Establishment risk modelling of invasive organisms: Solenopsis invicta as a case study

1. lnvasive species are a growing threat to global biodiversity and economic systems. Over the past decade, researchers have become increasingly interested in predicting sites at risk of establishment by invasive species. Establishment risk assessment can increase the efficiency and viability of sampling and control programs. However, the task of prediction is complex because of issues including: ecological interpretation, data availability, statistical model design, model evaluation, and working within complex anthropogenic frameworks. Consequently, one cannot predict with certainty the probability that an invasive pest will establish a site. I examine methodological gaps contributing to this problem using the invasive organism, Solenopsis invicta, the red imported fire ant, as a case study. 2. 1 address the problems of building predictive models designed for sampling and control purposes across four experimental chapters. First, I propose methods for assessing risks using ranked probabilistic models and high resolution multispectral remote sensed imagery. I integrate empirical, mechanistic, and cross-evaluation methods to resolve biological issues involving variably scaled dispersal processes. Second, I devise a method for evaluating the performance of the statistical component of a species distribution model. I characterise similarities and differences in the predicted fits of a ranked probabilistic model and a generalised additive model. This provides information regarding how confident we can be in each model. Third, I propose methods for testing the performance of high resolution predictive models when applied to new locations (that is, generalisability testing). Despite the propensity of invasive pests to disperse great distances, researchers rarely evaluate invasive species models for long distance generalisability. I examine how well locally fitted ranked probabilistic and generalised additive models predict an invasion in Hong Kong, China; I compare this with the performance of models extrapolated from Queensland, Australia. I propose a method for applying models over large distances that can guide control and eradication programs whilst assisting in the development of locally fitted models. Fourth, I identify 'sites' at risk of establishment from an anthropogenic viewpoint relevant to control and eradication programs. Specifically, I create a vector-based very high resolution establishment risk model from Quickbird imagery that is compatible with cadastral data. 3. 1 demonstrate numerous original outcomes. First, I present the first high resolution predictive model for the red imported fire ant. I demonstrate that a two phas