A real-time forecasting framework for emerging infectious diseases affecting animal populations
7 affiliations
- 1School of Mathematics and Statistics, The University of Melbourne, Melbourne, Victoria, Australia
- 2Melbourne Centre for Data Science, The University of Melbourne, Melbourne, Victoria, Australia
- 3Melbourne Veterinary School, The University of Melbourne, Melbourne, Victoria, Australia
- 4Epidemiology, Surveillance and Laboratory Section, Australian Government Department of Agriculture, Fisheries and Forestry, Canberra, Australian Capital Territory, Australia
- 5School of Veterinary Science, University of Queensland, Brisbane, Queensland, Australia
- 6Biosecurity & Food Safety, New South Wales Department of Primary Industries, Orange, New South Wales, Australia
- 7Department of Mathematics and Statistics, Lancaster University, Bailrigg, Lancaster, United Kingdom
Abstract
Infectious disease forecasting has become increasingly important in public health. However, forecasting tools for emergency animal diseases, particularly those offering real-time decision support when parameters governing disease dynamics are unknown, remain limited. We introduce a generalised modelling framework for near-real-time forecasting of the temporal and spatial spread of infectious livestock diseases using data from the early stages of an outbreak. We applied the framework to the 2007 equine influenza outbreak in Australia, generating forecasts at three timepoints across four regional clusters. Prediction targets included future daily case counts, outbreak size, peak timing and duration, and spatial distributions of future spread. We evaluated how well the forecasts predicted daily cases and the spatial distribution of case counts, using skill scores (a measure of probabilistic forecast accuracy) as a benchmark for future model improvements. Forecast accuracy, certainty, and skill improved after formation of the outbreak’s peak, while early forecasts were more uncertain or prone to overestimation, highlighting the need for caution when interpreting pre-peak predictions, particularly when the impacts of control policies on future transmission are not adequately represented in the model. Spatial forecasts of broad, relative risk patterns were more robust than precise predictions of risk at the individual premises level or exact daily cases counts, supporting geographically targeted response strategies. Overall, this framework supports real-time decision-making in livestock disease outbreaks when applied with appropriate consideration of uncertainty, and establishes a foundation for future refinements and applications to other animal diseases. Author summary: Infectious disease outbreaks in livestock can spread rapidly, making timely and well-informed decisions critical for limiting their impact. However, predicting how an outbreak will evolve is difficult, especially early on when little is known about how the disease spreads. In this study, we developed a flexible modelling framework that uses data collected during the early stages of an outbreak to generate real-time forecasts of how disease may spread over time and space. We applied our approach to the 2007 equine influenza outbreak in Australia and assessed how well it could predict future cases and identify areas at higher risk. We found that forecasts became more accurate as more data became available, and that short-term predictions were generally more reliable than those made further into the future. Importantly, while precise predictions of case numbers or individual affected farms were often uncertain, the model was better at identifying “hotspot” areas of higher risk. These findings suggest the model is most useful for guiding targeted surveillance and control to these higher-risk zones, particularly when its uncertainty and assumptions are carefully considered.
Related
- Integrating pangenome and imputation framework reveals structural variants affecting stature and milk composition traits in French dairy cattle
- A Context-Aware Framework for Biological Health Index Harmonization and Machine Learning-Aided Health Forecasting of Waterbodies
- GaitScreen: a ground reaction force-based framework for prescreening gait abnormalities in orthopaedic populations
- A Unified Unsupervised Framework for Genome-Wide Association Studies in Heterogeneous Populations
- The evolution of sex for artificial intelligence: a population-genetic framework for multigenerational model populations
BibTeX
@article{theng2026realtime,
title = {{A real-time forecasting framework for emerging infectious diseases affecting animal populations}},
author = {Meryl Theng and Simin Lee and Andrew C. Breed and Sharon Roche and Emily Sellens and Catherine Fraser and Kelly Wood and Chris P. Jewell and Mark A. Stevenson and Chris Baker and Simon M. Firestone},
journal = {PLOS Computational Biology},
year = {2026},
month = sep,
doi = {10.1371/journal.pcbi.1014716},
url = {https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014716},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A real-time forecasting framework for emerging infectious diseases affecting animal populations
AU - Meryl Theng
AU - Simin Lee
AU - Andrew C. Breed
AU - Sharon Roche
AU - Emily Sellens
AU - Catherine Fraser
AU - Kelly Wood
AU - Chris P. Jewell
AU - Mark A. Stevenson
AU - Chris Baker
AU - Simon M. Firestone
PY - 2026
DA - 2026/09/10
JO - PLOS Computational Biology
DO - 10.1371/journal.pcbi.1014716
UR - https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1014716
AB - Infectious disease forecasting has become increasingly important in public health. However, forecasting tools for emergency animal diseases, particularly those offering real-time decision support when parameters governing disease dynamics are unknown, remain limited. We introduce a generalised modelling framework for near-real-time forecasting of the temporal and spatial spread of infectious livestock diseases using data from the early stages of an outbreak. We applied the framework to the 2007 equine influenza outbreak in Australia, generating forecasts at three timepoints across four regional clusters. Prediction targets included future daily case counts, outbreak size, peak timing and duration, and spatial distributions of future spread. We evaluated how well the forecasts predicted daily cases and the spatial distribution of case counts, using skill scores (a measure of probabilistic forecast accuracy) as a benchmark for future model improvements. Forecast accuracy, certainty, and skill improved after formation of the outbreak’s peak, while early forecasts were more uncertain or prone to overestimation, highlighting the need for caution when interpreting pre-peak predictions, particularly when the impacts of control policies on future transmission are not adequately represented in the model. Spatial forecasts of broad, relative risk patterns were more robust than precise predictions of risk at the individual premises level or exact daily cases counts, supporting geographically targeted response strategies. Overall, this framework supports real-time decision-making in livestock disease outbreaks when applied with appropriate consideration of uncertainty, and establishes a foundation for future refinements and applications to other animal diseases. Author summary: Infectious disease outbreaks in livestock can spread rapidly, making timely and well-informed decisions critical for limiting their impact. However, predicting how an outbreak will evolve is difficult, especially early on when little is known about how the disease spreads. In this study, we developed a flexible modelling framework that uses data collected during the early stages of an outbreak to generate real-time forecasts of how disease may spread over time and space. We applied our approach to the 2007 equine influenza outbreak in Australia and assessed how well it could predict future cases and identify areas at higher risk. We found that forecasts became more accurate as more data became available, and that short-term predictions were generally more reliable than those made further into the future. Importantly, while precise predictions of case numbers or individual affected farms were often uncertain, the model was better at identifying “hotspot” areas of higher risk. These findings suggest the model is most useful for guiding targeted surveillance and control to these higher-risk zones, particularly when its uncertainty and assumptions are carefully considered.
SN - 1553-7358
ER -