Research Paper

INTRODUCTION

Over the past 50 years, populations of birds in North America declined by about 30%, representing a net loss of almost three billion birds (Rosenberg et al. 2019). Shorebirds, grassland birds, and aerial insectivores showed the steepest declines, whereas waterfowl and birds of prey had positive trends (NABCI 2020). Particularly strong declines were found for some species of Nearctic-Neotropical migrant forest-breeding birds (NABCI 2020) that migrate to Central and South America for the non-breeding seasons.

Survival of migratory birds depends upon having suitable conditions and habitat at all stages of their life cycle including during the breeding season, migration periods, and non-breeding stationary periods (Runge et al. 2015). Birds may be limited by different factors at different times of year, but relatively few studies have examined factors at all stages of the full annual cycle (Faaborg et al. 2010), especially during the non-breeding periods (Marra et al. 2015, Imlay and Leonard 2019). Habitat loss and climate change have been considered important drivers of population declines for migratory bird species (Bairlein 2016, Kentie et al. 2018). For example, both factors have been linked to declines in Wood Thrush (Hylocichla mustelina) populations during the breeding season (Rushing et al. 2016a). Deforestation in non-breeding areas has been linked to declines in several Neotropical songbird migrants (La Sorte et al. 2017, Ruiz-Sanchez et al. 2017). Lower habitat quality on non-breeding grounds can also result in carry-over effects of lower bird reproductive fitness on breeding grounds, as found for American Redstarts (Setophaga ruticilla; Rushing et al. 2016b). Similarly, Rockwell et al. (2012) reported lower reproductive success for Kirtland’s Warblers (Setophaga kirtlandii) in years with drier weather in non-breeding stationary areas the preceding winter.

Much of the annual mortality for some migratory bird species occurs during migration (Sillett and Holmes 2002, Klaassen et al. 2014). Migratory birds may be killed by storms (Wiedenfeld and Wiedenfeld 1995), or by predators (Woodworth et al. 2014), or may die from exhaustion on long flights if their energy reserves are reduced (Alerstam 2011). Indirectly, predation risk (Pomeroy et al. 2006), competition (Moore and Yong 1991), or poor-quality stopover habitats (Rushing et al. 2016b) may impact their foraging ability and affect survival. Many migratory birds depart at night after dusk and land before dawn (Alerstam 2009) and may face risks from collisions with human-made structures (Gehring et al. 2009). Although most nocturnal migrants are thought to fly mainly at altitudes of 1–2 km (Gauthreaux 1991), they are still attracted to artificial lights and susceptible to collision mortality at take-off or landing (McLaren et al. 2018). Landing pre-dawn may also make it more challenging to find good foraging habitat in areas where habitat has been heavily altered by humans.

Reductions in availability of suitable stopover habitat could also be a limiting factor on migration (Faaborg et al. 2010). Migratory landbirds spend up to 95% of their migration period at stopover sites (Buler and Dawson 2014), where they refuel and build fat stores for the high energy expenditure of long migratory flights (Alerstam 2011). Stopover habitats during fall migration may be especially important because fall migration includes a new cohort of juveniles (Dokter et al. 2018). If insufficient high-quality habitats are available, juveniles, with less experience or lower social status than adults, may be forced to use suboptimal habitats that may reduce their survival (Yong et al. 1998).

For migrating landbirds, the importance of a stopover site likely depends not only on local habitat, but also on surrounding habitat. Buler et al. (2007) found higher Nearctic-Neotropical migrant landbird density in stopover sites surrounded by more deciduous forests. They measured the influence of forest cover at multiple spatial extents and found the strongest relationship, i.e., the “scale of effect” (Jackson and Fahrig 2015), to be within 5 km. Birds may use the forest cover at the landscape scale as a cue when selecting where to land at dawn, or they may move to areas of higher forest cover after landing. Some species are thought to move relatively little after landing (e.g., Ovenbirds, Seiurus aurocapilla; Buler 2006), whereas others may move several km over the day (Aborn and Moore 1997, Chernetsov 2011). In other situations, limited forest cover in the surrounding landscape may concentrate birds in the remaining forest. For example, Buler and Dawson (2014) found higher landbird density in stopover sites surrounded by developed lands or by water bodies than those surrounded by forest. Similarly, McCabe and Olsen (2015) found migrating birds were more often captured at stopover sites with relatively little woody vegetation in the surrounding landscape.

The type of forest around a site might also influence its use as a stopover site, possibly due to variation in the suitability of different forest types as food sources. Different North American songbird migrants use different types of forests at different times of the year (Blondel et al. 1991, Wilkin et al. 2009, Krikun et al. 2018). McLaren, Smolinsky, Buler (2016, unpublished manuscript) used weather surveillance radar to estimate density of nocturnal songbird migrants departing from sites in southern Ontario, and found that migrant density increased with the amount of deciduous forest, but was unaffected by the amount of coniferous forest, in the surrounding landscape within 5 km. Although the link between forest type and food resource availability during migration has been relatively little studied, Cohen et al. (2014) found that deciduous forest cover in the landscape surrounding stopover sites was correlated with fuel deposition rates of Red-eyed Vireos (Vireo olivaceus) during spring migration.

