Bird-window collisions in the summer breeding season

Study site

We assessed BWCs at Augustana College in northwestern Illinois, USA, for nine weeks of each breeding season (June–early August) from 2009 to 2012. The campus was constructed on 0.65 km² within the bluffs of the Mississippi River (90°33′W, 41°30′N) and located in the Eastern Tallgrass Prairie Bird Conservation Region (Sauer, Fallon & Johnson, 2003). Bluff faces contained moderately disturbed deciduous hardwood forest (‘wooded bluff faces’), and terraces above and below bluffs were similar in structure to grassland savanna with scattered woody trees and shrubs and an open understory of landscaped grass (‘landscaped savanna’). Work was completed at four low-rise commercial buildings: Westerlin Residence Hall (height =3 stories, building footprint = 5,690 m²), Hanson Hall of Science (height = 5 stories, building footprint = 2,432 m²), Thomas Tredway Library (height = 5 stories, building footprint = 1,837 m²), and Swenson Hall of Geosciences (height = 3 stories, building footprint = 745 m²).

Carcass surveys

In 2009 and 2010, we completed carcass surveys at intervals of 1–3 days at Hanson Hall of Science, Swenson Hall of Geosciences, and Thomas Tredway Library. During each survey, a trained fieldworker walked a complete transect around a building’s footprint within a 2-m buffer. This buffer accommodated the perpendicular distance from external walls at which most carcasses are located (95% CI [93–127 cm]; N = 51 carcasses; S Hager, 2013, unpublished data). A bird carcass consisted of a full body, partial carcass, or feather pile (Hager & Cosentino, 2014). We assumed high average detection probability of bird carcasses (0.88, SE = 0.01) based on Hager et al. (2013).

In 2011 and 2012, we assessed the daily pattern of BWCs by surveying for carcasses for four consecutive days per week at Hanson Hall of Science, Swenson Hall of Geosciences, and Westerlin Residence Hall, which was included because of construction activities that prevented access to Thomas Tredway Library. Within each day-building combination, five surveys were completed at sunrise, 0900 h (CDT), 1230 h, 1600 h, and one hour prior to sunset. We assumed that a carcass found during a survey died in the interval between that time and the previous survey. A ‘clean-up survey’ was conducted at sunset the day before the first sunrise survey for each sampling week (Hager & Cosentino, 2014). This survey removed all carcasses that accumulated between survey-weeks, which may have otherwise introduced detection bias on the first day of weekly sampling.

Carcasses and corresponding identification tags were placed in zip-lock plastic bags and later identified to species in the laboratory. Birds were classified as adult or juvenile based on plumage, degree of cranial pneumatization, and, in hummingbirds, pattern of bill serration (Pyle, 1997). We consulted Fair et al. (2010) for recommendations related to collecting procedures of bird carcasses. Carcasses were salvaged under state Scientific Permit (NH11.0313), Illinois Department of Natural Resources, and federal Salvage Permit (MB08907A-0), U.S. Fish and Wildlife Service.

Point count surveys

We used point counts to rapidly assess the population sizes of all species present on site in 2009 and 2010 (Johnson, 2008). Our goal for point count information was to describe the community of breeders in close proximity to study buildings (Hager et al., 2013). Thus, we established one 50-m radius count circle each in wooded bluff face and landscaped savanna, and count circle edges were within 50–125 m of each of study buildings. We reasoned that this information would most closely approximate the number and types of birds at risk of hitting windows at the buildings we were monitoring. Two surveys/week in June and July were completed during appropriate weather conditions and within 4 h after sunrise (Bled et al., 2013).

We identified and counted all birds seen and heard within a 5-min count period. Species richness was the total number of species observed, and abundance was the maximum number of individuals for each species, which is an appropriate community metric for repeated counts conducted at the same point location (Johnson, 2008). We assumed that the maximum abundance for each species was a reflection of the abundance of both adults and juveniles because we could not distinguish birds of various age classes during count surveys. The following species, guilds, and taxonomic groups were excluded from analyses: birds flying over the site, migratory flocks, waterfowl, and raptors (Kalinowski & Johnson, 2010; Hager et al., 2013). We followed the recommendations of Fair et al. (2010) in reducing impacts to birds resulting from investigator presence during point count surveys. All point count surveys were completed by SBH to reduce detection biases that may result from multiple counters.

We used North American Ornithological Atlas Committee (2012) to classify levels of each species’ breeding behaviors, which were documented opportunistically during point counts and at other times throughout the season. Species classified above the ‘observed’ level were considered part of the site’s breeding community.

