Radar Polarimetry- Potential for Geosciences Franziska Kersten Department of geology, TU Freiberg Abstract. The ability of Radar Polarimetry to obtain information about physical properties of the surface has led to many innovative applications in geoscientific research and concerning environmental issues. In the first part the advantages of polarimetric radar data are presented based on a review of conventional radar systems and single-channel Synthetic Aperture Radar (SAR). Radar Polarimetry allows the estimation of surface parameters of geoscientific interest such as surface roughness and soil moisture as illustrated in two presented case studies in the second part. Introduction Radar Polarimetry is a new development in the field of Remote Sensing. Providing information about physical parameters and geometric properties of the Earth surface it has become an important tool in Earth monitoring in recent years. Therefore it is already and will be used in the future in a vast variety of (geo-)scientific fields, such as geology, hydrology, cartography, oceanography, meteorology but also forestry and agriculture. This paper provides some basic knowledge about development, working principles and application possibilities of the currently used radar technique SAR and its extension POL-SAR (Polarimetric-SAR). Microwave remote sensing and SAR Radar is a technique operating in the microwave region of the electromagnetic spectrum, with wavelengths between 1cm and several meters (Fig.1). The signal is monochromatic and coherent.
2 Franziska Kersten Fig. 1 electromagnetic spectrum illustrating the wavelengths used in microwave remote sensing, the coloured areas indicate the microwave bands often used in SAR systems, modified after Oliver, Queguan (1998) Imaging radar systems can be either airborne or spaceborne radars. Short radar pulses are emitted to the ground. The received reflected electromagnetic waves contain information about the surface, which allows generation of twodimensional reflectivity maps of the illuminated terrain. Thus, radar being an active system, which is independent of the illumination by sunlight it can operate day and night. Another characteristic of the used electromagnetic waves not being disturbed by water drops makes radar imaging possible in all weather conditions. One big advantage in comparison to optical remote sensing lies in the high resolution that can be achieved. Since resolution in azimuthal direction depends on the size of the antenna footprint on the ground, it shows a significant decrease with increasing flight height. Conventional radar systems also require a rather large antenna to provide high resolution images. The SAR-System developed in the 1950 s based on the works of Carl Wiley (1951) overcomes those problems by generating a synthetically enlarged antenna using the Doppler-effect, thus making it possible to operate within high distances to the illuminated terrain. Spaceborne SAR Images show a resolution of about 5m, airborne SAR provides even higher resolution up to 30cm. SAR images contain information about morphology and ground conductivity of the observed terrain because the geometric and dielectric properties of the surface influence the reflection or backscattering of the emitted electromagnetic wave (Oliver, Quegan 1998). Fig.1 illustrates the wavelengths used in microwave remote sensing. Microwaves are capable of penetrating vegetation or soil up to a certain depth, depending on their wavelengths and certain physical parameters of the
Radar Polarimetry-Potential for Geosciences 3 observed targets, such as densitiy, conductivity and the complex dielectric constant (e.g. Sokol et al. 2004). Since dielectric properties correspond to land cover and vegetation SAR has proved being a good tool in forestry and agriculture (e.g. Ferro-Famil et al. 2003). Measurements are mainly carried out in the X-,C-, P- and L- band as emphasized in Fig.1. L- and P- band use longer wavelengths, which penetrate deep into the ground or vegetation, that is why the backscattering contains contributions from the entire volume. The shorter wavelenghts used in X-band lead to backscattering mainly from the surface or top of vegetation, thus providing information about the top layer. In processing and analysing SAR data it is of great importance to determine the different contributions to the backscattered waves. Since only the total backscattering is measured inversion to surface parameters of interest turns out to be problematic without a priori information. Polarimetry is a major extension of SAR developed to resolve these limitations. Polarimetric SAR The key difference in polarimetric SAR data in comparison to conventional SAR data is that measurements are carried out in multiple frequencies or polarizations. The obtained multidimensional information allows the identification of different scatterers via discrimination of different scattering mechanisms (Pottier et al. 2003). Polarimetric data are available since the early 1980 s but have only come in use in recent years because of the complex processing they require.there has been a growing number of airborne sensors capable of acquiring polarimetric data ever since then. The first spaceborne sensor SIR-C/X-SAR was launched in 1994. The polarization information contained in the backscattered wave is directly related to the scatterer s geometrical structure, its orientation and even more important for geoscientific and related applications its geophysical properties (Pottier et al. 2003). Polarimetric data has proven useful for surface roughness and soil moisture estimations, thus being applied not only in hydrology, land cover and land use studies but also in weathering erosion studies and geological mapping. In the field of remote sensing polarimetric data have led to a significant improvement in applying unsupervised classification methods to radar images. Acquiring, Processing, Analysing Since Radar Polarimetry deals with the vectorial nature of microwaves the polarization state of the original polarized transmitted wave can be described by an electromagnetic field vector.