We aimed to evaluate the influence of the surrounding landscape on the use of forest sites by migrating songbirds in southeastern Ontario, Canada. Specifically, we tested whether the density of migratory songbirds at forested sites during fall migration was related to the proportion of forest cover in the surrounding landscape. For this, we had two competing hypotheses. One possibility was that migrating songbirds may avoid landscapes with relatively little remaining forest, resulting in lower densities in the remaining forest sites in landscapes with a low amount of forest. Alternatively, they may still arrive in these landscapes, but be concentrated in the remaining forest in these landscapes, resulting in higher local densities. In addition to the amount of forest, we tested whether the density of migrating songbirds at stopover sites was influenced by the proportion of coniferous forest in the surrounding forest. Although not part of our original study design, as a post-hoc analysis to match previous studies, we also examined the relationship with the proportion of non-coniferous forest (i.e., deciduous and mixed forest) in the surrounding landscape. We analyzed the total number of nocturnally migrating songbirds as well as numbers of individuals of several distinct species groups.

METHODS

We assessed the influence of two landscape variables, the percent of the surrounding forest in the landscape and the percent of coniferous forest in that forest, on the average number of calling birds per minute during the day (as an index of bird density) at 37 bird sampling sites. We selected our sample sites to maximize variation in the two landscape variables of interest (% forest in landscape and % conifers in forest), while minimizing the correlation between them and spatial autocorrelation among sites. We first modeled the number of calling birds at a site on each landscape variable, measured within multiple spatial extents around the site, to find the scales of effect of the two landscape variables. We then created a model for each landscape variable at its maximum effect, including sample site, date, and time as covariates. We also tested for the effect of ratio of artificial to natural light as a potential confounding variable.

Selection of sampling sites

We selected sites to maximize variation in the two landscape variables of interest (% forest in landscape and % conifers in forest) and to minimize the correlation between them. To do this, we used a CanVec land cover map of southeastern Ontario to create a spatial dataset with the following information for each 30 x 30-m pixel: (i) whether the area within 100 m of the pixel was entirely forested, i.e., a “forested site,” (ii) the percent of forest in the surrounding landscape within 1 km and 5 km of the pixel, and (iii) the percent of coniferous forest in that forest within 1 km and 5 km of the pixel. We then selected a set of forested sites at least 5 km apart that maximized variation in % forest in landscape and % conifers in forest at each of these two scales in the surrounding landscape.

Our spatial data were based on the vector-based CanVec series land features from Natural Resources Canada (geographic coordinate system: NAD 1983 CSRS; scale 1:50,000), created between 1999 and 2002 with ortho-images from Landsat, and using an EODS land cover classification (Wulder and Nelson 2003). We selected sites from the area within 50 km of Franktown, Ontario (45.0455° N, 76.0617° W), which is the location of the nearest weather radar to Ottawa, chosen to enable future comparisons of our results with data on bird movements from weather radar. In ESRI Inc. ArcMap 10.5 (2016), we extracted polygons from the land-feature file “Wooded Area,” based on the “wood coverage descriptor” attribute, which listed land cover types based on multiple sources (see section “Wooded Area” in CanVec metadata catalog: NVCP 2018). We created two new layers: (i) one “all forests,” which included polygons representing deciduous/broadleaf forests, coniferous forests, and mixed forests, and excluded polygons representing shrubs and wetlands, and (ii) one including just the coniferous forest. The latter did not include forests that were a mixture of deciduous and coniferous trees (mixed forests). This is because identifying mixed forest from satellite imagery has high uncertainty, whereas coniferous forests can be easily identified (Lisein et al. 2015). We then converted these two vector layers to raster layers using 30 x 30-m pixels.

We used the raster layers to determine, for each pixel, the percent of forest cover in the surrounding landscape and the percent of coniferous forest in that surrounding forest. We did this at three spatial scales, 100-m, 1-km, and 5-km radii from each pixel, using a Moving Window Analysis. This is a technique that moves pixel by pixel and calculates a value of interest based on a defined “window” or spatial extent around the pixel. In this case, the desired values are the percent of forest in the window and the percent of conifer within that forest (within the window). As such, we obtained six raster layers, one layer for each of the two variables at each of the three spatial scales.

Using these raster layers, we then selected 41 bird sampling sites in southeastern Ontario, Canada, with varying % forest in landscape and % conifers in forest in the surrounding landscape. Potential sites were limited to forested pixels that were entirely surrounded by forested pixels within a 100-m radius. Although we modeled the landscape variables as continuous, for the purpose of site selection, we divided the range of available values for each variable (maximum range for both spatial scales of 20–100% for forest in landscape and 0–70% for conifers in forest) into five levels from lowest to highest (i.e., approximately 10–20% increments for each variable). To maximize the ranges of the landscape variables across sites at a given scale, and to minimize the correlation between them, we aimed to select four sites for each level of % forest in landscape combined with each level of % conifers in forest, leading to a maximum potential for 100 different sites. The range of available values and combinations were more constrained at the spatial scale of 5 km, so we focused on using the available values at the 5-km spatial scale to select sites. For each site selected, we then verified that range and variation in values were maintained at the 1-km spatial scale, and if not, we adjusted our selections to achieve the desired variation at the 1-km spatial scale. To minimize any potential effect of regional gradients in our landscape variables, we attempted to maximize their ranges in each of four geographic quadrants (i.e., NE, SE, SW, NW) centered on Franktown. To do this, we first selected a site in the first quadrant and the first level combination, and then a site in the second quadrant with that same level combination, changing the combination after the four quadrants had been considered at least once. We selected sites by visually examining the map and choosing an arbitrary pixel that met the desired criteria that was at least 5 km from any previously selected site to minimize spatial autocorrelation among sites. The manual selection resulted in 83 sites after we excluded some combinations because they did not occur in a quadrant. We also controlled for the area of open water as it might influence the location of migrating birds (Alerstam and Pettersson 1977, Buler et al. 2007). We removed sites with very high (over 20%) open water within 1 and 5 km of sites, and ensured there was no correlation between open water and our landscape variables of interest, % forest in landscape, and % conifers in forest (Appendix 1). We obtained permissions from landowners to sample birds at 41 of the selected sites. We verified that our landscape variables remained uncorrelated within these 41 sites.

