In this multi-scale assessment of temporal variation in bird-window collisions, our predictions related to the diel timing of collisions were only partly supported. We predicted more casualties would occur during morning than other times of day, which should have resulted in our detection of the greatest number of fatal collisions on midday surveys and more non-fatal collisions during morning surveys than midday and evening surveys. However, greatest numbers of both fatal and non-fatal collisions were observed on morning surveys, indicating that more collisions occurred during overnight and early morning periods than mid-to-late morning and afternoon. As predicted, this diel pattern was consistent across seasons. Our predictions about monthly and seasonal patterns were also only partly supported. Unexpectedly, total collision mortality was highest in the spring migration month of May, and avian residency status interacted with season such that roughly equivalent high numbers of resident and migrant collisions occurred in spring, more resident than migrant collisions occurred in summer (and overall from Apr to Oct), and more migrant than resident collisions occurred in fall. Further, unlike many past studies, collision mortality of migrants was roughly equivalent in spring and fall migration.
Diel collision patterns
We observed more fatal and non-fatal collisions in the morning than in midday and evening surveys combined, even though we included fewer morning surveys in diel analyses. These differences are likely conservative in that an even greater proportion of fatal collisions than we observed likely occurred overnight and in the early morning, but went undetected because of bias associated with observer detection and scavenger removal of carcasses. Concurrent work in our study area20 found that relatively inexperienced volunteers detected a slightly lower proportion of carcasses (0.69) than experienced surveyors (0.76). Because roughly 10% of morning surveys were conducted by less-experienced volunteers and all midday or evening surveys were conducted by full-time technicians or the authors, we likely missed more carcasses on morning surveys. Additionally, most scavenging events (68%) were at night20, so bird carcasses from overnight collisions were the least likely to persist until the subsequent survey. Thus, underestimation of fatal collisions in the preceding interval was almost certainly greater for morning surveys than for midday and evening surveys.
A prevailing hypothesis for why daytime bird-window collisions occur is that birds making local (e.g., foraging) movements fail to perceive a barrier when flying toward objects either on the other side of glass or reflected on a glassy surface8,15. Under this hypothesis, daytime collisions for both residents and migrating birds at stopover locations would be expected to occur most frequently when birds are most active, which is typically near dawn regardless of season. Our finding of the greatest number of collision fatalities on morning surveys circumstantially supports the above hypothesis and expectation, as do past descriptive studies of diel variation in bird-window collisions26,27,38,39. However, our study design did not allow differentiation between nocturnal and early morning collisions, and a nontrivial proportion of carcasses detected on morning surveys likely represented collisions from the preceding overnight period. Nighttime collisions may occur at any structural component not easily detectable at night (i.e., they are not limited to glass surfaces), and can be exacerbated by artificial light emission that attracts and disorients migrating birds36,37,44,45. Nonetheless, the observation of more non-fatal collisions (including directly witnessed collisions) during morning than midday or evening surveys does strongly suggest that morning carcass counts included many collisions that occurred near or shortly after dawn.
A potential limitation of our study regarding time-of-day analyses is the longer interval between evening and morning surveys than between other survey periods. If collisions occurred uniformly or randomly in time, we would expect to find more bird carcasses during morning surveys due to the longer preceding time interval. However, as described above, the early morning peak observed for non-fatal collisions (Fig. 1), which are less persistent than carcasses and therefore do not accumulate over time, suggests that collisions do not have a uniform or random temporal distribution and that a real peak in collision frequency occurs sometime shortly prior to when we conducted morning surveys. Another possible bias is that we conducted surveys during fixed time periods rather than adjusting them to seasonally varying sunrise and sunset times. This limitation would have most strongly affected our summer results, potentially inflating summer morning counts as a result of collisions for both early morning and late evening (the two periods when birds are most active) being grouped into the same survey period. However, this limitation was unlikely to greatly influence our conclusions about diel collision patterns because: (1) these patterns were fairly consistent across seasons, suggesting a relatively small influence of varying sunrise and sunset times, and (2) sunrise and sunset times vary by only a few hours over the course of the entire year whereas the broad time periods for which we compared collisions (overnight, morning, afternoon) consist of longer lengths of time. Further research is needed to identify the exact timing of collisions, including during overnight periods, and this could be accomplished with collision surveys conducted more frequently during the day and night or remote detection methods, such as video cameras, motion-triggered still cameras, microphones that record sounds of impact, and glass-mounted pressure sensors that detect vibrations from collision impact.