4 Franziska Kersten Measurements are usually carried out in the H-V-basis, meaning the wave being either horizontal (H) or vertical (V) linearly polarized. The backscattered wave will show a different polarization state due to depolarization effects occuring when a wave travels through a certain medium on the one hand. On the other hand waves backscattered from a building for example are completely polarized while backscattering from any kind of vegetation occurs partially polarized (Pottier et al. 2003). Comparison between emitted and received wave consequently leads to conclusions about the scattering mechanism and the type of scatterer. High performance polarimetric SAR systems are able to provide measurements of coherently transmitted and received signals on four channels. Following this, all four possible polarization combinations HH, VV, HV and VH between received (first letter) and transmitted (second letter) wave can be obtained at once. The scattering matrix S is then used to describe the backscattered wave for the case of deterministic scattering. In nature random scattering leading to partially polarized waves is observed in most cases. Non-deterministic scattering is described using the polarimetric covariance or coherency matrix. For completely polarized waves as a result of deterministic scattering, the scattering matrix S contains the scattering contributions of the two copolarized and the two crosspolarized signals. The elements of the scattering matrix are the polarimetric backscattering coefficients. Any combination of transmit and receive polarization can be computed using this scattering matrix. The four signals measured on H-Vbasis are sufficient to synthesize circular polarized signals. Such fully polarimetric data offer the highest information content and are provided for example by the U.S. spaceborne system SIR-C/X-SAR or the german airborne E-SAR. Generally the more information the scattering matrix contains, the more scattering mechanisms and scatterer properties can be identified. Fig. 2 the three main scattering contributions to the received signal, depending on the target, modified after Pottier et al. 2003 According to Pottier et al. (2003) there are basically three different scattering mechanisms influencing the polarization state of the backscattered wave. Single, double and volume scattering can occur depending on the observed target as illustrated in Fig. 2.
Radar Polarimetry-Potential for Geosciences 5 Applying a polarimetric target decomposition leads to discrimination of the different scattering contributions and inversion of surface parameters such as the dielectric constant. In the past various decomposition theorems have been proposed, such as model based or eigenvalue based decomposition (compare Freeman; Pottier et. al). Depending on the applicability of the used decomposition method in a certain case the results lead to a more or less successful extraction of physical parameters. Geoscientific Applications As already mentioned data acquired by polarimetric SAR-sensors show high potential for a wide range of applications in geoscience. The received radar wave can be described as a function of the geometric characteristics and physical properties of the surface on the one hand and of the frequency and incidence angle of the transmitted wave on the other hand (Mattia et al.1997). The fact that radar backscatter is sensitive to the dielectric constant allows the estimation of surface parameters like soil moisture content and surface roughness, which are related to dielectric properties (e.g. Huadong et al. 1997). Hajnsek et al. (2003) emphasize that these two parameters are closely linked to one another. It is one of the benefits of using POL-SAR data, that independent measurements of surface roughness and soil moisture content are possible. In the following, two case studies concerning inversion of the two mentioned surface parameters from polarimetric SAR data are presented. surface roughness estimation according to Dierking and Haack (1998) On the basis of fully polarimetric L-band data aquired by the danish airborne sensor EMISAR in August 1997 over an investigation area in the Northern Volcanic Zone in Iceland a surface roughness inversion was proposed by Dierking and Haack in 1998. Here surface roughness is used as a parameter for distinguishing between lava flows and surrounding terrain as well as for the identification of different lava facies within one flow. Using the fact that lava surfaces show typical radar scattering signatures, the authors have been able to draw conclusions which radar configurations are optimal for geological mapping in comparable terrains. Parameters like frequency, incidence angle, polarization of the radar waves and of course spatial resolution have great influence on the performance of the proposed surface discrimination method. With a given frequency range in the L-band and a spatial resolution of 10 by 10 m, variation of incidence angle and polarization state led consequently to knowledge about the variable information content in different radar configurations.