To test whether artificial light, which is known to attract nocturnal migrants (McLaren et al. 2018) could be influencing the results of our study, we developed an artificial light at night (ALAN) metric for each of our sites. We extracted the artificial-to-natural light ratio per km cell from the Light Atlas (Falchi et al. 2016). The values at our study sites ranged from 0.01 to 0.68, where values below one indicate more natural light than artificial light (Appendix 13). We did not detect an effect of the artificial light ratio on the dependent variable (Appendix 15).

Recording birds

We set up a Wildlife Acoustics SM4 autonomous recording unit (ARU) at the center of each of the 41 bird sampling sites to estimate the relative density of migrating songbirds from 16 August to 25 October 2018. ARUs have the advantage that they can sample birds at all sites throughout the season, rather than just once or twice per site on different days through the season as would be the case with human observers. Previous studies (Roark and Gaul 2021) and our own unpublished analyses (personal observation) have found that bird activity at stopover sites estimated using recorders was correlated with that estimated using human surveys on the same days, which supports the use of recorders to measure bird density during fall migration. All recorders were set up between 9 and 22 August, and retrieved between 27 and 30 October. Each ARU was scheduled to record for five minutes every 30 minutes between sunrise and sunset. We recorded only during daytime because we wished to measure on-the-ground site use by nocturnal migrants, which includes most migrating Nearctic-Neotropical forest songbirds (Alerstam 2009). Nighttime recordings would likely include mainly calls of birds migrating overhead. We obtained data from 37 of the original 41 recorders (Fig. 1; Appendix 2), due to human error and/or recorder failures at four sites.

Using the software Raven Pro (Center for Conservation Bioacoustics 2014), we listened to the recordings while watching the spectrogram to count individual calling birds. We estimated the minimum number of different individuals present at each site during each minute, using call characteristics (call pitch, intensity, frequency, and overlap as determined both by listening and by examining the spectrograms) to identify different individuals within each minute sampled (see Appendix 17 for examples). We recognize this represents an index of relative activity, rather than a complete count of birds at the site, as some individuals with similar call frequencies may have been overlooked, while the same individual may sometimes give different call types, and some birds present in the area may not call. We had initially planned to listen to one minute every hour from sunrise to sunset, every day for every site. However, acoustic processing took much longer than expected; thus, we had to reduce our effort, resulting in a variable number of minutes processed across sites (Appendix 2). At each site, we listened to recordings from between 5 and 29 days, evenly distributed across the sampling period, with a mean of 15.2 (±4.7 SD) days per site (Appendix 3). For each day sampled at a given site, we listened to at least one minute from each of the first eight hours recorded, for a mean of 129.1 (±65.2) min per site. The sampled minute for each hour was the first minute with unobstructed sound, i.e., calm wind and little rain.

Our predictor variables of interest were not correlated with the number of minutes processed per site (Appendix 4), the sampled dates (Appendix 3), or the sampled hours (Appendix 5). Furthermore, our dependent variable (the mean number of calling birds per minute) was not correlated with the number of minutes analyzed per site (Appendix 14).

We included birds from the following songbird families in our study, as most of their species are nocturnal migrants: Mimidae (mimids), Parulidae (warblers), Passerellidae (New World sparrows), Regulidae (kinglets), Thraupidae (tanagers), Troglodytidae (wrens), Turdidae (thrushes), Tyrannidae (flycatchers), and Vireonidae (vireos; Appendix 6).

Statistical analyses

To estimate the scale of effect for each of the landscape predictors, we modeled the mean number of calling birds per minute at a site in relation to each landscape variable separately, at each of 11 scales, i.e., 0.1, 0.5, 1, 2, 3, 4, 5, 6, 8, 10, and 15-km radius from the sample site, using a default, Gaussian generalized additive model (GAM; Appendix 7, Appendix 8). We considered the scale of effect to be the scale where the r-squared value was largest (Appendix 9).

To test our predictions about the effects of % forest in landscape and % conifers in forest on bird density at stopover sites, we fitted a negative binomial GAM (R Core Team 2017; mgcv R-package: Wood 2018) for each landscape variable separately, at its scale of effect. We evaluated the effects of the landscape variables in separate models because there was little or no correlation between them (Fig. 2). We included site and date (Appendix 10) as categorical random effects, and time of day (i.e., hours from sunrise; Appendix 11) as smoothed thin-plate splines with generalized cross-validation (GCV). Because the default GAM plots suggested a possible dome-shaped relationship between % forest in landscape and number of calling birds, we modeled this relationship as a smoothed three-knot, thin-plate spline (k = 3), allowing up to one peak in the model. The default GAM plots suggested a linear relationship between % conifers in forest and number of calling birds; therefore, % conifers in forest was modeled as a linear term only. As a post-hoc analysis to match previous studies, we also modeled % non-coniferous forest (i.e., deciduous and mixed forest) using the same scale of effect and model as those for % forest in landscape. From the models, we extracted the predicted number of calling birds per minute at each site after adjusting for site, date, and time of day sampled, and plotted them against the landscape variables. In addition to modeling the total number of calling birds per minute from all nocturnally migrating songbirds, we modeled several distinct nocturnal songbird species groups.

