Optimizing survey timing for detecting a declining aerial insectivore, the Black Swift ( Cypseloides niger borealis )

,


INTRODUCTION
The biology, distribution, and productivity of pan-American swifts (genera Cypseloides and Streptoprocne, subfamily Cypseloidinae) are poorly known throughout most of their range.These species are highly aerial (e.g., Hedenström et al. 2022) and fast flyers that typically only land during their nesting phase.They are associated with damp canyons and waterfalls (Chantler and Driessens 1995).Given swift nesting behavior and their inaccessible nest sites, it is challenging to determine breeding site occupancy.One of the best studied Cypseloidinae is the Black Swift (Cypseloides niger), specifically C. n. borealis; however, knowledge of its distribution and ecology is lacking across the Corresponding author: Paul G Levesque, paulglevesque@gmail.com majority of its range and a paucity of information collected at breeding sites is hampering efforts to conserve and restore populations.
The Black Swift is a long-distance migrant (Beason et al. 2012, Hedenström et al. 2022) that nests at breeding sites colonially or solitarily from temperate to tropical regions in western North America (south to Costa Rica, and in the West Indies; Villard andFerchal 2013, Gunn et al. 2023).Annual nest site fidelity is very high, with many sites used for decades (Kondla 1973, Collins and Foerster 1995, Hirshman et al. 2007, Levad et al. 2008, Levesque 2015).Despite high nest fidelity, relatively few Black Swift nest sites have been located in North America, especially in Canada (Gunn et al. 2023).Like other swift species, Black Swift nests are cryptic and inaccessible; they often are situated in rock crevices near waterfalls, wet canyons and caves, and sea cliffs (Legg 1956, Knorr 1961, Davis 1964, Gunn et al. 2023).Adults make long foraging trips (both temporally and spatially) that result in infrequent nest attendance during an extended nesting period (egg laying in June to fledging in late August to early October; Marín 1999, Hirshman et al. 2007, Gunn 2022).The provisioning rate of swift nestlings >9 days old is often infrequent (e.g., twice a day; Marín 1999, Gunn 2022), occurring primarily after dusk (Gunn 2022), and adult incubation bouts are long (>4 h, Marín 1997;>19 h, Gunn 2022).The difficulty in locating Black Swift nesting sites has long hindered our understanding of the basic ecology and reproductive needs of the species, including our ability to capture, tag, and DNA sample birds for monitoring and investigating research topics such as delineating subspecies, quantifying dispersal and gene flow, and understanding the full annual cycle (Committee on the Status of Endangered Wildlife in Canada (COSEWIC) 2015, Gunn et al. 2023)).
Most of the species breeds in Canada (C. n. borealis, ca.79,000 individuals; Partners in Flight 2023), mainly in British Columbia (BC), and to a lesser extent (<0.1%) in Alberta (AB) (COSEWIC 2015).The remainder of the C. n. borealis population (ca.10,000 individuals; Partners in Flight 2023) breeds in the USA (Alaska, California, Colorado, Idaho, Montana, New Mexico, Oregon, Utah, and Washington) and Mexico (COSEWIC 2015, Gunn et al. 2023).The other subspecies (C. n. costaricensis and C. n. niger) breed from Mexico to Costa Rica and the Caribbean, and no population estimates exist (Gunn et al. 2023).As with many aerial insectivores in North America (Nebel et al. 2010, Smith et al. 2015), the Black Swift has declined considerably.COSEWIC estimated a population decline of >50% in Canada between 1973and 2012(COSEWIC 2015).Black Swift is listed as Endangered on Schedule 1 of Canada's Species at Risk Act.The reasons for the population decline have not been confirmed, but they are thought to be related to changes in food availability and accessibility (COSEWIC 2015).Changes in the availability of high-quality prey at multiple periods during the annual cycle may contribute to aerial insectivore declines (English et al. 2017, Spiller andDettmers 2019).
Given documented population declines, the development of an effective survey methodology is an important first step toward recovery efforts for Black Swift in Canada and elsewhere in the species' range.An effective method to detect Black Swift site occupancy is required in order to develop and implement inventory and monitoring programs that can produce reliable estimates of Black Swift population size and trends, breeding distribution, and areas of occupancy.This information will inform recovery planning and implementation for the Canadian population of Black Swift and may have application in other parts of the species' range and to other Cypseloidinae species.
The most widely used method for determining Black Swift occupancy at seemingly suitable nesting habitat was developed in southern California (Foerster and Collins 1990) and refined in Colorado (Schultz and Levad 2001).The survey method, hereafter referred to as "evening surveys," involves monitoring waterfalls for Black Swifts returning to their nests for the night to incubate, brood young, or roost near or at the nest.Evening surveys have been used to estimate the number of nests at breeding sites and to estimate the size of regional populations (Foerster and Collins 1990, Levad et al. 2008); however, the utility of this survey method for estimating breeding populations is uncertain.Daytime surveys, originally recommended as a method for determining occupancy of known nests (Schultz and Levad 2001), were adapted for identifying new breeding sites with some success (Levad 2007, Levesque andRock 2017).However, because Black Swift nests are cryptic and difficult to access, extensive daytime survey effort often yields few results (i.e., Levesque and Rock 2017), and a standardized protocol has not been developed.
The reliability of breeding site occupancy surveys has not been formally tested for Black Swift, despite use of the survey methodology for nearly three decades.The effectiveness of evening surveys has been called into question for breeding populations in BC where nine evening surveys conducted during the nesting phase (between 13 June and 7 August) at five potential breeding sites from 2004 to 2013 resulted in no detections (Levesque and Rock 2017).One of the evening surveys was conducted at a site with a known Black Swift nest; despite the nest being active, observers did not detect adults visiting/returning to the nest before dark (Levesque and Rock 2017).Levad (2007) also described the failure of evening surveys in detecting the total number of birds returning to known nests at an active breeding colony in Colorado.In addition, only two breeding sites were confirmed among the more than 100 suitable waterfalls surveyed in Washington and BC during a citizen science inventory program using evening surveys (Altman 2003), which further suggests limitations to the method.
The evening survey method may be more effective if adapted for the dawn period, when Black Swifts leave their nest or roost sites to forage.In late July of 2016, we observed two adult Black Swifts departing from a waterfall before sunrise.In August of the subsequent year, Black Swifts were observed flying from nest habitat at the same site and at one additional site over four mornings, 3-7 min after sunrise.On the same days, the authors conducted evening surveys and never detected Black Swifts.These observations provided a promising indication that a more reliable method may exist for determining Black Swift breeding site occupancy and prompted the need to evaluate and quantify the robustness of dawn surveys relative to evening surveys.
In this study, we paired dawn and evening surveys at the same sites to compare the number of birds detected by each method.Furthermore, because dawn surveys lasted 1.0-1.5 hours, we could assess the time of the morning (before and after sunrise) that most birds were detected.Because light varies not only with time but can also be influenced by local site characteristics (e.g., canopy cover, site exposure/terrain) and weather (e.g., cloud cover), we also compared detection rates to ambient light levels.It is important to note that our goal was not to estimate population size.For that reason, we focused on relating the number of birds seen flying to or from a potential nest site to the survey timing (dawn or evening), light level, and time before or after sunrise at the moment the detection was made, while taking into account variation across months, years, and breeding sites.