6 Franziska Kersten Measurements of the crosspolarized signal at low incidence angles turned out to be optimal for the purpose in question. As was shown by Dierking and Haack (1998) 3-layer RGB images combined from backscattered intensities at HH-,VV- and HV-polarization are closely correlated with geologic maps. Because of the dominantly non-deterministic scattering a statistical approach for target description is necessary.the polarimetric parameters can then be calculated from the polarimetric covariance matrix. Plotted in images each polarimetric parameter is of different value to the task. The in general significantly rougher lava flows can be distinguished from the surrounding terrain by their depolarization ratio, a parameter derived from the backscattering coefficients σ at different polarization combinations. The backscattering coefficients are high at crosspolarization due to a significant contribution of multiple scattering. Containing even more information about the roughness the copolarization ratio allows identification of different lava units such as pahoehoe and a a lava within one flow. Fig. 3 average profile of backscattering coefficients across lava flows in the Northern Volcanic Zone, Iceland, modified after Dierking and Haack (1998) Fig.3 shows the variation of backscatter coefficients at HH- and HVpolarization over an average profile of the investigation area. In order to emphasize the different surface types and their typical σ values three examples are depicted coloured. Rough surfaces coincide with high σ values as can be seen in the case of the very rough a a lava (red). The backscattered intensity from the comparable smooth pahoehoe lava (blue) is significantly lower. The lowest backscattering coefficients belong to the surrounding terrain, which has been weathered intensely and therefore has the smoothest surface.
Radar Polarimetry-Potential for Geosciences 7 Although the correlation coefficient between HH- and VV-channels provides useful information concerning the contribution of different scattering mechanisms to the received signal, the analysis is widely based on the H/A/αdecomposition after Pottier and Cloude (1997). Where H stands for the scattering entropy, A for the scattering anisotropy and α is the mean scattering angle, which is significant for the main scattering process (Hajnsek et al.2003). Fig. 4 segmentation of the entropy/alpha space according to the H/A/α decomposition after Pottier and Cloude (1997), modified after Pottier, Cloude (2003) Fig. 5 three selected lava fields of the Northern Volcanic Zone, Iceland plotted in an entropy/alpha diagramm using the H/A/α decomposition method (Pottier, Cloude 1997), modified after Dierking, Haack (1998)
8 Franziska Kersten The entropy based classification is used to discriminate different scattering mechanisms as each field in the H/α space belongs to a certain target type or scattering contribution (Fig.4). Fig.5 shows that almost all lava fields plot in the field of medium entropy belonging to surface scattering behaviour. soil moisture estimation according to Sokol et al.(2004) The inversion of soil moisture from polarimetric data is of special interest to hydrological applications such as resource management, stream flow forecasting and hydrological modeling. In the presented case study based on the work of Sokol, McNairn and Pultz fully polarimetric C-band data aquired by Sir-C/X-SAR sensor during two measuring campaigns in April and October 1994 over the investigation area in Manitoba, Canada are used for soil moisture estimation. Additional sampling in the field has been carried out. For bare soils volumetric surface soil moisture, surface roughness and C-band SAR backscatter show a linear relationship. In presence of a significant vegetation cover the inversion of surface parameters is more complicated, since the short wavelenghts used in C-band are hardly able to penetrate into deeper layers The degree of interaction between biomass and radar wave depends on incidence angle and polarization state of the transmitted wave. Steep angles (greater 30 ) lead to higher penetration capability to the surface. Vertical polarized waves tend to interact more with the vertical structure of the vegetation, while horizontal polarized waves interact with the underlying soil surface. For comparison of surface sample and radar data field average statistics and copolarization signature plots are used. The correlation of mean backscatter for the horizontal copolarization state and field average values for soil moisture is depicted in Fig.6. With an r-value of 0,857 the correlation is statistically significant for p<0,05. The correlation between crosspolarized backscatter and soil moisture is weaker. Single polarization measurements at steep incidence angles are sufficient for bare soil surfaces, since use of co- and crosspolarization ratios lead to no significant improvement of the regression model used to predict the soil moisture. Data aquired at multiple polarizations and therefore containing more information are useful in case of shallower incidence angles for which roughness contribution increases. The effects of surface roughness and vegetation can be minimized by using the polarization ratios. By combining co- and cross-polarization ratios the equation becomes independent of surface roughness, the soil moisture can then be determined.
