以江西省崇义县5种主要造林树种(杉木、马尾松、木荷、苦楝和南酸枣)为研究对象,基于森林经理调查数据,采用随机森林算法构建了5种树种的适生性模型,并对不同立地条件下的造林地进行各树种的适生性预测。模型的输入变量为海拔、坡向、坡度、坡位、土壤类型、成土母岩、土层厚度、腐殖质层厚度,输出变量为树种生长适宜性。结果显示:1)随机森林算法构建的5种造林树种适生性模型的训练精度分别为88.69%,93.13%,95.54%,93.86%和98.92%,泛化精度分别为72.79%,84.18%,77.99%,81.22%和80.56%,具有较高的预测准确率;2)研究区内,对针叶树种适生性影响较大的立地因子均为腐殖质层厚度、海拔、土层厚度,而对阔叶树种影响较大的立地因子则因树种而异。基于随机森林算法构建的树种适生性模型可以较好地对造林树种的适生性进行判断,从而获得各树种的适生环境,可为适地适树和区域森林质量精准提升提供决策依据。

Based on forest management inventory data and five primary forestation species (Cunninghamia lanceolata,Pinus massoniana,Schima superba,Melia azedarach and Choerospondias axillaris),the adaptability models of forestation species were constructed to predict adaptability for afforestation sites in Chongyi County,Jiangxi Province. The input variables contained elevation,slope,aspect,position,soil type,parent rock,soil thickness,and thickness of soil humus,the output variable was growth adaptability. The results showed that the training accuracy of adaptability models for 5 forestation species were 88.69%,93.13%,95.54%,93.86% and 98.92%,respectively. In addition,the generalization accuracy of adaptability models for 5 species were 72.79%,84.18%,77.99%,81.22% and 80.56%,respectively. Site factors greatly affecting the adaptability of coniferous species were thickness of soil humus,elevation and soil thickness,while the driver factors for broad-leaved species depended on species. The species adaptability models based on random forest could analyze the adaptability for forestation species,and extract the comfortable growth environment. Thus,the established models could provide support to the problem of matching species to sites and the improvement of regional forest quality.