MEASUREMENT AND DISCRIMINATION

In microwave remote sensing the distinction between different subjects

sensed, usually areas on the ground, is primarily caused by the difference in the
signal strength received by the radar or radiometer. Hence, the measurement of
received signal strength is the most important measurement in remote – sensing
devices. This characteristic of remote – sensing microwave instruments is
different from the typical radar or radiometer used for other purposes, where a
measurement of position is often the most important characteristic of the system,
and measurement of the amplitude of the signal is unimportant.
Measurement of position represent discrimination for the microwave
remote sensor because the amplitudes of the signals received from different
areas must be discriminated from each other. The methods of distinguish among
different areas on the ground generally amount to measurements of angle,
distance, or speed. Speed measurement actually is used in microwave remote
sensing as a measurement of distance on the ground. Similarly, angle
measurement is used to discriminate a region on the ground in that the angle
from the axis of the sensor differs for different points on the ground.
The measurement of amplitude is highly dependent upon the statistics of
the received signals. Both for radars and radiometers, the usual signal has a
fading characteristic like that of noise. In the case of the radiometer this comes
about because the signal source itself is noiselike. In the case of an airborne
radar, it comes about because the Doppler frequency shift associated with each
point scatterer on the ground is different. This results in a signal made up of
many sine waves of different frequencies – a mode often used for noise. The
primary difference between the noiselike signals for the radiometer and radar is
that the bandwidth of the radiometer signal is very great (tens or hundreds of
MHz), whereas the Doppler bandwidth for the radar is much smaller and typically
does not exceed a few KHz.
THE RADAR EQUATION
The fundamental relation between the characteristics of the radar, the
target, and the received signal is called the radar equation. The geometry of
scattering from an isolated radar target (scatterer) is shown in Fig.1, along with
the parameters that are involved in the radar equation.

Fig.1 Geometry of and quantities involved in the radar equation.

Where a power Pt is transmitted by an antenna with gain Gt, the power per
unit solid angle in the direction of the scatterer is PtGt, where the value of Gt in
that direction is used. At the scatterer,

Power unit Power 1

area incident = transmitted Spreading Ss=(PtGt) ----------
on scatterer toward scatterer loss 4Rt2

(1.0)

Where Ss is the value of the Poynting vector (power density) at the

scatterer. The spreading loss 1 / 4Rt2 is the reduction in power density
associated with spreading of the power over a sphere of radius R surrounding
the antenna.
To obtain the total power intercepted by the scatterer, the power density
must be multiplied by the effective receiving area of the scatterer:
 Power incident effective
Intercepted = power / receiving Prs=(SsArst) (2.0)
by scatterer unit area area of
scatterer

Note that the effective area Ars is not the actual area of the incident beam
intercepted by the scatterer, but rather is the effective area; i.e., it is that area of
the incident beam from which all power would be removed if one assumed that
the power going through all the rest of the beam continued uninterrupted. The
actual value of Ars depends on the effectiveness of the scatterer as a receiving
antenna.
Some of the power received by the scatterer is absorbed in losses in the
scatterer unless it is a perfect conductor or a perfect insulator; the rest is
reradiated in various directions. The fraction absorbed is f a, so the fraction
reradiated is 1 – fa, and the total reradiated power is

Power power fraction of

reradiated = intercepted intercepted Pts= Prs (1 - fa)
by scatterer by scatterer power not (3.0)
absorbed

The conduction and displacement currents that flow in the scatterer result
in reradiation that has a pattern (like an antenna pattern); this situation is
illustrated in Fig.2. Note that the effective receiving area of the scatterer is a
function of its orientation relative to the incoming beam, so that Ars in (2.0) is
understood to apply only for the direction of the incoming beam. The reradiation
pattern may not be the same as the pattern of Ars, and the gain in the direction of
the receiver is the relevant value in the reradiation pattern. Thus,

Power / unit Power

area at = reradiated by gain toward spreading
receiver scatterer receive factor

1
Ss= PtsGts ---------- (4.0)
4Rr2
 Where Pts is the total reradiated power Gts is the gain of the scatterer in
the direction of the receiver, and 1 / 4Rr2 is the spreading factor for the
reradiation. Note that a major difference between a communication link and
radar scattering is that the communication link has only one spreading factor,
whereas the radar has two. Thus, if Rr = Rt, the total distance is 2Rt; for a
communication link with this distance, the spreading factor is only ¼(1 / 4Rt2),

Fig.2 Receiving, absorbing, and reradiating properties of scatterer.

Whereas for the radar it is (1/ 4) 2 (1 / Rt) 4. Hence, the spreading loss for
a radar is much greater than for a communication link with the same total path
length.
The power entering the receiver is
Received power / unit area effective
Power = at receiver receiving area Pr = SrAr, (5a)

Where the area Ar is the effective aperture of the receiving antenna, not its
actual area. Not only is this a function of direction, but it is also a function of the
load impedance the receiver provides to the antenna; for example, Pr would have
to be zero if the load were a short circuit or an open circuit.
The factors in (1.0) through (5.0) may be combined to obtain
1 1
Pr = (PtGt) -------- Ars (1 – fa) Gts ------- Ar
4Rt2 4Rt2

PtGtAr
= --------------- [Ars (1 – fa)Gts]  (5b)
(4)2 Rt2Rr2
 The factors associated with the scatterer are combined in the square
brackets. These factors are difficult to measure individually, and their relative
contributions are uninteresting to one wishing to know the size of the received
radar signal. Hence they are normally combined into one factor, the radar
scattering cross – section:
 = Ars (1 – fa)Gts  (6.0)

The cross – section  is a function of the directions of the incident wave

and the wave toward the receiver, as well as of the scatterer shape and dielectric
properties.
The final form of the radar equation is obtained by rewriting (5b) using the
definition of (6.0)

PtGtAr
Pr = ---------------   (7.0)
(4)2 Rt2Rr2

The most common situation is that for which receiving and transmitting
locations are the same, so that the transmitter and receiver distances are the
same. This is depicted in Fig. 3. Almost as common is the use of the same
antenna for transmitting and receiving, so the gains and effective apertures are
the same. That is,
Rt = Rr = R, Gt = Gr = G, At = Ar = A  (8.0)

Since the effective area of an antenna is related to its gain by

2G
A = -------  (9.0)
4

We may rewrite the radar equation (7.0) as

PtG22 PtA2
Pr = ------------ = ----------  (10.0)
(4)3 R4 42R4
 Where two forms are given, one in terms of the antenna gain and the
other in terms of the antenna area.
The radar equations of (7.0) and (10.0) are general equations for both
point and area targets. That is, the scattering cross – section  is not defined in
terms of any characteristic of a target type, but rather is the scattering cross –
section of a particular target. The form given in (10.0) is for the so-called
monostatic radar, and that in (7.0) is for bistatic radar, although it also applies for
monostatic radar when (8.0) is satisfied. With a remote – sensing radar an area
is usually to be studied. Therefore, a model suitable for area scattering is
appropriate. A similar model could be developed for volume scattering for the
remote sensing of, for example, clouds, which would lead to the expressions
given previously in Section 5 – 14.

