Forecasting Pathogen Dynamics with Bayesian Model-Averaging: Application to Xylella fastidiosa.

Area of Science:

• Ecology
• Epidemiology
• Computational Biology

Background:

• Accurate forecasting of invasive pathogen dynamics is crucial for effective eradication and containment strategies.
• Partial differential equation (PDE) models are commonly used for invasion modeling but can suffer from rigid behavior and data-model mismatches.
• Relying on a single model can lead to inaccurate forecasts due to inherent uncertainties.

Purpose of the Study:

• To develop an improved forecasting framework for invasive pathogens by integrating Bayesian model averaging (BMA) with mechanistic PDE models.
• To account for both parameter and model uncertainties in pathogen spread predictions.
• To assess the performance of BMA against traditional forecasting approaches using real-world data.

Main Methods:

• Proposed a set of competing PDE-based models to represent pathogen dynamics.
• Employed an adaptive multiple importance sampling (AMIS) algorithm for parameter estimation within a mechanistic-statistical framework.
• Evaluated model posterior probabilities and applied BMA to generate posterior distributions and forecasts.

Main Results:

• The BMA approach successfully integrated multiple PDE models, accounting for parameter and model uncertainties.
• Parameter estimation was performed using surveillance data for Xylella fastidiosa in South Corsica, France.
• The BMA forecast demonstrated superior performance compared to competing forecasting methods when validated against independent data sets.

Conclusions:

• Bayesian model averaging provides a robust method for forecasting invasive pathogen dynamics by synthesizing information from multiple models.
• This approach enhances prediction accuracy and reliability, crucial for managing emerging plant diseases like Xylella fastidiosa.
• The study highlights the value of combining mechanistic modeling with statistical uncertainty quantification for ecological forecasting.