Bird-window collisions: different fall and winter risk and protective factors

Bird-unfriendly windows

The dangers posed to birds by transparent, reflective, or mirrored glass have been demonstrated in both experimental tests and correlational field studies. Birds avoid certain window treatments when treated and control windows are hung at the end of flight tunnels (Klem, 2009; Klem & Saenger, 2013; Rössler, Nemeth & Bruckner, 2015) or in fields close to a wooded area (Klem, 1989; Klem, 1990; Klem, 2009). For existing buildings, the presence or extent of windows has correlated with greater bird mortality (Borden et al., 2010; Cusa, Jackson & Mesure, 2015; Hager et al., 2013; Klem, 1990; Klem, 2009). As Klem noted, “birds behave as if windows are invisible” (Klem, 2009). Mirrored glass has rarely been singled out for evaluation against other window designs, although past research has noted that mirrored windows pose high collision risks (Evans Ogden, 2002; O’Connell, 2001; Sheppard & Phillips, 2015). We examined whether a mirrored glass façade that partially covered one of eight monitored buildings had higher collision risks, which would replicate research from Brown et al. (2019).

Bird-friendly windows

Given the dangers of traditional reflective and mirrored windows, some buildings now have bird-friendly windows, defined as windows provided with permanent features visible to birds that deter collisions. These include fritted windows with small ceramic frits fused into the glass to reduce light and solar heat penetration (Wilson & Elstner, 2018). As a secondary benefit, these frits also are more visible to birds; for example in outdoor cage tests, birds avoided fritted windows in 62 of 70 trials (Klem, 2009). However, the frits are visible to humans, which may limit the acceptability of this solution.

Other permanent bird-friendly windows include embedded ultraviolet patterns, which are very unobtrusive to humans, such as the Mikado pick-up stick pattern from ORNILUX^®. In experimental tests these windows reduced collisions when the interior was dark (Klem & Saenger, 2013). However, bird species may vary in their abilities to detect these patterns (Håstad & Ödeen, 2014) and ultraviolet light is less available on overcast days or early mornings (Sheppard, 2019).

Trees create collision risks

Among the habitat factors that create window collision risks, trees and shrubs close to buildings or reflected in windows have also been found to increase collision risks (Borden et al., 2010; Gelb & Delacretaz, 2006; Klem et al., 2009; Kummer, Bayne & Machtans, 2016b). Although some past research focused on risks associated with particular trees, such as London plane trees (Platanus x acerifolia) (Gelb & Delacretaz, 2006), most studies do not specify particular species as creating risks. Pear trees became an initial focus for our study given early observations of the presence of Cedar Waxwings, who are especially dependent on fruit as a food source in winter (Witmer, 1996; Witmer, Mountjoy & Elliot, 2014). Because fruiting ornamental pear trees provide Cedar Waxwings with essential winter food and attract them to trees near buildings at the University of Utah, we focus on pear trees as a key risk factor.

Fall/winter replication rationale

In Brown et al. (2019), we examined how mirrored windows, pear trees near buildings, and bird-friendly buildings related to collision risks among eight buildings during winter season. We found that, when examined in single-variable models, pear trees near windows created an 8.5-fold increased odds of collisions and that mirrored windows created a 5-fold increased odds of collisions (Brown et al., 2019). Two buildings had bird-friendly windows, one with fritted and one with ORNILUX^®, and only one window collision was recorded across both buildings (a feather pile detected at the building with fritted windows). However, given that only two of eight buildings in that study had frequent window collisions in winter, the differences were statistically non-significant between the two buildings with bird-friendly windows and the other six buildings. Some of the six buildings without bird-friendly windows were known, from ad-hoc reports, to have experienced collisions in non-winter seasons. Research suggests that collisions are more numerous in migration seasons (Borden et al., 2010; Hager et al., 2013; Kahle, Flannery & Dumbacher, 2016; Schneider et al., 2018), especially fall (Loss, Will & Marra, 2015). We reasoned that an extended observation period, including fall and winter, could provide a more robust test of the efficacy of bird-friendly windows compared to other windows.

