Professional Documents
Culture Documents
Skalski 1994
Skalski 1994
REFERENCES
Linked references are available on JSTOR for this article:
http://www.jstor.org/stable/3783246?seq=1&cid=pdf-reference#references_tab_contents
You may need to log in to JSTOR to access the linked references.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted
digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about
JSTOR, please contact support@jstor.org.
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at
http://about.jstor.org/terms
Wildlife Society, Wiley are collaborating with JSTOR to digitize, preserve and extend access to Wildlife
Society Bulletin (1973-2006)
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
Wildl. Soc. Bull. 22:192-203, 1994
JOHN R. SKALSKI, Center for Quantitative Science, School of Fisheries, University of Wash-
ington, Seattle, WA 98195
Key words: density estimation, line-transect, mark-recapture, random sampling, sampling with pro
ability proportional to size, stratified random sampling, survey sampling
The application of mark-recapture or line are particularly suited for surveying big game
transect methods has been traditionally re- (White et al. 1989, Bothma et al. 1990, Drum-
stricted to surveying animal abundance at a mer et al. 1990) and bird populations (Gates
single locale. Often this single locale is but a et al. 1968, Ratti et al. 1983, Brennan and Block
small component of a species range or area of 1986). Burnham et al. (1980) provide guide-
interest. However, management decisions of- lines for design and analysis of line transect
ten are of regional scope, making the result of surveys. Precision of line transect surveys de-
a site-specific survey of little practical conse- pends on the number of animals seen and the
quence. My purpose is to present a quantitative spatial variability in abundance among repli-
framework for making sound statistical infer- cate lines within a site. Ramsey and Scott (1979)
ences to the wildlife populations in the region. modified visual survey techniques using vari-
To do this, both the sampling variability of the able circular plots to estimate density of forest
survey techniques and the heterogeneity of avifauna. Alternatively, mark-recapture tech-
wildlife numbers across the landscape must be niques have been adapted to estimate animal
accounted for. My approach combines finite densities (MacLulich 1951, Smith et al. 1971,
sampling theory (see Hansen et al. [1953], Swift and Steinhorst 1976, Otis et al. 1978).
Cochran [1977], Jessen [1978], for a review) Among these mark-recapture methods, the
with density estimation techniques. These nested grid technique of Otis et al. (1978:67-
methods are not new, but rather a compilation 74) based on the methods of MacLulich (1951)
of sampling techniques suitable for wildlife is logistically the easiest to implement. More
surveys. The statistical inferences sought are recently, the trapping web design of Anderson
with regard to total abundance (NT) or mean et al. (1983) has been proposed for estimating
density (15) of animals within a specified geo- the density of small mammals. However, eval-
graphic area. uations of density estimators by Wilson and
In order to make inferences to specific areas, Anderson (1985a,b) suggest continued use of
survey techniques must have well-defined geo- the nested grid design over that of the trapping
graphic boundaries. However, abundance es- web.
timates using mark-recapture techniques (see General inability to conduct mark-recap-
Seber [1982] for review) typically survey wild- ture or line transect studies over the entire
life over an unspecified area around a trap grid region of statistical inference necessitates the
making area assessments impossible. For this area be subsampled. Consequently, a regional
reason, only density estimation techniques will survey transforms into a 2-stage sampling pro-
be considered. Abundance estimates (N) are gram. The first stage is the drawing of a prob-
computed by multiplying the density estimatesabilistic sample of the spatial area. Any sam-
(D) by plot size (a). pling scheme that yields a probabilistic sample
Two general approaches have been devised can be employed (see Cochran [1977] for re-
for estimating animal densities. Line transect view). In this paper, I consider only simple
methods (Eberhardt 1978, Burnham et al. 1980) random sampling (SRS), sampling with prob-
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
AREA SURVEYS OF WILDLIFE * Skalski 193
be illustrated for Great Basin pocket mouse NT- aiDi Ni, (1)
1=1 i=1
(Perognathus parvus) populations using the
nested grid technique of Otis et al. (1978). The where
objective of the survey was to estimate mean
Ni = animal abundance on the ith study plot
density (5) and total pocket mouse abundance
(i = 1, ... IK),
(NT) on a 32-ha region. This contrived example
D, = animal density on the ith study plot (i
represents the results of trapping studies on six
= 1, . . .,K),
1-ha trapping grids established in big sage-
brush (Artemisia tridentata )-bluebunch
wheatgrass (Agropyron spicatum) habitat in Table 1. Density estimates (D) and measurement er-
rors for 6 hypothetical small mammal surveys on 1-ha
southeastern Washington. The 6 sites represent plots randomly selected in a 32-ha region (see Fig. 1)
a simple random sample of the 32 one-ha sites using the nested grid technique of Otis et al. (1978:67-
within the region (Fig. 1). Small mammals at 74).