RESULTS

We detected a mean of 0.67 (±1.42 SD) calling nocturnal songbird migrants per minute across 37 sites (Fig. 3), with the mean per site ranging from 0.21 to 1.78 calling birds per minute. The detected songbird migrants belonged to nine passerine families (Appendix 6).

Prediction 1: percent forest in landscape

The strongest relationship between % forest in landscape and number of calling birds (R² = 0.21, P = 0.042) was for % forest in landscape measured within 2 km of the sites (the scale of effect; Appendix 7, Appendix 9). The relationship between the number of calling birds per minute and % forest in landscape at this scale peaked in the middle, with the maximum number of calling birds per minute at 57% of the forest in landscape (Fig. 3). This maximum number of calling birds was 1.98 times the number of calling birds at the minimum forest cover (23%) and 2.24 times the number of calling birds at the maximum forest cover (94%).

For comparison with some previous studies that considered just deciduous forest, we also examined the relationship with % non-coniferous forest in landscape. We found that the mean number of calling birds weakly increased linearly with the amount of non-coniferous forest (R² = 0.15, P = 0.059; Fig. 4).

Prediction 2: percent conifers in forest

The scale of effect for % conifers in forest was 6 km (R² = 0.21, P = 0.003; Appendix 8, Appendix 9). The mean number of calling birds declined with increasing % conifers in forest. There was a decrease of 58% in the number of calling birds with a 50% increase in the % conifers in forest (Fig. 5).

To confirm that any correlation between the two landscape variables, each at their scale of effect, did not influence our conclusions, we also modeled both variables together and found similar relationships (Appendix 12, Appendix 16), although the strength of the patterns was slightly weaker, possibly because of a slight residual correlation between % forest in landscape and % conifer in forest.

To test whether the effects of landscape variables varied among species groups, we modeled six species groups that had over 100 calling individuals (Appendix 6), namely flycatchers, kinglets, sparrows, thrushes, vireos, and warblers. Although weaker, their trends were similar to those from analyzing calls of all nocturnal songbird migrants (Appendix 16).

DISCUSSION

Our study suggests that the density of songbird migrants during fall migration is highest at stopover sites surrounded by an intermediate amount of forest, i.e., ~55% of the landscape within 2 km of the site covered in forest. The species groups that we examined either had a similar peak, or were not affected by forest amount. We had hypothesized either (i) a positive effect of forest cover on bird density if birds selected landscapes with more remaining forests in the surrounding landscape, or (ii) a negative effect of forest cover on bird density if limited forest cover in the surrounding landscape concentrated birds in the remaining forests. It is possible that the combination of these could produce a peaked relationship if in the lower half of the curve, i.e., at low forest cover, migrating birds avoid landing in forested sites with low surrounding forest amount. This would lead to an increase in bird density with increasing forest in the surrounding sites. This effect might not continue into the upper half of the curve if all sites with more than 50–55% of surrounding forest are equally attractive to migrating birds. In this case, in the upper half of the curve, i.e., at high forest cover, a decrease in bird density with increasing forest could result from a given number of landing birds spread through more forest, reducing the density at any given sample site in the landscape.

We note that this is not the only possible explanation for the peaked relationship between bird density and percent in the surrounding landscape. For example, the amount of forest edge tends to be highest in landscapes with an intermediate amount of forest (Fahrig 2003). If birds seek out forest edges to forage (Terraube et al. 2016), where specific insect prey may be more abundant (De Carvalho Guimarães et al. 2014, González et al. 2017) or insect prey tend to be more diverse (De Carvalho Guimarães et al. 2014, González et al. 2017, Pinksen et al. 2021, Ruchin et al. 2023), this might lead to greater densities in landscapes with more edge, i.e., with intermediate forest cover. We cannot directly test this hypothesis because our observations were all made at least 100 m from an edge.

Few previous studies have looked at the influence of landscape attributes on migrants. Buler et al. (2007) found a positive relationship between landbird migrant density on forested survey transects and the proportion of hardwood forests in the surrounding landscape over a similar range of forest proportions to our study. In contrast, Buler and Dawson (2014) and McCabe and Olsen (2015) found higher densities of migrant landbirds at stopover sites with more development or agriculture in the surrounding landscape, suggesting birds may have been concentrated in the remaining forest. McLaren, Smolinsky, Buler (2016, unpublished manuscript) also found a positive relationship between the number of migrating birds departing from a site at dusk, as estimated by weather surveillance radars, and hardwood forest cover in the surrounding landscape. However, their study was not directly comparable to ours, because the authors measured average density over grid cells of 1 km² that were not entirely forested whereas our local sites of 100-m radii were entirely forested. Although both Buler et al. (2007) and McLaren, Smolinsky, Buler (2016, unpublished manuscript) measured the proportion of the landscape that was hardwood forest, while we measured total forest of all types, their “hardwood forest” included mixed forest and wetland (i.e., containing up to 75% conifers). We calculated a metric of non-conifer in forest post-hoc to better compare with their “hardwood forest,” and we indeed found a weakly positive relationship across species groups. Through weather surveillance radars in a large spatial and temporal study, Guo et al. (2023) found high bird density both at an intermediate amount of deciduous forest in the landscape (within 5 km of radars) when there was a high amount of regional deciduous forest (within 100 km) and at a high amount of deciduous forest in the landscape when there was a low amount of regional deciduous forest. Although not directly comparable to our study, our 50-km study area was about 47% forested, and we similarly found a slight increase in bird density with more non-conifer (“hardwood”) forest in the landscape, yet a peak in bird density at an intermediate amount of forest cover in the landscape, possibly also due to less forest edge or less bird congregation at higher forest amount.

