Anda di halaman 1dari 6

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 33, NO.

1, JANUARY 1995 195

A Comparison of Backscattering
Models for Rough Surfaces
K. S. Chen, Member, IEEE, and Adrian K. Fung, Fellow, IEEE

Abstract-The objective of this study is to examine the ease of 11. SURFACE MODELS
applicability of three scattering models. This is done by consid-
ering the time taken to numerically evaluate these models and This section gives the three-dimensional mathematical ex-
comparing their predictions as a function of surface roughness, pressions for the selected models. The complexity of these
frequency, incident angle and polarization with the moment expressions and how they reduce analytically to known models
method solution in two dimensions. In addition, the complexity are discussed.
of the analytic models in three dimensions and their analytic
reduction to high and low frequency regions are also compared.
The selected models are an integral equation model (IEM), a full A. Phase Perturbation Model
wave model (FWM), and the phase perturbation model (PPM). From [ 1, (31)], the backscattering coefficient is expressed as
It is noted that in three dimensions, the full-wave model requires
an evaluation of a 10-fold integral, the phase perturbation model
requires a 4- and 2-fold integral while the integral equation
model is an algebraic equation in like polarization under single
scattering conditions. In examining frequency dependence of
IEM and PPM in two dimensions numerically, the same model
expression is used for all frequency calculations. It is found that
both the IEM and PPM agree with the moment method solution
from low to high frequencies numerically. The FWM given in
[lo] agrees only with the Kirchhoff model. where
-
I. INTRODUCTION

I N the development of a theoretical model for wave scat-


tering from randomly rough surfaces much effort has been
devoted to broadening the ranges of validity over the classical
Kirchhoff and small perturbation models and to bridging
the gap between these two models. Three models that have
so broadened their ranges of validity are selected for com-
parison with the moment method solution. They are the
phase perturbation model (PPM) [l], [ 2 ] , a full wave model P(K;) = k , ( K i ) / k . (5)
(FWM) [3]-[6], and an integral equation model (IEM) [7],
W(*;)is the normalized spectrum of the surface roughness
[8]. The comparisons are aimed at the case of application
such that
of these models by examining their numerical evaluation
and mathematical complexity. In Section 11, the three model
expressions in three dimensions are summarized and their
reductions in the high and low frequency regions are examined. -
Due to the complexity of some of these models only the In the equations above, K;= (kiz,kiz) is the incident wave
2 2 -

two-dimensional versions of these models are numerically vector, IC is wavenumber, and R d = R I - R2 is position vector
evaluated in Section 111. The relative amount of time taken to representing distance between two points on the surface.
do these numerical computations and the associated stability The PPM is originally derived for scalar waves. In Wine-
problems are discussed. Conclusions are drawn in Section IV. brenner’s dissertation, only the HH polarization case was
derived. Hence, it does not include polarization dependence.
Calculations of the backscattering coefficient using (1) in-
Manuscript received June 25, 1993; revised June 9, 1994. This work was volves a 4-fold integral and a 2-fold integral.
supported by the National Science Council of Taiwan.
K. S . Chen is with the Center for Space and Remote Sensing Research, Next, consider the reduction to known models. In the low
National Central University, Chung-Li, Taiwan, Republic of China. frequency limits where k a is small, one can expand the
A. K. Fung is with the Wave Scattering Research Center, Department of exponential in (1) in a power series and keep up to second
Electrical Engineering, University of Texas at Arlington, Arlington, TX 76019
USA. order terms. Equation (1) can be shown to reduce to the
IEEE Log Number 9406430. small perturbation model in horizontal polarization [ 11. In
0 196-2892/95$O4..OO 0 1995 IEEE
196 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 33, NO. 1, JANUARY 1995

IO IO

0
h
0
3- -10
C
-10
g -20
8 -8- IEM (HH)
U -30 -20
M
'C

-
$ 4 -4- PPM -30
3
0
x EWM (ref 10)
-50 A M M (HH)
m -40
PPM
-60
\,
-70 " " " " ' 1 " " 1 " " " " " " ' -50
0 10 20 30 40 50 60 70 0 IO 20 30 40 50 60 70
Incidence Angle (deg) Incidence Angle (deg)
(a) (b)

IO IO

unction of Incidence Angle h


h
0 = 4.188,ko= 0.6,HH Pol.) 0
3 3
C
5
.- .-
g -IO g -IO
8 8
U U
M

-
1,
M
'6c

-
'6 -20 -20 A IEM (HHI
I SIEM (HH)
-
Y
2
- SPM (HH) 42 -30
- SPM
PPM
(HH)

