FIG. 1. The IDECS (Image Discrimination, Enhancement, Combination and Sampling) DataHandling System developed at the University of Kansas.

S. A. MORAIN*

and D. S. SIMONETT

University of KansasLawrence, Kansas

K-Band Radar in Vegetation MappingDiscrimination studies of pine, fir, andhardwood forests, juniper woodland,sagebrush, chaparral, grassland, etc.

ANALYSIS TECHNIQUES

VEGETATION ANALYSIS METHOD WITH RADAR

I N AN EARLIER PAPER (Morain and Simonett,1966) we noted that a number of methods

help in the discrimination of plant commu­nities on radar imagery. These methods in-

* Presented at the Annual Convention of theAmerican Society of Photogrammetry, Washing­ton, D. c., March 1967 under the title "An Evalua­tion of K-Band Radar in Vegetation Mapping:Image Discrimination Studies at Horsefly Moun-

elude analysis of: (1) differences in averagegray scale values between communities, (2)textural differences, (3) edge effects, partic­ularly those which arise from different meansof gradation from one plant community toanother, (4) probability density functions perunit area, (5) analysis of spatial arrange-

tain, Oregon." The study was supported by theNational Aeronautics and Space Administrationunder Contract NSR-17-004-003 and NASA GrantNSG-298.

730

K-BAND RADAR IN VEGETATION MAPPING 731

ments, based on a knowledge of general dis­tribution of plant communities in an area,(6) the context of the communities, (7) imagetransformation, (8) differentiation, (9) in­tegration, (10) expansion of a portion of thedynamic range, and (11) level slicing. Items8 through 11 lend themsleves to two-di­mensional multi-color presentations involvinguse of the luminance and two-chrominancechannels in a color television system, and

handling system is the scanning of up to threeimages in a three-channel flying-spot scanner(FSS). The electrical analogs of the imagesplaced on the faces of the FSS are fed singlyor collectively to the matrixing unit and theoutputs of the matrix are presented to thethree (red, blue and green) electron guns ofthe cathode ray tube in a color televisionunit (CTV). Combinations of images canthus be reproduced in various colors to aid

ABSTRACT: Methods for the interpretation of vegetation from radar imageryhave been investigated through use of an image discrimination, enhancement,combination and sampling system (IDECS) developed at the Center for Re­search in Engineering Science at the University of Kansas. HH and HVpolarization K-band imagery of Horsefly Mountain, Oregon has been analyzedusing several electronic techniques aimed at improving our ability to discrimi­nate distributional patterns of vegetation. These techniques include the use oftri-color image combinations, the generation of probability-density functions toquantify variations in gray-scale level between types,. and the employment of adata space sensor to help distinguish between vegetation types.

photographic color methods involving mul­tiple projectors and color filters.

We also noted that although the above areconcerned with single images, it was alsopossible to obtain simultaneously recordedradar imagery which would give the fullpolarization matrix and/or a number of fre­quencies. Data of this type is potentiallysuitable for a number of additional types ofanalysis including polychromatic-displaytechniques.

It is with the polychromatic-display tech­niques that the present paper is primarilyconcerned, for the study deals mostly withthe various color combinations posbible withHH and HV polarization K-band imageryof a vegetated area at Horsefly Mountain,Oregon. With this color combined imageryvarious forms of level slicing, and data-spacesampling are employed.

ELECTRONIC IMAGE DISCRIMINATION

Equipment capable of performing thenecessary image discrimination, enhance­ment, combination and sampling (for whichwe use the acronym IDECS) has been as­sembled at the Center for Research in Engi­neering Science at the University of Kansas.The system, illustrated in Figures 1 and 2 isdescribed more fully in Simonett, (1966) andonly a brief outline of the operation of thesystem is given here.

The first operation in the IDECS data-

their interpretation. The CTV is synchro­nized with the FSS by synchronizing theirrasters (see Figure 2). An alternative tri­color oscilloscope (TCO) may also take theoutputs from the matrix or directly from theA, Band C channels of the FSS. This devicehas very high color fidelity, but poorer resolu­tion than the CTV. It may however be usedfor scanning images and handling radarscatterometer data.

