more consistent data than a range of different sensors from which an overview is derived. Secondly, international co-operation is commonplace in support of calibration and validation activities, leading to repeated calibration of instruments and observations. This latter point in particular helps to ensure that long time series of observations from a small number of sensors are internally consistent. Although for some instruments, and the information derived from them, a full and rigorous validation of the derivation process is ongoing and has not yet been resolved fully, in many cases partially reliable information can be used to good effect. Accepting that all observations may not be 100% reliable but the ensemble of many such observations may supply confidence in results and value to the project, several climatologies have been studied and are presented here.

5.3.1 Oil spills

During the first two project years, 220 SAR images have been acquired over the Baltic Sea test area, 207 SAR images over the North Sea test area, and 282 SAR images over the test area in the north-western Mediterranean Sea (Gulf of Lion). We have analysed every SAR image with respect to the occurrence of marine oil pollution. The locations and sizes of the detected oil pollution have been calculated and catalogued.

As a result, we included circles into the three maps shown in Figure 5-9 at the locations of the detected oil spills. The sizes of the circles are proportional to those of the oil spills, which range from 0.1 km² to 56 km² and the centre of the circles corresponds to the weighted centre ("centre of mass") of the oil spills. Especially for the Baltic Sea test area (left panel) it can clearly be seen that the oil spills follow the main ship traffic route going from south-west to north-east. However, the highest local occurrence of marine oil pollution was found in the north-western Mediterranean, south of Barcelona, where the outflow of the river Llobregat seems to cause a high incidence of pollution in that particular area. As already shown in Chapter 4.4.1, most of the pollution was observed during the summer period (April — September), see the green and yellow circles in Figure 5-9.

Figure 5-9. Maps of the three test areas Baltic Sea (left), North Sea (middle), and Mediterranean (right). The circles denote the locations of the detected oil spills, the circle sizes are proportional to the respective sizes of the detected oil spills, and the colour-coding gives the time of the year (namely the month) when the pollutioon has been detected.

Another finding is, that we detected more pollution during ascending (morning) passes than during descending (evening) passes. As an example, contour plots showing the mean spill-covered water surface in the Baltic Sea test area for morning (A.M., left column) and evening (P.M., right column) passes are shown in Figure 5-10, subdivided in the first year (December 1996 — November 1997, upper row) and the second year (December 1997 — November 1998, lower row). It is obvious that the amount of detected oil pollution is much larger for morning passes, which can be explained by the fact that oil is spilled from ships most often during night time. Since the oil spills may stay on the water surface for some hours (or even more, depending on kind and amount of the slick material, sea state, and wind conditions) they are still visible on SAR images acquired during morning passes.

There is no clear overall trend showing different marine oil pollution in the first and second years of this investigation. That is, in both project years (1996/97 and 1997/98) we found comparable values for the mean oil-covered water surface in each of the three test areas.

Table 5-2 summarises the results presented in this chapter more quantitatively. The upper section (upper two rows) shows the total number of analysed SAR frames as well as the total spill-covered area per test area (Baltic Sea, North Sea, and Gulf of Lion). From these data we can delineate that the mean (detected) spill-coverage per SAR image is lowest in the Baltic Sea (2.56 km²) and highest in the Gulf of Lion (3.94 km²).

Figure 5-10. Maps showing the mean oil-covered water surface in the Baltic Sea, as delineated from SAR imagery acquired during morning and evening passes (left and right column, respectively). Upper row: results for the first project year, lower row: results for the second project year.

In the second section of Table 5-2 the percentage of SAR images showing oil pollution is shown. It can be seen that we have detected the highest percentage of oil pollution in the Gulf of Lion test area (36% of the images, see the right-most column) and that about 20% of all images show radar signatures of large oil spills with sizes ranging from 1 km² to 5 km². In the Gulf of Lion the relative number of very large oil spills (larger than 5 km²) is higher than in the other two areas. This we attribute to the fact that the test site includes the sea area off Barcelona which is an area of high spatial density of oil pollution (e.g., in front of the harbour of Barcelona we often observed a large plume driven by the coastal current towards south).

The third section of Table 5-2 shows the total number of oil spills detected in the SAR images from all test areas. It can be seen that the highest occurrence is found in the Gulf of Lion test area. Even spills with a size larger than 1 km² have been found in the Gulf of Lion more often than in the North Sea and Baltic Sea.

