International Journal of Remote Sensing Vol. 30, No. 6, 20 March 2009, 1587–1602 On the co-polarized phase difference for oil spill observation M. MIGLIACCIO*, F. NUNZIATA and A. GAMBARDELLA Università degli Studi di Napoli Parthenope, Dipartimento per le Tecnologie, Centro Direzionale, Isola C4, 80143 Napoli, Italy (Received 11 July 2007; in final form 5 April 2008 ) In this study dual-polarized synthetic aperture radar (SAR) measurements were used to enhance oil spill observation. The co-polarized phase difference (CPD) was modelled and used to characterize the scattering return from oil spills and biogenic slicks. The model predicts, under low to moderate wind conditions, a larger CPD standard deviation (s) for oil with respect to the sea, while for biogenic slicks a s value similar to that for the sea is obtained. Experiments accomplished with multilook complex (MLC) C- and L-band SAR data show that the model predictions are confirmed and that the C-band is, as expected, to be preferred to the L-band. 1. Introduction Sea oil pollution is a matter of great concern, with about 180 millions of gallons of pollution estimated to be spilled every year (Delilah 2002, ITOPF 2007). Oil discharged into the sea enters into the fish life cycle, alters it and subsequently affects human health (Delilah 2002). An oil spill monitoring system, with all-weather day and night capability, is fundamental to support law enforcement and to minimize the ecosystem impact of such polluting events. For this purpose, airborne and spaceborne remote sensing is a key tool and synthetic aperture radar (SAR) is the most useful sensor for the detection of sea oil pollution, under low to moderate wind conditions (Brekke and Solberg 2005). SAR is an active, coherent, bandlimited microwave high-resolution sensor that provides valuable measurements at both day- and night-time and almost independently of atmospheric conditions. Physically, oil spill detection is possible because an oil slick dampens the short gravity and capillary waves that are responsible for the backscattered electromagnetic field at the SAR (Brekke and Solberg 2005). As a consequence, an oil spill generates a low backscatter area, that is a dark area in SAR images. However, oil spill detection in SAR images is not an easy task. Other physical phenomena can also generate dark areas and SAR images are affected by multiplicative noise known as speckle (Brekke and Solberg 2005). Dark areas not related to oil spills are known as look-alikes. Phenomena that give rise to look-alikes include biogenic films (e.g. slicks produced by animals and plankton), low-wind areas, areas of wind-shadow near coasts, rain cells, currents, zones of upwelling, internal waves and oceanic or atmospheric fronts (Brekke and Solberg 2005). Oil spill detection can be divided into three phases: dark area detection, features extraction, and oil spill/look-alike classification. Dark area detection algorithms are generally based on filtering *Corresponding author. Email: maurizio.migliaccio@uniparthenope.it International Journal of Remote Sensing ISSN 0143-1161 print/ISSN 1366-5901 online # 2009 Taylor & Francis http://www.tandf.co.uk/journals DOI: 10.1080/01431160802520741 1588 M. Migliaccio et al. techniques accomplished on multilook single-polarization SAR data. While dark area detection algorithms yield the area location and the segmentation of suspected polluted areas, the extraction of features (e.g. geometric, radiometric and texture related) is necessary to perform the classification. On the basis of the estimated features and some a priori knowledge it is possible to assign a probability that a dark area is an oil spill (Brekke and Solberg 2005). To enhance the ability to distinguish oil spills and look-alikes, the usefulness of additional external data is nowadays recognized; for instance, optical data to identify biogenic films (Fingas and Brown 1997). Although there is a general consensus that radar polarimetry is able to provide additional information for environmental applications, the real benefit of SAR polarimetric information, which can be extracted once a suitable electromagnetic model is available, has been demonstrated