A Training Program for Men Measuring TreeHeights with Parallax Instruments

EVERT W. JOHNSON,

Department of Forestry, Agricultural Experiment Station,Alabama Polytechnic Institute, Auburn, Ala.

ABSTRACT: A program for the training of men in the use of parallaxinstruments for tree height measurement is described. This program iscontroUed by constant testing to establish when an operator has reachedhis natural level of competence, i.e. he is no longer able to systematicallyreduce his errors. The utility of several common statistical measures forcontrol of such a program is examined. The measures showing thegreatest promise are: the mean of the absolute errors, the mean of thealgebraic errors, and the standard deviation of the algebraic errors. Thelatter two are particularly valuable since they provide data for the evalu­ation of subsequent work. The mean algebraic error provides a measureof operator bias which can be used as a correction factor. The standarddeviation of the algebraic errors provides a measure of the residual ran­dom error which cannot be compensated without repeated measurements.

INTRODUCTION

T HE ability to control a training pro­gram is essential whenever a manual

skill is involved. In the organized buildingtrades this is accomplished by making eachbudding carpenter, plumber, or electricianundergo a relatively lengthy apprentice­ship during which he learns procedures andgains skill by constant practice under closesupervision. In the case of individuals withgreat inherent ability this period of ap­prenticeship is excessively long, but, sinceno absolu te standards have been estab­lished in these fields, it is necessary tomake all candidates complete the full pe­riod of training in order to protect thereputation of the trade. This is uneconomicuse of manpower. I t would be more desir­able to set up standards and to promote themen as soon as they are capable of meetingor exceeding these standards.

The problem of training in a specificskill was encountered in the course of a re­search project initiated in the Departmentof Forestry, Agricultural Experiment Sta­tion of the Alabama Polytechnic Institute.The project involved the measurement oftree heights on vertical aerial photographsusing parallax methods. Since skilled pho­to-interpreters were not available, it wasnecessary to train student volunteers tocarry out the work.

50

I n the initial planning phases of this proj­ect, it was felt that supervised practice fora specified number of hours would produceoperators with the required degree of skill.This idea corresponds with the apprentice­ship plan used in the building trades. Sub­sequently, however, it became evident thatthe main project would consume a veryconsiderable amount of time and, if theteam was not to be broken up by gradua­tion before the project was completed, thetraining program would have to be as shortas possible. As a result, a testing programwas developed that provided a runningcheck on the operators' progress, makingit possible to move the men to the mainstudy as soon as they gained a satisfactorylevel of proficiency. This program is de­scribed in the remainder of this paper.

REVIEW OF LITERATURE

A thorough search of the literaturefailed to yield much information thatwould help in setting up a training pro­gram. Sharp (4), in his discussion of educa­tion and training in photogrammetry, in­dicated that training in the use of instru­ments is usually received on the job orthrough short courses; he did not discusssuch training programs in detail. Custerand Mayer (1) stated that military photo­interpretation courses provide 12 to 52

TRAINING PROGRAM FOR MEASURING TREE HEIGHTS 51

hours of training in photo metrics, bu t hedid not specify how this time is allocatedor give any details on the training.

Getchell a nd You ng (2) have reportedon an experiment designed to determinethe length of ti me needed to reach a satis­factory level of proficiency with a parallaxbar instrument and with the parallaxwedge. This report provides some usefulinformation. Apparently only one opera­tor was used so that it is impossible todraw general conclusions safely from thisexperience. The training program used inthis experiment consisted of a short famil­iarization period with one of the instru­ments, and then 10 cycles of measurementson 20 trees of known height on a set ofvertical aerial photographs at a scale of1: 15,840. These 10 cycles required approxi­mately 18 hours to complete. At the end ofthis period, the operator was considered tobe reasonably proficient in the use of theinstrument. He was then given a shortfamiliarization period on the second instru­ment and the 10 cycles were repeated. Thesecond 10 cycles required only 12 hours tocomplete. The operator's proficiency at theend of the training program was tested bythe measurement of 21 new trees. In theirreport Getchell and Young did not analyzethese measurements statistically. The fol­lowing statistics can, however, be com­puted from the published data:

1) the mean error for the parallax bar in­strument was -0.8 feet, while that for theparallax wedge was -0.7 feet;

2) the standard deviation of the errorswith the parallax bar instrument was±2.72 feet, and that with the parallaxwedge was ± 3.78 feet.

Getchell and You ng were probably cor­rect in their assumption that the operatorwas sufficiently trained for most tree heightmeasuring operations. Tn personal corre­spondence wi th Getchell and Vou ng, theevidence resulting from this experimentwas confirmed from past experience by Dr.R. N. Colwell, of the School of Forestry ofthe University of California and formerlywith the U. S.. Jav)' Photo InterpretationCenter, and by K. E. Moessner of the na­tion-wide Forest Survey conducted by theU. S. Forest Service.