The bottom section of Table 5-2 shows the distribution of the detected oil spills on the different overpasses (that is, descending (morning) and ascending (evening)) as well as on the different seasons (summer and winter). In all three test areas the occurrence of oil pollution detected on SAR images acquired during descending passes is much higher than during ascending passes. This observed difference is most pronounced during summer months and less pronounced during the winter months, because of the earlier sunset (the evening passes are between 21:00 and 22:00 UTC). Moreover, the finding that more pollution is detected during summer months (between April and September) may be due to the fact that the mean wind speed in all three test areas is lower during summer. Thus, any oil pollution might be easier to detect because of the wind speed dependence of the visibility of oil spills on SAR images (see section 4.4.1).

Table 5-2. Results from the statistical analyses of the ERS SAR images.

For the analysis of the fractal dimension of various different signatures, which are visible in SAR imagery and which may help to distinguish between oil pollution and so-called look-alikes, we used a box-counting algorithm to detect the self-similar characteristics for different SAR-image intensity levels. This purely geometrical description may be related to dynamic (oceanic and atmospheric) processes assuming that an energy transfer generated at a certain range of scales will affect the ocean surface at the same scales. The best geometrical characterisation of a multi-fractal set showing different fractal dimensions D for different SAR-image intensities i is given by the maximum fractal dimension; however, relevant information may also be obtained from the complete function D(i). In Figure 5-11 we show five different sections of ERS-2 SAR images corresponding to different oceanic phenomena. Also plotted are the respective functions D(i) without (purple) and after applying a speckle-noise (Kuan) filter (pink): a,b) anthropogenic oil spills, c) atmospheric convective cells, d) biogenic sea slicks, and e) rain cells. The anthropogenic oil spills tend to exhibit a second sharp peak at lower backscatter values and with a smaller (maximum) fractal dimension, whereas the wind-induced patterns and natural slicks show a broad (larger) maximum of D(i). The irregular-shaped rain cells again cause a sharp peak, but at higher backscatter values. The application of a speckle filter more clearly shows the differences in maximum fractal dimension.

There seem to be differences in the maximum fractal (box counting) dimension of oil spills compared to other types of oceanic phenomena, probably due to the short span of the spills. In particular, the maximum fractal (box counting) dimension of the detected oil spills and of the rain cells is about 1.3, whereas it is larger than 1.5 for the atmospheric features and for the slicks. As a result of this difference we believe that by using fractal analysis techniques existing oil spill detection algorithms may be significantly improved.

The results of our comprehensive statistical analyses show the capability of the ERS SAR system for detecting marine oil pollution, but also its limitations. For example, the poor temporal coverage of the satellite (same path every 35 days) makes an effective oil spill surveillance very difficult, if not impossible. Moreover, our results show a good negative correlation between the amount of detected oil pollution and the mean wind speed, i.e. we have found less pollution in areas of higher wind speeds. However, it could not clearly be found out whether this is an artefact of the measurement mechanism (i.e. of the radar backscattering from the water surface) or whether the north-western Mediterranean Sea is more polluted than the Baltic and North Sea.

By introducing new image processing techniques, namely the calculation of fractal dimensions, like the box counting dimension, an improvement of the classification of different signatures in SAR images seems to be possible.

Figure 5-11. Results of the fractal analysis of different SAR images. For each SAR image section on the righthanded side fractal dimensions were calculated as functions of image intensity (NRCS). a), b) oil spills, c) wind variations, d) natural slicks, e) rain cell. The purple curves correspond to the raw images without filtering, the pink curves correspond to the same sections after having applied a speckle (Kuan) filter.

The above results have been presented at the EARSeL 1998 Symposium (Enschede, The Netherlands (Gade et al., 1999)), at the International Symposium on Marine Pollution 1998 (Monaco), at the IGARSS conferences 1998 (Seattle, USA (Gade and Ufermann, 1998)) and 1999 (Hamburg, Germany (Gade and Redondo, 1999a)), and at the Oceans 1999 (Seattle, USA (Gade and Redondo, 1999b)). Furthermore, they are included in a peer-reviewed paper to be published in Science of the Total Environment (Gade and Alpers, 1999).

