This thesis sets out the results of work done in preparation for a South African Airborne Synthetic Aperture Radar (SAR) measuring campaign envisaged for 1994/5. At present both airborne and spaceborne SARs have found a niche in remote sensing with applications in subsurface mapping, surface moisture mapping, vegetation mapping, rock type discrimination and Digital Elevation Modelling. Since these applications have considerable scientific and economic benefits, the Radar Remote Sensing Group at the University of Cape Town committed themselves to an airborne SAR campaign. The prime objective of the campaign is to provide the South African users with airborne SAR data and enable the RRRSG to evaluate the usefulness of SAR as a remote sensing tool in South Africa.
In this thesis the details of how SAR images are formed are not important. Instead, the emphasis is on how the electromagnetic wave interacts with the surface because this determines how and what a SAR can measure. To this end, theoretical models, namely, physical optics, small perturbation and geometric optics models were investigated since these models enable the backscatter from a randomly rough surface to be determined. The Bragg resonance phenomena, which accounts for the relatively high backscatter from periodic surfaces was also investigated since this model is important in many sea surface imaging applications. Because the theoretical models are difficult to use, empirical and self-empirical models are popular. In subsurface applications, the penetration depth is determined by the microwave attenuation in the sand medium. The factors affecting the microwave attenuation in sand are the frequency of the radar signal, and the real and imaginary part of the permittivity. The lower the frequency, and the lower the loss tangent (defined as the ratio of imaginary part of the permittivity to the real part of the permittivity), the lower the microwave attenuation in the sand medium.
Both attenuation and scattering models are functions of the complex dielectric constant of the medium. This necessitated the need for an investigation into the factors influencing the complex permittivity of soil. These factors, listed in order of greatest influence are soil moisture, soil water salinity, soil type, soil density and soil temperature. The relative permittivity of dry soil ranges from 3 to 4, independent of frequency. As water is added, the relative dielectric permittivity increases rapidly as pure water has a value of approximately 80 for frequencies below 1 GHz. It is therefore obvious that the dielectric constant of a soil is greatly influenced by the volumetric water content. The water content of soil generally consists of *free* and *bound* water, where the amount of *bound* water is determined by the surface area of the soil particles. The smaller the individual soil particles the greater the total surface area. Thus, the amount of *bound* water in the soil is dependent on the soil textural composition. The real part of the complex dielectric constant of soil water is unaffected by salt content, whereas the imaginary part of the complex dielectric constant is affected, especially at low frequencies where ionic conductivity dominates. The remaining factors have negligible influence on the permittivity. Theoretical and empirical models for soil permittivity are available but the latter is more popular because the theoretical models are complicated and require too many input variables. The empirical models are not without problems. The real part of the permittivity can be successfully modelled from about 1 MHz to 18 GHz, whereas the imaginary part of the permittivity can only be successfully modelled from about 1 GHz to 18 GHz because the effects of ionic conductivity, which cannot easily be nicluded in the model, dominate at frequencies lower than 1 GHz.
Polarimetric imagery is a relatively new development in radar imagery and is at present an active research area. Polarimetric radars have the capability of identifying a scatterer from its polarization signature, which is obtained by a technique known as polarization synthesis. For example, a dihedral corner reflector signature can be used to identify buildings in an image, or a large smooth dielectric surface signature indicates the locations of clear-cut areas in an image of forest vegetation.
Next, the intended applications, namely, subsurface geological mapping, surface rock type discrimination, near-surface soil moisture content mapping, vegetation mapping, and Digital Elevation Modelling are discussed. Since the prime objective of this thesis is the planning of an airborne SAR campaign, the most important part of each application discussed is the optimum parameters for that application.
There are numerous geological applications of SAR which generally fall into one of two groups, namely surface and subsurface applications. The latter, which is a unique feature of radar imaging will be discussed first. Low loss material, such as dry sand, which covers most of the surface of arid regions can easily be penetrated with low frequency radar. Thus, the images formed reveal the subsurface geology. The optimum parameters for subsurface imaging are chosen to minimize the microwave attenuation or maximize the microwave penetration depth in the medium. For a given frequency ad soil moisture content, sand consistently has the lowest attenuation. However, the greatest influence on microwave attenuation is the moisture content of the soil. Greater microwave penetration is observed for soils with lower moisture content. In addition, lower radar frequencies yield greater microwave penetration of the obscuring medium. The optimum incidence angle for subsurface imaging is a compromise between minimizing the path length in the sand medium and avoid incidence angles at which specular reflections at the air-sand interface dominates the radar backscatter. Thus the optimum incidence angle for subsurface applications is between 10° and 20°.
SAR has been used very successfully in rock type discrimination since it is uniquely sensitive to the roughness or texture of the surface being imaged. The growing trend to use multifrequency and multipolarization radars has enabled considerably more information about the surface to be extracted from the radar image. The most powerful means of distinguishing between surfaces with different degrees of roughness is to exploit the wavelength dependence of the Rayleigh criterion. In this way roughnesses can be used to aid in the mapping of different lithologies and superficial deposits in terrain that is bare of vegetation. Also, the use of multipolarization has enabled rock types to be mapped with greater accuracy since each rock type can be matched to a particular polarization signature which is obtainable via a technique known as polarization synthesis.