Scavenger study

We minimized bias associated with imperfect detection of carcasses by using a sampling design that was informed by estimates of carcass persistence before removal by scavengers and decomposers and detection by field workers (Hager & Cosentino, 2014). Persistence or survival of whole and partial carcasses, i.e., feather piles (Fig. 1B), was monitored at four campus buildings (Hanson Hall of Science, Swenson Hall of Geosciences, Thomas Tredway Library, and Olin Center for Education Technology) during a 7-d study period (2–8 June 2010). Two of eight whole, intact bird carcasses were randomly placed below different facades of each building (see Hager, Cosentino & McKay, 2012 for details on carcass placement and daily monitoring). Each carcass was a different species ranging in size from Tennessee Warbler (Oreothlypis peregrina; 9 g, Sealy, 1985) to Brown Thrasher (Toxostoma rufum; 70 g, Cavitt & Haas, 2000).

Data analysis

For the carcass detection study, we estimated mean survival times (t) for complete and partial (i.e., feather piles) carcasses at each building using the exponential model r = e^−d/t, where r is the probability of survival for d = 1 day (Huso, 2011; Hager, Cosentino & McKay, 2012).

We used a Fisher’s Exact Test to examine differences in the number of BWCs among species considered Short-distance Migrant, Long-distance Migrant, and Permanent Resident.

To determine whether species abundance was related to the number of collisions, we used a generalized linear model, specifying a Poisson distribution with a log link function. We included maximum abundance, carcass age (juvenile or adult), the interaction between carcass age and maximum abundance, and year as predictors of carcass counts, but did not retain year as it was not significant.

We used a general linear model to evaluate the relationship between BWCs and time-of-day. Because inclement weather prevented some surveys from being conducted and because survey intervals were not equivalent, carcass data were standardized and are reported as the number carcasses survey^-1 hour^-1. Survey time and year were included as predictors of the number of carcasses detected per survey; year was not significant and was removed from the final model.

Scavenger assessment

During the 7-day scavenger study, the probability of daily carcass survival was high (0.8) and average carcass persistence at buildings was 6.25 days. A scavenger removed one carcass within 24 h after initial placement. Within 5 days, decomposers gradually transformed the remaining seven carcasses into feather piles, which were detectable and persisted until the end of the study (Fig. 1B).

Bird density and mortality

In 2009, we documented BWCs in 4 of 25 (16%) species of the breeding community, which we determined using the breeding behavior criteria established by North American Ornithological Atlas Committee (2012) (Fig. 2A). Two species not observed during point counts, and thus not considered part of the breeding community, were collected as carcasses: Belted Kingfisher (Megaceryle alcyon) and Yellow-billed Cuckoo (Coccyzus americanus). In 2010, we documented BWCs in 7 of 21 (33%) breeding species (Fig. 2B). An additional three species were collected as carcasses, but were not considered part of the breeding community, including Ruby-throated Hummingbird (Archilochus colubris), Baltimore Oriole (Icterus galbula), and Indigo Bunting (Passerina cyanea). There was no significant difference in the number of carcass species vs. non-carcass species (N = carcass, non-carcass) among Short-distance Migrants (1, 9), Long-distance Migrants (4, 3), and Permanent Residents (4, 6) (Fisher’s Exact Test, P = 0.15).

We determined the age classes of 24 carcasses, which were 67% juveniles and 33% adults (Fig. 3). Adults (all Long-distance Migrants) were affected by window collisions only at the beginning of the breeding season (Fig. 3A); species (N) included Yellow-billed Cuckoo (1), Ruby-throated Hummingbird (2), Eastern Wood-Pewee, Contopus virens (1), Red-eyed Vireo, Vireo olivaceus (1), Gray Catbird, Dumetella carolinensis (1), Indigo Bunting (1), and Baltimore Oriole (1). Juveniles were documented from the third week in June to the end of July (Fig. 3B), and were principally composed of Short-distance Migrants (N = 10; all American Robins) followed by Long-distance Migrants (N = 2 Ruby-throated Hummingbirds and N = 1 Gray Catbird) and Permanent Residents (N = 1 each of Black-capped Chickadee, Poecile atricapillus, Northern Cardinal, Cardinalis cardinalis, and House Finch, Carpodacus mexicanus).

