Radio Propagation and Remote Sensing of the Environment - Chapter 15 (end) pot

© 2005 by CRC Press 405 15 Researching Land Cover by Radio Methods 15.1 GENERAL STATUS Land covers are various — open soil, open water, vegetation, forests — and are characterized by both geometrical structure and electro-physical properties that produce a variety of electromagnetic wave interactions with natural objects, thus providing the background for research by microwave remote sensing. Generally, the infrastructures of land objects are comparatively small in geometric size, perhaps a few tens of meters (farming fields, for example); therefore, one of the main problems here, especially in the case of observation from space, is the problem of spatial resolution, which we have mentioned repeatedly. In the case of airborne instruments, the problem of space resolution is not as important for such observations. For observation from space platforms, the synthetic-aperture radar (SAR) systems are more effective for obtaining images with high spatial resolution. Other microwave instruments (for example, radiometers) are widely used, as a rule, for the sounding of statistically homogeneous areas, such as forest tracts, steppes and deserts, and some tundra regions. The situation can be improved by carrying out joint processing of data provided by instruments with different degrees of resolution. Some study could verify the effectiveness of several processing procedures, but this technology has still not found wide application. We will use the following classification of the land covers when we discuss radio methods for remote sensing: • Bare terrain and geological structures • Hydrology structures •Vegetation canopy • Internal basins • Snow cover and ice 15.2 ACTIVE RADIO METHODS Active radio methods are synonymous with radar technology, which is widely employed now due to the development of spaceborne radar techniques. SAR systems have become instrumental for systematic monitoring of the surface of Earth. Radar images reflect many peculiarities of the researched area, such as landscape elements, hydrology network, vegetable canopy, and artificial construction. Gathering this complicated information allows us to assemble special thematic maps: topographic, geologic, hydrologic, forestry, etc. One of the main requirements for radar images is an accurate tie-in to the terrain which is based partly on navigation data and partly TF1710_book.fm Page 405 Thursday, September 30, 2004 1:43 PM © 2005 by CRC Press 406 Radio Propagation and Remote Sensing of the Environment on position data of the fixed points. Certainly, knowledge of the appropriate antenna orientation is included on a list of information required for correct radar data interpretation. In fact, the radar image is a map of the backscattering coefficient, which depends not only on surface scattering of the radiowaves by soil but also on volumetric scattering by elements of a vegetable canopy. To begin, we will address scattering by bare soil and primarily consider the inclined incidence of radiowaves typical for SAR and scatterometer systems. In this case, the separation of large- and small- scale roughness is not as clear as for the ocean; therefore, it is difficult to distinguish between specular and resonant scattering. This is one reason why empirical or semi- empirical models have found wide application for the interpretation of experimental data. Theoretical models and experimental data show the distinct dependence of soil scattering intensity on surface roughness parameters and on soil permittivity, which, in turn, is strongly dependent on soil moisture. The simplest model of soil permit- tivity can be described by the so-called refractive formula : 43,116 . (15.1) Here, ε w is the water permittivity, described by Equations (14.1) and (14.2); ε g is the dry ground dielectric constant; and ξ is the water volumetric content (i.e., the volume part occupied by water in mixture). Numerically, this value coincides with volumetric soil moisture m v (g/cm 3 ). The absence of a numerical difference indicates that we should not separate these terms. We must be careful when comparing our moisture definition to moisture determined by the gravimetric method with oven drying. The full water content determined by the gravimetric method includes both free water and bound moisture, while electromagnetic waves react only to free moisture. The quantity of bound moisture depends on soil type; it is about 2 to 3% in sandy soils and can reach values of 30 to 40% of dry soil mass in clay and loess grounds. The soil permittivity depends weakly on the soil moisture at small concen- tration. Equation (15.1) has no theoretical background and is a suitable