A Site-Specific Model of Radar Terrain Backscatter and ...

C. C. LIN AND J. P. REILLY

A

A Site-Specific Model of Radar Terrain Backscatter andShadowing

Christopher C. Lin and J. Patrick Reilly

site-specific approach is presented to characterize terrain and target visibilityand terrain clutter as seen by a shipboard radar in a coastal environment. The methodtakes into account the location of the ship, the particular terrain topography, the radarparameters, and the propagation effect. The method incorporates atmospheric refrac-tive index conditions surrounding the radar, an optical ray-trace model, an electro-magnetic parabolic equation model, a database of terrain elevations, and a cluttermodel. The model can simulate illuminated and shadowed regions of both surfaceclutter and elevated targets. Simulated clutter results are shown to compare favorablywith clutter data measured at S-band, X-band, and Ku-band. This correspondence isevident in geo-graphic patterns and statistical distributions of clutter on directlyilluminated terrain surfaces.(Keywords: Littoral model, Radar clutter, Radar modeling, Radar propagation, Terraineffects.)

INTRODUCTIONA shipboard radar in a coastal region is subject to

complications not encountered in the open ocean.These complications include terrain shadowing of airtargets over land and clutter returns from terrain. Ter-rain shadowing occurs in geographic patterns dictatedby specific terrain contours, resulting in regions ofattenuated radar signal strength that can compromisedetection and tracking. Terrain clutter can obscure atarget’s signal, even when it is not in shadow as wellas the clutter at a range distant from the target. Both

432 JOH

terrain shadowing and clutter are significantly affectedby the characteristics of electromagnetic propagation,which can be complicated in the coastal region.

Existing models for land clutter are typically empir-ical. For instance, tables and empirical formulas areavailable in which reflectivity for generic types of ter-rain is related to various radar parameters.1–6 Typically,some statistical measure is given, such as mean ormedian reflectivity (s0), often along with parametersof some statistical model. Such models suffer from a

NS HOPKINS APL TECHNICAL DIGEST, VOLUME 18, NUMBER 3 (1997)

RADAR TERRAIN BACKSCATTER AND SHADOWING

number of deficiencies, including the inability to accountfor site-specific terrain features, geographic patterns ofclutter, propagation conditions, or target shadowing.Other researchers have developed models that accountfor site-specific terrain features,7,8 but these models failto account for arbitrary propagation conditions and donot represent target shadowing.

The model described here overcomes these deficien-cies. Features of the model include three-dimensionalrefractive index specifications, an optical ray-trace model,an electromagnetic parabolic equation model, and adatabase of terrain elevations called DTED (DigitalTerrain Elevation Data), which is published by theDefense Mapping Agency. The database provides ter-rain elevations on a 100-m grid for much of the Earth’sland mass. The model can accept refractivity inputsthat vary in three dimensions if such detailed data areavailable. Alternatively, a single profile of refractiveindex versus altitude may be used to represent a uni-formly stratified atmosphere, i.e., a condition in whichthe refractive index versus altitude is constant over thearea covered by radar.

The terrain effects model is currently configuredwith various levels of complexity and fidelity. A TerrainVisibility Routine (TEVIR) calculates regions of theterrain or of the air space above the terrain that aresubject to direct illumination by the radar. TEVIR-Iperforms such calculations for an atmosphere charac-terized by a linear, nonducting refractivity profile, ofwhich the standard atmosphere is a specific case.TEVIR-I performs calculations using straight-line raytrajectories over a round Earth having an equivalentEarth radius dictated by the refractivity slope. For thestandard atmosphere, the equivalent Earth radius is afactor of 1.33 greater than the true Earth radius.TEVIR-II can use arbitrary refractivity profiles, includ-ing profiles that vary in both range and azimuth fromthe radar location. TEVIR-II makes use of an opticalray-trace routine. A third variant called the RAD-SCAT (radar scattering) model determines propagationeffects using an electromagnetic parabolic equationmethod. This method involves a full-forward-wave cal-culation of the electromagnetic field. The parabolicequation is numerically solved in RADSCAT by theFourier split-step method, using the computer modelcalled TEMPER developed at the Applied PhysicsLaboratory.9,10 The RADSCAT model with TEMPERprovides numerical calculations of the propagationfactor and accounts for radar refraction, multipath, anddiffraction. The more complex RADSCAT model isused to calculate quantitative radar clutter data. LikeTEVIR-II, RADSCAT can accept three-dimensionalrefractivity inputs.

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 18, NUMBER 3 (

VISIBILITY MAPS

General ConsiderationsWe determine terrain visibility by one of two

methods: an optical ray-trace program or an electro-magnetic parabolic equation model (TEMPER). Inboth methods, electromagnetic energy is assumed to bereflected from water surfaces but absorbed by terrain.

The optical ray-trace (geometric optics) methodintegrates propagation differential equations along a raytrajectory. In the ray-trace method, rays are launchedat closely spaced elevation angles. Typically, both pos-itive and negative angles are included; the negativeones are reflected from the water surface. The range ofangles needed to simulate radar visibility depends onthe beamwidth of the radar, the maximum altitude ofterrain peaks, and the range of the terrain peaks to theradar. In a typical simulation of a shipboard radar, maxi-mum ray angles of ±2° are usually sufficient to com-pletely account for all rays that intersect the terrain.The trajectory of each ray is calculated for a particularrefractivity profile. Once a ray encounters a land surface,that surface is considered illuminated for subsequentranges along an equal azimuth slice until the terrainslope becomes negative. For ranges beyond the negativeterrain slope, the terrain is considered to be in shadowuntil another ray intersects the terrain surface. By re-peated application of this algorithm, one can identifydirectly illuminated regions along a radial slice; withslices along various azimuths, one can identify directlyilluminated areas of terrain. The accuracy of the result-ing plot improves as we increase the density of rayswithin the elevation launch angles and as we increasethe density of azimuth slices. For a typical application,we obtain satisfactory results with elevation ray spacingof about 0.02° and azimuth spacing of about 0.5°.

The number of ray calculations can become quitelarge. For example, using ±2° elevation coverage with0.02° spacing and 90° azimuth coverage with 0.5° spac-ing, there are 36,000 ray trajectories to be computed ifthe refractivity profiles are unique along each azimuthslice. For a uniformly stratified atmosphere, however,the number of calculated rays need be only 200 by usingthe same set of trajectories at each azimuth angle.

One can often evaluate the potential severity ofterrain clutter problems by examining a terrain visibil-ity map showing the regions of terrain that are directlyilluminated (i.e., not in shadow). Often the distribu-tion and extent of visible terrain are sufficient to es-timate the probable impact of clutter. In such cases, aray-trace solution obtained by TEVIR-I and -II may besufficient for a qualitative radar performance evaluation.