By extending the observation times to fall and winter seasons, we were able to address several research aims. First, we tested whether buildings with mirrored windows experienced more collisions, a finding that would replicate the original study (Brown et al., 2019). Second, we expected that the addition of fall migration data would show that the buildings with bird-friendly windows would have fewer collisions than other buildings. Third, we expected that the danger of pear tree proximity would only emerge as a significant factor in winter (a pear tree by season interaction), when birds eat the fruit.

Study site

The University of Utah is located along semiarid foothills of the northeastern edge of Salt Lake City, Utah, at the foot of the Wasatch mountains (interior US Environmental Protection Agency Ecoregion 19, Environmental Protection Agency, 2018). Its grounds are designated as a state arboretum. Therefore, it has a number of native species, such as Blue Spruce (Picea pungens), white fir (Abies concolor), and Gambel oak (Quercus gambellii), as well as introduced species, such as English hawthorn (Crataegus laevigata). Fruiting trees such as ornamental pear trees (Prunus calleryana) and flowering crabapples (multiple Malus species) are common on campus.

Building risk and protective factors

We selected buildings and operationalized their three risk and protective factors as previously described in Brown et al. (2019). Table 1 shows the eight buildings, their collision totals, and their risk and protective factors, along with window areas for monitored facades. Three buildings (1, 2, and 4) had mature fruiting pear trees within 4.9 m of the building façades (Brown et al., 2019). Building 1 had the risk factors of a mirrored façade on a 1980s extension to a 1901 building. It had 119 m² of mirrored surface, of which 30% or approximately 36 m² were protected with Feather Friendly^® Symmetry mitigation treatment. In addition, it had 114 m ²of traditional reflective windows on the other facades. The 1980s-era mirrored glass was more than just reflective; it included large mirrored sheets that reflected images even as natural light was dimming and regardless of the amount of sunshine. The mirrored sheets spanned across structures in the building, such as floors, and these structures were invisible to a person standing outside the building. The mirrors blocked so much light coming into the building that even on a bright day the building has overhead lights on inside to illuminate the hallway behind the mirror.

Finally, two buildings had bird-friendly glass, as shown in Fig. 1. Building 8 had windows with grey circular frits, approximately 3.2 mm in diameter, spaced at approximately 4 dots per 25 mm, arranged in staggered rows, covering approximately 40% of the surface. The ORNILUX^® Bird Protection Glass Mikado pattern looks like faint images of pick-up sticks criss-crossing in the glass, which had Polaris low-e coating (Lisa Welch, personal communication, Arnold Glas). The ultraviolet coating is designed to be visible to birds without obstructing human views. The three other buildings were selected because they had anecdotal reports of collisions, large amounts of traditional reflective glass, or proximity to many tree species.

Collision monitoring

Observer training and procedures

Observers received on-site training and collected inter-rater reliability data. Four observers collected most observations, supplemented in January by five volunteers from a conservation biology class. Training included walking along and, for the main observers, detection of carcasses placed below windows as training exercises, and instruction to alternate clockwise and counter-clockwise directions around buildings across monitoring days. In some cases, consistent with past research (Borden et al., 2010), the full building circumference was not monitored. Building 8 had fritted windows only on the west façade, so we monitored that façade only. Part of Building 1 was obstructed by a fenced in playground and part of Building 6 was inaccessible due to a rocky hillside. Figure 2 shows the monitored buildings, including designation of monitored facades, which were identical to Brown et al. (2019). Observers were instructed to double-bag carcasses or feather piles for delivery to an ornithology lab, which had an appropriate salvage license (United State Fish and Wildlife Service permit # MB836059-0, and Utah Division of Wildlife Resources Certificate of Registration # 5COLL3669) for research and disposal (Hager & Cosentino, 2014).

Outreach

To assure that we did not miss any hotspots of bird window collisions, we installed posters around campus with our contact information and published a newspaper editorial part way through the study. These outreach measures allowed us to share our iNaturalist site with campus users and to keep us apprised of any other problem areas on campus and provide public documentation of the study, as recommended by other researchers who focus on window collisions (Winton, Ocampo-Peñuela & Cagle, 2018). We also posted to iNaturalist and the local paper one striking photo of the results from the deadliest day of observations in order to sustain interest and concern for mitigation solutions (Fig. 3).

