Forest Ecology and Management 150 (2001) 313±322

Inventory time-cost and statistical power:

a case study of a Lao rattan
Tom D. Evansa,*, Oulathong V. Viengkhamb
a
Department of Plant Sciences, Oxford Forestry Institute, University of Oxford, South Parks Road, Oxford OX1 3RB, UK
b
Forestry Research Centre, National Agriculture and Forestry Research Institute, Box 7174, Vientiane, Laos PDR
Received 22 February 2000; received in revised form 25 July 2000; accepted 26 July 2000

Abstract

A pilot study was conducted to determine the optimal survey design for an inventory of the commercially important rattan
Calamus poilanei in central Lao PDR. Simulations were used to estimate the coef®cients of variation (CVs) of population
density estimates for alternative sampling designs, a model was developed to determine the optimal survey design which
minimised survey time for a given survey precision and a statistical power analysis was used to determine whether these
surveys were suf®ciently sensitive to form part of a monitoring programme.
Line survey and plot enumeration times were quite high due to moderately dif®cult terrain. The CV (135% for a 5 m  50 m
plot) was markedly higher than in some comparable studies. The CV could be predicted well from plot size but plot shape
explained little additional variance. The optimum plot size varied but the 5 m  50 m plot was close to the optimal solution
over a wide range of situations. Surveys with a precision of 20% or better were predicted to be very time consuming. Power
analysis indicated that a survey precision of 5 or 10% was likely to be required if an annual monitoring programme was to be
useful in detecting a decline in the population. To achieve this precision, surveys would have to be even more time consuming,
and many `optimal' solutions were clearly impractical. For a resource of moderate value such as rattan cane, very prolonged or
costly surveys such as these are most unlikely to be economically justi®ed. # 2001 Elsevier Science B.V. All rights reserved.

Keywords: Survey techniques; Population density; Non-timber forest products; Power analysis; Lao PDR; Calamus poilanei

1. Introduction and Gullison, 1995; Peters, 1996). An essential part of

model design and subsequent monitoring is to survey
There is increasing pressure to develop rational, the population density.
sustainable management systems for timber trees and Vegetation density surveys using ®xed-size plots
non-timber forest products in tropical moist forests. are often preferred because they are unbiased for
One recommended approach is to predict the optimal non-randomly distributed populations (Greig-Smith,
harvest regime from mathematical models of growth 1983). When time or money is limited, the precision of
and yield (Hall and Bawa, 1993; Alder, 1995; Boot such surveys depends on a trade-off between using
larger plots (which have a lower per-plot variance) and
* using more plots (which tends to reduce the standard
Corresponding author. Present address: c/o 11a, Yeoman Lane,
Bearsted, Maidstone, Kent ME14 4BX, UK.
error of the mean).
Tel.: ‡44-1865-275000; fax: ‡44-1865-275074. Such surveys can be time-consuming and it is
E-mail address: [email protected] (T.D. Evans). important that they are designed to yield statistically

0378-1127/01/$ ± see front matter # 2001 Elsevier Science B.V. All rights reserved.
PII: S 0 3 7 8 - 1 1 2 7 ( 0 0 ) 0 0 5 8 9 - 2
314 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322