Data Collection
We compared Black Swift detection rates at dawn and evening in 2018.To initially identify active sites for the 2018 study, we visited sites in mid-June to conduct dawn and evening surveys and search for nests at six historically known Black Swift breeding sites (sites that previously had Black Swift activity based on observations of adults, eggs, or nestlings at one or more known nests, or adult swift detections at the site during a dawn or evening survey) and at seven sites with potentially suitable nesting habitat.Potential habitat was assessed according to Black Swift nest site requirements identified in Colorado (Knorr 1961(Knorr , 1993a)): (1) flowing water; (2) high relief (commanding position above the surrounding terrain); (3) suitable nest niches that are inaccessible to ground predators; (4) unobstructed aerial access to suitable nest niches; (5) shading of nest niches; and (6) presence of moss at the nest site.We proceeded with paired surveys at nine active sites, after excluding four where no activity was detected in June.Selected sites were distributed across southern BC to reflect coastal and interior nesting sites.Site topography, aspect (i.e., sun exposure; Levad 2007), waterfall type (Conly 1993), waterfall/ canyon height and aspect, canyon length, degree of canopy cover, and substrate type varied by site.
In order to expand the number of known breeding sites in Canada, we modified the study in 2019 and 2020 to search new sites in both BC and AB using the same criteria as 2018.In addition to surveying the nine breeding sites from 2018 (eight of which continued to be active in all subsequent years), eight new sites with potentially suitable habitat were surveyed in 2019.Four of these new sites were determined to be active (Append.1).In 2020, 12 new sites with potentially suitable habitat were surveyed at dawn only; six of these new sites were determined to be active (Append.1).An additional 11 active sites surveyed in 2018 and/ or 2019 were re-surveyed in 2020 (Append.1).
During evening and dawn surveys, one to three observers, two of whom had more than 15 yr of experience surveying Black Swifts, positioned themselves downstream of the nesting habitat, outside of the spray zone, with an unobstructed view of the waterfall or canyon to enhance the detectability of the dark birds against the lighter sky and the white of the waterfall.At four survey sites (Deadman Falls, Canim Falls, Marble Canyon, and Two Valley Canyon), terrain and access limited observers' ability to position downstream from the nesting habitat feature.Instead, surveyors positioned themselves along the canyon rim at good vantage points such as viewing platforms or bridges.Surveyors continuously monitored the feature and adjacent habitat and the sky directly above the feature for Black Swift movement by uninterrupted scanning of the area with the naked eye.At some sites, surveyors were spaced >100 m from one another to cover the length of the canyon.The number and timing (departure from sunrise or sunset) of Black Swifts flying into or from potential nesting sites was recorded from each surveyor's vantage point.To avoid potentially double counting the same birds at sites where surveyors were independently conducting surveys within 50-100 m of one another (five dawn and five evening surveys, 2018 only), we removed one surveyor's data, chosen randomly, prior to analysis.Sites where Black Swift were observed flying into or from potential nesting sites were considered occupied.A light meter (Sekonic L308-SU) set to International Organization for Standardization (ISO) value of 100 was used to measure ambient light levels at the surveyors' locations at the time of detection and in between detections (approximately every 5-10 min).Factors that potentially influence detections such as local sunrise or sunset time, light availability, weather conditions (cloud cover, precipitation, wind), and surveyor were also recorded at the start and end of the survey.To determine the timing of sunrise and sunset, we referred to a handheld Global Positioning System (GPS) device or we downloaded the information from the National Research Council of Canada website, if a GPS device was not available.The National Research Council defines sunrise and sunset as the appearance and disappearance (respectively) of the upper limb of the sun as observed at sea level on a refracted (apparent) sea horizon (Government of Canada 2023).To reduce observer error, surveyors used digital audio recorders during the survey and transcribed results to a data form after conducting the survey.
The survey timing (dawn vs. evening) differed across the study period.In 2018, we tested the number of Black Swift detections during paired evening and dawn surveys (two surveys within a 24h period per site) repeated over the nesting phase (documented in Canada as beginning in early June and ending in mid-September; Rock et al. 2021) at each of the nine active breeding sites.Each paired survey was separated by at least 6 d to account for potential differences in activity across the nesting phase.Dawn surveys began 30 min before sunrise and concluded an hour after sunrise.Evening surveys began at least an hour before sunset and were terminated once it became too dark to observe or accurately identify Black Swifts.In 2019 and 2020, the study focused on identifying new breeding areas to expand the dawn sample size and evaluating the ambient light levels at which adults were detected.

