Early detection of emerging forest disease using dispersal estimation and ecological niche modeling.
Journal Article
Overview
abstract
Distinguishing the manner in which dispersal limitation and niche requirements control the spread of invasive pathogens is important for prediction and early detection of disease outbreaks. Here, we use niche modeling augmented by dispersal estimation to examine the degree to which local habitat conditions vs. force of infection predict invasion of Phytophthora ramorum, the causal agent of the emerging infectious tree disease sudden oak death. We sampled 890 field plots for the presence of P. ramorum over a three-year period (2003-2005) across a range of host and abiotic conditions with variable proximities to known infections in California, USA. We developed and validated generalized linear models of invasion probability to analyze the relative predictive power of 12 niche variables and a negative exponential dispersal kernel estimated by likelihood profiling. Models were developed incrementally each year (2003, 2003-2004, 2003-2005) to examine annual variability in model parameters and to create realistic scenarios for using models to predict future infections and to guide early-detection sampling. Overall, 78 new infections were observed up to 33.5 km from the nearest known site of infection, with slightly increasing rates of prevalence across time windows (2003, 6.5%; 2003-2004, 7.1%; 2003-2005, 9.6%). The pathogen was not detected in many field plots that contained susceptible host vegetation. The generalized linear modeling indicated that the probability of invasion is limited by both dispersal and niche constraints. Probability of invasion was positively related to precipitation and temperature in the wet season and the presence of the inoculum-producing foliar host Umbellularia californica and decreased exponentially with distance to inoculum sources. Models that incorporated niche and dispersal parameters best predicted the locations of new infections, with accuracies ranging from 0.86 to 0.90, suggesting that the modeling approach can be used to forecast locations of disease spread. Application of the combined niche plus dispersal models in a geographic information system predicted the presence of P. ramorum across approximately 8228 km2 of California's 84785 km2 (9.7%) of land area with susceptible host species. This research illustrates how probabilistic modeling can be used to analyze the relative roles of niche and dispersal limitation in controlling the distribution of invasive pathogens.