representative data and make best use of limited and semi-evergreen forests in Lao PDR (personal
resources. This is usually done by carrying out pilot observations) and also occurs in Vietnam (Gagnepain
surveys and then performing an optimisation analysis and Conrard, 1937) and Thailand (specimens at
to select the best survey design (Prodan, 1968). The Royal Botanic Gardens, Kew). Voucher specimens
optimum varies according to the terrain, how patchily Oulathong 211 and Tom Evans 29 from the study site
a species is distributed and other factors. Examples of are held at both Kew and the Forest Research Centre,
such studies are common in the literature (Freese, Vientiane, Lao PDR.
1961; Arvanitis and O'Regan, 1967; Micosa-Tandug, A survey line 600 m long was marked in the forest,
1978; Taafe, 1979; Zeide, 1980; da Silva and de running perpendicularly away from a major stream
Vasconcelas, 1996; Stockdale and Wright, 1996). and along a slight ridge. The line was divided into
However, it is less often stated how much time the 100 m sections. At a random point along each 100 m
optimally designed survey would require or how section, a single inventory plot measuring 20 m
sensitive it would prove as part of a post-harvest 100 m was marked perpendicular to the survey line
monitoring programme. These are especially impor- and running away to either the left or right. Each plot
tant considerations for non-timber forest products was subdivided into 40 subplots of 5 m  10 m in
(such as rattan canes) since they usually have a much which all stems of Calamus poilanei longer than
lower value per hectare than timber. This means that 1 m (from the ground to the base of the petiole of
an expensive survey programme could jeopardise the the last fully expanded leaf) were counted. The time
overall pro®tability of the harvest. taken to mark out and then enumerate each plot was
This paper describes a pilot study of rattan popula- recorded. The survey team consisted of ®ve people,
tion density in a forest in central Laos. Optimal survey two to mark out the plots and three to enumerate
design and required survey times were predicted and a rattans. The total area enumerated was 1.2 ha which
power analysis was used to predict the sensitivity of was 10% of the total area sampled (12 ha).
possible monitoring programmes. The pilot survey
was conducted as part of the design process for a 2.2. Simulation of other sampling designs
larger survey in the same area. The detailed results will
prove valuable to others planning such surveys in the Alternative sampling designs were simulated using
region. More importantly, the general conclusions are the data from smaller portions of each 20 m  100 m
relevant to anyone planning inventory and monitoring plot. The sizes and shapes of the simulated plots are
of tropical non-timber forest products, since they shown in Table 1. The measure of plot shape used was
suggest that traditional density surveys can be an rectangularity, de®ned as the ratio of plot length to
inappropriate tool in some situations. width. Subplots with their long axis parallel to the
main transect line had a rectangularity <1. This is
appropriate because such plots ran across, not along
2. Methods the dominant habitat gradient and thus were expected

2.1. Pilot survey Table 1

Pilot survey for enhanced survey design of a solitary-stemmed
The pilot survey was carried out during 4±11 Jan- rattan in central Laosa
uary 1999 near That Thon waterfall, Ban Naphong, Length (m) Width (m)
Pakkading District, Bolikhamxay Province, Laos PDR
(188230 2500 N 1048230 0000 E). The habitat is tall, pri- 5 10 20
mary, dry evergreen forest at 650±750 m altitude lying 10 0.005 (2) 0.01 (1) 0.02 (0.5)
on a mixture of steep and gentle slopes. The study 20 0.01 (4) 0.02 (2) 0.04 (1)
species was the commercially important rattan Cala- 50 0.025 (10) 0.05 (5) 0.1 (2.5)
100 0.05 (20) 0.1 (10) 0.2 (5)
mus poilanei Conrard (locally called wai toon). It is a
solitary, high-climbing plant with a naked stem dia- a
Plot sizes (ha) included in the study, with plot rectangularity
meter of 3±4 cm. It is typical of mature evergreen in parentheses.
 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322 315

to capture a different amount of variation and so give a monitor a possibly declining population, Gerrodette
different degree of sampling error. (1987) gives an approximate relationship (his Eq. (20))
Each large 20 m  100 m plot was divided into between the rate of population decline, survey preci-
an integer number of smaller subplots (e.g. two sion, the frequency of surveys and the power. This
20 m  50 m plots, four 10 m  50 m plots or 20 equation was used to estimate how sensitive the
10 m  10 m plots), all of the same shape and area. optimal survey designs selected above would be in
Then one subplot was picked at random from each of various situations.
the six large plots. The set of rattan counts from the six
subplots was used to estimate the mean rattan density
and its standard deviation. This simulation process 3. Results
was repeated eleven times for each subplot size except
the two largest where only one simulation was done. 3.1. Pilot survey
The eleven estimates of density and standard devia-
tions were averaged and then used to estimate the The population density of Calamus poilanei stems
coef®cient of variation (CV) for individual plots. over 1 m tall was estimated as 20.0 stems/ha with 95%
Regression was used to determine an empirical con®dence limits of 8.70±31.3 stems/ha. Plot enu-
relationship between plot size, plot shape and CV. meration time averaged 600 min/ha, and the time to
Using only a single CV estimate for each plot size and measure and travel lines between plots averaged
shape avoided pseudoreplication. 18.5 min/100 m.

