东北林业大学学报
東北林業大學學報
동북임업대학학보
JOURNAL OF NORTHEAST FORESTRY UNIVERSITY
2015年
5期
33-37
,共5页
肖锐%陈东升%李凤日%孙楠
肖銳%陳東升%李鳳日%孫楠
초예%진동승%리봉일%손남
两水平混合模型%杂种落叶松%胸径生长%树高生长
兩水平混閤模型%雜種落葉鬆%胸徑生長%樹高生長
량수평혼합모형%잡충락협송%흉경생장%수고생장
Two-level mixed effect model%Hybrid Larch%DBH growth%Height growth
以江山娇实验林场的16块杂种落叶松密度对比标准地为研究对象,使用非线性最小二乘法,对4种常用胸径和树高生长模型拟合优度进行比较,最终选择Schumacher 方程作为混合模型的基础模型。考虑林分效应和林木效应,采用两水平嵌套混合模型的方法,通过R软件进行不同参数混合作用的模拟,选择收敛的AIC、BIC和对数似然值最小的混合模型作为最优模型,最后将其与传统模型进行拟合优度的比较。结果表明:胸径生长模型中将林分效应作用于参数a1、林木效应作用于参数a0和a1,树高生长方程中将林分效应和林木效应作用于参数a0时的混合模型最优,可将胸径和树高生长传统模型的确定系数从0.7510和0.9008提高到0.9463和0.9474,均方根误差也大幅缩小,并且较好地消除了异方差现象,说明通过构造方差协方差矩阵可以校正随机参数以反映树木之间生长的差异。
以江山嬌實驗林場的16塊雜種落葉鬆密度對比標準地為研究對象,使用非線性最小二乘法,對4種常用胸徑和樹高生長模型擬閤優度進行比較,最終選擇Schumacher 方程作為混閤模型的基礎模型。攷慮林分效應和林木效應,採用兩水平嵌套混閤模型的方法,通過R軟件進行不同參數混閤作用的模擬,選擇收斂的AIC、BIC和對數似然值最小的混閤模型作為最優模型,最後將其與傳統模型進行擬閤優度的比較。結果錶明:胸徑生長模型中將林分效應作用于參數a1、林木效應作用于參數a0和a1,樹高生長方程中將林分效應和林木效應作用于參數a0時的混閤模型最優,可將胸徑和樹高生長傳統模型的確定繫數從0.7510和0.9008提高到0.9463和0.9474,均方根誤差也大幅縮小,併且較好地消除瞭異方差現象,說明通過構造方差協方差矩陣可以校正隨機參數以反映樹木之間生長的差異。
이강산교실험림장적16괴잡충락협송밀도대비표준지위연구대상,사용비선성최소이승법,대4충상용흉경화수고생장모형의합우도진행비교,최종선택Schumacher 방정작위혼합모형적기출모형。고필림분효응화림목효응,채용량수평감투혼합모형적방법,통과R연건진행불동삼수혼합작용적모의,선택수렴적AIC、BIC화대수사연치최소적혼합모형작위최우모형,최후장기여전통모형진행의합우도적비교。결과표명:흉경생장모형중장림분효응작용우삼수a1、림목효응작용우삼수a0화a1,수고생장방정중장림분효응화림목효응작용우삼수a0시적혼합모형최우,가장흉경화수고생장전통모형적학정계수종0.7510화0.9008제고도0.9463화0.9474,균방근오차야대폭축소,병차교호지소제료이방차현상,설명통과구조방차협방차구진가이교정수궤삼수이반영수목지간생장적차이。
Four DBH and height growth equations were evaluated using nonlinear least square method to develop the model with the highest precision for Hybrid Larch stands of different density in Jiangshanjiao Experimental Forest Farm of Heilongjiang Academy of Forestry , and the Schumacher model was chosen for a higher precision .Considering the stand effect and the trees effect at the same time , the mixing effects with different parameters were simulated based on two-level mixed effects model using R software , and the best mixed model was developed for the smallest AIC , BIC and loglikelihood .When a1 with the stand effects and tree effects , a0 with the tree effects were considered , the DBH growth model was best .When a0 with the stand effects and tree effects were considered , the height growth model was best .The coefficients of determination could be reached 0.946 3 and 0.947 4 from 0.751 0 and 0.900 8, and RMSE were also significantly decreased .The mixed model showed more homogeneous residual variance than the traditional models .Therefore, the mixed model not only showed the mean trends of DBH and height , but also showed the individual difference based on variance-covariance structure .