The yellow flower thrips, Frankliniella occidentalis, is a global invader and a polyphagous pest of key agri- and horticultural crops in diverse field and greenhouse environment because F. occidentalis damaged the crop directly by feeding and oviposition and indirectly through transmission tomato spotted wilt virus (Brødsgaard, 1989;Ullman et al., 1997;Schneweis et al., 2017). Compared with other thrips, F. occidentalis has the highest transmission rate of tomato spotted wilt virus with 72.9% (Yoon et al., 2020). Tomato spotted wilt virus has a broad host range, infecting over 1000 plant species, including hot pepper, and causing economic loss (Pittman, 1927). Hot pepper is one of the 7 major fruit vegetables in Korea. In 2022, the total production area is 29,770 ha, and the output amounts to 68,984 tons (KOSIS, 2022). Therefore, it is essential to control F. occidentalis for reducing infection rate of tomato spotted wilt virus. However, F. occidentalis is difficult to control with only chemical control because of its strong resistance to insecticides. Therefore, it is recommended to apply integrated pest management strategies for controlling F. occidentalis density. For integrated pest management strategies, the density investigation should be carried out first. However, there is a limit to sampling total density data in the greenhouse because it is too labor-intensive, time- and cost-consuming. Therefore, the binomial sampling method is proposed as a method to overcome this problem. The binomial sampling method is based on the statistical relationship between the mean density per sample unit with at least one individual (Binns and Bostanian, 1990) and the accuracy increases when the proportion of sample unit is calculated with more than one tally threshold value. However, since the binomial sampling plan is less accurate than the actual density analysis, a validation test using the independent data should be performed (Naranjo et al., 1996). In this study, a binomial sampling survey of F. occidentalis in the hot pepper greenhouse was conducted and the goal was to establish the optimal monitoring method for F. occidentalis in the commercial greenhouse with a minimum effort.
Material and Methods
Data collection
Data were sampled weekly from April 9, 2021 to July 7, 2021 in a commercial hot pepper greenhouse in Daegok-myeon, Jinju-si, Gyeongsangnam-do. The field’s total area was covered 2,940 m2 (Greenhouse 1, width x length, 32 m × 92 m), and hot peppers were sown in a single row, spaced at intervals of 1 m in width and 20 cm in length. Sampling was carried out in 3 plots (12 m × 20 m) in consideration of the entrance and side windows effects. Within each plot, 20 plants were selected, spaced at intervals of 4 m in width and 5 m in length. A total of 62 weekly sampling was conducted. To sample the Frankliniella occidentalis adult, the plants were divided into upper, middle, and lower parts. The upper part was divided into more than 1.8 m, the middle part was 1.2-1.6 m, and the lower part was 0.7-1.1 m given the plant height is 2 m from the ground to the canopy. 9 flowers were sampled at each location (3 flowers x 3 times) by tapping them into a vial containing 70% alcohol.
Spatial distribution analysis
The spatial distribution per each stratum (upper, middle, and bottom part) was analyzed using Taylor’s power law (TPL; Taylor, 1961). Taylor showed that a relationship between variance (s2) and mean (m) in many species can be expressed by following regression equation (PROC REG; SAS Institute, 2010):
where a, exponential of the intercept, is a sampling factor, and the slope b is an index of aggregation, which is constant for a species and/or stage (Southwood, 1978).
The homogeneity test of each regression line was performed using Analysis of Covariance (ANCOVA; Proc GLM; SAS Institute, 2010) to compare slope and intercept values (Sokal and Rohlf, 1981;Park et al., 1999;Choi and Park, 2015).
Development of binomial sampling plan
There are multiple algebraic forms of the binomial sampling method including an empirical model that employs linear regression to portray the relationship between the proportion of sample units with infestation and the mean density (Wilson and Room, 1983;Binns and Bostanian, 1990). In particular, the relationship between the mean density (m) of F. occidentalis adult and the proportion (PT) in which adults occurred more than T individuals (Kono and Sugino, 1958;Gerrard and Chaing, 1970) is given by equation (2):
where α and β are parameters estimated through linear regression using SAS institute software (SAS Institute, 2010). The coefficient is estimated from the tally threshold (T) 1, 3, 5, which is the minimum number of F. occidentalis adults in 3 flowers. The variance of the estimated mean (υar(ln(m))) from the proportion of sample units infested is calculated to determine the best binomial sampling model for estimating the F. occidentalis density based on the different tally thresholds (Binns and Bostanian, 1990). The variance (υar(ln(m))) is calculated using the method of Schaalje et al. (1991), which is one of several methods for estimating the variance of the mean from the ratio of infested sample units. This variance is used to evaluate and compare the accuracy of the binomial sampling method according to the tally threshold.
