Remote islands are vulnerable to non-indigenous species: Utilization of data analytics to investigate potential modes of introduction and pest interceptions

Barbara Kachigunda

Biosecurity in Australia and globally is based on understanding and protection of our national health, economy, industries, and environment from the negative effects associated with invasive pests and pathogens. The biosecurity continuum includes pre-border preparedness, border protection and post-border management, eradication, and control. The biosecurity system in Australia aims to manage risks and reduce the likelihood and adverse consequences of pest and disease incursions on human, animal and plant health, the environment, and the economy. To identify biosecurity risks and solve pertinent issues in biosecurity, analysts must gather and collate information for multiple factors and from a variety of sources in areas including agriculture, the environment and public health. The amount and complexity of biosecurity data have exposed the limitations in traditional statistical methodologies in addressing issues in biosecurity management. Biosecurity surveillance data is challenging in terms of non-normality, over-dispersion and typically zero-inflated. This type of data follows a natural process rather than a pre-specified process or distribution models, and often contains a large proportion of zeros. Application of appropriate statistical models to analyse these unique data sets is essential to effective biosecurity decision-making. The data used throughout this thesis were typically characteristic of biosecurity data, containing a large proportion of zeros, non-normality, and over-dispersion. Data used were collected as part of a biosecurity program implemented on Barrow Island, a remote island off the western Australian coastline, prior to and during the development of an industrial project on the island. In the following research, the first step encompassed evaluation of a range of candidate statistical models for describing biosecurity border and post-border detection of terrestrial non-indigenous species. The dataset was fitted with a variety of models including lognormal linear model, Poisson and negative binomial generalized linear models, zero-inflated model, a three-component mixture mode and a clustering analysis approach. A clustering analysis approach was adopted using a generalization of the popular k-means algorithm appropriate for mixed-type data. The analysis approach involved determination of the most appropriate number of clusters using just the numerical data, then subsequently including covariates to the clustering. Based on the counts alone, three clusters gave an acceptable fit and provided information about the underlying data characteristics. Incorporation of covariates into the model suggested four distinct clusters dominated by physical location and type of detection. Though the three-component log-normal mixture model provided detailed insight into the distribution of the data by dividing the data according to their distinct characteristic of numerical ordering, the clustering model was the preferred approach for this study. Availability of more relevant data would greatly improve the model. Broader use of cluster models in biosecurity data is recommended, with testing of these models on more datasets to validate the model choice and identify important explanatory variables. Investigation of the diverse routes by which non-indigenous species can be introduced is also of key importance to biosecurity. A gap in many introductory pathway studies is the limited consideration given to multiple introduction pathways occurring simultaneously. Multiple pathways of non-indigenous species introduction to Barrow Island were investigated and fifteen potential modes of introduction were identified in association with importing location and personnel required for the project. Three-way management prioritisation using boosted regression modelling to determine the most important factors influencing the detection of non-indigenous species at the biosecurity border was assessed. Factors considered in detecting non-indigenous species included potential modes of introduction, detection type, border inspection point (physical location on Barrow Island), phase of industrial development, year, and month of detection, of which detection type, border inspection point and potential modes of introductions were key factors. Cargo vessel and inward bound passenger numbers peaked during the construction period and were associated with an increase in the number of live non-indigenous species detections. Exposed potential modes of introductions (e.g. flat racks and vessel topsides) contained a more diverse species assemblage, while potential modes of introductions associated with human habitation and activity had the highest likelihood of introducing live non-indigenous species. The nature of these potential modes of introductions potentially allowed non-indigenous species habitation of niche areas and/or provided a suitable food supply. Invertebrates comprised 73% of the detections, with 43% live non-indigenous species. Structures such as landings and jetties were recorded as invasion hotspots, consistent with being the first point of entry for arriving vessels. Human-inhabited environments reported abundant commensal non-indigenous species. Our study indicates that biosecurity surveillance programs need to prioritise management of specific species, potential modes of introductions, and sensitive and susceptible sites to target potential invasions. Biosecurity managers should prioritise potential modes of introductions with the highest likelihood of live non-indigenous species detection based on specific potential modes of introductions characteristics, including niche availability and habitat suitability. The study provided insight into how biosecurity surveillance programs need to assess current data and adapt management strategies appropriately. Evaluation of the predictive performance of models used in biosecurity surveillance is integral to subsequent management decision-making. This includes assessing the suitability of the model for specific applications, i.e. identifying important potential predictors, undertaking a comparative assessment of competing models and modelling techniques, and identifying aspects of the model that might need improvement. Opportunities to use statistical science for biosurveillance are vast, as are the challenges associated with available data related to biosecurity. This thesis explored a variety of analytical statistical methods to enhance interpretation and decision making in biosecurity, while also acknowledging the challenges associated with this type of data. There is a growing need to leverage scientific models and predictive analytics to improve decision making in the context of biosecurity management.

Remote islands are vulnerable to non-indigenous species: Utilization of data analytics to investigate potential modes of introduction and pest interceptions

Files and links (1)

Abstract

Details

UN Sustainable Development Goals (SDGs)

Metrics