SINGLE-LOOK SAR IMAGES AND DETECTION OF SEA DARK AREAS Attilio Gambardella(1), Ferdinando Nunziata(2), Antonio Sorrentino(2), Giuseppe Ferrara(2) and Maurizio Migliaccio(2) (1) Università degli Studi di Cagliari, Dipartimento di Ingegneria Elettrica ed Elettronica, Piazza d’Armi 19, 09123 Cagliari, Italy. e-mail: attgamba”at”diee.unica.it. (2) Università degli Studi di Napoli “Parthenope”, Dipartimento per le Tecnologie, Via Medina 40, 80133 Napoli, Italy. e-mail: nunziata”at”uniparthenope.it. ABSTRACT In this paper we present a novel approach for the detection of low backscatter areas and areas with strong scatterers in marine SLC SAR images. The approach is based on the use of the three parameters of the generalized K probability density function. This speckle model embodies the Rayleigh scattering scene, which is descriptor of scenes dominated by Bragg scattering, the Rice scattering scene, which is descriptor of areas in which a dominant scatter is present, and the K-distribution, which is descriptor of non-Gaussian signal statistic which normally characterize the sea echo from free-area illuminated by an high-resolution radar. A large data-set of ERS 1/2 SLC SAR images, provided by the ESA under the CAT-1 Project C1P-2769, is employed. Results show the effectiveness of the approach. 1. INTRODUCTION Synthetic Aperture Radar (SAR) is the key sensor for several marine remote sensing applications. In this study the stochastic nature of a SAR image is investigated as an additional source of information to be exploited to learn more about the underlying physics. Classical Rayleigh distribution model for the amplitude of backscattered signal well fits homogeneous land scenes and the backscattered signal from sea surface when a large area is illuminated by the radar. It represents a reference since it is associated to the so-called fully-developed speckle [1]. Moreover, when an high-resolution radar illuminates sea surface large deviations from Rayleigh statistic are often found. The K probability density function (pdf) has been found to be a particularly useful model for sea amplitude statistics [2]. It reduces to the Rayleigh pdf under appropriate circumstances [2]. Moreover, the K pdf model has been generalized to model the amplitude of backscattered signals with a non-uniform phase distribution [3]-[4]. The proposed model is based on the GK pdf, which ensures a continuous and physically consistent transition among different scattering scenarios [3]. This _____________________________________________________ Proc. ‘Envisat Symposium 2007’, Montreux, Switzerland 23–27 April 2007 (ESA SP-636, July 2007) speckle model embodies the Rayleigh scattering case, which is descriptor of scenes dominated by Bragg scattering, the Rice scattering case, which is descriptor of areas in which a dominant scatter is present, and the K-distribution, which is descriptor of non-Gaussian signal statistic which normally characterize the sea echo from free-area illuminated by an high-resolution radar [4]. The rationale of this study is to characterize the electromagnetic return of selected regions of interest (ROIs) over Single-Look Complex (SLC) SAR images by means of the three characteristic parameters of the GK distribution α, η and ν. Instead of suppressing the speckle by means of appropriate averaging, which makes the SAR image spatial resolution coarser, it is possible to extract useful information from SAR fullresolution images once that the speckle is properly modeled. As a matter of fact, GK pdf is able to describe the high-resolution SAR statistical behavior and is advisable when small low backscatter area (i.e. dark areas) and strong scatterers need to be detected. On the applicative view point, this study can assist classical oil-spill detection techniques. Results show the effectiveness of the approach once that the differential GK parameters are considered. 