2.3. Optimisation of sampling design 3.2. Simulation of other sampling designs

The optimal design is here considered to be the one Various linear combinations of explanatory and
which attains a given precision in the minimum total response variables were tried to determine which
survey time. The optimum design depends on a trade equation for predicting CV under other sampling
off between number of plots and plot CV. The opti- designs gave the best ®t to the data (Table 2). The
mum also varies depending on the total size of the quadratic combination preferred by Arvanitis and
forest, the survey precision required and the time O'Regan (1967) was not tried because it gave such
needed to survey lines and enumerate plots at a a modest improvement in ®t for their data. The pre-
particular site (Prodan, 1968). This is due to the ferred equation selected was
varying balance between total plot enumeration time
log…CV† ˆ 0:292 0:263 log…area†;
and travel time between plots as parts of the total
2
time budget. The optimum also depends on whether …R …adj† ˆ 91:5%† (1)
the shortest direct route is followed from plot-to-plot
or whether a set of parallel transect lines is laid out Table 2
and plots accessed along these. All these relationships Pilot survey for enhanced survey design of a solitary-stemmed
can be modelled mathematically as explained in rattan in central Laosa
Appendix A.
Explanatory variablesb Dependent R2(adj)
variable (%)
2.4. Power analysis
Area CV 63.5
Area, rectangularity CV 59.9
The power of a survey depends on the likelihood of log(area) CV 90.5
Type II errors or `false negatives' during statistical log(area), log(rectangularity) CV 91.0
testing. A higher likelihood of false negatives implies log(area) log(CV) 91.5
a lower power. This likelihood varies with the strength log(area), log(rectangularity) log(CV) 92.3
of the effect (thus a gentle decline in population a
Alternative explanatory variables for the coef®cient of
density is more likely to be overlooked than a steep variation in density estimates.
one). In the case of transect surveys being used to b
Area: subplot area.
316 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322

This was the second best equation by a small margin approach recommended by Cohen (1988) and Rotten-
but was selected for ease of comparison with the berry and Wiens (1985). There is no standard accepted
results of Arvanitis and O'Regan (1967); Taafe level of power, perhaps because the measure is so
(1979) and Stockdale and Wright (1996). scarcely used and because the choice depends on the
relative costs of Type I and Type II errors (Peterman,
3.3. Optimisation of sampling design 1990). Three illustrative levels were chosen here, 0.95,
0.90 and 0.80 because these are the levels used in
The optimum plot size which minimised survey examples given by Steel and Torrie (1960); Gerrodette
time in each case was determined graphically. The (1987); Judd and McClelland (1989); Maxwell and
optima were not sharply de®ned in some cases and so a Delaney (1989) and Snedecor and Cochran (1989).
range of plot sizes gave very similar survey times. Calculations were performed for various possible
Examples of the optimum plot sizes, numbers of plots rates of decline and for series of up to ®ve surveys
and survey times required are given in Table 3 for four (Table 4). These results are not speci®c to the results of
sizes of forest, four levels of survey precision and two the present pilot survey and can be applied to any
alternative survey layouts. Survey times are those similar series of surveys if the precision of each survey
required by a team of ®ve people working 7 h/day. is known. The survey interval is not speci®ed but note
Precision is half of the 95% con®dence interval that the rate of decline is quanti®ed per survey interval.
expressed as a percentage of the survey mean. It is perhaps easiest to understand the data if one
considers ®rst an interval of 1 year. Thus, Table 4
3.4. Power analysis shows that, for example, if we require a power of 95%
then given a programme of two surveys, 1 year apart,
Results are presented as the minimum rate of each with a precision of 10%, the minimum rate of
decline which could be detected by a given survey decline which could be detected reliably is estimated
design at a given level of reliability (power), an as 23% per annum. In another example, at the end of a

Table 3
Pilot survey for enhanced survey design of a solitary-stemmed rattan in central Laosa

Precision (%) Forest area (ha) Transect lines Direct routes

Optimal Plot no. Time Optimal Plot no. Time

plot size (team-days) plot size (team-days)

