摘要
【目的】胸径生长率模型是研究林分生长变化、森林生长收获预估以及生物量和碳储量动态变化等的基础支撑,对于森林资源管理具有重要意义。探索利用多层感知机神经网络技术建立上海市单木胸径生长率模型,为上海市森林资源年度监测数据更新提供技术支撑。【方法】利用第六次至第九次全国森林资源连续清查上海市1999、2004、2009、2014年4期固定样地调查数据,对复位样木按照两倍标准差法进行胸径生长异常值剔除,再按照树种和前期胸径分组进行数据合并,分组计算样木胸径生长量的算术平均值,然后按复利式计算出相应的生长率,进而分别建立水杉、樟树、女贞、木兰、杨树5个树种的传统非线性回归和人工神经网络多层感知机的单木胸径生长率模型。使用确定系数(R2)和估计值剩余标准差(SEE)进行模型评价,使用确定系数(R2)、估计值剩余标准差(SEE)、平均预估误差(MPE)和平均百分标准误差(MPSE)进行预估评价。【结果】5个树种单木胸径生长率建模时,非线性回归模型的确定系数(R2)达到0.854、0.790、0.691、0.641和0.608,多层感知机模型确定系数(R2)达到0.903、0.863、0.802、0.684和0.650,后者确定系数(R2)较非线性回归模型分别提高0.049、0.073、0.111、0.043和0.042,多层感知机模型的估计值剩余标准差(SEE)较非线性回归模型均有所下降,5个树种分别下降0.13、0.26、0.32、0.12和0.10;预估后期胸径时,非线性回归模型确定系数(R2)达到0.880、0.832、0.526、0.860和0.799,多层感知机模型确定系数(R2)达到0.883、0.839、0.561、0.862和0.803,后者确定系数(R~2)较非线性回归模型分别提高0.003、0.007、0.035、0.002和0.004,2种模型的平均预估误差MPE均在2%以内,平均百分标准误差(MPSE)均在20%以内,多层感知机模型的估计值剩余标准差(SEE)、平均预估误差(MPE)和平均百分标准误差(MPSE)较非线性回归�
[Objective]Tree diameter growth rate model is essential for studying the change of stand growth,the prediction of forest growth and harvest,and the dynamic change of biomass and carbon storage,which is of great significance for forest resource management.This study aims to provide technical basis for implementing annual updating of forest resource monitoring in Shanghai,With the use of multi-layer perceptron neural network technology,individual tree diameter growth rate models of major tree species were developed.[Method]Based on the data of permanent plots measured in 1999,2004,2009,2014 in Shanghai from the 6th to 9th national forest inventories of China,this study removed diameter growth abnormal value from the trees according to the double standard deviation method and grouped the data by the main species and the previous diameter as well as calculated the arithmetic mean value of the diameter growth increment and the corresponding growth rate of sample trees with compound interest formula by species and the previous tree diameter.Individual tree diameter growth rate models of 5 major tree species(Metasequoia glyptostroboides,Cinnamomum camphora,Ligustrum lucidum,Magnolia liliflora and Populus spp.)were developed using artificial neural network multilayer perceptron and nonlinear regression method.The coefficient of determination(R2)and the residual standard deviation of the estimate(SEE)were used for model evaluation,and the coefficient of determination(R~2),the residual standard deviation of the estimate(SEE),the mean predictive error(MPE),and the mean percentage standard error(MPSE)were used for estimation evaluation.[Result]When modeling the diameter growth rate of the five tree species,the determination coefficients(R2)of the nonlinear regression model reached 0.854,0.790,0.691,0.641,0.608,and the determination coefficients(R2)of the multi-layer perceptron neural network model reached 0.903,0.863,0.802,0.684,0.650.Compared with the nonlinear regression model,the determination coefficient(R2)of the multi-
作者
肖舜祯
刘强
徐志扬
刘龙龙
朱海伦
XIAO Shunzhen;LIU Qiang;XU Zhiyang;LIU Longlong;ZHU Hailun(East China Survey and Planning Institute,National Forestry and Grassland Administration,Hangzhou 310019,China)
出处
《江西农业大学学报》
CAS
CSCD
北大核心
2022年第5期1169-1176,共8页
Acta Agriculturae Universitatis Jiangxiensis
基金
上海市绿化和市容管理局2020年科学技术项目(G201209)。
关键词
单木模型
胸径生长率
非线性回归模型
多层感知机模型
tree-level model
growth rate of diameter
nonlinear regression model
multilayer perceptron(MLP)networks
