Sie sind auf Seite 1von 10

Published March, 1944




I N various investigations of the effects of grazing upon forage pro-

duction and utilization, one of the most difficult problems is an
accurate appraisal of the single factor of greatest importance, v/z., the
volumeof forage itself. For a reliable picture of .these effects it is
necessary to obtain accurate estimates of the actual amountof forage
produced or remaining after treatm6nt on each area subjected to
experimental control. This task is difficult simply because forage
varies considerably in the weight of plant material produced by each
species in a highly variable population.
Since all the forage cannot be harvested and weighed, it is necessary
to obtain a reasonable estimate of the true total weight by sampling.
Using some standard method, such as clipped plots, the. sampling
procedure is relatively simple in principle. It is only required to clip
enough plots, distributed over the pasture by some efficient scheme
of randomization, to provide an average forage weight which is ac-
curate within prescribed limits. The numberof clipped plots necessary
to provide a re]iable mean, however, is generally large. Beruldsen and
Morgan (I)a found that e S independent observations were a minimum
number per .sample for acceptable accuracy under their Australian
pasture conditions. Davies (4), who also worked in Australia, con-
cluded that the sampling errors of small samples are of considerable
magnitude. Ellenberger (S) and his associates in Vermont observed
that the weight of forage clipped from small plots varied greatly be-
tween pastures and from place to-place within pastures. Robinson,
Pierre, and Ackerman(7) used nine cages per pasture to protect plots
to be’ clipped and found the sampling errors of the means of these
nine. observations too large for dependable interpretation of results.
In actual practice, therefore, the sampling process is not an e.asy
task, particularly when field observations have to be taken within a
short period of time. Ordinarily, data on forage production, for ex-
ample, must be obtained within the space of a week or two; and data
on residual forage on summerrange land must be obtained after the
]ivestock have been removed and before autumn snows make the
sampling task impossible.
In experiments on summer ranges we have found it difficult to
reconcile these requirements and the limitations of available funds
and personnel with the demands imposed by a highly variable popula-
tion of forage. The number of plots which could be clipped simply did
not provide sufficiently reliable information; observed differences
amongpasture averages could be at.tributed to little more than the
chance variation contained in sampling errors. In such a stalemate we
1Contribution from the Rocky Mountain Foregt and Range Experiment Sta-
tion, Forest, U. S. Dept. of Agriculture, maintained in cooperation with
Colorado State College at Fort Collins, Colo. Received for publication August I I,
~Silviculturist, Forest Ecologist, and Senior Clerk, respectively.
3. in parenthesis refer to "Literature Cited." p. 2o

have been faced by two alternatives, either to abandon the forage

studies or to devise some short-cut method which would provide the
necessary information within the prescribed limits of time and funds.
l~or obvious reasons the second course was chosen, and it then be-
came necessary to select promising short-cut methods and test their
efficiency by actual field trial.

After consideration of a number of methods for measuring various
factors which might be associated with forage weight, we finally
selected two short-cut methods as being the most likely to fit our
needs. One, the line-transect, has been used in various fields and was
adapted to forage measurements by Canfield (2, 3). The other, which
simply involves the estimation of forage weight on sample plots (5),
has been tried by a number of investigators.
If used without quantitative control, neither of these methodspro-
vides data on forage weight directly, and they are not capable of
rigorous test to determine the accuracy of information on forage ob-
tained by their use. By the clipping of forage on someof the estimated
plots and in a belt around some of the line transects, however, the
data obtained by these short-cut methods could be converted to
quantitative weight estimates. Both of the methods seemed to prom-
ise considerable reductions in field time requirements, and both ap-
peared simple and easy to apply. And’finally, quantitative estimates
of the error of forage weights calculated from the field data could be
obtained by double-sampling analysis.


As the name implies, double-sampling involves the sampling of any
population by two methods, one which yields data directly on any
desired factor such as forage weight, but is laborious and expensive
to use; and another (short’-cut) which yields data on some factor
which is highly correlated with the desired factor (8) and is much
cheaper to use. Use of any short-cut method in the sampling of forage
yield causes a loss in precision per plot, as it can provide only an
approximatipn of the actual weight of forage on each plot through a
regression of forage weight on the factor observed. Since this method
is cheaper to apply in the field, however, so many more sample
observations may be taken that a net gain in precision or efficiency
may result. The amount of gain, if any, depends on the relative cost
of double-sampling as compared to clipping, and on the relative
accuracy with which forage weight can be calculated from the re-
In field forage sampling a number of randomized observations is
taken by someshort-cut method, and in addition the forage is actually
clipped at a relatively small number of sampling points, taken at
random from the larger sample. Thus we have available two sets of
data, a large sample containing only observations taken by the short-
cut method, and, within the large sample, a small sample containing

