Declines in migratory landbird populations require targeted research to identify population-limiting factors acting throughout the annual cycle. Tracking studies provide data on migration timing and speed, which have been used to describe carryover effects of nonbreeding habitat use. To explore how variation in precipitation and forest cover among nonbreeding sites influenced departure dates and migration speed in Vermivora chrysoptera (Golden-winged Warbler), we fitted 69 birds with radio transmitters in Costa Rica and tracked them via the Motus Wildlife Tracking System. Capture sites covered a broad gradient of precipitation and forest cover and were separated by <100 km. The datasets contained herein include departure dates and migration durations for birds tracked via Motus and related data for each individual, including capture location, age, sex, and Motus tag ID. For each capture location in Costa Rica, we extracted precipitation values from Worldclim and estimated forest cover around locations in QGIS (see methods below).

Study regions and sites

Five study regions were selected in central Costa Rica that were representative of the gradient of precipitation and forest cover within the nonbreeding range of V. chrysoptera. Within each of the five study regions, one to three discrete study sites were selected for capturing V. chrysoptera.

Captures and deployment of radio transmitters

V. chrysoptera were captured at each of the study sites between February 26 and March 31 in 2022 and February 24 and March 26 in 2023. Birds were captured using 9 or 12 m long mist nets (30 mm mesh) in combination with conspecific playback of V. chrysoptera song and chip calls to lure birds toward nets at ground-level or elevated 1-2 m using net pole extensions. On capture, birds were fitted with a numbered aluminum band (Porzana Ltd, reporting address www.aselva.co) and a unique color band combination for future resightings. Birds were aged and sexed following the criteria in Pyle (1997). For radio tracking, birds were fitted with 0.26 g NTQB2-1-M coded VHF nanotags (Lotek, Ontario, Canada) registered with Motus. Tags were set up with an interval of 15.9 seconds between pulses in 2022 and 15.7 s in 2023, and had an expected life span of 72-108 days, which was sufficient to cover the pre-migratory and migratory period between late February and May. Tags were attached using a leg-loop harness (Naef-Daenzer 2007) constructed with a 0.5 mm black polyester and polyurethane thread (Stretch Magic). Total harness length was adjusted in line with unflattened wing chord as follows: wing 56-60 mm - 68 mm harness, wing 61-64 mm - 70 mm harness, and wing >65 mm - 72 mm harness. Total tag and harness weight was ~0.30 g, equivalent to 3.0%-3.9% of the mass of birds at capture.

Defining departure dates using Motus automated telemetry data

To determine departure dates, birds were tracked using existing permanent Motus stations (4 stations: Finca Cristina; Caracara; Casa Carballo, Fortuna; Paraiso Quetzal) or through stations mounted temporarily within capture regions (7 stations: La Paz, San Ramon; Rio Macho; Naciente Mero; Jardin Botanico Lankester; San Antonio, Turrialba; Valle Los Quetzales; Finca Galeón). Permanent stations typically consisted of two 9-element Yagi antennas connected to a SensorStation (Cellular Tracking Technologies, New Jersey, U.S.) or a SensorGnome (www.sensorgnome.org), while temporary stations were fitted with one or two 3-element Yagi antennas to maximize local detections. At the end of each study period, detection data were downloaded from Motus and processed to identify individual departure dates following the recommendations in Taylor et al. (2017). We examined individual detection histories using the Motus package (Birds Canada 2024) for Program R (version 4.1.3; R Development Core Team 2017) and identified reliable departure dates using one or more of the following criteria (criteria are listed in order of decreasing confidence): A) Birds with continuous detections at nearby stations, followed by a clear nocturnal departure signal, e.g. a spike in signal strength as illustrated in Taylor et al. (2017) (n = 24); B) Birds with discontinuous or sporadic detections at a “local” station, followed by a clear nocturnal departure signal (n = 4). C. Birds with few or no detections on local stations but with a clear departure signal (n = 9). D. Birds with few or no detections on local stations but a clear nocturnal flyby signal on one or more stations <50 km to the north of the capture site (n = 7); E. Birds with regular daily detections for at least a month but no clear departure signal (n = 2). Twenty-three birds did not meet any of these criteria and were excluded from subsequent analysis.

Modeling the relationship between precipitation, forest cover, and departure date

We evaluated whether precipitation or forest cover was correlated with departure date by running a series of Generalized Linear Models, in which possible effects of age, sex, and year on departure date were controlled for (Turcotte-van de Rydt et al. 2023). For precipitation and forest cover, two variables were assessed for each as described below. We also included elevation in the model set given its relation to occupancy probabilities (Bennett et al. 2019a).

To create precipitation variables, monthly precipitation values were extracted from the WorldClim global precipitation 1 km resolution rasters (Fick and Hijmans 2017) for the capture location of each individual bird. Monthly values were then summed to create two new variables based on the assumption that resource availability (Wolda 1978) may be influenced by either Annual Precipitation (sum of values over 12 months) or by Nonbreeding Precipitation: sum of values between November and March (months in which V. chrysoptera were present at study sites). For forest cover, we estimated the area of forest within a 500 m and 1000 m buffer around individual capture locations using a 2023 forest layer developed specifically for Costa Rica (SNIT 2024, resolution 50 m) in the program QGIS version 3.40 (QGIS Development Team 2024). The layer classified forests into six types, and we selected those relevant to V. chrysoptera – mature forest and secondary forest (Chandler and King 2011).

Estimating migration speed based on detections in the Motus array

To create a variable representative of migration speed, we determined the number of days elapsed between a reliable departure signal from Costa Rica and the first reliable detection in a narrow migration pathway through the US Midwest bounded by 36° and 44° latitude and -88° and -96° longitude (note, our estimate of migration speed includes both flight and stopover phases following Hedenström and Alerstam (1998)). Given that the exact route taken and the number of kilometers flown cannot be directly estimated from detections in the Motus array, the time taken between point A and point B is considered an appropriate proxy for migration speed (Gómez et al. 2017).