Data Preparation
We ran three models to evaluate the optimal time to survey for Black Swifts.Our first model compared detections between dawn and evening surveys and across light levels.We used surveys from 2018 because few evening surveys were conducted in 2019 and none in 2020.The analysis included nine sites, with paired dawn and evening surveys repeated between two and four times at a site from June to August (Append.1).The data consisted of the number of swifts observed (0-4 birds) each time the light meter was read (12-79 times depending on site, survey timing, and date).
Ambient light meter readings were converted from exposure value (EV) to lux using the formula lux = 2 EV × 2.5 (1) The light meter did not specify a value (indicated as "error") for any reading less than EV = 0 (2.5 lux).Although actual light levels could be less than 2.5, we never conducted surveys in complete darkness.Thus, we replaced all error values (79 of 1,250 total observations) with 2.5 lux.Furthermore, the light meter did not always function correctly, which resulted in 117 missing values out of 1,250 total observations.We imputed the missing values by using predictions from a linear regression relating light levels to time of day within each survey timing (dawn, evening), survey date, and site.We ran the imputation five times to produce five data sets that were modeled separately but combined into a final product (see https://web.archive.org/web/20230702191454/https:/cran.r-project.org/web/packages/brms/vignettes/brms_missings.html).
We imputed the missing values using the mice package, version 3.13.0(van Buuren and Groothuis-Oudshoorn 2011) implemented in R version 4.1.2(R Core Team 2021).We also ran the analyses with the missing data removed and the results were similar.We report the results from the imputed data.
Our second and third models measured the effect of time of day on bird detections for the dawn surveys.We only registered detections/non-detections when we read the light meter, which was approximately every 5-10 min and at time of detections.However, because we were constantly viewing the habitat feature, we created data sets where every minute was an independent observation, knowing that any minute without a data point is a non-detection.Our data sets spanned from 40 min before sunrise to the survey's end, approximately 110 min after sunrise.
We used two different data sets to model the effect of time of day.
The first consisted of dawn surveys from 2018, 2019, and 2020, and only included sites with at least five independent surveys spanning the three years, which left eight sites (Append.1).Our second data set consisted of surveys from the last three weeks of August 2020, at six sites (Append.1); we analyzed the more limited data set to minimize any influence of intra-and inter-annual variation in bird detections.Moreover, we had another, partially independent data set, to evaluate the robustness of the results.

