Interpretation and
Comparision of Air Sar Quad-Polarised Radar Images D.Amarsaikhan, M.
Ganzorig Abstact Geographician (cartographer) Institute of Informatics and RS, Mongolian Academy of Sciences, Av. Enkhtaivan-54B, Ulaanbaatar-51, Mongolia The aim of this paper is to describe and analyses the basic scattering characteristics for five selected classes such as water, forest, grass, residential and high density urban areas and how they vary with wavelength, polarization, and incident angle. For this purpose, quad-polarised images of Syney area taken from airborne SAR have been used. The results are discussed and some explanations on findings are given. Introduction Generally, the radar backscatter received from a surface is determined by incident angle, wavelength, polarization, surface roughness and dietectric properties of the surface. At radar wavelengths, 3 types of scattering such as surface scattering, volume scattering, and corner reflector-like scattering can occur. If the surface is homogeneous then surface scattering will occur and it can be either specula or diffuse, or intermediate depending on the wavelength and surface roughness. If the surface is dielectrically inhomogeneous then from the underlying materials will occur. The depth to which the radiation penetrates is dependent on the wavelength and the water content of the volume material. In volume scattering when the density of scatterers is low, the dependence of the backscattering coefficient is only slight and as the average dielectric constant increases, the dependence on incident angle increases. Corner reflector-like scattering occurs in result of right angles formed between natural and artificial objects. Polarisation can be reflected in various ways from the natural objects. If the surface is sufficiently rough then both like and cross polarized fields can be received. HH polarized image will be dominated by reflectance coming from surface scattering mechanisms. HV and VV polarized data will display a greater component of volume scattering as a result of the signal penetrating to some depth below the surface. In general, the brighter the return on HV or VV images, the more likelihood the backscatter is coming from a three dimensional or heterogeneous layer [1,4]. Radar with its side viewing mode and which introduce different distortions and require various procedures for their correction. The aim of this paper is make judgment on the basis of interpretation of backscatter values and geometric problems will not be discussed. For the analysis, five classes such as water, forest, grass, residential and high density urban areas of Sydney area have been selected. The classes are compared using individual and group pixels. For analysis of individual pixels, the backscatter values of pixels selected from different parts of the image were used, whereas for group analysis, 101-159 contextually dependent pixels representing the selected classes have been selected and compared on the basis of mean value and standard deviation (SD). The Basic Theory of Scattering Mechanisms for the Selected Classes The selected classes have different backscattering properties. The following describes the basic theory of scattering mechanisms of each class [2,3,4,5]. The aim of radar RS at the water surface is to determine wave amplitudes, surface wind velocities and directions. A perfectly flat sea will behave as a specular reflector and will appear black in imagery for all incident angles except 0. To obtain some backscatter from a sea surface, it must by some mechanism, be made rough and the principal mechanism for the roughening the surface of the sea is the generation of waves. Different types of waves having different means for generation and characteristics are given in Richards et al. (1987). Data at P(75cm), L(18.7cm), C(6.3cm) and X(3.3cm) reported by Guinard and Daley in ( Manual of RS : 1391) appear to saturate for wind velocities in excess of 10knots; however, data recorded at the higher frequencies varied more with wind speed than did the lower frequency data. Open grass will act as a mixture of grass and soil and the backscatter will depend on the volume of either of them. Plant geometry, density and the water content are the main factors influencing on the backscatter coming from the vegetation cover. Most probably, grass and extended vegetated surfaces could have components of all (ie, diffuse, specular and intermediate), reflection depending on the wavelength and incident angle. At angles close to nadir and frequencies below about 8GHz (ie, C,L,P bands) the presence of vegetation cover (crops and grass) has a minor influence on the backscatter (Manual of RS). The backscattering of soil will depend on the surface roughness and incident angle. The presence of water strongly affects the microwave emissivity and reflectivity of a soil layer. At low moisture levels there is a low increase in the dielectric constant. Above a critical amount the dielectric constant rises rapidely. This increase occurs moisture is directly related to the texture and structure of the soil. Therefore, the backscatter values from open grass will not be as high as in the case of pure volume and corner reflector-like scattering. In case of forest areas trunk-ground double bounce scattering, branch-ground double bounce and branch-direct backscattering, crown volume backscattering and crown volume attenuation and ground backscattering can occur, ie, backscatterfrom forest will be volume scattering derived from multiple-path reflections from leaves, twigs and trunks. Considering a mixture of different grass and forest types as vegetation, for its mapping Ulaby (1982) recommended the use of frequencies greater than 8GHz and moderate depression angles. This is probably based on the increasing penetration and sensitivity to soil properties underlying underlying vegetation with lower frequency and near nadir incident angles. Thus, short wavelength can be used for top layer study, whereas long wavelength for lower layer study. When incident angle increases in the far range, more volume scattering should be expected due to path difference of radiation. The blackscatter from urban areas will contain information about street alignment, building size, density, roofing material, its orientation, vegetation and soil resulting in all kinds of scattering. Roads and buildings in urban areas can reflect a larger component of radiation if they are aligned at right angles to the incident radiation. Here the intersection of a road and a building tends to act as a corner reflector. The amount of blackscatter is very sensitive to street alignment. The variation in return between neighbouring areas of streets and buildings aligned at right angles to the incident radiation will have a saturated very bright appearance and non-aligned areas will have a more speckled, bright/dark appearance in the resulting image. Volume and surface scattering will also play an important role in the response from urban areas. Using L (23.5) band and Rayleigh's criterion of surface roughness the boundary between diffuse and specular relection can be determined at about 3cm when incident angle is 60 digress. Many urban surfaces have variations that approximate, are greater than or fall below these values. For example, bitumen and concrete surface would always have variations of less than 3cm and would generate a specular response, and appear dark at any incident angle, while roofing material, grass broken soil and extended vegetated surfaces could have both diffuse and specular reflection depending on the incident angle [4]. Analysis and Discussions The observed brightness values of individual pixels representing the selected classes are given in table 1. As seen from the table water has the lowest values in comparison with other classes. This is due to a specular reflection of seawater. Generally, in case of equal calibration more backscatter is expected in C band because of its wavelength nature in which more diffuse reflection could be expected than in L or P bands. The pixel -5 selected at the Coogee Bay has more higher values than the others do. According to the theory, in a far range more specular reflection, which results in lower backscattering should be expected. However, this is most probably because of a fact that in this area wind is much stronger than in the Sydney harbour area (where the other 4 pixels were selected). The strong wind makes the sea surface much rougher, which can result in more diffuse scattering. The statistics of the brightness values of pixel groups is given in table 2. As seen from table 2, in case of case of calm sea the backscatter means are lower than the rough sea surface and like polarized fields in all bands have higher mean values than cross polarization but they from more compact clusters than the other two polarization.