Fig.3 Geometry of and quantities involved in monostatic radar equation

The basic scattering model for an area used in determining the radar
equation is based on the assumption that the area observed at any instant
consists of many point scattering targets whose locations are fairly random.
Their amplitudes also may be random. The normal assumption is that no single
amplitude is comparable with the overall signal that is received; i.e., no single
scatterer dominates. With these assumptions, at any particular point in space
and instant in time the phase relations among the scatterers from the different
point targets are fixed, and the electric fields received can be written as
N
Er =  Eri
i =1
 N
Er =  Eri ejie – j2kRi,  (11.0)
i =1

where I is the phase associated with the scattering process from a

particular point target i, and Ri is the exact distance from the radar to that
particular point target. Thus, the electric field at a particular point in time and
space has a well – determined value and is the result of a phasor summation
process indicated by (11.0). In general, one is concerned with the average
received power, rather than the field received at a particular instant. When the
averaging is performed, the phase effects cancel out as they do with noise [the
noise voltage at any particular instant is the result of a summation similar to
(11.0), but without the factor 2kRi]. When we perform an ensemble average over
all, space and time, the average received power Pr is given by
N
Pr =  Pri  (12.0)
i=1
Since the power received from an individual target is given by (10.0), we
may express (12.0) as
2 N PtiGi i
2

Pr = ------  --------  (13.0)

(4)3 i=1 Ri 4

The reason that Pti is included in the summation as well as Gi and Ri is

that the power illuminating different parts of the area may not be the same.
Further more, since different scatterers are at slightly different directions from the
radar, they may experience different gain, and the distances will be slightly
different also.
The next step in the procedure is to reexpress the scattering cross –
section of the individual subtarget:
i  Ai
i = ----------
 Ai
 The numerator of this equation simply represents the magnitude of the
scattering cross – section multiplied by the area on the ground associated with it,
and the denominator is the area on the ground. We now define the differential
scattering coefficient as the average value of the scattering cross – section per
unit area, given by
i
 = ----------
0
 (14.0)
 Ai

If each relevant  Ai over which Pt, G, R are nearly constant contains

enough scattering centers so that a reasonable average may be performed over
it, one may replace i in (13.0) by 0  Ai in the average, obtaining for the radar
equation, as a substitute for (13.0),
2 N
2
PtiGi (
0
 A i)
Pr = ------  --------------  (15.0)
(4)3 i=1 Ri4

If the condition allowing the use of 0 is satisfied, one may pass to the limit
for the summation and substitute an integral, namely:
2 PtG  dA
2 0

Pr = ------ area --------  (16.0)

(4) illuminated R4
3

This is the area – extensive form of the radar equation and is the form
most widely used or remote – sensing radars.
Note the limitations on this form of the radar equation:
1. Many individual point scatterers must exist in a region over which G,
Pt, and R are essentially constant.
2. Many more scatterers must exist in the total illuminated area at any
particular instant.
For example, if the illuminated area is only a few meters on a side and
contains a large point scatterer such as an automobile that has only two or three
major scattering centers, the return will be dominated by the automobile and the
very notion of a differential scattering coefficient does not apply. On the other
hand, an area of the same size containing stalks of grass probably would contain
more than enough individual scatterers to allow the differential scattering
coefficient to be used. Thus, one must be careful in using this concept to assure
that it is applicable in any particular case.

DISCRIMINATION IN ANGLE, RANGE, AND SPEED

Spatial discrimination between signals received from different parts of an
area may be achieved with microwave remote – sensing systems by methods
involving separation of the return from different angles, from different distances,
and from regions whose speed relative to the location of the sensor varies.
Spatial discrimination capability is usually described in terms of resolution.
Strictly speaking, resolution is a measure of the ability to determine whether only
one or more than one different target elements are being observed. Two objects
are said to be resolved if the system or observer can tell that, in fact, there are
two rather than one.
Consider, for an example, the illustration shown in Fig.4. A Radar
transmits a rectangular pulse. In Fig.4 (a) objects 1 and 2 give radar echoes that
cannot be resolved because the objects are too close together. At time t 1 the
pulse has not returned from either object. At t2 enough time has elapsed so that
a return has started from object number 1. At t3 a return has started from object
2, but object 1 still gives a return. At t4 no return is coming from object 1, but a
return continues from object 2. And finally, at t5 the pulse has passed both
objects and no return is observed. The radar output shown below exists, then,
for a long period, but there is no gap between the signals returned from object 1
and those from object 2. The output as shown is constant, but in fact it may be
different during the time when both objects are illuminated simultaneously. The
return from the two objects together may be either larger or smaller than that
from a single object, because the signals may either add or cancel, depending
upon their relative phases. This variation in amplitude, however, will not in
general allow one to determine that two objects are present. It might just as
easily be an indication that a single object is present that has a time – varying
return.

Fig.4 Examples of range resolution – rectangular pulse.

In Fig.4 (b) the objects are far enough apart so that they can be resolved
by this pulse. That is, at t2 the signal has started to return from object 1 and at t3
it has ceased to return from object 1, so a single unique return pulse is obtained
from object 1. Later, at time t4, we find part of the discrete return pulse from
object 2, and the output shown below indicates that one can readily determine
that two objects are present. Thus, in Fig.4 (a) the objects cannot be resolved,
whereas in Fig.4 (b) they are resolved.
A more realistic example is shown in Fig.5, where the radar response
(which may be in the angle, range, or speed domain) is shown as a curve. On
the left the two objects are so close together that the returns overlap as
indicated. A small decrease in the power at the midpoint is indicated, as would
be found if the two returns were in phase with each other. However, the
presence of two targets in this situation can only be established if the noise level
is so small that the very small dip between the two returns can indeed be reliably
interpreted. In the middle example the targets are somewhat farther apart, and
the effects of the potentially different phase returns from the different objects are
both small enough compared with a, one can determine that two objects are
present. For the size of a shown in the sketch, however, this condition is seldom
met. On the other hand, on the right side of the figure the objects are far enough
apart so that the response goes essentially to zero in the region between them,
and one can clearly establish that two objects are present.

Fig.5 More realistic example of range resolution.

The situations illustrated in Figs. 4 and 5 are the simplest cases. The
objects give signals of equal strength, and no phase differences between them
need be considered. In the more complex and therefore more realistic situation,
two objects to be resolved may give returns of different strengths. In this
situation, the middle example in Figure.19 would yield a response that would be
much harder to interpret as coming from two objects. On the other hand, the
right – hand example could probably still be recognized as being from two
objects.
 In optics, the standard definition for resolution is based on the ability to
determine the presence of line pairs, one of which gives a strong (bright)
response, and the other a weak (dark) response. The definition is based on a
particular contrast ratio, and for this ratio and line pairs with equal contrast
adjacent to each other, the resolution is said to be a certain number of line pairs
per unit distance on, say, a film (Slater, 1980).

Fig.6 Radar definition of resolution – the half – power width

The practice in microwave sensing is different. In microwave sensing the

problem of defining the contrast is avoided by stating that the resolution is the
distance (expressed in terms of angle, range, or speed) between the points at
which the response power is half the peak – power response. That is, the
resolution is defined as the half – power width of the response. This is illustrated
in Fig.6, where the resolution W is shown as the width between the half – power
points in the response from the target sensed.
One must be careful in comparing resolutions of optical systems with
those of microwave systems to ascertain the definition used for each system‟s
resolution. Although straightforward optical systems use the line pair method,
some scanning systems use a method comparable with that used in the
microwave community. Therefore, any comparison of the resolution of two
systems in the two frequency ranges must be made with care.
The microwave systems obtain discrimination and, hence, resolution by
measurement of one or more of the quantities angle, range, and speed. Angular
discrimination is achieved in the microwave system by the beam width of the
antenna. If the antenna has a narrower beam, the discrimination is better and
the resolution distance is smaller; we say the resolution is “finer”. The range
resolution, or distance resolution, is obtained by a time – delay measurement,
which is equivalent to a range measurement because of the known constant
speed of the electromagnetic wave. Many different kinds of time – delay
techniques may be used in radar systems. The most commonly used is the
transmission of a pulse as indicated in the example shown in Figs.4 and 5. The
time resolution is therefore approximately the duration of the pulse. Frequency
modulation is also widely used, and the relation of the parameters of the
modulations to the resolution will be described later. Other types of modulations,
such as use of binary sequences of phase, frequency, or amplitude, may also be
used for this measurement. The resolution is the obtained in the receiver by a
correlation process between the received signal and the modulation pulse
sequence. Speed measurement depends upon the Doppler frequency as
discussed later. The Doppler frequency shift of the received carrier frequency is
directly proportional to the relative speed between the object sensed and the
radar system. Thus, a filter discriminating the different frequencies received is
capable of discriminating the different speeds. The geometry for a radar system
traveling over the earth is such that different points on the earth have different
relative speeds, so these filters that distinguish relative speeds may be used to
discriminate between signals from different parts of the surface of the earth.
Fig.7 illustrates the different techniques and their combinations for a
microwave system carried on an aircraft traveling horizontally above the surface
of the earth. In Fig.7 (a) angle measurement alone is used; that is, the
beamwidth of the antenna establishes the resolution. The area shown cross –
hatched on the ground lies within the half – power contours of the antenna
pattern and therefore is defined as the resolvable are or “resolution cell” for an
angle – only system. Passive microwave (radiometer) systems depend strictly
on angle resolution, because both the range and speed measurement are based
on comparison of the received signal with modulation imposed at a transmitter.
Since the radiometer has no transmitter, these methods are not available to it.
In Fig.7 (b) the combination of range (distance) and angle resolution
techniques is shown. If only the range measurement were to be used (if the
antenna beam were to illuminate the entire surface of the ground), the resolution
cell would be a ring lying between the two half – power range response contours.
By combining the angle measurement and range measurement, the resolvable
region or resolution cell is limited to the smaller cross – hatched area. The
typical “ standard radar” uses this kind of discrimination system. The antenna
pattern is narrow in one direction and wide in another direction. The range
resolution allows breaking up the wide direction of the illuminated ground into
smaller segments. The narrow direction of the antenna beam is used for
resolution in angle.

