Spatiotemporal interactions of a novel mesocarnivore community in an urban environment before and during SARS‐CoV‐2 lockdown

Data collection The camera trap study of urban mammals (www.wildtierforscher-berlin.de) is one of the scientific projects conducted by citizen scientists within the knowledge transfer project WTimpact (http://www.wtimpact.de). We divided the area of Berlin into a regular grid of 287 2 x 2 km and accepted around 200 Berlin citizens per sampling phase with private gardens (either adjunct to their residential area or within an allotment), trying to get at least one participant per each 2 x 2 km grid per sampling phase to ensure spatial independence of the data. For each new sampling phase, we selected new citizen scientists while respecting this spatial grid. The camera traps took three consecutive pictures when triggered. We repeated this study for five sampling phases: first; October 7th - November 4th 2018, second; April 1st - April 28th 2019, third; September 30th - October 27th 2019, fourth; from March 30th - April 26th 2020 and fifth; September 28th - October 26th 2020. The mesocarnivore community was composed of the native red fox (Vulpes vulpes) and native martens (Martes foina and Martes martes), the invasive raccoon (Procyon lotor) and the feral/domestic cat (Felis catus), directly associated with human activities. We excluded badgers (Meles meles) as their presence was rare. Species’ spatial analyses We first checked for spatial overlap of mesocarnivore species by modelling mesocarnivore community assemblage in response to environmental covariates as well as the species associations using Joint Species Distribution Models (JSDM) in a hierarchical Bayesian framework using the R package Hmsc. JSDMs are a multivariate method that analyse the response of multiple species to environmental drivers and allow to assess species associations in the residual variance after accounting for the environmental effects. We analysed the urban mesocarnivore community spatial patterns using three complementary approaches: (i) a binary detection-non detection model (‘detection’ hereafter) based on the detection of each species at least once at a camera trap location during a sampling phase, (ii) a relative use intensity model (‘use intensity’ hereafter) based on the number of independent pictures (i.e. filtered with a time difference of 30 minutes) of each species at a camera-trap location during each sampling phase, and (iii) a nocturnality model (‘nocturnality’ hereafter) based on the proportion of independent pictures taken at night over the total number of independent pictures taken at a camera-trap location, per species and sampling phase. Only for the nocturnality model did we restrict the data set to the night time between 6 pm and 6 am, corresponding to when wild species were mostly active, according to the activity pattern analyses. In all three models, we included as fixed effects environmental variables related to four main groups: sampling phase, garden characteristics, local urban environmental variables, and the effects of cats. Given that cats are attached to the households they belong to (with the exception of stray cats), we considered them as explanatory variable associated with the environmental conditions. We therefore included cat presence (detection model), cat use intensity (use intensity model) or cat nocturnality (nocturnality model) as explanatory variable in the respective models. Finally, season was included to account for variability of mesocarnivores’ activity within the year, as a binary categorical variable spring/fall. During our study, the epidemic of the Novel Coronavirus SARS-CoV-2 reached Berlin. The Berlin Senate established several contingency measures, resulting in lockdowns during spring and fall 2020. Consequently, human activities drastically decreased during this global shutdown, leading to an increase in wildlife sightings, possibly representing a change of activity patterns of urban wildlife. To account for a possible change in urban mesocarnivores’ space use and activity pattern in Berlin gardens we created a binary variable of the SARS-CoV-2 lockdown, denoted covid/ no_covid, referring to low (covid)/ high (no_covid) human disturbance, respectively. Temporal analyses To test for temporal partitioning between the mesocarnivores of Berlin, we first filtered the pictures of the same species with a minimum time difference of 30 minutes to consider independent presence events. Using the R package camtrapR, we compared the activity patterns of all four species by assessing the temporal overlap D1 between each species. The coefficient ranges from 0 (no overlap) to 1 (complete overlap) and refers to the area under both density curves resulting from the activity patterns of each species. To test for avoidance or attraction we measured the time interval between the last picture of a species and the first picture of the focal species, hereafter called ‘time of delay’, for gardens where both species were detected. For the temporal analyses, we restricted the pictures to when wild species were mostly active, according to the activity pattern analyses, i.e. between 6 pm and 6 am. The time of delay for the red fox, as focal species, for instance, would be the time difference between the last picture of a raccoon, a cat or a marten, and the first picture of a fox. In our study we then considered that the focal species would avoid another species if the time of delay was significantly greater than for its own species. In this case, we also analysed the time of delay of cats as a response variable, in contrast to the spatial analyses. Finally, to account for differences of probability of presences in gardens, we ran pair-wise regression of time of delay: we restrained the data to gardens where only the two species occurred and ran a similar regression with only one variable; the species after which the focal species was detected.