5.3.2 Algal blooms

Algal blooms, i.e. mass occurrence of algae is a phenomenon common to many waters surrounding Europe and other areas. The frequency and intensity of these blooms is a matter of concern. Many blooms, such as the spring bloom in cold and temperate waters, do not often directly affect human activities, even though intensified blooms leads to higher oxygen demand during the decomposition phase, thereby leading to possible anoxic deep water conditions. The marine spring blooms, often dominated by various dinoflagellate and diatom species, have not been exclusively studied within the frame of the Clean Seas project. The focus has been on the summer algal blooms in the Baltic Sea.

Late summer blooms of nitrogen-fixing, filamentous cyanobacteria (blue-green algae) are a regular phenomenon in the Baltic Sea. Nodularia spumigena (Figure 5-12), which is facultatively hepatotoxic, often dominates the blooms in the open sea. Apart from its toxicity, it also affects the environment by its ability to use dissolved molecular N₂ as an additional nitrogen source, thereby increasing the available nitrogen compounds in the water. The intensity and spatio-temporal variability of the blooms are reported to show considerable variations (Kononen, 1992).

It is generally considered that Nodularia blooms in the Baltic are initiated if sufficient phosphorus is available and the water temperature exceeds 16°C. These conditions, even if generally valid, are hardly sufficient to predict or model the actual bloom areas. Other environmental factors which controls the occurrence and triggering of cyanobacterial blooms are salinity, irradiance and wind speed (Wasmund, 1997).

Other algal blooms being identified within the Clean Seas project includes the Coccolithophore blooms in the North Sea and Skagerakk areas. Several blooms in the Skagerakk have been detected during the project time by imagery from the Advanced Very High Resolution Radiometer (AVHRR). In 1999 data from the Sea-viewing Wide Field-of-view Sensor (SeaWiFS) was also used to monitor blooms in the north-eastern Atlantic.

Apart from the cyanobacteria in the Baltic, and especially the toxic Nodularia spumigena blooms, other toxic blooms affect neighbouring waters. Examples of such blooms are the Gyrodinium aureolum bloom during the fall of 1981-1984 in the Skagerakk and the Chrysochromulina polylepis bloom in Kattegat in 1988. The latter of the two blooms was monitored by satellite remote sensing technique. Recent research has found ways to biologically control toxic algal blooms. The dinoflagellate Dinophysis norvegica, often responsible for DSP in mussel farms, is one species that might be controlled by the parasite Parvilucifera infectans. The parasite also attacks Alexandrium, known to cause severe damage to aquacultures around Europe and elsewhere. With a possible tool to control directly the influence of harmful algal blooms, the need for proper algal bloom detection, monitoring and prediction in the near future will further increase.

Although mass blooms of cyanobacteria in the Baltic have been known since the mid- 19th century, it has been suggested that the extent and intensity of the blooms has increased because of anthropogenic sources of eutrophication. However, due to the high spatial and temporal variability, and the scarcity of data from the open sea, long-term changes in the extent of cyanobacterial blooms are very difficult to prove using data from conventional shipboard monitoring. At some stage of the bloom cyanobacterial filaments become positively buoyant and aggregate in the surface layer in a process of inverted sedimentation (Horstmann, 1986). The filaments often occur in large agglomerates giving the impression of yellow snowflakes (Figure 5-12). At low wind speeds the agglomerates accumulate at the surface to an extent that they become visible even on low-sensitivity satellite imagery. Various satellite imagery of the Baltic Sea is available since the mid-1970s and can be used for mapping near-surface algal blooms. In principle, this would allow detection the inter-annual dynamics of cyanobacterial surface accumulations during the last 20 years. Although frequent cloud cover over the Baltic Sea severely limits the amount of available satellite data, the accumulations of cyanobacteria usually co-occur with periods of sunny and calm weather when good imagery is also available. The AVHRR sensor has been flown on the NOAA series of satellites with small modifications since 1978. Due to its wide swath width (over 2500 km), frequent coverage (several passes per day), and availability, the AVHRR is especially suitable for monitoring purposes. On the other hand, the relatively low spatial resolution (approximately 1 km at nadir) and coarse spectral information (two broad bands in the visible-near infrared domain) may limit the sensor's capability to distinguish accumulations of cyanobacteria from certain types of clouds and other atmospheric disturbances. By processing and analysing a very large number of NOAA AVHRR images, the variations of cyanobacterial surface accumulations during the last 18 years (1982-1999) have been found to be considerable. In Rud and Kahru (1995) a time series of cumulative surface accumulations clearly illustrates the high inter-annual variations. The spatial distribution of surface accumulations during the 1997 bloom season is illustrated in Figure 5-13. In 1997 the area affected by the cyanobacteria exceeded 100 000 km², a number probably to be surpassed by the result from 1999.