We suggest that future studies evaluate the effects of forest cover along other bird migration routes. Most stopover habitat studies in North America have focused on eastern migration routes (e.g., Mississippi coast: Buler et al. 2007, Buler and Moore 2011, and Northeast America: Buler and Dawson 2014; McLaren, Smolinsky, Buler 2016, unpublished manuscript), where land cover is more forested than central migration routes (Smith et al. 2012). Along the central routes, stopover sites with more agriculture and less forest in the surrounding landscape are more common (Smith et al. 2012). We speculate that whenever the range of surrounding percentage forest is below 50%, an occurrence likely more common in central North America, the influence of surrounding forest on bird density would be positive instead of peaking, as all areas would fall in the lower half of the relationship in Fig. 3. Bird surveys using weather surveillance radars (e.g., McLaren, Smolinsky, Buler 2016, unpublished manuscript), could be used to study larger geographic areas encompassing multiple migration routes.

We found a lower density of nocturnal songbird migrants during fall migration at stopover sites where the proportion of the surrounding conifers in forest was higher. This decrease was consistent across species groups that we studied. This supports our hypothesis that deciduous forests are more attractive to migrants, possibly because they contain more prey insects than coniferous forests (Blondel et al. 1991, Wilkin et al. 2009, Cohen et al. 2014). At least some of the coniferous forests in our region are monoculture plantations, which may be particularly poor areas for migrants to forage.

In apparent contrast to our result, McLaren, Smolinsky, Buler (2016, unpublished manuscript) found no effect of the amount of coniferous forest in the surrounding landscape on landbird migrant density, using land cover data from the Agriculture and Agri-Food Canada (AAFC) Annual Crop Inventory. However, McLaren, Smolinsky, Buler (2016, unpublished manuscript) measured the total area of coniferous forest in the landscape surrounding sampling sites, while we measured the proportion of the surrounding forest in coniferous forest. We speculate that the proportion of conifers in forest did not vary sufficiently across the sites in McLaren, Smolinsky, Buler (2016, unpublished manuscript) for them to observe a response. In contrast, our site selection was designed to maximize the variation across sites in percent conifers in forest. McLaren et al. (2018) was a larger scale study and found a decrease in migrant density with more coniferous forest within 2 km, consistent with our results although their conifer metric only considered patches with over 75% conifers. Interestingly, Cohen et al. (2021) found a positive effect of all forest types within 5 km of 1 km² cells on bird density from weather radars in the southeastern U.S., the strongest being evergreen forests, and a stronger effect of coast proximity in the spring than fall. We speculate that coast proximity may be prioritized over forest type by songbird migrants, especially when tired during long-distance migration, such that any forest type has a positive association with stopover. We may have found a negative effect of conifer amount because our study site was far from the coast, or used as a stopover earlier during migration (northern range for many migrants), i.e., when medium- to long-distance songbird migrants have less need to land from tiredness (Buler and Moore 2011) and can afford to be more selective.

We found a scale of effect of 2 km for the percent of the surrounding forest in landscape and of 6 km for the percent of conifers in forest. These scales of effect were far from the lowest and highest scales examined of 0.1 km and 15 km; therefore, it is likely that we did not miss the relevant scales of effect (Jackson and Fahrig 2015). Buler et al. (2007) found a scale of effect of 5 km for the proportion of deciduous forest cover in the landscape surrounding transect surveys of migrant birds. McCabe and Olsen (2015) measured the effect of the ratio between landscape-scale and local-scale proportion of vegetation on capture rate of migrating birds. They found that the strongest influence of this ratio occurred when the landscape-scale vegetation proportion was measured within 4 km of the bird-capture sites. Overall, these and our observed scales of effect indicate that effects of landscape structure on migrating birds occur within several kilometers of a potential stopover site.

We speculate that these scales of effect are related to the behavior of migrating birds as they evaluate or explore stopover sites based on landscape structure before or after landing, or both. Migrant species vary in after-landing dispersal distances, some exploring up to several km away from the initial landing site (e.g., Aborn and Moore 1997, Buler 2006, Chernetsov 2011). Many may evaluate the landscape from above, land, and readjust if forest cover or food availability does not meet expectations. We do not know the extent to which migrants disperse on the ground or can identify more dense forests from above. We encourage further evaluations of both in-air and on-ground stopover movements at larger spatial scales to understand when and why different scales of effect are relevant to the influence of landscape structure on bird density.