with
where MSE is the mean square error from Equation (2), N is the number of data in the regression used to estimate α and β from Equation (2), P is the average value of ln(-ln(1-PT)), is the sample estimate of variance of β, n is the number of sample taken from a population (Kuno, 1986;Gerrard and Chaing, 1970;Nachman, 1984;Binns and Bostanian, 1990), and a and b are parameters estimated from Taylor’s power law (Taylor, 1961). The linear regression was determined for all the data of all the plants. The ratio of the standard error (SE) to the mean (m) is the precision (d=s2/n), and if s2/n is replaced with the values of υar(ln(m)) obtained in Equation (3), it is as follows:
Validation analysis of the sampling plan
The independent data set was obtained for validation analysis of the binomial sampling plan (Naranjo and Hutchison, 1997). The validation data were collected from March 11, 2022 to May 06, 2022 in the same greenhouse as the development data. The Fixed Sample Size (FSS) method and Wald’s Sequential Probability Ratio Test (SPRT) method of Resampling Validation for Sampling Program (RVSP) were evaluated using independent data not used in development for the validation of the binomial sampling method.
To verify that the data fitted well within 95% confidence limits, the equation from Jones (1994), , and the predicted mean on the logarithmic scale were used. The FSS method was used with a fixed sample size of 50 with replacement of observations, and simulation was repeated 1000 times. In the FSS, the tally threshold (T) value was set to 3, based on the parameters of an empirical binomial sampling model. For the SPRT method, the upper and lower bounds were established as 0.55 and 0.32, respectively. The type Ⅰ and Ⅱ errors were set at α and β, both at 0.1.
Result and Discussion
Population dynamics of Frankliniella occidentalis in pepper greenhouse
Based on visual data, F. occidentalis adults showed a higher density in the upper stratum than other strata due to the behavioral characteristic of moving towards the upper part of the plant. The highest density was observed on April 23 and gradually decreased thereafter (Fig. 1).
This decrease is caused by the weekly chemical control using a pesticide with emamectin benzoate and flometoquin. In commercial greenhouses, insecticide use prioritizes the quality and yield of the produce, with economical and quick-acting insecticides being the primary choice for controlling pest density. However, because F. occidentalis populations inhabiting greenhouses exhibit more active detoxification enzyme activities than those in open fields or agroforestry interaction areas, they demonstrate strong resistance to various classes of insecticides, including organophosphates, carbamates, pyrethroids, neonicotinoids, and spinosyns (Cloyd, 2016;Fan et al., 2023). Once resistance has been acquired, it can be persisted for a considerable amount of time even without prolonged exposure to the insecticide. This makes it difficult to control the pest density through chemical control alone (Demirozer et al., 2012). As the importance of personal health and environmental issues continue to rise along with economic growth and national income, interest in eco-friendly foods is also increasing (Jung et al., 2017). Therefore, developing an efficient control decisionmaking model using a binomial sampling model can help reduce insecticide usage.
Spatial distribution analysis
Upon examining the spatial distribution of F. occidentalis adult per stratum using TPL, it was discovered that the coefficients of intercept (lna) for F. occidentalis adult were 0.92, 1.01, and 0.92 at the top, middle, and bottom, respectively. Additionally, the slope values that represented concentration (b) were 1.31, 1.35, and 1.25, respectively (Table 1). The derived TPL coefficients indicated that all b values were larger than 1 (P < 0.05) and showed an aggregated distribution.
Through ANCOVA analysis, it was observed that the intercept (F=0.06, df =2, 34, p=0.94) and slope (F=0.14, df=2, 32, p=0.87) did now show any significant differences. Consequently, the TPL coefficient was derived by pooling data and was used to develop a binomial distribution model.
