Study area
The Charlotte Metropolitan Area (CMA) is composed of 12 counties in North and South Carolina, USA surrounding the city of Charlotte, North Carolina (35.22° N, 80.86° W) and covering an area of 8,280km2. The CMA human population is estimated at 2.8 million, with rapid growth since 2010 (U.S. Census Bureau 2019). The population of Mecklenburg County, within which Charlotte is located, is expected to grow by over 570,000 between 2010 and 2040 with an annual growth rate of 2.3% (Charlotte Future 2019).
The Charlotte city center consists of a concentration of developed land cover types, but development across the entire CMA is fairly sprawling and dominated by developed open space and single-family housing. Forested land, primarily deciduous, accounts for nearly 50% of the CMA, whereas pasture and hayfields account for just over 20%. Wetlands, natural grasslands, and open water are relatively rare. Elevations in the region range between 93 and 780 meters and the climate is humid continental, with average summer highs of 32° C, average winter lows of -1° C, and average annual precipitation of 116 cm (U.S. Climate Data 2021).
Roosting sites
We identified 15 black vulture and turkey vulture communal roosting sites within the CMA in 2019 and an additional 14 in 2020 for a total of 29 roosting sites (Figure 1). Selected roosting sites were located by the lead author using a list compiled by Mecklenburg Audubon Society (MAS) members, eBird reports, and personal observations of vulture movements across the CMA at sunrise and sunset (totaling approximately 80 hours). Our use of multiple sources of information to locate roosting sites helped to maximize roosting site sample size and minimize any potential spatial bias in roost location due to the preferential selection of easily visible sites near roads and/or other development. Prior to being included in the study, all reported and observed roosting sites were checked for vulture presence and type of roost. We selected roosting sites that hosted vultures overnight and excluded temporary roosts used during the day as resting and congregating sites prior to or immediately following overnight roosting.
Twenty-seven roosts were located on transmission or cellular towers. Roosts on transmission towers (22) were situated in right-of-way (ROW) corridors with forest on either side, whereas roosts on cellular towers (5) were not located in right-of-way corridors and were adjacent to a variety of land covers. The two roosts not located on towers were on small clusters of deciduous trees within 1km of transmission towers and adjacent to residential development.
Roost surveys
We counted the number of vultures at each roost once a month from November 2019 to March 2020 and from November 2020 to March 2021, for a total of five surveys per roost per year. Thirteen roosts were surveyed in both years and 16 roosts were surveyed in a single year. Survey periods coincided with the non-nesting season during which vultures use roosts most consistently by returning to the same roost night after night and in the largest numbers (Sweeney 1984). During each survey, we counted vultures at roosts between 30 minutes before and 30 minutes after sunrise when individual vultures could be distinguished as they became more active and spread out on the roosting structure but before they left the structure to forage in the surrounding landscape (Sweeney 1984). During the second year of surveys, the count period was shortened to only the 30 minutes before sunrise as a result of observations in the previous year that vultures often left roosting sites earlier than expected (see the Analyses section below for the correction applied to first year data to account for this difference in survey methods). Trained volunteers assisted the authors in conducting surveys in both years.
Explanatory variables
Local variables We measured local weather and habitat variables at each roosting site that may be important in explaining vulture attendance including air temperature, wind speed, cloud cover, roosting structure height, surrounding vegetation height, and open space (Table 1). Vultures generally remain at or near roosting sites longer during colder temperatures or inclement weather (Sweeney and Fraser 1986). However, on windy mornings, vultures may be more likely to leave the roost earlier, perhaps to take advantage of wind currents for early-morning foraging flights (Davis 1979). We measured roosting structure height, surrounding vegetation height, and open space to account for the physical characteristics of roosting sites that may aid in arrival, departure, and flight ease. Open fields or corridors surrounding roosting sites allow unobstructed arrival and departure and may provide upward air currents (Coleman and Fraser 1989; Davis 1979). The same benefits may be provided by the height of the roosting structure, with structures above tree level providing easier access.
Table 1
Potential explanatory variables of black vulture (Coragyps atratus) and turkey vulture (Cathartes aura) roost attendance in the Charlotte Metropolitan Area, USA. Developed land cover, Open Water land cover, Developed-Forest edge density, and deer carcass density were measured for landscapes with radii of 0.4, 0.5, 1, 2, 3, 4, 5, 10, 15, and 20km.