Overall, the model containing maximum abundance, carcass age, and the interaction between carcass age and maximum abundance explained significant variation in bird mortality (χ² = 36.3, df = 3, P < 0.001, McFadden’s r² = 0.26). The relationship between BWCs and abundance depended on age (interaction: χ² = 12.7, df = 1, P < 0.001, Fig. 2). For adults, bird abundance was correlated negatively with BWCs (β = −0.80 ± 0.22). That is, BWCs were highest in the least abundant species, e.g., Red-eyed Vireo and Eastern Wood-Pewee, and lowest in species observed with high abundance values, e.g., House Sparrow (Passer domesticus), Chipping Sparrow (Spizella passerina), and American Goldfinch (Spinus tristis). For juveniles, mortality was positively related to abundance (β = 0.27 ± 0.05), which was mainly driven by the American Robin (Fig. 2).

Daily patterns of mortality

There was a quadratic relationship between the frequency of BWCs and survey time (r² = 0.48, F_1,7 = 3.2, P = 0.10, Fig. 4). Although this relationship was not significant at α = 0.05, we interpret the model fit as marginally significant at α = 0.1 because our likelihood of detecting a signifcant effect was limited by a small sample size of carcasses (N = 10). Sixty-six percent of birds died between sunrise and 1600 h and the remaining mortality occurred before sunrise. No carcasses were observed between 1600 h and sunset. Carcass species documented in 2011 and 2012 included: American Robin, Ruby-throated Hummingbird, Black-capped Chickadee, and four carcass species that were not observed in 2009 and 2010: Norther Flicker (Colaptes auratus), Ovenbird (Seiurus aurocapilla), White-breasted Nuthatch (Sitta carolinensis), and House Sparrow (Passer domesticus).

Species vulnerability

Eleven of 17 (65%) carcass species observed in all four breeding seasons (2009–2012) were recently listed by Loss et al. (2014) as being highly vulnerable to striking windows in the United States (Table 1). Of these, we found that Ruby-throated Hummingbirds and Gray Catbirds died in the greatest numbers and in at least two of the breeding seasons. However, two species of the breeding community, Cedar Waxwing (Bombycilla cedrorum) and Blue Jay (Cyanocitta cristata), considered vulnerable to BWCs were never found as carcasses. Generally, bird groups estimated to be at high risk and low risk for window collisions corresponded to the bird groups we observed as carcasses (Table 2). For example, Hummingbirds and Swifts and Kingfishers are listed as high-risk groups, which we documented with carcasses from Ruby-throated Hummingbird and Belted Kingfisher. Inconsistent with published vulnerability estimates (Loss et al., 2014) were observations of no mortality in high-risk groups, e.g., Waxwings, and documented mortality in low-risk groups, e.g., Flycatchers and Vireos.

Conservation implications and future research

Generally, the carcass species we found conformed to the species and species groups considered to be highly vulnerable to BWCs, such as Ruby-throated Hummingbird and Nuthatches, tits, and chickadees (e.g., Black-capped Chickadee) (Loss et al., 2014). Moreover, we observed limited mortality in birds whose species groups are considered to be at low risk of collisions, e.g., Orioles (Baltimore Oriole), Vireos (Red-eyed Vireo), and Flycatchers (Eastern Wood-Pewee). However, we found no carcasses of some high-risk species, such as Cedar Waxwings.

A more comprehensive understanding of the magnitude of mortality for species and ages affected in this region should be derived from studies that systematically sample a large number of study buildings and that employ a standardized carcass survey protocol (Hager & Cosentino, 2014). Standardized mortality data would allow for direct comparisons among sites and, collectively, to local population estimates from, for example, the North American Breeding Bird Survey (Loss, Will & Marra, 2012). Moreover, these data can be incorporated into models of survival for adults and post-fledglings, which when combined with estimates of nest productivity, would shed important light on population trajectories (Balogh, Ryder & Marra, 2011).

At broad scales, the nature of bird communities among urban areas should reflect landscape structure and functional connectivity (Chace & Walsh, 2006; Ramalho & Hobbs, 2012), which would result in variation in the species and magnitude of BWCs. For example, the American Robin (an urban adapted species) responds positively to urbanization throughout much of its range including desert scrub, closed canopy forest, and grasslands (Blair, 2004; Chace & Walsh, 2006), and high levels of juvenile mortality documented in this study may be occurring in urban landscapes throughout the range of this species. However, differences in avian communities among the major habitat types (i.e., desert scrub, closed canopy forest, and grasslands) should yield unique suites of urban sensitive species that may be vulnerable to window collisions. Future work should assess variation in species affected by BWCs for breeding bird communities across multiple spatial scales, which could then inform studies on demography and population health of those species.