-30
?
:;,
(L;
,
MM (HHI
, , ,,, I , , ,,I\, , , ,,, , ,,
x
A
F ~ M (ref
MM (HH)
10)

-40 -40
0 IO 20 30 40 50 60 70 0 IO 20 30 40 50 60 70
Incidence Angle (deg) Incidence Angle (deg)
(c) (d)
Fig. 1. Model predictions of roughness behavior as compared with exact moment method simulation ( k L = 4.188); (a) ko = 0.2, (b) k o = 0.4,
(c) k a = 0.6, (d) k o = 0.8,

the high frequency limit, according to [l] the PPM differs where subscripts 1 and 2 denote two points on the surface, and
from the Kirchhoff model by a factor of cos4 8. However, in p , q denote the polarization of the scattered and incident waves,
[2] it is shown that PPM reduces numerically to KM for the respectively. A , is the projected area on the reference plane.
two-dimensional problem. According to [2], the discrepancy Bahar commented in [5] that "the numerical evaluation of the
between the analytical and numerical results is due to the 10-fold integral is, in general, too time-consuming for practical
necessity of retaining higher order terms in the expansion of purpose." In this paper he also stated that "in order to consider
the p function. These terms are responsible for the reduction problems that are numerically tractable (using minicomputers)
of PPM to KM. only one-dimensionally rough surfaces are considered." A
way to simplify (7) is to assume that the surjke slopes are
B. Full-Wave Model independent of the heights [5] and
The full-wave model given in [5, (9)] is

Under these assumptions (7) can be reduced to two double


integrals if the characteristics functions of the rough surface
height have analytic forms.
The reduction of the full-wave solution in the low frequency
region has been discussed in [lo] where it is indicated that
Bahar's analytic argument for the reduction of the full-wave
model to the perturbation model is based on an inconsistent
expansion in small surface height. They further reported that
"for backscattering with shadowing neglected, the full-wave
CHEN AND FUNG A COMPARISON OF BACKSCATTERING MODELS FOR ROUGH SURFACES 197

20 ~ ' " ' ~ " ~ ' ~ " ~ ' ~ ~ ' ~ ~ 1 ~ ' " 1 ~ ' ~
Frequency Behavior of Models I
as Function of Incidence Angle -
h
0
--. (kL=3.14,k0=1.0,HHpol) 1
+
3
ss o
U
8

-
--8- IEM IHHI
M lo
2
' C -20
2 PPM

B -30
x
A
FWM (ref 101
MM ( H H )

-30

-40 -40
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
IncidenceAngle (deg) Incidence angle (deg)
(a) (a)

10
- j
- - . - . -:.,
-
i 1 .j
j
--
A
..................: -:-- 4 .._...........I..................
- -. - :................. .- . . . . . .
: - '
:
c - - i -- -
.........i. ............ . -........3-.+
.................................. i................

4 -_i
........ . ; ..+
0 1 ~ -!.-
:: ,.->
.......... .................................. <......................................................... .<. ............
% . :
.:.
.:

, , I , Frequency
, , , I , Behavior
, ,, , , of Models
I , , Y
...... -1EM (Wl ......i.................... j .................... ?..:..-...._..... i.................
_ _ -KM 2 _ _ - KM
...... as Function of Incidence Angle ........ 'C
spMlw) .......- - . SPM ( W )......
MM IW)
I (kL = 1.57, ko = 0.5, Wpol.)
i

........................................................................................................ .
.
.
..
.
.
..
.
.
..
.
.
..
.
.
..
. .................
-30 1 .....
. ...
i
j
-
-ICY Behavior of Models ........
C-^^..^

as Function of IncidenceAngle
(kL =.....................................................
3.14, ko = 1.0, Wpol.)

-40
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
IncidenceAngle (deg)
IncidenceAngle (deg)
(b) (b)
Fig. 2. Model predictions of frequency behavior as compared with exact Fig. 3. Model predictions of frequency behavior as compared with exact
moment method simulation (kL = 1.57, ku = 0.5; (a) HH polarization, (b) moment method simulation (ICL =3.14, ku =1.0, (a) HH polarization, (b)
VV polarization. VV polarization.

predictions agree exactly with the Kirchhoff results." In re- or


sponse, [5] showed that his full-wave model can reduce to
Kirchhoff and small perturbation model when conditions he
deemed appropriate are imposed. However, we are unable
to reproduce a FWh4 equation from [5] because not all the
symbols are defined in the expression provided.