Variable | Type | Spatial scale | Data source |
Number of vultures | Response | NA | Sunrise roost surveys |
Air temperature | Explanatory | Local | Weather Underground Personal Weather Stations (Weather Underground, 2021) |
Wind speed | Explanatory | Local | Weather Underground Personal Weather Stations (Weather Underground, 2021) |
Cloud cover | Explanatory | Local | Visual estimate of percent cloud cover during roost surveys |
Roosting structure height | Explanatory | Local | Clinometer and rangefinder measurements |
Surrounding vegetation height | Explanatory | Local | Clinometer and rangefinder measurements |
Open space | Explanatory | Local | Google Earth and rangefinder measurements |
Developed land cover | Explanatory | Landscape | 2016 National Land Cover Database (Yang et al., 2018) |
Open Water land cover | Explanatory | Landscape | 2016 National Land Cover Database (Yang et al., 2018) |
Developed-Forest edge density | Explanatory | Landscape | 2016 National Land Cover Database (Yang et al., 2018) |
Deer carcass density | Explanatory | Landscape | North Carolina Department of Transportation and Wildlife Resources Commission (NCDOT, 2018), South Carolina Department of Public Safety (SCDPS, 2018), and 2018 TigerLine road shapefile (U.S. Census Bureau, 2018) |
Survey date | Explanatory | NA | NA |
We collected data on air temperature and wind speed during each survey from Weather Underground Personal Weather Stations (PWS) (Weather Underground 2021) located near roosting sites. For each roosting site and survey, we used the PWS located nearest to the site and the air temperature and wind speed measurements at the start of the survey. During each survey, we also visually estimated percent cloud cover using a scale from 0-100% in 10% increments.
In January of the year each roost was first surveyed, we measured the height of each roosting structure and surrounding vegetation using a Suunto PM-5/360 clinometer and Nikon Aculon rangefinder. Using the clinometer, we measured the angle of elevation from the viewpoint to the top of the roosting structure or the top of the trees surrounding the roosting structure. We defined the top of the roosting structure as the typical location of the highest roosting vulture. For roosts located on transmission towers, we took the average of the angle to the trees on either side of the ROW corridor. For all other roosts, we used the trees or other vegetation closest to the roosting structure. With the rangefinder, we measured the distance from the viewpoint to the base of the roosting structure or the base of the vegetation sampled with the clinometer. Heights were calculated by multiplying the angle tangents by distances.
We measured open space using Google Earth and confirmed the distance with a Nikon Aculon rangefinder. For roosts on transmission towers, open space was the width of the ROW corridor at the level of the tower. For other types of roosting structures, we measured multiple locations around the roosting structure and calculated the average width of open space surrounding the structure.
Landscape variables We chose four landscape variables to explain vulture attendance at roosts: amount of Developed land cover, amount of Open Water land cover, Developed-Forest edge density, and deer carcass density (Table 1). Developed land cover provides roosting and nesting structures and food sources that appear to attract vultures (Campbell 2014; Novaes and Cintra 2015; Thompson et al. 1990). Open water provides food and water for vultures. Vultures have also been found to frequently use edge habitat between open and forested areas, perhaps to benefit from the proximity of cover and foraging opportunities (Coleman and Fraser 1989; Novaes and Cintra 2015). Finally, previous research has found food availability to be a main predictor of the location of vulture roosting sites (Campbell 2014; Novaes and Cintra 2015), with roadkill potentially being an important food resource for vultures in urban landscapes (Thompson et al. 1990).
Each variable was measured in landscapes with radii of 0.4, 0.5, 1, 2, 3, 4, 5, 10, 15, and 20km centered on roosting sites. Landscape scales encompassed variation in estimates of space use by black and turkey vultures. Holland (2015) reported a maximum core area size of 0.45km2 for black vultures and 0.42km2 for turkey vultures, maximum home range size as 30-60km, and Houston et al. (2011) reported turkey vulture home ranges over 900km2.
We measured the amount of developed land cover and the amount of open water land cover in landscapes as the proportion of each landscape covered by the Developed classes or the Open Water class of the 2016 National Land Cover Database (Yang et al. 2018). We measured the Developed-Forest edge density of each landscape as the length of edge between any Developed class (Open space, Low intensity, Medium intensity, and High intensity) and any Forest class (Deciduous, Evergreen, Mixed) of the 2016 National Land Cover Database divided by landscape area.