only for land applications and initial investigations on sea oil spill observation were generally unsatisfactory (Gade et al. 1998). Only in a recent paper (Migliaccio et al. 2007) was it shown that fully polarimetric features can assist in distinguishing between oil spill and biogenic lookalikes. On similar physical guidelines, in this paper the usefulness and the significance of a dual-polarized SAR sensor is explored. Although a fully polarimetric SAR is to be desired, there may be hardware and budget considerations that require the design and implementation of a simpler polarimetric SAR configuration (Raney 2007). For instance, the SAR sensors onboard RADARSAT-2 and ALOS PALSAR operate in fully dual- and single-polarization modes while the SARs onboard ENVISAT and COSMO-SkyMed operate only in single- or dual-polarization modes. The use of dual-polarized SAR sensors for oil spill observation is operationally important. Many studies on the use of partially polarimetric SAR data for geophysical remote sensing have been addressed (e.g. Lee et al. 2001, Raney 2007). Among them, for the purposes of this paper, some theoretical and land application studies (Ulaby et al. 1987, Boerner et al. 1987, Ulaby et al. 1992, Kuga and Zhao 1996, Lee et al. 2001) using the co-polarized phase difference (CPD), that is the phase difference between HH and VV channels, are emphasized. Only a few preliminary studies have been published regarding sea application of the CPD (Schuler et al. 1993), and none on oil spill observations. In Schuler et al. (1993) a relationship between the CPD and sea surface roughness and local incidence angle was reported using JPL AIRSAR C-, L- and Pband polarimetric data. In Lee et al. (1994) a closed form for the multilook CPD probability density function (pdf) was derived and compared successfully with JPL AIRSAR polarimetric ocean surface SAR data. In this paper a new paradigm is stated: dual-polarimetric SAR data, once properly interpreted by means of a tailored electromagnetic model, are more useful than the single polarization data for oil spill observation. A model that relates the CPD to sea surface scattering mechanisms with and without slicks is first developed. It is shown that, under low to moderate wind conditions, the broadening of the CPD pdf is sensitive to the presence of a low backscattering area. The theoretical considerations predict a different sensitivity of oil slicks and biogenic look-alikes because of their different damping effects. Thus, a novel and very effective filtering technique, based on CPD, has been implemented and tested on multilook complex (MLC) C- and L-band SAR data. The results confirm that the C-band is better than the L-band (Gade et al. 1998) and that the CPD is useful both to assist oil spill observation and to distinguish between oil spills and biogenic look-alikes. This CPD for oil spill observation 1589 filtering technique can be used to improve classical oil spill detection procedures. The paper is organized as follows: in section 2 the theoretical background is reported; in section 3 the experiments are presented and discussed; and conclusions are drawn in section 4. 2. Theory A full-polarimetric SAR measures the 262 scattering matrix S, which relates the electromagnetic field scattered by the observed scene Es to the incident one Ei (Guissard 1994): Es ~ e{jkr i SE r ð1Þ where j is the imaginary unit, k is electromagnetic wavenumber and r is the distance. The scattering matrix, considering the horizontal and vertical linearly polarized electric fields, can be expressed by: : ! : S hh S hv : S~ : ð2Þ S vh S vv where each complex element of the scattering matrix, called the scattering amplitude, can be written as: : Spq ~Spq ejQpq p, q~h, v ð3Þ Thus, invoking reciprocity, equation (2) can be written as: ! " Shh ejQc Shv ejQx S~ejQvv Shv ejQx Svv ð4Þ where Qx ~Qhv {Qvv ~Qvh {Qvv ð5Þ Qc ~Qhh {Qvv ð6Þ and are the cross-polarized phase difference (XPD) and the CPD, respectively. It is well known that in natural surface scattering the Qpq are uniformly distributed and contain no useful information (Ulaby et al. 1992). This is generally untrue for the CPD and XPD. In fact, as the