Hot stops: Timing, pathways, and habitat selection of migrating Eastern Whip-poor-wills

Although miniaturized data loggers allow new insights into avian migration, incomplete knowledge of basic patterns persists, especially for nightjars. Using GPS data loggers, this study examined migration ecology of the Eastern whip-poor-will (Antrostomus vociferus), across three migration strategies: flyover, short-stay, and long-stay. We documented migration movements, conducted hotspot analyses, quantified land cover within 1-km and 5-km buffers at used and available locations, and modeled habitat selection during migration. From 2018-2020 we captured breeding whip-poor-wills from three study sites in Massachusetts and programmed GPS tags to collect data during fall and spring migration periods. Across 19 individual males (nine of them with repeated years of data), GPS tags collected 479 locations, where 30% were classified as flyover points, 33% as short-stays, and 37% as long-stay locations. We documented seasonal flexibility in migration duration, routes, and stopover locations among individuals and between years. Analyses identified hotspot clusters in fall and spring migration in the Sierra de Tamaulipas in Mexico. Land cover at used locations differed across location types at the 5-km scale, where closed forest cover increased and crop cover decreased for flyover, short-stay, and long-stay locations, and urban cover was lowest at long-stay locations. Discrete choice modeling indicated that habitat selection by migrating whip-poor-wills differs depending on the scale and migration strategy. For example, at the 5-km scale birds avoided urban cover at long-stay locations and selected closed forest cover at short-stay locations. We suggest that whip-poor-wills may use land cover cues at large spatial scales, like 5-km, to influence rush or stay tactics during migration.

Methods
From 2018-2020, we captured breeding whip-poor-wills from three study sites in Massachusetts and programmed GPS tags to collect data during fall and spring migration periods. Across 19 individual males (nine of them with repeated years of data), GPS tags collected 479 locations, where 30% were classified as flyover points, 33% as short-stays, and 37% as long-stay locations.

Data processing

We filtered and retained migration data points when loggers connected to ≥ 4 satellites and points had dilution of precision values < 5 to ensure a 3D fix of the location (Forrest et al. 2022, Bakermans et al. 2022). Using 30-m USGS DEM (digital elevation model; http://ned.usgs.gov) data, we generated the altitude of each point by converting the GPS tags’ altitude to altitude above sea level and then subtracted the local elevation (from the DEM) from the bird’s altitude (A. Korpach, pers. communication). Next, we classified migration points based on altitude and number of points at a single location as either flyover, short-stay, or long-stay. Long-stays were locations with ≥ 2 GPS points within the same vicinity (i.e., < 10 km). Short-stay and flyovers consisted of one GPS point at a single location. We differentiated short-stay versus flyover points by altitude based on the altitudes of birds at long-stay locations (mean = 17 m, range = 121 m). Short-stays were locations with elevations < 100 m (mean = 15 m), and flyover locations had an altitude ≥ 100 m above the ground (mean = 800 m).

Hotspot Analyses

To identify areas of high or low use during migration, we ran an optimized hotspot analysis in ArcGIS 10.8.2 to identify statistically significant spatial clusters of high (hotspot) and low values (coldspot) of migration locations using the Getis-Ord Gi* statistic (Sussman et al. 2019). This tool can “aggregate data, identify an appropriate scale of analysis, and correct for both multiple testing and spatial dependence” (ESRI 2021). 

Land cover classification

We used ArcGIS and quantified land cover types from 2019 data using the 100-m Copernicus Global Land Service layer (Buchhorn et al. 2020). Land cover types were classified as (a) closed forest, (b) open forest, (c) shrubland, (d) herbaceous vegetation (hereafter, grassland), (e) herbaceous wetland, (f) cropland, (g) bare, (h) fresh- or saltwater, and (i) developed land (Buchhorn et al. 2020). Using the geoprocessing features of ArcMap, we quantified land cover at 5-km and 1-km circle at an actual migration location (i.e., used) and random locations (i.e., available).

Habitat selection

We used discrete choice modeling to determine habitat selection of Eastern whip-poor-will during migration. Discrete choice models examine the probability that an individual chooses a location based on a choice set of alternative available locations (Cooper and Millspaugh 1999). Choice sets included one used location based on the GPS fix and ten available locations. We constructed separate models for each type of migration point (i.e., flyover, short-stay, and long-stay) and spatial scale (i.e., 1 km and 5 km) with individual as a random effect. We used package jagsUI (Kellner 2021) with the software JAGS 4.3.1 (Plummer 2003).

