基于机器学习算法的广西桉树适宜性研究  

Eucalyptus suitability in Guangxi based on machine learning algorithms

作　　者：杜雨菲 吴保国[1] 陈玉玲 DU Yufei;WU Baoguo;CHEN Yuling(School of Information Science and Technology,Beijing Forestry University,Beijing 100083,China)

机构地区：[1]北京林业大学信息学院

出　　处：《浙江农林大学学报》2020年第1期122-128,共7页Journal of Zhejiang A&F University

基　　金：“十三五”国家重点研发计划项目(2017YFD0600906)

摘　　要：[目的] 探索立地因子与桉树Eucalyptus适宜性之间的关系,开展树种适宜性研究,为桉树适宜性研究提供新思路,为科学造林提供支持。 [方法] 以广西桉树人工林为研究对象,选取广西国有高峰林场的1 883个森林资源小班调查数据,分别运用朴素贝叶斯、支持向量机、随机森林算法作为树种适宜性评价方法,构建桉树适宜性分类模型。输入为地貌类型、海拔、坡向、坡位、坡度、凋落物厚度、腐殖质层厚度、土层厚度、石砾含量、成土母质,土壤类型等11个立地因子信息,输出为桉树适宜性。 [结果] 3种算法构建的模型拟合精度依次为63.18%、69.73%、78.03%,泛化精度依次为64.33%、67.93%、78.18%。相比于朴素贝叶斯、支持向量机算法,随机森林算法分类效果更好。立地因子重要性排序由高到低依次为:海拔、土层厚度、坡向、坡度、石砾含量、凋落物厚度、坡位、腐殖质层厚度、土壤类型、地貌类型、成土母质。200~350 m海拔、80~100 cm土层厚度的地区比较适宜桉树生长。 [结论] 基于机器学习算法构建的桉树适宜性评价模型可以较好地对桉树的适宜性做出预测。Objective The aim is to provide a new idea for tree species suitability evaluation, provide a support for scientific afforestation, and explore the relationship between site factors and tree suitability. Method Take a Eucalyptus plantation in Guangxi as the research object, 1 883 forest resource sub-compartment survey data of Guangxi state-owned Gaofeng Forest Farm were selected. Then, Naive Bayesian, Support Vector Machine, and Random Forest algorithm were used to evaluate the suitability of tree species and to construct a suitability classification model for Eucalyptus. Eleven site factors, namely, landform type, elevation, aspect, slope position, slope, litter thickness, humus layer thickness, soil layer thickness, gravel content, parent material, and soil type were input with the output being Eucalyptus suitability. Result The fitting accuracy of the three models was 63.18% for Naive Bayesian, 69.73% for Support Vector Machine, and 78.03% for Random Forest algorithm with a generalization accuracy of 64.33% for Naive Bayesian, 67.93% for Support Vector Machine, and 78.18% for Random Forest algorithm. The order of importance for site factors was elevation > soil layer thickness > aspect > slope > gravel content > litter thickness > slope position > humus layer thickness > soil type > landform type > parent material. Overall, Eucalyptus was more suitable for growth in areas of 200-350 m altitude and 80-100 cm soil layer thickness. Conclusion Thus, machine learning classification algorithms could be used to fit the non-linear relationship between tree species suitability and site factors.

关 键 词：森林测计学 适宜性 机器学习 朴素贝叶斯 支持向量机 随机森林 桉树 

分 类 号：S758.8[农业科学—森林经理学]