Fig.7 Methods for resolution in microwave sensing: (a) angle only,

(b) angle and range, (c) angle and speed, and (d) range and speed.

Figure.7(c) shows a combination of angle and speed measurement used

for resolution. Here, half – power speed response contours are illustrated on the
surface. If the antenna beam illuminated the entire surface, these contours
would define completely the limits of the resolution cell. For horizontal motion
above a plane surface the contours are hyperbolas, so the resolution cell would
be a hyperbolic – shaped strip like that shown in the figure. By combining the
angular and speed response, a smaller resolvable area, as illustrated by the
cross – hatching, can be obtained. This method has been used in several
airborne and spaceborne scatterometer systems.
Figure.7 (d) shows a combination of range and speed discrimination to
establish the small resolution cell shown by the cross – hatching. Since with
readily available techniques the half – power speed – response and half – power
range – response contours may be made closer together than the beamwidth,
the resolution obtained by combining range and speed techniques is usually finer
than that obtained where an angle measurement must be used. This method is
the basis for synthetic – aperture radar, which is widely used in remotely sensing
the surface of the earth when fine resolution is required.
Since the passive radiometer must depend on angular resolution alone, its
ability to obtain a small resolution cell on the ground is limited. Hence, when fine
resolution is required with an airborne or spaceborne system, radar is needed.
For some applications, particularly over the ocean, the requirement for resolution
is not so stringent, and the selection between the sensors may be made on the
basis of the ability of one or the other (or a combination) to best determine the
parameters of the surface being observed.

SIMPLE PHYSICAL MODELS

Detailed mathematical scattering models are treated in Chapters 12 and
13 (Volume III), but an understanding of many of the physical scattering
mechanisms can be based on relatively simple concepts. This section treats
several widely used models; other models may be appropriate for certain
applications.

Point Scatterers
The concept used in Section 7 – 1 for defining the scattering coefficient 
involved multiple independent scatterers, although the nature of the individual
elements of the surface was not mentioned. Some of the earliest models
assumed that surface scattering arose from many point scatterers, i.e., the
individual patterns were very broad. For vegetation – covered surfaces having
leaves small compared with a wavelength, such a model might be a reasonable
one, but it does not explain the behavior of scattering from many surfaces.
Clapp (1946) was one of the earliest writers to develop models of this
kind. His models were all based on collections of small spherical scatterers, and
gave the variation of  with incidence angle as
Model 1:
 () =  (0)  1.0
Model 2:
 () =  (0) cos   2.0
Model 3:
 () =  (0) cos2   3.0
Model 1 assumes that the spheres are independent and sufficiently
separated that they cannot shield each other; thus the isotropic reradiation
pattern of the spheres is preserved. For this model to apply, the spheres must
be arranged in a layer of enough depth so that, on the average, the same
number of spheres is illuminated by the beam at all angles. This is quite
unrealistic, so very few, if any, natural surfaces give experimental results in
accord with model 1.
For model 2, the spheres are more densely packed; consequently one
must consider the concepts of illuminated projected area and surface area, as
discussed briefly in Section 11 – 1. Figure 8 shows the geometry for an
incremental surface area illuminated by radiation at incidence angle . The area
is assumed to be rectangular with depth y. Two areas must be considered: the
surface area
As =  x y,  4.0
and the projected area (normal to the incoming rays)
Ap = ( x cos) y
= n y  5.0
 For a given illuminating beam of width in elevation , the power through
Ap is independent of  after the distance effect is removed, and the area is also
independent of . Hence the power per unit projected area is independent of .
On the other hand, the surface area As increases with increasing , so the
power per unit surface area decreases as  increases. This means that the
scattering coefficient, which is based on scattering per unit surface area, is not
independent of . Hence the power would be expected to decrease if a group of
spheres were closely packed near the surface, because the number of spheres
in a given area would remain the same while the power incident on this area
would decrease with increasing ; this is the basis for model 2.

Fig.8 Geometry for relation between surface area x and projected area n

Model 3 is based on Lambert‟s law for optics. Here the assumption is that
reradiation is in accordance with a cosine law rather than being isotropic for each
sphere. Since the incident power per unit area also decreases according to a
cosine law, the combination is as in (3). Rough physical surfaces do indeed
sometimes show this behavior over the middle range of angles of incidence.
Mathematically we may derive model 2 by first considering a lossless,
perfectly conducting set of spheres and then extending this is to the lossy case.
For a projected area Ap, the incident power at distance R is given by
 P t Gt
Pi = ------- Ap  6.0
4R2

where Pt is the transmitter power and Gt is the gain of the transmitting

antenna. If we assume isotropic radiation into the upper hemisphere, but none
into the lower, the power received from Ap by an antenna with effective aperture
Ar is
P iA r 2PtGtAr
Pr = -------- = ----------- Ap  7.0
2R2 (4R2) 2

This power must be related to the surface area As to allow the writing of
the radar equation in a standard form like that given in Section 7-1. Thus

PtGtAr
Pr = ----------- 0As  8.0
(4R2) 2

By comparing (7.0) and (8.0), we can see that

0As = 2Ap
So that for this case
2Ap
0
= ------- 2 cos   9.0
As

Thus, for the lossless situation, the maximum value of 0 is 2 because of the
restriction that confines reradiation to the upper hemisphere. When loss is
present, the amount of power reradiated is reduced, so
0 () = 0 (0) cos ,  10.0

Where 0 (0) is less than 2.

Some workers in the remote – sensing field consider field consider model
2 to represent a perfectly rough surface, and choose to refer all scattering to the
scattering from such a surface. Another way to think of this is to consider
referring all scattering to the projected, rather than to the surface, area. The
scattering coefficient with this definition is designated  and is related to  by
cos, as indicted in (9.0). in terms of  (), model 2 is given by
 () =  () (perfectly rough surface).  (11.0)
Lambert‟s law (model 3) is based on a cosine pattern for reradiation into the
upper hemisphere, which means that  (), instead of being constant, is given by
 () =  () cos  (Lambert‟s law).  (12.0)
Although actual scattering – coefficient variation seldom approximates closely
either model 2 or model 3, sometimes either model, or something in between,
may be used to represent vegetation scattering.

Facets
One approach different from those used by Clapp involves the use of facet
models. The concept of scattering from facets in really more like a specular –
reflection than a scattering model, but is often described as scattering. Figure. 9
Illustrates how one can approximate a rough surface through a series of small
planar “ facets,” each tangential to the actual surface. Facet models treat the
scattering or reflection from the assemblage of such facets by taking into account
both their reradiation patterns and the distribution of their slopes.

Fig.9 Representing a rough surface as a collection of facets

Fig.10 Normal – incidence reradiation patterns of facets.

 The scattering from an individual plane facet depends on its size.
Figure.10 shows the concept graphically. In Fig.10 (a) the scattering from an
infinite plane is given by a delta function; that is, the surface is a specular
reflector, so the only way reradiation can come back to the point source is for the
ray to strike it at normal incidence. Figure. 10 (b) shows the narrow reradiation
pattern for a facet that is wide but finite in extent, and Fig.10 (c) shows the wider
pattern for a narrower facet. Side lobes are present for both finite facet sizes, but
their amplitude is low compared with the main lobe.
The qualitative terms used to describe the facet sizes depend on the
wavelength. For a facet to be “wide” it must be many wavelengths across. The
concept is the same as that for the radiation pattern of an antenna aperture.
Thus, a facet that has a narrow reradiation pattern at a short wavelength may
have a wide one at a longer wavelength.
Figure.11 illustrates the concept graphically. The reradiation patterns of
six facets are shown, along with both the maximum possible contribution they
would make if properly oriented, and the contributions they make in actuality with
their true orientations. The large facets could make larger contributions to the
signal if they were properly oriented, but in fact, their major reradiation is in the
specular direction away from the radar. Because the reradiation from the smaller
facets is not as directional, they contribute out of proportion to their maximum
potential.