1997) 433


In this article, we will examine propagation effectsusing the refractivity profiles shown in Fig. 1. ProfilesA and B were measured by the Pacific Missile TestCenter off the coast of California over a 3-day periodin June 1990. Strong surface-based ducts were persis-tent during this period. Profile C was measured off thecoast of California by APL personnel during the sum-mer of 1992. Surface-based duct heights of 500, 600,and 1000 ft apply to profiles A, B, and C, respectively.The profiles include evaporation ducts at the surface.Profile D was measured in the Arabian Gulf in June1995. This profile includes an 461-m elevated duct anda 38-m surface-based duct. An additional profile rep-resentative of a standard atmosphere is also shown inFig. 1. When using the profiles of Fig. 1, we retain theevaporation duct for propagation over the sea. Howev-er, we assume that the evaporation duct does not existover land. For overland applications, we delete theevaporation duct and extrapolate the refractivity pro-file above that duct to the surface. The refractivityprofiles shown here have been constructed from tem-perature, pressure, and humidity constituents, usingrelationships given in Reilly et al.11 The evaporation

0

200

400

600

800

1000

1200

1400

300 324 348 372 396 420

Alti

tude

(ft)

Refractivity (M units)

ASTD B C D

Figure 1. Measured refractivity profiles off the coast of California(A, B, and C) and in the Arabian Gulf (D). A and B were measuredby the Pacific Missile Test Center in June 1990; C was measuredby APL in the summer of 1992; D was measured by APL in June1995. (STD = standard atmosphere profile.)

Figure 2. Terrain and ray-trace profile with standard atmospherepropagation.

1500

1125

750

375

00 6030

Range (nmi)

Terrain

Illuminated surface

Alti

tude

(ft)

6 12 18 24 36 42 48 54

434 JOH

duct profile has been constructed using a surface bound-ary layer model with a neutrally buoyant condition.12

Terrain Visibility–Ray-Trace Solutionsfor Stratified Atmosphere

Figure 2 is an example of a DTED terrain profile andray trajectories with standard atmosphere propagation.The highlighted areas are those in direct illumination,according to the procedure described under “GeneralConsiderations.” Figure 3 is a terrain visibility diagramfor a radar located in the Red Sea with standard atmo-sphere propagation. The region shown is characterizedas “high-relief terrain,” having terrain peaks of 8200 ft.Figure 4 is a ray diagram for a refractivity profile involv-ing a uniformly stratified, 600-ft surface-based duct(identified as profile 6 in Ref. 11). Figure 5 shows thesame area as in Fig. 3, but with a 600-ft surface-basedduct.

By comparing Figs. 3 and 5, one sees significantdifferences in the patterns of visible terrain under stan-dard atmosphere and ducting conditions. Clearly, muchmore terrain is directly illuminated with the 600-ft duct

Figure 3. Terrain visibility diagram determined by the ray-tracemethod for a radar located in the Red Sea with standard atmo-sphere propagation.

Figure 4. Terrain and ray-trace profile with a 600-ft surface-basedduct.

34 35 3627.50

28.05

28.60

Longitude (deg)

Radar20 nmi

Latit

ude

(deg

)

3

0 12 24 36 48 60

1500

1200

900

600

300

0

Range (nmi)

Terrain

Illuminated surface

Hei

ght (

ft)



than with the standard atmosphere. However, the max-imum extent of the clutter is similar in the two cases.The increased density of illuminated terrain underducting may be explained by the downward refractionof radar energy, which illuminates features that mightotherwise be hidden, as suggested in Fig. 4. The terrainat great distances is illuminated primarily by rays thatescape the duct and intersect high-altitude terrain fea-tures. For that reason, there is not a large difference inthe maximum extent of visible terrain in the two casesshown here.

Figures 6 through 9 illustrate terrain visibility withpropagation via a 1000-ft surface-based duct (profile Cof Fig. 1) near the coast of Saudi Arabia. In theseexamples, the terrain is characterized as low-to-mediumrelief, with peaks of about 1000 ft. With the radarsituated 70 nmi from the coast (Fig. 7), the directlyilluminated terrain is confined to a narrow band about20 nmi wide along the coast. However, by moving theradar to a location that is 25 nmi more distant (Fig. 8),the directly illuminated terrain extends to about 75 nmifrom the coast (Fig. 9)—a significant increase relativeto Fig. 7. The extended visibility in Fig. 9 occurs

34 35 3627.50

28.05

28.60

Longitude (deg)

Radar

20 nmi

Latit

ude

(deg

)

3

Figure 5. Terrain visibility diagram determined by the ray-tracemethod for a radar located in the Red Sea with a 600-ft surface-based duct and uniformly stratified atmosphere.

Figure 6. Ray diagram for operation near Saudi Arabia with a1000-ft surface-based duct. The radar was 70 nmi from the coast.

0 75 150Range (nmi)

0

1.5

3.0

Terrain

Illuminated surface

Hei

ght (

ft 3

103

)


because of the opportunity for reflected paths from theocean surface to intersect distant terrain via thesurface-based duct model as indicated in Fig. 8.

Modification of Refractivity Profiles by TerrainThe atmospheric constituents that govern refractivity

are temperature, pressure, and humidity—parametersthat are subjected to terrain influences. Consequently,a realistic atmospheric model would include terrain-related modifications to the refractivity profiles, incontrast to the simple stratified atmospheric assump-tions used in the previous examples. To illustrate theability to represent terrain influences, we simulated anadiabatic sea breeze (ASB) model in which a sea breezetransports the air mass from the sea to the land.11 Weassumed that as the air mass is raised in altitude, itundergoes adiabatic expansion (i.e., heat is neitheradded nor subtracted). We assumed knowledge of tem-perature, pressure, and humidity versus height over thesea, and that the temperature, pressure, and humidityconstituents are transformed via an adiabatic process asthe air mass is transported over the land. As the air mass

30

29

2847.00 48.85 50.70

Longitude (deg)

Latit

ude

(deg

) Radar

40 nmi

Figure 7. Terrain visibility diagram for operation near Saudi Arabia,1000-ft surface-based duct. The radar was 70 nmi from the coast.

Figure 8. Ray diagram for operation near Saudi Arabia with a1000-ft surface-based duct.The radar was 95 nmi from the coast.

00

75 150

1.5

3.0

Range (nmi)

Hei

ght (

ft 3

103

)

Terrain

Illuminated surface

1997) 435


moves inland, we calculated new constituent profiles atrange increments DR. We then interpolated the con-stituents between calculated profiles using a linear in-terpolation procedure. In Fig. 5 of Ref. 11, the ASBmodel was applied to profile A of Fig. 1 with a shipposition in the Red Sea as in Fig. 5. When the visibilitydiagrams generated by the ASB and stratified atmo-sphere models are compared, one can observe differenc-es in specific regions. However, the overall density andrange extent of illuminated terrain are similar in thetwo cases considered. For other classes of terrain relief,however, one might expect greater differences betweena stratified atmosphere and an ASB process.

It is not our intention to represent the ASB modelas a realistic case, but rather to demonstrate the abilityof our visibility and clutter models to incorporate theinteraction of terrain and atmospheric processes. Al-though the ASB process introduces additional com-plexities into the refractivitymodel, it is nevertheless a simpli-fication of the physical processespresent in coastal meteorology.We have ignored a number ofphenomena likely to be important,such as boundary layer phenome-na, heat inputs from the terrain,and more complex air movementpatterns.

Visibility of Air TargetsBesides the clutter processes

discussed here, terrain shadowingcan limit the ability of the radarto detect and track overlandtargets. It is often useful to eval-uate target shadowing limitationsusing a target visibility diagram asillustrated in Fig. 10. This exam-ple applies to air targets over

30

29

2847.00 48.85 50.70

Latit

ude

(deg

)Radar

Longitude (deg)

40 nmi

Figure 9. Terrain visibility diagram for operation near SaudiArabia, 1000-ft surface-based duct. The radar was 95 nmi fromthe coast.