SAR has the ability to measure the near-surface soil moisture content remotely. Radar backscatter is influenced by physical parameters (surface roughness and surface permittivity) and radar parameters (incidence angle, frequency and polarization). The soil moisture is inferred from the surface permittivity, which is in turn deduced from the measured radar backscatter. To determine soil moisture content from radar backscatter successfully requires the effects of vegetation, surface roughness and local incidence angle on the observed backscatter to be minimized or removed completely. This is accomplished by choosing an optimum frequency, incidence angle and polarization for imagery from which the soil moisture content will be extracted. Several researches have shown that the optimum parameters for soil moisture measurement is an incidence angle of 10°, a frequency of 4 GHz and HH or VV polarization.
The feasibility of vegetation type discrimination, particularly for crop type distribution mapping was also investigated. To distinguish between two crops in a radar image requires that the backscatter from these crops differ in intensity. This backscatter intensity is influenced by vegetation parameters such as the canopy density and crop height. Unfortunately, the measured backscatter is also greatly influenced by the soil moisture and roughness. Hence, successful vegetation mapping depends on the degree to which the effect of the soil moisture can be minimized. Researchers have shown that the soil moisture influence is minimized by using incidence angles of approximately 40° and frequencies in the vicinity of 8 GHz.
Two sea surface imaging applications were investigated namely, oil pollution and fish shoal monitoring. Detecting, monitoring and mapping of oil spills on the sea surface are applications based on the Bragg scattering phenomenon which accounts for the observed sensitivity of imaging radars to the amplitude of ocean capillary waves. The presence of oil on the ocean surface significantly reduces the amplitude of capillary waves which in turn significantly reduces the radar backscatter from that surface. Thus, oil on the sea surface corresponds to dark patches. Fish shoal monitoring is also based on the Bragg scattering phenomenon. Researchers have shown that certain fish such as bluefin tuna break the water surface by repeatedly jumping out of the water. This behaviour, which is associated with feeding, produces a rough surface which results in greater backscatter and thus appears as a bright feature in the radar image. Similarly, net floats cause a roughening of the surface which enables the location of the nets to be mapped. The optimum parameters for sea surface imaging applications are difficult to determine because of insufficient data on the subject. Nevertheless, an estimate of the optimum frequency and incidence angles is gleaned from the limited literature on the topic. Since Bragg scattering is the theoretiacl basis for both air pollution and shoal monitoring, these applications are probably optimum at high frequencies (X-band), large incidence angles (40°) and VV polarization.
Digital Elevation Models (DEM) can be generated by extracting topographic information from SAR data using Shape-from-shading, Stereoscopic imaging and Interferometric SAR techniques. In Shape-from-shading, the radar backscatter is assumed to be proportional to the local incidence angle which is a function of the terrain slope and incidence angle. Since the incidence angle is fixed by the sensor hardware, the slope of the terrain can be inferred from the backscatter. The DEM is then constructed from the terrain slope information. The accuracy of the DEM is extremely sensitive to the accuracy with which the backscatter intensity versus local incidence angle can be modelled. The accuracy of the model is in turn dependent on the roughness of the terrain because rougher surfaces yield better models. Since roughness is frequency dependent, higher frequencies yield better models. An exact frequency can only be determined if the surface roughness is known but C-band frequencies will probably be high enough. The optimum incidence angle is chosen to minimize the likelihood of layover and shadowing. Without a priori knowledge of the topography, the optimum incidence angle is 45°.
Stereoscopy is another technique whereby topographic information can be extracted from two overlapping images. The method is based on the apparent movement or parallax of features in the stereo image pair. The elevation of a feature is proportional to the observed parallax of the feature. The method requires two images of similar image quality, tone and texture, but different imaging geometry to present parallaxes for the height perception. The optimum stereo geometry is a compromise between two equally important criteria. Firstly, the two images forming the stereo pair must be very similar in image quality, terrain illumination, tone and texture so that the stereo pair correlates well. Secondly, the two images forming the stereo pair must be sufficiently different in geometry to present parallaxes for height perception. Since radar actively illuminates the terrain, significant differences in viewing geometry also imply illuminations differences. In addition, the optimum incidence angles and intersection angles are also dependent on the terrain relief. For relatively flat terrain, the incidence angles should range from about 20° to 60° with an intersection angle greater than about 30°. For relatively high relief terrain, the incidence angles should range from 40° to 70° with intersection angles ranging from 15° to 30°.
Due to the coherent nature of SAR, interferometric principles can be used to extract height information from raw SAR data. If two receiving antenna are placed in the range-height plane, then the phase difference of the echoed power can be determined. This means that differences in height, modulo the radar wavelength, can be detected. In this way three dimensional terrain maps can be constructed. The accuracy of the DEM is maximized through minimization of phase errors, wavelength, and slant range and maximization of the baseline distance. However, the level of phase noise in the system increases with increasing baseline distance up to a critical baseline distance where the signals are no longer correlated. It can be shown that the optimum baseline distance ranges from 0.2 to 0.8 of the critical baseline distance. These baseline distances are limited by aircraft dimensions when both antennas are mounted on the same aircraft which implies that the frequency can be chosen so that the baseline distance is optimum.
Ideally, one imaging radar could be configured to have frequency and incidence angle ranges that would contain the optimum parameters for all the envisaged applications. Unfortunately, a combination of financial and technical limitations made this ideal imaging radar unrealistic. Hence, some compromises had to be made which automatically implied implementing some applications with less than optimum imaging parameters.
The details of the proposed South African airborne SAR campaign are contained in the appendix.