Arendt et al.: Temporal trends and influences on fishery-independent catch rates for Caretta caretta in an important coastal foraging region 473 
years. Catch was examined annually for the overall 
data set and for 5-cm size classes between 40.1-45.0 cm 
and 100.0-105.0 cm SCLmin that had at least 100 total 
captured loggerhead sea turtles. For 23 turtles with 
posterior carapace injuries, ad hoc assignment to 5-cm 
size classes was made with the use of paired (SCLmin 
and straight-line carapace width [SCW]) measurements 
for loggerhead sea turtles (1230) captured in this sur- 
vey. The paired measurements were used in this calcu- 
lation: SCLmin=(1.42xSCW) - 10.5 (F= 10,579; P<0.001; 
coefficient of determination [r 2 ] = 0.90). Catch data were 
fitted to a negative binomial distribution and analyzed 
through the use of a generalized linear model with a 
log link function, after exclusion of 2% of attempted 
sampling events because of tow times (55 events) that 
were not ±95% of the target trawl duration (20 min in 
2008-09, 30 min in all other years) or where suspect 
trawl start or end locations could not be resolved (42 
events). Thirteen additional sampling events were ex- 
cluded because data were missing for them for at least 
1 of 25 model terms. 
Two of the included model terms were temporal: year 
and time of day at the start of each trawling event 
(1=<0959 h local standard time [ LST] ; 2 = 1000-1259 
h; 3 = 1300-1559 h; 4 = >1600 h). However, two stan- 
dard temporal terms (i.e., season and day of year) were 
not included in the model. Such temporal terms were 
excluded because this survey was conducted within 1 
month of the summer solstice (i.e., a peak and stable 
photoperiod) and nearly 2 months after juvenile log- 
gerhead sea turtles return to nearshore coastal waters 
in this region (Arendt et al., 2012b). Another reason for 
their exclusion was a spatiotemporal bias in sampling 
due to staggered vessel start dates. 
Six spatial model terms were used. One of these pa- 
rameters consisted of geographic subregions: l=Win- 
yah Bay to Charleston, South Carolina; 2 = Charleston, 
South Carolina, to Savannah, Georgia; 3=Brunswick to 
Savannah, Georgia; 4=St. Augustine, Florida, to Bruns- 
wick, Georgia. The other spatial terms were minimum 
distance from shore (in kilometers) at the start of each 
trawling event (determined with ArcGIS Arclnfo 10.0); 
distance (in kilometers) and bearing (in degrees) from 
the closest of 31 estuary inlets within the study area; 
trawl transect bearing (in degrees), computed with 
Pythagorean theorem; and seafloor type assigned by 
co-occurrence of <1, 2, or >3 of 56 hard-bottom indica- 
tor species 2 ’ 3 . 
Six environmental parameters were measured in situ 
at the start of each trawling event, several of which are 
known to influence spatial distributions of loggerhead 
sea turtles in pelagic habitats (Baez et al., 2007; Man- 
sfield et al., 2009; Kobayashi et al., 2011). Sea-surface 
temperature (SST, in degrees Celsius) was measured 
by the ship’s transducer or, in 2000 and 2001, read by 
a digital thermometer for a bucket of surface water. 
Mean water depth (in meters) was recorded by a fa- 
thometer at the start and end of each trawling event, 
after which the relative change (in percentage) between 
the start and end locations was recorded. Wind velocity 
(in knots) and direction were recorded with a shipboard 
anemometer. The relative distribution of cloud cover (in 
percentage) across the entire dome of sky also was esti- 
mated. Wind direction was recorded as text at sea but 
later converted to numeric values in this manner: north 
(0°), north-northeast (22.5°), northeast (45°), etc. Where 
wind velocities were recorded as calm, wind direction 
was assigned to the direction recorded just before winds 
became calm. Hourly SST data from the buoy at Gray’s 
Reef National Marine Sanctuary (GRNMS) (station 
41008; http://www.ndbc.noaa.gov;) were used for substi- 
tutions for 539 trawling events with missing SST given 
±10% agreement for 95% of 3100 paired observations 
from both data sets. 
Seven model terms were generated through the use 
of external data sets and included 8-day compilations 
of chlorophyll-a (Chl-a, in miligrams per cubic mililiter; 
http://disc.sci.gsfc.nasa.gov/giovanni/overview/index. 
html#) at resolutions of 9 km (Sea-viewing Wide Field- 
of-view Sensor [SEAWIFS]; 2000-02) and 4 km (Moder- 
ate Resolution Imaging Spectroradiometer, Aqua satel- 
lite [MODIS-A]; >2003) for the observation closest (8.5 
±18.9 km; mean ±standard deviation [SD]) to the trawl- 
ing event midpoint; daily mean and change in baromet- 
ric pressure (milibars) recorded hourly at the GRNMS 
buoy; monthly North Atlantic Oscillation (NAO) in- 
dex values from the NOAA Climate Prediction Center 
(http://www.cpc.ncep.noaa.gov/products/precip/CWlink/ 
pna/nao.shtml); tide stage (0=ebb, l=flood) and range 
(in meters) in water level between high and low water 
during the tidal event when sampling was conducted, 
determined from hourly data at National Ocean Service 
gauges near Winyah Bay, South Carolina (8662245), 
Charleston, South Carolina (8665530), Savannah, Geor- 
gia (8665530), and Mayport, Florida (8670870); and 
random scrambling of event order (H 6 = 2.64, P= 0.852 
by year) to evaluate the model (Kobayashi et al., 2011). 
In addition to the model and offset terms described 
above, null models also contained up to 4 interaction 
terms (mean depth vs. distance from shore; mean depth 
vs. distance from inlet; distance from inlet vs. distance 
from shore; year vs. NAO index) identified with a cor- 
relation test (Pearson’s coefficient of correlation [/'] >0.4) 
performed in R. As such, terms in the null model were 
analyzed in the following order: Loggerhead count=year 
+ NAO + ( year*NAO ) + subregion + time of day + mean 
depth + distance from shore + trawl depth change + 
distance from inlet + ( mean depth* distance from shore ) 
+ (mean depth* distance from inlet) + ( distance from 
shore* distance from inlet ) + bottom type + cloud cover + 
wind velocity + wind direction + daily mean barometric 
pressure + interdaily change in daily mean barometric 
pressure + bearing from inlet + tide stage + tide range 
+ transect bearing + Chl-a + SST + random order + 
log(transect length). 
Final model selection was accomplished through step- 
wise regression based on the lowest Akaike’s infor- 
mation criterion (AIC) score. A chi-square analysis of 
deviance was performed in R to assess the statistical 
significance of variables retained in the final model. 