Fig.11 Facet Model

 For the situation in Fig.10 (a) the facet is infinite in extent (relative to a
wavelength). For this to be true with realistic – sized facets, one must assume
that the wavelength is zero. This assumption is also implicit in geometric optics,
so these facet models are known as “geometric – optics” scattering models. With
this assumption, only those facets whose slope is appropriate to beam the signal
back to a part of the receiving antenna contribute to the return signal. This point
is illustrated in Fig.12, where the receiving antenna subtends an angle2 m at
the facet, and a range of facet slopes m wide contributes all of the signal
received. If the probability density function (PDF) of the facet slopes is P s (),
the received power is given by
+m/2
Pr = A Ps () d  (13.0)
 - m/2
Where A is a constant that contains all the radar and distance parameters
in the radar equation.
The geometric – optics model I relatively easy to handle mathematically
because only the slope distribution and antenna geometry are needed. When
the finite sizes of the facets and the finite wavelength are considered, both the
slope distribution Ps () and a PDF for facet size must be used, and it is difficult
to determine the latter. In fact, one expects the facet size and the slope to be
related, so a joint distribution Ps (, size) is needed – and this only applies to the
one – dimensional case. Where the slopes may be in any direction, one must
consider the joint distribution Ps (x,y, size) as well as the full facet pattern (a
function of facet size). Although this approach is intuitively satisfactory, it is of
little practical value because of the difficulty o establishing meaningful PDFs.
The approximation of a rough surface by a set of tangent planes may be
improved by the use of formulas for scattering from curved surfaces. The
“curved facets” that may then be used can represent the actual rough surface
better than the tangent planes can, but the problem of modeling them is at least
as difficult, if not more so. Thus, although some theories based on this approach
have been developed, the approach has not achieved wide acceptance.
 Fig.12 Geometry of zero – wavelength facet reflection (geometric optics)

Since specular reflection can occur only when the ray is normal to the
target, it is confined to angles of incidence near vertical except in special cases
where smooth surfaces with steep slopes are present. True specular reflection
(where the signal is the same as from an infinite plane) can only occur when the
flat area is two or more Fresnel zones across. In the rare situations when the flat
area is two or more Fresnel zones across. In the rare situations when this
condition occurs, the signal looks as if it has come from an image point located
the same distance below the surface as the radar is above the surface. The
usual radar equation is then replaced by
PtGtArTsp
Pr = -------- =  14.0
4(2h) 2

where „h‟ is the height above the surface and Tsp is the Fresnel power reflection
coefficient.
Al though this situation is rare, reasonably flat surfaces smaller in extent
than a Fresnel zone and corresponding to the large facets in Fig.4(b) often are
found near vertical incidence. Frequently the scattering coefficient varies as
shown in Fig.7, and one can consider the variation as a superposition of the delta
function of specular reflection with another function characterized by a gentler
decrease in  with angle similar to that of one of the Clapp models. When such
a surface is observed by a narrow – beam antenna of nonzero beamwidth, the
averaging over the beamwidth results in the smooth curve joining the specular
and rough – surface responses. Since quasispecular facets [Fig.4 (b)] are more
common, the curve in Fig.7 may also be used to join the response for a facet
model near vertical with a rough – surface model at large incidence angles.

Fig.13 The Backscattering coefficient consists of a coherent specular component

around nadir and an incoherent scattering component (dashed curve) that varies
slowly with . The solid, smooth curve is the sum of the two components,
convolved with a relatively narrow antenna pattern

Bargg Resonanace
Surfaces may be described in terms of the spectrum of their vertical
displacements from the mean. Often the response at the larger angles of
incidence (beyond 30) appears to arise primarily from resonant components in
this spectrum, even though the energy in the resonant – frequency range is quite
small compared with the total energy. This mechanism has been used to
describe scattering from the ocean in terms of resonance with the tiny capillary
and short gravity waves, even when these waves have a height measured in
tenths of a millimeter whereas they ride on waves that are meters high. This
phenomenon is described by the term Bragg resonance by analogy with the
Bragg resonances used in spectroscopy.
Figure.14 illustrates the Bragg resonance phenomenon. A sinusoidal
component of the surface spectrum is shown, along with an incoming plane wave
at angle of incidence . The wavelength of the surface component is L, and the
radar wavelength is . If the excess distance from the source to each successive
wave crest is /2, the round – trip phase difference between the return signals
from successive crests is 360, so the signals add in phase; if R is any other
distance (except a multiple of /2), they add out of phase.
If one postulates that the signal from each of the Bragg – resonant waves
comes from the same part of the wave, the received voltage is the phase –
coherent sum of the voltages from the individual waves; thus
N
Vr =  Voe-j2kRoe – j2klR,  15.0
l=0

Fig.14 In – phase addition for Bragg scattering when R = n/2

where N is the total number of wavelengths of the resonant component within the
illuminated area. Since (15.0) is a geometric progression, its sum is easily
obtained as in calculations for antenna arrays:

sin [k (N + 1) R]
Vr = Voe-jkr0 ------------------------  (16.0)
Sin [k R]
The maximum voltage is then proportional to N+1, and the maximum power is
proportional to (N+1) 2. Since the average power for the nonresonant scatterers
is proportional to their number rather than to its square, the effect of very small
resonant scatterers in large numbers can exceed that of much larger ones with
random spacing but comparable numbers.
The Bragg resonance condition is
2
kR = ----- R = n, n = 0,1,2,…..  (17a)


In terms of the spatial wavelength L and the angle of incidence  (Fig.8.0), the
Bragg resonance condition is written in the form
2L
---- Sin  = n, n = 0,1,2,…..  (17b)


To illustrate the magnitude of this effect, consider an example for which  = 30,
 = 3 cm and L = 3 cm, which satisfies (17b). If the illuminated area is 3 m
across in the direction of the resonant surface – wave normal, then N = 100 and
Vrmax = 101 V0. The resonant power received is then 10, 201 V 02. A single large
scatterer with 20 dB more signal results in a power of only 100V02, or 20 dB less
than that from 100 smaller resonant scatterers. Even if there are 100 of the
20 dB – larger scatterers, the resulting mean power is only about the same as
that for the 100 small but resonant scatterers. If the illuminated area were 30 m
across, 1000 of the resonant scatterers would be present and the signal power
from them would be 106V02, so 1000 nonresonant 30 – dB – higher scatterers
would be needed to give the same return. For airborne and space- borne
scatterometers, whose footprints are measured in hundreds or thousands of
meters, the resonance effect is so powerful that it can dominate the signal even if
the resonant scatterers are very small indeed. This resonance effect is similar to
that of coherent integration discussed in Chapter.7 and in connection with SAR
image formation in Chapter.9.
Hard Targets
Most of the area in an image is dominated by surface scattering as
discussed above, but important features in radar images are caused by isolated,
very strong scatterers, often referred to as hard targets. Some of these large
signals come from essentially normal flat or slightly rounded metal surfaces.
Others are from resonant elements and appropriately oriented linear conductors.
Many of them are caused by the corner – reflector effect.
Resonant elements are metallic or high – dielectric – constant objects
whose lengths are integer multiples of  /2. Such objects have a reradiation
pattern the same as that for an antenna of the same dimensions. For example, if
a half – wavelength metal object is in the field of view, its reradiation pattern is
like the square of the gain pattern for a  /2 antenna;
 = max cos4  (18.0)
where  is measured from the normal to the axis of the object. The polarization
of the incoming wave need not be parallel to the length of the object, provided a
component of the electric field is parallel; if the two are not parallel, the scattered
field is differently polarized than the incoming field, as shown in Fig.15. In the
figure, the incoming E is vertical, the object (a wire) is titled, and the scattered
field has the same polarization as the field that would be radiated by the wire had
it been used as an antenna (both vertical and horizontal components).

