95th ESA Annual Meeting (August 1 -- 6, 2010)

COS 19-9 - A total systems approach to modeling and pest management in wheat cropping systems

Tuesday, August 3, 2010: 10:50 AM
333, David L Lawrence Convention Center
Ilai N. Keren, Department of Math Sciences, Montana State University, Bozeman, MT, Fabian Menalled, Land Resources and Environmental Sciences, Montana State University, Bozeman, MT, David Weaver, Department of Land Resources and Environmental Sciences, Montana State University, Bozeman, MT, Alan Dyer, Plant Sciences & Plant Pathology, Montana State University, Bozeman, MT and Jim Robison-Cox, Mathematical Sciences, Montana State University, Bozeman, MT
Background/Question/Methods The concentration of wheat production in the Northern Great Plains has resulted in the influx of specialized multitrophic pest complexes whose members interact in both positive and negative ways.  Thus, management recommendations based on the traditional single-species pest control paradigm may lead to undesirable responses. To develop a predictive model based on a total system approach to pest management, we evaluated interactions between wheat stem sawfly, Fusarium crown rot, and cheatgrass and their responses to simple cultural practices, namely varying seeding rates and wheat cultivars. Because field observations may be confounded by the occurrences of multiple interacting pests, the ‘true’ effect on yield of any single member of the complex is not estimable. Also, the effects of management scenarios on any single pest cannot be readily and fully interpreted in the presence of multiple pests. Structural equation models (SEM) have been suggested as appropriate tools for ecological path analysis but parameter estimation in a fully observed network has received less attention. We adopted a Bayesian network approach to simultaneously model pest interactions with each other, and their residual effect on yield, using Gibbs sampling.

Results/Conclusions The Bayesian network in this study extends the classical SEM beyond strictly linear models and allows coefficients of observed nodes to share information during the model fitting process. Parameter estimates obtained from the posterior remain unbiased with variance in the appropriate range when using diffused priors. Analysis of our empirical data indicates increasing cheatgrass cover significantly reduced yields in most cultivar seeding density combinations. Moreover cheatgrass enhances Fusarium infection intensity and changes sawfly preferences for certain cultivars, significantly increasing or decreasing the odds of sawfly attack. These biologically sensible results help validate adopting a Bayesian network model to evaluate the impact of a multitrophic pest complex and are a first step in providing a total systems approach to pest management in wheat cropping systems.