90 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO.

I , JANUARY 1994

A Terrain Parabolic Equation Model

for Propagation in the Troposphere
Amalia E. Barrios

Abstract- A method to model tropospheric radiowave propa- Recently, a finite-difference radiowave propagation model
gation over land in the presence of range-dependent refractivity for arbitrary terrain paths, called FDPEM (Finite-Difference
is presented. The Terrain Parabolic Equation Model (TPEM), is Parabolic Equation Model), has been developed 161. Another
based on the split-step Fourier algorithm to solve the parabolic
wave equation, which has been shown to be numerically efficient. PE model for propagation over terrain has been developed
Comparisonsbetween TPEM, other terrain models (SEKE, GTD, by Marcus 171, incorporating a hybrid finite-differencehrface
FDPEM), and experimental data show predominantly excellent Green’s function. Two PE terrain models currently exist that
agreement. TPEM is also compared to results from an experi- use the split-step algorithm. One is by McArthur 181, which
ment in the Arizona desert in which range-dependent refractive is based on similar principles of what will be presented in
conditions were measured. Although horizontal polarization is
used in the implementation of the model, vertical polarization is this paper. His approach, however, differs slightly-more will
also discussed. be said about this in Section 111. The other PE terrain model
was developed by Ryan 191. Though the model presented here
I. INTRODUCTION was developed independently, it is theoretically equivalent
to Ryan’s model. However, there are differences in imple-
0 SUCCESSFULLY model tropospheric radiowave prop- mentation which, it is assumed, will give somewhat different
agation over terrain, it is necessary to properly estimate results. These differences will be touched upon in Section 111.
the effects of reflection, refraction, and diffraction. Many Comparisons with the above two models will not be given
models currently exist that use a combination of spherical earth since results were not available.
diffraction, multiple knife-edge diffraction, wedge diffraction, The model presented here, called TPEM (Terrain Parabolic
and geometrical optics to arrive at a solution for the field for Equation Model), is based on a modification to the smooth
ii given transmitterlreceiver geometry and a specified terrain
earth PE and uses the split-step Fourier algorithm [lo], [ 111.
path. One model, called SEKE (Spherical Earth Knife Edge), This is a numerically efficient model because of the use of
was developed at Lincoln Laboratory [l]. This model is fast Fourier transforms (FITS) in its implementation. Since
based on the assumption that the propagation loss over any only a minor modification to the smooth earth PE is required
path (in the frequency range from VHF to X-band) can be to include terrain effects, a brief description of the derivation
approximated by one of the multipath, multiple knife-edge and implementation will be given in Sections I1 and 111,
diffraction, or spherical earth diffraction losses alone, or a respectively. Validation of TPEM will be given in Section IV
weighted average of these three basic losses. Another model using comparisons with some of the aforementioned models
is, based on the geometrical theory of diffraction (GTD) 121 and measured data.
and works by determining what ray paths exist, for a given
heightheceiver geometry and terrain profile, from a family of
16 ray types. The total field at the target is then found by 11. TERRAIN MODEL
adding the ray amplitudes from each possible ray. While these In the following formulation, the atmosphere is assumed
models may adequately account for reflection and diffraction, to vary in range and height only, making the field equations
they lack a proper accounting for range-dependent atmospheric independent of azimuth. Also, there is an assumed time
environments. SEKE allows for a variable earth radius factor, dependence of e-iwt in the field components. We begin with
but this assumes a constant gradient and horizontal homogene- the parabolic wave equation for a flat earth [ 5 ] , [12]:
ity. The GTD model will allow inhomogeneous environments
but problems still arise in regions of caustics.
For many years now the parabolic equation (PE) method has
d2?b(X,
dz2 + d?b(Xc, 2 )
2ik,----
dX

h:en used to model radiowave propagation in the troposphere +k,2[n2(X,z ) - l ] $ ( X , z ) = 0 ( I )