Figure 5-13. Number of days with detected surface accumulations of cyanobacteria in the Baltic Sea during July-August 1997. Image compiled by Dr Mati Kahru.

In June 1997 an unusually early bloom was detected in the AVHRR imagery. This bloom was earlier than a ‘regular’ Nodularia bloom, which usually starts early to mid July depending on weather conditions. Field samples showed that the bloom instead was caused by Aphanizomenon flos-aquae, another cyanobacteria, also abundant in the Baltic Sea, but normally not so close to the sea surface. During the long time series of AVHRR imagery, very few blooms have been detected in the Bothnian Sea. Exceptions to this occurred in both 1997 and 1999.

Regular monitoring of algal blooms in the Cleans Seas test sites has been performed during the two first years of the project time. In the Baltic Sea such monitoring was extended into the final year as well. Before the 1999 season all regular algal bloom monitoring within the project was based on AVHRR data, a data source easily available through the receiving station at Stockholm University. During the 1999-bloom season, SeaWiFS data were utilised for the first time on a regular base for the monitoring in the Baltic Sea (Figure 5-14).

Figure 5-14. SeaWiFS image from July 11, 1999 indicating several algal blooms in the Baltic (cyanobacteria), Skagerakk and the North Sea (Coccolithophore). A smalller bloom also occurs west of Esbjerg in Denmark.

As the algal blooms pose environmental concern and even health hazard, the information request on ongoing algal blooms is intense. The current status of the monitoring can be described as pre-operational as some adjustments are needed for the implementation within a regular monitoring scheme outside a university. The benefits from such regular monitoring and statistics are the better insight into the temporal and spatial dynamics of the algal blooms. With these insights accurate guidelines for in situ measurements and the localisation of monitoring stations can be designed.

Currently the procedure of monitoring the algal blooms, outside the scope of Clean Seas, uses data from both the AVHRR instrument and the ocean colour sensor SeaWiFS. Information about the spatial distribution of a bloom is normally published on the Internet within 45 minutes after a satellite pass. The information includes processed satellite images with annotations indicating affected areas and levels of confidence in the interpretation of the data. Accompanying the image is also a text note describing the features present in the image. The Information Office for the Baltic Proper in Stockholm directly uses the satellite derived information in their work to compile all available data about an ongoing bloom. Other agencies and institutions, both national and international, also uses the distributed satellite ‘bloom maps’ in their work.

5.3.3 Mapping of hydrodynamic dispersion areas

We will present here some of the advanced methods used to extract relevant geometrical information from the SAR images that may be used to characterise the local horizontal diffusivity of the ocean surface. Two basic measures are needed as a function of the position (longitude and latitude). The fractal dimension and the integral or correlation length-scale of the SAR intensity signal.

Calculation of Fractal Dimensions

Some of the theory relating fractal analysis to the turbulence is presented in Redondo(1990, 1995,1996 a) b)). This is a rapidly evolving research field both in the mathematical theory and in the experimental applications. Here we present the basic method.

Fractal analysis was used to identify different dynamic processes that might influence the radar backscattering from the ocean surface. We used a box-counting algorithm that is able to detect the self-similar characteristics for different SAR-image intensity levels.

The fractal dimension D (i) as a function of intensity i may be calculated using

where N(i) is the number of boxes of size e needed to cover the SAR contour of intensity i. The algorithm operates dividing the digitised 2D surface into smaller and smaller square boxes and counting the number of them which have values close to the SAR radiation level under study.

Let us assume a convoluted line, which is embedded in a plane (that is why it is usually referred to as D2, or fractal dimension within an Euclidean plane of dimension 2). If it is a single Euclidean line, its (non-fractal) dimension will be one. If it fills the plane its dimension will be two.