co- and cross-polarized scattering amplitudes are almost uncorrelated for most natural surfaces, the XPD follows approximately a uniform distribution (Ulaby et al. 1992) (see also experimental pdf shown in figure 1) whereas, because of the correlation between the copolarized scattering amplitudes, the CPD is generally non-uniformly distributed (Ulaby et al. 1992) and has a mathematical expression that is given by (Joughin et al. 1994, Lee et al. 1994): # $l ! "ðlz1=2Þ=2 1{r2 C ð2l Þ 1 {l{1=2 pl ðQc Þ~ lz1=2 pffiffiffi Pl{3=2 ð{bÞ ð7Þ 2 pC ðl Þ 1{b2 with 1590 M. Migliaccio et al. Figure 1. Plots of the theoretical XPD pdf (dashed line) and the experimental one relevant to a transect obtained from the C-Band. SIR-C/X SAR data acquired on 4 October 1994 at 04:37 UTC (solid line). b~r cosðQc {QÞ ð8Þ where l is the number of looks, P(?) is the associate Legendre function of the first – are the modulus and the phase of kind and C(?) is the gamma function. r and Q the complex correlation coefficient between HH and VV channels, respectively, : and r is given by (Joughin et al. 1994, Lee et al. 1994): : :# ·Shh Svv ¶ : ð9Þ r~ qffiffiffi&ffiffiffiffiffiffiffiffiffi&ffiffiffiffiffiffiffiffi&ffiffiffiffiffiffiffiffiffi&ffiffiffiffiffi ~re j Q 2 2 ·& hh & ¶·& vv & ¶ : The width of (7) depends on l and on r. In particular, the pdf becomes narrower – (Joughin et al. 1994, when either l or r increase. The peak of the pdf is at Qc5Q Lee et al. 1994). Moreover, when r tends to 0 (total decorrelation between HH and VV channels), the pdf becomes uniformly distributed between [2p, p), while for r approaching 1 (HH and VV totally correlated) the pdf tends to a Dirac delta – function. For 0,r,1 the pdf resembles a Gaussian bell with a mean value m5Q and a standard deviation s that is inversely related to r (see figure 2). It is now important to read the CPD pdf in marine physical terms. Its behaviour can be explained, under low to moderate wind conditions and for incidence angles far from the grazing angle, by considering some key reference scenarios. In the case of an oil-free sea surface, the small-scale scattering is well modelled by the Bragg scattering mechanism, which is characterized by low polarimetric entropy (Schuler and Lee 2006) and the HH–VV phase difference around 0u (Guissard 1994). Two cases must be considered. CPD for oil spill observation 1591 Figure 2. Plots of the theoretical CPD pdf for |r|50.1, 0.7, 0.9 and its phase –Q50u, with l54. When the long-wave structure is weak, the backscatter follows the Small Perturbation Model (SPM). Because the first-order SPM does not show any depolarization, the cross-polarized scattering amplitudes vanish and the scattering matrix S is diagonal. In practice, because of the high HH–VV correlation (Van Zyl 1989), we expect a narrow CPD pdf whose width is mainly related to the system noise (Freeman 1993). When the long-wave structure is present, the backscatter calls for a two-scale model. Depolarization effects, as well as an increasing of the amount of polarimetric entropy, are expected (Schuler et al. 1993, Schuler and Lee 2006), and the full S matrix needs to be considered. In practice, the cross-polarized terms are neglected because they are smaller than the co-polarized ones and very close to the noise floor (Freeman et al. 1994). In real measurements, the two former oil-free sea scattering cases are almost indistinguishable in terms of CPD pdf. The main effect of an oil slick over the sea surface is to dampen the small-scale structure. A low backscattered signal and a high polarimetric entropy, which indicate that a non-Bragg scattering mechanism is in place (Schuler and Lee 2006), are experienced (Migliaccio et al. 2007). As the polarimetric entropy measures the randomness of the complex polarimetric scattering processes (Cloude and Pottier 1996), a low correlation between the HH and VV backscattered signals, and thus a broadening of the CPD pdf, is expected for the oil-covered area. In other words, oil slicks, which are generally characterized by strong damping properties (Gade et al. 1998), are expected to be distinguishable from the surrounding sea because of their different scattering mechanisms. The case of a biogenic slick is very different. The presence of a biogenic slick over the sea surface, because of its weak damping property (Gade et al. 1998), still calls for a Bragg scattering mechanism, that is a Bragg scattering with a lower backscattered signal. Therefore, similar CPD pdfs for the biogenic-free and biogenic-covered sea surface are to be expected. 