We acknowledge that the percent forest in landscape and percent conifers in forest are not the only cues used by birds when deciding where to stop during migration. Neotropical migrants show seasonal plasticity and their behavior may be influenced by developed land cover types (Zuckerberg et al. 2016). Migratory landbirds near urban areas, distracted by artificial lights, potentially land in sites with suboptimal forest cover (McLaren et al. 2018). Similarly, migratory landbirds near coastal stopovers may be less influenced by surrounding forest cover and more by a need to land as a result of depleted energy reserves or adverse weather conditions (Buler and Moore 2011).

Our results suggest that to conserve night-migratory songbirds, we should ensure the availability of sufficient stopover sites that are surrounded by about 50% forest or more that is predominantly deciduous, within about 5 km of the site. Meeting the target of 50% forest will become more challenging over time, because many forests continue to be lost on account of agriculture or urban expansion. Focusing reforestation efforts on mixed, species-rich forests, rather than conifer plantations (Parker et al. 2001), should create landscapes more suitable as stopover sites for migrating passerines.

LITERATURE CITED

Aborn, D. A., and F. R. Moore. 1997. Pattern of movement by Summer Tanagers (Piranga rubra) during migratory stopover: a telemetry study. Behavior 134:1077-1100. https://doi.org/10.1163/156853997X00412

Alerstam, T. 2009. Flight by night or day? Optimal daily timing of bird migration. Journal of Theoretical Biology 258:530-536. https://doi.org/10.1016/j.jtbi.2009.01.020

Alerstam, T. 2011. Optimal bird migration revisited. Journal of Ornithology 152:5-23. https://doi.org/10.1007/s10336-011-0694-1

Alerstam, T., and S.-G. Pettersson. 1977. Why do migrating birds fly along coastlines? Journal of Theoretical Biology 65:699-712. https://doi.org/10.1016/0022-5193(77)90016-9

Bairlein, F. 2016. Migratory birds under threat. Science 354:547-548. https://doi.org/10.1126/science.aah6647

Blondel, J., A. Dervieux, M. Maistre, and P. Perret. 1991. Feeding ecology and life history variation of the Blue Tit in Mediterranean deciduous and sclerophyllous habitats. Oecologia 88:9-14. https://doi.org/10.1007/BF00328397

Buler, J. J. 2006. Understanding habitat use by landbirds during migration along the Mississippi gulf coast using a scale-dependent approach. Dissertation. University of Southern Mississippi, Hattiesburg, Mississippi, USA.

Buler, J. J., and D. K. Dawson. 2014. Radar analysis of fall bird migration stopover sites in the northeastern U.S. Condor 116:357-370. https://doi.org/10.1650/CONDOR-13-162.1

Buler, J. J., and F. R. Moore. 2011. Migrant-habitat relationships during stopover along an ecological barrier: extrinsic constraints and conservation implications. Journal of Ornithology 152:101-112. https://doi.org/10.1007/s10336-010-0640-7

Buler, J. J., F. R. Moore, and S. Woltmann. 2007. A multi-scale examination of stopover habitat use by birds. Ecology 88:1789-1802. https://doi.org/10.1890/06-1871.1

Center for Conservation Bioacoustics. 2014. Raven Pro: interactive sound analysis software (version 1.5). The Cornell Lab of Ornithology, Ithaca, New York, USA.

Chernetsov, N. 2011. Daytime movements of nocturnal migrants at stopover between two nearby capture sites. Journal of Ornithology 152:1007-1011. https://doi.org/10.1007/s10336-011-0690-5

Cohen, E. B., K. G. Horton, P. P. Marra, H. L. Clipp, A. Farnsworth, J. A. Smolinsky, D. Sheldon, and J. J. Buler. 2021. A place to land: spatiotemporal drivers of stopover habitat use by migrating birds. Ecology Letters 24:38-49. https://doi.org/10.1111/ele.13618

Cohen, E. B., S. M. Pearson, and F. R. Moore. 2014. Effects of landscape composition and configuration on migrating songbirds: inference from an individual-based model. Ecological Applications 24:169-180. https://doi.org/10.1890/12-1867.1

De Carvalho Guimarães, C. D., J. P. Rodrigues Viana, and T. Cornelissen. 2014. A meta-analysis of the effects of fragmentation on herbivorous insects. Environmental Entomology 43(3):537-545. https://doi.org/10.1603/EN13190

Dokter, A. M., A. Farnsworth, D. Fink, V. Ruiz-Gutierrez, W. M. Hochachka, F. A. La Sorte, O. J. Robinson, K. V. Rosenberg, and S. Kelling. 2018. Seasonal abundance and survival of North America’s migratory avifauna determined by weather radar. Nature Ecology and Evolution 2:1603-1609. https://doi.org/10.1038/s41559-018-0666-4

Esri Inc. 2016. ArcGIS Desktop - ArcMap (Version 10.5). Esri Inc, Toronto, Ontario, Canada. https://www.esri.com/en-us/arcgis/products/arcgis-desktop/resources

Faaborg, J., R. Holmes, A. Anders, K. Bildstein, K. Dugger, S. Gauthreaux, P. Heglund, K. Hobson, A. Jahn, D. Johnson, S. Latta, D. Levey, P. Marra, C. Merkord, E. Nol, S. Rothstein, T. Sherry, T. Sillett, F. Thompson, and N. Warnock. 2010. Recent advances in understanding migration systems of New World land birds. Ecological Monographs 80:3-48. https://doi.org/10.1890/09-0395.1

