摘要
【目的】杨树是我国栽培数量最多的阔叶树种,其生长快、易繁殖、适应性强、轮伐期短等特点对解决木材供需平衡、促进碳汇、实现碳循环等方面至关重要,探究杨树生长环境影响机制对实现森林资源高效管理、推动生态文明建设具有重要意义。【方法】本研究利用全国森林资源连续清查部分固定样地杨树实测数据,构建未考虑环境因素和考虑环境因素的杨树胸径生长率多元回归模型,结合随机森林(RF)、梯度提升机(GBM)和支持向量机(SVM)算法,对RF、GBM和SVM算法以RMSE最小完成模型最优参数确定,并通过平均绝对误差(MAE)、均方根误差(RMSE)和决定系数(R^(2))进行模型评价,实现环境因子对杨树生长的重要性程度知识挖掘。【结果】杨树胸径生长率主要受其自身胸径大小的影响,且随着胸径的增大而减小,呈现反“J”型趋势;考虑环境因素的回归模型较未考虑环境因素的回归模型R^(2)从0.066提高到了0.403;机器学习算法预测效果明显优于回归模型算法,其中以RF算法精度最高,R^(2)达0.730,预测结果和实际值基本一致;多元回归模型、RF和GBM对模型重要性解释程度规律基本一致,SVM存在微小差异。【结论】回归模型精度虽略低于机器学习算法,但其“白箱”优势可为未来森林资源调查工作中判定其胸径是否存在异常提供依据;杨树生长受环境影响,与地理空间位置关系紧密,温度适宜、降水充沛的低海拔地区以及坡度平缓、坡位较低的北坡区域更适宜杨树生长,密度越大越不利于其生长;在杨树林的营造过程中,应首先考虑造林地理位置、气象气候等因素;其次,考虑林分结构,特别是林分密度合理性;最后,考虑地形结构是否适宜进行杨树林营造工程建设。
[Objective] Forest is known as the “lung of the earth”, which is the material and spiritual basis for promoting ecosystem circulation and human survival. Populus spp. is the broadleaved tree species with the largest cultivated quantity in China. Its characteristics of fast growth, easy reproduction, strong adaptability, and short rotation period are important for keeping the balance of wood supply and demand,promoting carbon sequestration, and realizing carbon cycle. Exploring the environmental impact mechanism on its growth is of great significance to realize the efficient management of forest resources and promote the construction of ecological civilization. [Method] In this study, a multiple regression model of Populus spp.DBH growth rate without considering or having considered environmental factors was established based on the continuous inventory data of Populus spp. in forest resources in China. Random forest(RF), gradient boosting machine algorithm(GBM) and support vector machine(SVM) were established. The optimal parameters of these algorithms were determined by the minimum RMSE. These models were evaluated by the mean absolute error(MAE), root mean square error(RMSE) and coefficient of determination(R^(2)). The level of importance of meteorological climate, terrain, stand structure and other environmental factors on the growth of Populus spp. was explored. [Result] The DBH growth rate of Populus spp. was mainly affected by its own DBH, decreasing with the increase of DBH, and showing an obvious inverse “J” shaped trend.The model considering environmental factors had higher accuracy than that without considering. Especially,for the growth rate model of Populus spp., the R^(2) was increased from 0.066 to 0.403. The prediction effect of the machine learning algorithm was obviously better than the regression model algorithm. Among them, the accuracy of RF algorithm was the highest, with R^(2) reaching 0.730. The prediction result was basically the same as the actual value. The multivariate regression model, RF and GBM were basically consistent in explaining the importance of the model, while SVM had slight differences. [Conclusion] Although the accuracy of the regression model is slightly lower than that of the machine learning algorithm, it is a white box, which can provide a basis for determining whether the DBH is abnormal in the future inventory work.The growth of Populus spp. is affected by the growth environment and closely related to the geographical location. The low altitude areas with suitable temperature and abundant precipitation and the north slope areas with gentle slope and low slope position will be more suitable for its growth. The higher the density is,the less the conducive to its growth is. In the process of Populus spp. planting, the factors of afforestation location, weather and climate should be considered first. Secondly, the rationality of stand structure,especially stand density should be considered. Finally, we should consider whether the topographic structure is suitable for afforestation projects.
作者
张瀚月
冯仲科
黄国胜
杨雪清
冯泽民
Zhang Hanyue;Feng Zhongke;Huang Guosheng;Yang Xueqing;Feng Zemin(Precision Forestry Key Laboratory of Beijing,Beijing Forestry University,Beijing 100083,China;Key Laboratory of Genetics and Germplasm Innovation of Tropical Special Forest Trees and Ornamental Plants of Ministry of Education,Hainan University,Haikou 570228,Hainan,China;National Forestry and Grassland Bureau Survey Planning and Design Institute,Beijing 100714,China;Key Laboratory of Earth System Modeling of Ministry of Education,Tsinghua University,Beijing 100084,China)
出处
《北京林业大学学报》
CAS
CSCD
北大核心
2022年第11期50-59,共10页
Journal of Beijing Forestry University
基金
海南省重点研发计划项目(ZDYF2021SHFZ256)
中央高校基本科研业务费专项(2015ZCQ-LX-01)。
关键词
杨树生长
回归模型
机器学习
胸径生长率
环境影响机制
growth of Populus spp.
regression model
machine learning
DBH growth rate
environmental impact mechanism