for over-ocean paths 131. The biggest advantage to using the PE
method is that it gives a full-wave solution for the field in the where k , is the free-space wavenumber, TL is the index of
presence of range-dependent environments. Two methods may refraction, $ represents a scalar component of the electric field,
be used to solve the PE. One uses finite-difference techniques and x and z are the spatial Cartesian coordinates corresponding
[4],and the other uses the split-step Fourier algorithm 151. to range and height, respectively.
The field from either a horizontal or vertical electrical
Manuscript received April 16, 1993; revised August 12, 1993.
The author is with NCCOSC RDTE DIV 543, San Diego, CA 92152-7385. dipole source satisfies the same parabolic differential equation
IEEE Log Number 921503. (1). The type of source one wants to model determines the
0018-926W94$04.00 Q 1994 IEEE
BARRIOS: PROPAGATION IN THE TROPOSPHERE
boundary condition that is applied at the earth's surface. For
the present, only horizontal polarization will be addressed.
More will be said about the vertical polarization case in
Section V. For a perfectly reflecting surface, the horizontal
polarization boundary condition is $ ( x , z ~ ( z )=) 0, where
z ~ ( x represents
) z = T ( x ) and T ( x ) is a general height
function describing the terrain. The application of the above
boundary condition to an arbitrary terrain height function
adds to the complexity, or difficulty, of the problem when
propagating over terrain. The fact that the boundary condition
is range-dependent makes a straight-forward solution very
difficult.
A transformation is made according to the method first
presented by Beilis and Tappert [13], in which they used this
technique to model rough surface scattering for underwater
acoustic fields. The original coordinate system is transformed
such that a simpler boundary condition is obtained in the
new coordinate system and a new PE is derived. This same
method was arrived at independently by Abarbanel [ 11 and the
concept of transforming the coordinate system was introduced
years earlier by Neviere, et al. [15]. The method by Beilis
and Tappert will be briefly described here, but the reader is
referred to [13] for a more detailed derivation.
We now introduce a change variables. Let the new height
and range variables be represented by
x = x
c =z - T(z)
where
X2
T ( x )= t ( x ) - -
2a
and define the scalar component of the field in terms of the
new coordinate system
~ ( zz ,) = ~ ( xc ) ,e i B ( x l C ) .
The function t ( x ) describes the actual terrain and can be any
digitized set of heighurange points. x2/2a (where a is the
earth's radius) takes into account the earth's curvature. (3)
is then substituted into (1) to obtain (omitting arguments for
simplicity)
l32Q a i a*
t 2 i ( g +ko& +2ik,- ax
ai2
~
+ [kz(d- 1) - 2k,-at? - 2k,--i36 a<
as ax
826
+ 2- -
as2
($)
dX
'1
* = 0.
The objective is to now recast (4) so that it has the same
form as (1). Certain conditions are imposed such that terms
involving first derivatives in 9 and 6 are made to disappear.
We start by equating the two terms in the coefficient of a9/a<
in order to eliminate this first derivative term. This yields a
form for 6 which we can now substitute into the derivative
91
terms in the bracketed quantity. We end up with the modified
PE
where
from (3), with the term in brackets representing the final form
for 6. The refractive index n in (5) is written as a function
of the new height and range variables. T " ( x ) represents the
second derivative with respect to 2:. Using (2), (5) can be
written as
@ W x , 5)
as2 + 22k0-d QdX( x , i)
+ k , 2 [ 4 C+ T ( X ) )- 1
{
- 25 t"(X) - -
31 9 ( x , C) = 0.
Comparing (7) above with [lo, eq. (27)], one can see that the
(7)
only difference between the two equations is the additional
term 2ct"(x). Therefore, to take into account terrain effects
one only needs to include the second derivative of the terrain
with respect to range. The new boundary condition for (7)
becomes Q ( x , 5 = 0) = 0. With this simpler boundary
condition the problem becomes easier to solve and, in fact, can
be solved by the same split-step PE algorithm as described in
references [3], [ 5 ] , [lo].
(2)
111. IMPLEMENTATION
A. Split-Step Algorithm
Following the method described in [IO], the split-step PE
(3) algorithm can be written from (7) as
9 ( x+ ax,C) = e"o*X["-6~(C)-Ct"(X)IS--1
. { e i A X ( e - k " ) S { q ( x5))).
, (8)
The field ,J,(~, is used to detemline the field at the next
incremental range step, A x . Qualitatively, (8) represents the
aQ field propagated in free space over a range interval A x ,
then attenuated, or modulated, by the actual environment. The
refractive index is included in the modified refractivity unit,
or M-unit, defined by
(4)
M ( < )= I. + kJ
-I
*_I
x 106.
The transform variable p is a function of the propagation
- _ -
angle 8 : p = IC, sin 8. 8 is defined as the angle relative to
the horizontal. The bracketed quantity ([ 1) containing the
second derivative of the terrain can be considered as equivalent
to defining a new modified refractivity. Conceptually, the
modified refractivity, which normally accounts for the earth's
curvature, is now used with an additional term to take into
92 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. I , JANUARY 1994

iiccount the radius of curvature of each segment ( A x ) of the the equation can no longer be simplified such that a new
lerrain. The second derivative, t ” ( x ) ,is determined using the Helmholtz equation results [13]. The fact that this model
second-order central difference formula with the range interval works extremely well using the wide-angle propagator, when
corresponding to the PE range step. compared to other models and measured data (as will be
Referring back to the approach by McArthur, his method shown shortly), may lead one to assume that this term can
consists of adjusting the field in transform, or angle ( p ) , space be considered very small so as to be ignored. No detailed
while the method presented here adjusts the field in height (C) analysis has been done here to see if this is the case.
space. That is, in McArthur’s method, slopes are modeled by The numerical implementation of (8) is fairly straightfor-
“tilting” the field via the Fourier shift theorem, so the wave ward. The Fourier transform is implemented using a mixed
front remains perpendicular to the boundary, which has the radix FFT algorithm developed by Bergland [ 171, combined
effect of flattening the surface and skewing the beam direction with algorithms for sine and cosine transforms by Cooley,
[ 81. Whereas here, flattening of the boundary is done explicitly Lewis, and Welsh [18]. A cosine-taper (Tukey) window [19]
be the change of variables described in Section 11, and in is used to attenuate the field smoothly at large heights and
which beam direction is now affected by t { ‘ ( x )in the resulting large propagation angles that are near the maximum set by
modified refractive index. , , ,C and p,,. In the implementation by Ryan, no filtering
The forward and inverse Fourier transforms are defined by is done in this “buffer” region, but a complex absorber, or
“sponge,” is included in the split-step algorithm [9] to satisfy
-
the radiation condition.
-5max