each site were captured using a 13- x 13-trap Site N = D 1 Var(N, I N.)
grid with Sherman traps (6.5 x 16.5 X 5 cm)
1 63.8 135.7
spaced 7.7 m apart. Trapping was conducted 2 47.1 83.9
for 10 consecutive days and multiple mark- 3 28.7 50.1
4 52.3 94.6
recapture methods were used to survey pop-
5 19.3 43.4
ulations. Density was estimated independently 6 38.4 67.6
at each site using the most appropriate mark-
recapture model. When 2 trap grids were ad-
N = 41.6 V
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
194 Wildl. Soc. Bull. 22(2) 1994
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
AREA SURVEYS OF WILDLIFE * Skalski 195
N
AA
= AT=Klk A (6) NT = 32(41.6) = 1,331.2
on the 32-ha site. The estimated variance (5)
with variance estimator
of total Great Basin pocket mouse abundance
Var(NT) is calculated to be
Var(D) = 2 (7)
Var(NT)
The estimates of mean density (D) and total
abundance (NT) can be assumed to be approx-
imately normally distributed for moderate
sampling effort within and between plots. Fol- = 2{_- 6(262.304) +7~7
lowing recommendations of Cochran (1977:
27), a (1 - a)100% confidence interval (CI)
[( 62)X 32
estimate of total abundance can be computed = 38,907.904.
using variance formula (5) as
Using these results, an approximate 95% CI
NT ? t1 a/2,k-1\/ai estimate of total Great Basin pocket mouse
abundance then would be
Similarly, a (1 - a)100% CI estimate of mean
density is computed using variance formula CI(1,331.2 - 2.571V3
(7) as
< NT < 1,331.2 + 2.571V3W8) =
D ? t1_a/2,k-1 )' CI(824.1 < NT < 1,838.3) = 0.95.
where tlIa/2k - is a critical value and follows Similarly, an estimate of mean density of Great
a Student's t-distribution satisfying the rela- Basin pocket mice on the 32-ha site is
tionship P( I t I < t a1/2,kI) 1 - a. For small 1,331.2
sample sizes, the estimators NT and D will tend D - - =41.6/ha
to be right-hand skewed. Both Ricker (1958) 32
and Cormack (1968) recommended using an with estimated variance
inverse transformation (e.g., 1/D) to eliminate
skewness when it occurs, in which case, CI _ : - 38,907.904 - 38.0.
Var(D I D) =-380
estimation should be based on the expression (32)2
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
196 Wildl. Soc. Bull. 22(2) 1994
It then follows that an approximate 95% CL (i.e., Pi = k/K), estimator (8) reverts to equa-
estimate of mean pocket mouse density is tion (4). When sample size is predetermined
(i.e., k fixed) the Horwitz-Thompson estimator
CI(41.6 - 2.571VX86
(8) has the variance (Yates and Grundy 1953)
< D < 41.6 + 2.571V'_i87d) = 0.95
Var(NT)
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
AREA SURVEYS OF WILDLIFE * Skalski 197
NT ? Z-ia/2\/r), (11) B
where D
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
198 Wildl. Soc. Bull. 22(2) 1994
Stratified Sampling
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
AREA SURVEYS OF WILDLIFE * Skalski 199
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
200 Wildl. Soc. Bull. 22(2) 1994
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
AREA SURVEYS OF WILDLIFE * Skalski 201
provides the basis for weighting the relative SPPS typically will result in a smaller variance.
contributions of the spatial variation (i.e., S2N) However, when plots are of equal size, SPPS
and sampling variability of wildlife census and SRS are equivalent. Stratified sampling
techniques (i.e., Var(D, I Di)). Through proper
designs also are suggested to help reduce the
variance of area survey results. Appendices A
design and analysis of regional surveys, reliable
and B show variance formulae for finite sam-
inferences to wildlife resources are possible with
ascribed error rates. Furthermore, the human pling methods must be modified to account for
resources necessary to conduct useful wildlife error in animal densities measurements. Unless
surveys for management purposes can be ob- adjustments in the variance estimates are in-
jectively assessed using sample size calcula- corporated, classical finite sampling proce-
tions. dures will underestimate the actual variance
The formulae presented here appear com- in area surveys of wildlife. The consequence
plicated at first inspection. However, computer would be a false sense of precision and the
software such as CAPTURE (White et al. 1982) failure to appreciate the uncertainty of re-
is well suited to density estimation using mark-source management decisions.
recapture data. Similarly, DISTANCE (Laake
et al. 1993) conveniently estimates density us-
ing line-transect data. The remainder of the LITERATURE CITED
formulae use sample means and sample vari-
ANDERSON, D. R., K. P. BURNHAM, G. C. WHITE, A
ances from among replicate sites to estimate D. L. OTIS. 1983. Density estimation of small
regional totals and associated variance esti- mammal populations using a trapping web and
mates. Virtually any scientific hand-held cal- distance sampling methods. Ecology 64:674-680.