The output of either the red, blue or greenchannels from the matrix unit or the A, B,

D. S. SIMONETT

732 PHOTOGRAMMETRIC ENGINEERING

3-Channel Flying

Spat Scanner

(FSS)

Sync &Interlace

Generator

(SYNC)

Black & White

Television

Color

Television

(CTV)

(BWTV)

R

BG

Video

AmplifiersMatrix Unit

to A, B, Cor to R, 1or G to A, ~ and C~to A, Band C

Matrix Unit Matrix outputs FSS Ttputs FSS outputs

r rTri-Color Pulse Height Differentiation Data Space

Osci lIoscope Analyzer Unit Sensor (DSS)

(TCO) (PHA) (DU) and SchmittTrigger (S1)

FIG. 2. System diagram of the IDECS data handling system.

and C channels of the FSS may also be fedinto a pulse height analyzer to produce prob­ability density functions and other statisticalmanipulations. The three channels of theFSS may also be sampled with a data spacesensor (DSS), including a number of SchmittTriggers used in concert; and the outputsfrom these devices in turn may be fed to theCTV to selectively enhance the data sampledby the DSS or the Schmitt Triggers. Finally,a differentiation unit is included which en­ables one or more images to be differentiatedand presented in different colors to the CTV.

Using the IDECS system we have studiedthe degree to which it is possible to extract se­lectively various levels of information on nat­ural vegetation employing HH and HV K­band images of Horsefly Mountain, Oregon.The Horsefly Mountain site was chosen be­cause the radar imagery displays notable dif­ferences in texture between and within differ­ent plant communities, and wide variations inthe gray scale response occur on the twopolarization images. In addition detailedfield studies on vegetation in this area and itsrelation to the radar image has recently beenconcluded by the senior author (Morain,1967).

BACKGROUND OF THE HORSEFLY

MOUNTAIN SITE

PATTERNS OF VEGETATION TYPES

In Figure 3 are shown the major vegetationtypes of Horsefly Mountain, based upon fieldstudy in August 1966 and extensive use of theU.S. Forest Service Timber Type Map, 1962,

of the area. The seven major vegetation typesare ponderosa pine forest, juniper woodland,white fir forest, hardwood forest, sagebrushon basalt rubble, chaparral shrub (old burn),grassland, and recently-burnt areas whichare almost entirely vegetation free. Figure 4is the HH and HV polarized K-band radarimagery used in this study.

Virgin ponderosa pine in this area forms avery open forest with a tendency towardsover-mature trees and a grass understory.However, virgin forest is rare in the Horseflyregion which has been extensively logged.What is left are a few relic large trees ofponderosa with various mixes of saplings andmedium-height pole timber growing uparound the remnants of the old forest. As thearea is a portion of Fremont National Forest,the young growth is protected. Consequentlythe amount of young growth in the area isprobably more than the normal. The denseyoung growth seems to cause slightly moreuniform gray tones on the radar image of theforest than would be true in the virgin ponde­rosa pine.

White fir tends to occur in monospecificstands only in the most mesic sites; however,it is widely scattered in various mixes between10 and 30 per cent in the understory of partsof the pine forest. Incense cedar is ubiquitous,but rarely constitutes more than 5 per cent ofa stand.

Juniper woodland tends to be a lower ele­vation community associated with morexeric spots, although pine and juniper mixin some areas. Juniper is notably free in as-

K-BAND RADAR IN VEGETATION MAPPING 733

sociating with other dominant forms ofvegetation in the area. It may have a sage­brush understory, a grass understory, nounderstory at all, or it may mix with pine.Some juniper stands may be very dense,covering as much as 50 percentof the ground,and yet in others the density may be 10 percent or less. All gradations may be seen andit is therefore difficult to establish what onemight call the normal structural form forjuniper woodland.

