论文部分内容阅读
1972和1974年分别在黑龙江省江山娇林场及孟家岗林场设置10块长白落叶松人工林固定样地(8块抚育间伐样地、2块对照样地),采用连年复测数据,分析抚育间伐对人工长白落叶松样地枯死与单木枯死的影响.基于二分类变量Logistic回归,建立了样地枯死及样地内单木枯死概率的两阶段模型(Ⅰ:抚育间伐后样地水平枯死概率模型;Ⅱ:枯死样地中单木水平枯死概率模型),采用广义估计方程(GEE)方法对模型参数进行估计.根据敏感度和特异度曲线相交点确定枯死概率最优临界点.结果表明:样地数据按照抚育间伐次数分为4组分别建模(模型1~模型4).在模型1中,地位指数、林分年龄的自然对数、抚育间伐年龄及强度为显著自变量;模型2~模型4采用主成分分析法建模,主成分包含林分年龄、每公顷株数、平均胸径及抚育间伐因子,说明抚育间伐因子对样地枯死概率有显著影响.抚育间伐对枯死样地中单木枯死概率无显著影响,单木枯死概率模型中显著性自变量为林分初植密度、年龄、林木胸径的倒数及林分中大于对象木的所有林木断面积之和.样地枯死概率模型及单木枯死概率模型Hosmer和Lemeshow拟合优度检验均不显著,模型AUC均在0.91以上,估计正确率均超过80%,说明模型拟合效果较好.
In 1972 and 1974, 10 permanent or permanent Larix gmelini plantations were set up in Jiao Linchang and Mengjigang Forest Farm, Heilongjiang Province, respectively (8 tending nursery sites and 2 control sites) Based on the dichotomous variables Logistic regression, a two-stage model of dead wilt and the probability of single tree dead in the plots was established (Ⅰ: Probability model of horizontal withering after tending of thinning plots; Ⅱ: the probability of single-tree horizontal withering in the dead land), the generalized estimation equation (GEE) method was used to estimate the model parameters, and the critical point of dead probability was determined according to the intersection point of the sensitivity and the specificity curve.The results showed that: The data were divided into four groups according to the number of tending thinning (model 1 to model 4). In model 1, the position index, the natural logarithm of stand age and the age and strength of tending thinning were significant independent variables. Model 2 ~ The principal component analysis included the age of the forest, the number of trees per hectare, the average diameter at breast height and the tending and thinning factors, which indicated that tending thinning factors had a significant effect on the probability of death from the plots. There was no significant effect of cutting on the probability of single tree dead in the dead land. The significant independent variable of single tree dead probability model was the initial planting density, age, reciprocal of DBH and all the forest area larger than the target wood And the probability of death from the plots and the probability of single-tree dead probability model Hosmer and Lemeshow were not significant, the AUC of the model was above 0.91, and the correct rate was over 80%, which showed that the model fitting effect was better.