Quantitative and theoretical analysis of species distribution models for invasive species risk assessment and management

Abstract

Biological invasions are a major component of global change leading to numerous impacts on biodiversity, natural and managed ecosystems and natural resources. Despite being more vulnerable to biotic exchange, freshwater ecosystems have tended to receive less attention in invasive species research. There is an urgent need to maintain freshwater ecosystem ecological quality which requires a deeper knowledge about the species that may cause a biological invasion, as well as the process of invasion. To prevent and mitigate the effects of invasive species, biosecurity systems have been implemented in all trading nations around the world. In such systems, the estimation of species potential distribution is key to invasive species risk assessment, and decision making around that risk. Species distribution models (SDM) are therefore important tools used for early detection and are helpful for the design of improved programs for management of such species. However, these models rely on a number of assumptions that may not be valid for all species and all taxa, and there are concerns about the reliability of model predictions in new areas. The main aim of this research was to investigate the key assumptions that underlie the use of species distribution models, to understand their impact on predictions, and to determine the reliability of those predictions.

SDM characterize the species niche to infer predictions in new areas by relying on two main assumptions: the species niche remains unchanged during the invasion (niche conservatism) and that target species occupies all the suitable environments (species are at equilibrium with the environment). This thesis investigated whether these assumptions hold and, if they do not, how they could impact species distribution predictions. To further determine the reliability of SDM predictions in new areas, I investigated the impact of extrapolation when models are required to project into new areas, and the impact of evaluation methods on model performance on the reliability of predictions. In contrast with previous research, and, in an attempt to achieve some generality, the current global distributions of a large number of invasive freshwater invertebrate (22), along with global climate records were used to test assumptions of niche conservatism, species equilibrium and the impact of extrapolation and methods of evaluating model performance.

The analysis of the global distributions of the 22 selected species in this research showed that 90% of the species did not conserve their native niche in invaded areas and were able to establish in novel environments in the invaded range. Contrary to other studies on other taxa, this result indicates that niche conservatism may be rare in invasive freshwater invertebrates and suggests that current predictive tools may underestimate the potential range of freshwater invertebrates in new areas.

Using nine species for which there were sufficient data over the invasion process, the equilibrium assumption was also challenged by niche analysis of each species in relation to residence time. As might be expected, species were found to progressively fill their niche in the invaded range (from 21 to 195 years) with increasing residence time. For the selected freshwater invertebrate species, the average number of years to reach equilibrium in the invaded range was 122 years which is faster than the time shown for other taxa. Moreover, using early invasion records I found that the selected species colonized environments different from those occupied in their native niche. Such results suggest that models constructed at earlier stages of invasion using only native information are likely to underestimate the species potential distribution.

Current concerns regarding the impact of extrapolation on model predictions were also confirmed where such predictions had high uncertainty. However, contrary to expectations, similar performing models showed high levels of uncertainty when predictions were interpolated. Of key interest, was the finding that most of the uncertainty was explained by how each model characterized the species response to the environment. Additionally, the limitations of current evaluation methods were demonstrated, which for some models tended to inflate their performance, thereby increasing uncertainty. In particular, the impact of initial steps in model building (pseudo-absences generation), on the ability to estimate model performance, was demonstrated. This research highlights the importance and value of investigating species niche dynamics and their assessment before implementing species distribution models. The results showed that characterising the niche of species and niche dynamics to investigate SDM assumptions, could improve invasive species surveillance tools and contribute to invasive species risk assessments. For example, I demonstrated that information about niche conservatism can assist the prioritization of surveillance areas. Additionally, the study of niche dynamics enables the identification of species of concern that are able to occupy novel environments as well as those that are able to spread rapidly across the invaded range. However, the analysis of the impact of extrapolation as well as interpolation showed the existing challenges of prediction in novel areas. As well, the analysis demonstrated the importance of validating models using multiple approaches to identify models that provide more reliable predictions that can be used in the early detection and management of invasive species. Conversely, because each model characterises the species response to the environment differently, a potential solution could be to combine predictions based on models having similar behaviours. Additionally, this work identified for a set of SDM, a range of model behaviour with various degrees of complexity. As well the trade-off between using complex and simple modelling techniques to characterize species distribution, was discussed.

Additionally, the finding that current evaluation methods can inflate model performance has led a proposal in this thesis, of a framework to obtain a more rigorous evaluation by stratifying datasets to evaluate model performance and generate pseudo-absences in a masked geographic background.

As a body of work this thesis clearly illustrates some of the opportunities for using more detailed analyses of species distribution data for invasive species risk assessment but also highlights the challenges associated with predicting the potential distribution of freshwater invertebrates, in particular, and thus the vulnerability of freshwater ecosystems. To reduce the uncertainty associated with our knowledge and understanding of freshwater species niche and to reduce uncertainty of prediction, this research highlights the urgent need for greater availability of more appropriate global datasets for further freshwater ecosystems studies.