Directory UMM :Data Elmu:jurnal:A:Agricultural Systems:Vol66.Issue2.Nov2000:

Agricultural Systems 66 (2000) 99±113
www.elsevier.com/locate/agsy

Quantifying trade-o€s between pest
sampling time and precision in commercial
IPM sampling programs
E.M. Cullen a,*, F.G. Zalom a, M.L. Flint a, E.E. Zilbert b
a

Department of Entomology, University of California, One Shields Avenue, Davis, CA 95616, USA
b
Department of Agronomy and Range Science, University of California, Davis, CA 95616, USA
Received 13 January 2000; received in revised form 30 July 2000; accepted 7 August 2000

Abstract
A key component of Integrated Pest Management (IPM) in agricultural systems is ®eld
monitoring for pests prior to reaching a management decision. The study objective was to
quantify a common design constraint of commercial IPM sampling programs: the trade-o€
between sampling time and accuracy of the resulting pest management decision. Using consperse stink bug (Euschistus conspersus Uhler) in processing tomatoes (Lycopersicon esculentum Miller), this paper develops a sample size model based on pest spatial distribution and
response from a pest control advisor survey. Results identify a gap between pest sampling
programs developed by University researchers and sampling methods adopted at the commercial ®eld level. The sample size model permits variation in pest treatment thresholds and

treatment decision accuracy. Conclusions support use of this model to satisfy commercial time
constraints while maintaining a reliable level of sampling precision. This study introduces a
novel approach to transferring IPM sampling programs from University research to commercial ®eld adoption. # 2000 Elsevier Science Ltd. All rights reserved.
Keywords: Integrated Pest Management; Pest sampling; Sample size; Decision making

* Corresponding author. Tel.: +1-530-752-4785; fax: +1-530-752-1537.
E-mail address: [email protected] (E.M. Cullen).
0308-521X/00/$ - see front matter # 2000 Elsevier Science Ltd. All rights reserved.
PII: S0308-521X(00)00038-X

100

E.M. Cullen et al. / Agricultural Systems 66 (2000) 99±113

1. Introduction
A key component of Integrated Pest Management (IPM) in production agriculture systems is ®eld monitoring for pests prior to reaching a management decision
(Stern et al., 1959; Flint and van den Bosch, 1981; Kogan, 1988, 1998; Prokopy et
al., 1994). Insect pest sampling programs quantify pest population abundance. Only
after an insect pest is known to be present and an actual or potential economic
threat should a decision be made to apply insecticide(s).

Pest ®eld distribution is an important determinant of the number of samples
required for a precise population estimate. An insect pest with uniform ®eld distribution lends itself to a random sampling program because every sample will have
an equal chance of recovering the pest. For this reason, uniformly distributed pests
require a smaller sample size than pests with an aggregate distribution. An aggregated pest population is better sampled using a strati®ed random sampling program
(Dent, 1991), where the ®eld is divided into sections based on previous knowledge of
pest distribution (e.g. particular locations on plants or areas within a ®eld). Strati®ed random sampling can only be properly implemented when proportions of the
pest population in di€erent ®eld sections are known (Dent, 1991).
The number of samples drawn from an insect pest population relates directly to
the precision of the population estimate. Too few samples will reduce the reliability
of the estimate; however, more samples will increase the cost of the sampling program, where cost is measured in terms of time, labor, equipment or ®nancial outlay
(Dent, 1991; Schea€er et al., 1986). Choosing the number of sample sites within a
®eld involves some risk. If a decision is made to apply insecticide, and treatment is
not economically necessary, ®nancial and biological costs of insecticide application
will be incurred. Growers must also accept a level of risk when a decision not to
apply insecticide is made that treatment will be necessary and economic crop loss
due to pest damage will occur.
Agricultural scientists have two options: (1) they can develop pest monitoring programs using sampling patterns and techniques appropriate for commercial use; or (2)
after developing a research monitoring program, they can modify it for commercial
use (Dent, 1991). IPM sampling programs developed by a combination of basic
and applied University research are reliable in yielding accurate pest-management

decisions. The problem, with regard to IPM implementation, is that University pestsampling protocols are often too time consuming to be practical under commercial
®eld conditions. For example, the need to monitor thousands of hectares and coordinate IPM sampling and treatment timing with equally important crop production
priorities (e.g. irrigation and cultivation schedules) means that growers and pest
control advisors are often left to adapt University monitoring programs to meet
their commercial time constraints (Cullen, 1999).
Objectives of this study were to: (1) quantify a common design constraint of
commercial IPM sampling programs: the trade-o€ between time spent sampling and
accuracy of the resulting pest management decision; (2) develop a model for pest
sampling programs to integrate commercial expectations of reliability and mitigate
the apparent con¯ict between a cost-e€ective sample size and precision of the pest

E.M. Cullen et al. / Agricultural Systems 66 (2000) 99±113

101

population estimate; and (3) introduce a novel approach to transferring IPM sampling programs from University research to commercial ®eld adoption.