approxima- tion only in the microwave region; 116 other approximations can be found in Ulaby et al. 90 The permittivity of dry soil depends only on its density in the first approach. This dependence can be expressed in the form: 118 , (15.2) where ρ g is the ground density (g/cm 3 ). For our discussion here, we will use the following values for the various numer- ical calculations: ρ g = 2 g/cm 3 ; t = 20°C; S = 2‰; ε g = 4, ε w ≅ 80, and σ ≅ 2.4 · 10 9 CGSE. For many practical applications, however, ρ = 1.5 g/cm 3 is more realistic. These values indicate that the behavior of various types of ground is our primary practical independence of soil permittivity at the C- and L-bands is apparent. This is understandable, as the water permittivity real part is practically constant at these εξε ξε=+− () wg 1 ερ gg =+105. TF1710_book.fm Page 406 Thursday, September 30, 2004 1:43 PM focus. The soil permittivity dependence on moisture is plotted in Figure 15.1. The © 2005 by CRC Press Researching Land Cover by Radio Methods 407 frequencies and the imaginary part is small. The frequency dependence occurs closer to millimeter-wavelength bands which is reflected by the 37.5-GHz curve. More detailed analysis can be found, for example, in Ulaby et al. 90 and Shutko. 116 The predominance of specular or diffuse scattering mechanisms is primarily determined by the radio-wave frequency. The analysis of experimental data provided by Shi et al. 120 gives the values cm for roughness amplitude and l = 20 to 30 cm for correlation length. The data of Dierking 121 provided values of 1 to 7 cm for field roughness amplitude and 2 to 37 cm for correlation length. Profilometer measurements allow us to conclude that the exponential autocorrelation function is the best approximation of experimental data. Shi et al. 120 tested the autocorrelation function in the form , where the most probable value of index ν is again unity. More exactly, about 76% of the measured profiles of roughness could be described by the correlation function with ν 1.4. The index difference produces a difference in the spatial spectrum which is proportional to the function: . (15.3) on the index ν value; therefore, it would be enough to be confined by the case ν = 1. This means that we are dealing with fractal surfaces of Brownian type, and the spectrum given by Equation (14.40) is suitable for our purposes. The data reported by Dierking 121 suggest that many natural surfaces have stationary random processes with a power-law spatial spectrum of the form , where 3 α 3.7, which indicates the fractal character of surfaces FIGURE 15.1 Average values of soil permittivity dependence on moisture: (1) 1.3 GHz; (2) 5.3 GHz; (3) 37.5 GHz. 0 0 6 12 18 24 0.1 0.2 0.3 0.4 ξ ε soil permitivity volumetric soil moisture 1 2 3 ζ 2 23= to ρ() expxxl= − () ρ ν () expxxl= − ()       Hq eJqs s dsq kl s (,) , sinν ν θ νν = () = −− ∞ ∫ 1 2 0 121 0 i  Kq q ζ α ()≈ 1 TF1710_book.fm Page 407 Thursday, September 30, 2004 1:43 PM The curves of Figure 15.2 demonstrate the weak dependence of the spatial spectrum © 2005 by CRC Press 408 Radio Propagation and Remote Sensing of the Environment bounded by natural media. The available analytical approximation of these spectra with conservation of roughness magnitude and correlation length can be written as: , (15.4) Brownian type of spectrum is from this family. Now, we are ready to analyze the processes of scattering by the terrain. First, we will consider the P-band waves scattered by bare soil. The parameters specified above allow us to employ the perturbation method approximation; that is, we will angular dependence of backscattering coefficients for horizontally and vertically polarized waves of the P-band. It was assumed for our computations that ξ = 0.2, , and l = 20 cm. Calculation of the backscattering coefficient for higher frequencies cannot be based on an approximation of the perturbation method, because, for example, product kl > 1 for the C-band. Unfortunately, no theoretical models have been developed that produce good numerical agreement of computed and experimental data. All models, regardless of soil type, describe the wave scattering processes in forms that allow us only to explain qualitatively some aspects of the phenomenon, primarily because the radio wavelength does not essentially differ from the correlation length of soil surfaces. For this reason, we cannot apply any modern asymptotic approaches of scattering theory. One of the best alternatives is to use semi-empirical models that approximate the experimental data. One of these models is the Oh, Sarabandi, and Ulaby (OSU) model, 122 which provides an expression for the cross-polarized ratio : FIGURE 15.2 Graphics demonstrate the weak dependence of spatial spectrum on the index value: ——, v = 1; ······, v = 1.2; – – – –, v = 1.4. 