The relative influence of history, climate, topography and vegetation structure on local animal richness varies among taxa and spatial grains

Understanding the spatial scales at which environmental factors drive species richness patterns is a major challenge in ecology. Due to the trade-off between spatial grain and extent, studies tend to focus on a single spatial scale, and the effects of multiple environmental variables operating across spatial scales on the pattern of local species richness have rarely been investigated.

Here, we related variation in local species richness of ground beetles, landbirds, and small mammals to variation in vegetation structure and topography, regional climate, biome diversity, and glaciation history for 27 sites across the USA at two different spatial grains.

We studied the relative influence of broad-scale (landscape) environmental conditions using variables estimated at the site level (climate, productivity, biome diversity, and glacial era ice cover) and fine-scale (local) environmental conditions using variables estimated at the plot level (topography and vegetation structure) to explain local species richness. We also examined whether plot-level factors scale up to drive continental scale richness patterns. We used Bayesian hierarchical models and quantified the amount of variance in observed richness that was explained by environmental factors at different spatial scales.

For all three animal groups, our models explained much of the variation in local species richness (85-89%), but site-level variables explained a greater proportion of richness variance than plot-level variables. Temperature was the most important site-level predictor for explaining variance in landbirds and ground beetles richness. Some aspects of vegetation structure were the main plot-level predictors of landbird richness. Environmental predictors generally had poor explanatory power for small mammal richness, while glacial era ice cover was the most important site-level predictor.

Relationships between plot-level factors and richness varied greatly among geographical regions and spatial grains, and most relationships did not hold when predictors were scaled up to continental scale. Our results suggest that the factors that determine richness may be highly dependent on spatial grain, geography, and animal group. We demonstrate that instead of artificially manipulating the resolution to study multi-scale effects, a hierarchical approach that uses fine grain data at broad extents could help solve the issue of scale selection in environment-richness studies. 

Methods

We estimated landbird richness at the plot level using the NEON dataset of breeding bird point counts (Carrasco et al. 2022: National Ecological Observatory Network 2016a). Birds associated with terrestrial habitats were sampled during the breeding season using point counts performed by one or more expert observers who recorded all species seen or heard within a 125 m radius during a 6-minute period. To allow comparison across plots with different number of point counts, we only used the central point count for plots with multiple point counts. To standardise the sampling effort among plots, we aggregated data from two survey dates for each plot, carried out during the years 2016, 2017 or 2018, which resulted in 192 plots across 25 NEON sites for our analysis (Carrasco et al. 2022: Supplementary Material Table S1). We selected survey dates based on their temporal proximity to the available lidar data (for calculating topographical and vegetation structure indices).

We obtained data on ground beetles from the NEON’s Ground Beetles Sampled from Pitfall Traps dataset (Carrasco et al. 2022: National Ecological Observatory Network 2016c), which provides counts of ground beetles (Coleoptera: Carabidae). The NEON sampling protocol uses four pitfall traps (473 mL deli containers filled with 150 or 250 mL of propylene glycol) placed 20 m from the centre of the plot in the four cardinal directions. Sampling occurred in a two-week duration (i.e., a bout) during the growing season. We used 10 bouts per year in each plot in order to standardise the sampling efforts. Unlike for landbirds and small mammals, we only selected plots that were sampled in 2018 (n = 110 plots across 19 NEON sites; Carrasco et al. 2022: Supplementary Material Table S1). We made this decision because the number of traps per bout changed from four to three from 2018 on, and therefore using data from different years would have led to inconsistent sampling efforts.

We aggregated taxa identified to species across the 10 bouts to calculate plot-level species richness. For specimens sent for taxonomic validation, we used the species ID assigned by the expert taxonomist. For ground beetles, we used lidar data that were collected temporally closest to 2018.