2. THEORY BACKGROUND According to the two-scale model the surface height is partitioned into a large-scale and a small-scale displacement [5]-[7]. Thus, over any patch of the surface, large compared with small-scale length but small respect to the large-scale length, the scattering can be generally modeled as first-order Bragg scattering from the small inhomogeneities [5]-[7]. The effect of the large-scale structure is to modulate, both linearly and non-linearly, the small-scale scattering [2], [8]. When the dominant mechanism is Bragg scattering it is possible to assume a fully developed speckle model for the statistical fluctuations of the backscattered signal. This is the case of sea surface when a weak large-scale modulation occurs or the area illuminated by the radar is too large with respect to the large-scale structure [2]-[8]. In this case the scattered field can be written as the sum of N elementary fields; mathematically it can be seen as a N-step twodimensional random walk in a complex plane [3]. When N is a given number, assuming statistical independence between the n-th step length {an} and the associated phase shift {φn}, if {φn} are uniformly distributed, the central-limit theorem, for asymptotically N large, can be exploited and fully developed speckle regime is achieved [4]. In this case the scattered field can be described by a zero-mean circular complex Gaussian process. Hence, the field amplitude and intensity are Rayleigh and exponentially distributed, respectively [3]-[4]. However, the above mentioned assumptions of randomness and statistical independence, although likely in the case of a confused short-crested sea, may fail in presence of a large-scale modulation due to a long-wave field [2]. Thus, under these conditions, large deviation from Gaussian statistic is often found, particularly when an highresolution radar, such as SAR, is considered [2]. In this case the, modulation of small-scale inhomogeneities by the larger ones leads to the well-known bunching phenomena, which characterizes sea surface [4], [9][10]. This phenomena can be included in the random walk taking N as a random number fluctuating according to a negative binomial (NB) distribution, ruled by a non-negative bunching parameter α. For asymptotically large mean step number and under the hypothesis of strong scattering a non-fully developed model based on the K-distribution is achieved [3]-[4]. The K pdf, relevant to the intensity of the scattered field (i ≡ |E|2), is given by [2]: f (i ) = 2 α Γ(α )η α +1 α +1 2 ⎞ ⎛2 Kα −1 ⎜⎜ α i ⎟⎟ ⎠ ⎝η , (1) where Γ(·) is the Eulerian gamma function, Kα-1(·) is the modified Bessel function of the second kind of order α-1 and η is the slope parameter. This is a twoparameters function which has shown a good agreement in fitting experimental data in a wide range of non-Gaussian scattering configurations, including microwave sea echo [2], [4], [8], [11]-[12]. Eq. (1) is a function of the modified Bessel function of the second kind and as the order of Kα-1(·) gets larger the NB distribution, associated to N, tends to a Poisson distribution and the shape of (1) finally becomes an exponential pdf, i.e. a Rayleigh pdf for the field amplitude [3]-[4]. α is unambiguously referred as shape parameter and can be used as indicator of the degree of non-Gaussian signal statistics [12] while the η parameter, under uniform large-scale conditions, is mainly related to the intensity of the local backscattering, due to the small-scale inhomogeneities [8], [11]. According to [3]-[4] it is possible to generalize this model considering a biased two-dimensional random walk (i.e. a non-uniform phase distribution). In this case, for large mean step number, a GK pdf is obtained. This is a three parameters pdf whose expression, for the field intensity, is given by [3] : ⎞ 2 ⎛ ⎟ ⎜ α −1 ⎟ ⎜ α 2α 2 f (i ) = i Io ⋅ ⎟ ⎜ Γ(α )η α +1 ⎜ ⎛ ν 2 ⎞ ⎟ ⎜ ⎟ 1+ ⎜ ⎜ 4α ⎟ ⎟ . ⎠⎠ ⎝ ⎝ 1 ⎫ ⎧ ⎞ ⎛ν ⎪ 2 ⎡⎛ ν 2 ⎞ ⎤ 2 ⎪ ⎟α i ⎥ ⎬ i ⎟⎟ Kα −1 ⎨ ⎢⎜⎜1 + ⋅ ⎜⎜ 4α ⎟⎠ ⎥⎦ ⎪ ⎝η ⎠ ⎪η ⎢⎣⎝ ⎭ ⎩ α −1 (2) α and η are the same parameters shown in (1) and ν deals with the presence of a dominant scatterer. The ν parameter