B. Initial Field
+
As (8) shows, the field at range x A x is dependent on the
field at the previous range step. Therefore, one must begin with
a source field at range 0 in order to propagate the field forward.
where here, the transforms are written in continuous form, but
From (6), the source field in the new coordinate system is
with limits of integration corresponding to the “bandlimits”
written in terms of the source field for propagation over a flat
placed upon C and p (since the discrete Fourier transform,
by way of the FFT, is actually used). , , ,<and p,, are
earth
determined by Nyquist’s criteria: Cmaxpma,=7rN, N being the
transform size.
The free-space propagator, which is the exponential term The source field is easily determined using image theory and
Fourier transform properties. One first begins with noting
in the bracketed quantity ({ }) in (S), is the wide-angle
propagator originally developed by Feit and Fleck [16]. The that the field at range 0 is essentially the antenna aperture
distribution, and that the far-field antenna pattem and its
wide-angle propagator is used here because of the necessity
aperture distribution are a Fourier transform pair
of including large angles when propagating over terrain. The
propagation angle 0 becomes large for the terrain case because
the slope of the terrain needs to be accounted for. The steeper
tlhe terrain, the larger the propagation angle upon reflection. Applying the boundary condition that the field vanish at the
A special note should be made here on the use of the surface, we use image theory to obtain
wide-angle propagator. As described in [lo], the derivation
of the split-step algorithm from the PE of the form of (7),
$(O, Z) = A(. - 20) - A*(z + ZO)

leads to the standard propagator. Computationally, the wide-
angle propagator is no more time-consuming than the standard
propagator, and it has the advantage that it is more accurate at
higher angles. However, there is no derivation known to this
author to obtain the wide-angle propagator from the PE of
the form of (7). The wide-angle propagator is derived from
an approximation of the “pseudo” operator resulting from
factoring out the elliptical wave equation. In order to derive
diis propagator one must begin with the elliptical Helmholtz
equation [5], [16]
where 11, is written as the sum of the source and image fields,
and zo represents the antenna height. Since the antenna pattern
is what is normally dealt with, one can simply transform (9)
to obtain
Two antenna patterns are currently used in TPEM. One is
a truncated omnidirectional pattern [ f ( p )= 11-truncated since
0 only extends to the propagation angle specified by the user
and the maximum angle corresponding to the largest slope of
the terrain profile. The other is a normalized Gaussian antenna
pattern:

and follow the same procedure as in Section I1 to obtain

a “modified” Helmholtz equation. From this new equation where
the split-step algorithm using the wide-angle propagator can
then be derived. The problem in doing this is that a second
derivative term in Q, coupled in x and C, appears so that
 ;
BARRIOS: PROPAGATION IN THE TROPOSPHERE 93

Standard Atmmrhm - k i s c k a N55 U35

888
FRB9 llltr 435 Fnm f k 167
POMRIZAYIOW HOR WLARlwlTlOll HOR
lAM HT 18.3 TRAN HI 18.3
RANGE kn 54.5 IN6B la 35.5
ANT TYPE OM1 lull ? W E OWN I
umw dcg WA UER BU deg N/A
KLEU AN6 leg N/A ELEU AN6 deg w n

-Ipm
328 xxxxxxx SENE

16-38

7
-28 -18 E 18 28 n 28
P R O ~ T I O NFACTOR dB FREE SPACE - - -- PROPRGAIIWI FETOR dB FREE SFME -

8
I 000
11 22 33 11 55 8
Ranma km 36
R a w - km

Fig. 1. Terrain profile for north (55 h)Beiseker Pa* With height-gain Plot Fig. 2. Terrain profile for west (35 h)Beiseker path with height-gain plot
showing TPEM, SEKE, and measured signal. showing TPEM, SEKE,and measured signal.