The box-counting algorithm divides the embedding Euclidean plane in smaller and smaller boxes (e.g., by dividing the initial length L₀ by n, which is the recurrence level of the iteration). For each box of size L₀/n it is then decided if the convoluted line, which is analysed, is intersecting that box. The number N(i) is the number of boxes which were intersected by the convoluted line (at intensity level i). Finally, we plot N versus L₀/n (i.e., the size of the box e) in a log-log plot, and the slope of that curve, within reasonable experimental limits, gives the fractal dimension. Note that the sign of the fractal dimension is not relevant.

This would need to be done for different contour-levels corresponding to different SAR intensity levels i. For practical purposes it is enough to check at the frontiers of the boxes, whether there is any pixel with the desired intensity level, except in the very fragmented convoluted lines (from a topological point of view).

The underlining characteristic of a fractal set is the self-similarity of scales in the sense that there are smaller and smaller scales which maintain some relation between them. This purely geometrical description may be related to the dynamic processes assuming that an energy input generated at a range of scales will effect the ocean surface at those scales. This relationship corresponds to an exact fractal set, such as the Koch curve, the interface looks exactly the same when looked under different magnifications. In a statistical fractal set, the interface only looks statistically similar when the scale is reduced. For different parts of the interface, the fractal dimension changes slightly, but we can define a mean or average fractal dimension. The best geometrical characterisation of a multi-fractal set that shows different Fractal dimensions for different intensities is given by the maximum fractal dimension, but relevant information may be obtained by the complete function D(i).

Figure 5-15. Eddy structures detected in the Gulf of Lion during August 1997

Integral length-scales

The satellite-borne SAR is able to detect oceanic features with a range of scales as seen in Figure 5-15, which shows several eddy structures in the Gulf of Lion area during August 1997. The spatial cross correlation may give an indication of the length over which such features are correlated.

Let i(x) be the intensity of the SAR backscatter at point x and i(x+r) the intensity at a point separated a distance r from the first one. The normalised average

Represents the cross-correlation of i(x) over the area where the average is taken. Dividing the average by the variance s i forces the value of R(0) to be one.

The integral length-scale is defined as

And indicates the spatial scale where the SAR intensities are correlated.

Eddy Diffusivity

The eddy diffusivities in the ocean exhibit a large variation and show a marked anisotropy, not only horizontal values are much larger than vertical ones but there is a strong dependence on the spatial extent of the tracer dye or pollutant and at larger scales the topology of the basic flow is very important. In the case of oil spills, these are strongly influenced by the buoyancy and horizontal diffusion depends on ambient factors such as wave activity, wind and currents.

Measurements have been made near the coast for a variety of weather conditions and these values have been compared with cruise measurements and with estimates from satellite observations.

There is a strong dependence of horizontal eddy diffusivities with the Wave Reynolds number as well as with the wind stress measured as the friction velocity from wind profiles measured at the coastline. These results have been published recently in Bezerra et al. (1998). Both effects are important and give several decades of variation of eddy diffusivities measured near the coastline (between 0.0001 and 2 m²s^-1). Longshore currents are also important near the coast. Experiments of dye diffusion such as those performed filming the evolution of slicks allow to characterise the ranges of Kx and Ky as a function of the distance to the coast and other environmental factors (Wave height and frequency, wind stress and mean current).

A good estimate of the eddy diffusivity comes from a scaling that includes the thickness of the surf zone as well as the depth and the wave period. Measurements in the Mediterranean are almost two orders of magnitude smaller than in the Pacific coast. On a larger scale, and further away from the coast the relevant eddy diffusivities are much larger, because large eddies, that often scale on the Rossby deformation radius :

disperse further oil or tracer slicks in the sea surface. Here N is the local Brunt-Vaisalla frequency, f is the Coriolis parameter and h is the relevant depth.

Figure 5-16. Vorticity mapping of the NW Mediterranean Sea from SAR.

We will present as examples the state of the ocean surface revealed by SAR on the month of August in both 1997 and 1998. The different descending passes of the SAR are separated three days, but we consider that the basic dynamic features do not change much during that time. The low wind areas, seen in black preclude any analysis, the same is true for very strong winds, so our analysis is valid under a certain window of atmospheric conditions.