1592 3. M. Migliaccio et al. Experiments In this section we describe and discuss the experiments used to demonstrate the capability of the CPD both to observe an oil spill and to distinguish between oil spills and biogenic look-alikes, under low to moderate wind conditions. A total of 25 C- and L-band MLC SAR images were processed. In detail, seven acquisitions include scenes in which biogenic look-alikes were present, four acquisitions were relevant to oil spills and in one an oil spill and six different biogenic look-alikes were present. In total, 40 low backscattering areas were analysed. The data set was acquired by the sensor SIR-C/X-SAR during the missions STS-59 (April 1994) and STS-68 (September and October 1994). Further details on the data set can be found in table 1 (Gade et al. 1998). The radar was designed and built to make eight different measurements at the same time: L- and C-band backscatter at four different polarization combinations: HH, HV, VH and VV (multifrequency and fully polarimetric SAR). The noise floor at the L-band and the C-band was 236 and 228 dB, respectively. The incidence angle varied between 20u and 55u and the SAR swath width on the ground varied between 15 km and 90 km. SAR data were processed by means of a simple and effective filtering technique that, applied over the CPD image, estimated m and s through an N6N moving Table 1. C- and L-band SIR-C/X SAR data set. Processing number Type and band 11587 11588 41466 41467 41369 41370 11351 11352 11438 11439 11585 11586 41464 41465 12815 12816 17040 17041 44326 44327 49938 49939 40385 40386 MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC MLC L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band L-band C-band Area of acquisition Pacific Ocean (Japan) Pacific Ocean (Japan) Pacific Ocean (Japan) Pacific Ocean (Japan) Pacific Ocean (Japan) Pacific Ocean (Japan) Pacific Ocean (Japan) Azores North Sea North Sea English Channel North Sea Date, time UTC Surfactant 15 April, 02:14 4 October, 04:37 1 October, 05:33 12 April, 03:12 16 April, 01:53 8 April, 21:38 2 October, 05:15 12 April, 07:21 11 April, 10:49 1 October, 08:14 8 October, 05:57 11 October, 08:12 Wind speed (m s21) OLA 8.8 OLA 9.3 OLA 5.7 OLA 9.0 OLA 6.8 OLA 4.2 OLA 5.2 Oil Low to moderate* Oil Low to moderate* Oil Low to moderate* Oil Oil and look-alikes 4.0 12.0 *In this case no detailed information on the wind speed is available. Only the reference wind condition is given (Gade et al. 1998). 1593 CPD for oil spill observation window. In this study a window size equal to 7 was used because it represents a good compromise between speckle smoothing and texture preservation in polarimetric SAR images (Lee et al. 1997). Seven of the experiments are shown and discussed in detail. Other results are summarized in table 2. The first experiments concern the processing of SAR data in which only oil spills are present. Then, the same analysis is accomplished over SAR data in which Oley alcohol (OLA) slicks are present. OLA forms a monomolecular surface film that simulates well a biogenic surface slick (Gade et al. 1998). The last experiment is with regard to a SAR acquisition, under high wind speed, in which both biogenic look-alikes and an oil spill are present. The first data set is relevant to the acquisition of 1 October 1994, 8:14 UTC [processing number (p.n.) 44326]. Figure 3(a) shows an excerpt of the L-band VV power SAR image in which an oil spill is clearly visible. The estimated m and s images are shown in grey tones in figures 3(b) and 3(c), respectively. They do not reveal any features related to the oil spill. This result can be quantitatively confirmed by analysing the measured CPD pdfs, relevant to both the oil slick and the surrounding sea surface, see figure 3(d). In fact, the CPD approach is not able to discriminate