Fig.15 The Field Es of the wave scattered by a wire has x – and z – components
even though the incident field was polarized in the z – direction only.
 This resonant effect can be very powerful. For example, in one K a – band
image of an airport, some very bright signals were observed for points in an
aircraft parking area, but it was found that no aircraft had been parked there
during the production of the image. Closer examination on the ground showed
that the bright returns were from resonant elements in a small storm drain.
Long linear conductors, even though not resonant, may give strong
echoes. The strongest echoes occur when the direction of the incoming beam is
normal to the conductor axis. If the conductor length is many wavelengths (up to
a Fresnel zone), the signal is like the one that would be received from a linear
antenna of the same aperture.
Corner –reflector effects are especially important for buildings. The
enhancement that corner reflectors give to echoes over a fairly wide range of
incidence angles was discussed in Chapter 10 in connection with their use as
calibration targets. With buildings, the angle between the vertical wall and the
ground forms a corner reflector. Other corner reflectors can be found where
walls (or walls and roofs) join at right angles. Street curbs also may serve as
corner reflectors.
Most of the corner reflectors discussed in Chapter 10 had three sides, so
their patterns were similar in any plane through their axes. With the corner
reflectors found in cities only two sides are usually present, so the incoming
signal has to be nearly normal to the line of the corner for maximum echo
enhancement. For cities laid out in a grid pattern this results in the “cardinal –
point effect”. When the radar flies parallel to either direction of the grid , the
beam is normal to the corners formed by the walls parallel to the flight direction,
and the signal from the buildings is high. When the radar flies in another
direction, the corner – reflector enhancement cannot take place, so the signal is
much lower. With modest resolution systems this can cause small gridded
villages to be missed unless the flight direction is parallel to the grid, even though
such villages show up strongly when the flight direction is proper.
A fine example of the cardinal – point effect is shown in Fig.16, an image
of Sun City, Arizona. This planned city is laid out in concentric circles, so that on
a single pass the radar observes all of the different azimuth angles relative to the
walls of the houses. The brighter echoes from those parts of the circles where
the house walls are parallel to the flight line stand out in the image.

Fig.16 Radar image of Sun City, Arizona

(Courtesy of the U.S Strategic Air Command)

THE NATURE OF VOUME SCATTERING

The nature, media are inhomogeneous, although some of them may be
treated as though they were homogeneous for a given incident frequency, angle,
or medium condition. In view of the discussion in Section 11- 1, both surface and
volume scattering usually are present in scattering or emission from terrain.
However, it is convenient to ignore one or the other of these whenever is it
practical to do so. For example, in the microwave region, the sea has a large
dielectric constant and is treated as a homogeneous medium capable of surface
scattering only. At optical wavelengths, however, the dielectric constant of
seawater is much smaller and the dust particles in water can make volume
scattering important.
Since volume scattering is caused mainly by dielectric discontinuities
within a volume and, in general, the spatial locations of discontinuities are
random, intuitively we expect the scattered waves within the volume to be in all
directions. For an inhomogeneous medium with a small average dielectric
constant, the angular backscattering curve should be fairly uniform. For a larger
average dielectric constant, the angular backscattering curve drops off faster with
increasing incidence angle. A qualitative illustration of these backscattering
curves is shown in Fig. 17.

Fig.17 Dependence of the backscattering coefficient on the average dielectric

constant of a scattering volume.
To determine the presence of volume scattering, we need to know (1) if
the medium is inhomogeneous and (2) what its effective depth of penetration is.
Usually, item (1) is known from physical properties and item (2) is estimated from
the dielectric properties of the medium. When both volume and surface
scattering are present, the factors involved may be summarized as follows. In
surface scattering, (a) the scattering strength is proportional to the relative
complex dielectric constant of the surface, and (b) its angular scattering pattern is
governed by the surface roughness (see Section 11-2). In volume scattering
(a) the scattering strength is proportional to the dielectric discontinuities inside
the medium (below the boundary surface) and the density of the embedded
inhomogeneities (or the variance of the fluctuating dielectric function for a
continuous random medium), and (b) its angular scattering pattern is determined
by the roughness of the boundary surface, the average dielectric constant of the
medium, and the geometric size of the inhomogenerities relative to the incident
wavelength (or the correlation length of the fluctuating dielectric function for a
continuous random medium). These relationships are translated into
mathematical models in Chapter 13 (Volume III).
To obtain an expression for the depth of penetration, consider a wave
incident upon a soil surface from the air in the direction z: part of its power is
scattered back into the air medium and the remainder is transmitted across the
boundary into the soil medium. If the transmitted power at a point just beneath
the surface (z = 0+) is P (0 +), the power at a depth z is given by
z
P(z) = P(0 +) exp (- ke (z‟) dz‟),  (19.0)
0

where ke (z) is the extinction coefficient of the soil medium at depth z. The
penetration depth p may be defined as the depth z at which
P (p) 1
--------- = -----  (20.0)
P (0 +) e
Or
p

 ke (z) dz = 1.  (21.0)
0

If scattering in the soil medium is ignored,

ke  ka = 2  (22.0)
where ka is the (power) absorption coefficient and  is the field attenuation
coefficient [see (4.66) and (4.114)]. If, additionally, (z) is approximately
constant as a function of z, (11.41) reduces to
1
p = ------  (23.0)
2

where
2
= ---- Im []  (24.0)
 is the wavelength in free space, and  = ‟ - j” is the relative complex
dielectric constant. The penetration depth should not be confused with the skin
depth s = 1/. For materials with ” / ‟ < 0.1, the above two expressions lead to
the approximate formula:
‟
p  -------  (25.0)
2”

which is applicable to most natural materials except water.

In what follows we shall discuss the relative importance of surface and
volume scattering for sea ice, snow, vegetation, and soil. Where possible,
sample calculations of the effective depth of penetration are provided.

Sea Ice
Sea ice is a mixture of ice, salt, brine pockets, and air bubbles. The air –
sea – ice boundary may be smooth or rough. Its average relative dielectric
constant is around 3.5, and its loss tangent varies widely depending upon ice
type (see Appendix E of Volume III). In backscattering within 20 of nadir,
surface scattering is expected to dominate. At large incidence angles, volume
scattering is most important, but its angular shape is modified by the interaction
between surface and volume scattering. The loss tangent of sea ice varies with
depth. Hence, it is important to examine the depth of penetration in each case to
determine which are the dominant inhomogeneities responsible for volume
scattering. Figure.18 shows typical curves for the penetration depth in pure ice
and in first – year and multiyear sea ice. The curve for pure ice was calculated
from dielectric – constant data at T = - 10C (these data are summarized in
Appendix E of Volume III). Dielectric – constant data reported in the literature for
first – year and multiyear sea ice cover a relatively narrow range for ‟, but ” can
vary over a very wide range of values depending upon the ice density, salinity,
temperature, and geometrical configuration of the air and brine pockets in the ice
volume. The corresponding ranges of values for the penetration depth p are
shown as shaded areas in Figure.18. It should be noted that these are typical
values for T in the neighborhood of - 10 C; ” usually decreases with decreasing
temperature, which would result in greater penetration depths. If the temperature
were decreased from 10 C to - 30 C, for example, p of first – year and
multiyear sea ice could increase by a factor of 3 – 10, depending on the specific
ice type.

Fig.18 Penetration depth in pure ice and in first – year and multi – year sea ice.
Shaded areas correspond to the range of values of dielectric constant usually
reported in the literature. Which arises in part from the range of salinities and
densities commonly encountered in sea ice (from Appendix E of Volume III).

