Lear et al.: Fine-scale behavior and habitat use of Pristis pect/nata 
351 
traces (Fig. 3A) were removed from the 
burst event mask for each individual, 
leaving only burst events designated by 
large-amplitude and high-frequency lat¬ 
eral body movements for further analysis 
(Fig. 3, B and C). Mean depth, mean tem¬ 
perature, percentage of time spent active, 
and total number of burst events were 
summarized for each hour of the deploy¬ 
ment for each individual. 
Generalized additive mixed models 
(GAMMs) were built by using the mcgv 
package (vers. 1.8-28; Wood et al., 2017) 
in R, vers. 3.5.2 (R Core Team, 2018), and 
were used to analyze patterns in activ¬ 
ity and bursting with respect to 5 fixed 
effects: depth, temperature, time of day, 
tide, and age class. Tidal period was split 
into 6 discrete phases—high, low, first 
ebb, second ebb, first flood, and second 
flood—by using tide data reported by 
NOAA’s National Ocean Service for the 
Peace River (available from the Tides 
and Currents website). Periods of low 
and high tide were designated as the 2 h 
surrounding the minimum and maxi¬ 
mum tide height within a cycle, and each 
bridging ebb and flood period was evenly 
divided into a first and second phase, 
with each tidal phase lasting approx¬ 
imately 2 h. Smalltooth sawfish were 
split by age class into either a group for 
young of the year (YOY; <1.5 m STL) or 
a group for individuals age 1 or older 
(>1.5 m STL) (Scharer et al., 2012), with 
4 individuals in each group (Table 1). 
To determine informative drivers of 
activity, a series of GAMMs were built for 
predicting the percentage of time saw¬ 
fish were active by using different com¬ 
binations and interactions of the 5 fixed 
effects. The Akaike information criterion 
corrected for small sample sizes (AICc) 
and log likelihood of each model were 
used to determine the best-fit model 
and the predictors that should be main¬ 
tained in the model. The number of burst 
events observed in each hour was highly 
zero inflated; therefore, to describe pat¬ 
terns in burst events, a 2-step process 
was used following methods described 
by Gleiss et al. (2017). First, a series of 
GAMMs with binomial distributions was 
built to determine which fixed effects 
affected the probability of a burst event. 
Similar to what was done for the model 
for activity, the fit of different combina¬ 
tions of these predictor variables was 
assessed by using AICc, and the model 
£? B 
13 
co 
0) 
Q 
CO 
33 
CO 
CL 
CD . 
□ 
< 
m 
Q 3 
O 
w 
CO 
-Q 3} 
1 
08:08:30 
1 
08:08:45 
1 
08:09:00 
1 1 1 
■ ■ 
■1 
1.0 
22:29 
22:30 
22:31 
22:32 
22:33 
Q _ 
wwAJL/iAv I 
VW^WlAs%vX» 
i r 
Figure 3 
(A) Dynamic acceleration traces and overall dynamic body acceleration 
(ODBA) data that show a typical chafe event (dorsal rubbing) for the small¬ 
tooth sawfish (Pristis pectinata), characterized by a distinctive W shape in the 
heave acceleration axis. These events were excluded from the burst event mask 
for each individual. Burst events included in the mask were generally charac¬ 
terized by either (B) short bursts accompanied by sharp ascents or (C) longer 
bursts without distinct changes in depth. Data used in these graphs were col¬ 
lected from acceleration data loggers deployed on smalltooth sawfish caught 
and tagged in the Peace River, Florida, between May 2014 and November 2015. 