Statistical Analysis
We modeled Black Swift detections using a Bayesian hierarchical zero-inflated Poisson additive model implemented with the brms package (Bürkner 2017(Bürkner , 2018) ) in R version 4.1.2(R Core Team 2021).The results were nearly identical if we used a zero-inflated negative binomial distribution instead (results not presented).We chose a zero-inflated model because non-detections far exceeded detections.The zero-inflated structure mixes two processes to account for non-detections, one where birds are not present and one where birds are present but not detected.We ran additive models because we expected detections to peak at low light levels or around dawn, which precludes the expectation of a linear change in detections with light level or time before/after sunrise.We used random effects to account for non-independence among temporally replicated surveys within a site, with each group corresponding to a Monday to Sunday week (Append.1).
Consequently, we could assess the variation in relationships (detections vs. survey-type, detections vs. time) across the season and across sites.For the second model that used data from 2018, 2019, and 2020, we did not explicitly consider year as a random effect because there were some sites with only one or two surveys in a year (Append.1).Thus, we constructed independent survey periods that combined survey and year, i.e., the first survey made in 2018 was a distinct group from the first survey made in 2019.
During some parts of some surveys, multiple observers were watching for swifts, thereby increasing the probability of detecting a bird.We accounted for such varying effort by modeling the expected number of birds detected as a rate In a linear model, the rate becomes an offset variable whose coefficient is set to 1 We included the offset variable in all our models.
We ran all models with the following priors: the intercept and slope coefficients were modeled as a normal distribution with a mean of zero and a standard deviation of 10; the standard deviation of the random effect was modeled as a Cauchy distribution with a location parameter of zero and a scale parameter of one; the zero-inflated term-the probability a bird was on site even if not detected-was modeled with a beta distribution with both shape parameters set to two (Kurz 2023).
We ran models with four chains and 5,000 iterations with 1,000 tossed away as a warm-up.We set all initial values to zero.
We evaluated the fit of our models using diagnostics available in the brms (Bürkner 2017(Bürkner , 2018) ) and bayesplot (Gabry and Mahr 2022) packages: trace plots and R-hat values to test for chain convergence, effective sample size to quantify autocorrelation, and posterior predictive checks to compare predicted and observed values.For the posterior predictive checks, we took 1,000 draws of the posterior distribution to quantify bird detections, the number of zeros in each draw, and the maximum number of birds predicted to be detected in each draw.