20 200 0.01±0.05 282±121 13±15 0.01±0.04 282±136 15

1000 0.01±0.05 282±121 30 0.025±0.1 174±84 24
10000 Any 406±58 229 0.2 58 50
20000 Any 406±58 450 0.2 58 64
15 200 0.025 310 19 0.01 502 21
1000 0.025±0.05 310±215 37 0.025±0.05 310±215 36
10000 Any 722±104 235 0.1±0.2 149±104 75
20000 Any 722±104 456 0.2 104 93
10 200 0.025 697 37 0.005±0.01 1625±1129 37
1000 0.025±0.05 697±484 55 0.02±0.025 784±697 61
10000 0.025±0.05 697±484 253±255 0.1 336 129
20000 0.025±0.05 697±484 473 0.1±0.2 336±233 162
5 200 0.025 2788 134 0.005 6500 96
1000 0.025 2788 152 0.005±0.01 6500±4514 158
10000 0.025 2788 351 0.04 2177 330
20000 0.025 2788 571 0.05 1936 412
a
Optimal survey designs for a range of situations.
 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322 317

Table 4 (indicating very good ®t). Thus, the rule-of-thumb

Pilot survey for enhanced survey design of a solitary-stemmed
would have given poor results in the present case
rattan in central Laosa
which shows that conducting a new pilot survey
Power Precision of Number of surveys in the was necessary. Boon (1962, cited in Arvanitis and
(%) single survey (%) series O'Regan, 1967) also considered that the rule-
2 3 4 5 of-thumb was not a good alternative to a pilot survey.
95 20 45 25 16 11
15 34 18 12 9 4.1.2. The effect of forest size on optimum plot size
10 23 12 8 6 Table 3 shows that for smaller forest areas, smaller
5 11 6 4 3 plots are predicted to be most ef®cient because plot
90 20 41 22 14 10 enumeration is more time consuming than travel
15 31 17 11 8 between plots. For larger forest areas, the travel time
10 20 11 7 5 between plots becomes more important and fewer,
5 10 6 4 3 bigger plots make best use of time. Interestingly, for
80 20 35 19 12 9 very large forest areas (20,000 ha‡, 200 km2‡) and
15 26 14 9 7 precision of 15 or 20%, the plot size is predicted to be
10 17 10 6 4
irrelevant because it has negligible effect on the total
5 9 5 3 2
survey time.
a
Minimum rate of decline per survey interval (%) which can be
detected at the end of a series of surveys of different individual 4.1.3. The choice between parallel survey lines
precision and different total number; the probability of Type I error and direct routes
is set at a ˆ 5%.
Table 3 also shows that using parallel survey lines
was predicted to be much slower than following direct
routes between plots for areas of 1000 ha or greater
programme of ®ve surveys, also spaced annually, each (for example, 24 versus 32 days for 1000 ha, 50 versus
with a precision of 5%, one would reliably be able to 229 days for 10,000 ha) so it would be better to use
detect sustained declines lower than 1% per annum. direct routes. This result is relevant to survey design
but is not stressed by standard sources (Loetsch
and Haller, 1964; Husch et al., 1972; Philip, 1994).
4. Discussion However, since survey lines are the preferred forest
inventory method in Laos and other South-East Asian
This section covers ®rst the particular implications countries, the following discussion assumes this
of the results for survey design and then discusses the method will be used.
wider question of whether such surveys are actually
appropriate in all situations. 4.1.4. The optimum plot size
Table 3 also shows that quite small plots appeared to
4.1. Implications for survey design be preferable in most cases. It seems that a 0.025 ha
would probably be suitable over a wide range of forest
4.1.1. The need for a pilot survey sizes. Zeide (1980) recommends a more elaborate way
In this study, an empirically determined regression of calculating the time to enumerate plots, allowing for
was used to relate CV to plot size and shape. Freese the fact that smaller plots have proportionately more
(1961) and Zeide (1980) recommended the use of a perimeter. This alternative method is detailed in
rule-of-thumb formula which obviated the need to do Appendix A. If this approach had been used to calcu-
this. However, that formula gives a very poor ®t to late the ®gures in Table 3, conclusions would have
the current data set. Paired t tests to indicate goodness been unchanged for forests larger than 10,000 ha but
of ®t for the two alternative ways of predicting CV different for smaller areas as shown in Table 5. The
give P < 0:001 for Freese's method (indicating very optimum plot size becomes larger and survey times
poor ®t) and P ˆ 0:978 for the empirical equation shorter. In the current survey, no ®eld data were
318 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322