between the oil spill and the surrounding sea surface. Similar results were obtained for all the L-band SAR images (not shown because of limited space). The results are consistent with those reported by Gade et al. (1998) and show that the CPD approach is not useful for oil spill observation when applied to L-band data. The second data set is relevant to the same acquisition but at the C-band, p.n. 44327 (figure 4(a)). In this case the CPD output images clearly show features related to the oil spill (figures 4(b) and 4(c)). Analysis of the m and s images shows that oil spill is clearly distinguishable, and s provides the key information. This result confirms the theoretical model proposed in section 2, which predicts a different scattering mechanism in the presence of an oil spill and this is revealed by the Table 2. C- and L-band SIR-C/X SAR data set. Processing number 11588 41467 41370 11352 11439 11586 41465 12816 17041 44327 49939 40386 Surfactant OLA OLA OLA OLA OLA OLA OLA Spill Spill Spill Spill A B C D E F G Mean s slick (u) Mean s sea (u) c VV image cs image 4.5 4.0 5.1 12.3 6.3 5.7 10.4 19.3 60.0 60.0 34.3 17.7 19.0 17.9 21.1 19.6 16.2 16.0 3.5 3.6 4.0 10.3 4.7 4.7 8.7 11.2 11.2 17.0 18.0 13.8 13.8 13.8 13.8 13.8 13.8 13.8 0.348 0.390 0.326 0.330 0.436 0.215 0.131 0.336 0.477 0.378 0.299 0.282 0.358 0.292 0.376 0.367 0.386 0.275 0.0832 0.0572 0.0691 0.172 0.166 0.109 0.0784 0.583 0.833 0.620 0.217 0.100 0.105 0.0870 0.127 0.186 0.115 0.0970 1594 M. Migliaccio et al. Figure 3. L-band SAR data relevant to the acquisition of 1 October 1994 at 08:14 UTC (p.n. 44326). (a) An excerpt of the L-band VV power SAR image in which an oil spill is clearly visible. (b, c) The estimated m and s images, respectively, shown in grey tones. (d) The measured CPD pdfs relevant to both the oil-covered and the surrounding sea surface. analysis of the measured oil spill CPD pdf, which is much broader than the surrounding sea surface (figure 4(d)). The mean oil spill s value is 60.0u while that for the surrounding sea surface is only 17.0u. The behaviour predicted by the theoretical model is confirmed by the analysis of the polarimetric entropy, estimated according to the Cloude and Pottier (1996) eigenvector decomposition theorem. The entropy values for the oil-covered and oilfree sea surface are 0.89 and 0.61, respectively. Note, however, that in this case the estimation procedure needs fully polarimetric data. It should also be noted that, because of the very different s values exhibited by the oil-covered sea surface and its surroundings, the CPD for oil spills acts as an emphasis filter. To quantitatively estimate the emphasizing capability of the CPD, the contrast parameter (c), defined according to Blacknell (1994) as the standard deviation to mean ratio, was evaluated. This is a texture-related parameter that is largely used for choosing the adaptive threshold in dark patch detection procedures CPD for oil spill observation 1595 Figure 4. C-band SAR data relevant to the acquisition of 1 October 1994 at 08:14 UTC (p.n. 44327). (a) An excerpt of the C-band VV power SAR image in which an oil spill is clearly visible. (b, c) The estimated m and s images, respectively, shown in grey tones. (d) The measured CPD pdfs relevant to both the oil-covered and the surrounding sea surface. (Solberg et al. 2007) and, combined with other geometrical and radiometric features, for discriminating between oil spills and look-alikes (Solberg et al. 1999). The c parameter was evaluated for the VV power image and the corresponding s image. In this latter case, c was equal to 0.378 and 0.620 for the VV and s images, respectively. Thus, the contrast was approximately doubled after the CPD-based filtering. The third data set is relevant to the acquisition of 11 April 1994, 10:49 UTC (p.n. 17041), see figure 5(a). The processing results are in agreement with what was experienced previously (figure 5(b)). The fourth data set is relevant to the acquisition of 8 October 1994, 5:57 UTC (p.n. 49939), see figure 6(a). A typical oil spill pattern due to a ship (see elongated shape related to the spilling mechanism) is present. No information about the type of oil is reported