Study area
We conducted this project at the two known long-term active breeding sites of COTE in Maryland, USA during the 2017 nesting season (May‒August). The first of our two sites was the Paul S. Sarbanes Ecosystem Restoration Project at Poplar Island (38.762° N, 76.384° W; hereafter Poplar). The restoration of Poplar is the product of a partnership between the U.S. Army Corps of Engineers and the Maryland Department of Transportation Maryland Port Administration (Maryland Environmental Service 2017) focused on the beneficial use of clean dredged material to restore remote island habitat. The breeding population of COTE on Poplar has been the highest, on average, in the state for the past several years. While terns bred at various locations on Poplar, our work (2017) focused on the largest breeding site located in the northwest corner. This colony encompassed ~ 0.45 ha, with 227 marked nests from an estimated 182 breeding pairs. While the colony was long and narrow, it was a clearly defined colony with no nests located in outcroppings away from the main colony.
Our second study location was Skimmer Island (38.336° N, 75.094° W; hereafter Skimmer), a small uninhabited island in the Isle of Wight Bay along the Maryland portion of the Atlantic seaboard. The island has been an important nesting site for COTE in recent decades but has a history of colony collapse due to extensive predation by Great Horned Owls (GHOW) (Bubo virginianus). Additionally, Skimmer has been slowly eroded by waves and boat wake action (Maryland Department of Natural Resources 2016). During our study (2017) the colony encompassed ~ 0.04 ha, with 151 marked nests from an estimated 123 breeding pairs. Terns occupied the southern portion of this island in a circular colony containing all nests on the island, though the colony occupied only ~ 1/3 of the available sand habitat.
Nest success
As part of routine colony monitoring, we marked and monitored all nests within the COTE colonies associated with this project during the incubation/hatching period. Colony monitoring consisted of researchers walking through the colony (2‒3 times weekly for Poplar; once weekly for Skimmer) in a line abreast formation identifying and marking new nests, recording the number of eggs and their condition by nest, and capturing chicks for banding with plastic field readable bands and metal U.S. Geological Survey bands. Unfortunately, due to the low sampling interval at Skimmer, we were not able to estimate the hatching success of each nest within the colony, instead determining fate only for camera-monitored nests. However, determining hatching success of monitored nests was possible on Poplar. We considered a nest on Poplar to have likely hatched if eggs were no longer found in the nest within 19‒31 days after the clutch initiation date, unless (1) sign of failure was present (i.e., sign of predation, nest wash out, etc.) in which case it was considered confirmed failed, or (2) a chick was captured or observed in which case it was considered confirmed hatched. Eggs gone from the nest prior to 19 days or remaining after 31 days of clutch initiation were considered likely to have failed. While incubation length has been reported to vary between approximately 21‒29 days (Hays and LeCroy 1971; Burger and Gochfeld 1991; Arnold et al. 2006), we used a slightly wider range as nests were not monitored daily. Since hatching and fledging success is known to be lower for nests established later in the season when less experienced nesters generally breed and environmental conditions are harshest (Arnold et al. 2008), we differentiated between original and re-nesting attempts based upon the date at which a large number of individuals were documented arriving at the colony following the collapse of a nearby colony (a separate sub-colony on Poplar Island, ~ 3 km from focal colony) due to predator pressure. Additionally, we examined the maximum number of eggs in completed clutches on Poplar, though a comparison with nests on Skimmer was not performed due to the longer interval between nest monitoring. Only nests where eggs were present and appeared viable to observers during two consecutive observations were included to avoid including nests where egg laying may have been interrupted by predation or weather events and thus may show an artificially low number of eggs. Hatching success and number of eggs were separately examined based on the type of equipment at the nest (camera + iButton, iButton only, or no equipment) via linear regression in R 3.3.3 (R Core Team 2018). A Tukey test was used to make comparisons within groups of all nests or original nests only, with no between group comparisons. Though the location of nests within a colony is known to impact nest survival (Hunt and Hunt 1975; Antolos et al. 2006), it is not accounted for in this analysis as the colony is long and narrow making differentiation between nests on the interior versus edge of the colony difficult and likely arbitrary.
