Albert Loren P, Keenan Trevor F, Burns Sean P, Huxman Travis E, Monson Russell K
Department of Ecology and Evolutionary Biology, University of Arizona, Tucson, AZ, 85721, USA.
Lawrence Berkeley National Laboratory, Berkeley, CA, 94709, USA.
Oecologia. 2017 May;184(1):25-41. doi: 10.1007/s00442-017-3853-0. Epub 2017 Mar 25.
Eddy covariance (EC) datasets have provided insight into climate determinants of net ecosystem productivity (NEP) and evapotranspiration (ET) in natural ecosystems for decades, but most EC studies were published in serial fashion such that one study's result became the following study's hypothesis. This approach reflects the hypothetico-deductive process by focusing on previously derived hypotheses. A synthesis of this type of sequential inference reiterates subjective biases and may amplify past assumptions about the role, and relative importance, of controls over ecosystem metabolism. Long-term EC datasets facilitate an alternative approach to synthesis: the use of inductive data-based analyses to re-examine past deductive studies of the same ecosystem. Here we examined the seasonal climate determinants of NEP and ET by analyzing a 15-year EC time-series from a subalpine forest using an ensemble of Artificial Neural Networks (ANNs) at the half-day (daytime/nighttime) time-step. We extracted relative rankings of climate drivers and driver-response relationships directly from the dataset with minimal a priori assumptions. The ANN analysis revealed temperature variables as primary climate drivers of NEP and daytime ET, when all seasons are considered, consistent with the assembly of past studies. New relations uncovered by the ANN approach include the role of soil moisture in driving daytime NEP during the snowmelt period, the nonlinear response of NEP to temperature across seasons, and the low relevance of summer rainfall for NEP or ET at the same daytime/nighttime time step. These new results offer a more complete perspective of climate-ecosystem interactions at this site than traditional deductive analyses alone.
几十年来，涡度协方差（EC）数据集为了解自然生态系统中净生态系统生产力（NEP）和蒸散量（ET）的气候决定因素提供了见解，但大多数EC研究都是以系列方式发表的，以至于一项研究的结果成为后续研究的假设。这种方法通过关注先前得出的假设反映了假设演绎过程。这种顺序推理的综合重申了主观偏见，并可能放大过去关于生态系统代谢控制的作用和相对重要性的假设。长期的EC数据集有助于采用另一种综合方法：使用基于归纳数据的分析来重新审视对同一生态系统的过去演绎研究。在这里，我们通过在半日（白天/夜间）时间步长使用人工神经网络（ANN）集合分析来自亚高山森林的15年EC时间序列，研究了NEP和ET的季节性气候决定因素。我们在极少先验假设的情况下直接从数据集中提取了气候驱动因素的相对排名以及驱动因素与响应的关系。ANN分析表明，当考虑所有季节时，温度变量是NEP和白天ET的主要气候驱动因素，这与过去研究的汇总结果一致。ANN方法发现的新关系包括融雪期土壤湿度对白天NEP的驱动作用、NEP随季节对温度的非线性响应，以及在相同白天/夜间时间步长下夏季降雨对NEP或ET的低相关性。这些新结果比单独的传统演绎分析提供了该地点气候与生态系统相互作用更完整的视角。