in the literature (Fortuny-Guasch 2003). As in the previous cases, the output image (figure 6(b)) clearly shows features related to the oil spill. The 1596 M. Migliaccio et al. Figure 5. C-band SAR data relevant to the acquisition of 11 April 1994 at 10:49 UTC (p.n. 17041). (a) An excerpt of the C-band VV power SAR image in which two oil spills are present. (b) The estimated s image shown in grey tones. measured oil CPD pdf is broader than the sea one (figure 6(c)) and in fact the oil spill and sea mean s values are very different (34.3u, 18.0u). Concerning c, it is equal to 0.299 and 0.217 for the VV and the s images, respectively. Such an unexpected result is most probably due to the ageing and weathering processes that generate greater image heterogeneity. Before proceeding further, it is important to note that it might be thought that the ability to detect an oil slick by measuring the CPD s values is related to the low SIRC signal-to-noise ratio (SNR). However, this is not the case as shown, for instance, in the worst case (p.n. 44327), where the oil-covered CPD pdf was plotted only considering the pixel values above the noise floor (figure 7). The s values are now 17.0u and 44.0u (instead of 17.0u and 60.0u) for the oil-free and oil-covered areas, respectively. This behaviour confirms the physical consistency of the proposed approach. The following experiments were performed over SAR images in which biogenic look-alikes were present. The fifth data set is relevant to the acquisition of 15 April 1994, 2:14 UTC (p.n. 11588), in which an OLA is clearly visible (see figure 8(a)). The s image (figure 8(b)) does not show any features related to the OLA, that is it is not possible to discriminate the OLA. This result confirms the proposed theoretical model that predicts, for a weaker damping, a scattering mechanism similar to the sea surface, as witnessed by the measured CPD pdfs that are almost overlapping (figure 8(c)). In detail, the mean s value relevant to the OLA is 4.5u while the surrounding sea surface one is 3.5u. CPD for oil spill observation 1597 Figure 6. C-band SAR data relevant to the acquisition of 8 October 1994 at 05:57 UTC (p.n. 49939). (a) An excerpt of the C-band VV power SAR image in which an oil spill is clearly visible. (b) The estimated image shown in grey tones. (c) The measured CPD pdfs relevant to both the oil-covered and the surrounding sea surface. The behaviour predicted by the theoretical model is confirmed by the analysis of the polarimetric entropy estimated for both the slick-covered (0.54) and the surrounding sea surface (0.53). Moreover, the c values are 0.348 and 0.0832 for the VV power image and the s image, respectively. This shows that the CPD, for OLA, acts as a de-emphasis filter. The sixth data set is relevant to the acquisition of 1 October 1994, 5:33 UTC (p.n. 41370), in which an OLA is present (figure 9(a)). This result is in agreement with what was experienced previously (see figures 9(b) and 9(c)). The only additional comment relevant to the acquisitions made over the area of Japan (see table 1) is that in these cases a peak shift is exhibited on both the sea and the OLA CPD pdfs. Its physical meaning is not related to the presence of OLA but more insights are presently unavailable. It should be noted, however, that this does not affect the CPD filtering capabilities. The last data set concerns the experiment conducted by the University of Hamburg described in Gade et al. (1998). During this experiment artificial biogenic 1598 M. Migliaccio et al. Figure 7. Measured CPD pdfs relevant to both the oil-covered and the surrounding sea surface only considering the pixel values above the SIR-C noise floor. films and a mineral oil spill consisting of heavy fuel (IFO 180) were deployed. The corresponding SAR data were acquired on 11 October 1994 (p.n. 40386) over the German Bight of the North Sea, under high wind conditions. The area of interest in the experiment is shown in figure 10(a), where the seven small slick-covered areas are present. From top to bottom the surface films are due to IFO 180, OLA, oleic acid methyl ester (OLME), triolein (TOLG), TOLG spread with the help of n-hexane, OLME