Data analysis

To assess interrater reliability we used the intraclass correlation coefficient (ICC), in which a score of 1.0, on a 0 to 1.0 scale, indicates perfect agreement (Shrout & Fleiss, 1979). The ICC scores within 95% confidence intervals are interpreted as evidence of good reliability if values are between 0.75 and 0.90 and excellent reliability if values are above 0.90 (Koo & Li, 2016). We used a one-way random effects test to derive the ICC, which is appropriate when different observers contribute to reliability observations. We selected the value suitable for reliance on a single rater’s score rather instead of an average score across observers (Koo & Li, 2016). All analyses used IBM SPSS Statistics 25 software (IBM, 2017, Armonk, NY, USA).

To assess the effects of risk and protective factors, we used the daily count of bird window collisions for each of the eight buildings. We employed Generalized Estimating Equations (GEE), which are useful for analyses of correlated count data across repeated measures, such as our repeated assessments of each building (Norusis, 2008). Our model used a Poisson distribution, log-link function, and significance levels of p < 0.05 (Ma, Mazumdar & Memtsoudis, 2012). After preliminary testing of an autoregressive structure across repeated weeks, we selected an independent correlation structure instead, given slightly better model fit results. The fit statistic used to compare models was a quasi-likelihood under independence model criterion (QICC); smaller scores indicate better fit (Norusis, 2008). We entered the three risk and protective factors as dummy variables in an analysis that included both fall and winter data. Tested models replicated the past analyses of Brown et al. (2019). These included a test of all three risk and protective factors, followed by each factor tested alone, followed by the best fit model from all 1-, 2-, and 3-way tests. To evaluate the expectation that pear tree proximity was especially deadly for birds in winter, in a new analysis we added a main effect for season (0 = fall and 1 = winter) and the key season by pear tree interaction term. If our hypothesis is correct, there should be more collisions counted in the winter season for buildings near pear trees than for other combinations of seasons and pear tree proximity (i.e., fall/pear trees, fall/no pear trees, and winter/no pear trees). We also presented simple descriptions of window areas and counts of collisions by buildings and season in Table 1.

Bird-friendly windows

We found that two buildings with bird-friendly windows had fewer collisions than other buildings when data were collected for fall and winter seasons. Although various bird-friendly glass designs are recommended in the American Bird Conservancy web site based on data from experimental tests (American Bird Conservancy, 2020), very few peer-reviewed evaluations of their performance in buildings in situ have been published. Both the ORNILUX^® UV protection and the fritted windows have been found to protect birds in experimental trials where the windows are tested without being installed in actual buildings (Klem, 2009; Klem & Saenger, 2013). Our study represents the first of many studies needed to evaluate how these bird-friendly windows perform in practice. As research demonstrates, buildings can have a wide variety of environmental features that influence their likelihood of window collisions, such as nearby trees, siting for shadows and light, species abundance (Nichols et al., 2018; Wittig et al., 2017), or shape of the building (Riding, O’Connell & Loss, 2019). Thus, it is important to test bird-friendly windows across a wide variety of design and habitat features. Although bird-friendly windows in building applications receive anecdotal support, such as the reduced window collisions reported by the Detroit Zoo (Van Dam, 2018), this study provides more systematic support for the efficacy of ORNILUX^® and fritted windows.

Two forms of bird-friendly windows were evaluated in the current study. Tunnel testing of fritted glass has suggested that frit patterns applied as stripes were more effective than dots for the same area of glass covered (Sheppard, 2019). However, the dotted frits in the present study were effective. Fritted windows had no collisions in the current study and only one collision (a feather pile) in the 2019 study. It is not clear why dotted frits are so effective in the current study. The (Sheppard, 2019) study had tested white 3.2 mm dots that were spaced to cover only 20% of the window surface. The current study tested “warm grey” 3.2 mm dots were spaced approximately five per 25 mm, with approximately 40% coverage. Another anecdotal report has indicated that earlier tests of 3.2 mm white dots spaced to cover 40% of the surface did deter collisions (Leonard, 2013). The closely spaced dots in the present study might be perceived as a solid surface, not individual dots. This suggests it would be fruitful to test differences across frit shapes, colors, and coverage.