in addition the actual weights of forage clipped at a portion of the

sampling points observed in the large sample.
Then, using only the small-sample data, a regression is calculated
to showthe relation of actual forage weight (Y) to the factor observed
by the short-cut method (X). In itself this regression adds nothing
to the information on forage weight provided by the small sample
alone. Wehave also, however, a relatively precise estimate of X,
provided by the mean of the large sample (including values for
derived from the small sample) ; and by solving the regression equa-
tion for the large-sample mean of X, a correspondingly more precise
estimate of average forage weight may be derived. Also, since both
the large and small samples were taken by randomization, an error
variance can be calculated for each portion of the regression equation;
and thence can be obtained the error of the estimated forage weight.
This method can be substantially refined by segregating the forage
into individual species or plant classes (groups of similar species), and
making a separate analysis for each. Estimated mean forage weight
can be calculated for each plant class from its ownregression equa-
tion, and the results summedto give the estimated weight of all
classes of forage on the pasture. The analytic procedures required
for obtaining the regression equations and calculating the error
variances of the resulting estimates and of their sums are nicely dis-
cussed by Schumacher and Chapman(8, chapters VIII and XI).
As indicated above, the efficiency of double-sampling is affected
by the relative cost of clipping as comparedto double-sampling, and
by the accuracy of the regression equation. Hence it may correctly be
reasoned that efficiency depends also on the numberof plots that are
clipped as compared to the total number of sample observations. If
too many plots are clipped, the cost of-sampling becomes unneces-
sarily high, while the use of too few clipped plots results in an un-
reliable regression equation. Thus, it is desirable to estimate the
proportion of clipped plots which may be expected to provide maxi-
mumprecision with minimumwork.
For a regression involving a single, independent variable, the
optimum proportion of the total number of plots to the number
clipped may be calculated if the following factors can be estimated
with reasonable precision, viz., the variance of estimate of the re-
gression equation (Vy.x); the regression coefficient (b); the variance
of X (VxL); the approximate costs of obtaining the values of X and
Y at each sarnpling point (Cx and %); and the cost of travel between
sampling points (cd, including setting up equipment at any new
point. Then the optimum ratio of the total number of observations
(nL) to the number of clipped plots (ns) may be estimated approxi-
mately from the equation

For this approximate calculation the travel cost per plot may be
taken as essentially constant with varying numbers of sample observa-
tions, within the limits imposed by a fixed total cost of sampling.

In order to fit the studies of sampling methods into active range experiments
with maximumefficiency, we tried out the two methods at two different locations.
The line-transect trial was made in experimental pastures at the Manitou Experi-
mental Forest, in the headwaters of the South Platte River, while the efficiency
of weight estimates was tested in a study of the influence of gopher removal on
forage production, located on the Grand Mesa National Forest in western Colo-
rado. Thus, line transects and weight estimates could not be compared directly
with each other, but the efficiency of each method could be compared with that
obtained by clipping quadrats.

Experimental area.--At the Manitou Experimental Forest six pastures, each
25o to 3oo acres in size, have been established to study the influence of three in-
tensities of grazing by cattle on forage production, beef yields, erosion, and in-
The forage on these pastures is primarily a bunchgrass type existing as an
understory in an open ponderosa pine stand. Park-like areas from a few square
rods to several acres in extent are dominated by bunchgrasses. In areas occupied
by open stands of ponderosa pine the same species persist but in lighter density
than in the park-like areas. In spots where the canopy is dense, and directly be-
neath the pines, a few sedge plants are the only species present.
Field procedure.--In ascertaining forage production, a total of 36 plots was
sampled in each pasture. At each plot were measured the ground-level diameter
(to the nearest o.o~ foot) and the average height (to the nearest o.Io foot) of
portion of every plant which touched a 3o-foot cable stretched between two iron
stakes. These plants were classified as to whether grazed or ungrazed, and segre-
gated into four classes, v/z., tall bunchgrasses, short bunchgrasses, single-stemmed
and sod-forming grasses and grasslike plants, and weeds (forbs). Browse species
were ignored, as they form a minute part of the palatable forage in this cover type.
Six of thh 36 transects were selected at random for clipping. After the line-
transect data were obtained, the vegetation was clipped from a 6-inch by 3o-foot
belt transect surrounding the 3o-foot line-transect. The forage was sorted by
classes, air-dried, and then weighed to the nearest gram. In addition to these data,
each crew kept accurate notes on the amount of time consumed in clipping, tally-
ing, travel between plots, and handling the clipped forage.

