PMN Proeedings: 2011 Field Crops Rust Symposium

2011 Field Crops Rust Symposium:
Abstract

Poster #17

Identifying Highly Connected Counties Compensates for Resource Limitations When Sampling National Spread of Soybean Rust

Presenting Author: K. A. GARRETT (1)

Coauthors: S. Sutrave (1, 2), C. Scoglio (2), S. A. Isard (3), J. M. S. Hutchinson (4)

Affiliations: (1) Dept. of Plant Pathology, Kansas State University, Manhattan, KS, USA; (2) Dept. of Electrical and Computer Engineering, Kansas State University, Manhattan, KS, USA; (3) Dept. of Plant Pathology/Dept. of Meteorology, Pennsylvania State University, University Park, PA, USA; (4) Dept. of Geography, Kansas State University, Manhattan, KS, USA

Resource limitations for maintaining sentinel plot networks are an important incentive for making sampling as efficient as possible. We developed a dynamic network model for U.S. soybean rust epidemics, with counties as nodes and link weights a function of host hectarage and wind speed and direction. The network model framework supports evaluation of the predictions based on information from different subsets of the nodes where soybean rust incidence data were available. We compared four strategies for selecting an optimal subset of sentinel plots, listed here in order of increasing performance: random selection, zonal selection (based on more heavily weighting regions nearer the south, where the pathogen overwinters), frequency-based selection (based on how frequently the county had been infected in the past), and frequency-based selection weighted by the node strength of the sentinel plot in the network model. Reducing the sentinel plot set to 10% of the original set increased the typical error to 20% under random selection, but to only 6% under zonal selection. When the sentinel plot set was reduced to 2.5% of the original set, typical errors for frequency-based selection were 10%, while for frequency-based selection weighted by node strength, typical errors were only 5%. When dynamic network properties for invasive species such as rusts are characterized, this information can be used to reduce the resources necessary to survey and predict invasion progress.