Fahrig, L. 2003. Effects of habitat fragmentation on biodiversity. Annual Review of Ecology, Evolution, and Systematics 34:487-515. https://doi.org/10.1146/annurev.ecolsys.34.011802.132419

Falchi, F., P. Cinzano, D. Duriscoe, C. C. M. Kyba, C. D. Elvidge, K. Baugh, B. Portnov, N. A. Rybnikova, and R. Furgoni. 2016. Supplement to: The New World atlas of artificial night sky brightness. GFZ Data Services. http://doi.org/10.5880/GFZ.1.4.2016.001

Gauthreaux, S. A. 1991. The flight behavior of migrating birds in changing wind fields: radar and visual analyses. American Zoologist 31:187-204. https://doi.org/10.1093/icb/31.1.187

Gehring, J., P. Kerlinger, and A. M. Manville. 2009. Communication towers, lights, and birds: successful methods of reducing the frequency of avian collisions. Ecological Applications 19:505-514. https://doi.org/10.1890/07-1708.1

González, E., A. Salvo, and G. Valladares. 2017. Natural vegetation cover in the landscape and edge effects: differential responses of insect orders in a fragmented forest. Insect Science 24:891-901. https://doi.org/10.1111/1744-7917.12377

Guo, F., J. J. Buler, J. A. Smolinsky, and D. S. Wilcove. 2023. Autumn stopover hotspots and multiscale habitat associations of migratory landbirds in the eastern United States. Proceedings of the National Academy of Sciences 120(3):e2203511120. https://doi.org/10.1073/pnas.2203511120

Imlay, T. L., and M. L. Leonard. 2019. A review of the threats to adult survival for swallows (Family: Hirundinidae). Bird Study 66:251-263. https://doi.org/10.1080/00063657.2019.1655527

Jackson, H. B., and L. Fahrig. 2015. Are ecologists conducting research at the optimal scale? Global Ecology and Biogeography 24:52-63. https://doi.org/10.1111/geb.12233

Kentie, R., T. Coulson, J. C. E. W. Hooijmeijer, R. A. Howison, A. H. J. Loonstra, M. A. Verhoeven, C. Both, and T. Piersma. 2018. Warming springs and habitat alteration interact to impact timing of breeding and population dynamics in a migratory bird. Global Change Biology 24:5292-5303. https://doi.org/10.1111/gcb.14406

Klaassen, R. H. G., M. Hake, R. Strandberg, B. J. Koks, C. Trierweiler, K.-M. Exo, F. Bairlein, and T. Alerstam. 2014. When and where does mortality occur in migratory birds? Direct evidence from long-term satellite tracking of raptors. Journal of Animal Ecology 83:176-184. https://doi.org/10.1111/1365-2656.12135

Krikun, R. G., J. L. McCune, E. M. Bayne, and D. T. T. Flockhart. 2018. Breeding habitat characteristics of Canada Warblers in central Alberta. Forestry Chronicle 94:230-239.

La Sorte, F. A., D. Fink, P. J. Blancher, A. D. Rodewald, V. Ruiz‐Gutierrez, K. V. Rosenberg, W. M. Hochachka, P. H. Verburg, and S. Kelling. 2017. Global change and the distributional dynamics of migratory bird populations wintering in Central America. Global Change Biology 23:5284-5296. https://doi.org/10.1111/gcb.13794

Lisein, J., A. Michez, H. Claessens, and P. Lejeune. 2015. Discrimination of deciduous tree species from time series of unmanned aerial system imagery. PLoS ONE 10:e0141006. https://doi.org/10.1371/journal.pone.0141006

Marra, P. P., E. B. Cohen, S. R. Loss, J. E. Rutter, and C. M. Tonra. 2015. A call for full annual cycle research in animal ecology. Biology Letters 11:20150552. https://doi.org/10.1098/rsbl.2015.0552

McCabe, J. D., and B. J. Olsen. 2015. Landscape-scale habitat availability, and not local geography, predicts migratory landbird stopover across the Gulf of Maine. Journal of Avian Biology 46:395-405. https://doi.org/10.1111/jav.00598

McLaren, J. D., J. J. Buler, T. Schreckengost, J. A. Smolinsky, M. Boone, E. Emiel van Loon, D. K. Dawson, and E. L. Walters. 2018. Artificial light at night confounds broad-scale habitat use by migrating birds. Ecology Letters 21:356-364. https://doi.org/10.1111/ele.12902

Moore, F. R., and W. Yong. 1991. Evidence of food-based competition among passerine migrants during stopover. Behavioral Ecology and Sociobiology 28:85-90. https://doi.org/10.1007/BF00180984

North American Bird Conservation Initiative (NABCI). 2020. The state of Canada’s birds 2019. NABCI, Gatineau, Québec, Canada. http://nabci.net/resources/state-of-canadas-birds-2019/

NVCP. 2018. ISO 19110 feature catalogue vers. 1.1.0. ftp://ftp.geogratis.gc.ca/pub/nrcan_rncan/vector/canvec/doc/CanVec_Catalogue_50K_Land/SRDB_EXPL_ESSIM_50K_Land-sd-en.html

Parker, W. C., D. C. Dey, S. G. Newmaster, K. A. Elliott, and E. Boysen. 2001. Managing succession in conifer plantations: converting young Red Pine (Pinus resinosa Ait.) plantations to native forest types by thinning and underplanting. Forestry Chronicle 77:721-734. https://doi.org/10.5558/tfc77721-4

Pinksen, J., E. R. D. Moise, J. Sircom, J. J. Bowden. 2021. Living on the edge: effects of clear-cut created ecotones on nocturnal macromoth assemblages in the eastern boreal forest, Canada. Forest Ecology and Management 494:119306. https://doi.org/10.1016/j.foreco.2021.119309

Pomeroy, A. C., R. W. Butler, and R. C. Ydenberg. 2006. Experimental evidence that migrants adjust usage at a stopover site to trade off food and danger. Behavioral Ecology 17:1041-1045. https://doi.org/10.1093/beheco/arl043

R Core Team. 2017. R: a language and environment for statistical computing. R Foundation for Statistical Computing, Vienna, Austria.

