A response to 'Trends in tropical tree growth: reanalysis confirms earlier findings'.
A response to 'Trends in tropical tree growth: reanalysis confirms earlier findings'.
复制标题
对“热带树木生长趋势：重新分析证实了早期发现”的回应。
DOI：
10.1111/gcb.13605
复制
发表时间：
2017
影响因子：
11.6
通讯作者：
Brienen RJ
中科院分区：
文献类型：
--
作者：
Brienen RJ
我们最近证明了Van Der Sleen等人（2015）和Groenendijk等人（2015）的树木年轮的生长趋势受到人口统计学偏差的影响。特别是，聚集性年龄分布导致其增长趋势呈负偏倚。在回应中，他们对我们的分析提出了质疑，并提出了另一种修正方法。我们在这里表明，他们的论点是不正确的，是基于对我们分析的误解，他们的替代方法不起作用。首先，他们认为我们的修正方法导致了虚假的正增长。这是一种误解。是的，在我们对修正方法的测试中，我们发现了正增长趋势（见Brienen等人，2016年的SI图3），但它们并不是虚假的，因为它们是预期的，并且具有正确的幅度。正如Brienen et al.（2016）所解释的那样，我们的方法并没有纠正所有偏差，也没有消除生长缓慢的生存偏差的影响（Brienen et al., 2012）。作者将这种趋势错误地解释为校正过程中的错误，而实际上，这证实了我们的方法是完美的。关于我们的洗牌方法，作者提出的另一点是，它通常会产生微不足道的结果。然而，我们只建立了一个由其物种不规则年龄分布引起的预期生长趋势的零模型，这是一种有效的，公认的方法。其次，作者声称我们不必要地移除了物种。然而，这背后有明确的逻辑。为了确定非均匀年龄偏差对趋势的影响，必须从数据集中删除受其他偏差影响的物种。因此，我们首先删除了作者自己认为存在死亡率偏差的三个物种（Groenendijk et al., 2015）。然后，我们使用两种不同的校正程序对所有剩余的9个物种（包括那些具有非均匀年龄偏差的物种）测试了非均匀年龄偏差的影响。作为最后的测试，我们还删除了年龄分布最不均匀的三个物种，使用Van Der Sleen et al.（2015）的原始方法来估计剩余六个物种随时间的生长趋势，该方法没有纠正任何偏差。最后，作者建议从某些物种中删除最近的生长数据作为一种替代的校正方法。这个程序是有缺陷的。首先，通过删除最近的增长数据，人们无法再测试最近是否增长了！其次，这种方法错误地假设，只有在缺乏新员工的情况下，这种偏见才会出现。然而，这个问题的产生不仅仅是因为缺乏最近的新兵，任何不均匀的年龄分布都可能导致偏差，即使有最近的新兵（见图1）。最后，他们的子集方法并没有消除如图1所示的非均匀年龄偏差的影响，因此对这个问题没有用处。我们的结论是，Van der Sleen等人（2016）提出的观点都是无效的，不幸的是，他们的树木数据仍然无法检测到过去几十年的生长变化。
We recently demonstrated that growth trends from tree rings from Van Der Sleen et al.(2015) and Groenendijk et al.(2015) are affected by demographic biases. In particular, clustered age distributions led to a negative bias in their growth trends. In a response, they challenge our analysis and present an alternative correction approach. We here show that their arguments are incorrect and based on misunderstanding of our analysis and that their alternative approach does not work. Firstly, they argue that our correction methods result in spurious positive growth increases. This is a misinterpretation. Yes, in our test of the correction method, we find positive growth trends (see SI Fig. 3 in Brienen et al., 2016), but they are not spurious as they are expected and of the correct magnitude. Our approach does not correct for all biases and does not remove the effect of slow-grower survivorship bias (Brienen et al., 2012), as explained in Brienen et al.(2016). The authors misinterpreted the trend as a fault in the correction procedure, while in fact, it is confirmation that our methods work perfectly. Another point the authors raise with regard to our shuffling approach is that it would yield often insignificant results. However, we only establish a null model of expected growth trends arising from the irregular age distributions of their species, which is a valid, accepted approach. Secondly, the authors claim we unnecessarily removed species. There is however clear logic behind this. To identify the effect of the nonuniform age bias on trends, those species affected by other biases had to be removed from the dataset. We thus first removed three species which were identified by the authors themselves to be biased by mortality biases (Groenendijk et al., 2015). We then tested the effect of the nonuniform age bias using two different correction procedures for all remaining nine species, including those with nonuniform age biases. As a final test, we also removed the three species with the most nonuniform age distributions to estimate growth trends over time for the remaining six species using the original method of Van Der Sleen et al.(2015) which does not correct for any biases.Finally, the authors propose to remove recent growth data from some species as an alternative correction approach. This procedure is flawed. Firstly, by removing recent growth data, one cannot any longer test whether growth increased recently! Secondly, the approach erroneously assumes that the bias only occurs when there is a lack of recent recruits. However, the problem not only arises because of the lack of recent recruits, and any nonuniform age distribution may result in biases, even if there are recent recruits (see Fig. 1). Finally, their subsetting approach does not remove the effect of the nonuniform age bias as shown in Fig. 1 and thus is of no use for this problem. We conclude that none of the points raised by Van der Sleen et al.(2016) are valid, and their tree data unfortunately still preclude detection of growth changes over the last decades.