Thesis
Spatio-temporal modelling of detection and spread of invasive plant pests
- Creator
- Rights statement
- Awarding institution
- University of Strathclyde
- Date of award
- 2025
- Thesis identifier
- T17240
- Person Identifier (Local)
- 201955859
- Qualification Level
- Qualification Name
- Department, School or Faculty
- Abstract
- Invasive plant pests significantly threaten agriculture, ecosystems and economies. Effective control requires early pest detection and reliable spread assessment, which can be addressed through statistical analysis. This thesis develops novel statistical approaches to data analysis and modelling for two invasive insect species: Fall Armyworm (FAW) in India during the period 2018-19, and the Emerald Ash Borer (EAB) epidemic in the USA from 2002 to 2020. The decision-making for prevention and control is often hampered by the lack of data or their low quality. Novel detection methods have been developed, but they have not yet been analysed rigorously. The FAW data are based on a citizen science approach utilising Plantix, an innovative method that integrates artificial intelligence (AI) with mobile technologies. Constructed by Progressive Environmental and Agricultural Technologies (PEAT) GmbH, it can provide comprehensive monitoring of geographical areas and early detection of pest invasions. However, there is no gold standard, and the data need to be statistically interpreted before they can be used to estimate prevalence. For the EAB data, a different approach is needed as only an initial true positive case was provided from each observed infested county in the USA. However, we also have data on the host (ash trees) density and climate forcing. For the two cases, the main research objectives are: (i) developing a rigorous framework for estimating FAW prevalence and using it to estimate the true prevalence in different parts of India, and (ii) developing a continental-level model for the spread of EAB in the USA. Both approaches apply frequentist and Bayesian techniques, using classification methods and several diagnostic performance tools to compare model outputs with data. A classification model, a bi normal mixture, was used to estimate the True and False FAW observations, using the data classification by the Plantix mobile app and our assumptions. A Bayesian meta-analysis estimates pooled test sensitivity and specificity, assuming the logit sensitivity and specificity follow a multivariate normal distribution. Four distinct methodologies were implemented to select the most appropriate model for estimating FAW prevalence, including frequentist methods and the Bayesian metaanalysis with stochastic sensitivity and specificity. In the case of the EAB, a colonisation-dispersal model was adapted and utilised to include climatic (annual average of growing degree day), non-climatic (ash intensity habitat) conditions, and dispersal mechanisms. The model was fitted to the best available data, quantifying the uncertainty in the model and its predictions and assessing its performance in tracking the spread of EAB over two decades. The thesis analysis yields key findings for both FAW and EAB. These findings classify positive and negative Plantix app observations as True or False, evaluate app accuracy and enable estimation of FAW prevalence. Additionally, the evaluation of the data sensitivity and specificity for each maize season is more accurate than for the entire period. Significant factors for EAB colonisation are ash species availability, and the adult EAB flights dispersal distance. The results also highlight that the citizen science and mobile technologies can aid the government in early pest detection for effective spread control of FAW, EAB, and similar pests, and may even combined with inspector monitoring to limit the EAB spread.
- Advisor / supervisor
- Kleczkowski, Adam
- Resource Type
- DOI
Relations
Items
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
PDF of Thesis T17240 | 2025-03-12 | Public | Download |