The pixels representing the forest trees wee in areas situated along the northern coast of Sydney harbout. The area might have more soil moisture, which can contribute to high backscatter. Here, X band can be used for the study of top leafy layer of the trees, whereas L and P bands for the study of the lower layer of the trees. The areas in total will behave as a diffuse reflector due to volume scattering although some other scattering could also be expected. In the case of pixel-2 more volume scattering will occur, because it is standing further away from the other pixels (i.e., increase in (-)). As seen from both tables, the average brighness value of forest is close to each other in like polarized fields of C and P bands but for the contextually dependent pixel group it is similar in L band even in cluster formation. This could have been due to the backscatter from 3D or heterogeneous layer. HV polarization has the lowest values in all cases but it creates more compact clusters in C and L bands. It can be very useful if one is doing a land cover mapping using one of the existing pattern recognition techniques. It we need to create more compact cluster in C and L bands. It can be very useful if one is doing a land cover mapping using one of the existing pattern recognition techniques. If we need to create a FCC image, which indicates the most tonal differences among the forest classes then, a combination of P-HV, L-VV or HH and C-HH are preferable. The pixels representing grass were selected at the Randwick area (expect pixel-5). Here, the grass could have components of diffuse and specular reflection depending on the wavelength and incident angle. It is interesting to note that the average backscatter values in HH polarization of L and P bands are higher than in other polarization. Further, it can also be seen from the tables that the selected samples form more scattered clusters. Pixel-5 was selected in an area situated close to a pond, which should have more soil moisture than others. In Manual of RS (1983:1587) it is argued that vegetation with high moisture content tends to reflect rather than transmit incident energy, the return may be brighter for such surfaces. As seen from table 1, pixel-5 has the highest values almost in all polarization fields. If one wants to see the best FCC images indicating the tonal differences in grass. If one wants to see the best FCC image indicating the tonal differences in grass class, then a combination of P-HH or VV, L-HV and C-HH can be used. The mean backscatter values representing the residential areas have the highest values in HH polarization but they form very scattered clusters (i.e., high SDs). Here, it is interesting to see another parameter, which is the orientation f the objects influencing the radar return in such an area, eg, some pixels selected in residential areas have very high brightness values in comparison with other pixels representing the same class. It might occur due to the fact that the metallic roofs of the buildings in this area are facing the radar beam. Pixel-3 selected at road has high values in all cases. This is probably due to the fact that it is aligned at right angles to the radar signal. Otherwise, it should have had low brightness value. Pixel-2 was selected on the border between some trees and buildings. It has high values in HH polarization in C band and VV polarization of L band. It might be due to interaction between 2 scattering mechanisms; e.g., trees act as a volume scatterer, while buildings bordering or lying beneath the trees act as corner reflectors. As seen from the tables there is a very high contrast between the values representing the residential class. Therefore, it would be not easy to draw a decision boundary between this and other classes because of its highly scattered values unless there is a purification of the selected samples. However, a good image representing the tonal differences among the features can be created by combination of P-HV or VV, L-VV and C-HH. As seen from the tables, a high density urban area reflects very high values due to a corner reflector effect. It is interesting to note that pixel-5 which was selected on the border between the high density urban and the residential areas has lower values in P band than other pixels. It can be seen from table 2 that like polarized fields in all bands form compact clusters while cross polarized fields create very scattered clusters. Furthermore, it is possible to judge that HH and VV polarization can be used in a classification (surely, excluding pixle-4) for land cover mapping. Unlike other classes, here only one band can be used to distinguish the high -density urban area from other classes. Conclusions The aim of this study was to compare and analyses the backscatter values of quad-polarized images on the basis of the wavelength, polarization and incident and incident angle. As seen from the analysis, the backscatter values in the cross polarized field has lower values than the like polarized field. Also, it can be seen that if one type of scattering is dominated for a specific class it created more compact cluster, while if it is a mixture of different scattering mechanisms then rather scattered clusters are created. This kind of study might be very useful for preliminary data analysis before statistical pattern recognition of image segmentation are applied. Acknowledgement The authors are very grateful to Prof. B. Forster, School of Geomatic Engineering, UNSW, Australia for providing relevant data for this study. References
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||