Figure 10. Visibility of standard atmosphere p

43.9

43.0

42.116

200-ft

500-ft

1000-

10,000

Latit

ude

(deg

)

436

air targets for a ship position off the coast of former Yugoslavia, withropagation.

former Yugoslavia, for a ship position approximately25 nmi off the coast; the calculated visibility assumeda standard atmosphere. It is assumed that the air targetflies at a constant height above the terrain. The shadedregions indicate where an air target at several differentheights would be directly illuminated by the radar. Thecoded regions should be interpreted as being cumula-tive, e.g., the colored region applying to 10,000 ft alsoincludes the regions for lower- altitude targets. Thisvisibility diagram was produced by calculating ray tra-jectories and applying a visibility algorithm similar tothe procedure used to determine terrain visibility asdescribed under “General Considerations.”

Application of Electromagnetic PropagationRoutine

The ray-trace methods discussed earlier provide rela-tively fast qualitative solutions. The TEMPER programprovides more detailed quantitative calculations of thetotal electromagnetic field using the Fourier split-stepmethod.9,10 Inputs to TEMPER include the radar fre-quency, polarization, antenna elevation beam pattern,antenna elevation pointing angle, and refractivityprofiles (index of refraction versus height). The refrac-tivity profiles may vary with range if such detailedinformation is available. In our applications, the rele-vant output of TEMPER is the one-way or two-waypropagation factor F2 or F4, where F = E/Eo, Eo is thefree-space field, and E is the field under the assumedconditions. We have adapted TEMPER to simulate alower terrain boundary by setting the calculated fieldto zero at and below the terrain boundary at each rangeincrement.13 The first nonzero field value above theterrain surface is identified with field strength incident

18 20Longitude (deg)

*

Radar

target

target

ft target

-ft target

Sarajevo

20 nmi

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 18, NUMBER 3 (1997)


on the surface at that range. This procedure essentiallyrepresents the terrain as a series of knife edges, whichbehaves as an approximation to a perfectly absorbingboundary. With this method, the TEMPER calculationwill be unique for each azimuth slice.

Figure 11 illustrates a TEMPER solution using aparticular terrain slice as a lower boundary; F2 has beencoded on a color scale. The solution includes diffrac-tion energy in regions that would be considered inshadow according to the ray-trace algorithm. Thisexample was generated using profile B of Fig. 1 tospecify refractivity—the same profile used in the raydiagram of Fig. 4. Other relevant parameters for Fig. 11are S-band, vertical polarization, 1.5° beamwidth, and0° antenna pointing angle.

By applying the TEMPER program to each azimuthslice, a terrain visibility map may be obtained. Al-though the TEMPER method does not explicitly de-termine shadowed terrain, one can obtain a terrainvisibility map by applying a threshold to F2. In Ref. 13,a visibility map was generated by applying a thresholdof –6 dB to F2. By comparing that visibility diagramwith the one determined with the ray method (Fig. 5),one sees very little difference.

Terrain Boundary AssumptionsBoth ray-trace and TEMPER methods make use of

perfectly absorbing boundaries. However, the propaga-tion models have the capability to include reflectingboundaries. For the ray method, one might calculate areflected path after a ray encounters the terrain, takinginto account the terrain slope in determining the re-flected launch angle. For TEMPER, one might considerusing a finite impedance lower boundary, such as hasbeen studied elsewhere.14,15

Not only is our implementation of an absorbingboundary much simpler than alternatives using reflect-

Range (nmi)600 30

0

1500

1200

600

300

900

Hei

ght (

ft)

F 2 (dB):0 to –2.99 –6.00 to –8.99 <–12.00

> 0 –3.00 to –5.99 –9.00 to –11.99

Figure 11. Illustration of terrain and propagation factor profile witha 600-ft surface-based duct. F 2 = one-way propagation factor.

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 18, NUMBER 3 (1

ing boundaries, but one can advance arguments forpreferring an absorbing boundary. One argument is thatreflected rays are unlikely to return energy to the sur-face at a more distant range. In most cases of practicalinterest, rays launched at more than 1° penetrate mostpractical ducts and are therefore not refracted back tothe surface. Therefore, terrain slopes of more than 1°would not return reflected energy to more distant sur-face locations. Second, energy will tend to be diffuselyreflected because realistic terrain is typically very roughcompared with radar wavelengths. Consequently, aspecular reflection or smooth surface calculation isusually unrepresentative. Thirdly, absorption at theboundary because of finite impedance will further di-minish the reflected energy.

To adequately assess the impact of terrain boundaryassumptions, more modeling efforts are required. Workcontinues at APL to develop finite impedance bound-ary applications that we will eventually apply to ourterrain effects models.

CLUTTER CALCULATIONS

General RelationshipsClutter magnitude depends on both radar system

and terrain parameters. Radar system parameters in-clude transmitter power, frequency, antenna gain, andresolution size (both range and azimuth). Terrain pa-rameters comprise type, roughness, reflectivity, andlocation relative to the radar. Clutter power can beexpressed by

PP G F A

R Lc

t c

( ),=

2 20

4

3 44

l s

p(1)

where Pc is returned clutter power, Pt is transmittedpower, G is antenna gain, l is radar wavelength, s0 isthe average clutter reflectivity of a radar cell, F4 is thetwo-way propagation factor, Ac is the area of an illu-minated cell, R is the range to the clutter cell, and Lrepresents various system losses.

Using the “constant-gamma” model, the clutterreflectivity is defined as

s s g c0 01 1= =∑ ∑n ni

i

n

ii

nsin , (2)

where n is the number of DTED facets (defined in Ref.11) within a radar cell, s0i (m2/m2) is the reflectivityof the ith facet, ci is the grazing angle at the ith facet,and g is the normalized reflectivity (m2/m2), which is

997) 437


predefined and depends on the radar frequency and thetype of terrain. We use the following frequency rela-tionship in our model:

g g=

r

r,

ff

k

(3)

where gr is a reference value applicable to a particularradar frequency fr; according to Ref. 1, we use the fre-quency scaling factor k = 0.5. Whereas some data sug-gest that g and k vary with both terrain relief andcultural development,1,7 we currently use the samevalues for both high- and low-relief terrain, namely,gr = 0.17 at fr = 3.0 GHz, and k = 0.5. The gr value hasbeen increased relative to our previous estimates12,13 toconform with the measurements presented in the suc-ceeding section. We have not varied gr with terrainrelief, reasoning that terrain relief factors due to large-scale shadowing may be adequately accounted for in ourDTED-based method. We recognize, however, that thespecification of gr for our model as a function of terraintype and frequency requires further investigation andclarification. In addition, grazing angle and radar fre-quency relationships are discussed further in the follow-ing section.

Generally one characterizes clutter in terms of thereflectivity parameter s0, which is expressed as a unit-less quantity (m2/m2); s0 is usually determined by mea-suring returned clutter power and solving for s0 usingEq. 1. Since one typically lacks detailed knowledge ofthe propagation factor at the clutter source, it is cus-tomary to assume F4 = 1 in this calculation. Recogniz-ing the difficulty of separating reflectivity and propa-gation factor, we will characterize clutter reflectivity interms of the combined parameter s0F

4.

Parametric VariationsMany clutter models assume that the backscatter

coefficient s0 increases with increasing grazing angle,1–5

a trend in accordance with rough surface scatteringtheory.16–18 Others have proposed a relationship basedon depression angle of the radar antenna pattern ratherthan on grazing angle of the incident energy.6,7 Accord-ing to depression angle advocates,7 the backscatter fromterrain is dominated by processes involving verticallyoriented features (e.g., trees, cultural features) andassociated “microshadowing,” rather than the roughsurface mechanisms of theoretical models, and thatgeneral rough surface theory does not apply to mostterrain. There are difficulties in defining appropriateangles in either view. Both depression and incidence

438 JOH

angles are affected by atmospheric refractivity proper-ties, and grazing angles are further affected by the finestructure of terrain slopes. In general, such details maynot be accurately known.