RESULTS
All our models converged with R-hat = 1.00 and robust posterior predictive checks.The model diagnostics are detailed in Append.2.

Dawn vs. Evening Surveys
For a light level of 2.5 lux and for the population of nine sites, we detected 5.38 times more birds for an observation made during the dawn survey than the evening (95% credible interval of the difference: [3.14, 9.98]), which translated into an average of 0.199 [0.0800, 0.457] birds detected for a dawn observation and 0.0369 [0.0151, 0.0808] birds detected for an evening observation.
Although the credible intervals overlapped, the higher upper bound of the dawn surveys reflects the fact that 115 of 619 (18.6%) dawn observations detected one or more birds whereas only 32 of 523 (6.12%) evening observations detected one or more birds.Across sites, the ratio of birds detected between a dawn and evening observation ranged from 8. 93 [3.91, 23.4] at Marble Canyon to 3.22 [0.515, 10.0] at Clinton Falls (Fig. 1).
For the dawn surveys, bird detections declined non-linearly with increasing light levels (Fig. 2): the parameter describing the nonlinear relationship between birds and light (6.41 [0.640, 17.2]) was higher than the value for a linear relationship (1.00).Meanwhile, for evening surveys, bird detection changed little with increasing light levels (Fig. 2): the non-linear parameter (1.49 [0.0500, 5.39]) was close to the value for a linear relationship (1.00).Bird detections were higher on dawn than evening surveys until approximately 300 lux, at which point detections were higher on evening surveys (Fig. 2).
For dawn surveys, spatial variation in detection was greater than temporal variation: the variation in bird detections across sites (standard deviation mean and 95% credible interval: 1.19 [0.636, 2.11]) was greater than variation in detections across survey periods within a site (standard deviation mean and 95% credible interval: 0.347 [0.0241, 0.806; Fig. 3).However, the variation in effectiveness of dawn over evening surveys was similar across space and time: the ratio of detections for an observation made in the dawn vs. evening survey across sites (0.507 [0.0304, 1.26]) was similar to the ratio in detections across survey periods within a site (standard deviation mean and 95% credible interval: 0.494 [0.0253, 1.21]) (Fig. 3).

The Effect of Dawn Survey Time
Our analysis associating bird detections with time since sunrise for dawn surveys clarified the previous findings: detections peaked slightly (approximately 10 min) before sunrise (Fig. 4a, b) when light levels would be relatively low (45.5 ± 34.6 lux, average ± standard Fig. 3. Spatial and temporal variation (standard deviation (±95% credible interval) of group level terms) in Black Swift detection among dawn detections and between dawn and evening surveys.The variation in bird detections across sites was greater than variation in detections across survey periods within a breeding site.However, the variation in effectiveness of dawn over evening surveys was similar across space and time.The ratio of detections for an observation made in the dawn vs. evening survey across sites was similar to the ratio in detections across survey periods within a site.
deviation over all sites and survey dates; Append.3).Moreover, detections declined more rapidly after sunrise than they had accumulated prior to sunrise (Fig. 4a, b).The results were consistent for the two data sets: the first consisting of eight sites with data taken from June to August 2018-2020 (Fig. 4a) and the second with a partially overlapping set of seven sites with data from August of 2020 (Fig. 4b).For the first data set, bird detection peaked at approximately 7.5 min before sunrise with 0.0528 [0.0136, 0.170] birds per observation interval (minute).A survey conducted 30 min before sunrise detected 0.0265 [0.00632, 0.0936] birds in a minute, but a survey conducted 30 min after sunrise detected 0.00733 [0.00188, 0.0260] birds.For the second data set, bird detection peaked at approximately9 min before sunrise (0.0843 [0.0161, 0.300] birds).However, detections per minute were low through most of the morning, surpassing 0.0200 birds 20 min before sunrise (compared with approximately 37 min before sunrise in the larger data set) and dipping below 0.0200 again 2 min after sunrise (compared with approximately 16 min after sunrise in the larger data set).The difference between data sets is not altogether surprising given that the larger data set samples more months and years.