and Obw is the 3 dB beamwidth. To include elevation angles
one can simply replace f(p) with f(p - p , ) , where p , =
IC, sin O e l e v .
C. Environment
The program accepts height versus refractivity profiles
at specified ranges. At every range step Ax the split-step
algorithm requires a refractivity profile as a function of height.
For a range-independent case only one profile is required. If
several profiles are entered to simulate a range-dependent en-
vironment, then interpolation among the profiles is performed
according to the method described by Barrios [ l I]. In Ryan’s
model the interpolation is done using a bivariate surface fitting
algorithm [20].
The terrain profile is entered in much the same way as the
refractivity profiles. All that is required is a series of data
points corresponding to height versus range to describe the
terrain. As mentioned in Section 11, the ground is considered
to be smooth and perfectly conducting. Therefore, no finite
conductivity results will be presented, in which varying per-
mittivity and conductivity values are used. Also, no effects
due to vegetation or forested areas will be considered. For the
results presented in Section IV, only horizontal polarization
will be addressed.
IV. VALIDATION
In this section all measurement and prediction results will
be displayed as height vs. one-way propagation factor (field
strength relative to free space) in dB. “Height” within these
plots refers to height above the ground at the particular receiver
range shown.
Propagation measurements were made over several sites in
Canada by Lincoln Laboratory [l]. Comparisons will be pre-
sented for one site in particular, the Beiseker area in Alberta,
Canada. The terrain is considered to be intermediate rolling
farmland with negligible vegetation. A standard atmosphere
of 118 M - u n i t s h for TPEM was used and a 4/3 earth radius
factor was used for SEKE. Fig. 1 shows the 55 km north
terrain profile (Beiseker N55) along with the height-gain plot
comparing SEKE, TPEM, and the propagation measurements.
The frequency is 435 MHz, the transmitter height is 18.3 m
above the ground, and the receiver range is 54.5 km. For this
case TPEM and SEKE agree fairly well with the measured
data. Fig. 2 shows the same comparison for a frequency of 167
MHz along the 35 km west path (Beiseker W35). Here, both
TPEM and SEKE agree well with the data, however, TPEM i s
also able to capture the multipath pattern at the higher altitudes.
Validation of the GTD model was made using the VHF
propagation measurements taken at Gardner, MA [21]. Mea-
surements were made at low altitude over hilly, forested
terrain. The frequency used was 110.6 MHz with the antenna
located at one half-wavelength (approx. 1.4 m) above a ground
plane located 4.6 m above the ground. Fig. 3 shows the Natty
Pond terrain profile, along with the height-gain plot displaying
the measured data with predictions by TPEM and the GTD
model. The receiver range for this case is at 6.6 km with the
environment being standard atmosphere for TPEM, and 4/3
earth radius for the GTD model. Both models agree very well
with the measured data.
94 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. 1, JANUARY 1994

East finglla, Standard Mwsphare

488 TPM uer: 1.2

WUIRIZMION HOR
M T HT ta 45.8
3a M T TYP @WS
H UER BU des 3.8
e ELU &NG dag 8.8
I 40 I
g 248
h
t

e 168

-,
-38 -28 -ie 0
PROPAGATION FACTOR dB FREE SPACE - --- e-
# 10
-- U”

36 54 72 98

Fig. 4. Coverage diagram showing propagation loss contours for East Anglia

i “1
588
terrain path, standard atmosphere.

Eaxt Anglia. Standard Atmosphere

188
P 1 RE9 HHz 2318
R \ POLARIZATION HOR
0 \ IWV(HY1 45
P 114 REC H I i 358
n MITYPE GAUSS
z88k i 4
Rang- k-
6 e
I
10 6
n
WER BU deg
BLEuIy16deg
3
E
1 128
I
Fig. 3. Terrain profile for Natty Pond with height-gain plot showing TPEM, 0
GTD, and measured signal. N
i42
L
0
S
One of the biggest advantages of the split-step PE algorithm S 156
is that it is particularly useful in creating coverage diagrams. d
0
!Since the solution is determined recursively, one can always 178
find the field for any given set of receiver heights at every 1 18 36 54 72 90
WINGE km FREE SPICE - - - -
range increment along the path. An example of such a coverage
diagram is given in Fig. 4. The gray shades correspond to Fig. 5. Propagation loss vs range plot for TPEM and FDPEM at 350 m
different values of propagation loss as shown in the legend. receiver height for East Anglia termin with TPEM offset by 10 dB to
distinguish between the two curves.
Areas in the coverage diagram with no shading (upper left and
lower right comers) are areas in which the loss is greater than
155 dB. The terrain is a 90 km long path in East Anglia, U.K. inherently neglects backscatter. Although not shown here, a
‘The antenna pattern used for this case is the Gaussian pattem, comparison was made for a flat-topped block between TPEM
with the frequency at 2.31 GHz. A comparison with FDPEM and FDPEM and excellent agreement was found.
on a loss vs. range plot is shown in Fig. 5 for a receiver In 1946, the Navy Electronics Laboratory (now Naval
height of 350 m. TPEM has been offset by 10 dB in order Command, Control and Ocean Surveillance Center) conducted
to distinguish between it and the result from FDPEM, since an experiment over the Arizona desert to study atmospheric
the two curves, when overlayed, appear as one. While there inhomogeneity and the irregularity of the terrain [22]. The
may be small differences in detail, TPEM shows excellent transmitter was located at Gila Bend with two receiving
agreement with FDPEM. As an example of the numerical stations along the path. One was located at Sentinel, 26.7
efficiency of the split-step algorithm, the computation time miles away, and the other was located at Datelan, which was
needed to create the coverage diagram for Fig. 4 was just over 46.3 miles from the transmitter. Meteorological measurements
3.5 minutes on a 486/50 MHz IBM compatible PC. For this were made at these stations and at several stations along the
case, a 2048 point transform was used with a range step of path. Meteorological and radio measurements were taken at
aipproximately 180 m. Losses were computed for a maximum regular intervals throughout the day, however, in the following
cif 280 receiver heights at every range step. comparisons, only those measurements made at 0300 on
Notice that (7) does not allow for propagation over vertical February 6, 1946 are used. The reason being that it is during
obstacles, such as cliffs or buildings (t”(x)will be undefined the night and early morning hours that surface-based ducts
for such cases). A special case can be made by simply associated with temperature inversions occur because of the
eliminating the field immediately adjacent to such obstacles cooling of the ground. The terrain profile and the location
and propagating the field as usual. This does not violate any of the measuring stations, along with the refractivity profiles
conditions in the split-step model as the PE approximation are shown in Fig. 6. Each profile contains a small surface-
BARRIOS: PROPAGATION IN THE TROPOSPHERE 95