Table 5
Pilot survey for enhanced survey design of a solitary-stemmed rattan in central Laosa

Forest area (ha) Plot size, time (non-Zeide) Plot size, time (Zeide)

200 0.01±0.025 ha, 20 days 0.05 ha, 15 days

1000 0.01±0.04 ha, 38 days 0.04±0.1, 32±33 days
a
Comparison of optimal plot sizes and survey times with and without the modi®cations of Zeide (1980) for a transect-line design.

collected to determine whether Zeide's approach is not practical with wider plots. Long narrow plots are
more accurate. also preferable because they are less disorientating in
Micosa-Tandug (1978) identi®ed 0.01 ha as the dense forest (personal observation).
optimal plot size in her study of rattans in the Phi-
lippines, somewhat smaller than the optimum in Laos. 4.1.6. The time estimates are probably minima
Stockdale and Wright (1996) concluded that for a If multiple resources were surveyed or if more detail
200 ha forest block optimum plot sizes were was required (e.g. the stem lengths or sexes of rattans)
0.0025±0.025 ha for estimating rattan densities; the the survey time would increase further. It is also
Lao optimum falls in this range. important to note that the pilot survey covered an
The optimum plot sizes given in Table 3 are appro- area of only 12 ha. Rattans are often said to be patchily
priate for the particular resource studied. Other distributed on scales larger than this (J. Drans®eld,
resources are likely to show different degrees of personal communication, 1999), so the average CV for
clumping and occur in terrain of varying dif®culty, surveys over wider forest areas would be higher and
so it would be best to repeat the calculations using data the expected time to achieve a given accuracy perhaps
from a short pilot study such as this before conducting even higher than estimated here. However, for such
surveys over a very large area. large areas it might be more ef®cient to stratify the
area on the basis of topography and local knowledge
4.1.5. The effect of plot shape and sample each block separately.
Stockdale and Wright (1996) found that increased
rectangularity was an important predictor of decreased 4.2. Are traditional, plot-based inventories always
CV. Their results indicated that arranging plots so that appropriate?
they run downslope rather than across slope improved
precision at their study site because it increased 4.2.1. Inventories can be very time-consuming
within-plot variation and reduced between-plot varia- An important result of this study is that the time
tion. In contrast, this effect was not shown in the required to estimate the population density of this Lao
current study. Table 2 shows that adding log(rectan- rattan to a reasonable degree of accuracy is impracti-
gularity) as a variable explains only 0.8% more of the cally high (Table 3). This was not expected from a
variation in log(CV). This may be because the area had reading of the literature, although the problem is
irregular topography so some plots included very little brie¯y mentioned by Loetsch and Haller (1964).
habitat variation and others would have included much For example, in the study area a density estimate with
variation whatever their shape and size. the relatively low precision of 20% is predicted to
Given that 0.025 ha plots are required we recom- require 30 days of work in the forest for a team of 4±5
mend the dimensions 5  50 m. Such a plot has the people to survey 1000 ha (10 km2) and 229 days for an
advantage that it is narrow enough that only one side area of 10,000 ha (100 km2). Many individual Lao
of the plot needs to be precisely measured. People villages use areas of forest greater than 100 km2
counting plants along this marked line can simply (personal observation and IUCN-NTFP Project, Lao
estimate for most plants whether they fall in the plot, of®ce, unpublished data) but are most unlikely to be
and in cases of doubt can measure 5 m to one side or able to invest this much time in a single task involved
2.5 m to either side, as appropriate, to check. This is in improving the management of a single resource.
 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322 319

Table 6
Pilot survey for enhanced survey design of a solitary-stemmed rattan in central Laosa

Study Inventory subject Unit recorded Country CV (%)

Taafe (1979) Trees Not specified Paraguay 13.5

Arvanitis and Regan (1967) Conifers Individuals USA 28
Micosa-Tandug (1978) Rattan Individuals Philippines 50
Stockdale and Wright (1996) Rattan Clumps Brunei 65b
Freese (1961) Douglas fir Volume USA 88
Present study Rattan Individuals Laos 135
a
CV for 0.025 ha (5 m  50 m) plots recorded by some other published studies.
b
Estimated graphically.