DISCUSSION
Our data show that, at breeding sites in western Canada, dawn surveys are more effective than evening surveys for detecting Black Swifts.Dawn survey observations detected 5.38 times the number of birds, on average, compared with evening surveys made within the same 24-h period at the same site.We also identified that, within the dawn period, detections declined with increasing light levels and that most bird detections occurred slightly before sunrise and declined rapidly thereafter.Although evening surveys have been used to determine occupancy at nesting sites (e.g., California (Foerster and Collins 1990), Colorado (Schultz andLevad 2001, Levad 2007), and the Pacific Northwest (Altman 2003)), we do not recommend using them as a substitute for dawn surveys.The higher detection rates during dawn surveys result in fewer false negatives for occupancy at a given site.Because detection rates are higher during dawn surveys, counts of swifts may more accurately determine the number of breeding or roosting individuals at a given site; however, this should be further tested using occupancy modeling.
Dawn surveys may outperform evening surveys because of differences in flight dynamics as birds leave vs. return to their nests.For example, we observed adults taking various flight approaches when returning to their nest (typically in the evening), including flying downward vertically to land within the nest crevice, making their flight path less obvious to surveyors.Adults may be more easily observed when they depart from their nest because they are constrained by the surrounding terrain and available flight corridor as they transition from a resting position (zero velocity) and convert the potential vertical component to gain the altitude and horizontal velocity necessary to fly above the canopy or terrain.This was most pronounced at Marble Canyon where the swifts appeared to be laboring to increase horizontal velocity as they climbed out of the deeply incised narrow canyon.A similar flight behavior was noted by Knorr (1993b.)who described swifts leaving a waterfall in California as following the stream bed or performing an "orbiting climb" to reach sufficient altitude to clear the terrain and rise above the canopy.Flight mechanics combined with terrain constraints may also reduce the velocity of flight during nest departure, potentially enabling higher detections by observers at dawn; however, flight speed was not measured in this study.
The effectiveness of dawn surveys is contingent on being able to observe birds flying from their nest sites.Although canyon and surveyor orientation relative to the sun could limit light entry and obscure flights at dawn, this site-dependent factor was mitigated by positioning oneself at the base of the waterfall or canyon, surveys made once in 2020 at seven different sites.The thick line is the expected value of the posterior distribution, and the shading is the 95% credible interval.We ran the 2020 data to discover whether the detection-light relationship was similar irrespective of whether the data was taken from across weeks and years or limited to one specific moment in time.
downstream of the suitable nesting habitat, facing a direction that maximizes the surveyor's ability to see the dark bird against the lighter waterfall or sky.Levad (2007) noted the importance of accessing the base of a waterfall to conduct evening surveys and suggested birds could go easily undetected, especially when the surveyor's vantage point is poor.When nest sites are known, daytime surveys could be an additional method to determine annual nest occupancy, as one can get a view of the nest and its contents, and not depend on detecting flying birds (Schulz and Levad 2001).
In the current study, there was a high degree of variation in the number of birds detected across breeding sites.Historically, in western Canada, Black Swift nesting colonies have been as small as one active nest (i.e., Clinton Falls; Levesque 2015) and as large as 12 nests (i.e., Johnston Canyon, Holroyd 1993).The number of nests possibly correlates with the number of roosting birds returning to or departing the site, which is the behavior most likely to be observed by surveyors assessing occupancy.Consequently, even dawn surveys cannot make up for imperfect detection of small populations.For example, at Clinton Falls in 2018 and 2019, our dawn surveys never detected a bird, despite there being a known, active nest in both years and one evening detection in 2018.Moreso, effective detection requires having trained observers conducting surveys.Black Swifts are a fast-flying species (7.91 m/s flapping flight; Marín and Stiles 1992), so a bird could be missed or the species could be incorrectly identified if the observer does not have sufficient focus and experience.In our study, the two lead observers had 15 or more years of experience surveying swifts annually.The other swift species whose Canadian breeding range overlaps with Black Swift, Vaux's Swift (Chaetura vauxi) and White-throated Swift (Aeronautes saxatalis), were not observed leaving waterfalls.
