The Application of Bayesian Model Averaging in Compatibility of Stand Basal Area for Even-Aged Plantations in Southern China
The Application of Bayesian Model Averaging in Compatibility of Stand Basal Area for Even-Aged Plantations in Southern China
复制标题
贝叶斯平均模型在南方同龄人工林林分断面积兼容性中的应用
DOI：
10.5849/forsci.13-034
复制
发表时间：
2014
期刊：
影响因子：
1.4
通讯作者：
Jianguo Zhang
中科院分区：
文献类型：
--
作者：
Xiongqing Zhang;A. Duan;Leihua Dong;Q. V. Cao;Jianguo Zhang
林分生长与产量模型包括整林模型、单树模型和直径分布模型。以江西省分宜县杉木生长数据为基础，采用贝叶斯平均模型（BMA）对林分断面积进行预测，将三类模型结合为一个预测模型。 BMA 是一种统计方法，通过根据后验概率对各个预测进行加权来推断一致预测，表现较好的预测比较表现较差的预测获得更高的权重。此外，BMA 还考虑了方差所反映的模型不确定性。 BMA的方差可以分解为反映模型一致性的模型间方差和反映数据变异性的模型内方差。结果表明，所有林分断面积预测的模型间方差均远大于模型内方差。由此产生的模型产生了准确可靠的预测，BMA 预测的 95% 置信区间很好地涵盖了观察结果。 BMA方法为三种模型提供了一致的林分断面积预测，从而提高了这些模型之间的兼容性。
Stand growth-and-yield models include whole-stand models, individual-tree models, and diameter distribution models. Based on the growth data of Chinese fir (Cunninghamia lanceolata [Lamb.] Hook.) in Fenyi County, Jiangxi Province, in southern China, Bayesian model averaging (BMA) was used to forecast stand basal areas by combining these three types of models into a single predictive model. BMA is a statistical method that infers consensus predictions by weighting individual predictions based on their posterior probabilities, with the better performing predictions getting higher weights than the poorer performing ones. Furthermore, BMA accounts for model uncertainty as reflected by the variance. The variance of BMA can be decomposed into a between-model variance that reflects the model’s consistency and a within-model variance that reflects the data variability. Results showed that the between-model variance was much greater than the within-model variance for all the stand basal area predictions. The resulting model produced accurate and reliable predictions, and the 95% confidence interval of BMA predictions encompassed the observations very well. The BMA method provided a consistent prediction of stand basal area from three types of models, thus improving compatibility among these models.