We used 2018 TigerLine road data (U.S. Census Bureau 2018) and 2018 white-tailed deer (Odocoileus virginianus) collision data (NCDOT 2018; SCDPS 2018) to estimate deer carcass density in landscapes. For landscapes in North Carolina where deer collisions are tracked by road segment, we first calculated the number of deer collisions and total road length, using only road types on which deer collisions were recorded. The number of deer collisions was then divided by total road length and then again by landscape area. In contrast, the South Carolina Department of Transportation tracks the number of deer collisions by county. We therefore had to estimate the number of deer collisions in landscapes in South Carolina by multiplying the number of deer collisions in each county in a landscape by the areal proportion of the county overlaid by the landscape, summing the resulting deer collisions across counties, and dividing by landscape area. Deer carcass densities for landscapes that overlapped the state boundary were multiplied by the proportional area of the landscape in the state prior to averaging. Landscape variables were calculated using ArcGIS Pro, v2.5.0 (ESRI 2020) and FRAGSTATS, v4.2.1.603 (McGarigal et al. 2015).
Analyses
During the 2019-2020 roost survey season, we observed that vultures often left roosting sites earlier than expected – in some cases, most vultures were gone by 30 minutes after sunrise. Starting in March of 2020, we conducted surveys within only 30 minutes before sunrise in order to ensure an accurate count of roosting vultures for the remainder of the study.
We corrected for any bias resulting from vultures leaving roosting sites earlier than expected by conducting additional counts of roosting vultures in relation to time before and after sunrise to estimate the rate at which vultures leave roosting sites. The lead author counted the number of vultures roosting every minute from 60 minutes prior to sunrise to 60 minutes after sunrise at eight different roosting sites over 16 days from September 2019-March 2020. We then used a 6th order polynomial regression to model the proportion of the roost remaining with respect to time before and after sunrise (F6,1042 = 262.30, p < 0.001, adjusted R2 = 0.60; Figure A1). All roost counts from 2019-2020 and 2020-2021 were adjusted by dividing the count by the predicted proportion of the roost remaining at the time the count was conducted.
We identified the local and landscape features associated with the adjusted number of vultures at roosts using repeated measures, generalized linear mixed-effects models, multi-model inference, and simultaneous autoregressive models. All models included a random site effect to account for the non-independence of observations from the same roosting site. Models also included survey date, measured as Julian date, i.e., the number of days since the beginning of the Julian period in 4713 BC, to account for variation in roost attendance across months and years (Sweeney and Fraser 1986). We restricted landscape variables in models to those from the same spatial scale in order to minimize high levels of collinearity among explanatory variables. Collinearity among local and landscape variables in models was below thresholds above which levels are deemed unacceptable (r < 0.70 and VIF < 10 (Dormann et al. 2013); Appendix A), except for the correlations between Developed-Forest edge density and Developed land cover within 10km, 15km, and 20km (0.72 ≤ r ≤ 0.79) (Tables A1-A11). We divided explanatory variables by their partial standard deviations (standard deviations divided by VIF values, sample size, and the number of predictor variables (Cade 2015)) and we also standardized explanatory variables. Finally, we log-transformed the response variable in models to address heteroskedasticity in residuals.
We evaluated models containing all possible combinations of explanatory variables, with the exception that models contained only landscape variables from the same spatial scale, using AICc. As roosting sites were spatially clumped in our study area, we tested for spatial autocorrelation in the residuals of the best models (ΔAICc < 2) (Burnham and Anderson 2002) using correlograms, Moran’s I, and Bonferroni-corrected p-values. Most residuals exhibited significant spatial autocorrelation (Figures A2-A8). To account for this source of variation, we re-analyzed the top models as simultaneous autoregressive models of the spatial error type (SARerr) using the distance at which spatial autocorrelation was most pronounced in correlograms to define neighborhoods.
We calculated model-averaged effect sizes of explanatory variables in SARerr models and their unconditional variances (Burnham and Anderson 2002). All effect sizes for each explanatory variable were unimodal (Cade 2015). We used RStudio v1.3.1073 (RStudio Team 2020) and the MuMIn v1.43.17 (Barton 2020), car v3.0-11 (Fox and Weisberg 2018), spdep v1.1-11 (Bivand and Wong 2018), ncf v1.2-9 (Bjornstad 2020), lme4 v1.1-27.1 (Bates et al. 2014), spatialreg v1.1-8 (Bivand and Piras 2015) and ggplot2 v3.3.5 (Villanueva and Chen 2019) packages to carry out analyses.