0 1 2 3 4 q wave parameter 0 0.2 0.4 0.6 0.8 H spectral function  Kq l ql ζ ν νζ αν() ,= − () + () = 1 1 2 22 22 ζ 2 2= cm q = σσ hv 0 vv 0 TF1710_book.fm Page 408 Thursday, September 30, 2004 1:43 PM base all of our computations on Equations (6.48) and (6.49). Figure 15.3 shows the which is similar to the spectra used to describe turbulence (see Chapter 7). The © 2005 by CRC Press Researching Land Cover by Radio Methods 409 (15.5) We must use a two-term subscript now in order to emphasize the fact we are dealing with matched and cross-polarized components of the scattered signal. An equation that applies to the copolarized ratio p = is: , (15.6) where the incident angle θ i is expressed in radians. The backscattering coefficient of vertical polarization is approximated by the formula: . (15.7) incident angle at the L- and C-bands. The volumetric moisture value is chosen to be equal to 0.2. The “m” added to the subscript reflects the fact that this model was developed at the University of Michigan. We can see that the backscattering coef- ficients of horizontal and vertical polarizations differ slightly at the C-band. This weak difference takes place at the chosen parameter of roughness . Obviously, this difference will be bigger for a smoother surface. This small difference is emphasized by the Kirchhoff (or geometrical optics) approximation FIGURE 15.3 Angular dependence of backscattering: (1) σ vv ; (2) σ hh . 20 40 θ −40 −36 −32 −28 −24 σ 2 1 incident angle, degrees backscattering coefficient, dB qF k== −−           σ σ ζ hv 0 vv 0 023 0 1 2 .() exp . σσ hh 0 vv 0 pk F ==−       −    σ σ θ π ζ hh vv i 0 0 033 0 2 1 2 2 .() exp(        σ θ θθ vv 0 i hi vi = () + ()       = − g p FF g cos ,.ex 3 2 07 1 pp. . −                       065 2 18 k ζ ζ 2 2= cm TF1710_book.fm Page 409 Thursday, September 30, 2004 1:43 PM The plots of Figure 15.4 show the dependence of backscattering coefficients on the (see Chapter 6), which reflects its qualitative correctness. Figure 15.5 shows the © 2005 by CRC Press 410 Radio Propagation and Remote Sensing of the Environment dependence of values of the cross-polarization coefficient on the incident angles for the L- and C-bands. The fact that backscattering coefficients are determined relative to only one parameter is an advantage of the OSU model. Recall that, in geometrical optics approximations, the backscattering coefficient depends on only one parameter: the slope. Large elements of the terrain caused by changes in slope and variations in the roughness parameters are distinguished on radar images by varying brightness. Radar images provide a good representation of the peculiarities of a landscape. This is one of the reasons why radar mapping has found application in geology. The specific method of observation at normal viewing angles from the Earth’s surface allows us to detect faintly marked relief elements of slightly rugged terrain such as hills, valleys, etc. Radar maps often have more contrast compared to aerial photographs due to their employment of polarization methods. It is important to note that the use of radar mapping overcomes the screening effect of vegetation to reveal various features of geological structures, including lineaments and circular structures. FIGURE 15.4 Dependence m of backscattering coefficients at the L-band (——) and C-band (– – –) on the incident angle: (1) and (3) σ vv ; (2) and (4) σ hh . FIGURE 15.5 Cross-polarization coefficient dependence on the incident angle for the (1) L- band and (2) C-band. 10 20 30 40 50 θ −30 −25 −20 −15 −10 −5 σ incident angle, degrees backscattering coefficients, dB 1 2 3 4 1 2 10 20 30 40 50 θ incident angle, degrees −40 −35 −30 −25 −20 −15 −10 −5 σ hv cross polariz. backscattering coef., dB TF1710_book.fm Page 410 Thursday, September 30, 2004 1:43 PM © 2005 by CRC Press Researching Land Cover by Radio Methods 411 Interferometry technology is very effective for mapping landscape details. A brief explana- tion of this technology is as follows. Imagine that similar radar scenes are obtained by subsequent flights over a neighboring orbit separated by dis- tance d , as shown in Figure 15.6. Assume that the satellite orbits lie at planes parallel to the x -axis and that base d is oriented along the y -axis. Then, let us also assume summation of the signals reflected by any pixel of the surface which is possible due to the high coherency of the radar system itself. The intensity of the summarized signals will depend on their phase difference, which occurs because of the different radar positions. Each reflected signal is noise- like, but, in the case of small