Roark, E., and W. Gaul. 2021. Monitoring migration timing in remote habitats: assessing the value of extended duration audio recording. Avian Conservation and Ecology 16(1):21. https://doi.org/10.5751/ACE-01885-160121

Rockwell, S. M., C. I. Bocetti, and P. P. Marra. 2012. Carry-over effects of winter climate on spring arrival date and reproductive success in an endangered migratory bird, Kirtland’s Warbler (Setophaga kirtlandii). Auk 129:744-752. https://doi.org/10.1525/auk.2012.12003

Rosenberg, K. V., A. M. Dokter, P. J. Blancher, J. R. Sauer, A. C. Smith, P. A. Smith, J. C. Stanton, A. Panjabi, L. Helft, M. Parr, and P. P. Marra. 2019. Decline of the North American avifauna. Science 366:120-124. https://doi.org/10.1126/science.aaw1313

Ruchin, A. B., L. V. Egorov, and A. A. Khapugin. 2023. Edge effects in the distribution of Coleoptera in the forests of the center of the European part of Russia. Insects 14(4):371. https://doi.org/10.3390/insects14040371

Ruiz-Sanchez, A., K. Renton, and R. Rueda-Hernandez. 2017. Winter habitat disturbance influences density and territory size of a Neotropical migratory warbler. Journal of Ornithology 158:63-73. https://doi.org/10.1007/s10336-016-1368-9

Runge, C. A., J. E. M. Watson, S. H. M. Butchart, J. O. Hanson, H. P. Possingham, and R. A. Fuller. 2015. Protected areas and global conservation of migratory birds. Science 350:1255-1257. https://doi.org/10.1126/science.aac9180

Rushing, C. S., P. P. Marra, and M. R. Dudash. 2016b. Winter habitat quality but not long-distance dispersal influences apparent reproductive success in a migratory bird. Ecology 97:1218-1227. https://doi.org/10.1890/15-1259.1

Rushing, C. S., T. B. Ryder, and P. P. Marra. 2016a. Quantifying drivers of population dynamics for a migratory bird throughout the annual cycle. Proceedings of the Royal Society B: Biological Sciences 283:20152846. https://doi.org/10.1098/rspb.2015.2846

Sillett, T. S., and R. T. Holmes. 2002. Variation in survivorship of a migratory songbird throughout its annual cycle. Journal of Animal Ecology 71:296-308. https://doi.org/10.1046/j.1365-2656.2002.00599.x

Smith, W. K., C. C. Cleveland, S. C. Reed, N. L. Miller, and S. W. Running. 2012. Bioenergy potential of the United States constrained by satellite observations of existing productivity. Environmental Science and Technology 46:3536-3544. https://doi.org/10.1021/es203935d

Terraube, J., F. Archaux, M. Deconchat, I. van Halder, H. Jactel, and L. Barbaro. 2016. Forest edges have high conservation value for bird communities in mosaic landscapes. Ecology and Evolution 6:5178-5189. https://doi.org/10.1002/ece3.2273

Wiedenfeld, D. A., and M. G. Wiedenfeld. 1995. Large kill of neotropical migrants by tornado and storm in Louisiana, April 1993. Journal of Field Ornithology 66:70-80.

Wilkin, T., L. King, and B. Sheldon. 2009. Habitat quality, nestling diet, and provisioning behavior in Great Tits Parus major. Journal of Avian Biology 40:135-145. https://doi.org/10.1111/j.1600-048X.2009.04362.x

Wood, S. 2018. mgcv: mixed GAM computation vehicle with automatic smoothness estimation.

Woodworth, B. K., C. M. Francis, and P. D. Taylor. 2014. Inland flights of young Red-eyed Vireos Vireo olivaceus in relation to survival and habitat in a coastal stopover landscape. Journal of Avian Biology 45:387-395. https://doi.org/10.1111/jav.00276

Wulder, M. A., and T. A. Nelson. 2003. EOSD land cover classification legend report. Version 2. Natural Resources Canada, Canadian Forest Service, Pacific Forestry Centre, Victoria, British Columbia, Canada.

Yong, W., D. M. Finch, F. R. Moore, and J. F. Kelly. 1998. Stopover ecology and habitat use of migratory Wilson‛s Warblers. Auk 115:829-842. https://doi.org/10.2307/4089502

Zuckerberg, B., D. Fink, F. A. La Sorte, W. M. Hochachka, and S. Kelling. 2016. Novel seasonal land cover associations for eastern North American forest birds identified through dynamic species distribution modelling. Diversity and Distributions 22:717-730. https://doi.org/10.1111/ddi.12428