Although there was a high degree of variation in bird detections across sites, detections were relatively more stable across the nesting phase, per breeding site.Consistency in detections across the nesting phase may have occurred because both adults typically leave their nest or roost site at dawn, except during incubation or brooding.Correlating detection to occupancy requires further investigation, because the presence of helpers could bias occupancy estimates; however, nest helpers have never been observed for Black Swifts (Gunn et al. 2023).Early failed breeders could also visit or roost on cliff walls at breeding sites to prospect for available nests for use in future years, as has been reported in other long-lived colonial species (i.e., seabirds; see Cadiou et al. 1994) although, to the best of our knowledge, prospecting behavior has not been documented in Black Swifts.
Although we detected more birds during dawn than evening surveys, detections were confined to a relatively short period of the morning.More than 30 min prior to or after sunrise resulted in very low detection rates.Potentially surveyors were unable to detect birds prior to 30 min before sunrise due to low light conditions.Hence, an effective Black Swift survey does not need to be long, but it must target sunrise.The temporal pattern likely reflects the fact that swifts depart nest/roost sites around dawn to begin foraging once morning temperatures allow for airborne insects to emerge into air columns and insect activity peaks (Chapman et al. 2011).Research in California and Colorado has shown that one or both parents brooded or roosted at the nest site overnight (Marín 1999, Gunn 2022, Hedenström 2022).Young nestlings (<30 d old) were fed at two or more time periods during the day, typically once in the mid-morning (0830-1230) and again in the evening (after 1830) during one or more evening trips to the nest (Marín 1999, see also Collins andPeterson 1998, Gunn 2022).Morning provisioning by parents was common when the nestling was small (<30 d old), switching to dusk or nocturnal provisioning thereafter (Marín 1999, Gunn 2022).In addition to coinciding with insect pulses, leaving the nest to forage at relatively low light conditions may reduce the risk of predation.Predator avoidance has been suggested as the primary reason for nocturnal food deliveries to the nestling (Collins and Peterson 1998) and may explain why relatively few provisioning prey delivery flights are made to the nest in daylight.We found that adults returning to the nest did so over a broad period around dusk, consistent with Marín's (1999) observation of evening provisioning between 1830 and 2000 and Gunn's (2022) observation that evening provisioning began when adults returned to the nest between 1800 and 2200.Because most swifts depart at dawn, a better count of individuals at a colony may result from dawn surveys relative to evening surveys.
Currently, surveying breeding sites is the most effective tool to assess Black Swift breeding population size as detection rates are low among standardized surveys such as the Breeding Bird Survey (BBS), breeding atlases, and migration monitoring (COSEWIC 2015).We have gone a step further, demonstrating that observations should be made around sunrise.The ability to determine site occupancy based on the higher detection rates during dawn surveys could increase the number of known breeding sites, and thereby allow surveyors to conduct more focused nest searches at occupied sites.For example, at breeding sites identified during dawn surveys as well as unsurveyed potential sites, ground-based nest searches using an infrared camera (Forward-looking Infrared; FLIR) have proved useful for detecting active nests and roosting birds and nonbreeding season remotely piloted aircraft/drone imagery have led to the location of nests at known and potential breeding sites (W.E. Gross, unpublished data).The brief dawn departure period may equate to greater mist net captures for work requiring birds in hand, e.g., attaching tracking devices and/or enabling genetic or morphometric studies.Additional studies should be conducted to determine whether dawn may also be the best time to work on other swift species in the same subfamily as Black Swift.Our results provide a key first step to developing an effective monitoring technique tailored to Black Swift biology that can help develop accurate data on numbers, status, and population trends for this avian insectivore, which faces population declines (Sauer et al. 2017) and is consequently listed as "Endangered" in Canada.For each original data point, the average of the 1000 posterior draws for that data point is given.