An unreported training program hasbeen used by Professor John C. Sammi, ofthe College of Forestry, State Universityof New York, while teaching a course con­cerned with the application of photogram-

me try to forest mensuration (3). The au­thor knows of this program from personalexperience. In this course a portion of thelaboratory work consists of making a tim­ber inventory using Spurr's (5) "plot ap­proach." Tn order to obtain reasonablysatisfactory resu I ts wi th th is proced ure,the photo-interpreter must be skilled intree height measurement. To gain a modi­cum of this skill Professor Sammi in­structed his students to measure theheights of six to eight objects, located onthe Syracuse University campus, for whichheight data were available. These measure­ments were repeated five or more timeswithin a given "set." The standard devia­tion of the height measurements on eachobject was then computed. Standard devi­ations for the several objects were thenaveraged to obtain the mean standard de­viation for the set. These mean standarddeviations were plotted over the nu mberof the cycle. Theoretically a trend shouldha ve developed that flattened ou t as theprogram developed, and finally showed amore or less constant value, indicating thatthe operator had reached his natural levelof proficiency. In practice several draw­backs to this plan developed. The mostcritical of these was that too few objectswere measured in a single cycle within aset. Several cycles could be completed in asingle laboratory period, and consequent­ly memory bias became a very significantitem. However, this general method, ifproperly administered, has the great ad­vantage that development of operator skillis under constant surveillance; and, whenthe operator reaches a natural level ofcompetence, he can be graduated to moreserious work.

PROCEDURE

Because it provided a guide to the de­velopment of operator skill, Professor Sam­mi's basic idea was used as the foundationof the training program. It was modified insome respects in an attempt to make itsomewhat more effective. In addition, itwas expanded in order to provide informa­tion concerning the relative merit of sev­eral common statistical measures as guidesto operator competence.

A group of six students began the pro­gram. These men met as a class at the be­ginning of the indoctrination phase. In­struction in properly setting up a stereo­scopic pair of photographs and in operat-

52 PHOTOGRAMMETRIC ENGINEERING

\

- OPERATOR I'-- - OPERATOR II:----- OPERATOR m

,I \

\

//

/

6

0:o

2 0:tr

"'3 z.."'.4 :>;

°

- 8 L-'--;;-~3'--~4.-.--,T-'i~FIA=-L'7'-'8"9'''10''111--T:12

FIG. 2. Learning curves. Mean algebraic errorsfor the parallax wedge.

were used as a guide to the operator's de­velopment (see Figures 1 through 6). Thesetrials were repeated, with at least one daybetween successive trials, using the sameinstrument, until the operator was judgedto be ready for serious work. The strict se­quence and the required time lapse be­tween successive trials with the same in­stru men ts were specified in the programdesign in order to reduce, as much as pos­sible, the effect of memory bias_

In order to keep the development ofskill with one instrument approximatelyparallel with the second, a trial was firstmade with the parallax wedge on a givenset of 10 trees, and then this was followedby a trial with the Contour Finder. Thenboth instruments were moved to a new setof trees. This, of course, meant that thesame trees would be measured twice inclose succession, but, since the instrumentswere so different in design and operation

8~---------------,

7~z

+4r-:----~:_:_:::__----------,;-- OPERATOR I.- ~- OPERATOR II;-- OPERATOR m

II 12108

.- OPERATOR I--- OPERATOR II-_..- OPERATOR m

TRIAL5 6 7432

ing the instruments was provided. The in­struments used were U. S. Forest Serviceparallax wedges, printed in red on a flex­ible film base, and an Abrams ContourFinder. Every technique known to the au­thor concerning use of these instrumentsfor tree height measurement was demon­strated and explained. The operators wereurged to try these techniques and to usethose that seemed most effective. Thisfamiliarization period required approxi­mately 3 hours and the time seemed to bewell-spent.

Three sets of vertical aerial photographswere provided for the experience-gainingphase of the program. These were standardU. S. Department of Agriculture photo­graphs taken in late winter of Lee andMacon Counties, Alabama. The sets cov­ered areas near Auburn, Opelika, andTuskegee. Ten trees, whose heights hadbeen measured with an Abney level, weremarked on each set. An attempt was madeto secure a wide range of species andheights. True average photographi.c scaleswere determined from measured distancesor from com pu ted distances betweenbench marks. From these data the truedifferential parallax for each tree was com­puted. These true differential parallaxesprovided the standard for subsequent test­ing of the operators.

I nstructions to the operators stated thatthe three sets of 10 trees each were to bemeasured one after the other in a strict se­quence. A group of 10 tree height measure­ments made with one instrument from asingle set of photographs was calle~ a"trial" and the statistics for such tnals

10

9

8'; \<.> \z \

7: \00

6~

0:50

0:0:

4'"z..