Radar Polarimetry-Potential for Geosciences 9 Fig. 6 correlation between HH backscatter and surface volumetric soil moisture, modified after Sokol et.al. (2003) Conclusion The additional information obtained by polarimetric SAR data is a major contribution to Earth Sciences. The above presented case studies are only two examples chosen from an ever increasing number of projects and studies based on Radar Polarimetry. A lot of research has already been done on the relationships between surface parameters and the polarization state of electromagnetic waves used in SAR systems. New possibilities of acquiring information from the surface using Remote Sensing have been established in recent years thanks to Radar Polarimetry. This is of special value in regions difficult to reach and areas with unsuitable weather conditions for optical sensors. As has been shown there is no universal valid procedure in interpreting polarimetric data. In the past various models have been proposed in order to describe targets and their dominant scattering behaviour and extract physical parameters from measured data. A lot of work still needs to be done in order to improve inversion algorithms. As mentioned by Pottier et al. (2003) a major objective of current research is to extend the qualitative analysis to a quantitative one concerning the extraction of phyical parameters. This will result in even wider application possibilities for radar polarimetry, thus making it a promising technique for the future.
10 Franziska Kersten References W.Dierking, H.Haack (1998) L-Band Polarimetric SAR-Signatures of Lava Flows in the Northern Volcanic Zone, Iceland. IEEE Trans. Geosci. Remote Sens. 3:1339-1341 L.Ferro-Famil, E.Pottier, J.S.Lee Unsupervised classification of natural scenes from polarimetric interferometric SAR data In: C.H.Chen (editor) Frontiers of remote sensing. World Scientific 2003 I.Hajnsek, E.Pottier, S.R.Cloude (2003) Inversion of Surface Parameters From Polarimetric SAR. IEEE Trans. Geosci. Remote Sens. vol.41, no.4: 727-744 G.Huadong, L.Jingjuan, W.Changlin, W.Chao, T.G.Farr, D.L.Evans (1997) Use of Multifrequency, Multipolarization Shuttle Imaging Radar for Volcano Mapping in the Kunlun Mountains of Western China. Remote Sensing of Environment 364-374 J.S.Lee, D.L.Schuler, T.L.Ainsworth A review of polarization angle estimation from polarimetric SAR data In: C.H.Chen (editor) Frontiers of remote sensing. World Scientific 2003 F.Mattia, T.Le Toan, J-C.Souyris, G. De Carolis, N.Floury, F.Posa, G. Pasquariello (1997) The Effect of Surface Roughness on Multifrequency Polarimetric SAR Data. IEEE Trans. Geosci. Remote Sens. vol.35, no.4 : 954-966 C.Oliver, S.Quegan (1998) Understanding Synthetic Aperture Radar Images. Artech House remote sensing library J.Sokol, H.McNairn, T.J.Pultz (2004) Case studies demonstrating the hydrological applications of C-band multipolarized and polarimetric SAR. Can. J. Remote Sensing vol. 30, no. 3: 470-483 M.Zribi, N.Baghdadi, N.Holah, O.Fafin, C. Guérin (2005) Evaluation of a rough soil surface description with ASAR-Envisat radar data. Remote Sensing of Environment 96: 67-76 M.Zribi, N.Baghdadi, N.Holah, O.Fafin (2005) New methodology for soil surface moisture estimation and its application to ENVISAT-ASAR multi-incidence data inversion. Remote Sensing of Environment 96: 485-496