以江西省大岗山实验局不同初植密度的杉木林为研究对象,选择修改的 Richards 模型形式,考虑样地效应,采用 SAS 软件进行非线性混合效应模型的模拟,利用 AIC 和 BIC 值评价模型模拟效果.在此基础上考虑优势木平均高连续观测数据的时间序列相关性,并把初植密度以哑变量形式考虑进去,再进行混合模型的模拟.最后,利用验证数据对混合模型方法与传统的非线性回归模拟方法进行精度比较.研究结果表明,修改的 Richards 形式的优势木平均高与林龄关系的非线性混合效应模型,其估计精度比传统的回归模型估计精度明显提高,增加随机效应参数个数能够提高模型的估计精度.一阶白回归误差结构矩阵模型在解释优势木平均高的时间序列相关性时不仅提高了混合模型的模拟精度,而且能够很好的表达连续观测数据间误差分布情况;同时考虑样地的随机效应、观测数据的时间序列相关性及不同初植密度的混合模型模拟精度比传统的非线性回归方法模拟精度高.%The improvement on the dominant height growth implies in better productivity estimation due to the forest height growth is directly related with the site characteristics and forest productivity. A modified Richards growth model with nonlinear mixed effects is simulated used SAS software for modeling fir plantation dominant height growth in conjunction with different plantation density in Dagangshan Experiment Bureau of Jiangxi Province. Akaike Information Criterion(AIC) and Bayesian lnformation Criterion(BIC) were used in model performance evaluation. Within-plot time series error autocorrelation of dominant height growth data and different plantation density expressed with dummy variable were taken into account in mixed model. Finally, the precision of mixed models was compared with the precision of conventional nonlinear ordinary regression analysis method based on validation data. The result showed that the precision of modified Richards forms nonlinear mixed effect model which takes into account plot' s random effect was better than that of conventional regression model. Increasing the number of random effect parameter can increase the precision of model. First-order autoregressive error model in explaining time series error autocorrelation of dominant height growth not only improved simulated precision, but also can describe error distribution of sequence observation data; The precision of mixed model considering plot random effects, time series error autocorrelation and different plantation density at one time is better than that of ordinary regression analysis method.
展开▼