Experimental area.--The Grand Mesa experiment is factorial in design. Its
objective is to discover .the influence of grazing by cattle and gophers, separately
and together, on forage production and related factors. Sixteen I-acre areas have
been established, four in each of four locations or blocks. The four treatments,
cattle and gophers, cattle alone, gophers alone, and no grazing, have been assigned
at random to the 4 acres in each block.
The study area is on comparatively level to rolling land at an elevation of ap-
proximately lO,5OOfeet, where the vegetation consists of open grassland and sage-
brush interspersed with stands of Engelmann spruce. All observations were con-
fined to the park-like areas of sagebrush and grass-weed communities, since the
ground beneath the spruce trees is practically devoid of forage. Grasses constitute
approximately 2o% of the vegetal cover, varying from 8 to 3o%, depending

largely on the intensity of use in the past by grazing animals. Weedsare abundant,
constituting from 55 to more than 90% of the cover. On approximately one half
of the study area sagebrush is the dominant plant, the understory consisting of
grasses and weeds.
Field procedure.--For sampling the forage, each I-acre area.was subdivided into
nine strata and two sample plots were assigned at random to each stratum. These
plots, each 12. 5 square feet in drea, provided the large sample. For the small
sample, five plots were drawn from the 18 plots in each acre.
Estimates of green forage weight were obtained by one man, who had previously
trained himself by estimating the forage 6n a number of plots and checking his
estimates by clipping and weighing the forage. In sampling the experimental plots,
each species was estimated separately; then the species estimates were combined
into three classes, grasses, weeds, and shrubs.
After the plots had been estimated on a single acre, a~lother worker clipped the
forage on the five plots forming the small sample, segregating it into the three
classes and weighing the clip in green condition. The estimator was then permitted
to compare his estimates with the actual green w.eights as a running control on
subsequent estimates. Finally, the clipped forage was air-dried and reweighed ~o
the nearest gram.

In Table 1 are presented," together with their standard errors,
numerical estimates of the air-dry weight (in pounds per acre) of each
class of forage on the six pastures as calculated by the double-sampling
method. For simplicity and because they were similar in magnitude,
the separate pasture standard errors were pooled to provide a single
value for each forage class.
I.--Average estimated forag~ weight in pounds per acre per pasture,
by forage classes.

Forage class
Pasture Total
No. forage

37.6 I54.6 19.9 41.3 253.4

26.0 174.o 13. 48.3 262.2
44-9 167.6 16.4 57.0 285.9
36.6 112.2 14. 4o.6 2o4.I
42.9 I75.
5 46.6 283.1
50.3 162.9 23.5 37.1 273.8

238.3 946.8 I06.

5 270.9 1,562.5
Mean ................ 39.7 157.8 17.7 45.2 260.4
Standard error ......... 4-13.
4 4-24.0 4-4.6 =t=8.2 =1=29.o

From these standard errors it is evident that the information on

the separate classes is not very precise, although the mean values per
pasture for total forage are considerably more accurate. These, in
fact, with a standard error of about i i%, may be considered for our
purposes to provide satisfactory information on the amount of forage
remaining on these pastures at the end of the grazing season.

Several pertinent facts resulted from the double-sampling analysis.

