Abstract Title
Uncovering the contribution of environmental and ecological factors to spatial risk of 2.3.4.4b HPAI in wild birds
Abstract
A wide diversity of wild birds may act as hosts of avian influenza, from waterfowl reservoirs (typical sources of livestock incursions) to atypical wading bird and shorebird hosts, many of which have suffered catastrophic die-offs as a result. Despite the devastating impact the 2.3.4.4b outbreak of HPAI has had, its dynamics within wild bird hosts remain poorly understood.
We use species distribution models to estimate spatial risk of 2.3.4.4b H5N8 and H5N1 in wild birds across Europe, Asia, and the Americas. We identify high-risk areas and explore the relative importance of environmental and ecological traits as potential drivers.
Machine learning is used to model presence of HPAI at a 10km resolution. Models are trained on reported geospatial records of HPAI adjusted for national sampling strategies and use both variables capturing environmental characteristics (e.g., altitude and temperature) and ecology of wild bird populations (e.g. abundance of high-risk taxa, species richness, foraging behaviours). Separate models are used to reflect seasonality within bird population movements and models are validated on data periods excluded from training.
Mapped risk predictions in wild birds with associated uncertainty are generated. Model outputs align with current understanding in highlighting coastal regions and inland waterways as key areas. Both wild bird ecology and environment contribute to predicting spatial patterns of 2.3.4.4b though we find intercontinental differences in explanatory features.
The models presented here offer a framework that can both uncover the determinants of spatial HPAI spread and contribute to surveillance strategies for future HPAI epizootics.
We use species distribution models to estimate spatial risk of 2.3.4.4b H5N8 and H5N1 in wild birds across Europe, Asia, and the Americas. We identify high-risk areas and explore the relative importance of environmental and ecological traits as potential drivers.
Machine learning is used to model presence of HPAI at a 10km resolution. Models are trained on reported geospatial records of HPAI adjusted for national sampling strategies and use both variables capturing environmental characteristics (e.g., altitude and temperature) and ecology of wild bird populations (e.g. abundance of high-risk taxa, species richness, foraging behaviours). Separate models are used to reflect seasonality within bird population movements and models are validated on data periods excluded from training.
Mapped risk predictions in wild birds with associated uncertainty are generated. Model outputs align with current understanding in highlighting coastal regions and inland waterways as key areas. Both wild bird ecology and environment contribute to predicting spatial patterns of 2.3.4.4b though we find intercontinental differences in explanatory features.
The models presented here offer a framework that can both uncover the determinants of spatial HPAI spread and contribute to surveillance strategies for future HPAI epizootics.
Co-Author(s)
Sarah Hayes (University of Oxford),
Joe Hilton (University of Manchester),
Karan Pattni (University of Liverpool),
Joaquin Mould-Quevedo (CSL Seqirus USA),
Christl Donnelly (University of Oxford),
Matthew Baylis (University of Liverpool),
Liam Brierley (University of Glasgow)
Abstract Category
Avian influenza in mammals, pandemic preparedness, and one health