Our current model uses a grazing angle relationshipaccording to Eq. 2, where the grazing angle is definedby the slope of a ray from the ray-trace method and theterrain slope of a directly illuminated DTED cell. Forshadowed regions, the grazing angle is approximatedusing the incidence angle of the previous ray and thelocal terrain slope. If the grazing angle in the shadowzone is negative (i.e., negative terrain slope exceeds rayinclination), we replace ci with a minimum value in Eq.2; this approximation is required to account for finitescattering due to surface roughness. In this article, theDTED terrain slope is determined as the slope along atwo-dimensional slice. A more accurate definition ofterrain slope would consider the three-dimensionalslope of terrain. However, the error in backscattercoefficients by using a more simply determined two-dimensional slope is not excessive.11

Clutter reflectivity is often characterized using theparameter g of Eqs. 2 and 3; g is usually determinedexperimentally and may be separately specified forvarious terrain relief and cultural development classi-fications.1 Past descriptions of g can be only an initialguide for our applications for two reasons. First, mostexperimental data on g apply to average reflectivityover an entire radar resolution cell, whereas in ourapplications, we apply g to smaller DTED cells withina radar resolution cell. As a result, we would expect theappropriate g value in our model to be different fromthe larger scale value determined in previous studies.Second, previous descriptions of g include only genericclassifications of terrain, whereas in our model theterrain slopes and shadowing features are separatelydetermined for every DTED cell. Consequently, wehypothesize that aspects of terrain roughness might beaccounted for in a DTED-based approach using DTEDterrain slopes, and the same value of g might apply tovarious terrain relief classifications. This hypothesiswill have to be examined in future experimental studiesof the sort presented in the “Experimental Verification”section of this article.

As discussed in that section, our experimental dataare best fit using g = 0.17 at S-band and 0.20 at Ku-band. The experimental value at Ku-band exceeds thatat S-band by only 0.7 dB, whereas the difference ex-pected from Eq. 3 is 3.25 dB. It will be necessary toconduct additional measurements before we can deter-mine the experimental error in the estimate of g or canconfidently specify g for other radar frequencies andterrain types. Hence, when making predictions of clut-ter strength, we use Eq. 3 with k = 0.5.



Example Calculation of Radar ClutterFigure 12 shows the results of the RADSCAT land

clutter model for an S-band radar with the followingparameters: 0° antenna elevation pointing angle, 1.5°beamwidth (both azimuth and elevation), horizontalpolarization, 1-ms pulse width, and 62-ft antennaheight. Figure 12a is the land profile at a particularradar azimuth. The highlighted surfaces are visible byradar, as determined by the ray-trace program. Notethat the land profile presents only one slice of terrain.Within the azimuthal radar beamwidth, the terrain pro-files could vary, especially at large distances from theradar. In our simulations, we generally compute at leasttwo slices within each beamwidth. Figure 12b shows theone-way propagation factor F2 computed by TEMPERwith profile B from Fig. 1. Figure 12c shows s0F

4 versusrange. When s0 is computed by Eq. 2, the incident anglewithin a shadowed region is assumed to be the same asthat of the last ray on the illuminated surface. Figure12d depicts the returned clutter power calculated by

P kF A

Rc

c .= s04

4(4)

For convenience of calculation and plotting, we eval-uate Eq. 4 using k = 1010, determined with Ac havingunits of square nautical miles, and R having units ofnautical miles. Figure 12d shows clutter power not onlyfrom directly illuminated terrain but also from shad-owed regions due to diffraction.

One can construct a clutter map by repeating thecalculations shown in Fig. 12 at incremental azimuthangles. Figure 13 is a radar clutter map (s0F

4); theassumed propagation conditionand radar location are those usedin the visibility diagram of Fig. 5.The radar parameters in this ex-ample are S-band, 3° beamwidth,0° elevation pointing angle, 2-mspulse width, and 62-ft antennaheight. In Fig. 13, the magnitudeof s0F

4 is indicated with a grayscale. The azimuth increment is0.5°.

Statistical RepresentationThe Weibull distribution is

often used to represent the statis-tical distribution of terrain clut-ter. Statistical representations ofterrain clutter can, however, beambiguous and lead to misinter-pretation. Therefore, one must

s0F 4 (dB)

Latit

ude

(deg

)

34

28.60

28.38

28.16

27.94

27.72

27.50

10 nmi

Figure13. Radar clutte

JOHNS HOPKINS APL TECHNICAL DIGEST

5000

4000

3000

2000

1000

0

10.0

–12.5

–35.0

–57.5

–80.0

0

–20

–40

–60

–80

30

10

–10

–30

–500 10 20 30 40 50

Range (nmi)

Alti

tude

(ft)

s0F

4 (

dB)

Pow

er (

dB)

F2

(dB

)

(a)

(b)

(c)

(d)

Figure 12. Land clutter return of an S-band radar simulation undera heavy surface duct propagation condition. (a) Terrain profile atradar beam center; (b) one-way propagation factor (F 2); (c) radarclutter reflectivity (s0F 4); (d) returned clutter power.

:

Longitude (deg)35 36

–18 <

–22 < ≤ –18–26 < ≤ –22–30 < ≤ –26

–34 < ≤ –30≤ –34

Radar

r map (s0F 4).

, VOLUME 18, NUMBER 3 (1997) 439


be circumspect in interpreting and applying statisticalclutter models. A major reason for this ambiguity is theinfluence of the minimum threshold used in the statis-tical database, a parameter that is usually unstated indescriptions of experimental clutter data.

The selection of a lower limit in simulated data isarbitrary. Since the model has no absolute lower limiton s0F

4, one could set a lower threshold to excludeweak clutter data. The effect would be to alter thestatistical parameters of the resulting database. As apractical matter, a similar issue will be present withmeasured data. Radar measurements will be limited toclutter cells that exceed a minimum signal-to-noiseratio (SNR), and this minimum will be a function ofthe sensitivity of the radar, including transmitter power,antenna gain, noise figure, and other parameters. Toillustrate this point, imagine that clutter measurementsare made by two radars that differ in sensitivity but areidentical in all other respects. If we set a threshold atsome multiple of the SNR, the more sensitive radarwould include more data points with small values ofs0F

4 and would have a smaller average s0F4.

We illustrate this situation in Figs. 14 and 15, whichgive simulated cumulative count distributions of s0F

4

for low- and high-relief terrain, respectively, includingthe frequency scaling law of Eq. 3. Figures 14 and 15were determined with Eq. 3 and a hypothetical value

0

500

1000

1500

2000

2500

–110 –90 –70 –50 –30 –10

Nu

mb

er

of co

un

ts

s0 F 4 (dB)

X-bandS-bandL-band

Nu

mb

er

of

cou

nts

s0 F 4 (dB)

0

200

400

600

800

1000

–100 –80 –60 –40 –20 0

X-bandS-bandL-band

Figure 14. Cumulative distribution of radar clutter reflectivity (s0F 4)in low-relief terrain.

Figure 15. Cumulative distribution of radar clutter reflectivity(s0F 4) in high-relief terrain.