distances between the two satellite passes, a high level of coherency between the signals reflected by the same pixel is maintained, and the phase difference has a definite value. More correctly, the discussed phase difference is stochastic but its mean value is not zero. 9 The latter depends on the pixel position and the base size. If the investigated surface is flat, on average, then the lines of constant phase difference will be straight along the flight direction. The values of the x -coordinate are assumed to be much less than platform height H and horizontal distance y from the flight trace and observation point (point O in Figure 15.6). When observing some hill elements above a flat terrain, the equiphase will differ from a straight line and its curvature will depend on the hill topography. It is possible to show (neglecting small values) that equiphase lines are described by the equation: . (15.8) Here, h is the hill height, and f ( x,y ) is the function describing the hill shape. The case h = 0 corresponds to a flat surface. Having subtracted the first term from the experimental data, we can obtain the equiphase lines related directly to the hill topography. An example of equiphase counters is given by Figure 15.6. The two-pass positions of the radar antenna can be compared to the two-antenna inter- ferometer system, and we can talk about a synthetic interferometer. The real one was realized during the shuttle radar topography mission (SRTM) when the second antennae of the C- and X-band radars were situated at the end of a 60-m boom. This mission provided interferograms within Earth latitudes ±60°. The processing of interferogram data permits retrieval of a terrain topography with high accuracy. This accuracy is due to the high interferometry sensitivity, particularly when the interferometer lobe angular width is much greater than the angular size of the investigated pixel; that is, , where θ is the incident angle. This inequality can be rearranged into , where D is the synthesis length of the SAR. ∆ Λ ϕ kd y yH H d hf x y yH == + − () + 22 22 2 , λθ ρ Ldcos >> dDcosθ << 2 FIGURE 15.6 Interferometry technology for mapping land- scape details. z d y Hill o x TF1710_book.fm Page 411 Thursday, September 30, 2004 1:43 PM © 2005 by CRC Press 412 Radio Propagation and Remote Sensing of the Environment Now, we will turn our attention to the problem of surface parameter measurement by means of radar systems. As backscattering coefficients depend only on surface roughness and on the permittivity of the sounded medium, our discussion will focus on measurement of the roughness parameters and the permittivity value. Defining surface roughness parameters is a very important problem for many reasons. For example, pedology is an area where information about the roughness properties is important for our understanding of many processes, such as flooding, infiltration, erosion, etc. Surface roughness is connected with the properties of some materials and this information is of value in geology. The bare soil backscattering coefficient is determined by both roughness and moisture. One problem with the radar data processing procedure is separation of these effects, but this can be accomplished by polarization measurements. One way to select the roughness effect is to define the correlation coefficient between the signals of orthogonal polarizations. 120 If U p is the complex amplitude of the received signal of matched p polarization, then the unknown correlation coefficient is defined as: , (15.9) where U q is the signal of the other orthogonal matched polarization. Shi et al. 120 used ρ pq for their calculations, although the value 1/2Re ρ pq is more logical for such applications. Their reason for using ρ pq was that the magnitude of the copolarized correlation coefficient is weakly affected by calibration errors. Analysis of experi- mental data and theory 120 suggest that the coefficient of correlation between right and left circularly polarized signals depends on the roughness amplitude and weakly reacts to changes in soil moisture. This assumption allows us to consider the corre- lation coefficient mentioned as being representative of bare soil roughness. Another parameter frequently under discussion is soil moisture. It is a very important value that plays an essential role in the various phenomena of hydrology, meteorology, climatology, agronomy, etc. Such areas as meteorology and climatol- ogy require moisture data on a large spatial scale (even global). Moreover, the soil moisture content is a changeable parameter that must be monitored periodically. These data cannot be obtained by onsite measurements; therefore, remote sensing technology becomes particularly significant. The