Submitted to: Pest Management Science
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 11/21/2012
Publication Date: 8/28/2013
Publication URL: http://handle.nal.usda.gov/10113/57833
Citation: Elmouttie, D., Keirmeier, A., Flinn, P.W., Subramanyam, B., Hagstrum, D., Hamilton, G. 2013. Sampling stored product insect pests: a comparison of four statistical sampling models for probability of pest detection. Pest Management Science. 69(9):1073-1079. DOI: http://dx.doi.org/10.1002/ps.3469. Interpretive Summary: Sampling stored grain for insect pests is critical for maintaining grain quality and to determine if insect control is necessary. To develop the best sampling protocol for stored grain it is important to select a statistical model that explains how the insects are distributed in the grain, because the model can be used to predict the number of samples necessary to detect insects in the grain. In collaboration with scientists from Queensland University of Technology and Kansas State University, we compared the accuracy of four different statistical models to detect insects in grain. Of the four models, the compound model performed the best under both high and low insect densities. The findings from this study will be used to improve insect pest management programs for stored grain.
Technical Abstract: Statistically robust sampling strategies form an integral component of grain storage and handling activities throughout the world. Developing sampling strategies to target biological pests such as insects in stored grain is inherently difficult due to species biology and behavioral characteristics. The design of sampling programs should be based on an underlying statistical distribution which is sufficiently flexible to capture variations in the spatial distribution of the target species. In this paper we compare the accuracy of four sampling approaches, the negative binomial model, the Poisson model, the double logarithmic model and the compound model, to detect insects over a range of insect densities. Although the double log and negative binomial models performed well under specific conditions, we show that, of the four models examined, the compound model performed the best over a broad range of distributions. Unlike the double logarithmic approach and negative binomial approach, the compound model was shown to be robust when insects were restricted to few samples but density was high. The Poisson approach consistently underperformed, failing to detect insects at the 0.95 detection threshold in comparison to the other three models examined. This paper illustrates that the compound model is robust over a broad data range and leads to substantial improvement of detection probabilities within highly variable systems such as grain storage.