The New Zealand Government has committed to a 250,000 ha expansion of plantation forests by 2020 in order to diversify the forestry sector and capture carbon to mitigate climate change. Whilst introducing novel alien species can bring economic benefits, the risks of future invasion problems have not been fully quantified at the appropriate scale for many species. This is because invasions are a complex mix of species-traits, biogeographic factors, human actions, and also because many long-lived woody species have lag-phases between initial introduction, naturalisation and invasion. This thesis investigates why some species become invasive whilst others do not using the genus Pinus as a model system, and New Zealand (NZ) and Great Britain (GB) as study regions. I improve on previous studies that have addressed this question by accounting for successes and failures across the entire invasion process (which incorporates the stages introduction, naturalisation and invasion).

I compare four methods of quantifying invasion risk by: (a) testing how robust the Australian weed risk assessment tool (WRA) is to methodological issues including taxonomic range, region and knowledge of invasive behaviour elsewhere; (b) quantifying the relative contribution of species, biogeographic, and human factors to invasion success using boosted regression trees (BRT); (c) assessing whether phylogenetic relationships can predict invasion risk, and whether control- ling for phylogeny in Markov chain Monte Carlo generalised linear mixed models (MCMCglmm) changes the importance of species, biogeographic and human factors in invasion success; and (d) dissecting the causal relationships between species, biogeographic and human factors using a novel Bayesian method for exploratory path analysis.

I found that the WRA performed well at discriminating between successful and failed species at the introduction and naturalisation stages (AUC >= 0.80) but not at the spread stage, and these results were consistent between NZ and GB. When I repeated the procedure without information of species' prior invasion behaviour, the WRA was less accurate at distinguishing among species (area under the reciever operating characteristics curve or "AUC" <= 0.73). Thus the WRA may not be a viable approach to risk assessment when this crucial information is unavailable. Boosted regression tree analysis indicated that human (high forestry use index) and biogeographic factors (closer climate match; NZ only) were the strongest predic- tors of introduction success. Human (a high forestry use index, large area planted and longer residence time) and biogeographic attributes (a close climate match and larger native range size) were the strongest contributors to naturalisation (NZ and GB). Species attributes (including the Z-score, a composite measure of pine invasiveness) contributed relatively little compared to other factors at all stages. The BRT method was reliable (introduction stage AUC >= 0.86; naturalisation stage AUC >= 0.98), relatively straightforward, and could be used as an alternative approach to risk assessment when the WRA may fail. I found that there was no phylogenetic signal in introductions, naturalisations, invasions, or in any traits that might determine success for Pinus. Consequently, phylogeny may not be a useful predictor of invasion risk for pines. Phylogenetically controlled MCMCglmm produced the same results as non-phylogenetically controlled models with a similar level of reliability (introduction AUC = 0.92; naturalisation AUC = 1.00). These results suggest that non-phylogenetic models produced reliable …