尽管微型数据记录仪（GPS data logger）为鸟类迁徙研究带来了全新视角，但学界对其基础迁徙模式的认知仍存在不足，对夜鹰类（nightjars）而言尤为如此。本研究借助GPS数据记录仪，针对东部鞭夜鹰（Eastern whip-poor-will，学名：Antrostomus vociferus）的迁徙生态学展开研究，涵盖三种迁徙策略：飞越型、短暂停留型与长期停留型。研究人员记录了该物种的迁徙活动，开展热点区域分析，量化了使用位点与可用位点周边1千米及5千米缓冲区内的土地覆盖类型，并对迁徙过程中的栖息地选择进行建模分析。

2018至2020年间，研究人员在马萨诸塞州的三个研究区域捕获了繁殖期的东部鞭夜鹰，并为其安装GPS标签，设定在秋季和春季迁徙期间收集数据。在19只雄性个体（其中9只提供了跨年度的追踪数据）中，GPS标签共采集到479个定位位点，其中30%被归类为飞越位点、33%为短暂停留位点、37%为长期停留位点。研究发现，不同个体间以及不同年度之间，东部鞭夜鹰的迁徙时长、迁徙路线以及中途停留位点均存在季节性灵活性差异。分析结果显示，墨西哥塔毛利帕斯山脉（Sierra de Tamaulipas）是秋季和春季迁徙过程中的热点集群区域。在5千米尺度下，不同类型位点的土地覆盖类型存在显著差异：飞越型、短暂停留型与长期停留型位点的封闭林覆盖占比依次升高，农田覆盖占比依次降低；而长期停留位点的城市用地占比最低。研究推测，东部鞭夜鹰可能会借助5千米级别的大空间尺度土地覆盖信号，来调整迁徙过程中的快速飞越或停留策略。

## 研究方法
2018至2020年间，研究人员在马萨诸塞州的三个研究区域捕获了繁殖期的东部鞭夜鹰，并为其安装GPS标签，设定在秋季和春季迁徙期间收集数据。在19只雄性个体（其中9只提供了跨年度追踪数据）中，GPS标签共采集到479个定位位点，其中30%被归类为飞越位点、33%为短暂停留位点、37%为长期停留位点。

### 数据处理
研究人员对迁徙定位数据进行筛选与保留：仅保留数据记录仪连接卫星数≥4颗、位置精度稀释因子（dilution of precision, DOP）<5的位点，以确保获得三维定位结果（Forrest等，2022；Bakermans等，2022）。借助30米分辨率的美国地质调查局数字高程模型（USGS DEM, digital elevation model；http://ned.usgs.gov）数据，研究人员将GPS标签记录的海拔转换为海平面海拔后，减去该位点的当地地形高程（源自DEM数据），从而得到鸟类的实际飞行高度（A. Korpach，私人通信）。随后，研究人员根据单个位点的GPS点数与鸟类飞行高度，将迁徙定位位点划分为飞越型、短暂停留型与长期停留型三类。长期停留位点指同一区域内（即距离<10千米）存在≥2个GPS定位点的位点；短暂停留与飞越位点均仅包含单个GPS定位点。研究人员以长期停留位点的飞行高度为基准（平均值为17米，范围为121米），区分短暂停留与飞越位点：短暂停留位点的地面海拔<100米（平均值为15米），飞越位点的地面海拔≥100米（平均值为800米）。

### 热点区域分析
为识别迁徙过程中的高频与低频使用区域，研究人员借助ArcGIS 10.8.2开展优化型热点区域分析，利用Getis-Ord Gi*统计量（Sussman等，2019）识别迁徙位点的高值（热点）与低值（冷点）空间集群，且结果具有统计学显著性。该工具可实现"数据聚合、确定适宜的分析尺度，同时校正多重检验与空间依赖性偏差"（ESRI，2021）。

### 土地覆盖分类
研究人员借助ArcGIS平台，基于2019年的100米分辨率哥白尼全球陆地服务图层（Copernicus Global Land Service layer；Buchhorn等，2020）量化土地覆盖类型。土地覆盖类型共分为9类：(a) 封闭林、(b) 开放林、(c) 灌丛、(d) 草本植被（下文简称草地）、(e) 草本湿地、(f) 农田、(g) 裸地、(h) 淡水或咸水水域、(i) 开发用地（Buchhorn等，2020）。利用ArcMap的地理处理工具，研究人员分别量化了实际迁徙位点（即使用位点）与随机生成的可用位点周边5千米和1千米圆形缓冲区内的土地覆盖占比。

### 栖息地选择分析
研究人员采用离散选择模型（discrete choice modeling）分析东部鞭夜鹰迁徙过程中的栖息地选择偏好。离散选择模型通过构建可供选择的可用位点集，计算个体选择某一位点的概率（Cooper与Millspaugh，1999）。本研究的选择集包含1个基于GPS定位的实际使用位点与10个随机生成的可用位点。研究针对三种迁徙位点类型（飞越型、短暂停留型、长期停留型）以及两种空间尺度（1千米、5千米）分别构建模型，并将个体作为随机效应项。分析采用JAGS 4.3.1软件与jagsUI工具包（Kellner，2021）完成。