Path Profile Gila Bend to Sentinel

I
leel FREP WIZ 178
WULRlZATIMl HOR
TIM HT f t SE
RRWEE am 26.7
MT TYPE OM1
VER By delr Wtl

I
I
I
E I
-BE -68 -4E -28 8 2 8
PROPMGRIION FACTOR dB FREE SPACE ----

M - "NITS Gila Bend to k b l a n

!
I
FRgp HHz 178
Fig. 6. Terrain profile from Gila Bend to Datelan showing meteorologi- l I
I
POLARIZATION HOR
cal and receiver stations with heighdrefractivity profiles measured at 0300 it" HI rt SE
I RllNGE U 46.3
February 1946. I MT TYPE olpll
I UEA BY deg wn
I ElEu AWG deg WII
I 1ZE I
based duct. Since the ducts were created by the dissipation of I
heat from the ground, interpolation between the profiles was I
I
performed such that the trapping layers followed the contour I

1
of the terrain. - TPEH 1
OmDDD dserued I
Figs. 7(a) and 7(b) compare TPEM with measured data at I

the Sentinel and Datelan receiving stations, respectively, for a I

frequency of 170 MHz and antenna height of 50 ft. For this - I

-88 -68 -4E -28 E 2 B

case the predictions give an excellent match. Figs. 8(a) and
8(b) show the same comparisons for a frequency of 520 MHz
with an antenna height of 190 ft. Although good agreement is
shown also at this frequency, for the Sentinel station TPEM
slightly underestimates the measurements. Comparisons for
1 GHz are shown in Figs. 9(a) and 9(b). Here, very good
agreement is shown for the Sentinel terminal, but predictions
are greater than the measured data by about 20 dB at some
receiver heights for the Datelan path. At 3.3 GHz, TPEM
again matches measurements very well for the short path
[Fig. 10(a)], but greatly overestimates the measured field by
about 25 dB at some heights for the Datelan path, Fig. 10(b).
The reasons for the large discrepancies in Figs. 9(b) and
10(b) are not known. Though not very well documented,
it was assumed that the antennas used in the experiment
were omnidirectional. Using narrow beam antenna patterns for
TPEM did not improve the comparisons.
V. DISCUSSION
In the determination of the initial field in Section 111-B
it as assumed that the ground was a perfect conductor and
the antenna was horizontally polarized. The assumption of
infinite conductivity is not realistic. However, the comparisons
given in the previous section showed excellent agreement
between TPEM predictions and measured signals. In fact,
both S E E and the GTD model included reflection coefficient
calculations that took into account relative permittivity and
conductivity of the ground. There are two explanations for the
PROPAGATION FACTOR dB FREE SPACE ----
(b)
Fig. 7. Height-gain plot showing TPEM and measured signal for frequency
of 170 MHz, at receiver ranges (a) 26.7 mi and (b) 46.3 mi.
the favorable comparisons between all three models with the
observations. One is simply that for horizontal polarization,
a value of -1 (infinite conductivity) for the magnitude of the
reflection coefficient is a very good approximation for all radar
frequencies from VHF (100 MHz) to K band (20 GHz) for
almost all types of land. In the present application, for land-
based radars, one is usually dealing with small grazing angles.
For frequencies ranging from 100 MHz to greater than 5000
MHz, Reed and Russell [23] show high reflection coefficient
magnitudes for grazing angles up to 10" for several different
ground types, including 0 conductivity.
For vertical polarization and infinite conductivity, mathe-
matically, the boundary condition for (7) becomes aQ(x,0)/
= 0. The infinite conductivity assumption for vertical
polarization, however, is not a good approximation. Reflection
coefficient magnitudes start to fall rapidly at grazing angles
from 2' to IO'. One can apply the Leontovich boundary
condition for finite conductivity:
96 IEEE TRANSACTTONS ON ANTENNAS AND PROPAGATION, VOL. 42, NO. 1, JANUARY 1994

