Scale dependency of joint species distribution models challenges interpretation of biotic interactions

Aim: Separating the biotic and abiotic factors controlling species distributions has been a long-standing challenge in ecology and biogeography. Joint species distribution models (JSDMs) have emerged as a promising statistical framework towards this objective by simultaneously modeling the environmental responses of multiple species and approximating species associations based on patterns in their (co-)occurrences. However, the signature of biotic interactions should be most evident at fine spatial resolutions. Here, we test how the resolution of input data affects the inferences from JSDMs.

Location: Switzerland

Taxon: Birds

Methods: Using standardized survey data of 43 woodland bird species and eight climatic, topographic and vegetation structural predictors, we fit JSDMs at different spatial resolutions (125m to 1000m) and sampling periods (1 and 5years). In addition, we calculate functional similarity among all species as an independent proxy of biotic interactions, specifically competition. We then assess how JSDM performance and estimates vary with the spatial resolution of the input data and test whether species associations are consistent across grain sizes and with the alternative approach based on functional similarity.

Results: Our results show better model performance at coarser spatial resolutions and for longer sampling periods. Although pairwise species associations estimated in JSDMs were generally shifted towards positive values, we found a higher proportion of negative associations at fine spatial resolutions. Strikingly, estimates were not consistent across spatial scales and frequently switched between positive and negative values. Moreover, estimated species associations tended to be more positive for functionally similar species.

Main conclusions: Our results show that species associations are more differentiated, i.e. cover a broader range of values, at finer spatial resolutions. Yet, their positive correlation with functional similarity and the general over-representation of positive associations suggest that shared responses to unobserved environmental predictors rather than biotic interactions underlie these scaling effects, cautioning against a naïve interpretation of species associations estimated by JSDMs at any spatial scale.

Methods
Data collection and processing is described in detail in the article.

研究目的：分离调控物种分布的生物与非生物因子，一直是生态学与生物地理学领域长期存在的核心挑战。联合物种分布模型（Joint Species Distribution Models, JSDMs）作为极具潜力的统计框架应运而生，其可同时模拟多物种的环境响应，并基于物种的（共）发生模式近似推断物种间的关联。然而，生物相互作用的信号在精细空间分辨率（spatial resolution）下应最为显著。本研究旨在探究输入数据的空间分辨率如何影响联合物种分布模型的推断结果。
研究区域：瑞士
研究类群：鸟类
研究方法：本研究采用43种林地鸟类的标准化调查数据，以及8项气候、地形与植被结构类预测变量，在125米至1000米的不同空间分辨率，以及1年、5年的不同采样周期（sampling periods）下拟合联合物种分布模型。此外，我们计算所有物种间的功能相似性（functional similarity），将其作为生物相互作用（biotic interactions，特指种间竞争）的独立替代指标。随后，我们评估联合物种分布模型的性能与参数估计值如何随输入数据的空间分辨率发生变化，并检验物种关联在不同空间粒度（spatial grain sizes）下是否一致，以及其是否与基于功能相似性的替代分析方法结果相符。
研究结果：结果显示，在较粗的空间分辨率与更长的采样周期下，模型性能更佳。尽管联合物种分布模型所估计的物种成对关联整体偏向正值，但我们在精细空间分辨率下发现了更高比例的负关联。值得注意的是，参数估计值在不同空间尺度下并不一致，且常在正负值间切换。此外，功能相似性更高的物种，其估计得到的物种关联往往更为正向。
主要结论：本研究结果表明，在更精细的空间分辨率下，物种关联的分化程度更高，即取值范围更广。然而，物种关联与功能相似性的正相关关系，以及正关联的整体过度表征，提示这些尺度效应（scaling effects）的本质是物种对未观测环境因子的共同响应，而非生物相互作用。这警示我们，切勿对任意空间尺度下联合物种分布模型估计得到的物种关联做出未经审慎考量的解读。
方法补充：数据收集与处理的详细描述见原文。