Beaver activity and red squirrel presence predict bird assemblages in boreal Canada

Wetlands and predation in boreal ecosystems play essential roles throughout the breeding season for bird assemblages. We found a positive association of beaver activity and a negative influence of American red squirrels (Tamiasciurus hudsonicus) on bird assemblages. We used a multispecies hierarchical model to investigate whether bird communities differ between two major wetland habitats in boreal Canada: beaver ponds and peatland ponds. In addition to including variables such as forest cover and latitude, we adopted a structural equation model approach to estimate the occupancy of American red squirrels and its potential influence on bird communities. Using automated recording stations deployed at 50 ponds, we detected 96 bird species in 2018 and 2019. Bird species were grouped into four taxonomic guilds according to their habitat successional requirements: early successional species, late successional species, generalists, and wetland species. Beaver ponds harbored higher species richness, a pattern driven primarily by early successional species. The occupancy of almost a quarter of the species was lower in the presence of red squirrels. Late successional species responded positively to the cover of forest surrounding the pond. Our results highlight the value of considering acoustic data of red squirrels to quantify habitat quality in boreal forests. We conclude that beaver activity shapes bird assemblages through modification of their habitat, and that some bird guilds are associated negatively with the presence of American red squirrels.

Methods
Acoustic sampling and covariates.

We used a SM4 Song meter (Wildlife Acoustics Incorporation, Concord, Massachusetts, USA) at each pond to record bird and squirrel vocalizations. During a given sampling period, ARUs were set with both microphones facing upward for four consecutive days to record a 5-min sample at 00600 h. Each file was encoded at a sampling rate of 44.1 kHz and a 16-bit resolution. Each pond was sampled for four consecutive days during two periods in both years, yielding an average of 80 minutes per pond. All recording data were analyzed by an ornithologist with extensive field experience in the study region to identify species based on their vocalizations or other aural cue. Species were categorized into four guilds based on their requirements for various successional stages of black spruce: (1) early successional species, associated with open areas of low-vegetation such as young forests or regenerating shrublands; (2) late-successional species preferring older forest stands; (3) generalists, consisting of species that use a wide range of successional stages; and (4) wetland and aquatic species. In addition to birds, we identified the vocalizations of the American red squirrel from recordings for use in occupancy analysis. We included four candidate variables as potential predictors of boreal bird occupancy: pond type, red squirrel latent occupancy, proportion of all forest types over 2 m (deciduous, mixed, and coniferous forest) within circular buffers of 1000 m centered on acoustic recorder locations, and we included latitude to reflect habitat changes in the study area. We considered two parameters that potentially influenced detection probability during acoustic surveys: the quality of the recording and the number of days after snowmelt.

Data processing and analysis.

We investigated occupancy patterns of the American red squirrel, and the entire bird community detected at least once in our recordings. We prepared detection histories for each species at each pond, where we indicated detection (1) or non-detection (0) in each recording of the two periods of four consecutive days in each year. Each pond in each year was considered as an independent site and we included a fixed year effect as well as a site random effect to account for potential differences in occupancy between years. Thus, detection histories for a given species were arranged in a matrix of 100 rows (50 sites x 2 years) and 8 columns (4 recordings x 2 periods). We estimated bird community structure with a multispecies site-occupancy model. We adapted a structural equation strategy to our multispecies occupancy model. Specifically, one component of our model estimated the occupancy of American red squirrels as a function of different explanatory variables, and then used the latent squirrel occupancy as an explanatory variable in the multispecies occupancy model for the bird communities. We used the multispecies model to derive the species richness from the posterior distribution of the true occurrence of each species at each site, expressed as the mean of the posterior distribution. We investigated the relationship between species richness against the explanatory variables pond type, forest cover, latitude, and the posterior mean of the squirrel occupancy state in a linear mixed-effect model.