With these limitations in mind, it is nevertheless possible to analyse in a statistical integrated way all available frames. This is done in Figure 5-16 for the NW Mediterranean frames 8,9,10,19 and 20 (see Figure 3-1). There we show in Figure 5-16a the probability of detecting the vortical slicks as a feature akin to a vortex. In Figure 5-16b the submarine canyon presence (green lines) and the situation of the detected vortices and their distribution of angles, shapes and size are presented over the whole measurement period.

Figure 5-17. Percentage of detected vortices as a function of their ellipticity.

There are several apparent relationships between the number of detected eddies and their size (Figure 5-17a)), their ellipticity measured as the ratio between the ellipse semi-axes ratio a/b shown in Figure 5-17b) and the fact that the observed eddies are larger near cape Begur in the NE. We show in Figure 5-17c) the area of the observed vortical slicks and the distance (in km) from Cape Begur.

Relevant geometrical information of different areas is also given by the maximum fractal dimension, which is related to the energy spectrum of the flow. Using all the available information it is possible to investigate the spatial variability of the horizontal eddy diffusivity K(x,y). This information would be very important when trying to model numerically the behaviour in time of the oil spills.

A method for evaluating the relevant meso-scale eddy diffusivity is straight forward using dimensional analysis (as a velocity times a length scale) from the measured distribution of integral length scales and the eddy turnover times associated to inertial oscillations associated to the local Coriolis parameter

where W is the rotation of the earth and q is the latitude.

Using the integral length-scale distribution l(x,y) as a function of longitude and latitude we may calculate the eddy diffusivity as

where K(x,y) is the horizontal diffusivity at latitude y and longitude x

Another method, still under investigation, that takes into account the fractal dimension of the SAR images and the relation between fractal dimensions and velocity spectra described by Redondo (1990,1994,1996) is used to provide more detailed estimates of the seasonal variations of eddy diffusivities. The theory provides a relationship between the geometrical self-similarity of the area detected by the fractal dimension of the SAR image intensity which is detected by the fractal dimension D relevant in the spectral range between the integral length scale l(x,y) and the maximum of either the Kolmogorov length-scale or (as is the case with the present resolution) the pixel resolution of the images.

Redondo (1990) found that on a 2D surface, a relationship between the power spectrum of velocity, which in general is

With e the turbulent energy dissipation, and the fractal dimension may be found as b = 5 - 2 D. The fractal dimension is calculated as the maximum fractal dimension of all possible intensity contours of the SAR reflected intensity, which exhibit a complex geometry. The main hypothesis is that the topological complexity of the SAR image is induced by the dynamics of the velocity field at the sea surface. It is not thought that a precise distinction of the features produced by the wind and those produced by the wave field or currents is needed, because as shown by Bezerra et al. (1998) both effects produce turbulent diffusion at the sea surface.

Here an alternative method is based on an integration of the derived fractal spectra

used to estimate the relevant velocity

and then an alternative way to estimate the spatial distribution of horizontal eddy diffusivity taking into account the flow structure is:

The two methods of deriving eddy diffusivity maps from SAR image information give realistic estimates and these values may be used to parameterise sea surface turbulence. The method involving the fractal dimension measurement is much more elaborated but seems to have a better theoretical justification.

Figure 5-18. Map of averaged horizontal eddy diffusivity of the Gulf of Lion area obtained by SAR analysis during 1997 and 1998. X is longitude and Y latitude. The Eddy diffusivity average values are in (m2/s) calculated according to Richardson's 4/3 law fo a 10km spill.

In Figure 5-18 we show the distribution of horizontal average eddy diffusivity in m²s^-1 relevant to a 10 km spill. In order to relate diffusivities at different length-scales Richardson´s law should be used as there is a dependence of turbulent diffusivities on length-scale as

These techniques are helpful in providing more realistic estimates of spatial and temporal variations of the horizontal dispersion in the Gulf of Lion, to be used in numerical models and in prevention of accidents such as oil spills.

Conclusions

The satellite-borne SAR seems to be a good system for oil spills detection, both in coastal areas and along ship routes, these may be easily identified as described in Gade and Redondo (1999). It is also a convenient tool to investigate the eddy structure of a certain area and the effect of bathymetry and local currents are important in describing the ocean surface behaviour. In the example presented near Barcelona, the maximum eddy size agrees remarkably well with the limit imposed by the local Rossby deformation radius using the usual thermocline induced stratification, Redondo and Platonov (1999). The Rossby deformation radius, defined as

where N is the Brunt-Vaisalla frequency and h the thermocline depth, Rd is about 20 km for the area near Barcelona.

