Use of geospatial methods to characterize dispersion of the Emerald ash borer in southern Ontario, Canada

摘要

Since the introduction of the Asian Emerald Ash Borer beetle (EAB, Agrilus planipennis) to Southern Ontario in 2002, all species of ash trees (Fraxinus) in the province are currently at risk. Due to the aggressive nature of this beetle, early detection is critical in its eradication and can be facilitated by species distribution modelling. That said, several issues need to be addressed in order to increase the predictive accuracy. In this study, the effects of sampling bias such as positive spatial autocorrelation and data prevalence (i.e., proportion of presence to absence points) were investigated in an EAB dataset. A filtering distance threshold approximating the EAB's dispersal range was used to minimize the effects of autocorrelation and the most appropriate prevalence was determined during the modelling process. To analyze the impact of environmental and anthropogenic predictors on the distribution of the EAB, logistic regression, Random Forest (RF) and a hybrid of Random Forest and generalized linear models known as the Random Generalized Linear Model (RGLM) were applied to EAB data from 2006 to 2012 across Ontario. Approximately 80% of the EAB samples were used as training, with 20% as validation. Ultimately, three risk maps were created from the 2006-2012 EAB data by using the coefficients from logistic regression as weights and an automated risk map tool for the RF and RGLM models. High-risk areas were identified from the risk maps for species prevalence and distribution monitoring. From these, precautionary measures can be implemented to stem the expansion of the beetle and thus reduce the destruction of the Ash tree species. All models identified June wind speed as the most important predictor variable followed by population centres. Lastly, Random Forest had the best sensitivity (86%), followed by stepwise backward logistic regression (82%), and RGLM (77%) for the 2013 prediction dataset.