is expected to reveal the presence of a nonnegligible coherent component in the backscattered sea surface signal, due, for example, to a ship. Since here main concern is on the analysis of the consistency of the physical approach, only a suboptimal and computer time effective parameter estimation procedure is considered [13]. According to such approach ν and η parameters are estimated by means of the Rice factor, i.e. the coherent to incoherent received power ratio. The α parameter is estimated by means of a numerical procedure, based on the χ2 test, which minimize the L2 error norm between the measured pdf and the theoretical GK pdf, where the α parameter is left free [5]. 3. EXPERIMENTAL RESULTS In this section the behavior of the three GK distribution parameters, as descriptors of marine dark areas and strong (although small) scatterers over SLC SAR VV polarized C-band images, is presented and discussed. The reference data set, i.e. the ERS 1/2 SLC SAR VV polarized C-band images, has been provided by the European Space Agency (ESA) under the CAT-1 Project C1P-2769. The nominal resolution is 10 meters in range and 5 meters in azimuth. The study has been conducted evaluating the GK parameters relevant to the different ROIs. In order to better appreciate the parameters sensitivity, each ROI is coupled with a reference area. Such area is identical in size, is a non-dark area and without dominant scatterers. To further assist the interpretation of the results, the wind speed is evaluated by the SAR images [14]-[16]. The evaluation of the wind speed by SAR data, although sub-optimal [14], allows overcoming the incorrect time/spatial co-location between SAR data and operational wind data and retrieves the wind speed at a finer spatial scale (1-2 Km) [15]-[16]. The first data set is relevant to the acquisition of 26 July 1992, 9:42 UTC (ERS-1, SLCI, orbit 5377, frame 2889, descending pass) off the coast of Malta. Fig.1 shows the quick-look of the SAR image. The ROIs, which are manually selected in the SAR image, are shown in Fig. 2 which is made by two sub-images excerpted by Fig. 1. In the right and bottom-right side of the image (see Figs. 1 and 2a) several oil slicks are present. On the left side of the image (see Figs. 1 and 2b), near the coast of the isle of Malta, a low wind area is present, in the top-middle of the image some ships and small oil slicks are present. The ROIs relevant to low-wind areas are labeled as “LW”, oil slicks are labeled as “Oil” and the ROIs including the ships are labeled as “S”. In Fig. 2c a zoom of the SAR image relevant to Oil4 area is shown. The same format is used in all subsequent experiments. The SAR image regards a sea surface area characterized by a low-to-moderate wind speed. In Tables 1 the GK parameters are listed for all ROIs and corresponding reference areas. In all cases the χ2-test of hypothesis with a 95% confidence level was applied and passed. Analysis of the three GK parameters shows that the absolute value of η is useless, the absolute value of α is of some interest and the absolute value of ν is useful. Therefore the differential parameters η% and Δα have been introduced. They are defined as follows: ⎛ η ROI − ηRef ⎞ ⎟⎟ ⋅ 100 , ⎝ η Ref ⎠ η% = ⎜⎜ (3) and Δα = α ROI - α Ref . (4) With respect to η, all these experiments show that the dark areas are characterized by a negative η% value generally greater than 50%. This is untrue for the Oil3 and LW2 ROIs, where η% value is negative but higher. This can be physically explained by the sea-oil mixing occurring in the Oil3 area and in general by the heterogeneity of these two ROIs, see Figs. 1 and 2. Physically, the characterization of dark areas in terms of negative η% value is due to a smaller roughness in dark areas. It is also important to note that the S1 and S2 ROIs, characterized by the presence of a ship, exhibit a very high positive η% value. With respect to α it must be noted that the notation “∞” is used to emphasize that a high value (>60) is measured, since