Beason et al. • BLACK SWIFT MIGRATION AND WINTERING AREAS 
3 
encapsulated in a water-resistant housing with 
two external terminals for commands and data 
transfers. 
We designed a backpack harness system 
modified from Buehler et al. (1995) because 
Black Swift legs are too attenuated for the leg- 
loop harness often used for geolocators on 
passerines (Rappole and Tipton 1991). The 
harness material. 5 mm tubular Teflon ribbon 
(Bally Ribbon Mills. Bally. PA, USA), was 
attached at four points to the geolocator and 
crossed under the keel. We secured the free ends 
of the ribbon with size 69 bonded right twist 
Kevlar thread (The Thread Exchange Inc., Wea- 
verville. NC. USA) and stitched the ribbon where 
ii crossed the keel to avoid shifting. We applied 
cyanoacrylate glue on all stitches and cut ends to 
prevent fraying. 
We attached geolocators to four adult Black 
Swifts in August 2009. three al Fulton Resurgence 
Cave (2 females. I male) and one at Box Canyon 
Falls (male); the birds weighed 49.5 51.5 g. Each 
geolocator, including harness materials, weighed 
1.5 g. representing 2.9-3% of body weight, well 
within guidelines suggested by Caccamise and 
Hedin (1985). 
Delta Analysis .—We conducted pre-deployment 
calibration for ~9 days and post-deployment 
calibration for ~7 days by placing them at a 
known location with a dear view' of the sky. We 
retrieved Ihree of Ihe four geolocators in July and 
August 2010. 
We used software programs (BASTrak) devel¬ 
oped by BAS to download, process, and interpret 
data archived by the loggers, each of which had 
collected data throughout their entire deployment. 
We rejected latitude data gathered ~30 days 
before and after the equinoxes since day lengths at 
the equinoxes are equal at all latitudes, resulting 
in poor location fixes. Internal clocks maintained 
accuracy during deployment and there was no 
need to correct for clock drift. 
Two values are required for analyzing and 
plotting the geolocator data: the dusk/dawn light 
transition threshold and the corresponding sun 
elevation angle at this threshold. We chose a 
sensitive light transition threshold value of two to 
reduce variation in day length due to the effects of 
shading which influences the resulting distribution 
of location fixes. We used static pre-deployment 
calibration to calculate the corresponding sun 
elevation angles (-6.4 , -6.5 , and -6.6 ). We 
calculated times of sunrise and sunset using 
TransEdit2; positions were calculated with Bird- 
Tracker which derives longitude from absolute 
time of local midday/midnight and calculates 
latitude by comparing day/night length, a tech¬ 
nique which provides two geographical positions/ 
day. Wc used only midnight fixes to produce 
maps based on the assumption that sw'ifts were 
roosting at night and migrated during die day. We 
identified days with irregular shading events, 
resulting in shorter day lengths or anomalous 
transition limes, by visually inspecting sunrise and 
sunset times and excluded them from the analysis. 
An average of 145 fixes for each bird remained to 
map wintering range and an average of 26 fixes 
remained to map the spring migration path. 
Mapping of fall migration was not possible due 
to overlap with the fall equinox. 
We calculated kernel density surfaces using the 
wintering area data from each geolocator with the 
Spatial Analyst Kernel Density function (ESRI 
2009). This function calculates density of fixes in 
a search radius around those fixes. These densities 
lit a smoothly curved surface over each location. 
The surface value was highest at the location of 
the point and diminished with increasing distance 
from the point. We used a fixed kernel with a 
search radius of 185 km to compensate for the 
approximate average error in latitude and longi¬ 
tude known to occur in geolocator data (Phillips et 
al. 2004). The kernel function is based on the 
quadratic kernel function described in Silverman 
(1986). We calculated the density surfaces at 
1-km resolution as this is adequate to capture 
density at a small scale over a large geographic 
area. We calculated 90%. 75%. and 50% density 
polygons from the kernel density surfaces to 
enhance graphic displays ot higher use density 
areas. We used the average nearest-neighbor 
distance function in Arclnfo Spatial Statistics 
(ESRI 2009) to characterize the spatial point 
pattern of winter locations. This function quanti¬ 
fies and characterizes the spatial pattern of each 
geolocator and indicates if the pattern is evenly 
dispersed, random, or clustered compared to a 
spatial random distribution. We estimated approx¬ 
imate migration duration, arrival, and departure 
events from plotting longitude and date. We used 
the 50% kernel density polygons for 
all three swifts to describe land cover use and 
overlaid those with a global land cover layer usine 
2009 satellite imagery at a 300-m resolution 
produced by the European Space Agency Glob- 
Cover 2009 Project (Bontemps et al. 2010). 
