Inferring species interactions in ecological communities: a comparison of methods at different levels of complexity

1. Natural communities commonly contain many different species and functional groups, and multiple types of species interactions act simultaneously, such as competition, predation, commensalism or mutualism. However, experimental and theoretical investigations have generally been limited by focusing on one type of interaction at a time or by a lack of a common methodological and conceptual approach to measure species interactions. 2. We compared four methods to measure and express species interactions. These approaches are, with increasing degree of model complexity, an extinction-based model, a relative yield model and two generalized Lotka-Volterra (LV) models. All four approaches have been individually applied in different fields of community ecology, but rarely integrated. We provide an overview of the definitions, assumptions and data needed for the specific methods and apply them to empirical data by experimentally deriving the interaction matrices among 11 protist and rotifer species, belonging to three functional groups. Furthermore, we compare their advantages and limitations to predict multispecies community dynamics and ecosystem functioning. 3. The relative yield method is, in terms of final biomass production, the best method in predicting the 11-species community dynamics from the pairwise competition experiments. The LV model, which is considering equilibrium among the species, suffers from experimental constraints given the strict equilibrium assumption, and this may be rarely satisfied in ecological communities. 4. We show how simulations of a LV stochastic community model, derived from an empirical interaction matrix, can be used to predict multispecies community dynamics across multiple functional groups. 5. Our work unites available tools to measure species interactions under one framework. This improves our ability to make management-oriented predictions of species coexistence/extinction and to compare ecosystem processes across study systems.1. 自然群落通常蕴含丰富的物种与功能群，且多种类型的物种相互作用同步发生，例如竞争、捕食、偏利共生或互利共生。但现有实验与理论研究普遍存在两方面局限：要么仅聚焦单一类型的物种相互作用，要么缺乏统一的方法论与概念框架以量化物种相互作用。 2. 本研究对比了四种用于量化与表征物种相互作用的方法。随着模型复杂度依次提升，这四种方法分别为：基于灭绝的模型、相对产量模型，以及两类广义洛特卡-沃尔泰拉（Lotka-Volterra, LV）模型。上述四种方法此前已分别应用于群落生态学的不同分支领域，但极少被整合统一。我们首先系统梳理了各方法的定义、前提假设与所需数据，并通过实验获取隶属于3个功能群的11种原生生物（protist）与轮虫（rotifer）的种间相互作用矩阵，随后将这四种方法应用于该实证数据集。此外，我们还对比了各方法在预测多物种种群动态与生态系统功能方面的优势与局限性。 3. 就最终生物量产出而言，相对产量法是基于两两竞争实验预测11物种种群动态的最优方法。而考虑物种间平衡的洛特卡-沃尔泰拉模型，因严格的平衡假设受到实验条件约束，该假设在自然生态群落中往往难以满足。 4. 我们展示了如何借助从实证相互作用矩阵衍生得到的洛特卡-沃尔泰拉随机群落模型的模拟结果，预测跨多个功能群的多物种种群动态。 5. 本研究将现有用于量化物种相互作用的工具整合至统一框架中，这一工作有助于提升我们开展物种共存与灭绝相关管理导向预测的能力，同时也便于在不同研究系统间对比生态系统过程。