C. Integral Equation Model


From [7] integral equation model for like- and cross-
polarized back-scattering are

x w(")(Ic
sin o - U ,w)dudw (12)
where 0 is the incidence angle and
I, = (kacos 0 ) n [ 2 n e - ~ ~ ~ f
~ 2sin2 01 (13)
and the plus sign is for ww polarization while the minus sign
n=l
is for hh polarization.

(10) is the Fourier transform of the nth power of the surface


correlation function p. Equation (10) only accounts for single
198 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 33. NO. 1, JANUARY 1995

A. Phase Perturbation Model


From [2] the backscattering coefficient for horizontal polar-
0 ization is
%
L
00

.I &, = k cos2 O e - 2 w [ N z l dx explj2kx sin 91


g -10
8
U x {exP[(k.>2~11(x>l- 11 (15)
J
L+
00
i .................. i

-
(HHI .,.; ...................i.....................
-20 _ - - KM
SPM (HHI 1i .. Fn&ency Behaviorof Models 1
Nz = 2k2cos6 p[(u + ksinO)/k]W(u)du (16)
8
#
- PPM
..:..+
as Function of Incidence Angle - 00

-30 -
- A
FwM (ref.lo)
M M (HH) .
.....

.
W- = 6 . 2 8 . k ~=2.0. HH pol.) i
. 7
J_,
~ ~ ~= ( x 1kcos0
) kp[(u +1csin~)/k]l2~(u)ei"~du.
-40
(17)
0 10 20 U) 40 50 60 70 The calculation using (17) requires a double integral and
Incidence Angle (deg)
a single integral; both integrands vary from non-oscillatory
(a)
to highly oscillatory [2]. The degree of oscillation depends
on the incident angle, surface roughness, and the surface
correlation function. The larger the incident angles, the greater
the oscillation. A Gaussian spectrum may be regarded as the
easiest to compute because it has a narrower bandwidth than
other surface roughness spectra such as a sea spectrum.

......
-
................ ....................

- - - KM
. - SpM
IW

(W)
:.1 _
.................. 1: ........ FGuency
i \

i. \ . j.
Behavior of Models
as Function of Incidence Angle
(w) i , (E= 6 . 2 8 . k ~= 2.0,VVp01.)
....... i ........+. ..........i................... i................. . j..................
B. Full-Wave Model
The FWM backscattering model is developed by Bahar
but not all symbols are defined in his expression [ 5 ] . For a
two-dimensional problem, we adopt the expression given in
reference [ 101 which is for horizontal polarization
.
:
.
,: :

-40 t , l , , I l , , , I , , , , I , , , , l , , , ~ l , , , , l , , , ~ '
0 10 20 30 40 50 60 70
IncidenceAngle (deg)
(b)
Fig. 4. Model predictions of frequency behavior as compared with exact In (18), p(x) is the correlation function and wx =
moment method simulation (kL =6.28, ku =2.0 (a) HH polarization, (b)
VV polarization. 2ksin0,vy = -2kcosO .

C. Integral Equation Model


scattering in like polarization while (12) represents multiple The E M backscattering coefficient in two dimensions is
cross polarized backscattering. given by
Note that (IO) consists of three sums. The reduction of IEM
to small perturbation model (SPM) occurs when ka is small
and only the n =1 terms are kept. At the high frequency end,
only the first sum in (10) is significant and it is equal to the (19)
Kirchhoff term. The other two sums increase slower than the 1, = (kacos 6)n[2ne-k2u2'Os' e f 2 sin2 01.
decrease due to exponential decay. Thus, they approach zero
as Ica tends to infinity. The reduction of (12) to second order (20)
SPM occurs when Ica is small and only the first significant
The use of (19) involves only an algebraic expression which
term is kept [7]. The corresponding reductions for dielectric
is easy to evaluate.
surfaces have been given in [8].
To study the model predictions as surface roughness and
frequency are changed, we compare the backscattering coef-
ficients with exact moment method (MM) simulations [9] for
the incident angles from 0 to 60' using a Gaussian surface
roughness spectrum. One possible criterion for quantitative
111. NUMERICAL RESULTS comparisons is that the total deviation of models from MM
In order to evaluate all three models we restrict ourselves to at all incident angles does not exceed 4 dB. The KM and
two-dimensionalproblems. First, we shall give the backscatter- SPM backscattering coefficients are also plotted here to serve
ing coefficients in two dimensions and then show the computed as references. Fig. 1 is for a fixed parameter kL of 4.188,
results. where L is the surface correlation length, and for a k a
CHEN AND FUNG: A COMPARISON OF BACKSCATTERING MODELS FOR ROUGH SURFACES 199