BOTHMA, J. D., M. J. S. PEEL, S. PETTIT, AND D. GROSSMAN.
culator can be used to calculate these summary 1990. Evaluating the accuracy of some commonly
values and tally them to produce the desired used game-counting methods. South Afr. J. Wildl.
Res. 20:26-32.
regional statistics.
BRENNAN, L. A., AND W. M. BLOCK. 1986. Line tran-
sect estimates of mountain quail density. J. Wildl.
SUMMARY Manage. 50:373-377.
BURNHAM, K. P., D. R. ANDERSON, AND J. L. LAAKE.
Area surveys of animal resources are typi- 1980. Estimation of density from line transect
sampling of biological populations. Wildl. Mono
cally based on a 2-stage nested sampling de- 72.202pp.
sign. The first stage is the selection of a subset COCHRAN, W. G. 1977. Sampling techniques. John
of possible study plots within the region of Wiley and Sons, New York, N.Y. 428pp.
CORMACK, R. M. 1968. The statistics of capture-re-
inference. The second stage then consists of capture methods. Oceanographic Mar. Biol. Annu.
surveying animal densities at the plots selected Rev. 6:455-506.
in the first stage. Consequently, both sampling DRUMMER, T. D., A. R. DEGANGE, L. L. PANK, AND L.
L. McDONALD. 1990. Adjusting for group size
error associated with subsampling the region influence in line transect sampling. J. Wildl. Man-
and measurement error associated with the age. 54:511-514.
survey method contribute to the variance of EBERHARDT, L. L. 1978. Transect methods for pop-
ulation studies. J. Wildl. Manage. 42:1-31.
estimates of total abundance and mean den- FELLER, W. 1968. An introduction to probability the-
sity. Sample size calculations need to take into ory and its applications. Vol. I. John Wiley and
account both of these error sources in the de- Sons, New York, N.Y. 509pp.
GATES, C. E., W. H. MARSHALL, AND D. P. OLSEN. 1968.
sign of area surveys. Line transect method of estimating grouse popu-
Use of sampling with probability propor- lation densities. Biometrics 24:135-145.
tional to size (SPPS) is recommended over sim- HANSEN, M. H., W. N. HURWITZ, AND W. G. MADOW.
1953. Sample survey methods and theory. Vol. I.
ple random sampling (SRS) when study plots John Wiley and Sons, New York, N.Y. 638pp.
are of unequal size. When plots vary in size, HORWITZ, D. G., AND D. J. THOMPSON. 1952. A gen-
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
202 Wildl. Soc. Bull. 22(2) 1994
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms
AREA SURVEYS OF WILDLIFE * Skalski 203
+ 2 K K (P i
f-l j=l j ij
k
Var(i) = - ______ K
_ ,>i
K (1 - P,)Var(i, I x1)
where
k
i=1 Pi
_ Var(2I x,) resulting in a negative bias of
Var(i ) = i=X k K
i(j )2
Var(X) = + 2 2; "pip-PjP)ii
i=1 i J=1 j j t;
j>1
Sg2 = -
x~i (k-1 k Var(ix I xi)
i=1 P
APPENDIX B because
i=l i
APPENDIX C
with unbiased variance estimator
Preliminary survey data can be used to estimate the
k I- P.) variance components used in equation (15) for sampl
Var(Xi) p xi size calculations. From a preliminary sample of n plots
i=l
the average sampling error can be estimated as
k kc (P i - PjP,)x'xj
+ 2; 2; (8P P )
i=l j=l i ij
j>i _ Var(N, I Nj)
for k random. When the observations are measured Var(N, I N,) = i='
n
with error [E(i I xi], the estimator and the spatial variance can be estimated from the
k Xi
relationship
i=1 i
il Pi
However, the expected value of (18) when the xi are
measured with error is
K (1 _Fp)
E[Var(X)]= p2 xi2
i=
This content downloaded from 128.111.121.42 on Mon, 19 Sep 2016 14:08:42 UTC
All use subject to http://about.jstor.org/terms