Abstract
The Banana Bunchy Top Virus (BBTV) is one of the most economically important vector-borne banana diseases throughout the Asia-Pacific Basin and presents a significant challenge to the agricultural sector. Current models of BBTV are largely deterministic, limited by an incomplete understanding of interactions in complex natural systems, and the appropriate identification of parameters. A stochastic network-based Susceptible-Infected model has been created which simulates the spread of BBTV across the subsections of a banana plantation, parameterising nodal recovery, neighbouring and distant infectivity across summer and winter. Findings from posterior results achieved through Markov Chain Monte Carlo approach to approximate Bayesian computation suggest seasonality in all parameters, which are influenced by correlated changes in inspection accuracy, temperatures and aphid activity. This paper demonstrates how the model may be used for monitoring and forecasting of various disease management strategies to support policy-level decision making.
Introduction
Banana Bunchy Top Virus (BBTV), transmitted by the banana aphid (Pentalonia nigronervosa), represents a critical threat to food security and the banana industry, particularly in the Asia-Pacific region. Management typically involves "rogueing"—the physical removal of infected plants—to halt transmission. However, existing epidemiological models often rely on deterministic frameworks that fail to capture the inherent randomness of biological spread or the spatial complexities of real-world plantations. There is a pressing need for flexible models that account for "noisy" field conditions and provide actionable forecasts for biosecurity.
Methods
The researchers developed a stochastic network-based Susceptible-Infected (SI) model to simulate BBTV dynamics within a 12-hectare plantation in Newrybar, NSW, Australia.
Network Structure and Data
Instead of modeling individual plants, the plantation was divided into "nodes" based on physical subsections separated by paths. A node is classified as "Infected" if it contains at least one symptomatic plant. The study utilized 38 months of GPS-tagged longitudinal data for model calibration and an additional 7 months for validation.
Model Parameters
The model tracks three primary transition probabilities:
- $\theta_0$ (Recovery): The probability an infected subsection returns to a susceptible state through successful rogueing.
- $\theta_1$ (Neighboring Infectivity): Local spread between adjacent subsections, representing short-distance aphid movement.
- $\theta_2$ (Distant Infectivity): Long-range spread to non-adjacent subsections, representing wind-assisted or long-distance aphid flight.
Approximate Bayesian Computation (ABC)
Due to the computational complexity of the likelihood function in this stochastic system, the authors employed Approximate Bayesian Computation via Markov Chain Monte Carlo (ABC-MCMC). This method estimates parameters by comparing millions of simulations to observed field data, retaining those that best replicate the historical infection patterns. To account for environmental influences, parameters were estimated separately for summer and winter cohorts.
Key Insights and Results
The model revealed significant seasonal dependencies in disease dynamics.
- Seasonality and Inspection Paradox: Both neighboring ($\theta_1$) and distant ($\theta_2$) infectivity increased during summer, likely due to higher aphid activity and storm-facilitated dispersal. Interestingly, the model showed higher "recovery" ($\theta_0$) in winter. The authors attribute this to an "inspection paradox": slower plant growth in winter makes symptoms harder to detect, leading to fewer reported infections rather than actual virus eradication.
- Parameter Identifiability: A negative correlation was found between neighboring and distant infectivity, suggesting that the total infection count can be explained by either frequent short-range jumps or fewer long-range events.
- Predictive Performance: The model achieved an Area Under the Curve (AUC) of 0.65, representing a "fair" predictive capability and a 40% improvement in binomial deviance loss over a random baseline. Forecasts suggested a steady-state infection probability between 45% and 57%.
Discussion and Future Work
While the model provides a robust framework for within-field dynamics, it has limitations. The current 2D representation ignores topography and specific wind vectors, which are known to influence aphid flight paths. Furthermore, the binary resolution (infected vs. susceptible) does not capture the intensity of the outbreak within a subsection.
Future research could enhance the model by integrating real-time meteorological data and accounting for human factors, such as inspector experience. Despite these limitations, this stochastic network approach offers a valuable tool for evaluating disease management strategies and supporting policy-level decision-making in banana biosecurity.
Summary updated