multiplicity of different terrain areas in the landscape requires relatively high spatial resolution for the sounding tools. Among microwave devices, only SAR, as we noted earlier, satisfies this condition, especially in the case of observation from space. This explains the great interest in soil moisture estimation on the basis of SAR data. The complexity of the radar signal returning from a rough surface makes this a difficult estimation problem. , (15.10) ρ pq pq pq UU UU = ∗ 22 ηξθ σ ξ h,v hh vv (,,)f d d , = 0 TF1710_book.fm Page 412 Thursday, September 30, 2004 1:43 PM Figure 15.7 demonstrates the soil moisture sensitivity: © 2005 by CRC Press Researching Land Cover by Radio Methods 413 at the L-band for both polarizations as a function of the incident angle. The men- tioned sensitivity is expressed in decibels and corresponds to the point ξ = 0. This sensitivity is sufficient, especially in the case of vertical polarization. At vertical polarization, a weak maximum of the sensitivity occurs at an incident angle of 40°. Such dependence is the basis for development of soil moisture measurement by radar technology; however, it cannot be the basis for an algorithm to solve the inverse problem (i.e., soil moisture retrieval) because the backscattering coefficient depends on both soil electrophysical properties and the roughness parameters. It is difficult to state the cause of changes in the backscattering coefficient, as they can be the result of a change in roughness or variations in moisture. It is necessary to know in advance the terrain roughness characteristics in order to evaluate the radar data against the soil moisture value. It is impossible to have such preliminary information on a large scale, so such methods can hardly be considered successful. An investigation conducted at test areas to determine rough- ness parameters is more likely to help determine the correctness of various scattering models than develop a retrieval algorithm. It is important, then, to have an algorithm of radar data processing that does not take into account the roughness parameters. This is a reason why algorithms based on polarimetric analysis of radar data are more effective. It is easy to see that, within the framework of the perturbation method, the ratio of the backscattering coefficient does not depend on the roughness spectrum (see Equations (6.48) and (6.49)); that is, the ratio depends only on soil permittivity and angle of incidence: (15.11) This means that P-band radar, for which the perturbation method can be valid, can of volumetric soil moisture at the P-band for incident angles of 30° and 50°. FIGURE 15.7 Angular soil moisture sensitivity: (1) η v ; (2) η h . 20 40 θ η 40 20 0 1 2 incident angle, degrees soil moisture sensitivity γξθ σ σ θfC,, () () ==+ vv 0 hh 0 2 12 TF1710_book.fm Page 413 Thursday, September 30, 2004 1:43 PM be used for this kind of measurement. This ratio is shown in Figure 15.8 as a function © 2005 by CRC Press 414 Radio Propagation and Remote Sensing of the Environment Obviously, the ratio under discussion is more sensitive to moisture change at large incident angles. Another way to estimate soil moisture is to determine the cross-polarization ratio (see Equation (15.5)). This ratio vs. volumetric soil moisture is represented in Figure 15.9 for the C- and L-bands. It does not depend on the incident angle in the approximation given by the OSU model. At the C-band, this ratio is more sensitive to moisture content change compared to the L-band. However, this advantage is the seeming one, in the general case, taking into account the scattering and screen effects of vegetation. Before investigating this problem further, we should note that the procedures of polarization data processing presented here reflect only basic approaches to the problem solution; other procedures can be found in the litera- ture. 90,120,124 For our discussion of soil covered by vegetation, we will first consider grassland. The backscattered signal, in this case, consists of at least five components. The first of them is directly scattered by the soil roughness component (the ground-bounce term) and is attenuated by extinction due to vegetation elements (absorption and spatial scattering). This component can be represented in a very simplified form: FIGURE 15.8 Soil moisture sensitivity at the L-band for both polarizations as a function of incident angle: (1) 30°; (2) 50°. FIGURE 15.9 Volumetric soil moisture ratio for the (1) L-band and (2) C-band. 2 1 0 0.1 0.2 0.3 0.4 ξ volumetric soil moisture 1 1.5 2 2.5 γ VV/HH ratio cross-polarization ratio 0.16 0.12 0.08 0.04 0 0 0.1 0.2 0.3 volumetric moisture 0.4 2 1 q ξ TF1710_book.fm Page 414 Thursday, September 30, 2004 1:43 PM [...]