Buras Allan, van der Maaten-Theunissen Marieke, van der Maaten Ernst, Ahlgrimm Svenja, Hermann Philipp, Simard Sonia, Heinrich Ingo, Helle Gerd, Unterseher Martin, Schnittler Martin, Eusemann Pascal, Wilmking Martin
Institute of Botany and Landscape Ecology, Ernst-Moritz-Arndt-Universität, Soldmannstraße 15, 17487 Greifswald, Germany.
GFZ German Research Centre for Geosciences, Section 5.2, Telegrafenberg, 14473 Potsdam, Germany.
PLoS One. 2016 Jul 28;11(7):e0158346. doi: 10.1371/journal.pone.0158346. eCollection 2016.
This paper introduces a new approach-the Principal Component Gradient Analysis (PCGA)-to detect ecological gradients in time-series populations, i.e. several time-series originating from different individuals of a population. Detection of ecological gradients is of particular importance when dealing with time-series from heterogeneous populations which express differing trends. PCGA makes use of polar coordinates of loadings from the first two axes obtained by principal component analysis (PCA) to define groups of similar trends. Based on the mean inter-series correlation (rbar) the gain of increasing a common underlying signal by PCGA groups is quantified using Monte Carlo Simulations. In terms of validation PCGA is compared to three other existing approaches. Focusing on dendrochronological examples, PCGA is shown to correctly determine population gradients and in particular cases to be advantageous over other considered methods. Furthermore, PCGA groups in each example allowed for enhancing the strength of a common underlying signal and comparably well as hierarchical cluster analysis. Our results indicate that PCGA potentially allows for a better understanding of mechanisms causing time-series population gradients as well as objectively enhancing the performance of climate transfer functions in dendroclimatology. While our examples highlight the relevance of PCGA to the field of dendrochronology, we believe that also other disciplines working with data of comparable structure may benefit from PCGA.
本文介绍了一种新方法——主成分梯度分析（PCGA），用于检测时间序列种群中的生态梯度，即源自种群中不同个体的多个时间序列。在处理来自表现出不同趋势的异质种群的时间序列时，检测生态梯度尤为重要。PCGA利用主成分分析（PCA）得到的前两个轴上载荷的极坐标来定义具有相似趋势的组。基于平均序列间相关性（rbar），通过蒙特卡罗模拟对PCGA组增加共同潜在信号的增益进行量化。在验证方面，将PCGA与其他三种现有方法进行比较。以树木年代学实例为重点，结果表明PCGA能够正确确定种群梯度，并且在特定情况下比其他考虑的方法更具优势。此外，每个实例中的PCGA组都能够增强共同潜在信号的强度，并且与层次聚类分析效果相当。我们的结果表明，PCGA有可能更好地理解导致时间序列种群梯度的机制，以及客观地提高树木年代气候学中气候传递函数的性能。虽然我们的实例突出了PCGA在树木年代学领域的相关性，但我们认为其他处理具有可比结构数据的学科也可能从PCGA中受益。