E
a "1
1 CiIa Bond to Scntinsl

I:
WWGE
MI TYPE
UERW
ELEU
HT ft
gm

deg
Iv(6 deg
198
26.7
OnNI
WA
Wh
H

6
I

1
i
PREP
RIv(6E

I1 TYPE
UER W
MI 1000
POlARIzAIlON HOR
TAIW HI em

des
iBB
ft 26.7
WNI
WR

H
T
88-
f
t
I
40-

-m E 28
FREE SPWE

:j
6i1a Bend to Datslan
288 I

i
I
I FREP f k 528 I FREP IYlr 1000
POUIRlZATIMI WR POlARIZIITIR HOR
I I
IRM HT f t 1% I TWVl HT ft 108
wmoo obssrusd I RIIWGE sm 46.3 RMGE em 46.3
I MI IYPE OM1 I PWT TYPE oIp(I
I UER W deg Wh I WBY deg WR
I KLEV ffl6 deg WII I BLEU IG deg N/R
I I
I
ft I
I I
I I
I % I I

-Ipm
I
I
I
I
} I
I
I

I
E! I 1
-BE -68 -40 -28 0 2 8
PROPAGAIION FACTOR dB FREE SMCE --- -
(b)
Fig. 8. Height-gain plot showing TPEM and measured signal for frequency
of 520 MHz, at receiver ranges (a) 26.7 mi and (b) 46.3 mi.

where the split-step algorithm for finite conductivity requires the

CYh 2 ikon, for horizontal polarization
a, E for vertical polarization
n c
Here, nc is the complex refractive index, E is the relative
permittivity, U is the conductivity, and X is the wavelength in
meters. To obtain the boundary condition in the new coordinate
system the same change of variables is used and (3) is
substituted into (9) to obtain
As was done for the parabolic equation, the only modification
to the boundary condition is to the alpha term, or equivalently,
defining a new complex refractive index. One can then use the
modified PE along with the new boundary condition above to
model finite conductivity using the mixed Fourier transform
in the split-step algorithm described in [lo].
The question is now posed: for land-based radars, is this
really necessary? The mixed Fourier transform required in
use of sine and cosine FFTs. This will add to the execution
time and to memory storage requirements. Even for grazing
angles up to 2", the reflection coefficient magnitude for vertical
polarization is very close to that for horizontal polarization
[ 2 3 ] .In tropospheric propagation, angles most affected by the
environment are within 0.5" of horizontal, with angles between
0.5" and 2" somewhat less affected. Microwave radars are
more commonly used for long range (> 10 km) and below line-
of-sight operations where energy is incident and diffracted at
small grazing angles.
To demonstrate the actual differences between the two types
of polarization for over-land propagation, a comparison is
made between TPEM using horizontal polarization, infinite
conductivity, and FDPEM, using vertical polarization and
ground constants taken from the CCIR curves for permittiv-
ity and conductivity for very dry ground. FDPEM correctly
models the Leontovich boundary condition and would give an
accurate prediction using numerical values of permittivity and
conductivity. A frequency of 100 MHz will be used, since it
is at the lower frequencies that the biggest difference between
reflection coefficient magnitudes for horizontal and vertical
polarization arise. The value used for the relative permittivity
 ~

BARRIOS: PROPAGATION IN THE TROPOSPHERE 91

Gila Bend to Sentinel East nwlia, Ducting Rtmorphnra

I
FRE9 I%

4
P WuullZllTIoW 188
UER
I
lwyl Hl m 45

!
0
1 m o ~ o ohrued P RE HI I 288
- m n MT TYPE WSS
6 UKR W dog 4
n KLEUVNC deg 98
1
I
0
n
L
0
S
40 S
d
B
1 I I I I
-€a -68 -498 -28 B 28 f i 1 8 3 6 5 1 n s a
PROPIIGMIWI FWIOR dB FREE SPllCK - - - - Rnw kn IRE SPKE - - - -

