Analyzing the shape of observed trait distributions enables a data-based moment closure of aggregate models
Analyzing the shape of observed trait distributions enables a data-based moment closure of aggregate models
复制标题
DOI：
10.1002/lom3.10218
复制
发表时间：
2017-12-01
影响因子：
2.7
通讯作者：
Klauschies, Toni
中科院分区：
文献类型：
--
作者：
Gaedke, Ursula;Klauschies, Toni
性状分布的形状可以告知结构生态群落的选择力。在这里，我们提出了一种新的基于矩的方法来分类观察到的生物量加权性状分布的形状为正常，峰值，偏斜，或双峰，有利于时空和跨系统的比较。我们观察到的浮游植物性状分布表现出很大的差异，大多是偏态或双峰，而不是正常的。此外，平均值、方差、偏度和峰度都有很强的相关性。这与基于特质的聚合模型相冲突，后者通常假设正态分布的特质值和小方差。考虑到我们的数据和一般模型假设之间的这些差异，我们使用观察到的性状分布来测试不同的聚合模型与一阶或二阶近似和不同类型的矩封闭预测生物量，平均性状和性状方差动态使用弱或中度非线性适应度函数。对于弱非线性适应度函数，具有二阶近似和基于数据的矩闭合的聚合模型依赖于偏度和平均值之间的观察到的相关性，峰度和方差预测生物量，并且通常也平均性状变化相当好，并且比具有一阶近似或基于正常的矩闭合的模型更好。相反，没有一个模型可靠地反映了性状方差的变化。对于中度非线性适应度函数，聚合模型的性能通常也很差。这对基于正态的方法的普遍适用性提出了质疑，特别是在预测方差动态确定性状变化速度和生物多样性维持方面。我们详细评估如何以及为什么可以获得更好的近似。
The shape of trait distributions may inform about the selective forces that structure ecological communities. Here, we present a new moment-based approach to classify the shape of observed biomass-weighted trait distributions into normal, peaked, skewed, or bimodal that facilitates spatio-temporal and cross-system comparisons. Our observed phytoplankton trait distributions exhibited substantial variance and were mostly skewed or bimodal rather than normal. Additionally, mean, variance, skewness und kurtosis were strongly correlated. This is in conflict with trait-based aggregate models that often assume normally distributed trait values and small variances. Given these discrepancies between our data and general model assumptions we used the observed trait distributions to test how well different aggregate models with first- or second-order approximations and different types of moment closure predict the biomass, mean trait, and trait variance dynamics using weakly or moderately nonlinear fitness functions. For weakly non-linear fitness functions aggregate models with a second-order approximation and a data-based moment closure that relied on the observed correlations between skewness and mean, and kurtosis and variance predicted biomass and often also mean trait changes fairly well and better than models with first-order approximations or a normal-based moment closure. In contrast, none of the models reflected the changes of the trait variances reliably. Aggregate model performance was often also poor for moderately nonlinear fitness functions. This questions a general applicability of the normal-based approach, in particular for predicting variance dynamics determining the speed of trait changes and maintenance of biodiversity. We evaluate in detail how and why better approximations can be obtained.