北方寒带生态系统中的湿地与捕食作用，在鸟类群落的整个繁殖季发挥着至关重要的作用。我们发现，河狸活动与鸟类群落呈正相关关系，而北美红松鼠（Tamiasciurus hudsonicus）则对鸟类群落产生负面影响。我们采用多物种层级模型，探究加拿大寒带地区两种主要湿地生境——河狸塘与泥炭塘之间的鸟类群落差异。除纳入森林覆盖度、纬度等变量外，我们还采用结构方程模型方法，估算北美红松鼠的占据率及其对鸟类群落的潜在影响。

通过部署在50个水塘的自动录音站，我们在2018年与2019年共检测到96种鸟类。依据鸟类对生境演替阶段的需求，将其划分为4个功能类群：早期演替物种、晚期演替物种、广适性物种以及湿地物种。河狸塘的物种丰富度更高，这一模式主要由早期演替物种驱动。近四分之一的物种，其占据率在红松鼠存在时显著降低。晚期演替物种对水塘周边的森林覆盖度呈正响应。我们的研究结果凸显了利用红松鼠的声学数据量化寒带森林生境质量的价值。综上，河狸活动通过改变生境塑造鸟类群落，而部分鸟类功能类群与北美红松鼠的存在呈负相关关系。

方法
声学采样与协变量
我们在每个水塘部署SM4型声学记录仪（Song Meter，Wildlife Acoustics公司，美国马萨诸塞州康科德市），以录制鸟类与松鼠的鸣叫声。在单次采样周期内，自动录音设备（ARUs，Automated Recording Units）的两个麦克风均朝上设置，连续4天每日在06:00录制5分钟的音频样本。所有音频文件均以44.1kHz采样率、16位分辨率进行编码。我们在两年内的两个采样周期中，对每个水塘连续4天进行采样，平均每个水塘累计采样时长约80分钟。所有录音数据均由一位在研究区域拥有丰富野外经验的鸟类学家，通过鸣叫声或其他听觉线索完成物种识别。

依据对黑云杉不同演替阶段的生境需求，我们将物种划分为4个功能类群：（1）早期演替物种，与低植被覆盖的开阔区域（如幼龄林或更新灌丛）相关联；（2）晚期演替物种，偏好成熟林分；（3）广适性物种，可利用广泛的演替阶段生境；（4）湿地与水生物种。除鸟类外，我们还从录音中识别出北美红松鼠的鸣叫声，用于占据率分析。我们纳入4个候选变量作为寒带鸟类占据率的潜在预测因子：水塘类型、红松鼠潜在占据率、以录音站为中心的1000米圆形缓冲区内所有高度超过2米的森林类型（落叶林、混交林与针叶林）的占比，同时纳入纬度变量以反映研究区域的生境变化。我们还考虑了声学调查期间可能影响检测概率的两个参数：录音质量以及融雪后的天数。

数据处理与分析
我们探究了北美红松鼠的占据模式，以及在录音中至少被检测到一次的整个鸟类群落。我们为每个水塘的每个物种构建了检测历史，其中在每年两个周期、连续4天的每次录音中，记录检测（1）或未检测（0）结果。我们将每年的每个水塘视为独立样点，并纳入固定年份效应与样点随机效应，以解释不同年份间占据率的潜在差异。因此，某一物种的检测历史被整理为100行（50个样点×2年）×8列（4次录音×2个周期）的矩阵。我们采用多物种样点占据模型估算鸟类群落结构，并将结构方程策略适配至多物种占据模型中。具体而言，模型的一个组分以不同解释变量为基础估算北美红松鼠的占据率，随后将潜在的松鼠占据率作为解释变量纳入鸟类群落的多物种占据模型。我们通过多物种模型，从每个样点各物种真实出现情况的后验分布中推导物种丰富度，该丰富度表示为后验分布的均值。我们采用线性混合效应模型，探究物种丰富度与解释变量（水塘类型、森林覆盖度、纬度以及松鼠占据率的后验均值）之间的关系。