Snow
Dry snow is a mixture of air and ice; consequently, its relative permittivity
is a function of the snow density s and the relative permittivites of air (unity) and
ice. The relative permittivity of pure ice is ‟ = 3.15 and is independent of
temperature and frequency in the microwave region. Hence, in effect, the
permittivity of dry snow, ‟ds‟ is governed by the density s, which, under natural
snowpack conditions, usually varies over the range 0.2 – 0.5 gcm-3. The
corresponding range for ‟ds is 1.4 – 2 (see Appendix E of Volume III). If the
snow contains water in liquid form (in which case it is referred to as wet snow),
the snow permittivity becomes an increasing function of the liquid – water content
m (percent by volume), but its magnitude rarely exceeds 4 (and usually does not
exceed 3) for the range of values of m commonly encountered under natural
conditions.
In contrast to the relatively weak dependence of ‟ on m, the imaginary
part, ”, is extremely sensitive to the presence of liquid water in the snow mixture,
and the greatest relative change in the magnitude of ” occurs over the range of
m between 0 and 0.5 percent. At 9.375 GHz, for example, Cumming (1952)
– 3
found that ” of dry snow (with density s = 0.28 gcm-3) is about 1.2 x 10 at
T= -1C. Upon increasing m from zero to 0.5 percent, “ increased by more than
one order of magnitude, which corresponds to a decrease in the penetration
depth by the same ratio. This dependence of p on m is illustrated in Fig.19 for
4, 10, and 20 GHz. The curves shown were calculated using, in part, measured
dielectric data and, in part, frequency extrapolations based on measured data.
The use of extrapolation is necessary due to the limited availability of
experimental data on the microwave dielectric properties of wet snow.

Fig.19 The penetration depth of snow decreases rapidly with increasing liquid
water content, especially in the immediate vicinity of M = 0
(from Appendix E of Volume III).
 The power transmission coefficient as () of the air – snow boundary is
governed by the relative permittivity of the snow, the angle of incidence , and
the polarization configuration (assuming the snow surface to be relatively
smooth). As was stated earlier, ‟ of snow usually is less than 3, even for wet
snow conditions. This means that significant transmission across the boundary
can take place over a wide range of angles , and for both horizontal and vertical
polarizations. At nadir, for example as  0.98 for dry snow with s = 0.4, and for
snow with m as high as 5 percent, as  0.9. Volume scattering within the snow
medium is caused mainly by the ice particles, whose relative dielectric constant
is around 3.15. In backscattering, near – nadir scattering is expected to be
dominated by the air – snow boundary (or the ground surface, if the snow is
sufficiently dry). At larger incidence angles, volume scattering should dominate,
but the angular shape of the scattering curve is affected by other factors
mentioned earlier in this section. When a snow medium is very wet, so that
penetration is negligible, volume scattering can be ignored.

Vegetation Canopies
Estimates of the penetration depth (via the dielectric constant) in
vegetation are difficult to determine, due to the difficulty of establishing the
effective dielectric constant of a medium in which the size of the inclusions
(leaves, stalk, etc.) usually is of the order of magnitude of the wavelength or
larger. Indirect estimates of p, however, have been obtained through
comparisons of the radar backscattering coefficient at angles close to nadir for
bare soil with vegetation – covered soil. These estimates indicate that p of
mature crops in the green stage of growth is typically several meters at
frequencies around 1 GHz, and decreases to one meter or less at frequencies
above 10 GHz. Key factors that govern the penetration depth at a given
frequency include the geometry of the canopy parts, the moisture content of the
canopy parts, and the vegetation volume fraction in the canopy (volume of
vegetation material per unit volume of canopy, which is mostly air). If the plant
moisture content is low, as is the case for many types of crops during the fruit –
filling stage prior to harvest, the penetration depth can be substantially higher
than it is when the crop is green and lush.
In vegetation scattering, the air – vegetation boundary usually is
unimportant. Hence, in general, we have a combination of volume scattering in
the vegetation layer and surface scattering by the underlying ground. The
relative importance of these two contributions depends on several factors
including the penetration depth, the height of the canopy, the microwave
frequency, and the angle of incidence. The angle of incidence  plays an
especially important role in radar backscattering. It was noted earlier that for
surface scattering,  usually decreases rapidly with increasing  for angles close
the nadir (unless the surface is electromagnetically very rough), while  due to
volume scattering decreases very slowly with increasing  (see Fig. 17). Hence,
in general, we expect the backscattering of the soil to be at least as important as
the volume backscattering contribution of the canopy for angles close to nadir,
while at large angles of incidence the volume backscattering contribution of the
canopy is more likely to be the dominant contribution, unless the penetration
depth is significantly larger than the slant height of the canopy (which may be the
case if the plant water content is small and / or the frequency is in the lower part
of the microwave spectrum).

Soil
Calculated curves for the depth of penetration in loamy soil are shown in
Fig.20 as a function of the volumetric moisture content m for three microwave
frequencies. It is observed that if m is high, the p is less than 1 cm at 10 GHz
and less than 10 cm at the lowest frequency, 1.3 GHz. Moreover, it should be
noted that the values of p shown are somewhat larger than those for real soil
conditions because the computation using (23.0) doest not take account of
losses due to scattering in the soil medium. Hence, it usually is assumed that
because the penetration depth in wet soil is so small, the inhomogeneities within
the soil medium cause negligible amounts of volume scattering in comparison to
scattering from the air – soil surface. The 4 – and 10 – GHz curves of Fig. 20.
show that p  5 cm for values of m as low as 0.05 gcm-3, which means that the
above assumption (that volume scattering in the soil medium may be ignored in
favor of surface scattering) may be extended to cover all moisture conditions
above 0.05 gcm-3 at frequencies above 4 GHz. In sufficient evidence exists at
the present time to support the applicability of the assumption at frequencies in
the neighborhood of 1 GHz for dry soil conditions.

Fig.20 Penetration depth of loamy soil as a function of moisture content

(Appendix E of Volume III).
GENERAL CHARACTERISTICS OF TERRAIN AND SEA BACKSCATTERING
Relating the backscattering coefficient  to the physical parameters of a
particular type of target is an important step toward acquiring, from a remote
platform, information concerning the target. In earlier sections, we discussed
the relationships between  and (a) volumetric water content for vegetation
canopies, (b) soil-moisture content for bare and vegetation – covered soil, (c)
snow water equivalent, and (d) wind speed and direction. These and other
similar application – oriented relationships focus on the variation of  within a
given target class such as vegetation or snow. Equally important is the variation
between different classes and the statistical properties of  for individual
classes. Such information is useful for discriminating between classes and for
system – design considerations.
In this section, we shall provide information on the general characteristics
of the backscattering coefficient for individual terrain classes and for the sea.
The level of detail and the degree of completeness will vary from one class to
another, corresponding to the availability of published data. One common
problem in remote sensing, and to some extent in all experimental research,
relates to the absolute calibration of measurement instruments. When a sensor
is used to observe a given target or class of targets, either as a function of time
or of some target parameter, the information derived from the sensor generally is
useful, even though the sensor‟s measurement accuracy (absolute level) may not
be very good. To establish the statistical distribution of a quantity, such as the
backscattering coefficient , for a give target class, it is important that the
distribution be based on a large volume of measurements. This can be
accomplished through the compilation of pertinent data reported in the literature.
The drawback to this approach, however, is that the distribution resulting from a
data base consisting of data reported by different investigators, using different
instruments with different accuracies, will not be a true representation of the
distribution of  for that class, due to the biases between the absolute levels of
the various instruments. Hence, the approach taken here is to select a data base
generated by a single sensor for the presentation of information on the variability
of  for an individual target class rather than to combine data bases, even
though the results may not represent as wide a range of conditions of that target
class as one would like.

Average  for Terrain

Before we consider individual terrain classes, we shall first consider the
statistics of  based on the data recorded by the 13.9 – GHz scatterometer that
was flown aboard Skylab in 1973. The area illuminated by the scatterometer
varied between about 100km2 at the nadir and 360km2 at  = 48; thus, each
datum represents an average value of  for the various terrain classes included
in the illuminated area.
The results discussed below were derived from data acquired during
Skylab passes over North America in 1973, and more specifically, from a special
concentration of passes above the western portion of the U.S. during the periods
30 May to 14 June and 3 August to 21 September 1973.
Figure.21 shows a histogram of Skylab data acquired over land for  in the
32 - 34 range, with VV polarization. In view of the time of year and the
geographical locations of the passes, the target class may be described as a
general class encompassing agricultural terrain, natural vegetation, desert
surfaces, and a miscellany of cultural scenes, lakes, and rivers, although the
latter set of classes exercises a statistically minor influence on the measured
data because of their small area extent relative to the size of the illuminated area.
For smaller illuminated areas such as those of RARs and SARs, the
variation in  is much greater than that illustrated in Fig.21 for the data observed
by the Skylab scatterometer. However, no extensive study has been made that
would allow a statement of the types of distributions and ranges to be expected.
Presumably the range is closely related to the size of the illuminated area, but
the relationship has yet to be determined.
From histograms (similar to that shown in Fig.21) that were generated at
several angles  between 1 and 45, it was possible to generate the curves
shown in Fig. 22 for the mean value and the upper and lower deciles (levels
exceeded by 10 percent and 90 percent of the total number of measurements) of
. It is observed that the dynamic range of  (lower to upper decile) is largest at
angles close to nadir; it decreases with  until about 10 and then remains
approximately constant with increasing , except for a slight increase at 45.
Similar distributions also were reported for HH and HV polarizations
(Moore et al., 1975).