We predicted diel collision patterns would not vary seasonally because the pattern of birds being most active near sunrise and sunset is relatively consistent across seasons. Hager and Cosentino46 provide excellent guidelines for conducting bird-window collision surveys, but their recommendation to conduct surveys in mid-to-late afternoon is based on summer monitoring that found mortality to peak between late morning and early afternoon in Illinois, USA29. We suspect differences in diel patterns between that study and ours relate to geographic variation and/or seasonal sampling coverage, as our larger sample of surveys included spring and fall migration in addition to summer. Although many collision-prone species migrate nocturnally, the diel collision peak for migrants could still occur in early morning because nocturnally-migrating birds often set-down into stopover habitats during early morning47,48 and may be most susceptible to collisions at this time. There could be subtle seasonal variation in diel collision patterns that we failed to detect; however, the majority of collisions across seasons appear to occur near or before dawn (see also26,27,38,39). In combination with the previous study showing that scavenging peaks overnight20, we suggest that conducting daily collision surveys in the morning could result in the least biased mortality estimates, especially in urbanized areas where humans (e.g., cleaning crews) may remove carcasses in the early morning. As noted by Hager and Cosentino46, further research is needed to identify how the optimal survey time is influenced by factors such as geography, the bird community, animal scavengers, and removal of bird carcasses by humans.
Monthly and seasonal collision patterns
We expected more collisions in fall than other seasons because bird populations in North America are larger after summer breeding and include higher proportions of young birds that have less experience with flight, migration, and human structures. Also, numerous studies have found the greatest window collision mortality in fall, a pattern driven largely by migrant birds 12,18,23,27,31,40,49. Contrary to expectation, both raw counts and bias-adjusted estimates of collision fatalities were highest in the spring migration month of May and higher overall in spring than fall. This pattern resulted from a high number of both migrant and resident bird collisions. In fact, we found that roughly equal numbers of resident and migrant collisions occurred in spring. When considering migrating individuals only, we found roughly similar numbers of collisions in spring and fall migration, a finding that also is unexpected in light of past studies. Notably, two other studies that found that a large proportion of total collisions consisted of resident birds30,50 also documented a seasonal pattern of total collisions that was less skewed toward fall. An explanation for the relatively large amount of total spring mortality, and for our finding that migrant mortality was roughly similar in spring and fall, is that some long-distance migrants follow elliptical migration paths where migration routes in fall are farther east than in spring51,52, such that in central North America, numbers of some species are greatest during spring migration. This explanation is supported by our observation of some elliptical migrants colliding during spring but not fall (e.g., Swainson’s Thrush [Catharus ustulatus]). Our study adds further nuance to the understanding of seasonal variation in bird collisions and exemplifies the need to study bird-window collisions in a wider array of geographic contexts to allow region-relevant management recommendations.
Our predictions regarding avian residency status were only partly supported; more migrants than non-migrants were indeed killed during fall migration. In spring, however, roughly equal numbers of each group collided, and far more residents than migrants collided in summer (and overall from Apr–Oct). This result does not account for varying abundance (overall and seasonally) of each species group, so it does not necessarily imply that resident species are more vulnerable to, or at greater risk of, window collisions relative to their abundance or period of residency in our study area. However, the finding of a high proportion of non-migrant casualties was still unexpected given that previous similar studies have almost universally reported higher mortality among migrants15,25,26,39,49,53,54, although most sampled during migratory periods only. Even with our individual-based residency designations, we may have slightly underestimated migrant mortality because all individuals of some migratory species were classified as unknown due to overlapping resident and migratory periods. However, even if all unknown individuals were migrating, there were too few birds in this category (22 of 341 [7%] total carcasses) to change our conclusions regarding the migrant-resident comparisons. Anecdotally, many spring and summer collision fatalities were recently fledged juveniles (n = 26 [25%] in May; n = 17 [30%] in June), clearly indicating that some collision victims were indeed not migrating, and therefore, that the high number of resident collisions is not an artifact of our classification system. Moreover, we did not observe collisions of any migrant individuals during June, even though a few species migrate through our study area in small numbers during this period (e.g., shorebirds [order Charadriiformes] and some tyrant flycatchers [family Tyrannidae])55. It is possible, however, that some resident individuals were undergoing post-breeding dispersal at the time of collision, as evidenced by a small late-June peak of Tufted Titmouse (Baeolophus bicolor) and Carolina Chickadee (Poecile carolinensis) collision victims with brood patches (TJO unpublished data).