440 JOH

of g = 0.05 at S-band. Three radars are represented,differing only in frequency. The high-relief terrain (Fig.15) applies to the Red Sea area with the radar lookingeastward toward mountainous terrain in the azimuthsector 50–60° and a maximum range of 60 nmi. For low-relief terrain (Fig. 14), the radar is situated in theArabian Gulf 10 nmi from the coast at 29.1° latitudeand 48.3° longitude. The observed region covers thelow-relief terrain of Kuwait and Saudi Arabia in thesector 260–270°, with a maximum range of 60 nmi. Thesimulated data include shadowed terrain illuminatedthrough diffraction and do not have an absolute min-imum value. Because of enhanced diffraction at longerwavelengths, small values of clutter are enhanced as thefrequency band is lowered. However, the reverse is trueat large values of s0F

4, where clutter is enhanced asfrequency is increased because of the scaling law givenin Eq. 3.

Table 1 summarizes the averages of s0F4 and number

of cells (N) exceeding the threshold when variousthresholds are applied to the data of Figs. 14 and 15.From these data, one might infer that s0F

4 has a fre-quency scaling law that depends both on threshold andterrain relief. A terrain relief dependency was alsoindicated in the measurements of Billingsley,7 whofound s0F

4 to increase with increasing radar frequencyin low-relief farmland but to decrease with increasingradar frequency in high-relief mountainous terrain.

In comparing simulation results with measurements,it is important to understand the difference in statisticalproperties of simulated and measured data. Whereas themodel provides statistical expectations, i.e., statistical

Table 1. Statistics of radar clutter reflectivity (s0F4) withvariation of threshold and radar frequency (1.5° beam-width, 1-ms pulse width, 62-ft antenna height).

L-band S-band X-bandThreshold Mean Mean Mean

(dB) (dB) N (dB) N (dB) NHigh-relief terrain

40 –34.3 568 –32.0 551 –29.3 54930 –37.9 647 –36.4 641 –32.5 60820 –42.3 739 –40.5 721 –35.7 66010 –47.3 843 –44.8 804 –40.3 730

0 –50.6 912 –48.5 872 –45.8 817–10 –53.0 963 –51.6 929 –49.5 875

Low-relief terrain40 –42.0 418 –41.8 462 –39.1 48330 –48.3 683 –46.5 631 –43.1 62720 –53.9 971 –52.2 908 –47.7 76810 –59.7 1294 –57.9 1177 –54.0 992

0 –66.1 1684 –65.0 1549 –61.5 1286–10 –70.7 1980 –70.7 1889 –70.0 1678

Note: Threshold refers to clutter-to-noise ratio for a hypotheticalradar system. N = number of cells.



averages of reflectivity from individual radar cells,actual measurements include statistical fluctuationsabout these averages. The simulation results can be saidto belong to a simple statistical distribution, whereasmeasurements will follow a compound distribution (seeFig. 23, which is discussed later).

EXPERIMENTAL VERIFICATION

Data Collection and ProcessingTerrain clutter measurements were collected by a

shipboard radar equipped with a coherent data collector(CDC), which has a 40-dB dynamic range. To increasethe dynamic range, it was necessary to collect data witha series of attenuation as was done with S-band data.Data were collected near the west coast of the UnitedStates and near the Arabian Gulf in 1993 and 1995,respectively. The measured areas of the west coast of theUnited States are mountainous high-relief terrain; themeasured areas of the Arabian Gulf have both moun-tainous high-relief and desert low-relief terrain. Theterrain of both regions has a variety of features, includ-ing deserts, forests, vegetated land, urban areas, andrural areas. The tests did not include absolute calibra-tion of the radar power. In relating radar measurementsto absolute reflectivity, we relied on specified radarparameters.

Terrain clutter measurements were collected nearthe coast of southern California and northern Wash-ington in May and June 1993. In this article, we presentthe data of southern California only. Pertinent radarparameters are S-band, horizontal polarization, 1.6°beamwidth (azimuth and elevation), 0° elevationpointing angle, 3.0-ms pulse width, and 120-ft antennaheight. Although the S-band radar has a relativelystrong output power, clutter data were recorded by adata collector having a limited dynamic range. To coverthe full measurement range required for our tests, datawere sequentially collected with attenuation levels of80, 60, 40, 20, and 0 dB and later merged for expandeddynamic range. In the data merging process, a 17-dBlower limit and a 37-dB upper limit were used at eachattenuator setting to eliminate system noise and satu-rated data, respectively. The clutter-pulse-noise-to-noise [(C+N)/N] ratio of the merged data ranged from17 to 117 dB. The process of data collection at the fiveattenuator settings consumed approximately 4–6 min.In the merged data file, each clutter measurement hasbeen attributed to a fixed radar cell despite the fact thatthe ship speed was approximately 10 knots during thedata collection period. This extended data collectionprocess effectively creates a smearing effect. The datacollector can acquire measurements over an extendedazimuth and range sector. In these tests, the azimuthand range sectors were approximately 130° and 63 nmi,


respectively. Considering the range resolution (0.24nmi), azimuth resolution (1.6°), and attenuator set-tings (5), more than 105 measurements compose a sin-gle range/azimuth data set. Multiple data sets weretransferred to tape for later analysis.

Atmospheric data were collected using an instru-mented helicopter that measured temperature, pres-sure, and humidity versus height. The helicopter flewa series of sawtooth patterns from the ship over thewater, thereby allowing a sequence of profiles to beconstructed as a function of range from the ship. At-mospheric data were collected 10–15 nmi inland.These data were later processed to determine a seriesof profiles of refractivity index versus height. Duringdata collection, measured refractivity did not deviatesignificantly from a standard atmosphere condition. Asa general rule, however, propagation in marine or coast-al environments can deviate significantly from stan-dard atmosphere conditions, resulting in significantvariations in terrain clutter as seen by a shipboard radar(see preceding discussion).

A Ku-band and an X-band radar were used tocollect land clutter data in the Arabian Gulf in Feb-ruary and June 1995. The Ku-band radar parameters arevertical polarization, 63-ft antenna height, 1.6° azimuthbeamwidth, 10° elevation beamwidth (pointing at4.5°), and 200-ns pulse width. The Ku-band radar hasrelatively low power, and the full measurement rangecan be covered by the 40-dB dynamic range of theCDC without adding attenuation. The X-band radarparameters are horizontal polarization, 58-ft antennaheight, 1.5° azimuth beamwidth, 4.7° elevation beam-width (pointing at 2.3°), and 260-ns pulse. The X-bandmeasurements (using a 64-pulse compression/uncom-pression mode) were collected by the same type ofCDC without adding an attenuator.

Atmospheric data were not collected during the Ku-band radar measurements, but since the Ku-band radarwas very close to land, the radar measurements wouldnot be significantly affected by the atmospheric con-ditions. An assumption of a standard atmosphere isused in the simulation to be discussed. For the X-bandradar measurements, refractivity profile D of Fig. 1 wasmeasured approximately 8 h before the data were col-lected.

Comparison of Measurement and Model:Geographic Patterns

Figure 16 illustrates several features of measured andsimulated reflectivity data. Figure 16a shows a terrainprofile determined from the DTED database for a singleazimuth direction (106.34°) with the ship positionedin southern California. The highlighted regions repre-sent directly illuminated terrain, as determined withthe ray-trace algorithm. The terrain profile shown is

1997) 441


only one representation of the terrain contour withina radar beamwidth. Within a beamwidth, the terrainheight will vary at a fixed range because of the azimuthextent of the beam, especially at the greater limits ofrange. Figure 16b shows the one-way propagation factordetermined from the TEMPER program. Figure 16cshows measured and modeled values of s0F

4 versusrange, where the simulated data were obtained with theaverage of two azimuth slices per radar beamwidth. Themagnitude of s0F

4 necessary to equal system noise isalso shown. As a general rule, measurements below thenoise limit are not possible.