Fig.21 Histogram of Skylab’s scatterometer data over North America for the 32 -
34 angle of incidence range (Moore et al., 1975).

It is worth noting that the curve for the mean value of was found to have
an angular behavior very similar to the mean value for agricultural terrain based
on truck – mounted data acquired during the growing season (see Section
Agricultural Terrain) and the two curves were within 3dB of one another in level
(Moore et al., 1980).
 Fig.22 Angular patterns of the mean, upper decile, and lower decile of Skylab’s
scatterometer observations over North America during the summer season
(Moore et al., 1975).
Agricultural Terrain
The statistical behavior of for agricultural terrain was investigated by
Ulaby (1980) using a truck – mounted radar spectrometer covering the
1 – 18 – GHz frequency region. The targets (which were agricultural fields
planted with corn, milo, soybeans, alfalfa, wheat, clover, and grass) were
observed over a wide range of growth stages extending from preemergence (i.e.,
bare soil) to post harvest. The observations were made as a function of the
angle of incidence for each of 18 frequencies between 1 and 18 GH z and for
each of the three linear polarization configurations (HH, VV, and HV). A total of
285 observations were made for frequencies in the 1 – 8 GHz band and 567
observations for frequencies in the 8 – 18 GHz band (two separate subsystems
were used).
 Fig. 23 Angular variation of backscattering from vegetation at (a) 1.5 GHz,
(b) 4.25 GHz, (c) 8.6 GHz, and (d) 17 GHz. The curves shown are for the mean
backscattering coefficient -0, the median 050, and the levels exceeded 5% and
95% of the time,  05, and 095, respectively (Ulaby, 1980)

For each set of observations made using a given combination of angle of

incidence , frequency f, and polarization configuration p, a histogram similar to
that shown in the previous section (Fig.21) was generated. Four levels of were
computed from the histogram: (a) , the mean value of  (b) 5, the level
exceeded only 5 percent of the time, (c) 50, the median and (d) 95, the level
exceeded 95 percent of the time. Figure.23 shows the angular variation of these
four levels at each of four different frequencies. It is observed that (a) the mean
value, , is always higher in level than the median, 50, at angles close to nadir,
due o the asymmetry of the histograms in this angular range, (b) the mean and
the median are very close to each other at angles higher than 20, and (c) the 95
percent – t0 5 – percent dynamic range always decrease with increasing 
between  = 0 and  = 20, and then remains approximately constant for larger .
 The same four levels of  are presented in Fig.24 as a function of
frequency for  = 0, 20, and 60. Note the difference in behavior between the
nadir curves of Fig.24 (a) and those at the higher angles in (b) and (c). At  = 0,
all four levels of  indicate a weak dependence on frequency, while at  = 20
and 60, the curves show a rapid increase in level with increasing f up to about 6
GHz, followed by a slower rate between 6 and 18 GHz.

Fig. 24 Spectral variation of backscattering from vegetation at (a) 0, (b) 20, and
(c) 60. The curves shown are for the mean backscattering coefficient -0, the
median 050, and the levels exceeded 5% and 95% of the time,  05, and 095,
respectively (Ulaby, 1980)

The behavior depicted in Figs.23 and 24 is for HH polarization. Similar

curves are included in Ulaby (1980) for VV and HV polarizations, together with an
empirical model expressing  and 50 as functions of  and f for each of the
polarization configurations.
Snow
In comparison with agricultural terrain, few data are available on the back-
scattering coefficient of snow – covered ground. Figures.25 and 26 show curves
for 5, , and 95 as a function of frequency at  = 20 and 50, respectively.
These curves are based on experimental data acquired by the University of
Kansas truck – mounted radar spectrometer at 18 frequencies in the
1 – 18 – GHz band and at a single frequency at 35.6 GHz (Stiles and Ulaby,
1980b). The database consists of observations over a six – week period at a test
site in Colorado and includes several diurnal observations during which the snow
varied from a very dry state at night to a wet – snow state in the early afternoons.
Most of the data were made of man – packed snow up to a depth of 1.7 m.

Fig.25 Observed spectral variation of backscattering from snow at  = 20. The

curves shown are for the mean backscattering coefficient -0, and the levels
exceeded 5% and 95% of the time,  05, and 095, respectively
(adapted from Stiles and Ulaby, 1980)
 The general frequency behavior of the curves shown in Figs. 25 and 26 is
similar to that shown in the previous section for vegetation. Note that a different
scale is used on the abscissa for the frequency range below 17 GHz than for the
range between 17 and 35 GHz, which gives rise to an apparent increase in
slope with increasing frequency above 17 GHz. Had a linear scale been used
throughout the 1 – 35 GHz range on the abscissa, this artificial change in slope
above 17 GHz would not have occurred.

Fig.26 Observed spectral variation of backscattering from snow at  = 50. The

curves shown are for the mean backscattering coefficient -0, and the levels
exceeded 5% and 95% of the time,  05, and 095, respectively
(adapted from Stiles and Ulaby, 1980)
It should be noted that the curves shown in Figs. 5 and 6 merely give a
representative picture of the range of values that  would have for snow –
covered terrain; they are based on a data set of limited size and they by no
means cover all types or conditions of snow that may exist under natural
conditions.
Sea Ice
At a given combination of radar parameters (angle of incidence,
frequency, and polarization configuration), the backscattering coefficient of sea
ice has been observed to vary over a wide range of values, sometimes in excess
of 20dB, depending upon the ice type and condition. The factors governing of 
sea ice are discussed in Chapter 20 (Volume III), but to illustrate the range of
variability we show (Fig.27) curves of the upper and lower limits of  values for
first – year sea ice at 1.5 GHz. These curves, which were generated from
measurement conducted at sites off the coast of Alaska (Onstott el al., 1982)
during the spring season, show the variability due to ice conditions for only one
type of sea ice, namely first – year ice. Also shown is a curve representing  of
a pressure ridge, which is akin to a weak “ hard target”. Other measurements
also were reported by Onstott et al. (1982) for first – year and multiyear sea ice at
9 GHz; these fall in the range indicated in Fig. 28. Again, the reader is cautioned
against treating these typical ranges of  as absolute limits of its variability over
all possible conditions and types of sea ice; they are included here merely as
examples of the ranges observed in a measurement program in one location
during one season of the year.

Fig.27 Observed range of variation of 0 for first – year sea ice and one
observation of 0 for a pressure ridge, both at 1.5 GHz
(adapted from Onstott et al., 1982).
 Fig.28 Observed range of variation of 0 for different sea - ice and conditions at
1.5 GHz (adapted from Onstott et al., 1982).

For purposes of comparison, it is worth mentioning that  of lake ice is

usually much lower in magnitude than  of sea ice (at the same radar
parameters), unless the lake – ice surface is covered by a layer of snow, in which
case  increases significantly.

Fig.29 Observed ranges for the backscattering coefficient of several terrain

classes at 9.375 GHz (Newbry, 1961).
Comparison of Target Classes
The Goodyear Aerospace Corporation used an imaging radar operating at
9.375 GHz to obtain backscattering coefficient data for a variety of target classes
(Newbry, 1961). The antenna was designed to provide uniform ground
illumination over a wide range of incidence angles from 20 to 80. Widely
differing terrains representing four geographic areas in five states were studied.
The imaged areas included irrigated farmland in Arizona, forests and meadows in
Minnesota, trees and marshland along the New Jersey coast, mangrove islands
and swamps in the Florida Keys, dry pine forests and grassland in Arizona,
desert regions in both Arizona and California, and the Bristol dry lakebed in
California. The ground targets were identified by aerial photographs taken in
flight, and the radar imagery also was used for verification purposes.
Figure.29 (from Newbry, 1961) shows the range of values of  recorded
by the above system for each of several classes of targets. Terrain with heavy
vegetation cover is limited to a band about 5 dB wide, and the narrow band of
broken desert included the dry grassland of northern Arizona and the desert area
near Chandler, Arizona and Amboy, California. The sandy desert of Yuma,
Arizona contributed to the wide band of arid desert sand.