Use of direct routes rather than transect lines would One potential solution to the high CV of a conven-
make surveys more technically dif®cult but would tional survey might be to use an adaptive survey
reduce the times stated above to 24 and 50 days, method, for example Adaptive Cluster Sampling
respectively. Even this degree of effort is unlikely (Thomson, 1990, Roesch, 1993). Such methods have
to be practical in many situations, especially if been used rarely in forestry applications, and not at all
required repeatedly for monitoring purposes. Econo- in non-timber forest product studies (J. Wong, perso-
mising on survey effort could reduce the precision of nal communication) but they deserve consideration
the results to such a degree that the data cease to be and ®eld testing.
useful.
The CVs recorded in this study are markedly higher 4.2.2. Monitoring surveys using these designs
than those recorded in several other published studies would have limited power
(Table 6). This suggests that the study species is a The power analysis in Table 4 indicates what kind of
particularly dif®cult one to inventory and that time surveys would be able to detect steep or gentle popu-
limitations would thus be less severe in some circum- lation declines. If we take the view that the risk of false
stances. For example, Stockdale and Wright (1996) positives and risk of false negatives should be equally
found that CVs were about 50±70% for most plot sizes likely, we require a power of 0.95. It appears that a
and they projected that rattan clumps in a 200 ha forest series of two surveys 1 year apart, each with precision
could be surveyed with 20% precision by a team of of 20%, could only reliably be expected to detect an
unspeci®ed size in 15±20 h using direct routes annual decline of 45% or more. Such a decline would
between plots. The same survey in Laos is predicted leave the population extinct in less than 3 years. Sensi-
to take a team of ®ve about seven times longer. This tivity this poor would have very limited value if it was
difference may also be partly due to higher travel and the main means of monitoring population levels.
enumeration times in Laos, but Stockdale and Wright Even so, in the Lao example, an enormous effort is
(1996) do not specify these. likely to be required to demonstrate this decline with
The high CV is a result of patchy distribution at the statistical signi®cance (30 days of work in the forest,
scale of the range of plot sizes used. Low density is not twice, by a team of 4±5, even for a forest of only 10 km2).
a cause, since for a given pattern of patchiness CV is Surveys with an individual precision of 5% could be
independent of density. Note that this study treated expected to detect a decline of 11% in the space of one
plots spread across 12 ha whereas Stockdale and survey interval or if carried out again in the middle of
Wright (1996) placed all their plots within 1.5 ha the interval, a decline of only 6% per interval. This is
and Micosa-Tandug (1978) within 1 ha. Thus, the closer to the degree of sensitivity likely to be required
present study probably detected additional sampling for monitoring purposes. Table 3 shows the great
variation at a level missed by those two studies amount of effort required to carry out such precise
but more relevant to the problems of a full-scale surveys in the Laos example. The same 10 km2 forest
survey. block would require over 150 days of work by the
320 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322

team on each of the three surveys. This is clearly Acknowledgements

impractical.
A third approach would be to wait longer between Khamphone Sengdala, Banxa Thammavong and the
surveys Ð thus two surveys of 20% precision people of Ban Naphong village are thanked for their
conducted 5 years apart would be able to detect a assistance during the pilot survey. Nick Brown and
decline of 45% over that period or about 9% of the Joost Foppes gave helpful comments on the manu-
starting population per year. In one sense this would be script. Jenny Wong and Nell Baker kindly supplied a
a cheaper solution to increasing sensitivity. However, draft of their review of biometric methods for NTFPs.
evidence of this decline would still not be available The material of Calamus poilanei was identi®ed by
until the population had almost halved, a very late comparison with type material lent by the MuseÂe
stage at which to begin remedial action. Nationale d'Histoire Naturelle, Paris. This study was
Power analysis is all too rarely attempted in biolo- a part of the Darwin Initiative Project on the Diversity
gical surveys (e.g. Peterman, 1990 for ®sheries and Sustainable Management of Rattans in Lao PDR,
research and J. Wong, personal communication for funded by the UK government's Department of the
non-timber forest products) but where it has been used Environment. Fieldwork was done with the permis-
the results have shown a sobering lack of power sion and collaboration of the Forestry Research Centre
(Peterman, 1990; Taylor and Gerrodette, 1993; of the National Agriculture and Forestry Research
Strayer, 1999). Power analysis should be used as a Institute of the Lao Ministry of Agriculture and
matter of course in non-timber forest product work to Forestry.
avoid survey designs which have little chance of
detecting the phenomenon of interest.
Appendix A. Derivation of expression linking
plot size and survey time
4.2.3. Overall conclusions
The purpose of inventory and monitoring surveys is
Individual plot CV, overall survey precision and
to support improved management. This is only worth-
number of plots are linked by the equation
while if there are direct bene®ts due to improved
economic productivity or indirect bene®ts such as t CV
Eˆ (2)
preventing inadvertant extinction of the harvested n1=2
species. where E is the precision and t is the appropriate
This pilot survey suggests that for rattans in Laos, Student's t statistic.
the time costs of these surveys will outweigh any It follows that the number of plots required to
economic bene®ts they may offer. This cannot be produce a given precision is
explicitly calculated since at present there are too
 
