Investigating the use of invasion history, meta-analysis and niche-based models as tools for predicting the ecological impacts of introduced aquatic species

Biological invasions pose a major threat to global biodiversity. While there is increasing concern regarding the impacts caused by non-indigenous species (NIS), generalisable tools for predicting their ecological effects have yet to be developed. Several researchers have suggested that examining the previously documented effects of NIS, termed invasion history, can serve as a basis for forecasting future impacts. Yet, while predictive models for impact have been developed based on the invasion histories of several widespread invaders, the generalisability of such approaches has not been demonstrated. The severity of the impacts caused by NIS may vary as a function of their local abundance across invaded sites. Thus by estimating the expected abundance of introduced species, at potential recipient locations, we may be able to identify habitats which are particularly vulnerable to their effects. While ecological niche-based models (ENM) have often been used to predict the abundance of species within their native ranges, such approaches have rarely been applied to NIS. In this thesis I conduct an extensive literature review, using 19 aquatic species, and assess the utility of invasion history for predicting future impacts. I illustrate that, while for most NIS limited and heterogeneous data currently inhibits the development of quantitative predictions, invasion history can often reveal the type and direction of future impacts. Using one of these species, Cyprinus carpio, as a case study, I conduct a meta-analysis and demonstrate that, where data is available, models incorporating NIS biomass can explain a substantial amount of variation in the severity of impacts across invaded locations. I then develop neural network-based ENM to forecast both the occurrence and biomass of C. carpio in a portion of its invaded range, using monitoring data from Minnesota. I test the ability of the resulting models to generate predictions for new sites within the main study region and to extrapolate to geographically independent locations. My results demonstrate that such models have high predictive power, when applied within the region where they were parameterized and while accuracy is reduced in new locations; such models can nonetheless provide insight into the risk posed by NIS.