Integrating big data to establish an early warning for highly pathogenic avian influenza viruses
- Publication Date: 2019-12-23
| Execution Methods | Being on the flyway of migrating birds, Taiwan experienced epidemics caused by four different subtypes of highly pathogenic avian influenza (HPAI) virus clade 2.3.4.4 since 2015. Our projects include three independent groups corroborated to enhance our understanding of how wild birds, including migrating birds, contributed to the poultry farm outbreaks and the spreading of HPAI. The methods used in this study included statistical models for spatial-temporal clustering and machine learning. |
| Performance Evaluation | Study 1 was led by Dr. Chao. Her group identified four risk factors consistently showing strong association with the spatial clustering of poultry farms affected by H5N2 and H5N8 during 2015 and 2017, including high poultry farm density, poultry heterogeneity index, non-registered waterfowl flock density and higher percentage of cropping land coverage. Study 2 led by Dr. Hong-Dar Isaac Wu, who applied scan statistics to explore the temporal and spatial clustering of various outbreak farms. Meanwhile, the current study identified several high-risk migrant birds using the idea of nested case-control epidemiological study, equipped with a generalized estimating equation estimation. As a risk factor, bird abundance map was estimated by a generalized additive model. Further, we studied a second-order spatial clustering to match with the direction of disease transmission. Traditional standard deviational elliptic method was amended to fit a model allowing the exploration of local transmissions. Study 3 led by Dr. Tsung-Jung Liu, who applied machine-learning based methods to quickly identify the locations of the poultry farms from satellite images. Avian influenza is propagated by the migratory birds, and the influence is directly related to the density of poultry farms. However, marking the locations of the poultry farms manually is not time efficient. And the number of the poultry farms is also changing all the time. We need a faster solution to identify the locations of the poultry farms. In this project, we proposed a machine-learning based method to tackle this problem. The source data is provided by the Taiwan Agricultural Research Institute, Council of Agriculture, including 6 satellite images (resolution 0.5m x 0.5m) and their corresponding labelled data (both registered and nonregistered farms) at different time and areas of Taiwan. We modify the existing deep neural networks (U-Net & SegNet) to improve the location identification performance. |
| Conclusion & Suggestion | 1. The only ENABLE project engaged in social responsibility. 2. Identify several environmental factors contributing to the clustering of HPAI outbreak poultry farms. 3. First and the only team in Taiwan conducting HPAI-associated big data analysis. 4. Continue to utilize the dataset from remote sensing data and citizen scientist for animal disease surveillance. Continue to collaborate with UC-Davis, UC-San Diego, NASA, University of Oklahoma to exchange academic findings. |
| Appendix |