few data to estimate the productivity of wild rattans t CV 2
in Laos, but the results of Anders Bogh (unpublished nˆ (3)
E
data) from Thailand and studies currently underway in
Laos (authors' unpublished observations) suggest that Substituting the Eq. (1) into Eq. (3) gives Eq. (4), an
this will be modest. expression for n in terms of plot size.
Therefore, it is questionable whether standard, plot-  
t‰expf 0:292 0:263 log…area†gŠ 2
based, statistically rigorous resource inventories will nˆ (4)
E
be an affordable NTFP management tool in areas such
as Laos with large forests, dif®cult terrain and patchy To calculate the time required to survey n plots we
distributions of the target species. Management agen- need to estimate the time taken to mark and enumerate
cies should only attempt them if very detailed numer- each plot and to travel between plots, i.e.
ical data are required for a well-de®ned purpose, no
Ttotal ˆ Ttravel ‡ Tenumeration
simpler alternatives exist and pilot studies indicate that
suf®ciently precise data can be gathered at reasonable The travel times between plots depend on the number
cost. of plots, the total area of forest and the route taken
 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322 321

between plots. Stockdale (1994) made her calculations References

using the shortest direct route between successive
plots, a method which involves calculating the exact Alder, D., 1995. Growth Modelling for Tropical Mixed Forests.
bearing and distance from each plot to the next and Tropical Forestry Paper 30, Oxford Forestry Institute, Oxford,
UK.
following a zig-zag path. The expression for the total Arvanitis, L.G., O'Regan, W.G., 1967. Computer simulation and
time taken to follow this route is economic ef®ciency in forest sampling. Hilgardia 38 (2),
1=2 133±164.
Ttravel ˆ T1 …10;000 An† Boon, D.A., 1962. Plot Size and Variability. I.C.T. Information
Series B (17). Delft, The Netherlands.
where T1 is the time taken to measure and travel 1 m Boot, R.G.A., Gullison, R.E., 1995. Approaches to developing
and A is the total forest area in ha. sustainable extraction systems for tropical forest products.
Ecol. Appl. 5 (4), 896±903.
We have carried out the calculations using this
Cohen, J., 1988. Statistical Power Analysis for the Behavioural
method. However, it is an unrealistic approach in Sciences, 2nd Edition. Academic Press, New York, USA.
Laos. It is much more usual that foresters will take da Silva, J.A.A., de Vasconcelas, A.J.N., 1996. Application of the
the slower but more straightforward approach of lay- relative ef®ciency methodology to select plot area and shape in
ing out parallel survey lines and placing plots at offsets forest inventories of the caatinga of Pernambuco-Brazil.
Commonwealth For. Rev. 75 (3), 243±246.
to the left or right of these lines. Assuming that these
Freese, F., 1961. Relation of plot size to variability: an
lines are placed 200 m apart, the expression for the approximation. J. For. 59, 679.
time taken to locate all the plots will be Gagnepain, F., Conrard, J., 1937. Palmiers. In: Lecomte, H. (Ed.),
    Flore GeÂneÂrale de l'Indochine, Tome 6, Parties 7 and 8. Massie
10;000 A L and Co., Paris, pp. 946±1056.
Ttravel ˆ T1 ‡ n 50 Gerrodette, T., 1987. A power analysis for detecting trends.
200 2
Ecology 68 (5), 1364±1372.
Greig-Smith, P., 1983. Quantitative Plant Ecology, 3rd Edition.
where L is the length of a plot.
Blackwell, Oxford, UK.