Figure 17a shows wide-area views of measured clutterin southern California. In Fig. 17a, magnitudes of s0F

4

have been coded on a color scale; these magnitudeswere derived from measurements of radar power usingEq. 1. This picture provides one example out of manysuch data sets collected during the exercise. Corre-sponding modeled data are shown in Figs. 17b and 17c.Figure 17b was derived from the Terrain VisibilityRoutine (TEVIR) using the ray-trace algorithm. Figure17c was derived from the RADSCAT program.

By overlaying the measurement and simulationmaps, one observes very good correspondence betweenthe geographic patterns of measured and modeled data.The predicted areas of terrain illumination in theray-trace routine (Fig. 17b) correspond well with the

Figure 16. Simulated and measured land clutter reflectivity. (a)Terrain profile at radar beam center; (b) one-way propagationfactor (F 2); (c) radar clutter reflectivity (s0F 4).

(a)

(b)

(c)

0

1000

2000

3000

4000

5000

6000

–100

–80

–60

–40

0

20

–100

–80

–60

–40

–20

0

0 10 20 30 40 50

Range (nmi)

s0

F4

(dB

)

Modeled data

Measured data

Noise limit

–20

F2

(dB

)E

leva

tion

(ft)

Illuminated terrain

Terrain

442 JO

Radar

(c)

119.0 117.5 116.0Longitude (deg)

Latit

ude

(deg

)

Radar

(b)

Latit

ude

(deg

)

(a)

Radar

Latit

ude

(deg

)

20 nmi

33.80

33.05

32.30

33.80

33.05

32.30

–20 dB<≤ –20 dB

≤ –50 dB–108 dB<

–50 dB<

–20 dB<≤ –20 dB

≤ –50 dB–108 dB<

–50 dB<

33.80

33.05

32.30

20 nmi

20 nmi

Figure 17. Comparison of land clutter measured and modeleddata. (a) Measured radar clutter reflectivity (s0F 4) at the coast ofsouthern California within 63.0 nmi; (b) terrain visibility map pre-dicted with the ray-trace method (standard atmosphere; antennaheight = 120 ft); (c) map of simulated radar clutter reflectivity(s0F 4) using RADSCAT method.

contours of measured clutter above –20 dB in Fig. 17a.This correspondence suggests that the strong cluttermeasurements are associated with directly illuminatedterrain.

Figure 18 shows a terrain profile determined fromthe DTED database for a single azimuth direction(213.9°) and the corresponding measured data atKu-band. The highlighted points on the terrain profileindicate directly illuminated terrain as determined byTEVIR. The positions of strong measured clutter re-turns coincide exactly at illuminated terrain shown inthe figure. Figure 19 illustrates wide-area patterns ofmeasured and simulated clutter. In Fig. 19a, magnitudesof measured (C+N)/N are color coded between

HNS HOPKINS APL TECHNICAL DIGEST, VOLUME 18, NUMBER 3 (1997)


Range (nmi)

(b)

0 10 20 30 40 50

70

50

30

10

–10

(C +

N)/

N (

dB)

0

1000

2000

3000

4000(a)

Ele

vatio

n (f

t) Terrain

Illuminated terrain

Figure 18. Land clutter return of Ku-band radar for a singleazimuth direction (213.9°). (a) Terrain profile; (b) measured (C+N)/N(clutter-plus-noise-to-noise ratio).

193 to 247° in azimuth. By overlaying Figs. 19a and19b, one finds good correspondence between the geo-graphic patterns of measured data and simulated visibleterrain, especially at close ranges. Because of the lowpower of the Ku-band radar, only the strongest clutterreturns beyond 15 nmi can be represented. Conse-quently, the geographic size of measured land clutter issmaller than the simulated visible terrain at distantranges.

Figure 20 illustrates several features of the measuredand simulated clutter return, (C+N)/N, of an X-band(MK 92) radar. Figure 20a shows a terrain profile de-termined from the DTED database; Figs. 20b and 20cshow the measured data without and with pulse com-pression, respectively. Figure 20d shows the simulated(C+N)/N of land clutter. The strongest measurementswere seen to be saturated, as indicated by the flat topsin Fig. 20b. Since the pulse compression is a functionof uncompressed pulse, the magnitude of the com-pressed data (Fig. 20c) is saturated as well. Despite thesaturation of measured data, simulated data at all rangepositions agree with the patterns of illuminated terrain,which is highlighted in the terrain profile.

Figure 21a shows the terrain visibility map simulatedby TEVIR near the coast of Iran under a strong surface-based duct. The beach area in the figure is directlyilluminated due to the duct, but would not be illumi-nated under standard atmosphere propagation. Figure21b shows the land clutter map of the X-band measure-ments with uncompressed pulse mode. Although the(C+N)/N data are saturated, they are adequate for aqualitative analysis. By comparing Figs. 21a and 21b,we observe good correspondence between the geo-graphic patterns of measured data and simulated visibleterrain.


Figure 19. Comparison of measured data and simulated terrainvisibility map. (a) Measured (C+N)/N (clutter-plus-noise-to-noiseratio) with Ku-band radar at the coast of United Arab Emirates; (b)simulated terrain visibility map predicted with TEVIR (standardatmosphere; antenna height = 63 ft).

0

10

20

30

40(d)

20 30 40 50 60 70 80Range (nmi)

(C+

N)/

N (

dB)

0

10

20

30

40(c)

(C+

N)/

N (

dB)

(C+

N)/

N (

dB)

(b)

0

10

20

30

400

1000

2000

3000

4000

5000

Ele

vatio

n (f

t)

(a)

Figure 20. Land clutter return of an X-band radar. (a) Terrainprofile; (b) uncompressed data; (c) compressed data; (d) simulatedclutter.

25.2

24.9

24.6

Radar

42 dB< ≤ 70 dB

≤ 42 dB≤ 38 dB

≤ 34 dB

38 dB<34 dB<30 dB<

25.2

24.9

24.6

Radar

55.6 56.1 56.6

Longitude (deg)La

titud

e (d

eg)

Latit

ude

(deg

)

(a)

(b)

10 nmi

10 nmi

997) 443


Figure 21. Comparison of simulated terrain visibility map andmeasured data. (a) Terrain visibility map predicted with TEVIR, atthe coast of Iran (ducting atmosphere, antenna height = 58 ft); (b)measured (C+N)/N (clutter-plus-noise-to-noise ratio) collected withan X-band radar using 64-bit uncompressed pulse at the coast ofIran (data shown from 27 to 80 nmi in range and 0 to 90° in azimuth).

Statistical RepresentationFigure 22 illustrates the cumulative distribution of

measured and modeled reflectivity (s0F4) in southern

California. The vertical axis gives the number of radarcells exceeding the value of s0F

4 shown on the hori-zontal axis. One curve shows the measured data; theother lines show simulated data using the methoddescribed in the “Clutter Calculations” section withg = 0.17. The modeled data includes calculated effectsof various horizontal antenna sidelobe levels. Sidelobeeffects were determined by integrating clutter from cellsalong a constant azimuth slice, but where the cluttermagnitude was reduced by the assumed sidelobe level,and adding the integrated sidelobe clutter to the cal-culated main beam value. A sidelobe level of –80 dBapproximates that of the experimental system.