For the bird-friendly building with ORNILUX^® windows, there had been no window collisions in the 2019 study and there were two window collisions in the current study. The Lincoln Sparrow fatality was a clear window collision where windows were recessed under a substantial overhang, which lowers the visibility of the UV patterns and therefore effectiveness of the protection, according to the manufacturer’s web site (http://ornilux.com). The Mourning Dove fatality might have been a carcass cleaned by a predator or scavenger, given the condition of the carcass. We note that one published study has separated intact carcasses from those with evidence of dismemberment to include in separate analyses; dismembered carcasses might result from scavenger activity after a window collision or from predation events that deposited the carcass in a manner where it is indistinguishable from the defining features of a window collision (Loss et al., 2019). Future research on larger samples might benefit from this strategy of separate analyses, with groups of carcasses defined as clear window collisions versus potential window collisions (Loss et al., 2019). In either case, by keeping the Mourning Dove in the data as a window collision, we provided a conservative test of the efficacy of the ORNILUX^® glass. Interest in fritted glass and ORNILUX^® UV glass is likely to increase as building owners grow more aware of the need to protect birds to attain more sustainable buildings (Sheppard, 2019).

Seasonality of collisions

The current study provided insight into winter collisions, whereas past research has often focused on bird fatalities in migration or breeding seasons. Many studies find fall to be the season of most window collisions, given that fall migrations include many young birds that may not make a spring return (Zink & Eckles, 2010). Consequently, several studies have collected data during fall only (Hager et al., 2017) or fall and spring migration only (Bracey et al., 2016; Cusa, Jackson & Mesure, 2015). Sometimes the fall fatality rate is especially high; in Cleveland, Ohio there were 193 fall and 52 spring migrant collisions, although with more observation days in fall, and substantially fewer found in summer and winter (Borden et al., 2010). In Minneapolis, Minnesota there were 758 fall (2.5 months), 209 spring (2.5 months), and 33 window collisions in summer (June) (Loss et al., 2019). In Blacksburg, Virginia; Alberta, Canada (Kummer, Bayne & Machtans, 2016b), San Francisco, California (Kahle, Flannery & Dumbacher, 2016), and Rock Island and Moline Illinois (Hager et al., 2013), fall and spring migration as well as summer breeding seasons had more window collisions than winter (although the 5 winter collisions in Illinois were close in magnitude to other seasons: spring n = 7, summer n = 8, and fall n = 14). Not all studies share the pattern of fall prominence of window collisions. A Stillwater, Oklahoma study identified more spring than fall and summer window collisions (Riding, O’Connell & Loss, 2019) and a Xalapa, Mexico study found no seasonal variations (Gómez-Martínez et al., 2019). This short review suggests that there are likely more fatalities in migratory seasons, especially fall. However, when studies omit winter data collection, they may underestimate risks to resident birds who are present in winter.

Although many studies describe collisions at each season or analyze multiple seasons of collisions together, the current study represents one of several that has tested for particular seasonal differences in risk factors or species vulnerability. The season by pear tree interaction test provided more definitive data than past research. The pattern tentatively identified by Brown et al. (2019) was that winter proximity of fruiting pears to buildings with mirrored and reflective windows posed special risk. However, without data from another season, it was difficult to be definitive. The new data indicated that the combination of pear trees near buildings in winter posed a 40-fold increased risk of bird window collisions over other conditions. Other studies have taken different strategies to understand different signatures of risk across seasons. For example, Riding et al. conducted separate analyses by seasons and found that risk factors vary across seasons, with lawn cover, for example, important in summer and fall but not spring (Riding, O’Connell & Loss, 2019). Similarly, another study found night-lighted buildings to be more risky in spring and larger total glass area to be more risky in fall (Loss et al., 2019).