this means that a Gaussian scattering regime is practically due. Results listed in Tables 1 clearly show that the dark areas are characterized by an high negative Δα value witnessing a change into the scattering regime. When the scattering regime changes from the non-Gaussian to the Gaussian case Δα is equal to -∞, when scattering regime changes from different non-Gaussian cases Δα is finite and negative. With respect to ν, results show that, as expected, it is very low for all dark and reference areas. In S1 and S2 ROIs ν is much greater than usual values. To emphasize this it is appropriate to consider ν% defined analogously to (5). As result S1 and S2 ROIs are characterized by very high and positive ν% values. It must be noted however that, surprisingly, ν is not as large as one would expect. In order to physically explain such unexpected result a detailed analysis was accomplished. Relevant pdfs (not shown to conserve space) are indistinguishable from a K pdf, but for some strong isolated outliers. These outliers witness the presence of few strong dominating scatterers. As matter of fact a Rice pdf in not achieved but the GK parameters estimation procedure (section 2) fails to estimate the α value because the measured pdf shows some gaps and bins with very few number of samples [17]. The notation “NA” is used to emphasize this. In conclusion, the used GK parameters estimation procedure is able to emphasize ROIs in which a dominating scatterer is present. The second case regards a well-known oil spill accident which has been widely studied and documented, e.g. [18]. The data set is relevant to the acquisition of 02 June 2003, 21:02 UTC (ERS-2, SLCI, orbit 42439, frame 1107, ascending pass). Fig.3 shows the quick-look of the SAR image and the selected ROIs. In this case all ROIs are relevant to oil slicks. It is shown an oil slick of about 39 kmq offshore south of the Swedish coast caused by the Fu Shan Hai bulk carrier accident which occurred north of the Danish island of Bornholm in the Baltic Sea, on Saturday 31 May. More than 100 tonnes of heavy fuel oil began streaming from the ship hull, and currents carried the oil towards Sweden [18]. The interest on this data set is twofold: the dark area is relevant to a spillage 48 hours old and the scene is characterized by a high wind speed (8 m/s) [18]. Relevant results are shown in Table 2. Physical results are in agreement to what formerly experienced. However, it must be noted that the η% values are greater than former one, see Table 1. This is physically due to the oil spill aging process which is not balanced by the high wind speed regime [19]. 4. CONCLUSIONS In this paper it has been shown that full-resolution speckled SAR images can be effectively employed to highlight the presence of low backscatter areas and small strong scatterers. Classical approaches are not able to process such SAR images and therefore deal with multi-look SAR images where, at the expense of a poorer spatial resolution, the speckle is mitigated. The new approach is based on the GK model which is able to embody very different scattering cases. Therefore, it is possible to take advantage of such SAR fullresolution SAR images. Although this approach is not able to discriminate among the physical causes generating dark areas, the results demonstrates that GK parameters are useful and that the analysis is simple and effective. As a matter of fact, the approach can be envisaged as an additional tool to assist classic oil spill observation procedures. ACKNOWLEDGMENTS The authors acknowledge the ESA for providing the SAR data (C1P-2769), the NASA (POODAC) for providing the scatterometer wind data and the ECMWF data centre for providing the wind data. Figure 1. Quicklook image of the area of interest relevant to the acquisition of 26 July 1992, 9:42 UTC (ERS-1, SLCI, orbit 5377, frame 2889, descending pass) off the coast of Malta. (a) (b) Figure 2. Zooms of two part of SAR image of Fig.1. Table 1. Measured GK parameters relevant to ROIs of