3'"2

2

0

FIG. 1. Learning curves. Mean absolute errorsfor the parallax wedge.

FIG. 3. Learning curves. Standard deviations ofthe absolute errors for the parallax wedge.

TRAINING PROGRAM FOR MEASURING TREE HEIGHTS 53

Br------------------

FIG. 4. Learning curves. Standard deviations ofthe algebraic errors for the parallax wedge.

,\1\, \: \

I \I \I ' 0\I \ / \ r

/ \/

130'1-------:-~~-------__..- OPERATOR Ien OPERATOR n~- OPERATDR m120

110

100~

z90 ~

a::BO ~

70 zo

60~it'

50~

FIG. 5. Learning curves. Coefficients of variationof the absolute errors for the parallax wedge.

mean absolute errors. As can be seen inFigure 2, trends for the individual opera­tors became evident as the program pro­gressed. Theoretically, as the operatorsgain experience, these trend Jines shouldapproach, then flatten out, and finallyparallel, the line of zero error. The meandifferences between the trend lines and thezero error line are measures of the opera­tors' biases in making differential parallaxmeasuremen ts. If these differences aresubtracted algebraically from measure­ments made by the respective operatorsunder similar conditions, the effect of thebiases can be greatly reduced or possiblyeliminated. The residual errors would thenbe non-correctable variables best measuredby the standard deviation.

I n the same man ner as wi th the meanerrors, the standard deviations of the abso­lute and coded errors were computed andplotted over the number of the trial (seeFigures 3 and 4). Theoretically, the result­ing curves should have trended downwardand then leveled off indicating that as theoperators became more experienced theybecame less erratic. I n general, the actualcurves agree with this theory. They showthe downward trend but relatively littleevidence of leveling off. A larger number oftrials undoubtedly would have provided abetter example of this phenomenon. Afteran operator attains stability and his curveof the standard deviation of coded (alge­braic) errors remains relatively flat, themean difference between his trend line and

- OPERATOR [h_ OPERATOR n

OPERATOR m

£7°z

and since they were graduated in differentunits, it was felt that memory bias wouldhave little effect on the readings.

In an attempt to further reduce the ef­fect of memory bias, the operators were in­structed to record the instrument readingsat the top and bottom of each tree, but notto do any computation. The computationoperations were all carried out by the au­thor. The measured differential parallaxeswere compared with the computed values.If suggestions for improvement of tech­nique could be made, they were passed 011

to the operators. I n no case, however, didthe operators see their actual results untilafter the testing program was complete. Inthis way development of tree height meas­urement ability could be studied with rela­tively few photographs.

The first computational step was to de­termine differences between the computedand measured differential parallaxes. Thenthe mean absolute difference, or error, foreach set of 10 trees, \\"as computed andplotted over the number of the trial (seeFigure 1). As the operators became moreexperienced their mean absolute errorstended to become smaller un til they stabi­lized at relatively constant values. \iVhenthis occurred it was felt that the operatorshad probably reached their natural levelsof competence and were considered "ex­perienced."

The algebraic errors were then exam­ined. First the negative error values wereeliminated by the addition of a constant(coding). The mean algebraic error for agiven trial was then obtained by subtract­ing this constant from the mean codederror for the trial. The mean algebraicerror was then plotted over the number ofthe trial, as was done in the case of the

688

5 zo

4>=..;;

3~

o2 a::..

oI~

t-

o LI/l-.----~2-~3-A4-c:5--1.!TR~;.:~L'--.7,--.B,--.9,--7.10--71.,..1-----;:12

54 PHOTOGRAMMETRIC ENGINEERING

100,.------------------,

FIG. 6. Learning curves. Per cent correlation(100r2) between computed and measured dif­ferential parallaxes.

pattern is revealed in Figure 6. It is inter­esting to note that the levelling off portionof the curves for all the operators complet­ing the program lay between 60 and 70per cent. The correlation coefficient, whileinteresting, is, like the coefficient of varia­tion, quite sensitive to minor changes inthe error pattern and shows considerableinstability. As a result it was not used inthe evaluation of operator competence.

DISCUSSION

The con trolled program of trall1l11g asdeveloped for this project contained faultsand weaknesses, but it is believed thatthese were not due to the fundamental de­sign of the program. Instead, they weredue to the rather peculiar circumstancesunder which the program was carried out.The operators were senior students in theDepartment of Forestry of the AlabamaPolytechnic I nstitute. These men hadheavy academic loads and had to fit thephotogrammetry training into their sched­ules where it would disrupt their normalschool activities as little as possible. Thiscaused the training to be more erratic thanshould have been the case. This also madeit necessary to terminate the training be­fore the leveling out of the training curveswas clearly defined, in order to provide suf­ficient time for the operators to finish themain study before they graduated.