Sagebrush under normal conditions in thisregion would obtain heights of two feet or so,and would form a fairly continuous shrubstand. However, the forest service has

granted grazing rights to ranchers, and sage­brush as well as grassland has been heavilygrazed. Densities or the order of 10 to 25per cent are now common although in a fewsmall areas densities may reach up to 45per cent or even 50 per cent. But in any casemost of these bushes or individual shrubs aresmall.

Chaparral shrub normally occupies for­merly burned areas and consists of speciesof Arctostaphylos and Ceanosus. These shrubcommunities represent a successional stagefollowing forest fire, and for that reasonusually occupy well defined patches on pre­viously forested slopes. Chaparral may be

Major Vegetation Types at Horsefly Mountain, Oregon(boundaries corrected to geometry of radar imagery)

Reliability Diagram

USFS Timber~~, 1962;USGS 1:62,500 Quadrangle Map,

Bly, Oregon;field investigation conducted

August, 1966

Source:

[]ill]o

L

USFS Timber~M2Pand Field Investigation

Inferred

Lakes

Vegetation dominantly:

• Pine Forest

• JuniperWoodland

• White Fir Forest

§ Hardwood Forest

o sagebrush-basalt rubble

• Chaparral shrub- old burn

• Grassland~ Vegetation largely absent~ (recent burns)

• Unknown

FIG. 3. Major vegetation types at Horsefly Mountain, Oregon(boundaries corrected to geometry of radar imagery).

734 PHOTOGRAMMETRIC ENGINEERING

(a) HH Polarization

FIG. 4. HH and HV polarization, K-band radar imagery of Horsefly Mountain, Oregon.

mixed with areas of pine fe,;rest but this con­dition seems to be remnant of a previous suc­cession towards pine forest.

Essentially two kinds of grassland occur inthe Horsefly Mountain area: (1) mesic mead­ows with thick, continuous grass which mayoccasionally be marshy; and (2) less mesicforest openings with sparse and often shortgrass. Further details of these vegetationcommunities, and in particular the detailednature of the radar image texture, show asaverage gray scale values, and gray scalevariations within a single community will befound in Morain (1967).

SOME CONSIDERATIONS OF RADAR SIGNAL

BACKSCATTER FROM NATURAL PLANT COM­

MUNITIES

Cut-over ponderosa pine has a very markedradar image texture which normally permitsits rapid distinction from surrounding vegeta­tion types. In general, areas containing mixesof residual forest and an understory of sap­lings or pole timber rather consistently yield acoarser texture and a wider range of pointgray scale values than do young stands with­out residual forest. This may well arise fromincreased surface roughness which generatessignal returns from both tall and short trees.

K-BAND RADAR IN VEGETATION MAPPING 735

Juniper woodlands have no visibly distinctimage texture or gray scale associated withthem either on the HH or the HV images.The most characteristic expression on theHV polarization is a smear.

White fir forest has a texture and gray scaleclosely akin to the pine forest although thetexture of the whi te fir tends to be slightlyless coarse, and slight differences in the tworeturns on HH and HV imagery occasionallyappear.

Sagebrush shrub is difficult to distinguishon the radar from other non-forest types,including grasslands. Image texture is gen­erally quite fine in non-forest areas, andsimple visual inspection shows much thesame gray scale for the various vegetationtypes occupying them. The differences which

.exist are therefore subtle enough to requireelectronic and other data processing of theimages to see if some discrimination can beachieved. Much of the radar return in thesagebrush areas undoubtedly comes frombasalt rubble littering the surface, ratherthan from the spare, dry shrubs.

In general, sites with pure chaparral maybe recognized on HV imagery by their rela­tively fine texture in comparison to pine, theirmoderately high gray scale value, and bytheir position on slopes. Often the chaparralpatches are enclosed by pine forest and arenoted on the imagery mainly by breaks intexture. On HV imagery the image gray toneof chaparral lies between that of forest andmesic grassland. Gray scale cannot be used byitself, however, because it overlaps the grayscale values of a number of other vegetationtypes.