### 数据收集
本城市哺乳动物相机陷阱（camera trap）研究（www.wildtierforscher-berlin.de）是市民科学家参与的WTimpact知识转移项目（http://www.wtimpact.de）下设的科研项目之一。我们将柏林市域划分为287个2×2公里的规则网格，并在每个采样阶段招募约200名拥有私人花园（毗邻住宅或属于自留地）的柏林市民作为参与者，力求每个2×2公里网格在每个采样阶段至少有一名参与者，以保障数据的空间独立性。每轮新采样阶段均会重新招募市民科学家，并严格遵循该空间网格规则。
相机陷阱触发后将连续拍摄三张照片。本研究共开展五轮采样：第一轮：2018年10月7日—11月4日；第二轮：2019年4月1日—4月28日；第三轮：2019年9月30日—10月27日；第四轮：2020年3月30日—4月26日；第五轮：2020年9月28日—10月26日。
本研究的中型食肉动物（mesocarnivore）群落涵盖本土物种赤狐（Vulpes vulpes）、本土貂类（石貂Martes foina和松貂Martes martes）、入侵物种浣熊（Procyon lotor）以及与人类活动直接相关的家/流浪猫（Felis catus）。由于狗獾（Meles meles）的目击记录稀少，故将其排除在分析之外。

### 物种空间分析
我们首先通过联合物种分布模型（Joint Species Distribution Models, JSDM），结合R包Hmsc在分层贝叶斯框架下构建中型食肉动物群落组成对环境协变量的响应模型，并评估物种间关联，以此检验各中型食肉动物物种的空间重叠情况。联合物种分布模型是一种多元分析方法，可解析多物种对环境驱动因子的响应，并能在扣除环境效应后评估残差方差中的物种关联。
我们采用三种互补方法分析城市中型食肉动物群落的空间格局：（1）二元检测-未检测模型（以下简称"检测模型"）：基于某采样阶段内某相机陷阱点位是否至少一次检测到目标物种；（2）相对使用强度模型（以下简称"使用强度模型"）：基于某采样阶段内某相机陷阱点位处每个物种的独立有效照片数量（即按30分钟时间差进行过滤后的照片）；（3）夜行性模型（以下简称"夜行性模型"）：基于某采样阶段内某相机陷阱点位处每个物种的夜间独立有效照片占总独立有效照片的比例。仅在构建夜行性模型时，我们将数据集限定在野生物种主要活动时段的夜间（18:00至次日6:00），该划分基于活动模式分析结果。
在上述三种模型中，我们均纳入四类固定效应环境变量：采样阶段、花园特征、局地城市环境变量以及猫的影响。鉴于家猫（流浪猫除外）依附于所属住户，我们将其视为与环境条件相关的解释变量，因此在对应模型中分别纳入猫的检测情况（检测模型）、猫的使用强度（使用强度模型）或猫的夜行性特征（夜行性模型）作为解释变量。此外，我们将季节作为二元分类变量（春季/秋季）纳入模型，以控制年内中型食肉动物活动的变异性。
研究期间，新型冠状病毒SARS-CoV-2疫情波及柏林。柏林参议院出台多项应急措施，导致2020年春季与秋季实施封锁。受全球停摆影响，人类活动大幅减少，野生动物目击记录增多，这可能改变了城市野生动物的活动模式。为评估柏林花园内城市中型食肉动物的空间使用与活动模式是否发生变化，我们创建了"SARS-CoV-2封锁"二元变量，记为"新冠疫情封锁期（covid）/非封锁期（no_covid）"，分别对应低人类干扰（封锁期）与高人类干扰（非封锁期）。

### 时间分析
为检验柏林中型食肉动物之间的时间生态位分化，我们首先按至少30分钟的时间差对同一物种的照片进行过滤，以区分独立的出现事件。我们使用R包camtrapR，通过计算每两个物种间活动模式的时间重叠系数D1，比较四种物种的活动模式。该系数取值范围为0（无重叠）至1（完全重叠），对应两种物种活动密度曲线下的重叠面积。为检验物种间的回避或吸引关系，我们针对同时检测到两种目标物种的花园，计算某一物种最后一张照片与焦点物种第一张照片之间的时间间隔，以下简称"延迟时间"。
在时间分析中，我们将照片限定在野生物种主要活动时段（18:00至次日6:00）。例如，以赤狐为焦点物种时，延迟时间指浣熊、猫或貂类最后一张照片与赤狐第一张照片之间的时间差。本研究中，若延迟时间显著高于同物种自身的延迟时间，则认为焦点物种对另一物种存在回避行为。与空间分析不同，本次分析中我们同时以猫作为焦点物种来分析延迟时间。最后，为控制花园间出现概率的差异，我们开展了延迟时间的配对回归：将数据限定为仅存在两种目标物种的花园，并以焦点物种之后出现的物种作为唯一自变量进行回归分析。