Multivariate Bayesian analysis to predict invasiveness ofPhytophthorapathogens

Abstract

AbstractGlobal concerns are many for the invasive impacts ofPhytophthorapathogens on native vegetation, agriculture, nurseries, and urban parks and gardens. We compiled a database of 32 traits on 204 species ofPhytophthoraincluding data on each species' taxonomy (clade and subclade), historical knowledge (years since first described), impacted ecosystems, microenvironments inhabited, dispersal mode, physiology, and morphology. Drawing from approximately 11,394 unique host, pathogen, and country plant disease records from GenBank and other sources, we calculated potential invasiveness of 103 better studied species from cluster relationships. We used the species data to create a Bayesian network model predicting the degree and probability of invasiveness of individualPhytophthoraspecies. Model calibration testing resulted in <1% error rate in classifying invasiveness categories of well‐known species. We applied the model to predict the potential invasiveness of 101 other species with unknown invasiveness dynamics. The model can also be used to predict the invasive risk of other poorly studied and newly identifiedPhytophthoraspecies, and the general modeling approach can be used for other pests and pathogens, to advise land and resource managers to thwart potential invasions before they occur or intensify.