On HH imagery, burn patterns are noteasily detectable, whereas on the HV imagerythey normally are much more detectable. Forexample on the HH image it is not possible,by visual inspection alone, to disti nguishbetween most types of nonforest vegetation.In addition the textual detail on the HHimagery masks the occurrence of some of theolder burn areas because of the initiation of"forest texture" in places where tree regrowthhas progressed enough to affect the radarreturn. On the HH image alone it may bequite difficult to discriminate between thinspots in forest densi ty and burn areas in a latestage of succession. As this difference is moreecological than physiognomic (i.e. "geo­metric"), one would hardly expect any dis­crimination on the radar image.

Mesic and dry grassland types have low tomoderately low average radar returns respec­tively. Though they are often quite small

many mesic grassland spots are detectable onHH and HV imagery by their very darkappearance. Although the mesic grasslandsites have lush growth and hence a relativelyhigh dielectric (which would lead one toexpect a higher rather than lower radarreturn-see Simonett et aI., 1967) the radarreturns are low. We expect that this apparentanomaly arises from the tangled, mattedmatrix of the mesic grassland in which theradar signals bounce around, thereby sub­stantially decreasing the return.

The lighter gray tones (higher radar signalreturn) of the dry grassland may arise be­cause the grass blades are more vertical andless matted than in the mesic type. In ad­dition, part of the return may come frompebble-strewn bare ground between the grassculms. The influence of a uniform surface ofsmall pebbles on radar return in similar grass­lands in Escalante Valley, Utah is to increasethe amount of radar signal return (Morainand Simonett 1966).

INTERPRETATION AND DISCUSSION OF

IDEeS-PROCESSED IMAGERY

COLOR-COMBINED IMAGERY

A number of color combinations cannot beconsidered here because the subtle colorgradations are not reproducible in black andwhite. The four which are reproduced areexamples where the contrast in black andwhite is adequate for illustration of thetopics discussed.

Figure 5 shows a color-combined imagewith a forest-type map overlay which hasbeen adjusted to fit the geometry of the radarimage. The central medium gray area (whichappears violet on the color combination) issagebrush. The areas above and below withpaler tones (green on the color image) consistof ponderosa pine. Within the sagebrush arealittle pockets of pine forest are also discrimin­able. Areas of dark gray to the south of OttoBoy Flat and on the upper slopes of HorseflyMountain, (darker green on color combina­tion) correlate partially with areas of whitefir forest rather than pine. The area south ofOtto Boy Flat contains at least 50 per centpole timber of white fir coupled with someemergent relict pine. In this area of Oregonwhite fir frequently occurs on the more mesicnorth-facing slopes (as near Otto Boy Flat) orat higher elevations (as on Horsefly Moun­tain). The information on the upper slopes ofHorsefly Mountain is ambiguous because ofthe degradation in image quality along theedges of the color television system. I n ad­dition unremoved antenna side-lobing may

736 PHOTOGRAMMETRIC ENGIN EERING

FIG. 5. Color combination HH and HV radar image with map overlay.

also contribute to the return in this area.Finally, a number of very dark gray areas(deep blue on color combination) are mesicgrassland si tes.

In Figure 6 the map overlay is used as abase to display the highest part of the radarreturn signal on the HV polarization image.These areas of high intensity, shown as specksof dark gray (bright red on the color combina­tion), are dense sapling and pole timber in thepine forest. In a few instances emergent relictpine is also present. A number of small grayspots (red) just north of Otto Boy Flat are notshown on the forest service vegetation map.These may perhaps be pockets of densergrowth too small to show on the forest servicemap, although point-by-point stereoscopicexamination with air photographs indicatethat these sites could just as well be smalltopographic irregularities as actual dense

spot areas of regrowth. It is most difficult toreconcile each point on the radar image withthe aerial photographs because of the vastdifference in scale and information content.

In Figure 7 the clipping level on the grayscale of the HV image has been decreasedbut it is still apparently confined to ponderosapine-dominated regions, as no areas contain­ing pure stands of white fir or areas that havefairly heavy mixtures of white fir are empha­sized in gray (red on color slide). This is animportant distinction because where whitefir appears in monospecific stands it is usuallydense. On the HV image then it appears thatwhite fir gives fewer high-return point valuesthan do dense stands of ponderosa pine, ajudgement which is confirmed on inspectionof the probability density functions for fir andpine (Figure 9). Although in general the den­sity of vegetation directly affects the magni-

FIG. 6. HV radar image clipped at high intensity level.