TABLE I TABLE II
SAMPLES OF CPU TIMEREQUIRED BY MODELS SUMMARY OF MODELCOMPARISONS
(USING CONVEXSUPERMINI-COMPUTER)

FWM
IEM 1.25 0.84 0.95

varying between 0.2 to 0.8 in steps of 0.2. We choose these


sets of parameters because they lie between the regions of D=disagne with MM;A = a p with MM.
validity for the KM and the SPM. Note that only horizontal
polarization is considered here. As can be seen in the figures,
IEM and PPM accurately predict the scattering curve over all IV. CONCLUDING
REMARKS
angles, while the FWM gives a result in agreement with the Comparisons of three surface models for perfectly con-
K h t As lca increases, FWM is closer to MM because the ducting surfaces and HH polarizations are made from the
surface is closer to a Kirchhoff surface when L is comparable standpoint of mathematical complexity and numerical com-
to wavelength. Discussions on the range of validity can be putations. Among the three models the most complex is
found in [9]. It should be emphasized that the FWM used the FWM. From a computational standpoint the simplest
here is from [lo]. Figs. 2 4 present the frequency behavior model is the E M . The PPM is not fully polarimetric at this
of the backscattering coefficient as frequency changes from point. Comparisons for finitely conducting surfaces and other
kL =1.57,ka = O S to lcL =6.28,ka =2.0 in two steps each polarizations should be of interest for future investigations.
time by a factor of 2. Part (a) of each figure is for horizontal
polarization and part (b) is for vertical polarization. In Fig. 2(a) ACKNOWLEDGMENT
the surface parameters are larger than those appropriate for The authors are indebted to the anonymous reviewers for
the SPM. The IEM and PPM closely follow the MM results their penetrating stimulating comments that substantially in-
but FWM is on the high side. Fig. 2(b) shows a comparison fluenced this final presentation.
between IEM and MM for vertical polarization to demonstrate
the polarization dependence of E M . Similar illustrations are REFERENCES
plotted in Fig. 3 where k L =3.14 and ko =1.0. In this case, [I] D. P. Winebrenner and A. Ishimaru, “Application of the phase-
we are in the intermediate frequency region. As we can see perturbation technique to randomly rough surfaces,” J. Opt. Soc. Am,
from the figure, both IEM and PPM continue to match the A., vol. 2, no. 12, pp. 2285-2293, 1985.
[2] S. L. Broschat, L. Tsang, A. Ishimaru, and E. I. Thorsos, “A numerical
MM results for horizontal polarization and FWM also gets comparison of the phase perturbation technique with the classical field
closer to MM and is in close agreement with the Kirchhoff perturbation and Kirchhoff approximations for random rough surface
scattering,” J. Electro Waves Applic., vol. 2, no. 1, pp. 85-102, 1987.
model. Results based on a further increase in frequency to [3] E. Bahar, “Full-wave solutions for the scattered radiation fields from
kL =6.28 and lca =2.0 are indicated in Fig. 4. Here, IEM, rough surfaces with arbitrary slope and frequency,” IEEE Trans. Ant.
PPM, KM and FWM are all in agreement with the results of Propagat., vol. AP-28, pp. 11-21, 1980.
[4] -, “Scatteringcross sections for random rough surfaces: Full-wave
MM. analysis,” Radio Sci., vol. 16, pp. 331-341, 1981.
To test computational efficiency we use the cases in [5] -, “Full-wave analysis for rough surface diffuse, incoherent radar
Figs. l(a), 2(a), and 4(a). All calculations are performed on cross sections with height-slope correlationsincluded,” IEEE Trans. Ant.
Propagat., vol. 39, pp. 1293-1304, 1991.
CONVEX supermini-computer using double precision. The [6] E. Bahar and Y. F. Li, “Scattering cross sections for non-Gaussian
evaluation of the integrals in PPM and FWM are carried out surfaces: Unified full wave approach,” IEEE Trans. Ant. Propagat., vol.
using the Filon algorithm [ 121 which is suitable for cosine-sine 39, pp. 1777-1781, 1991.
[7] A. K. Fung and G. W. Pan, “A scattering model for perfectly conducting
integrals. First consider the case of Fig. l(a) where IEM takes random surfaces, I. Model development,” Int. J. Remote Sens. vol. 8,
1.25 s CPU time, PPM 33.4 s, and FWM 2.11 s to produce. no. 11, pp. 1579-1593, 1987.
[8] A. K. Fung, Z. Li, and K. S. Chen, “Backscattering from a randomly
To obtain the convergence results in Fig. 2(a), IEM takes 0.84 rough dielectric surface,” IEEE Trans. Geosci. Remote Sensing, vol. 30,
s CPU time, PPM 5.04 s, and FWM 1.07 s. When the surface pp. 35&369, 1992.
becomes rougher, as shown in Fig. 4(a), E M takes 0.95 s, [9] E. I. Thorsos, “the validity of the Kirchhoff approximation for rough
surface scattering using a Gaussian roughness spectrum,” J. Acoust. Soc.
PPM 112.9 s, and FWM 2.06 s. It is obvious that as the Am., vol. 83, pp. 78-92, 1988.
surface becomes rougher (Figs. 2(a) and 4(a)) or the surface [IO] E. I. Thorsos and D. Winebrenner, “An examination of the full-wave
slopes becomes smaller (Fig. 1(a)), the computational time method for rough surfaces scattering in the case of small roughness,”J.
Geophys. Res., vol. 96, no. c9, pp. 17107-17121, 1991.
required by PPM increases significantly as compared to IEM [ I l l E. Bahar, “Examination of full wave solutions and exact numerical
and FWM. It should be noted at this point that by introducing results for one-dimensional slightly rough surfaces,” J. Geophys. Res.,
vol. 96, no. c9, pp. 17123-17131, 1991.
appropriate asymptotic expansions, simplifying the evaluation [ 121 M. Abramowitz and I. Stegun, Handbook of Mathematical Functions.
of the PPM in the high roughness limit are reported in [ 131. New York: Dover, 1965.
Table I summarizes the computational efficiency. Finally, we [I31 K. Ivanova, M. A. Michalev, and 0. I. Yordanov, “Study of the phase
perturbation technique for scattering of waves by rough surfaces at
summarize our observations in Table I1 to indicate the major intermediate and large values of the roughness parameter,” J. Electro.
properties of the models. Waves Applic., vol. 4, no. 5, pp. 401414, 1990.
200 IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING, VOL. 33, NO. 1, JANUARY 1995

