| 基于贝叶斯累加回归树评估中国非洲猪瘟发生风险区域 |
| |
| 引用本文： | 王颢然,肖建华,欧阳茂霖,高宏岩,刘宇馨,高利,高翔,王洪斌.基于贝叶斯累加回归树评估中国非洲猪瘟发生风险区域[J].农业工程学报,2022,38(11):180-187. |
| |
| 作者姓名： | 王颢然 肖建华 欧阳茂霖 高宏岩 刘宇馨 高利 高翔 王洪斌 |
| |
| 作者单位： | 东北农业大学动物医学学院黑龙江省动物普通疾病防治重点实验室，哈尔滨 155030 |
| |
| 基金项目： | 国家自然科学基金项目（31802217） |
| |
| 摘 要： | 为探究中国非洲猪瘟发生的风险区域，对疫情的风险评估提供决策参考。该研究基于贝叶斯累加回归树（Bayesian Additive Regression Tree，BART）模型，应用2018年8月至2021年8月非洲猪瘟发生数据评估了中国非洲猪瘟的风险区域及相关的影响因素。结果表明：1）在影响非洲猪瘟风险区分布的环境因素中，归一化城市土地指数（0.213±0.026）贡献性最高，其次是归一化差分植被指数（0.207±0.028），年平均气温（0.199±0.025），家猪的分布（0.194±0.025）和最冷季降水量（0.187±0.026）。2）在评估模型中，小型城市相比于其他类型的城市发生非洲猪瘟的风险更高。非洲猪瘟的发生风险随着年平均气温、归一化植被指数、家猪的数量的升高而升高。最冷季降水量的升高会降低非洲猪瘟发生的可能性。3) 非洲猪瘟的风险区域主要集中在中国的东部和西南。4）在预测地图中，中国东南地区的不确定性较高，鉴于影响非洲猪瘟传播和发生的因素众多，未来需要对此区域保持重点关注。在模型的预测精准性评估中，模型曲线下面积（Area Under Curve，AUC）为0.90，证明风险地图的预测准确性较高。研究结果可为BART在动物传染病风险评估中的应用提供参考，同时为了解中国非洲猪瘟发生主要风险区域和影响因素，采取合理的预防和控制措施提供信息和建议。
|
| 关 键 词： | 病害 模型 非洲猪瘟 风险评估 贝叶斯累加回归树 影响因素 |
| 收稿时间： | 2021/9/18 0:00:00 |
| 修稿时间： | 2022/5/21 0:00:00 |
Assessing the risk areas for African swine fever in China using Bayesian additive regression trees |
| |
| Wang Haoran,Xiao Jianhu,Ouyang Maolin,Gao Hongyan,Liu Yuxin,Gao Li,Gao Xiang,Wang Hongbin.Assessing the risk areas for African swine fever in China using Bayesian additive regression trees[J].Transactions of the Chinese Society of Agricultural Engineering,2022,38(11):180-187. |
| |
| Authors: | Wang Haoran Xiao Jianhu Ouyang Maolin Gao Hongyan Liu Yuxin Gao Li Gao Xiang Wang Hongbin |
| |
| Institution: | Key Laboratory of the Provincial Education Department of Heilongjiang for Common Animal Disease Prevention and Treatment, College of Veterinary Medicine, Northeast Agricultural University, Harbin 155030, China |
| |
| Abstract: | African Swine Fever (ASF) has been one type of acute, febrile, and highly contagious infectious disease during pig breeding. This infectious disease can be characterized by respiratory impairment and severe bleeding of the skin and internal organs. Four levels can be classified for the severity of the ASF disease, including the most acute, acute, subacute, and chronic. Once the infection by the strong strains can cause up to 100% morbidity and mortality in pigs. As such, the ASF has been mandated as the reporting animal disease by the World Organization for Animal Health (OIE), where China has been classified as a risk area for Class I animal disease. Given that there is no effective vaccine or treatment available so far, the prevention and control of ASF rely mainly on some protective measures, such as strict monitoring and health management. Therefore, it is a high demand for an effective risk assessment model to deal with the spread of ASF. This study aims to determine the risk areas of ASF in China, particularly for the decision-making on the risk assessment of the epidemic. A Bayesian Additive Regression Tree (BART) model was also constructed to assess the risk areas and influencing factors using the ASF outbreak data and 33 environmental factors from August 2018 to August 2021. Some buffers were set at every 5 km for the collected points of the ASF outbreak, in order to eliminate the possible bias from the clustering during modeling. One distribution point of two intersecting buffers was removed using overlay analysis, where there was only one ASF outbreak point per 5 km × 5 km grid. A Variance Inflation Factor (VIF) analysis was performed to initially eliminate the highly correlated variables. The high correlation among environmental factors was avoided for the high accuracy of prediction. Specifically, the factors with the VIF values greater than 10 were considered as the multi-collinearity for the removal in the VIF analysis. Subsequently, the environmental variables were verified, according to the importance level of the factors in the BART model. Finally, five optimal environmental variables were identified to establish the BART model for the assessment of the ASF risk areas, including the distribution of normalized urban land index, the normalized differential vegetation index, the mean annual temperature, the distribution of domestic pigs, and the coldest season precipitation. The results showed that: 1) There was the highest contribution of the normalized urban land index distribution (0.213±0.026), followed by the normalized differential vegetation index (0.207±0.028), the mean annual temperature (0.199±0.025), the distribution of domestic pigs (0.194±0.025), and the coldest season precipitation (0.187±0.026). 2) There was a relatively higher ASF risk in the small cities, compared with the other types of cities. The ASF risk also increased with the average annual temperature, the normalized difference vegetation index, and the number of pigs. The high precipitation during the coldest season reduced the likelihood of ASF. 3) The risk areas of ASF were concentrated in eastern and southwestern China, including the parts of three northeastern provinces (Heilongjiang, Liaoning, and Jilin), Hebei, Shandong, Henan, Shanxi, Shaanxi, Anhui, Jiangsu, Hubei, Hunan, Guangxi, Guangdong, Hainan, Guizhou, Sichuan, Yunnan, and Chongqing. 4) In the risk map, there was a higher uncertainty in southeastern China. Therefore, much attention can be paid to this area in the future, due to various influencing factors for the spread and occurrence of ASF. Additionally, the Area Under the Curve (AUC) was 0.90 in the accuracy assessment of the model, indicating the high predictive accuracy of the risk map. The findings can provide a strong reference to determine the risk areas and influencing factors for the occurrence of ASF in China, thus formulating the prevention and control plans. |
| |
| Keywords: | disease model african swine fever risk assessment bayesian additive regression trees influencing factors |
|
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
| 点击此处可从《农业工程学报》下载免费的PDF全文 |
|