spread with the help of n-hexane and OLME spread with the help of ethanol, respectively (Gade et al. 1998). To identify the seven slicks, they are labelled as A, B, C, D, E, F and G, from top to bottom. Analysis of the s image (figure 10(b)) and the measured CPD pdfs relevant to slicks A and B (figures 10(c) and 10(d)) Figure 8. C-band SAR data relevant to the acquisition of 15 April 1994 at 02:14 UTC (p.n. 11588). (a) An excerpt of the C-band VV power SAR image in which an OLA slick is clearly visible. (b) The estimated s image shown in grey tones. (c) The measured CPD pdfs relevant to both the slick-covered and the surrounding sea surface. CPD for oil spill observation 1599 Figure 9. C-band SAR data relevant to the acquisition of 1 October 1994 at 05:33 UTC (p.n. 41370). (a) An excerpt of the C-band VV power SAR image in which an OLA slick is clearly visible. (b) The estimated s image shown in grey tones. (c) The measured CPD pdfs relevant to both the slick-covered and the surrounding sea surface. shows that, in this case, the filtering is not very effective. This result, confirmed by the c values analysis (table 2) and consistent to what was experienced in Gade et al. (1998), can be physically explained by considering the small size of all the slicks and the high wind conditions. The results relevant to all of the C-band data sets are summarized in table 2. Under low to moderate wind conditions the CPD approach is effective and the key physical results can be summarized as follows: N N N N N 4. The CPD s value is sensitive to the backscattering mechanism. Surface slicks characterized by different damping effects show different s values. Oil spills are characterized by s values larger than the sea surface; thus, the CPD acts as an emphasis filter. Biogenic slicks are characterized by s values similar to the sea values; thus, the CPD acts as a de-emphasis filter. The C-band is to be preferred to the L-band. Conclusions In this study a novel model, based on the CPD, was developed for oil spill observation under low to moderate wind speed conditions. We have shown that the approach is able to emphasize the presence of oil spills and de-emphasize the presence of biogenic look-alikes. The dual-polarimetric SAR data, once properly interpreted by means of a tailored electromagnetic model, are more useful than the single polarization data for oil spill observation. Acknowledgements We thank Dr Anthony Freeman, NASA Jet Propulsion Laboratory (JPL), for useful discussions at POLinSAR 2007, ESA-ESRIN, Frascati, Roma, Dr Joaquim Fortuny-Guasch at the DG Joint Research Centre, the JPL, and the United States Geological Services (USGS) for providing the SIR-C/X-SAR data used in this study. We also thank the anonymous reviewers for useful comments that enhanced the presentation of the results. 1600 M. Migliaccio et al. Figure 10. C-band SAR data relevant to the acquisition of 11 October 1994 at 08:12 UTC (p.n. 40386). (a) An excerpt of the C-band VV power SAR image in which seven small slicks are present. To identify the seven slicks, they are labelled as A, B, C, D, E, F and G, from top to bottom and correspond to IFO 180, OLA, oleic acid methyl ester (OLME), triolein (TOLG), TOLG spread with the help of n-hexane, OLME spread with the help of n-hexane and OLME spread with the help of ethanol, respectively. The estimated s image is shown in grey tones in (b) and (c). (d) The measured CPD pdfs relevant to the slicks A and B and the surrounding sea surface. References BLACKNELL, D., 1994, Comparison of parameter estimators for K-distribution. IEE Proceedings. Radar, Sonar and Navigation, 141, pp. 45–52. BOERNER, W.M., FOO, B.Y. and EOM, H.J., 1987, Interpretation of the polarimetric copolarization phase term in radar images obtained with the JPL airborne L-band SAR system. IEEE Transactions on Geoscience and Remote Sensing, 25, pp. 77–82. CPD for oil spill observation 1601 BREKKE, C. and SOLBERG, A.H.S., 2005, Oil spill detection by satellite remote sensing. Remote Sensing of Environment, 95, pp. 1–13. CLOUDE, R.S. and POTTIER, E., 1996, A review of target decomposition theorems in radar polarimetry. IEEE Transactions on Geoscience and Remote Sensing, 