6111 Bond to Datelan
IPW
m” obrarwd
I
I
I diffraction the dependence on polarization is very weak for
I
is 3.0, and for the conductivity, 1x S/m. A homogeneous,
low elevated duct was used for the refractivity and the terrain
path is the East Anglia profile shown in Fig. 4. The antenna
height is located within the duct. Fig. 11 shows a propagation
loss vs range plot for a receiver height of 200 m. As was
done with Fig. 5 , the curve for TPEM has been offset by
5 dB in order to distinguish between it and the result from
FDPEM, since they appear as one curve when overlayed on
top of each other. One can see that for all ranges, there is very
little difference between the two curves.
There is a second possible explanation for the favorable
comparisons between predicted fields using the perfect con-
ductor assumption, and measurements. Most rays that strike
the ground reflect at very sharp angles. These rays will undergo
very little refraction because of the very large propagation
angles associated with them. Rays may become reflected twice
or even three times. However, even within a surface based
duct, rays will not continually be reflected as in over-ocean
propagation, but will bounce off the ground and into space.
Therefore, when looking at long-range signals, cumulative ef-
fects from continual reflection (with less-than-unity reflection
coefficient magnitudes) will not be a factor.
Fig. 11. Propagation loss vs. range plot showing TPEM-horizontal polariza-
tion and FDPEM-vertical polarization for East Anglia terrain. TF’EM is offset
by 5 dB to distinguish between the two curves.
IRE9 M BEE
POLIIRIZATIffl HOR
TMtI HT ft 1998 A major concern with propagation over terrain is the de-
WYIGE sm 46.3 tectability of targets in the shadow zone, where targets lie
MI IYPK CM1
U E R I deg WA
ELKU M6 deg W11
beyond an obstacle such as a hill or building. The dominant
mechanism in this case is diffraction. Meeks [24] states
that in the theoretical analysis of diffraction, “for knife-
edge diffraction the results are rigorously independent of
polarization for small diffraction angles,” and for cylinder
permittivity and conductivity values at radar frequencies.
This independence of polarization on signals measured over
land has been reported (many times) experimentally. In a
study of nine over-land experiments ranging in frequency from
43 MHz to 10 GHz, and ranging from flat desert to moun-
tainous terrain, Kerr [25] reports several common features
resulting from the experiments, one being that vertically and
horizontally polarized fields showed similar behavior. Other
experiments leading to the same
result have been reported by Carlson and Waterman [26],
and Englund, et al. [27]. Even for small surface roughness,
where ground perturbations are electrically small, polarization
dependence will be weak [28]. However, polarization may
be a factor when surface perturbations are on the order of
a wavelength.
One last point to be addressed are the limitations within
TPEM. As already mentioned in many of the references
regarding the split-step PE algorithm, transform size is directly
proportional to frequency and propagation angle. When deal-
ing with terrain profiles that consist of relatively steep slopes,
transform sizes (and mn times) become so large that it makes
it quite impractical for an operational model. However, if one
does not need to be concerned with high-angle reflections, the
field can be handled as in the flat-topped block case, wherein
high-angle reflections will be ignored, and propagation angles
can be kept fairly low.
VI. CONCLUSION
A numerically efficient method has been presented to model
tropospheric radiowave propagation over irregular terrain in
the presence of range-dependent nonstandard environmental
98 IEEE TRANSACTIONS ON ANTENNAS AND PROPAGATION, VOL. 42. NO. 1, JANUARY 1994
conditions. While perfect conductivity and smooth surface
were assumed, results from this model were compared against
measured data and other existing models and were shown to
give predominantly excellent agreement.
Although a method was outlined to model vertical polariza-
tion and finite conductivity with the split-step PE algorithm,
doing so will needlessly create a larger and slower model.
It: has been shown that there is very little difference in
polarizations for transmittedreceiver geometries on land and
separated by relatively large distances. Since the final objective
in model development is to produce a real-time capability for
predicting signal levels for operational assessment, whether
it be for military or civilian requirements, the assumption
of horizontal polarization, perfect conductor, for land-based
transmitters and receivers should be adequate.
ACKNOWLEDGMENT
The author wishes to thank Dr. Kenneth Craig from Ruther-
ford Appleton Laboratory for supplying the East Anglia terrain
and environmental data.
REFERENCES
[ I ] S. Ayasli, “SEKE: A computer model for low altitude radar propagation