There are promising results in the possibility of using fractal techniques to distinguish between oil spills and natural slicks. There seems to be a reduction in the maximum fractal dimension of the oil spills compared to other types of natural slicks, probably due to the short time span of the spills that do not allow for the effect of all the self-similar eddy dynamics to affect the transported scalar (oil). There might also be some influence due to the spatial distribution of SAR reflectivity as a function of surface tension or of buoyancy effects in the ocean surface.

5.3.4 Ocean colour and ecosystem dynamics

The analysis of ocean colour data allows to evaluate the main characteristics of the marine environment at the scale of entire basins, and to put them into a climatological perspective. A regional assessment of the pigment field of the Mediterranean Sea, emerging from the analysis of the CZCS historical archive, points to the existence of geographical provinces, where a specific relationship seems to exist between remotely sensed indices and geographic/climatic features of the region (Barale and Zin, 1998). The provinces can be defined in terms of coastal processes and open sea conditions, as well as of transition conditions between these two extremes.

Coastal features such as river plumes, filaments and permanent gyres appear to be recurrent and to maintain their characteristics over the medium to long term (Barale and Larkin, 1998). A comparison of (monthly) mean pigment concentration for April and July in the Ligurian/Provençal/Balearic sub-basin, as derived from CZCS 1979-1985 data and from SeaWiFS 1998 data, is shown in Figure 5-19 and Figure 5-20. Both historical and current data present typical spring bloom conditions in April, when river plumes are larger, and high pigment concentration define sharp fronts separating the near-coastal area from open sea waters. Similarly, in July, concentrations are low in both cases, and both the Rhone and Ebro river plumes continue to influence the respective coastal zones with remarkably similar shape and extent. This comparison illustrates the capability of optical remote sensing to identify critical areas (i.e. river plumes) where environmental impact should be monitored on a synoptic basis, and to do so in a repetitive, consistent manner.

In the time domain, the seasonal trends, derived from mean values of the CZCS multi-annual archive, for the whole basin and on a monthly basis, point to a behaviour of the Mediterranean Sea similar to that of a subtropical basin (Barale et al., 1998). In such a scenario, higher pigment concentrations would be associated in winter to low temperatures and a high wind speed regime, due to a combination of runoff and vertical mixing, responsible for the cooling and nutrient enrichment of surface waters. Conversely, lower pigment concentrations would be associated in summer to high temperatures and a low wind speed regime, as the water column is strongly stratified and no cold waters or nutrient sources, from coastal zones or deeper layers, are readily available.

Some near-coastal areas, however, display a seasonality closer to that of a subpolar basin, with lower pigments in winter, because of reduced light and, more important, because of the very deep vertical mixing due to the prevailing wind field, which can prevent algae to stabilise in the upper well-lit layers. This seems to be particularly true for the northern part of the western basin, and in particular for the Gulf of Lion (Figure 5-7), where the lack of high pigments, and the very low temperatures, in winter might be linked to the extreme conditions generated by the overturning of the entire water column caused by the Mistral wind. The ensuing spring bloom would then be triggered by the relaxation of these conditions, when the wind regime weakens, the water column becomes sufficiently stable, and stratification occurs. Again, the comparison of historical and current data on the pigment field spatial and temporal patterns supports this interpretation, and suggests that geographical and meteorological factors of some coastal areas are critical in affecting both bio-geo-chemistry and dynamics of the entire basin.

The integration of geographical and environmental data, derived from ocean colour data for the Mediterranean Sea, and used here for the assessment of provinces and trends in the basin, is a powerful tool to understand, and follow, ecosystem dynamics in a medium characterised but very large space scales and very short time scales, difficult to monitor with conventional in situ techniques. The statistical results obtained should be considered with caution, due to the present sensor limitations in retrieving quantitative assessments of surface pigments, and to the still rather poor spatial and temporal resolution of the observations. However, the analysis of recurrent patterns derived from the long-term composite images, and the comparison with current data, showing the estimated mean conditions of the Mediterranean, can be interpreted in terms of the main environmental traits of the basin and its main geographical components.