臺灣博碩士論文加值系統

本研究利用2015年1月至2020年6月eBird資料庫與2015年至2017年台灣禽場資料，探討台灣禽流感爆發禽場與鳥種的關聯性。各別對於留鳥以及候鳥篩選前20% 觀察數量總和的鳥種，以及過往被驗出感染禽流感之鳥種當作分析重點，共36種留鳥與68種候鳥。運用邏輯斯迴歸模型與彈性網路法(Elastic net)篩選變數估算台灣3km ×3km 網格各鳥種的潛勢機率(傾向分數)圖，考慮有無納入環境因子兩種模型。接著使用肯德爾等級相關係數(Kendall rank correlation coefficient)、邏輯斯迴歸、零膨脹卜瓦松迴歸(Zero-inflated Poisson regression model，ZIP)分析各鳥種與爆發禽場的關聯性。
接著，我們利用傾向分數配對(propensity score matching，PSM)與McNemar檢定，探討各鳥種之出現與禽場爆發間之關聯性。在配對時，針對爆發禽場網格之傾向分數設立範圍，運用自助重抽樣法(bootstrapping)隨機抽樣1000次以探討各鳥種與禽場爆發之相關性。針對候鳥未納入環境因子的傾向分數分析，顯著次數100次以上且皆為正相關的鳥種有：高蹺鴴、太平洋金斑鴴、小鸊鷉、田鷸、白翅黑燕鷗、長趾濱鷸、家燕；候鳥納入環境因子的傾向分數分析，顯著次數100次以上且皆為正相關的鳥種有：高蹺鴴、紅嘴鷗、太平洋金斑鴴、尖尾鴨、小燕鷗、小鸊鷉、田鷸、灰鶺鴒、東方鵟、長趾濱鷸。針對留鳥未納入環境因子的傾向分數分析，顯著次數100次以上且皆為正相關的鳥種有：紅鳩、黃頭鷺、紅嘴黑鵯、紅冠水雞；留鳥未納入環境因子的傾向分數分析，顯著次數100次以上且皆為正相關的鳥種有：紅嘴黑鵯、家八哥、繡眼畫眉。最後我們定義具有相加性的機率模型，建立各網格內之禽場受各種鳥種感染而爆發禽流感之機率地圖，此地圖可以作為後續風險分析管理之用。

This study uses the eBird database from January 2015 to June 2020 and the data from Taiwan's poultry farms from 2015 to 2017 to explore the relationship between bird flu outbreaks and bird species in Taiwan. The analysis focuses on the top 20% of the total number of resident and migratory birds, as well as the bird species that have been found to be infected with avian influenza in the past. There are 36 resident and 68 migratory bird species. Considering whether there are two models of environmental factors, the logistic regression model and the elastic net method (Elastic net) are used to filter variables to estimate the potential probability (propensity score) map of each bird species on a 3km × 3km grid in Taiwan. Then, Kendall rank correlation coefficient, logistic regression, and zero-inflated Poisson regression model (ZIP) were used to analyze the correlation between bird species and outbreak poultry farms.
Next, we use propensity score matching (PSM) and McNemar test to explore the relationship between the emergence of various bird species and outbreaks in poultry farms. When pairing, a range was established for the propensity scores of the outbreak poultry farm grid, and the bootstrapping method was used to randomly sample 1,000 times to explore the correlation between bird species and poultry farm outbreaks. According to the analysis of propensity scores for waterbirds that are not included in environmental factors, the bird species that have been significantly more than 100 times and are all positively correlated are : Himantopus himantopus, Pluvialis fulva , Tachybaptus ruficollis, Gallinago gallinago, Chlidonias leucopterus, Calidris subminuta, Hirundo rustica; waterbirds are included in the environmental factor tendency score analysis, the significant number of more than 100 times and are all positively correlated bird species: Himantopus himantopus, Chroicocephalus ridibundus, Pluvialis fulva, Anas acuta, Sternula albifrons, Tachybaptus ruficollis, Gallinago gallinago, Motacilla cinereal, Buteo japonicus, Calidris subminuta.
According to the analysis of the propensity scores of resident birds that are not included in environmental factors, bird species that have been significant more than 100 times and are all positively correlated are: Streptopelia tranquebarica, Bubulcus ibis, Hypsipetes leucocephalus, Gallinula chloropus ; resident birds are not included in the propensity score of environmental factors According to the analysis, the bird species that have a significant number of more than 100 times and are all positively correlated are: Hypsipetes leucocephalus, Acridotheres tristis, Alcippe morrisonia. Finally, we define an additive probability model and establish a probability map of bird flu outbreaks in poultry farms in each grid by infection of various bird species. This map can be used for subsequent risk analysis and management.

