On studying the patterns of individual-based tree mortality in natural forests: A modelling analysis
On studying the patterns of individual-based tree mortality in natural forests: A modelling analysis
复制标题
DOI：
10.1016/j.foreco.2020.118369
复制
发表时间：
2020-11
影响因子：
3.7
通讯作者：
Christian Salas‐Eljatib;A. Weiskittel
中科院分区：
文献类型：
--
作者：
Christian Salas‐Eljatib;A. Weiskittel
树木死亡是影响森林生态系统动态、结构和组成的重要生态现象，其影响由于与森林条件和环境变化的关系而具有全球相关性。基于个体的死亡率数据存在一些挑战，特别是测量间隔不均匀的观察结果。在这里，我们开发并研究了几种常见的基于个体的死亡率建模策略，这些策略同时考虑了智利中南部混合nothofagusforests长期永久样地数据的不平等测量长度和数据的分层结构。这些策略取决于：(a)功能模型形式（logit和Gompit）， (b)周期长度调整方法（年化，协变量和暴露），以及(c)使用的数据结构（传统或所有时间序列的多种组合）。我们的研究结果表明，Gompit函数形式优于常用的logit链接函数。此外，在广义线性混合效应模型中考虑周期长度作为暴露比其他检验的周期长度调整提供更好的拟合优度。使用所有可能的动态数据组合并没有提高模型变量的预测能力，但在拟合模型的统计推断中发现了重要的差异。我们的研究结果强调，理解树木死亡率在很大程度上依赖于使用合适的建模策略，该策略既能捕获变化源，又能将其分配给相应的变量，在本分析中，最好使用多层次、二元和Gompit-exposure建模框架来完成。
Tree mortality is a critical ecological phenomenon shaping forest ecosystem dynamics, structure, and composition, while its effects are of global relevance due to its relationship with forest conditions and environmental changes. There are several challenges associated with individual-based mortality data, particularly observations with uneven measurement intervals. Here, we develop and examine several common individual-based mortality modelling strategies that simultaneously account for unequal measurement lengths and the hierarchical structure of the data in long-term, permanent plot data from the mixedNothofagusforests in south-central Chile. These strategies depend on: (a) the functional model form (logit and Gompit), (b) the period length adjustment method (annualized, covariate, and exposure), and (c) the data structure used (traditional or all multiple combinations of the time series). Our findings indicated that the Gompit functional form outperformed the commonly used logit link function. Furthermore, considering the period length as exposure in a generalized linear mixed-effects model offered better goodness-of-fit than the other examined period length adjustments. Using all the possible combinations of the dynamic data did not improve the prediction capabilities of the model variants, but important differences were found in the statistical inferences of the fitted models. Our results highlighted that understanding tree mortality strongly relies on using a suitable modelling strategy that is capable of both capturing and assigning the sources of variation to the corresponding variables, which was best accomplished using a multi-level, binary, and Gompit-exposure modelling framework in this analysis.