K-BAND RADAR IN VEGETATION MAPPING

FIG. 7. HV radar image clipped at moderate intensity level.

737

tude of radar return (Simonett et aI., 1967),obviously other factors than density alonemust be involved in the white fir and ponder­osa pine areas. In the pine forest on the north­ern flank of Horsefly Mountain a number ofopen patches in the forest are also discrimi­nated at this clipping level. If further studyshows that radar is commonly sensitive todifferences such as these, then it seems thatthis type of analysis could be quite helpful instudying radar imagery of areas of very gentleslopes where topographic complexities are notpresent.

Two environments where such discrimina­tion may be valuable are: gently sloping areasin tropical savannas (which have notabledifferences throughout the year in the stateand condition of plant communities), and in

previously glaciated sub-artie regions wheremany complex mosaics of plant communitiesalso occur on gentle topography.

In Figure 8 the clipping level has beendropped to the lowest gray scale levels of theHV image. All the higher returns are dis­carded. All pale gray areas lie above the clip­ping level, all dark gray areas (red on colorslide) lie below. I t is noticeable that consider­able discrimination within the low-returnsagebrush area is being achieved. A numberof pale gray (non-sagebrush) areas, whichwere not mapped by the U. S. Forest Service,are in eality small clumps of ponderosa pinewithin the sagebrush region. The radarimagery therefore makes a distinction whichis related to the actual vegetation distribu­tion. Another area of interest is in the north-

FIG. 8. HV radar image clipped at a low level

,'~ ~,

MESIC GRASSLAND/~'\,,,-,,; \.

../ '~---------

738

HH

PHOTOGRAMMETRIC ENGINEERING

HV

PINE

WHITE fIR

SAGEBRUSH

JUNIPER

HH-­HV _nnn_

Increasing intensity~

FIG. 9. Image textures and probability-density curves for natural vegetation, Horsefly Mountain, Oregon.

ern and eastern fringes of Otto Boy Flat. Thearea of dark gray (bright red on the slide) inthis region may well represent a zone of bare

basalt rubble lying between the forest and thetrue grass flats in the center of Otto BoyFlats. This basalt rubble zone was not

IJ Sapling and sparse youngL.:U growth pole timber

..Dense sapling and moderatelydense to dense pole timber

~ Residual old growth with or without~ understory of second growth pine

D Pine not dominant

o>:...... ....... ---"~ MilesDerived from: USFS Map of Timber Types inFremont National Forest. Sheet *439, 1062

FIG. 10. Structural Variation in Ponderosa Pine Forest between PaddockButte and Horsefly Mountain, Oregon.

K-BAND RADAR IN VEGETATION MAPPING

HH

739

FIG. 11. Image Textures and Probability-Density Curves for Structural Sub-Types in PonderosaPine Forest. 1. Dense saplings and pole timber. 2. Dense saplings with emergent relict pine. 3. Sparsesaplings. 4. Sparse saplings with emergent relict pine.

mapped by the U. S. Forest Service, but wasobserved in the field. Much of the fine mot­tling in the central portion of the sagebrushzone is a .very mixed area of little stringers ofpine mixed with the sagebrush. The pinetends to occur in small clumps beading acrosssmall washes.

PROBABILITY-DENSITY FUNCTIONS

In our earlier publication on HorseflyMountain (Morain and Simonett, 1966)probability-density functions were given fOl anumber of inferred vegetation types to indi­cate the possible utility of such frequency/density curves in studying plant communi­ties. In the present study probability-densityfunctions have been derived for mapped andfield-checked vegetation types. The HH andHV radar images were masked with blackphotographic paper and areas on the imagecorresponding to each vegetation type werecut out. The regions were then exposed to theflying-spot scanner and sample data was fed

to a pulse-height analyzer to produce theprobability-density functions. The curves inFigure 9, although uncalibrated in an abso­lute sense, are internally consistent becauseall were obtained without changing instru­ment parameters.