We extracted data on small mammals from the NEON’s Small Mammal Box Trapping dataset (Carrasco et al. 2022: National Ecological Observatory Network 2016d). NEON defines a small mammal as any rodent that is nonvolant, nocturnally active, and an aboveground forager weighing 5-600 g (e.g., cricetids, heteromyds, small sciurids and murids, etc.). Mammals were sampled using box traps, which were configured in a 10 m x 10 m grid (totalling 100 box traps) in most plots. Only species classified as “targeted” by NEON’s box traps were used in our analyses. To standardise sampling efforts, we included four bouts (each bout constitutes three consecutive nights of trapping) per plot and year. We estimated species richness based on individuals identified to the species level for each plot and year (2016-2018). We excluded opportunistic captures of non-targeted species. Lastly, for statistical modelling we retained from each plot only data from the year closest to the year of the lidar data (n = 89 plots across 23 NEON sites; Carrasco et al. 2022: Supplementary Material Table S1).

For all three taxonomic groups, our plot-level species richness metric is equivalent to the species density metric described by Gotelli and Colwell (2011). We used the term richness in our paper because we combined multiple surveys for birds and sampling bouts for ground beetles and small mammals (each of which aggregates multiple traps), to maximise sampling completeness within each plot, while ensuring that the sampling effort is the same by using the same number of surveys and bouts across all plots within each taxonomic group. Despite these measures, sampling completeness may nevertheless differ across plots. We examined the correlation between our raw species richness metric and the estimated asymptotic richness (Carrasco et al. 2022: Chao et al. 2014) (calculated using the iNext package in R (Carrasco et al. 2022: Hsieh et al. 2016)) to assess if differences in sampling completeness might affect our analyses and results from section 2.4 below. We found moderate to high positive correlation (landbirds: Pearson r = 0.60; ground beetles: r = 0.93; small mammals: r = 0.99) between our richness metric and estimated asymptotic richness (Carrasco et al. 2022: Supplementary Material Figure S1), suggesting that differences in sampling completeness is unlikely to affect the main conclusionsofouranalysis,especially for ground beetles and small mammals. Detailed information on animal surveys and richness estimation methodology can be found in Carrasco et al. 2022: Supplementary Material Appendix S1.

探明环境因子驱动物种丰富度格局的空间尺度，是生态学领域的核心挑战之一。由于空间粒度（spatial grain）与空间幅度（spatial extent）之间存在权衡关系，现有研究往往仅聚焦单一空间尺度，而针对跨空间尺度作用的多环境变量对局域物种丰富度格局的影响，相关探讨仍较为匮乏。

本研究以美国境内27个样点为研究对象，采用两种不同的空间粒度，将步甲（ground beetles）、陆禽（landbirds）与小型哺乳类的局域物种丰富度变异，与植被结构、地形、区域气候、生物群区多样性（biome diversity）以及冰川历史（glaciation history）的变异进行关联分析。

本研究分别采用样点尺度估算的变量（气候、生产力、生物群区多样性及冰期冰盖覆盖度）表征大尺度（景观尺度，landscape）环境条件，采用样方尺度估算的变量（地形与植被结构）表征微尺度（局域尺度，local）环境条件，以此解析二者对局域物种丰富度的相对影响；同时探讨样方尺度因子是否可向上推演，驱动大陆尺度的物种丰富度格局。本研究采用贝叶斯层级模型（Bayesian hierarchical models），量化不同空间尺度下环境因子对观测物种丰富度的变异解释量。

针对三类动物类群，本研究模型均可解释局域物种丰富度的大部分变异（85%~89%），但样点尺度变量对丰富度变异的解释比例高于样方尺度变量。温度是解释陆禽与步甲丰富度变异的最重要样点尺度预测因子；部分植被结构指标则是陆禽丰富度的主要样方尺度预测因子。环境预测因子对小型哺乳类丰富度的解释能力普遍较弱，而冰期冰盖覆盖度则是其最重要的样点尺度预测因子。

样方尺度因子与物种丰富度之间的关联在不同地理区域与空间粒度间存在显著差异，且当预测因子被推演至大陆尺度时，多数关联不再成立。研究结果表明，物种丰富度的决定因子可能高度依赖空间粒度、地理背景与动物类群。本研究证实，相较于人为调整分辨率以探究多尺度效应，采用大空间幅度下的微尺度数据的层级分析方法，或可解决环境-丰富度研究中的尺度选择难题。