... and from (–1.06 · 1 0-3 ) to (–4.54 · 10–3) ha/m3 for the L-band Finally, values of C3 range from 0.092 to 0.238 for the C-band and from (5.73 · 10–2) to (8.53 · 10–2) for the L-band The saturation levels of the steam value were 125 to 175 m3/ha for the C-band and 225 m3/ha for the L-band The universality of these parameters and their dependence on the incident angle is the subject of future investigation... Here, η is the degree to which vegetation covers the soil; ks, Ts, and Tos are the emissivity, the temperature, and the brightness temperature of the bare soil, respecˆ tively; Tv and A are the temperature and the albedo of the vegetation, respectively; and τ is the radiation attenuation in the vegetation layer The vegetation empty spaces are supposed to be much greater than the radiowavelength The experimental... Remote Sensing, 38(1), 349, 2000 128 Imhoff, M.L., A theoretical analysis of the effect of forest structure on synthetic aperture radar backscatter and the remote sensing of biomass, IEEE Trans Geosci Remote Sensing, 33(2), 341, 1995 129 Kurvonen, L., Pullianen, J., and Hallikainen, M., Retrieval of biomass in boreal forests from multitemporal ERS-1 and JERS-1 SAR images, IEEE Trans Geosci Remote Sensing, ... data 15. 3 PASSIVE RADIO METHODS Passive radio methods do not have as many applications for the investigation of land compared to remote sensing of the atmosphere and ocean The main reason, as we have said before, is the poor spatial resolution, which makes it difficult, for example, to employ space platforms and high-altitude aircraft The situation becomes more complicated by the fact that P- and L-bands... ERS-1 and JERS-1 images It was possible to compare the data obtained in different frequency bands, and this data processing has produced values of C1 that vary from (8.26 · 10–4) to (4.09 · 10–3) ha/m3 for the C-band and from (1.56 · 10–3) to (1.86 · 10–3) ha/m3 for the L-band, depending on the season Correspondingly, values of C2 range from (–2.51 · 10–3) to (–5.99 · 10–3) ha/m3 for the C-band and. .. by values of the order of 10 m at the C-band It means that lake ice does not fully absorb radiowaves longer than the C-band waves, and the thermal microwave radiation at these waves depends on the ice thickness © 2005 by CRC Press TF1710_book.fm Page 425 Thursday, September 30, 2004 1:43 PM Researching Land Cover by Radio Methods 425 Ice temperature changes with depth The temperature of the upper ice... Researching Land Cover by Radio Methods 0 −50 0.1 1 0.3 423 1 3 5 h, m 2 −100 150 ∆Tb, K FIGURE 15. 13 Correlation of increment of brightness temperature and stratification depth of groundwater: (1) clay loam; (2) sandy loam soil-vegetation system The emissions of the vegetation depend on the albedo of the scatterers Equation (9.89) represents a simplified description of the emission of a scattering layer of infinite... gives a value of 0.85; for the C-band, this value is about 0.65 This result verifies once more the advantage of the L-band for the measurements discussed here A more progressive way to measure biomass is to use a multifrequency, multipolarization system Mougin et al.130 reported that application of the AIRSAR system significantly improves biomass measurement The P-band HV and L-band HV have the largest... canopy changes the general picture of radiation The radio- wave attenuation in vegetation screens the soil radiation, sometimes partly and at other times fully The emissions of the vegetation are added to the emissions of the soil Some secondary effects (as described for calculation of the backscattering coefficient) contribute to the general picture of the microwave radiation of a © 2005 by CRC Press TF1710_book.fm... example of recognizing trees types has been provided by Nazarov.137,138 Their cluster analysis, based on the statistics of backscattering coefficients and their © 2005 by CRC Press TF1710_book.fm Page 420 Thursday, September 30, 2004 1:43 PM 420 Radio Propagation and Remote Sensing of the Environment standard deviations, has allowed identification of coniferous, deciduous, and mixed woods Another type of . Press 410 Radio Propagation and Remote Sensing of the Environment dependence of values of the cross-polarization coefficient on the incident angles for the L- and C-bands. The fact that. determination of the thick- depth is estimated by values of the order of 10 m at the C-band. It means that lake ice does not fully absorb radiowaves longer than the C-band waves, and the thermal microwave. covers the soil; k s , T s , and T os are the emissivity, the temperature, and the brightness temperature of the bare soil, respec- tively; T v and are the temperature and the albedo of the vegetation,