A time record was kept for wage pur­poses. This record substantiated Getchelland Young's (2) statement that it takesfrom 12 to 18 hours to become reasonablyproficient in the use of a parallax measur­ing instrument. In this study, however,the operators trained on both instrumentsat the same time, rather than completingthe training on one and then shifting tothe other. The three men completing thetraining program required a total of 21 to26 hours to attain reasonable proficiencywi th both instru men ts.

In summary, it is evident that a trainingprogram such as this would be very val­uable to any organization utilizing paral­lax instruments for measuring objectheights. With it a constant check is avail­able on the operator's development, andhis ultimate ability can be assessed. If heoperates with a constant bias, it is re­vealed, and compensating factors can beapplied. If a man requires an unusuallylong period of training to hpConme profi­cient, or if a man shows no talent for the

- OPERATOR I--- OPERATOR n

. OPERATOR m\\

\.l,1\,I"",I :\

J \/ -,f 1\": \ \

I "I \ \I I': \ \

I \ \

, \ \ i "./o'---;----,.----,\~.'r,'____.---'-'F'__.-.._____.,.___,=-__;_;____:!,

3 4 5 8 9 10 I I 12

10

90

80

70 ~

60 ~...llC

50~o

40 I­Z...

30 0

llC...20 "-

the zero error line would provide a measureof the random variability in measurementsto be expected from that operator, underthe specific conditions existing when thedata were obtained.

I n most cases, when the coefficien ts ofvariation of the absolute errors wereplotted over the trial numbers (see Figure5), the scatter of points was so great as toprevent detection of any but the crudestof trends. Most such trends appeared to beessentially horizontal indicating that, asexperience grew, the reduction in scatter oferrors tended to keep in step with the meanerror. There were, of course, exceptions tothis rule. For example, see Operator IIIon Figure 5. This man reduced his meanabsolute error quite rapidly compared tothe rate of reduction of his variability.This resulted in an upward trend of the co­efficient of variation. Actually this statis­tic is of minor importance when comparedwith the mean error and the standard devi­ation of the errors, for purpose of evaluat­ing operator competence. It is unneces­sarily sensitive to minor changes and doesnot reveal much information not alreadyavailable. For these reasons it was not usedas a judging tool.

The last statistic considered as a possiblegauge of operator competence was the cor­relation between the computed and meas­ured differen tial parallaxes. This was ex­pressed as the correlation coefficient. Theo­retically, when the correlation coefficientsare plotted over the numbers of the corre­sponding trials, there should be an up­ward trend and then a levelling off as theoperator becomes more experienced. This

TRAINING PROGRAM FOR MEASURING TREE HEIGHTS ss

job, it becomes evident as the program de­velops. Indeed, it appears to be a very use­ful tool for the training of personnel en­gaged in the application of photogram­metry to forestry.

LITERATURE CITED

1. Custer, S. A. and Mayer, S. R. 1955. "AComparative Analysis of Curricula andTechniques used in the Training of Photo­graphic Interpreters." Boston Univ. OpticalResearch Lab. Tech. Note No. 119.

2. Getchel, W. A. and Young, H. E. 1953.

"Length of Time Necessary to Attain Pro­ficiency with Height Finders on Air Photos."Univ. of Maine Forestry Dept. Tech. NoteNo. 23.

3. Sammi, J. C. 1957. Personal correspond­ence. Professor of Forest Mensuration, Col­lege of Forestry, State University of NewYork, Syracuse, N. Y.

4. Sharp, H. O. 1952. Education and training.MANUAL 01' PHOTOGRAMMETRY, 2nd. Ed.Amer. Soc. Photograml1letry, Washington25, D.C.

5. Spurr, S. H. 1948. "Aerial Photographs inForestry." Ronald Press Co., New York,N. Y.

~hlUp !B. 9(alf df~~oclate~DESIGNERS MANUFACTURERS1601 ELIOT STREET • DENVER, COLORADO

DOUBLE REFLECTING PROJECTOR

Wr3te for literature end prices.

, An accurate projectorbuill into a table

• Compact. low cost

• Range: Y3 reduction to3X enlargement

o 12" F:4.5 anastigmatcoated lens

• Dimensions: 35" wide.72" long. 36" high

Simple to operate andadjust

Fstahlishf'd 1910

JAMES W. SEWALL COMPANYOLD TOWN. MAINE

FOil EST ENGINEEI:ll"GTOPOGIlAPIIlC ~IAPPING

AERIAL SURVEYS

BRAl'lCH: James W. Sewall Co. (Canada) Ltd.Frednicron, N'ew Brunswick