Skip to Main Content
Modeling risk for SOD nationwide: what are the effects of model choice on risk prediction?Author(s): M. Kelly; D. Shaari; Q. Guo; D. Liu
Source: In: Frankel, Susan J.; Shea, Patrick J.; and Haverty, Michael I., tech. coords. Proceedings of the sudden oak death second science symposium: the state of our knowledge. Gen. Tech. Rep. PSW-GTR-196. Albany, CA: Pacific Southwest Research Station, Forest Service, U.S. Department of Agriculture: 333-344
Publication Series: General Technical Report (GTR)
Station: Pacific Southwest Research Station
PDF: Download Publication (190 KB)
DescriptionPhytophthora ramorum has the potential to infect many forest types found throughout the United States. Efforts to model the potential habitat for P. ramorum and sudden oak death (SOD) are important for disease regulation and management. Yet, spatial models using identical data can have differing results. In this paper we examine the results from five types of models generated from common input parameters, and investigate model agreement for distribution of risk for P. ramorum. We examine five models: (1) Rule-based, (2) Logistic regression, (3) Classification and Regression Trees, (4) Genetic Algorithm modeling, and (5) Support Vector Machines. The models differed in terms of parametric and non-parametric requirements, necessity for presence/absence data, and whether or not the explanatory variables were determined a priori or revealed during the model process. Nationwide input data included vegetation/host (hardwood diversity and hardwood density), topography, and climate (e.g. precipitation, frost days, temperature, and many other layers). We developed a risk map for the conterminous United States in which probabilities for P. ramorum disease establishment were based not on one model, but on agreement between multiple models. The five models were consistent in their prediction of some SOD risk in coastal CA, OR and WA. All models predicted some risk in the northern foothills of the Sierra Nevada mountains in CA. Outside of the west coast, the combined models predicted highest risk for SOD in a east-west oriented band including eastern OK, central AR, TN, KY, northern MI, AL, GA and SC, parts of central NC, and eastern VA, DL and MD. The paper also discusses issues of input data accuracy, coverage, availability of nationwide host datasets, data scale, and model computational requirements. Although theoretical in nature, the results of this paper have practical and applied value for managers and regulators of this disease.
- You may send email to email@example.com to request a hard copy of this publication.
- (Please specify exactly which publication you are requesting and your mailing address.)
- We recommend that you also print this page and attach it to the printout of the article, to retain the full citation information.
- This article was written and prepared by U.S. Government employees on official time, and is therefore in the public domain.
CitationKelly, M.; Shaari, D.; Guo, Q.; Liu, D. 2006. Modeling risk for SOD nationwide: what are the effects of model choice on risk prediction?. In: Frankel, Susan J.; Shea, Patrick J.; and Haverty, Michael I., tech. coords. Proceedings of the sudden oak death second science symposium: the state of our knowledge. Gen. Tech. Rep. PSW-GTR-196. Albany, CA: Pacific Southwest Research Station, Forest Service, U.S. Department of Agriculture: 333-344
Keywordsgeographic information systems, spatial modeling, sudden oak death
- Phytophthora ramorum canker (Sudden Oak Death) disease risk and progress in coast live oak, 2000-2012
- Mapping the risk of sudden oak death in Oregon: prioritizing locations for early detection and eradication
- The current state of knowledge on operational sanitation measures to lower risk of Phytophthora ramorum spread and the need for further study
XML: View XML