Fig.30 Angular patterns of the mean, upper decile, and lower decile,
of Skylab’s scatterometer observations over the ocean
during the summer season (Moore et al., 1975).
Average  for the Sea
Radar backscattering from the sea for the angles of incidence beyond
about 20 is governed by the Bragg scattering mechanism described in Section
11 – 5.1, while at near – vertical incidence angles the primary mechanism is a
quasispecular one in which geometric or physical – optics techniques can be
used to develop theoretical results. Because of the wind – speed dependence of
backscattering from the sea (described in Section 11 – 5.1), one cannot describe
average ocean radar scattering coefficients as readily as those for land. The
best way to handle the situation is to use the scattering models.

Fig.31 Angular variation of the mean ocean backscattering coefficient

observed by the Skylab 13.9 GHz scatterometer during the
summer of 1973 (Moore and Fung,1979).

The relative variations of the backscattering coefficient of the sea also

were described in Section 11 – 5.1, but two quantitative examples are presented
here in graphical form. These are based on the Skylab 13.9 – GHz scatterometer
Measurements. Figure. 30 (from Moore et al., 1975) shows angular curves of the
mean backscattering coefficient, , and the upper and lower deciles, 10 and
90, for HH polarization. The curves were generated from histograms of
observations made during the summer, primarily over the North Atlantic ad North
Pacific oceans. The angular dependence of  was fitted by a model of the form
 () = (0) e-/0 (m2 m-2)
and is shown in Fig. 31 (from Moore et al., 1976) for each of the three linear
polarization configurations. The e – folding angle  is observed to vary over a
narrow range between 5.35 and 6.99 among the three polarization
configurations. In contrast to the summer observations illustrated in Fig.31, in
observations made during the winter Skylab mission, which included a larger
number of high – wind cases,  exhibited a different angular dependence in the
0 – 15 region than it did in the 15 – 50 region, as illustrated in Fig. 32 for
vertical polarization. The e-folding angles corresponding to these two angular
regions are 7.4 and 18.1. These overall averages are not of value for wind -
speed determination, but do give an indication of the levels to be expected (for
radar design purposes), as long as adequate account is taken of the great
variation due to wind speed.

Fig.32 Empirical fits to the mean ocean backscattering coefficient observed

by the Skylab 13.9-GHz scatterometer during the winter season. Absolute level
questionable because damage to the antenna changed its gain
(Moore and Fung,1979).
GENERAL CHARACTERISTICS OF TERRAIN AND SEA EMISSION

Unlike the radar backscattering coefficient, whose magnitude can vary in

practice over many orders of magnitude depending upon the type of target being
observed and the set of radar parameters being used, the brightness
temperature of terrain and water is bounded by specific limits. At the low end,
TB = 0 K for perfectly conducting targets if no atmospheric radiation is incident
upon the target from above. At the high end, T B = T0, the physical temperature of
the target if the target is approximately in thermodynamic equilibrium. If the
target is a medium such as soil or water, T 0 is an effective value of the
temperature over the surface layer contributing the majority of the emission as
discussed in Section 4 – 9.
Most radiometric observations of terrestrial surfaces are made at angles of
incidence  between nadir and 60, and for most targets the brightness
temperature over this range either decreases or remains constant with increasing
 for H polarization, and either increases or remains constant with increasing 
for V polarization. For comparing the range of emission characteristics of
different terrain and ocean classes, we shall limit the discussion to nadir
observations. Figure.13 presents in bar – chart form the range of values that the
emissivity covers for each of several classes of targets. Theses ranges are
based on a combination of theoretical models and experimental observations
reported in the literature for frequencies in the 1 – 100 – GHz range, although the
majority of observations are confined to the 1 – 20 – GHZ range and the 35 –GHz
atmospheric – widow frequency. Part of the indicated range is due to variations
with frequency, convolved with variations due to conditions. For example, the
emissivity of smooth sea water varies from about 0.27 at 1 GH z to about 0.5 at
35 GHz, but if the surface of the water is roughened by wind action, and a layer of
foam and whitecaps forms on the water‟s surface, the emissivity can reach
values in the neighborhood of unity.
 Two points of caution should be stated. First, the ranges given in Fig.33
should be regarded in very general terms, and not used as absolute limits, since
the actual limits depend upon frequency. Second, the limits shown are for
emissivity; therefore they should not be used to estimate the corresponding
range for the apparent temperature that a radiometer would observe without
taking into account the downward – emitted atmospheric radiation reflected by
the target and the attenuation and emission characteristics of the atmosphere
between the target and the radiometer. This is particularly important for targets
with low emissivities such as metal objects; whereas the emissivity of a metal
object is zero, the observed radiometric temperature can range from about 10 K
at frequencies around 1 GHz up to about 250 K at frequencies near one of the
atmospheric absorption lines. Of course, radiometers are usually not used for
terrestrial observations in the absorption bands, but even in the atmospheric
window around 35 GHz, the sky radiation reflected by a metal object can be 30 K
or higher, depending on weather conditions.

Fig.33 Range of values that the microwave nadir emissivity

may cover for various classes of targets.
BACKSCATTERING FROM HARD TARGETS

The term “ hard target” usually is applied to strongly scattering objects, as

distinguished from area – extensive targets. Such objects include buildings,
bridges, power – line towers, vehicles, and other man – made structures.
Scattering from such objects must be described in terms of total radar cross –
section , rather than in terms of a quantity defined per unit area.
The radar signal received from a hard target usually (but not always) is
large compared with that received from the area – extensive background. The
reasons for this are various. Many such man – made objects contain metal
components whose high conductivity makes them good radar scatterers. They
usually have flat surfaces, as with building walls, or smooth, rounded surfaces,
as with automobiles. Many of these flat surfaces are vertical, so that, in
combination with the ground (often paved), they form corner reflectors. Other
corner reflectors abound in structures such as window frames, doorways, and
chimney – roof junctions. Sometimes sloping roofs themselves are at the proper
angle to give nearly specular returns to the radar.
Because their scattering patterns are highly directional, hard targets give
signals that depend more on aspect angle than do signals from typical terrain
surfaces. Thus, a building may appear as a very strong echo in one radar image
and may not appear at all in another image produced at a slightly different aspect
angle. This effect often causes difficulty in the interpretation of images from
areas containing many hard targets.
Large hard targets, such as industrial buildings and ships, often contain
many scattering centers. As a result, a single building or ship may appear as a
cluster of targets rather than as a single target. The interpreter of an image
containing such a cluster must therefore attempt to ascertain the shape of the
target from incomplete information. Use of frequency agility or panchromatic
illumination (Moore and Thomann, 1971) can improve this situation by averaging
out the fading associated with the fine structure of the scattering pattern, but the
cardinal – point effect (discussed in Section 11- 3) due to the directivity of flat
surfaces can only be combated by observing the target from several directions.
 Because of the frequent occurrence of nearly specular echoes from hard
targets, the dynamic range of the echoes is much greater than those for terrain or
even for the sea. Fine – resolution imaging radars often find apparent scattering
coefficients for hard targets that are as high as +40 dB, whereas the largest
values of  observed for terrain at the angles used in imaging are of the order of
10dB. Thus, hard targets often saturate a radar that has been adjusted for
terrain imaging.
By way of illustration, Fig.34 (b) shows a trace of the backscattered power
that was measured by a radar as its antenna beam scanned across the scene of
Fig.34 (a), which consisted of an asphalt runway, having grass on either side,
with a parked car in the middle. The ordinate is labeled  although, strictly
speaking,  only applies to the grass and asphalt. We observe that the average
of grass is about 9dB higher in level than the average of asphalt, and the peak of
the backscattering pattern due to the car is about 35dB higher than that of the
asphalt runway.
Fig.34 (a) Photograph of scanned scene: the ellipse is the area illuminated by the
antenna beam, (b) record of the backscattered power as the beam
scanned across the scene (Stiles et al., 1979).