2. Materials and methods
2.1. Case study: consperse stink bug in processing tomatoes
Consperse stink bug (Euschistus conspersus Uhler) is a key economic insect pest of

processing tomatoes (Lycopersicon esculentum Miller) in the Sacramento Valley and
Delta areas of California where about 50% of the state's crop is produced (University of California, 1998). Stink bugs feed with piercing-sucking mouthparts,
injecting a toxin that lique®es fruit tissue to aid digestion. Feeding scars appear as
yellow to white areas on the red fruit surface, with a white corky mass of tissue
beneath the peel. Stink bugs are capable of transmitting a pathogenic yeast (Nematospora sp.) between infected and uninfected fruit if the pathogen is carried on the
bug's mouthparts when feeding. Insecticide treatments are routinely applied for
stink bug control in processing tomatoes (Ho€man et al., 1987).
Several aspects of the stink bug pest problem in California processing tomatoes
qualify this system as a case study quantifying the dynamic relationship between
sampling e€ort and precision in commercial production systems:
1. Stink bug sampling methods are established for processing tomatoes. Growers
and their advisors have adopted a canopy shake sample method for stink bug
(University of California, 1998) and indicated interest in utilizing a commercially available pheromone trap as a monitoring tool (Cullen, 1999).
2. Research has established a correlation between stink bug population density in
processing tomato ®elds and fruit damage at harvest (Zalom et al., 1997b).
However, stink bug treatment decision making is complicated by lack of a ®rm
economic injury level because treatment thresholds vary with crop end use (e.g.
whole peel vs. paste). Although processing companies are known to have low
tolerance for stink bug feeding damage on tomatoes, none have established an
ocial tolerance level.

3. California pest control advisors (PCAs) are a group with valuable experience
implementing University-developed pest sampling programs at the commercial
®eld level. Licensed by the state of California, PCAs are the most signi®cant
source of pest control information for processing tomato growers (Flint and
Klonsky, 1985). PCAs monitor ®elds, obtain pest population estimates, determine treatment thresholds and provide growers with written recommendations
for pesticide use.
2.2. Survey of PCAs
A mail survey was conducted to determine how Yolo County, CA, processing
tomato PCAs implement University-developed stink bug sampling techniques at the

102

E.M. Cullen et al. / Agricultural Systems 66 (2000) 99±113

commercial level. The mailing list was compiled from the California Agricultural
Production Consultants Association (CAPCA) membership list. A Yolo County
Cooperative Extension farm advisor reviewed the CAPCA membership list, added
non-member PCAs and eliminated those not working in processing tomatoes. Surveyed PCAs were employed by farm supply companies or as independent consultants. Respondents voluntarily answered the questionnaire entitled, ``Developing
a stink bug monitoring program acceptable to processing tomato PCAs''.
The survey method was adapted from the Total Design Method (Dillman, 1978).

In March 1998 a cover letter, questionnaire and self-addressed, stamped return
envelope were mailed to all 42 Yolo County PCAs identi®ed as working in processing tomatoes. Two weeks later, a follow-up cover letter, replacement questionnaire
and self-addressed stamped return envelope were mailed to non-respondents only.
Two questionnaires were eliminated from the initial PCA population of 42 after
these individuals responded that they no longer worked in processing tomatoes, thus
reducing the total survey population to 40 individuals.
Overall survey response rate was tabulated as the percentage of completed and
returned questionnaires out of the total number of questionnaires mailed. Response
rate for each option per multiple choice question was tabulated as a percentage of
the total number of respondents answering that particular question.
Respondents were asked to identify how much time they spend per ®eld sampling
for all pests (insects, weeds and pathogens) and for stink bugs in particular; factors
leading them to spend more, or less time sampling; preferred stink bug sampling
time intervals throughout the season; and outcomes that would motivate them to
spend additional time sampling for stink bug.
Assuming that no one can be 100% accurate at insecticide treatment decision
making, and taking into account individual grower expectations, PCAs were asked to
indicate the level of accuracy they expect from a stink bug sampling program in terms
of whether or not an insecticide treatment is economically justi®ed. Respondents were
instructed to rate their preferred level of certainty that a decision to treat is correct