For example, the independent variable (summed diameter times
height) was well correlated with air-dry forage weight; only the short
bunchgrass and weed classes showed relatively poor correlation. In
the tall bunchgrass class, 78%of the variation in forage weight was
associated with the independent variable; corresponding figures for
short bunchgrass, single-stemmed grass, weeds, and total forage were
52, 77, 42, and 68%,respectively.
Also, in this study we found that the individual pasture regressions
were homogeneous, and that a single regression could be used with
equal accuracy for all six pastures. This fact contributed to the
efficiency of the double-sampling method, since the estimated forage
weights per pasture could be computedfrom a single regression based
on the 36 small-sample observations obtained in all six pastures
rather than from six relatively weak regressions based on only six
observations apiece. As a corollary of this fact, it turned out that we
employed almost the optimum number of clipped plots as compared
to the total numberin each pasture. As calculated by the equation on
page ~95, the optimumratio of the total number of observations to
the number of clipped plots was about L~3 to i.oo; and since the
regression was based on 35 observations, the 35 large-sample.observa-
tions per pasture gave an actual ratio of i.oo.
As another feature of the line-transect method, relative figures on
plant cover density and forage utilization could’be derived from the
large-sample data. In order to get similar figures from a sample
containing only clipped plots, we should have had to make separate
density estimates and segregate the clipped forage into grazed and
ungrazed plants. This process would have required at least 5 to
minutes additional time per plot.
Now,in examining the actual relative efficiency of double-sampling
as compared to clipping all plots, we can make comparisons based
either on field time requirements or on total time, depending on
whether or not field time limits the size of the sample. Considering
field time alone, 4 ~6 belt transects could have been ,clipped in the
time required for tallying 36 line transects and clipping six belts.
Comparing the resulting variances of mean values, the double-
sampling procedure provided an increase in information of about
~8%. If, on the other hand, the comparison is based on total time
expended, the double-sampling method provided only about
more information than clipping all plots. Actually, the office time
required for the rather complex double-sampling analysis largely
offset the savings in field time.
If we wish to apply these results to studies in other places, the
several components of field time becomeimportant in appraising the
relative economyof double-sampling. In our field samples the actual
tallying of line transects, including setting up and removingthe cable,
required I~.4 man-minutes per transect; clipping took up 3~.4 man-
minutes; and travel between plots, 22.2 man-minutes. If, in another

4Including training men, clipping, tallying, setting up and removing equipment,

weighing forage, and travel between plots.

study, the plots were much closer together, the double-sampling

procedure would be relatively more efficient because travel time
would be reduced. In the survey of large areas, on the other hand,
with plots relatively far apart or hard to locate, the advantage of
double-sampling with line transects might disappear entirely; travel
time would be so large that the additional cost of clipping all the
plots would form a very small part of the total field time.

AS shown in Table 2, the forage on these 16 l-acre areas was esti-
mated with satisfactory precision. Of the three forage classes, only
browse showed poor results; and the sampling errors for total forage
are considerably smaller (in percentages of forage weights) than those
observed in the line-transect method. By itself this observation does
not favor weight estimates, however, as the sampling was consider-
ably more concentrated than in the Manitou experiment.
2.--Average estimated forage weight per acre, by forage classes.

Forage classes
Acre desig- Total,
nation lbs.
Grasses, lbs. Weeds, lbs. Browse, ibs,

I-A 1;9;21" 3724-46 °

370±5 9Olq-86
I-B °5644-4 1,o9o-t-62
I924-2I 334±35
I-C 89-4-~3 5144-54 t54-4-39 756=k66
1354-23 8494-47 2754-73 ~,259-q-8~

2-A 494- 8 5404-36 i6-4-~6 6o4:t:41

723~4 764-22 9114-4o
~± 9
2-C 564- 6 6524-45 214-t
4 729+50
2-D x49±~8 919~33 54:t:I8 I~I224-40

3-A 954-20 244~:32 334-21 3734-46

3-B ~384-i.6 4614-43 414-16 64o=k59
3-C 95-4-I I 9oo~92 41±22 1,o36-+-1o
3-D 236-I-22 973=k8o 794-27 1,2884-98

4-A ~37±~5 4274-53 804-20 6444-61

4-B ~8q-~
7 549=k29 38-1-i2 7054-43
4-C ~58q-~ 6194-39 574-i7 834±49
4-D 1964-21 99o4-1o4 684-24 1,2534-I23
*Therlght-handfigure in each cell is thestandarderror of the averageforage weight,in pounds
per acre.

Perhaps a better criterion for judgment is provided by the relative

precision of regressions of air-dry forage weight on estimated green
weight. In genera], these correlations were somewhat better than the
line-transect results; 78% of the variation in the weight of total for-
age was associated with weight estimates, and corresponding data
for grasses, weeds, and browse were 79, 77, and 72%, respectively.
As-to relative time requirements, only 6.7 man-minutes per plot
were required for estimating green weights, as compared to 12.4 man-
minutes for tallying a line-transect; this considerably lower figure in
spite of the much heavier stands of forage on the Grand Mesa. Travel