Fig. 1 .
Fig. 1.The ratio of the expected number of birds detected on dawn surveys to evening surveys for nine sites in British Columbia.The points are the medians of the posterior distributions, and the lines are 95% credible intervals.

Fig. 2 .
Fig. 2. The predicted number of birds detected with increasing light for dawn and evening surveys.The lines are the median expected number of bird detections drawn from the posterior distribution and the shadings are the 95% credible intervals.

Fig. 4 .
Fig. 4.The relationship between survey time (minutes before and after sunrise, with zero indicating sunrise) and the number of Black Swifts detected, averaged over (a) surveys made between five and nine times between 2018 and 2020 at eight sites and (b) surveys made once in 2020 at seven different sites.The thick line is the expected value of the posterior distribution, and the shading is the 95% credible interval.We ran the 2020 data to discover whether the detection-light relationship was similar irrespective of whether the data was taken from across weeks and years or limited to one specific moment in time.

Figure S1. 1 .
Figure S1.1.Trace plots for the fixed effect parameter estimates (indicated by "b"), the variance of the smooth term (indicated by "sds"), and the zero-inflated probability term (indicated by "zi") for models relating Black Swift detections to light levels and survey timing.There are four chains for each of the five imputed data sets.The cloud of overlap indicate that (a) the chain converged, (b) swift detections to time since sunrise for eight sites spanning 2018 -2020, and (c) swift detections to time since sunrise for seven sites surveyed once in August 2020.

Figure S1. 2 .
Figure S1.2.Trace plots for the fixed effect parameter estimates (indicated by "b"), the variance of the smooth term (indicated by "sds"), and the zero-inflated probability term (indicated by "zi") for models relating Black Swift detections to time before and after sunrise for eight sites spanning 2018 -2020.

Figure S1. 3 .
Figure S1.3.Trace plots for the fixed effect parameter estimates (indicated by "b"), the variance of the smooth term (indicated by "sds"), and the zero-inflated probability term (indicated by "zi") for models relating Black Swift detections to time before and after sunrise for seven sites surveyed once in August 2020.

Figure S1. 4 .
Figure S1.4.Posterior predictive checks matching observed to expected swift detections for models relating detections to light levels and dawn vs. evening surveys.For each original data point, the average of the 1000 posterior draws for that data point is given.The expected and observed data are split into dawn and evening detections.

Figure S1. 5 .
Figure S1.5.Posterior predictive checks matching observed to expected swift detections for models relating detections to time before and after sunrise for eight sites spanning 2018 -2020.For each original data point, the average of the 1000 posterior draws for that data point is given.

Figure S1. 6 .
Figure S1.6.Posterior predictive checks matching observed to expected swift detections for models relating detections to time before and after sunrise seven sites surveyed once in August 2020.For each original data point, the average of the 1000 posterior draws for that data point is given.

Figure S1. 7 .
Figure S1.7.The distribution of the proportion of zero birds detected in 1000 data sets drawn from the posterior distribution of a model relating swift detections to light levels and dawn vs. evening surveys.The black line indicates the proportion of zero birds detected in the observed data.

Figure S1. 8 .
Figure S1.8.The distribution of the proportion of zero birds detected in 1000 data sets drawn from the posterior distribution of a model relating swift detections to time before and after sunrise for eight sites spanning 2018 -2020.The black line indicates the proportion of zero birds detected in the observed data.

Figure S1. 9 .
Figure S1.9.The distribution of the proportion of zero birds detected in 1000 data sets drawn from the posterior distribution of a model relating swift detections to time before and after sunrise for seven sites surveyed once in August 2020.The black line indicates the proportion of zero birds detected in the observed data.