For s0F4 greater than –40 dB, the measured and

simulated results in Fig. 22 agree well. For weaker clut-ter, however, the model predictions are significantlybelow measurements. It was suspected that the lack ofcorrespondence of weaker clutter might occur primarilyin shadow (diffraction) regions. To test this hypothesis,we examined the distribution of measured and simulat-ed clutter applying to radar cells that contained at least

27.6

26.9

26.251.9 53.3 54.7

Longitude (deg)

25.2

24.655.6 56.1 56.6

(a)

(b)

Radar

20 dB<18 dB< ≤ 20 dB

30 dB<≤ 30 dB

15 dB< ≤ 18 dB

Latit

ude

(deg

)

Radar

Longitude (deg)

Latit

ude

(deg

)

24.9

20 nmi

20 nmi

444 JO

Figure 22. Cumulative distribution of radar clutter reflectivity(s0F 4) for the coast of southern California; model with standardatmosphere, g = 0.174, two-way sidelobe = –60 to –80 dB.

a portion of directly illuminated terrain as indicated bythe ray-trace method. The resulting distributions areshown in Fig. 23. Two distributions are shown for themodeled data: one applies to the calculated values ofs0F

4, following the convention shown in Fig. 22; theother is obtained by random numbers drawn from aRaleigh power distribution, in which the mean valueis defined by the calculated s0F

4 for that cell. Thisdistribution simulates a compound distribution as de-scribed previously. When statistical distributions arecompared, the compound distribution is expected tomost nearly simulate the measured data.

It is seen that the measured and modeled distribu-tions in Fig. 23 correspond very well, suggesting thatthe model performs quite well when directly illuminat-ed surfaces are contained within the radar cell. Themedian value of s0F

4 is –23 dB for both measured andmodeled data; the Weibull a-parameter is 1.7 for themeasurements and the modeled compound distribu-tions. These values compare favorably with previouslypublished data at S-band.1

As indicated earlier, the X-band radar measurementswere saturated as a result of the limited dynamic rangeof the CDC measurements. Therefore, further quanti-tative analysis is not possible using the X-band data.Figure 24 illustrates the cumulative distribution of

1600

1200

800

400

0

Num

ber

of c

ells Measured data

Model (simple)

Model (compound)

–80 –70 –60 –50 –40 –30 –20 –10 0

s0F 4 (dB)

Figure 23. Cumulative distribution of radar clutter reflectivity(s0F 4) applied to radar cells containing directly illuminated terrainfor the coast of southern California.

Model with no sidelobeWith –80-dB sidelobe

Measured data

With –70-dB sidelobe

With –60-dB sidelobe

12

10

8

6

4

2

0Num

ber

of c

ount

s (in

thou

sand

s)

–80 –60 –40 –20 0 20s0F 4 (dB)

HNS HOPKINS APL TECHNICAL DIGEST, VOLUME 18, NUMBER 3 (1997)


measured and modeled reflectivity of the Ku-band radarin the Arabian Gulf. As noted in Fig. 18b, the averagevalue of (C+N)/N for the Ku-band data is 18 dB, whichis higher than most clutter returns in the diffractionzone. For a valid comparison, in Fig. 24, the cumulativeplot includes only the directly illuminated radar cellsdeterminated by TEVIR. As with the X-band data,good correspondence between measured and modeleddata is seen for strong returns. Unlike the S-band radar,no Ku-band data exist that can be used to validate themodel for weak clutter. The simulated data in Fig. 24were generated by the RADSCAT program, withg = 0.2. According to Eq. 3, g should equal 0.36 for aKu-band radar. If the clutter model used g = 0.36, thecumulative curve of the model would shift by 2 dB inFig. 24. It would require further study to determine acorrect g relationship for different radar frequenciesand terrain types.

Investigation of Differences BetweenMeasurement and Model

The statistical distribution of measured and simulatedclutter appears to correspond well for relatively strongclutter or for radar cells having directly illuminatedsurfaces. For weaker clutter or for clutter pertaining toshadow zones, the model appears to underpredict themeasured S-band data. We are investigating severalreasons for this discrepancy. The following describespotential sources of discrepancy that are being considered.

The sidelobes of an experimental system can in-crease the occurrence of relatively weak clutter. The S-band radar azimuth sidelobes, as seen in Fig. 22, doincrease relatively weak clutter but not to an extentsufficient to bring the measurement and model in com-plete correspondence. The experimental S-band systemalso had elevation sidelobes and cross talk betweenvertically separated beams. The effects of these eleva-tion sidelobes and cross talk were simulated and found

Figure 24. Cumulative distribution of radar clutter reflectivity (s0F 4)applied to radar cells containing directly illuminated terrain for thecoast of United Arab Emirates.

Measured data

Model

1400

1200

1000

800

600

400

200

0

Num

ber

of c

ount

s

–30 –25 –20 –15 –10 –5 0 5s0F 4 (dB)


to reduce the difference between measured and mod-eled data. We note that at the trailing edges of strongclutter peaks, the measured clutter tends to exceed thesimulated data (see Fig. 16). The effects of verticalantenna beam cross talk appeared to account for thetrailing edge effects. Despite this improvement, signif-icant differences remained between measured andmodeled data in the shadow zone (s0F

4 below –40 dB).The S-band data at five attenuator settings were

collected over a period of 4–6 min, during which timethe ship maintained a velocity of about 10 kt. To in-vestigate the possible smearing effects of positionchanges during the data collection period, we need tosimulate a database that contains such smearing effects.

As described in the section on visibility maps, ter-rain effects are incorporated into the TEMPER electro-magnetic routine using an approximation to a perfectlyabsorbing boundary. We are considering the possibilitythat the terrain boundary method implemented inTEMPER might not adequately represent diffractionzone effects. Another simplification in the model isthat the terrain boundary is two-dimensional, ratherthan three-dimensional, as in the real world. The useof a two-dimensional approximation will introduceerrors into the simulation. Another possible source oferror is that multiple scattering processes (as a result ofthe three-dimensional aspects of the terrain) are notincluded in the simulation.

SUMMARYIn this article, we discuss a model for terrain effects

on shipboard radar performance that accounts for site-specific terrain features and for propagation effects.Both terrain shadowing and clutter can be simulated.The method can accommodate atmospheric data thatvary in three dimensions, if such detailed informationis available. Site-specific terrain contours are describedthrough the DTED database, which is provided by theDefense Mapping Agency. The model is configuredwith various degrees of complexity. This article in-cludes previously published results at the S-band19 aswell as more recent data at the X-band and Ku-band.

Relatively simple, but fast, methods are providedwith TEVIR-I and -II. TEVIR-I computes terrain andtarget shadowing, where the propagation can be repre-sented as straight-line propagation over a round Earthwith an equivalent Earth radius factor. Atmosphericprofiles fitting this category can be represented by aconstant gradient of the index of refraction versusheight. The “standard atmosphere” is one example fit-ting into this category. TEVIR-II also determines ter-rain and target shadowing, but with arbitrary atmo-spheric inputs.

The RADSCAT method employs an electromagneticparabolic equation method to calculate the propagation

997) 445


factor. RADSCAT includes both refraction and diffrac-tion effects and can simulate terrain shadowing andbackscatter. This method can accept refractivity thatvaries in three dimensions. Although the RADSCATmethod provides much more detailed information, italso takes much more time to execute as compared withthe TEVIR methods.