over irregular terrain,” IEEE Trans. on Antenna Propagat., vol. AP-34,
no. 8, pp. 1013-1023.
[2] R. J. Luebbers, “Propagation prediction for hilly terrain using GTD
wedge diffraction,” IEEE Trans. Antennas Propagat., vol. AP-32, no. 9,
pp. 951-955, Sept. 1984.
[3] H. W. KO, J. W. Sari, and J. P. Skura, “Anamolous microwave
propagation through atmospheric ducts,” Johns Hopkins APL Tech. Dig.,
vol. 4, no. 2, pp. 12-26, 1983.
[4] S. T. McDaniel, “Propagation of a normal mode in the parabolic
approximation,” J. Acoust. Soc. Am., vol. 57, no. 2, pp. 307-31 1, 1975.
[ 5 ] F. D. Tappert, “The parabolic approximation method,” in Wave Propa-
gation and Underwater Acoustics Vol 70 (Lectures Notes in Physics), J.
B. Keller and J. S. Papadakis, eds. New York: Springer-Verlag, 1977,
pp. 224-287.
[6] M. F. Levy, “Parabolic equation modelling of propagation over irregular
terrain,” presented at the Seventh Intemat. Con5 Antennas Propagat.,
(ICAP Si), U.K., April 15-18, 1991.
171 S. W. Marcus, “A hybrid (finite difference-surface Green’s function)
method for computing transmission losses in an inhomogeneous atmos-
phere over irregular terrain,” IEEE Trans. Antennas Propagat., vol. 40,
no. 12, pp. 1451-1458, Dec. 1992.
R. J. McArthur, “Propagation modelling over irregular terrain using the
split-step parabolic equation method,” IEEE Int’l Conf: Radar ’92, no.
365, pp. 54-57, Oct. 12-13, 1992.
F. J. Ryan, “Analysis of electromagnetic propagation over variable
terrain using the parabolic wave equation,” Rep. NOSC Tr 1453, Oct.
1991.
J. R. Kuttler and G. D. Dockery, “Theoretical description of the
parabolic approximatiodFourier split-step method of representing elec-
tromagnetic propagation in the troposphere,” Radio Sci., pp. 38 1-393,
Mar-Apr. 1991.
[l 11 A. E. Barrios, “Parablic equation modelling in horizontally inhomoge-
neous environments,” IEEE Trans. Antennas Propagat., vol. 40, no. 7,
pp. 791-797, July 1992.
[12] V. A. Fnck, Electromagnetic and Propagation Problems. New York:
Pergamon, 1965, ch. 11-14.
[13] A. Beilis and F. D. Tappert, “Coupled mode analysis of multiple rough
surface scattering,” J. Acout. Soc. Am., vol. 66, no. 3, pp. 811-826,
Sept. 1979.
[14] H. D. I. Abarbanel, “Scattering from a random surface,” J. Acoust. Soc.
Am., vol. 68, no. 5, pp. 1459-1466, Nov. 1980.
[15] M. Neivere, M. Cadilhac, and R. Petit, “Applications of conformal
mappings to the diffraction of electromagnetic waves by a grating,”
IEEE Trans. Antennas Propagat., vol. AP-21, no. 1, pp. 37-46, Jan.
1973.
[16] M. D. Feit and J. A. Fleck, Jr., “Light propagation in graded-index
fibers,” Appl. Opt., vol. 17, pp. 399G3998, 1978.
1171 G. D. Bergland, “A radix-eight fast Fourier transform subroutine for
real-valued series,” IEEE Trans. Audio Electro-Acoust., vol. AU-17, pp.
138-144, 1969.
[18] J. W. Cooley, P. A. W. Lewis, and P. D. Welsh, “The fast Fourier
transform algorithm: Programming considerations in the calculations
of sine, cosine, and Laplace transforms,” J. Sound Kb., vol. 12, pp.
315-337, 1970.
[I91 F. J. Harris, “On the use of windows for harmonic analysis with the
discrete fourier transform,” Proc. IEEE, vol. 66, pp. 51-83, Jan. 1978.
[20] F. J. Ryan, “User’s guide for the VTRPE (variable terrain radio parabolic
equation) computer model,” Rep. NOSC TR 1456, Oct. 1991.
[21] M. L. Meeks, “VHF propagation over hilly, forested terrain,” IEEE
Trans. Antennas Propagat., vol. AP-31, no. 3, pp. 483489, May 1983.
[22] J. P. Day and L. G. Trolese, “Propagation of short radio waves over
desert terrain,” NEL Rep. 149, Nov. 3, 1949.
[23] H. R. Reed and C. M. Russell, Ultra High Frequency Propagation.
New York Science P, 1964, ch. 4.
[24] M. L. Meeks, Radar Propagation at Law Altitudes. New York: Artech,
1982.
[25] D. E. Kerr, Propagation of Short Radio Waves. MlT Radiation Labo-
ratory Series. New York: McGraw-Hill, 1951, ch. 4.
[26] A. B. Carlson and A. T. Waterman, Jr., “Microwave propagation over
mountain-diffraction paths,” IEEE Trans. Antennas Propagat., vol. AP-
14, no. 4, pp. 489-496, July 1966.
[27] C. R. Englund, A. B. Crawford, and W. W. Mumford, “Ultra-short-wave
transformation over a 39-mile optical path,” in Proc. I.R.E., Aug. 1940.
pp. 36G369.
[28] J. R. Wait, “Comment on ‘Microwave propagation over mountain-
diffraction paths’,’’ IEEE Trans. Antennas Propagat. Comm., p. 321,
1967.
Amalia E. Barrios was born in Mendota, Califor-
nia, in 1962. She received the B.S. degree in physics
from California State University, Fresno, CA, in
1983 and the M.S. degree in electrical engineering
from the University of California, San Diego, CA,
in 1989.
In 1983 she joined the Research and Develop-
ment, Test and Evaluation Division of the Naval
Command, Control and Ocean Surveillance Center
(formerly, Naval Ocean Systems Center) in San
Diego, Ca. She has been developing parabolic equa-
tion techniques to improve current modeling of tropospheric radiowave
propagation over variable terrain and is currently looking at methods to
account for rough surface effects.