and crop damage prevented. On a separate rating scale, respondents indicated their
preferred level of certainty that a decision not to treat was correct and insecticide
treatment costs saved. Respective rating scales ranged from 50 to 95% certainty. The
higher the percentage indicated, the more certain a respondent expected to be that a
decision to treat, or not to treat, for stink bug is economically justi®ed. PCAs were
then asked to indicate how much time they are willing to spend sampling for stink
bugs on an individual ®eld basis to meet their expected treatment decision certainty.
Stink bug insecticide options are restricted by regulations at the federal, state and
processing company levels. Additionally, few registered insecticides are e€ective in
controlling adult stink bugs (Ho€man et al., 1987). Presumably, insecticide ecacy
increases when used against the more susceptible nymphal stages. Referring to a
commercially available stink bug pheromone trap, PCAs were asked: ``If a stink bug
sampling program, based on pheromone trap catch, allowed you to target treatment
to the more susceptible nymphal stages, would you be interested in implementing the
program to increase insecticide ecacy?''. PCAs were then surveyed about costs
related to implementation of the pheromone trap as a monitoring tool.

E.M. Cullen et al. / Agricultural Systems 66 (2000) 99±113

103


2.3. Field sampling
Six commercial processing tomato ®elds, each 32±40 ha in size, were sampled in
the Sacramento Valley (two ®elds in 1996, four ®elds in 1997). Fields were sampled
weekly from mid-June ¯ower initiation to harvest (August±September). Samples
quanti®ed E. conspersus population abundance and ®eld distribution throughout the
season.
Tomato plants were sampled using the canopy shake sample (Zalom et al., 1998)
recommended by the University of California. One shake sample consisted of placing a 45.735.6 cm2 plastic cafeteria tray on the bed beneath a plant and shaking
the plant ®ve times over the tray to dislodge stink bug nymphs and adults from the
canopy, as well as scanning the ground under and around the tray for stink bugs.
Stink bugs dislodged from the canopy were recorded as number of bugs per tray.
In 1996, there were 22 sample sites per ®eld (Fig. 1). In 1997, 16 sample sites were
located in each ®eld (Fig. 1). At each site, ®ve shake samples were taken with 3 m
between samples within the same and adjacent rows. Data collected from each shake
sample included number of stink bugs per tray, genus and species, developmental
stage, sex and presence or absence of reproductive diapause as indicated by ventral
coloration (Toscano and Stern, 1980). Sampling, at the level of detail reported in
this study, required an average of 4.0 min per sample site.
A more feasible sampling routine for commercial implementation was estimated at

2.5 min (one to two shake samples) per site. Although a sample size of one to two
shake samples is less accurate than ®ve shake samples per site, this estimate was
based on the assumption that E. conspersus population densities approaching those
where control action would be taken will be similarly detectable by the more commercially acceptable, 2.5-min sample size.

Fig. 1. 1996 and 1997 stink bug ®eld sampling patterns. Each sample site (®ve samples per site) is represented by an oval. Sample distances into ®eld are labeled in meters. 1996 sampling grid contained 22 sites,
including inner locations extending into ®eld from edges adjacent to stink bug overwintering/alternate
host habitat (12.2, 24.3 and 97.3 m). 1997 sampling grid contained 16 sites, with inner locations removed.

104

E.M. Cullen et al. / Agricultural Systems 66 (2000) 99±113

2.4. Sample size model
Taylor's a and b coecients, taken from Taylor's Power Law (Taylor 1961),
describe the relationship between variance and mean (S2 ˆ a xb ) for individuals
distributed in a natural population (Southwood, 1978). For each ®eld, the mean and
variance of stink bugs caught in ®ve shake samples per site from all sample sites
within a ®eld were determined for each weekly sampling date. Zero counts were
omitted from mean and variance calculations. Taylor's a and b coecients were calculated for each ®eld by ln±ln linear transformation of the mean-variance data,

where b is the slope of the transformed data, and a equals the antilog of the transformed intercept (Taylor, 1961; Wilson, 1985).
An equation for estimating pest sample size that includes treatment decision certainty was developed by Karandinos (1976). Ruesink (1980), Wilson and Room
(1982) and Wilson (1985) incorporated Taylor's Power Law into Karandinos'
equation to form the sample size model used in this study:
n ˆ t2 =2 Dÿ2 a x

bÿ2

:

The model contains both variable and constant factors. The variable factors are:
1. n=sample size (sample sites per ®eld).
2. t2 =2 =standard normal variate for a two-tailed con®dence interval. Commercially, this is the percent certainty growers and PCAs expect from their stink
bug treatment decisions.
3. x=treatment threshold; the mean number of stink bugs per sampling tray at
which treatment is initiated. Zalom et al. (1997b) found that ®ve stink bugs per
2 m of row (0.33 bug per tray) results in 5% damage, the maximum damage
acceptable to many growers.
This study varied x to simulate two commercial scenarios:
I. x=0.33 bug per tray treatment threshold for whole peel canning ®elds

where stink bug feeding damage is readily apparent in the processed product. If growers require a highly precise stink bug population estimate to
assure minimal fruit damage, the number of sample sites per ®eld increases.
II. x=0.50 bug per tray treatment threshold for paste, sauce and dice ®elds
where stink bug damage is less apparent in the processed product. If growers can tolerate a less precise stink bug population estimate and a relative
increase in fruit damage, the number of sample sites per ®eld decreases.
4. D ˆ x ÿ xadj =xadj ; a ®xed proportion of the treatment threshold representing a
range of accuracy approximating the true stink bug population mean. The
value of xadj is the treatment threshold (x) adjusted by a percentage value.
For example, choosing a 10% accuracy range around a treatment threshold
of 0.33 bug per tray, x=0.33ÿ(0.330.10)=0.297. Then D=(0.33ÿ0.297)/
0.297=0.11.
The accuracy range (D) a grower or PCA accepts in the ®eld can be thought
of as the commercial con®dence interval. As D decreases, the commercial

E.M. Cullen et al. / Agricultural Systems 66 (2000) 99±113

105

con®dence interval narrows and the number of sample sites per ®eld necessary
for an accurate treatment decision increases.
Constant factors in this model are:
5. Taylor's coecients, a and b. Taylor's coecients were derived from 2 years of
E. conspersus mean and variance sampling data from six commercial ®elds.
The coecients, a and b, are population statistics representing a species' ®eld
distribution and degree of aggregation.

3. Results
3.1. Survey of PCAs
The questionnaire, ``Developing a stink bug monitoring program acceptable to
processing tomato PCAs'', yielded a response rate of 68%, with 27 of 40 questionnaires completed and returned.
Eighty-nine percent of respondents (37 and 52%, respectively) indicated they
spend 15 or 30 min in a typical (e.g. 40 ha) processing tomato ®eld per visit. One
®eld visit included a combination of insect, weed and pathogen sampling; soil fertility programs; monitoring crop stand, fruit ripening and other agronomic factors
as appropriate. It is important to note variation in total sampling time per ®eld as
100% of respondents adjust sampling time according to pest pressure and other
commercial production variables (Table 1).
Seventy-eight percent of respondents (22 and 56%, respectively) allocate 10 or 15
min, as a proportion of total ®eld monitoring time, to stink bug sampling when
necessary. However, 67% (18 of 27) of respondents do not spend an equal amount
of time sampling for stink bug in each of their processing tomato ®elds. Respondents answering this question selected commercial production variables causing
them to spend more time in some tomato ®elds than in others: a ®eld history of stink
bug damage (52%); ®elds intended for whole peel processing (41%); degree of
Table 1
Factors leading pest control advisors (PCAs) to increase or decrease total sampling time per ®elda
Factors leading PCAs to increase total sampling
time per ®eld

Factors leading PCAs to decrease total sampling
time per ®eld

Early season during stand establishment
Late season during fruit development
``High'' insect pressure, especially fruitworm,
lygus bug and stink bug
Fruit damage apparent. Further sampling
necessary to quantify pest abundance
Weather conditions present disease threat
Field size 40.5 ha or larger

Interval from stand establishment to fruit set
``Low'' insect pressure

Field size under 40.5 ha

a
Factors identi®ed by 100% of respondents (n=27) reporting variation in total sampling time per ®eld
throughout the growing season.

106

E.M. Cullen et al. / Agricultural Systems 66 (2000) 99±113

grower tolerance to stink bug presence (37%); processing company restrictions on
insecticide use (26%) and ®eld size (22%).
PCAs were asked whether they prefer one to two time-intensive stink bug samples
per ®eld, or several less time-intensive stink bug samples per ®eld, throughout the
growing season. Eighty-eight percent of respondents (44 and 44%, respectively)
chose several 15-min stink bug samples per ®eld or multiple ``quick checks'' (