Figs. 2(a)- 2(b). ROI Oil 1 Ref. Sea Oil 2 Ref. Sea η 0.0009486 0.0008612 0.001968 Oil 3 0.001568 0.002083 Oil 4 0.001083 Ref. Sea 0.002085 Oil 5 0.001122 Ref. Sea 0.002194 Ref. Sea Oil 7 Ref. Sea LW 1 Ref. Sea LW 2 Ref. Sea S1 Ref. Sea S2 Ref. Sea -52.9 0.002014 Ref. Sea Oil 6 η% 0.0007162 0.001520 0.0006750 0.001520 0.0002059 0.001093 0.0009361 0.001331 0.01387 0.001320 0.01225 0.001320 α 9 Δα -11 20 -56.2 -24.7 -48.1 -48.9 -52.8 -55.6 -81.2 -29.7 +950 +828 9 31 13 ∞ 4 ∞ 35 ∞ 6 40 3 40 41 54 1 ∞ NA ∞ NA ∞ ν 0.08 ν% Wind +14.2 Moderate -33.3 Moderate -60.0 Moderate -20.0 Moderate -57.1 Moderate -72.7 Low -81.8 Low -42.8 Low 0.0 Low +766 Low +1400 Low 0.07 -22 -∞ -∞ -∞ -34 -37 -13 -∞ - 0.02 0.03 0.02 0.05 0.04 0.05 0.03 0.07 0.03 0.11 0.02 0.11 0.04 0.07 0.04 0.04 0.26 0.03 0.45 0.03 Figure 3. Quicklook image of the area of interest relevant to the acquisition of 02 June 2003, 21:02 UTC (ERS-2, SLCI, orbit 42439, frame 1107, ascending pass) and sketch of the ROIs selected for parameters estimation. Table 2. Measured GK parameters relevant to ROIs of Fig. 4 ROI η Oil 1 0.001074 Ref. Sea 0.001343 Oil 2 Ref. Sea Oil 3a Ref. Sea Oil 3b Ref. Sea 0.0008146 0.001321 0.0007391 0.001185 0.0009366 0.001185 η% -20.1 α 5 2. 3. 4. 5. 6. 7. 8. 9. -33 38 -38.3 -37.7 -21.0 10 20 15 33 17 33 5. REFERENCES 1. Δα Beckmann, P. & Spizzichino, A. (1963) The Scattering of Electromagnetic Waves From Rough Surfaces, Artech House, Norwood, MA. Jakeman, E. & Pusey, P.N. (1976). A Model for Non-Rayleigh Sea Echo, IEEE Trans. Antennas Propag. 24(6), 806-814. Barakat, R. (1986). Weak-Scatterer Generalization of the K-Density Function with Application to Laser Scattering in Atmospheric Turbulence, J. Opt. Soc. Am. A. 3(4), 401-409. Jakeman, E. & Tough, P. (1987). Generalized K Distribution: a Statistical Model for Weak Scattering, J. Opt. Soc. Am. A. 4(9), 1764-1772. Jackson, C.R. & Apel, J.R., Ed. (2004) Synthetic Aperture Radar Marine Users Manual, NOAA, Washington, DC. Apel, J.R. (1990). Principles of Ocean Physics, Academic Press: International Geophysics Series, London, UK. Lemaire, D., (1998). Non-fully Developed Sea State Characteristics from Real Aperture Radar Remote Rensing, PhD Thesis, Université Catholique de Louvain, Louvain-la-Neuve, Belgium. Greco, M., Bordoni, F. & Gini, F. (2004). X-Band Sea-Clutter Nonstationarity: Influence of Long Waves, IEEE J. Oceanic Eng. 29(2), 269-283. Alpers, W., Ross, D. & Rufenach, C. (1981). On the Detectability of Ocean Surface Waves by Real and Synthetic Aperture Radar, J. Geophys. Res. 86(C7), 6481-6498. ν 0.01 ν% Wind -66.7 High -50.0 High 0.0 High -50.0 High 0.03 -10 -18 -16 0.01 0.02 0.02 0.02 0.01 0.02 10. Garello, R., Proust, S. & Chapron, B. (1993). 2D Ocean Surface SAR Images Simulation: A Statistical Approach, Proc.OCEANS’93 3(7-12). 11. Jakeman, E. (1980). On the Statistics of Kdistributed Noise,” J. Phys. A. 13(1), 31-48. 12. Eltoft, T. & Høgda, K.A. (2004). Non-Gaussian Signal Statistics in Ocean SAR Imagery, IEEE Trans. Geosci. Rem. Sensing 36(2), 562-574. 13. Corona, P., Ferrara, G. & Migliaccio, M. (2004). Generalized Stochastic Field Model for Reverberating Chambers, IEEE Trans. Electromagn. Compat. 46(4), 655-660. 14. Migliaccio, M. & Reppucci,A. (2006). A Review of Sea Wind Vector Retrieval by means of Microwave Remote Sensing, Proc. European Microwave Association 2(1), 136-140. 15. Fetterer, F., Generis, D., & Wackerman, C.C. (1998). Validating a Scatterometer Wind Algorithm for ERS-1 SAR, IEEE Trans. Geosci. Remote Sensing 36(2), 479-492. 16. Horstmann, J. & Koch, W. (2005). Measurement of Ocean Surface Winds Using Synthetic Aperture Radars, IEEE J. Oceanic Eng. 30(3), 508-515. 17. Corona, P., Ferrara, G. & Migliaccio, M. (1996). Reverberating Chambers as Sources of Stochastic Electromagnetic Fields, IEEE Trans. Electromagn. Compat. 38(3), 348-356. 18. http://ocean.dmi.dk/spill/FuShanHai/ 19. Espedal, H.A. & Wahl, T. (1999). Satellite SAR Oil Spill Detection Using Wind History Information, Int. J. Remote Sens. 20(1), 49-65.