Abstract:
Myrtle rust has been present in New Zealand since 2017 and represents a substantial threat to several important native plant species. Caused by the invasive fungal pathogen Austropuccinia psidii, myrtle rust infects the new growth of plant tissue in susceptible host species of Myrtaceae. Different species appear to vary in their susceptibility to infection by myrtle rust, and I investigated whether it was possible to predict the susceptibility of New Zealand species to infection based on observations from Australia where myrtle rust has been present since 2010. A trait-based framework was used to link species’ functional traits to their susceptibility, and I selected functional traits to use on the basis of expert opinion as to whether they would likely be influential in the infection process. Trait and myrtle rust susceptibility data were provided from Australia, but these datasets were variable in the extent of available trait values per species. I attempted to improve my test dataset by imputing values using two different algorithms, Phylopars and MissForest separately. I then used these two corrected datasets to construct models linking trait values to susceptibility, built with the machine learning algorithm Random Forests. These models were then applied to trait data from New Zealand species to obtain predictions of susceptibility. Key species including Lophomyrtus bullata and Metrosideros excelsa were predicted to be highly susceptible to infection by myrtle rust, matching observations from other studies. Other species, including Kunzea ericoides sensu lato and Syzygium maire were predicted to be less susceptible. Although the model suffered from problems stemming from issues with the original trait data, these predictions can still be used to inform Myrtaceae conservation and management of myrtle rust in New Zealand’s natural ecosystems. This study introduces a novel method of using traits to predict pathogen susceptibility, but better data is needed in order to improve prediction accuracy.