Probability-density functions for whitefir, ponderosa pine (mixed relict and regrowthtimber), sagebrush, juniper and mesic grass­land are presented in Figure 9. The curveshave some variations in shape for each of tHenatural plant communities involved. Thesediffering curve shapes indicate that radarinteraction with plant communities (andother objects present, such as the basaltrubble in the sagebrush region) is dissimilarfor these types. The probability-density func­tions thus confirm that data-space samplingon a single image, or in two-space on twoimages coupled with color combinations, is aproper (i.e., valid) discriminatory tool instudying natural plant communities. Theprobability-density functions and the color

740 PHOTOGRAMMETRIC ENGINEERING

slides thus lead us to essentially the same con­clusion: the radar image contains subtle differ­ences which the unaided or untutored eyeat first sight cannot distinguish. The IDEeScolor combining system brings these subtledistinctions to the attention of the inter­preter, and thereby expands his detection anddiscriminatory abilities.

A map of structural variation in ponderosapine forest between Paddock Butte andHorsefly Mountain, Oregon is given in Fig­ure 10. Using the masking procedures de­scribed above, three pine structural su b­types were sampled for probability-densityfunctions. The latter, shown in Figure 11,are internally consistent, but as they wereobtained with a different instrument settingthan the samples in Figure 9, a comparisoncannot be made. The curves in Figure 11show that sufficient differences exist betweenthe val ious structural sub-types to warrantfurther study, although in general Morain's(1967) earlier concl usions, based on visualinspection of the imagery, that "Few differ­ences between any of the pine types can bedetected on HH imagery" and "The orthog­onally depolarized component of signal back­scatter (HV) from sub-types within the pineforest is relatively uniform," are correct. Thesub-types should be further evaluated by adocumentation of the range of differences

detectable by eye and by electronic samplingtechniques. Such a comparative study isbeyond the scope of the present article, butwill be performed by us at a later date usingfully calibrated system.

REFERENCES

Morain, S. A., 1967, "Field Studies on Vegetationat Horsefly Mountain, Oregon and its Relationto Radar Imagery," CRES Report No. 61-22,pp. 19, The Remote Sensing Laboratory, Uni­versity of Kansas, Lawrence, Kansas.

---and D. S. Simonett, 1966, "Vegetation Anal­ysis with Radar Imagery," Fourth Symposiumon Remote Sensing of Environment, University ofMichigan Institute of Science and Technology,p. 605-622. Also issued as CRES Technical Re­port 61-9, University of Kansas, Lawrence,Kansas.

Simonett, D. S., 1966, "Application of Color-Com­bined Multiple-Polarization Radar Images toGeoscience Problems." In: Computer Applica­tions in the Earth Sciences: Colloquium onClassification Procedures (D. F. Merriam, Edi­tor). Computer Contribution No.7, State Geo­logical Survey, Lawrence, Kansas.

Simonett, D. S., J. R. Eagleman, A. Erhart,D. Rhodes and D. Schwarz, 1967, "The Poten­tial of Radar as a Remote Sensor in Agriculture:L A study with K-Band Imagery in WesternKansas," a joint study by the Center for Re­search, Inc., University of Kansas, and KansasState University Agricultural Experiment Sta­tion, CRES Technical Report 61-21.

U. S. Forest Service, 1962, Timber Type Map­Fremont National Forest, Oregon, 1:31,680,Bonanza Quadrangle #439.

*MULTI-FILM INSPECTION

3000 FOOT LAMBERTSBRIGHTNESS

TABLE

PHOTOGRAMMETRY, INC.STOCK NO. 29

CONTINUOUSLYVARIABLE ILLUMINATION

MOTORIZED FILM

TRANSPORTS

ADJUSTABLE HEIGHTAND TILT

SIMULTANEOUS VIEWINGOF FOUR FILMS

12230 WILKINS AVENUE

ROCKVILLE, MD.