34, pp. 498–518. DELILAH, A., 2002, Marine Oil Pollution: Technologies and Methodologies for Detection and Early Warning. European Commission JRC report, EUR 20231 EN (Ispra, Italy: European Commission). FINGAS, M.F. and BROWN, C.E., 1997, Review of oil spill remote sensing. Spill Science and Technology Bulletin, 4, pp. 199–208. FORTUNY-GUASCH, J., 2003, Improved oil slick detection and classification with polarimetric SAR. In Proceedings of POLinSAR 2003, ESA-ESRIN, Frascati, Italy, 14–16 January 2003. Available online at: http://earth.esa.int/workshops/polinsar2003. FREEMAN, A., 1993, The effects of noise on polarimetric SAR data. In Proceedings of IGARSS, 18–21 August 1993, Tokyo, Japan, Vol. 2, pp. 799–802. FREEMAN, A., VILLASENOR, J., KLEIN, J.D., HOOGEBOOM, P. and GROOT, J., 1994, On the use of multi-frequency and polarimetric radar backscatter features for classification of agricultural crops. International Journal of Remote Sensing, 9, pp. 1799–1812. GADE, M., ALPERS, W., HUHNERFUSS, H., MASUKO, H. and KOBAYASHI, T., 1998, Imaging of biogenic and anthropogenic ocean surface films by the multifrequency/multipolarization SIR-C/X-SAR. Journal of Geophysical Research, 103, pp. 18851–18866. GUISSARD, A., 1994, Phase calibration of polarimetric radars from slightly rough surfaces. IEEE Transactions on Geoscience and Remote Sensing, 32, pp. 712–715. ITOPF, 2007, Oil Tanker Spill Statistics: 2006. The International Tanker Owners Pollution Federation Ltd. Available online at: www.itopf.com/. JOUGHIN, I.R., WINEBRENNER, D.P. and PERCIVAL, D.B., 1994, Probability density functions for multilook polarimetric signatures. IEEE Transactions on Geoscience and Remote Sensing, 32, pp. 562–574. KUGA, Y. and ZHAO, H., 1996, Experimental studies on the phase distribution of two copolarized signals scattered from two-dimensional rough surfaces. IEEE Transactions on Geoscience and Remote Sensing, 34, pp. 601–603. LEE, J.S., HOPPEL, K.W., MANGO, S.A. and MILLER, A.R., 1994, Intensity and phase statistics of multilook polarimetric and interferometric SAR imagery. IEEE Transactions on Geoscience and Remote Sensing, 32, pp. 1017–1028. LEE, J.S., GRUNES, M.R. and DE GRANDI, G., 1997, Polarimetric SAR speckle filtering and its impact on classification. IEEE Transactions on Geoscience and Remote Sensing, 37, pp. 2363–2373. LEE, J.S., GRUNES, M.R. and POTTIER, E., 2001, Quantitative comparison of classification capability: fully polarimetric versus dual and single-polarization SAR. IEEE Transactions on Geoscience and Remote Sensing, 39, pp. 2343–2351. MIGLIACCIO, M., GAMBARDELLA, A. and TRANFAGLIA, M., 2007, SAR polarimetry to observe oil spills. IEEE Transactions on Geoscience and Remote Sensing, 45, pp. 506–511. RANEY, R.K., 2007, Decomposition of Hybrid-Polarity SAR Data. Available online at: http:// earth.esa.int/workshops/polinsar2007. SCHULER, D.L. and LEE, J.S., 2006, Mapping ocean surface features using biogenic slick-fields and SAR polarimetric decomposition techniques. IEE Proceedings. Radar, Sonar and Navigation, 153, pp. 260–270. SCHULER, D.L., LEE, J.S. and HOPPEL, K.W., 1993, Polarimetric SAR image signatures of the ocean and Gulf Stream features. IEEE Transactions on Geoscience and Remote Sensing, 31, pp. 1210–1221. SOLBERG, A.H.S., BREKKE, C. and HUSØY, P.O., 2007, Oil spill detection in Radarsat and Envisat SAR images. IEEE Transactions on Geoscience and Remote Sensing, 45, pp. 746–755. 1602 CPD for oil spill observation SOLBERG, A.H.S., STORVIK, G., SOLBERG, R. and VOLDEN, E., 1999, Automatic detection of oil spills in ERS SAR images. IEEE Transactions on Geoscience and Remote Sensing, 37, pp. 1916–1924. ULABY, F.T., HELD, D., DOBSON, M.C., MCDONALD, K.C. and SENIOR, T.B.A., 1987, Relating polarization phase difference of SAR signals to scene properties. IEEE Transactions on Geoscience and Remote Sensing, 25, pp. 83–92. ULABY, F.T., SARABANDI, K. and NASHASHIBI, A., 1992, Statistical properties of the Mueller matrix of distributed targets. IEE Proceedings-F, 139, pp. 136–146. VAN ZYL, J.J., 1989, Unsupervised classification of scattering behaviour using radar polarimetry data. IEEE Transactions on Geoscience and Remote Sensing, 27, pp. 36–45.