K. S. Chen (M’94) graduated from National Taipei Adrian K. Fung (S’6(rM’6&SM’7(rF‘85) was
Institute of Technology, Taipei, Taiwan and received bom December 25, 1936, in Liuchow, Kwangsi,
the Ph.D. degree in electrical engineering from the China. He received the B.S.E.E. degree from Taiwan
University of Texas at Arlington in 1990. Provincipal Cheng Kung University, the M.S.E.E.
From 1985 to 1990, he was with Wave Scat- degree from Brown University, Providence, RI, and
tering Research Center at the University of Texas the Ph.D. degree from the University of Kansas,
at Arlington. He is now an Associate Professor at Lawrence, in 1958, 1961, and 1965, respectively.
the Center for Space and Remote Sensing Research, He was a faculty member in the Electrical En-
National Central University, Taiwan. His major re- gineering Department of the University of Kansas
search has been in the areas of wave scattering from from 1965-1984, becoming a full Professor in 1972.
terrains and sea, radar signal and image simulation He has been a Professor of Electrical Engineering at
and analysis, and their applications to remote sensing. the University of Texas at Arlington since 1984. He is currently the Director
of the Wave Scattering Research Center. His research interests include wave
scattering and emission from irregularsurfacesand random media, radar image
simulation, numerical simulation of radar scattering, and radome analysis.
He is the author of Microwave Scattering and Emission Models and Their
Applicarions and co-author of a three-volume book on microwave remote
sensing.
Dr. Fung is a recipient of the 1987 Halliburton Excellence in Research
Award, the 1989 Distinguished Research Award from the University of Texas
at Arlington, and the 1989 Distinguished Achievement Award from the IEEE
Geoscience and Remote Sensing Society. He is a member of Sigma Xi and
U.S. Commission F of the Intemational Scientific Radio Union.

Anda mungkin juga menyukai