Video data collection
In order to continuously monitor both nest attentiveness and colony behavior without regular disruption, six cameras were placed in each colony, and each camera was connected to the same eight channel DVR. On Poplar, the wireless receivers for these cameras were secured to the wall of a waterproof container housing the DVR using duct tape. On Skimmer, receivers were secured to 3 m tall PVC pipes to provide a clear line of sight between the receivers and cameras over tall vegetation. At each site, five of the cameras were placed at individual nests and one camera was placed on an elevated post at the edge of the colony facing the length of the colony for overall predator detection. Cameras were scattered as much as possible to limit spatial correlation but kept to a confined portion of the colony to minimize the amount of wire running through the colony to power cameras. To ensure that cameras would not be used as a perch by potential predators, bird spikes were secured to cameras and the posts on which they were mounted. Video was recorded continuously from a 78° field of view in 720 p resolution at a frame rate of 25 fps. The camera’s built-in IR light was able to record video at a minimum illumination of 0 lx. All cameras within a colony recorded their data to a single 1 to 4 TB hard drive, capable of storing ~ 1.5 weeks of video data, within the DVR. These hard drives were replaced during colony surveys to avoid reaching storage capacity. Video-surveillance systems were installed at colonies on both Poplar and Skimmer once a majority of nests in the colony contained at least two eggs in an effort to provide birds the opportunity to acclimate prior to the start of incubation (Nisbet and Cohen 1975; Nisbet et al. 2017). Cameras were relocated to other nearby nests entering the incubation stage when nest failure or hatching was apparent to colony observers. A full description of the video-monitoring system design and installation can be found in Wall et al. (2018).
To review video, the hard drive from the field DVR needed to be removed and placed into an in-lab DVR for processing. Video was reviewed one channel at a time in 24-h segments at four times normal speed. Video reviewers documented each time an adult left and returned to the nest to the nearest minute resulting in a data file that listed the status of the nest (adult on or off the nest) at every one-minute interval for which video existed. Any change in status that lasted less than 1 min (i.e. nest left exposed during mate switching or upflights) was not included as we sought to view attentiveness as a status of the nest and not the individual parents. While behaviors at the nest such as mate switching were also documented during review, they were not further analyzed (see “Results” section). Finally, when reviewing colony camera video, reviewers noted the number of times adults flocked and, when determinable, the cause for the flocking. The limited range of IR lights (only a few meters in front of the camera were visible) prevented review of colony camera video between 22:00 and 04:00 nightly. It should be noted that this limitation on the review of nocturnal footage was present only for the colony camera, as nest cameras were close enough to nests to overcome this limitation.
iButton data collection
Similar to our placement of cameras, we placed Thermochron® iButtons (model Nos. DS1921G, DS1921H, DS1922L; Maxim Integrated, San Jose, CA; hereafter iButtons) throughout both colonies in a subsample of active nests containing two or more eggs (37 on Poplar and 39 on Skimmer; Nisbet and Cohen 1975; Nisbet 2017). All nests monitored by video cameras had iButtons, ensuring comparability of datasets. While video-monitored nests, and thus the iButtons within them, were spatially clustered to facilitate powering of the cameras, the remaining iButtons were dispersed throughout both colonies. We placed iButtons into a ring of craft foam to prevent the iButtons from damaging eggs as they were rotated by the incubating adult. To maximize likelihood of the iButton remaining within the nest we attached them to 15 cm plastic tent stakes using Velcro® and then placed them in the center of nests flush with the nesting substrate at the bottom of the nest cup. These methods are a modification of those described by Hartman and Oring (2006). Since iButtons cannot detect changes in temperature sooner than 3 min (Maxim Integrated, San Jose, CA) we set this as our sampling interval and interpolated between recorded temperatures through use of the “na.interpolation” function within R to reach the same minute by minute precision as video data. In instances when data was not recorded for > 5 min, due to an internal error causing a missed recording cycle or lost data, interpolation was not performed. It is critical to recognize that this approach is intended to monitor the temperature of the nest bowl in an effort to determine attentiveness, and will not provide exact egg temperatures.
Data manipulation and supplementation
Climatological data
While efforts were made to collect local climatological data directly at the study sites, complications with in-colony weather stations resulted in unusable data and required the use of external data sources. Thus, we gathered temperature and barometric pressure data for 1 June to 31 July 2017 from National Oceanic and Atmospheric Administration (NOAA) National Centers for Environmental Information stations WBAN:00124 (38.976° N, 76.333° W) and WBAN:93786 (38.308° N, 75.124° W) to approximate weather conditions on Poplar and Skimmer Islands, respectively (NOAA 2018). These stations were chosen as they were within ~ 10 km of their respective colonies. Since temperature and barometric pressure were recorded approximately every 20 and 60 min, respectively, these data were then interpolated for each 1 min interval between recorded observations of less than 59 min in length using the same method as described above and paired with corresponding video and iButton data.
Time-of-day bins
We assigned each minute increment of paired video and iButton data to a distinct time-of-day bin. The Morning bin ranged from civil dawn through 11:59 EDT, Peak from 12:00 through 16:00 EDT, Cooling from 16:01 EDT until civil dusk, and Night from civil dusk until civil dawn. Civil dusk and dawn were based on the times reported by the United States Naval Observatory weather stations in Cambridge, MD (38.561° N, 76° 76.079° W) and Ocean City, MD (38.406° N, 75.060° W) for Poplar and Skimmer, respectively (U.S. Navy Observatory 2016). The Peak bin was set based upon manual review of the data, with the time (s) of peak daily temperature per day recorded across the study period per site. We then found the nearest hours (noon to 16:00) that would contain > 95% of these peak times.