~cime was negligible on these small areas and was therefore included
in the other two components, while the actual clipping required 68.4
man-minutes per I2.S square-foot plot as compared to 32.4 man-
minutes per ~ 5 square-foot plot in the Manitou pastures.
As observed in the line-transect study, no significant differences
existed amongthe individual acre regressions. Therefore, it was pos-
sible to use a single regression (based on the pooled "within acre"
squares and products) for estimating the mean forage production on
each of the ~6 acres. Since, without previous experience, this fact
could not have been predicted beforehand, we used a rather inefficient
proportion of the total numberof plots per acre to the numberclipped
in each acre. As calculated by the equation on page 196, this ratio
should have been about 5.7 estimated plots for each clipped plot.
Although 18 plots were estimated and 5 of these plots were clipped
on each acre, we actually used a ratio of o.22 to 1.oo (that is, 18
estimated plots to 8o clipped plots in the regression) in estimating
the mean forage production per acre. The efficiency of double-
sampling would have been materially improved if only one or two
plots had been clipped on each acre and the time thus saved had been
used in estimating a larger numberof plots per acre.
Even with this relatively inefficient arrangement, double-sampling
with weight estimates provided a substantial increase in information
as comparedto clipping all plots. Onlyabout 6.8 plots could be clipped
in the field time required for double-sampling with 5 clipped and 18
estimated plots per acre. As a result, the latter methodprovided about
37%more information on total forage than that supplied by clipped
plots. Whenthe comparison is based on total time (field and office),
the gain in information dropped to about 14%. These gains are not
greatly different from the iine-transect results. If, however, only one
plot had been clipped on each acre and the surplus time used in
estimating about 4o additional plots per acre, the variance of the
calculated mean forage weight would have been reduced by about
one third, and the relative efficiency of double-sampling ~ould have
been correspondingly increased.

As a result of these analyses, we have a definite basis for quantita-
tive comparison of each double-sampling method with the cost of
clipping sample plots in our experiments, and a qualitative basis for
.comparing the two methods with each other.
As employed in these studies, both short-cut methods provided
a substantial saving in field time and some economyon the basis of
total time expended. The achievement of similar economies in other
studies would probably depend, however, on the relative amounts of
time required for clipping, traveling to and from the survey area and
between sample plots, and applying the short-cut method. In experi-
ments requiring intensive sampling by trained men, with sample
plots located close together, either of the double-sampling methods
maybe expected to provide substantial increases in efficiency as com-
pared to clipping all sample plots. In large-scale, extensive range

surveys, on the other hand, travel mayoften be the largest component

of total cost; especially in surveys of remote mountainareas, involving
transportation by horse and repeated camps in relatively inaccessible
areas. Under such conditions it may frequently be found that the
cost of clipping all sample plots forms a small part of total costs, and
yields rich returns in the precision of survey information as compared
to the usual qualitative estimates of forage density and similar factors.
Under critical scrutiny the sampling of such relative factors, without
the quantitative check provided by double-sampling, tells the in-
vestigator only that one sample average is larger or smaller than an-
other. It fails to give him quantitative information on the actual
amounts of forage present on each sampled area and on the error of
each sample average in terms of forage weight.
In comparing the two double-sampling methods v~ith each other,
our only definite statement can be that they showed no substantiM
difference in efficiency as used in our studies. Weight estimates may,.
however, be judged superior to line transects, simply because they
required less time in the field and seemed to give at least equally
precise estimates of forage weight. Also, as we used it, the weight-
estimate methodsuffered from the use of an inefficient proportion of
the total number of observations per acre to the number used in the
sr~all-sample regression. With a more efficient design, this method
might have turned out to be decidedly superior to the line-transect
In general, our feeling is that these studies have not solved the
problem of sampling range forage efficiently, although they have pro-
vided a step towardits solution. The principal difficulties still remain,
viz., forage is highly variable, with the result that adequate sampling
must still require a considerable expenditure of time.

Two double-sampling methods, using line-transects and forage
weightestimates,weretested to ascertaintheirrelative efficiency
estimatingthe amountof foragepresenton experimental areas,as
comparedto theclipping of vegetationon sampleplots.
Considering fieldworkalone,double-sampling withtheline-tran-
sectmethodprovided an increase in information of about~8~oas
compared withthe information whichcouldhavebeenobtainedby
clippingonly,duringthesameperiodof time.On thebasisof time
expendedin bothfieldandoffice,thedouble-sampling methodpro-
videdonly abouti~% moreinformation thancouldhavebeenob-
tainedby clippingalone.
The use of weightestimates in double-sampling providedabout
3 7~omoreinformation, thancouldbe obtained by straight clipping
inan equivalentamountoffieldtime.If fieldworkandoffice compila-
tionare bothconsidered the gainin information droppedto about
of intensive
samplingbothmethods provided
in fieldtimeandsomesavingin totaltimeex-
In otherstudies,
wouldbe consider-

ably affected by the relative amount of time consumed in field travel

as compared to the time requirements of clipping and double-samp-
ling. In large-scale extensive surveys, the clipping of all plots may
prove to be at least as efficient as any short-cut method.l