Appalachian-breeding Vermivora chrysoptera (Golden-winged Warbler) occur at very low densities in mid-elevation forests and agroforestry systems throughout the Andes and isolated massifs of northern Colombia

Vermivora chrysoptera (Golden-winged Warbler) is one of the most striking but also one of the most vulnerable migratory warblers on Earth. To understand its non-breeding ecology in northern South America, where the steeply declining Appalachian breeding population is known to spend the non-breeding season, we conducted 2,534 surveys between 2018 and 2023 in isolated massifs and the Andes of Colombia. Our goal was to evaluate the distribution, detection rates, elevational range, habitat preferences and microhabitat associations of the species. We obtained 54 detections, with a very low detection rate (2.13%) compared to rates reported from Central America (~10%). Detections covered a wide range of elevations (634 – 2,551 m a.s.l.), with support for a mid-elevation peak at around 1,750 m a.s.l. Importantly, detection elevation was negatively correlated with latitude such that peak abundances occurred at higher elevations closer to the equator. Most detections were in mature or secondary forests (72%), but birds also occurred along forest edges (17%) and in agroforestry systems (11%), such as shade-grown coffee. We found weak evidence for a female preference for lower elevations, as well as more frequent use of agroforestry systems relative to males. Logistic regressions provided some support for a preference for forest over agriculture in both sexes, as well as a higher detection probability at points with intermediate canopy height and canopy cover. Our results highlight the low density at which the species occurs across the northern Andes of Colombia and other isolated massifs. The association with forested habitats and the relatively narrow elevation range occupied by most birds, implies that focusing conservation efforts on protecting or restoring forests at elevations where detection rates peaked would likely be an effective strategy for ensuring that non-breeding habitat remains for Appalachian-breeding Vermivora chrysoptera.

Methods
Distribution and timing of surveys

We conducted surveys for Vermivora chrysoptera in Colombia during five nonbreeding seasons in 2018 – 2023 across a broad elevation gradient (350 – 2,650 m a.s.l.). Survey locations were established in six biogeographic regions: Magdalena Valley – departments of Santander and Tolima; Cauca Valley – departments of Caldas, Quindío, Antioquia, Valle del Cauca and Risaralda; Sierra Nevada de Santa Marta – department of Magdalena; Pacific slope – Antioquia, Chocó, Risaralda and Valle del Cauca; Serranía de Perijá – Cesar; and Orinoco-Amazonas – Boyacá (Figure 1). Locations within biogeographic regions were selected to maximize coverage of likely nonbreeding areas, however, certain regions could not be surveyed due Colombia´s internal conflict, including parts of northern Antioquia and eastern Norte de Santander. In any given season, surveys were conducted between November 10 and March 20 in order to avoid the main post-breeding and pre-breeding migration periods. Not all regions were surveyed in each year.

At each of 101 locations, 8 – 12 point count stations were established, resulting in 1,324 unique points (Table 1), with each point separated by at least 250 m from all the other points (except in 2020, when intermediate points were placed at locations with no detections to ensure that birds had not been missed between points). Within a season, individual points were surveyed once (647 points) or on at least two separate occasions (677 points). Points surveyed in more than one season (105 points), were treated as unique data points for each season, as they represented independent habitat selection processes by birds. Surveys were distributed among six different landcovers: mature forest, secondary forest, riparian forest, forest edge, agroforestry systems and live fences typically bordering cattle pastures (for detailed descriptions see Habitat Type below; see Table 3 for sample size by habitat). These were the most representative landcovers within our study region, however, points were not established in open, treeless, cattle pastures (the most extensive landcover), given that V. chrysoptera does not occur where there is a near or complete absence of trees, vines, hanging dead leaves and epiphytes (Bennett et al. 2016).

In two regions, a stratified survey design was implemented over two nonbreeding seasons in order to examine variation in elevation and habitat use: 1) in Santander, including the municipalities of Tona and Bucaramanga and 2) in Caldas, primarily Manizales municipality. In each region, points were divided nearly equally across six 250 m-wide elevation bands, spanning elevations from 1,000 to 2,500 m a.s.l., and also among landcover types. Each point was surveyed on three occasions during a given nonbreeding season.

Point-count protocol

Point counts had a duration of 10 minutes and followed the protocol tested and adopted by Bennett et al. (2019a) to maximize the detectability of V. chrysoptera. During each point, we broadcast a five-minute recording of V. chrysoptera songs and calls (periods A: 0 – 2.5 min and B: 2.5 – 5 min), immediately followed by a five-minute reproduction of an owl mobbing sequence including the vocalizations of Glaucidium brasilianum (Ferruginous Pygmy-Owl), Megascops asio (Eastern Screech-Owl) and alarm calls of several migratory landbirds (period C: 5 – 10 min). To broadcast recordings, a portable speaker (Xiaomi Compact Bluetooth Speaker 2) was placed at head height in a tree and speaker volume was set to one level less than the maximum. When a V. chrysoptera was detected during a survey, we recorded the period (A, B or C) of the initial detection, sex (e.g., the face mask and throat of males is black, while in females it is gray and indistinct), and signs of hybridization (e.g., yellow coloration in the breast or clear phenotypes of V. chrysoptera x cyanoptera [Brewster’s or Lawerence’s Warblers]), where they could be reliably determined. Points were surveyed primarily in the morning (82% of points; mean time = 09:47 local time).

At each point count station, we recorded geographic coordinates using a handheld GPS unit, elevation above sea level in meters, and time of day (24-hour format). We also recorded the following vegetation variables: Habitat type – Observers selected the habitat classification that best described the landscape within a 50-m radius around a point: mature forest (canopy > 15 m); secondary forest (canopy < 15 m); forest edge (clear boundaries between forest and other landcovers); agroforestry system (typically shade-grown coffee with canopy cover > 30%); riparian forest (narrow strips of forest bordering watercourses < 50 m wide); live fence (boundaries formed by trees and native vegetation); other (heterogenous agricultural areas that did not fall into the above categories). Canopy cover – estimated to the nearest 10% at four cardinal points 5 m from the center of the point, by looking directly up through an inverted binocular and estimating the percentage of the field of view covered by vegetation; Canopy height – Four trees within 25 m of the point and representative of the average canopy height (i.e., not the tallest tree) were selected and measured using the Measure Height application for smart phones; Hanging dead leaf index - was recorded on a four point scale: 1. No dead leaves; 2. 1-50 dead leaves; 3. 50-500 dead leaves; 4. >500 dead leaves; Epiphyte index - (including bromeliads, lichens and mosses) 1. No epiphytes; 2. 1-30% of trees covered with epiphytes; 3. 30-70%; 4. >70%; Vine index - included only vines in the midstory and upper forest levels: 1. No vines; 2. Few vines, thin, not forming tangles or clusters; 3. Medium vines, thicker, 1-2 vine tangles; 4. Many vines of all sizes, 3+ vine tangles. Values for the hanging dead leaf index, epiphyte index and vine index correspond to an estimate of all dead leaves, epiphytes and vines within a 50-m radius of a point.