Variations in atmospheric refractivity can signifi-cantly alter patterns of terrain clutter and shadowing.The effects will depend on the structure of atmosphericrefractivity as well as terrain relief. With a surface-basedduct, for instance, it is possible to markedly increase thedensity of directly illuminated terrain or to greatly extendthe range extent over which strong clutter is returned.

The RADSCAT and TEVIR models were comparedwith radar measurements taken off the west coast of theUnited States from an S-band radar. The correspon-dence between measured and simulated clutter is verygood for relatively strong clutter. This correspondenceis evident in the geographic patterns and statisticaldistribution of clutter returns in both northern Wash-ington and southern California. Although the terrainhas similar mountainous relief in the two locations, thecomposition of the terrain is quite different. In themeasurement area of northern Washington, the terrainis forested, with little cultural development. In south-ern California, the terrain is semiarid with significantcultural development, particularly along the coast. De-spite the differences in terrain composition, the clutterreturns on directly illuminated surfaces were similar inthe two locations.

Other data used in this study were collected by Ku-band radar under the standard atmosphere conditionnear the west coast of the United Arab Emirates andby X-band under a strong surface-based duct near thecoast of Iran. In both cases, excellent agreement be-tween measurements and modeled clutter was obtainedwhen the spatial patterns of strong clutter returns werecompared.

Within shadowed regions, the model predicts generallylower clutter strengths as compared with measurements.We are currently investigating several hypotheses thatmight explain the discrepancy. Primary considerationsinclude smearing due to platform motion during themeasurement interval, inadequate representation of dif-fraction, multiple scattering effects, data collection re-sponse dynamics, and refractive effects of vegetation.

446 JOH

Future improvements in the model will incorporatethe Digital Feature Analysis Data (DFAD) database ofterrain composition. This database, also published by theDefense Mapping Agency, indicates features such asvegetation, structures, roads, bridges, and power lines.With such data, it is possible to better predict backscatterand the effects of both natural and cultural features.

REFERENCES1Nathanson, F. E., Reilly, J. P., and Cohen, M. N., Radar Design Principles, 2nd

Ed., McGraw-Hill, Inc., New York (1991).2Long, M. W., Radar Reflectivity of Land and Sea, Artech House, Dedham, MA

(1983).3Ulaby, F. T., and Dobson, M. C., Handbook of Radar Scattering Statistics for

Terrain, Artech House, Norwood, MA (1989).4Barton, D. K., “Land Clutter Models for Radar Design and Analysis,” Proc.

IEEE 73(2), 198–204 (1985).5Henn, J. W., Pictor, D. H., and Webb, A., “Land Clutter Study: Low Grazing

Angles (Backscattering),” in Advances in Radar Techniques, J. Clark (ed.),Peter Peregrines Ltd., London, England, pp. 222–226 (1985).

6Currie, N. C., and Zehner, S. P., “Millimeter Wave Land Clutter Model,” inAdvances in Radar Techniques, J. Clarke (ed.), Peter Peregrines Ltd., London,England, pp. 227–231 (1985).

7Billingsley, J. B., “Radar Ground Clutter Measurements and Models, Part I:Spatial Amplitude Statistics,” AGARD Conference Proceedings on Target &Clutter Scattering and Their Effects on Military Radar Performance, AGARD-CP-501, NATO Electromagnetic Wave Propagation Panel Meeting, Ottawa,Canada, DTIC No. AD-P006373, pp. 1.1–1.15 (May 1994).

8Ayasli, S., “SEKE: A Computer Model for Low Altitude Radar PropagationOver Irregular Terrain,” IEEE Trans. Ant. Prop. AP-34(8), 1013–1023(1986).

9Dockery, G. D., “Modeling Electromagnetic Wave Propagation in theTroposphere Using the Parabolic Equation,” IEEE Trans. Ant. Prop. 36(10),1464–1470 (1988).

10Kuttler, J. R., and Dockery, G. D., “Theoretical Description of the ParabolicApproximation: Fourier Split-Step Method of Representing ElectromagneticPropagation in the Troposphere,” Radio Sci. 26(2), 381–393 (1991).

11Reilly, J. P., Lin, C. C., and Rudie, S. A., “Land Clutter and Shadowing WithConsideration of Propagation in Coastal Regions,” Proc. Radar ’92, Brighton,England, pp. 26–29 (Oct 1992).

12Paulus, R. A., “Practical Applications of an Evaporative Duct Model,” RadioSci. 20(4), 887–896 (1985).

13Lin, C. C., and Reilly, J. P., “Radar Terrain Clutter Model with Consider-ation of Propagation Effects,” Proc. 23rd European Microwave Conf., Madrid,Spain, pp. 478–482 (Sep 1993).

14Marcus, S. W., “A Hybrid (Finite Difference–Surface Green’s Function)Method for Computing Transmission Losses in an Inhomogeneous Atmo-sphere Over Irregular Terrain,” IEEE Trans. Ant. Prop. 40(12), 1451–1458(1992).

15Levy, M. F., “Parabolic Equation Modeling of Propagation Over IrregularTerrain,” Electron. Lett., 26(15), 1153–1155 (1990).

16Beckman, P., “Scattering by Composite Rough Surfaces,” Proc. IEEE 53,1012–1015 (Aug 1965).

17Barrick, D. E., “Rough Surface Scattering Based on the Specular PointTheory,” IEEE Trans. Ant. Prop. AP-16(4), 449–454 (July 1968).

18Beckman, P., and Spizzochino, A., The Scattering of Electromagnetic Wavesfrom Rough Surfaces, Artech House, Norwood, MA (1987).

19Reilly, J. P., and Lin, C. C., “A Propagation-Based Model of Terrain Effectsfor Shipboard Radar Applications,” Proc. NATO AGARD-SPP Meeting,Bremerhaven, Germany (Sep 1994).

ACKNOWLEDGMENTS: This work has been supported through the U.S. NavyAegis Shipbuilding Program, PMS-400.



THE AUTHORS

J. PATRICK REILLY is a member of APL’s Principal Professional Staff andsupervisor of the Environmental Modeling Section of the Sensor Signal and DataProcessing Group in the Air Defense Systems Department. He obtained a B.E.E.from the University of Detroit in 1962 and an M.S.E. from George WashingtonUniversity in 1965. Since joining APL in 1962, he has worked on a variety oftheoretical and experimental projects associated with radar, sonar, acoustics,infrared systems, and bioelectric phenomena. He supervised the Electromagneticsand Acoustics Section of the Environmental Assessment Group, and was thedirector of research programs on human reactions to electric currents andelectromagnetic fields. Mr. Reilly is a senior member of the IEEE and a memberof the Bioelectromagnetics Society. He is the author of a book entitled ElectricalStimulation and Electropathology, and also the co-author of another book entitledRadar Design Principles. His e-mail address is [email protected].

CHRISTOPHER C. LIN received his B.S. and M.S. degrees in electricalengineering from Drexel University in 1983 and 1985, respectively. He joinedAPL in 1985 and is a Senior Staff engineer in the Sensor Signal and DataProcessing Group of the Air Defense Systems Department. He has worked on avariety of projects for radar detection and tracking systems, and developed thealgorithm for multisensor tracking. Currently, Mr. Lin is developing a terraineffects clutter model for shipboard radar applications. His e-mail address [email protected].

JOHNS HOPKINS APL TECHNICAL DIGEST, VOLUME 18, NUMBER 3 (1997) 447

Date post:	19-Nov-2021
Category:	Documents
Upload:	others
View:	5 times
Download:	0 times

A Site-Specific Model of Radar Terrain Backscatter and ...

Documents