Although our sampling captured the peak months of spring and fall migration in our study region, we undoubtedly missed some early-spring migrants before April and late-fall migrants after October. However, greater than 10 years of near-daily collision surveys at one of the most collision-prone buildings in our study (TJO unpublished data) suggests this number of missed collisions was relatively small. Specifically, total collisions at this building (including residents and migrants) dropped from an average of > 8 birds in October to < 3 birds in November, with < 0.5 birds per month between December and March. Our surveys between November and March did document a non-trivial number of collisions (22 fatal and non-fatal collisions over 2 years) despite these surveys being conducted less frequently and at a smaller subset of buildings compared to monitoring from April to October. Based on these observations, future research is warranted to quantify relative numbers of collisions for migrant and resident individuals throughout all migration periods, and of total collisions throughout the year.
Factors related to avian reproductive phenology likely contributed to the monthly and seasonal collision patterns we observed. The relatively large number of collisions for resident birds from May to August likely reflects the main breeding period for these birds. During breeding, adult birds make foraging flights to supply food for nestlings and fledglings, and these frequent flights may increase collision susceptibility relative to non-breeding periods. Notably, adult fatalities experienced immediately before or during the breeding season may also have a disproportionate effect on population abundance because there is little time for population-level processes (e.g., density dependent increases in adult survival) to compensate for mortality prior to the breeding season21, and because dependent offspring may also die if adults are killed during the nestling or fledgling periods. In addition to collisions experienced by breeding adults, and as noted above, we observed many collisions of recently fledged juveniles and of adult birds likely undergoing post-breeding dispersal or dispersal between breeding attempts; these observations further support that breeding-related activities influence window collision risk.
Weather also was likely to contribute to the temporal variation we observed, especially for week-to-week variation in collisions. Weather conditions have been shown to influence building collisions for one species, the American Woodcock (Scolopax minor)56. Although formal analyses of weather effects on total bird collisions are lacking, weather appears to have contributed to several major bird collision events, including at buildings45,57,58 and other types of structures59. Weather conditions such as precipitation, cloud cover, and the presence and/or strength of headwinds or tailwinds—as well as extreme weather events like intense storms—are known to influence the timing and magnitude of bird migration35, and are thus likely to influence collision risk. Some of these factors (e.g., low cloud ceilings) may have especially strong effects on nocturnal migrants by exacerbating effects of nocturnal lighting and driving birds into areas of greater collision risk32,56. When considering the week-to-week variation we observed within seasons, the fine-scale peaks and lulls in collisions may respectively reflect weather conditions that favor or disfavor bird migration (e.g., strong tailwinds or headwinds, respectively) and/or elevate or reduce collision risk for migrating birds (e.g., low cloud ceilings or clear skies, respectively). A complementary explanation for weekly variation is the varying migration phenologies of different bird groups, e.g., fall collision peaks in late Sep and mid-Oct might reflect migration peaks for wood warblers (family Parulidae) and sparrows (family Passerellidae), respectively, in our study region. Finally, some of this short-term collision variation might have arisen simply due to stochastic factors unrelated to weather or the migration phenology of different species groups. Further research is needed to investigate correlations between weather and collisions at various temporal scales, and analyses such as ours that document multi-scale temporal variation in collisions contribute to clarifying these relationships.
Source: Ecology - nature.com