方法

本研究采用美国国家生态观测站网络（National Ecological Observatory Network, NEON）的繁殖鸟类点计数数据集（Carrasco等，2022：美国国家生态观测站网络，2016a）估算样方尺度的陆禽丰富度。研究针对陆地生境关联鸟类，于繁殖季开展点计数调查：由一名或多名资深观察者记录6分钟内、125米半径范围内所见或听闻的所有鸟类物种。为统一不同样点的点计数数量差异，对于设置多个点计数的样方，仅采用其中央点计数数据。为标准化各样方的采样努力，我们整合了2016、2017或2018年每个样方两次调查日期的数据，最终纳入25个NEON站点的192个样方用于分析（Carrasco等，2022：补充材料表S1）。调查日期的选择基于其与可用激光雷达（lidar）数据的时间邻近性，以用于计算地形与植被结构指数。

我们从NEON的地面陷阱诱捕步甲数据集（Carrasco等，2022：美国国家生态观测站网络，2016c）中获取步甲数据，该数据集记录了步甲（鞘翅目：步甲科，Coleoptera: Carabidae）的捕获数量。NEON的采样方案为：在样方中心四周20米处的四个方位放置4个地面陷阱（473mL的熟食容器，内装150或250mL丙二醇）。采样于生长季内开展，持续两周为一轮采样。为统一采样努力，每个样方每年采用10轮采样。与陆禽和小型哺乳类不同，我们仅选择2018年采样的样方（共19个NEON站点的110个样方；Carrasco等，2022：补充材料表S1），原因在于2018年起每轮采样的陷阱数量从4个调整为3个，若使用不同年份的数据会导致采样努力不一致。我们将10轮采样中鉴定至物种水平的类群数据整合，以计算样方尺度的物种丰富度；对于送至分类学验证的标本，采用专家分类学家给定的物种编号。步甲数据对应的激光雷达数据选取时间上最接近2018年的数据集。

我们从NEON的小型哺乳类笼捕数据集（Carrasco等，2022：美国国家生态观测站网络，2016d）中提取小型哺乳类数据。NEON将小型哺乳类定义为：非飞行、夜行性、地面觅食、体重5~600g的啮齿类（如仓鼠科、异鼠科、小型松鼠科与鼠科类群等）。多数样方采用10m×10m的网格布设笼捕陷阱（共100个）开展哺乳类采样。本研究仅纳入NEON笼捕方案中定义为“目标物种”的类群。为标准化采样努力，每个样方每年纳入4轮采样（每轮为连续3晚的诱捕作业）。我们基于每个样方每年（2016~2018年）鉴定至物种水平的个体数据估算物种丰富度，并排除偶然捕获的非目标物种。最后，为开展统计建模，我们从每个样方中仅选取与激光雷达数据年份最接近的年度数据，最终纳入23个NEON站点的89个样方（Carrasco等，2022：补充材料表S1）。

对于三类分类学类群，本研究采用的样方尺度物种丰富度指标与Gotelli与Colwell（2011）描述的物种密度指标等价。本研究使用“丰富度”这一术语，是因为我们将鸟类的多次调查数据、步甲与小型哺乳类的多轮采样数据（每轮整合多个陷阱数据）进行整合，以最大化每个样方内的采样完整性，同时确保每个类群的所有样方采用相同数量的调查与采样轮次，以此统一采样努力。尽管采取了上述措施，各样方的采样完整性仍可能存在差异。为评估采样完整性差异是否会影响本研究的分析与2.4节的结果，我们检验了原始物种丰富度指标与估算的渐近丰富度（Carrasco等，2022：Chao等，2014）之间的相关性——该渐近丰富度通过R语言的iNext包计算得到（Carrasco等，2022：Hsieh等，2016）。结果显示，三类类群的丰富度指标与渐近丰富度均呈中度至高度正相关（陆禽：Pearson相关系数r=0.60；步甲：r=0.93；小型哺乳类：r=0.99）（Carrasco等，2022：补充材料图S1），这表明采样完整性差异不太会影响本研究的主要结论，尤其是步甲与小型哺乳类类群。关于动物调查与丰富度估算方法的详细信息，可参见Carrasco等（2022）的补充材料附录S1。