Analysis
Natural bout characteristics
After all data supplementation and manipulation was complete we identified all bouts, or consecutive periods of time in which an adult was present (on-bout) or absent (off-bout) from the nest, based on the on/off statuses generated during video review. The minimum consecutive length of time required to be labelled a bout was 1 min, as this was the precision with which video data was recorded. Bouts before or after periods when video was not recorded or was not viable due to interference providing an “out-of-range” error, were discarded since one could not accurately confirm the actual duration of the bout. Similarly, bouts at the beginning or end of the entire sampling period, and bouts during or immediately surrounding in-colony researcher presence were discarded. This approach was designed to provide a basic understanding of nest attentiveness rates while also providing researchers with a general description of how adults behaved naturally. Since attentiveness and related behaviors are known to differ among egg laying, incubation, and hatching (Courtney 1979), we only included data from the date of clutch completion (when nest was observed to have all eggs eventually deposited) until the date of first chick hatch (successful nests; Courtney 1979) or 31 days after clutch completion (failed nests). For nests that had no new eggs deposited after cameras were in place (suggesting the clutch was already complete) the first 2 days of footage were discarded to allow for the birds to acclimate to the presence of the equipment.
We used three metrics to summarize nest attentiveness and the impact on the nest during natural bouts: bout duration, mean bout temperature, and overall nest attentiveness. Bout duration, or the time from beginning to end of a bout, was calculated for each natural bout documented in our study period without respect for time-of-day as there is no biological reason to expect these behaviors to fall smoothly within the artificially created bins. Instead, this metric was grouped by bin during which the bout originated. Since both data and residuals were non-normally distributed we calculated the mean, median inter-quartile range, and standard deviation of bout duration by time-of-day, nest success, and attentiveness status and compared these distributions via a pairwise Wilcoxon rank sum test using the Benjamini and Hochberg correction via the “pairwise.wilcox.test” function in R. Mean bout temperature, or the average of all iButton temperature data during natural bouts, was calculated by attentiveness status (on or off nest) and nest success for each time-of-day bin. Differences between mean bout temperatures were examined via linear mixed effects models using the ‘nlme’ package in program R (temp ~ time-of-day bin + nest attentiveness status + nest success + time-of-day bin: nest attentiveness status + time-of-day bin: nest success + nest attentiveness status: nest success, random = ~ 1|Nest ID). Least squares means and pairwise comparisons were calculated via the “emmeans” function in R.
Finally, to evaluate nest attentiveness, we calculated the percentage of time an adult was on the nest during natural bouts per bin by day and nest. Percent attentiveness was only generated for bins with at least half of their durations composed of natural bouts to avoid skewing data with short duration samples. We then used a Beta regression to evaluate differences in nest attentiveness (percent nest attentiveness ~ time-of-day bin + nest success + time-of-day bin:nest) success via the “betareg” function in R. Since beta regression requires data bounded between zero and one, bins with values of zero or one were transformed via the formula \(x' = \frac{{x\left( {n - 1} \right) + s}}{n}\) (Smithson and Verkuilen 2006) where n is sample size and s is a numerical constant set to 0.5. Least squares means and pairwise comparisons were then calculated via the “emmeans” function in R.
Bout status comparative modeling
We identified comparison bouts (bouts to use for comparing status assigned via video review and via iButton modeling) in the same fashion as natural bouts with the following exceptions: (1) bouts which occurred when researchers were within the colony were similarly discarded but bouts before or following research presence were not discarded and (2) bouts before video data went out of range were retained. Since comparison bouts were only for use in modeling the ability of iButton data, when paired with other covariates, to determine if an adult was on or off of the nest and not the description of bouts characteristics, the full restrictions set for natural bouts were not necessary here. Only footage determined to be within the incubation period was included in comparison bouts as thermal properties of the nest may change as chicks hatch and eventually leave the nest cup.
We tested a suite of six a priori logistic regression models against a subset of the comparison bouts dataset to determine the best-fit model for determining nest attentiveness based on iButton and climatological data. The subset of the comparison bout dataset was created such that it contained only records that had a value for all covariates used in the most complex model. This was completed to ensure that sample size would not change between models and invalidate AIC comparison (Konishi and Kitagawa 2008). We ran all models by time-of-day bin and compared them based on AIC score (Konishi and Kitagawa 2008). After best-fit models were selected for each time-of-day bin, we further split data into training (25%) and testing (75%) subsets via a stratified random sampling of the on vs off comparison bouts. The training subsets were then used to inform the previously determined best-fit models. Following model fitting, we used the “predict” function in R to determine the predicted probability of each record in the testing dataset being part of an on-bout. All points with a predicted probability of an adult being on the nest < 0.5 were considered off while the remainder was classified as on. We then compared outcomes to the status assigned via video classification (assumed as truth since status could be visually confirmed) and the percentage of data correctly predicted was calculated by probability bins and by the duration of the bout with which each record was associated.