Combining Tree- and Stand-Level Models: A New Approach to Growth Prediction
Combining Tree- and Stand-Level Models: A New Approach to Growth Prediction
复制标题
DOI：
10.1093/forestscience/54.5.553
复制
发表时间：
2008-10
期刊：
影响因子：
1.4
通讯作者：
Chaofang Yue;U. Kohnle;Sebastien Hein
中科院分区：
文献类型：
--
作者：
Chaofang Yue;U. Kohnle;Sebastien Hein
组合不同模型生成的信息的方法被认为可以提高预测质量。然而，这种方法尚未开发用于森林生长预测。因此，进行了一项研究来调查组合模型预测森林生长的潜力。通过结合从同一数据源参数化的树级和林分级生长模型生成的生长信息，为挪威云杉（Picea abies [L.] Karst.）开发了生长预测系统。树级模型预测累积直径分布的相对树直径增长率。林分水平模型预测林分断面积的绝对增长。组合预测系统由三个基本步骤组成：初步预测、组合和反馈修正。每年根据两种类型的生长模型生成林分断面积生长的初步预测。然后基于方差和协方差方法组合初步预测，并且组合的估计器用于在反馈过程中更新增长模型。在树级生长模型中，更新的林分断面积随后被分解为单棵树。在验证数据集的基础上，与单独使用树级模型的预测相比，使用组合预测系统进行的林分断面积预测的效率提高了 14％ 至 43％，与使用林分模型的预测相比，效率提高了 4％ 至 16％，具体取决于预测的长度（5-30 年）。
Approaches to combining information generated from different models are recognized for improving forecast quality. However, such an approach has not been developed for forest growth predictions. Therefore, a study was carried out to investigate the potential of combining models to forecast forest growth. A growth prediction system was developed for Norway spruce (Picea abies [L.] Karst.) by combining growth information generated from tree- and stand-level growth models parameterized from an identical data source. The tree-level model predicts relative tree diameter growth rates of a cumulative diameter distribution. The stand-level models predict absolute stand basal area growth. The combined prediction system consists of three basic steps: preliminary prediction, combination, and feedback modification. Preliminary predictions of stand basal area growth are generated annually from both types of growth models. The preliminary predictions are then combined on the basis of a variance and covariance method, and the combined estimator is used to update the growth models in a feedback procedure. In the tree-level growth model, the updated stand basal area is subsequently disaggregated to individual trees. On the basis of a validation data set, forecasts of stand basal area with the combined prediction system were shown to gain in efficiency from 14 to 43% compared with forecasts with the tree-level model alone and from 4 to 16% compared with projections with the stand-level model, depending on the length of the projection (5–30 years).