Cedar Waxwing vulnerability

Given the research field’s emphasis on migratory season data, risks for birds like Cedar Waxwings, who are drawn to winter fruiting trees, may also be underestimated. Despite relatively few studies of winter window collisions, it is notable that the review by Loss et al. (2014) found that Cedar Waxwings were the fifth most vulnerable to window collisions, with 3.6 times the collision risk as other species (Loss et al., 2014). However, the current study and the original Brown et al. (2019) study documented a larger percentage of Cedar Waxwing window collisions than found in other studies. The Brown et al. (2019) review of prior research found that Cedar Waxwings percentages of the total collisions were 15.7% in Durham, North Carolina (Ocampo-Peñuela et al., 2016) and 14% in Edmonton, Alberta (Kummer & Bayne, 2015) but less than 3% in seven other studies. One way to assess Cedar Waxwing vulnerability is to compare the collision data to the percent of eBird lists with Cedar Waxwing sightings (although this measure does not account for the numbers of birds seen per species, and Cedar Waxwings are often in flocks). The percentage of eBird lists with Cedar Waxwings in Salt Lake City is 9.1% for November through January, much lower than the 62.2% of collisions in the current study or the 91% found in the earlier winter-only study (Brown et al., 2019). For the other two studies with the highest percentages of Cedar Waxwings, Durham had 2.9% of eBird lists with waxwings in the target months (April and September/October) compared to 15.7% of waxwings in collision data; Edmonton had 8.8% in eBird lists (year-round) compared to 14% in collision data. These comparisons suggest that both Durham and Edmonton might also have local conditions that put Cedar Waxwings at greater than expected risk. Our research showed that the very local conditions of winter food from pear trees near mirrored or reflective window buildings create high risks for Cedar Waxwings. Pear trees are becoming more abundant across the country through invasive spread and inexpensive plantings (Culley & Hardiman, 2007; Tallamy, 2009), potentially multiplying the number of sites where pear trees are near windows. We urge other researchers to be alert to the possibility that their research sites might have this deadly combination of pear trees near windows and to collect data on whether other fruiting trees are also deadly to Cedar Waxwings when planted near windows. In addition, given that Cedar Waxwings flock all year and are very frugivorous, it is not clear whether fruit trees near windows pose risks to other birds; we witnessed European Starlings (Sturnus vulgaris) and American Robins (Turdus migratorius) in the trees but recorded only one robin fatality near the pear trees.

Targeted mitigation

As knowledge of bird window collision risk factors increases, it will be important to figure out how to apply this knowledge to enhance bird safety. For example, if winter deaths of Cedar Waxwings are found elsewhere, it might be possible to apply tempera paint to especially risky windows, washing it off after winter season (Sheppard & Phillips, 2015). Due to aesthetic concerns and challenges of reaching window surfaces, this proposed solution will not work for the buildings at the University of Utah, where we plan to mitigate the risky windows with applications of external dots (Feather Friendly^®). Other immediate solutions could involve replacing problematic trees. For long-term bird protection, planting guidelines could forbid planting fruiting trees near expanses of glass. Bird-friendly building standards that require permanent bird-friendly glass are becoming more popular but are not yet common. In addition to recommending such building standards at state or city levels, universities could also foster bird-friendly building requirements. Mirrored windows allow humans unimpeded views of the outdoors but, based on the evidence of their deadliness, we believe they should not be used in the future where birds might be found. The expansive views from mirrored windows, if motivated by a desire to see and appreciate nature, could be satisfied by bird-friendly windows instead.

Limitations

The study is limited by a small sample size of eight buildings and two seasons. Nevertheless, our targeted research questions could be addressed with a limited subset of conceptually- and empirically-based variables. In the future, it would be useful to collect data across larger samples that mix the risk and protective factors in various combinations and begin to single out differences among bird-friendly glass solutions. In addition, the study was consistent with a past published study, providing a rare opportunity to replicate past research, which has been called for by the bird window collision research community (Loss et al., 2019). We were also limited in that observer availability varied by time of day, although all observations were made during daylight hours. Like many studies of bird window collisions, we acknowledge that our detection of bird window collisions was unlikely to be perfect, so that the numbers should be considered underestimates.