摘要 i
Abstract ii
目次 iv
表目次 vi
圖目次 vii
第一章 緒論 1
第一節 研究背景 1
第二節 研究動機與目的 2
第三節 文獻回顧 3
第四節 研究架構 4
第二章 資料處理 6
第一節 資料描述 6
第二節 網格化資料呈現 8
第三節 傾向分數與原始數量分析 9
第三章 研究方法 11
第一節 以網格法估計傾向分數 11
第二節 彈性網路法(Elastic Net) 12
第三節 肯德爾等級相關係數(Kendall’s tau) 13
第四節 多變數邏輯斯迴歸 13
第五節 零膨脹卜瓦松迴歸 14
第六節 Matched-pair McNemar檢定 15
第七節 自助重抽樣法(bootstrap resampling method) 16
第八節 禽場爆發風險地圖 16
第四章 分析結果 18
第一節 以傾向分數作為鳥種分布之潛勢估計 18
第二節 傾向分數迴歸分析 : 未納入環境因子 20
第三節 McNemar配對分析與bootstrapping : 未納入環境因子 32
第四節 傾向分數迴歸分析 : 納入環境因子 38
第五節 McNemar配對分析與bootstrapping : 納入環境因子 48
第六節 禽場爆發風險地圖 52
第五章 結論與建議 56
第一節 結論 56
第二節 建議 57
參考文獻 58
一、中文部分 58
二、西文部分 58
附錄 60

一、中文部分
1陳宛均, 羅祈鈞, 蔡富安 & 張安瑜. 運用開放資料建置臺灣陸域環境因子 多時序資料集. 台灣生物多樣性研究 22, 13-44 (2020).
2 曾智一(2019)。〈台灣禽流感爆發與公民科學家鳥類觀測資料之關聯性分
析〉。
3 許家欣(2020)。〈以eBird資料庫估計鳥種數量分布並探討其與2015年台
灣禽流感爆發禽場之相關性〉。
4 曾子容、施浩榆、詹大千、李昌駿、萬灼華、顧家綺、顏慕庸、金傳春 (2016)。〈禽流感在人類的重大流行史觀與公共衛生〉。疫情報導 32(5)
5 劉正夫(2021)。〈以條件自相關迴歸模型估計全台禽流感風險鳥種之潛勢
分布圖〉。
二、西文部分
1Lupiani, B. & Reddy, S. M. The history of avian influenza. Comp Immunol Microbiol Infect Dis 32, 311-323, doi:10.1016/j.cimid.2008.01.004 (2009).
2Kung, N. Y. et al. Risk for infection with highly pathogenic influenza A virus (H5N1) in chickens, Hong Kong, 2002. Emerg Infect Dis 13, 412-418, doi:10.3201/eid1303.060365 (2007).
3Shinya, K., Makino, A. & Kawaoka, Y. Emerging and Reemerging Influenza Virus Infections. Veterinary Pathology 47, 53-57, doi:10.1177/0300985809354464 (2010).
4Bui, C. M., Gardner, L. & MacIntyre, C. R. Highly Pathogenic Avian Influenza Virus, Midwestern United States. Emerg. Infect. Dis 22, 138-139, doi:10.3201/eid2201.151053 (2016).
5Caliendo, V. et al. Enterotropism of highly pathogenic avian influenza virus H5N8 from the 2016/2017 epidemic in some wild bird species. Veterinary Research 51, 117, doi:10.1186/s13567-020-00841-6 (2020).
6Iglesias, I., Sánchez-Vizcaíno, J. M., Muñoz, M. J., Martínez, M. & de la Torre, A. Spatio-temporal model of avian influenza spread risk. Procedia Environmental Sciences 7, 104-109, doi:https://doi.org/10.1016/j.proenv.2011.07.019 (2011).
7An, M., Vitale, J., Han, K., Ng'ombe, J. N. & Ji, I. Effects of Spatial Characteristics on the Spread of the Highly Pathogenic Avian Influenza (HPAI) in Korea. Int J Environ Res Public Health 18, doi:10.3390/ijerph18084081 (2021).

9Kim, D. H., Pieper, C. F., Ahmed, A. & Colón-Emeric, C. S. Use and Interpretation of Propensity Scores in Aging Research: A Guide for Clinical Researchers. J Am Geriatr Soc 64, 2065-2073, doi:10.1111/jgs.14253 (2016).
10Efron, B. Bootstrap methods: Another look at the jackknife, 1979. The Annals of Statistics 7, 126.