The structure of global invasive assemblages and their relationship to regional habitat variables: converting scientifically relevant data into decision relevant information

Abstract

Quantitative methods for pest risk assessment combine sound statistical tools with sound ecological theory to convert scientifically relevant data into decision-relevant information. This thesis investigated a quantitative method for pest risk assessment called pest profile analysis (PPA). PPA is a new methodology that is based on the premise that the risk of invasion by crop pests into new areas can be predicted by analysing regional insect pest assemblages (also known as pest profiles). Regional pest assemblages comprise the presence or absence of recognised pest species in each region of the world. The analysis involves clustering these regions based on similarities between their pest profiles. PPA assumes that co-occurrence of pest species in a region is the outcome of a non-random structured process driven by biotic and abiotic characteristics of the region. The most commonly used clustering technique for grouping regional pest assemblages is a self-organizing map (SOM), which is an artificial neural network algorithm. Two other clustering methods that have also been used for PPA are hierarchical clustering (HC) and k-means. The main aim of this thesis was to perform a thorough validation test of the PPA approach. To do so, I first analysed the sensitivity of SOM PPA to changes in the number of species used as input data. The results showed that SOM PPA outputs (weight values that are interpreted as risk indices) were quite sensitive to changes in the input data. However, when the risk indices were transformed into ranked lists of species, the ranks were significantly less sensitive and hence potentially more useful for pest risk assessment. I assessed the validity of the groups (clusters) of regions obtained from a SOM PPA by applying an external validation measure, the ζ diversity metric. The ζ metric was used to quantify similarities between pest profiles within clusters. The results showed it can be used for assessing the uncertainty associated with PPA outputs. I also conducted a temporal study of distributional changes of crop pests worldwide to measure the degree of biotic homogenization that had occurred in the regional pest profiles over 10 years. The major findings were that homogenization is certainly occurring, but it is in an inceptive stage and pest assemblages still remain strongly regionalized. I made a detailed comparison between the SOM, HC and k-means clustering methods to identify the one that produced the most accurate predictions. Unexpectedly, HC performed best. This appears to contradict the main hypothesis behind clustering world's regions according to their pest profiles because the expectations were that since SOM and k-means create a higher number of highly similar clusters, they would provide better predictions. The results of this research showed that PPA can help to prioritise risks of invasion by insect pests. It provided a new measure of uncertainty to improve communication of model results to decision makers. The results highlighted the urgent need for research to identify the determinants of insect pest species' distributions around the globe, and to implement that knowledge into PPA and biosecurity decision making.