费腾, 叶江霞, 王敬文. 基于机器学习的宜良红火蚁入侵风险评估[J]. 西南林业大学学报(自然科学), 2024, 44(6): 1–9. DOI: 10.11929/j.swfu.202402014
引用本文: 费腾, 叶江霞, 王敬文. 基于机器学习的宜良红火蚁入侵风险评估[J]. 西南林业大学学报(自然科学), 2024, 44(6): 1–9. DOI: 10.11929/j.swfu.202402014
Fei Teng, Ye Jiangxia, Wang Jingwen. Risk Assessment of Solenopsis invicta Invasion in Yiliang County based on Machine Learning[J]. Journal of Southwest Forestry University. DOI: 10.11929/j.swfu.202402014
Citation: Fei Teng, Ye Jiangxia, Wang Jingwen. Risk Assessment of Solenopsis invicta Invasion in Yiliang County based on Machine Learning[J]. Journal of Southwest Forestry University. DOI: 10.11929/j.swfu.202402014

本文排版定稿已在中国知网网络首发,如需阅读全文请打开知网首页,并搜索该论文题目即可查看。

基于机器学习的宜良红火蚁入侵风险评估

Risk Assessment of Solenopsis invicta Invasion in Yiliang County based on Machine Learning

  • 摘要: 以滇中宜良县为研究区域,基于一系列可能影响因子对红火蚁的入侵风险进行了评估,旨在探索其空间扩散机制和格局。研究根据实地调查,通过遥感反演和地理信息系统(GIS)分析,对气象、植被、水域、地形地貌、人为活动等各种环境变量进行空间量化与栅格制图,并基于MaxEnt、随机森林和 Logistic三种机器学习方法构建了入侵风险模型,且分别进行精度评估及影响因子分析,最终以空间可视化地图表征红火蚁的入侵风险。结果表明:MaxEnt、随机森林和 Logistic 模型的AUC值分别为0.926、0.944和0.950,Logistic最优,模型的AUC值均大于0.9,表明了模型的可靠性;驱动因子分析显示,影响红火蚁入侵的因素复杂多样,最优Logistic模型的影响因子依次为土壤湿度、食量、海拔和坡向;红火蚁在滇中宜良的入侵风险格局表现出明显的空间异质性;红火蚁高风险区集中在水体周边、荒山荒地和新造林地,而难以侵入高海拔自然植被区;要特别重视区域苗木检疫工作和预防性消毒灭菌工作,加强森林经营管理,促进提高森林质量,构筑天然屏障。研究分析了红火蚁在滇中宜良的潜在入侵风险格局,为开展有效的防控提供了科学依据,同时也为其他入侵物种的风险评估提供了一定参考。

     

    Abstract: Using Yiliang County of Central Yunnan Plateau, China as the study area, the invasion risk of the Solenopsis invicta was assessed based on a series of possible drivers, aiming to explore its spatial dispersal mechanisms and patterns. Firsly, various environmental variables such as meteorology, vegetation, watershed, topography and geomorphology, and anthropogenic activities were quantified and raster mapped based on field work by remote sensing inversion and geographic information system (GIS) analysis. Three maching learning of MaxEnt, Random Forest and Logistic prediction models were constructed, also the accuracy was assessed, and a factor analysis was conducted on driving factors, respectively, so as to visualize the risk patterns of Solenopsis invicta invasion.The results show that the AUC values of MaxEnt, Random Forest and Logistic models are 0.926, 0.944 and 0.95 respectively, and the AUC values of the models are greater than 0.9, which indicates that the models are reliable; Factor analysis shows that the main driving factors of the model are multifaceted, and the dominant independent variables of all models are different. The driving factors of the optimal Logistic model are surface humidity, food, altitude and aspect;Through the prediction results of the model, it can be concluded that the invasion risk model shows strong spatial heterogeneity; The high-risk areas for Solenopsis invicta in the study area are concentrated in the surrounding water bodies, barren hills and wastelands and newly forested land, which are less likely to invade the natrual vegetation areas at high altitude; Special attention should be paid to the quarantine work and preventive disinfection and sterilisation of seedlings, and to strengthening the management of forest operations to promote the improvement of forest quality and the construction of natural barriers. The study complements the invasion pattern and potential risk of the red fire ant plague in central Yunnan, providing a scientific basis for effective prevention and control, as well as a reference meaning for the risk assessment of other invasive species.

     

/

返回文章
返回