The enumeration time depends only on the plot area Hall, P., Bawa, K., 1993. Methods to assess the impact of extraction
and was estimated from the survey times recorded of non-timber tropical forest products on plant populations.
during this survey. Survey speeds increased as the Econ. Bot. 47 (3), 234±247.
team became more experienced, so the time for the last Husch, B., Miller, C.I., Beer, T.W., 1972. Forest Mensuration, 2nd
Edition. Ronald, New York, USA.
few plots was probably most representative of future
Judd, C.M., McClelland, G.H., 1989. Data Analysis. Harcourt
speeds. This was an average of about 600 min/ha. The Brace Jovanovitch, San Diego, USA.
expression is Loetsch, F., Haller, K.E., 1964. Forest Inventory, 2nd Edition. BLV
Verlagsgesellschaft, Munich, Germany, p. 131.
Tenumeration ˆ naT2 Maxwell, S.E., Delaney, H.D., 1989. Designing Experiments and
Analysing data. Wadsworth Publishing Co., Belmont, CA, USA.
where T2 is the time to enumerate 1 ha and a is the area Micosa-Tandug, L., 1978. Sampling method for inventory of
of one plot. Philippine rattan and its distribution. Sylvatrop Philipp. For.
Res. J. 3 (3), 155±170.
As an alternative to treating enumeration time as Peterman, R.M., 1990. Statisical power analysis can improve
proportional to plot area, Zeide (1980) recommends ®sheries research and management. Can. J. Fish. Aquat. Sci. 47,
allowing for edge effects on survey time, using the 2±15.
equations T enumeration ˆ nTi and Ti ˆ bP0:75
i . For the Peters, C.M., 1996. The Ecology and Management of Non-Timber
current data set, setting Ti ˆ 3 min and Pi ˆ 0:005 ha, Forest Resources. World Bank Technical Paper 322. World
Bank, Washington, DC.
b would be estimated as 159.5. Data are presented in Philip, M.S., 1994. Measuring Trees and Forests, 2nd Edition. CAB
the results including and excluding this modi®cation. International, Wallingford, UK.
Finally, by combining the above expressions for Prodan, M., 1968. Forest Biometrics. Pergamon Press, Oxford, UK.
Tenumeration and Ttravel, it is possible to graph the Roesch Jr., F.A., 1993. Adaptive cluster sampling for forest
variation of Ttotal with plot size and total forest area. inventories. For. Sci. 39 (4), 655±669.
Rottenberry, J.T., Wiens, J.A., 1985. Statistical power analysis and
Note that this calculation assumes that the whole community-wide patterns. Am. Nat. 125, 164±168.
forest area has rattan distribution uniform with the Snedecor, G.W., Cochran, W.G., 1989. Statistical Methods. Iowa
12 ha area of the pilot survey. State University Press, Ames, USA.
322 T.D. Evans, O.V. Viengkham / Forest Ecology and Management 150 (2001) 313±322

Steel, R.G.D., Torrie, J.H., 1960. Principles and Procedure of Taafe, K.E., 1979. Computing Optimum Plot Size for Wildland
Statistics. McGraw-Hill, New York, USA. Inventories, Resource Inventory Notes BLM23, pp. 8±15.
Stockdale, M.C., Wright, H.L., 1996. Rattan inventory: determin- Taylor, B.L., Gerrodette, T., 1993. The uses of statistical power in
ing plot shape and size. In: Edwards, D.S., Booth, W.E., Choy, conservation biology: the vaquita and the northern spotted owl.
S.C. (Eds.), Tropical Rainforest Research Ð Current Issues. Conserv. Biol. 7 (3), 489±500.
Kluwer Academic Publishers, The Netherlands, pp 523±533. Thomson, S.K., 1990. Adaptive cluster sampling. J. Am. Stat.
Strayer, D.L., 1999. Statistical power of presence/absence data to Assoc. 85 (412), 1050±1059.
detect population declines. Conserv. Biol. 13 (5), 1034±1038. Zeide, B., 1980. Plot size optimisation. For. Sci. 26 (2), 251±257.