Binomial sampling model & tally threshold determination
In this study, a binomial distribution model of F. occidentalis adults was developed using data collected 13 weeks from April 9 to July 7, 2021. The tally threshold was set to 1, 3, and 5, and modeling was performed with 18 to 29 data points, excluding those with PT values of 0 and 1, according to the T value through Equation (1). The empirical model suitable for the experimental field was analyzed as 1,3 with an r2 value of 0.92 ~ 0.96, but the most stable T value was determined to be 3 through the MSE value (Table 2, Fig. 2). Choosing an appropriate tally threshold in the binomial distribution model is crucial for efficient pest management and reduction of influence of variability and bias in the estimated mean (Naranjo et al., 1996;Nyrop and Binns, 1991). According to Fig. 1, when T = 1, 3, and 5, the mean density per sample unit is 7, 10, and 19 when the infection rate is 95%. It is important to note that the mean density was analyzed based on F. occidentalis adult. A study conducted in Florida reported that the acceptable level of damage caused by F. occidentalis adults is 6 per pepper flower, and if 3 pepper flowers are used as a sample unit, the acceptable level of damage is 18. However, the allowable damage level of F. occidentalis larvae is only 2 per pepper flower, and the sample unit of this experiment is 6, which is higher than the expected damage level of adults (Demirozer et al., 2012). Furthermore, although F. occidentalis adults and larvae share the same food resources, there is little competition due to the presence or absence of locomotion, and both types transmit Tospowiltvirus (Northfield et al., 2011;Reitz et al., 2020). Therefore, considering the mean density level and the damage tolerance level according to the tally threshold, the appropriate T value was determined to be 3.
The analysis of a fixed sample size showed no significant difference between the infection rate and sampling precision based on the T value. Similar results were also found in other studies (Kang et al., 2016;Naranjo et al., 1996;Moerkens et al., 2018) However, there was a difference in precision according to the T value (Fig. 3). It was confirmed that the appropriate T value for the binomial distribution model for the F. occidentalis adults in the pepper greenhouse was 3, and the sampling precision level was lower than when T was set to 1 or 5. This indicates that control decisions can be made with a small number of samples using the binomial distribution model. However, according to Southwood (1978), the most appropriate precision level in the binomial distribution model is 0.25, and in comparison with the results of this study, the precision level of T=3 was 0.4 to 0.8. This difference can be attributed to the low density levels in the greenhouse, which were maintained through continuous chemical control using insecticides at weekly intervals. In April, the density of F. occidentalis adults in the pepper greenhouse was relatively high, but it gradually decreased and maintained an average of 0 to 2 from the end of May (Fig. 1). Population changes like these can affect the accuracy of the sample data, and thus, it is important to introduce integrated pest control measures in greenhouses. It is worth noting that a paper previously published in Andong showed a difference in the female ratio of F. occidentalis adults depending on the pepper cultivation period, with an increasing ratio reported (Kim et al., 2022).
Validation analysis of binomial sampling model
The precision of binomial sampling plans is crucially reliant on the relationship between the mean density and the proportion of 3-flower-unit infested at T=3 (Naranjo et al., 1996). In this study, all independent data fell within the 95% confidence limit (Fig. 4). The Fixed-Sample-Size (FSS) method and Wald's sequential precision ratio test (SPRT) method were employed to compare the predicted and actual mean density, and the infestation rate per 3-flower-unit. The predicted and actual means were found to be similar, and the infestation rate showed no significant differences (Table 3). These results indicate that a binomial sampling plan was effective in accurately estimating the mean density of F. occidentalis in the pepper greenhouse.
Sampling is a fundamental research activity in ecology and also serves as a critical foundation for Integrated Pest Management (IPM). IPM, which is an ecological approach to pest and crop management based on current information, aims not at complete eradication of pests but at maintaining pest populations below the economic injury level (Nyrop and Binns, 1991). This is achieved by implementing multiple control strategies, minimizing chemical control, and making use of pest population sampling, economic thresholds, and decisionmaking (Pedigo and Buntin, 1994).
Greenhouses create a high-risk environment for pest proliferation due to controlled temperature and humidity, making pest sampling labor- and time-intensive, particularly for small pests like F. occidentalis (Ugine et al., 2011). These practical challenges in greenhouse pest monitoring highlight the need for more efficient methods. Therefore, the binomial sampling plan is used as a viable alternative. Instead of counting pest numbers directly, this method observes the presence or absence of pests, significantly reducing time and cost associated with density assessment (Kuno, 1986). Additionally, the binomial sampling plan allows for threshold-based decision-making in pest control (Pedigo and Buntin, 1994).
By applying the binomial sampling model, greenhouse managers can make timely and accurate pest control decisions, optimizing pesticide use based on pest presence thresholds rather than exact counts. This approach enables greenhouse managers to reduce unnecessary pesticide use, thus minimizing environmental impact on greenhouse conditions and surrounding ecosystems. Therefore, it aligns with IPM's ecological framework by focusing on sustainable pest management within economic thresholds (Binns and Nyrop, 1992).