Demetras et al.: Use of underwater recorders to quantify predation of juvenile Oncorhynchus tshawytscha 
181 
trieving ten PERs by a 2-person boat crew ranged from 
90 to 115 minutes. 
Owing to extremely low flows during our study pe- 
riod in the spring of 2014, the lower San Joaquin River 
was under direct tidal influence over the course of the 
study and experienced a mixed, semidiurnal tidal pat- 
tern. The tidal nature of the San Joaquin River during 
this period required extra effort to determine the cor- 
rect mid-channel placement of the PERs so that they 
would remain within the site for approximately 45 
minutes or longer. If a PER did not remain within a 
study site for at least 45 minutes, or became beached 
or otherwise fouled, it was promptly retrieved, re-bait- 
ed and redeployed within the study site. 
Data processing and analysis 
PER GPS transponders recorded a location every 5 sec- 
onds, whereas predation timers recorded the timing of 
predation events. By cross-referencing predation data 
from the predation timer (time of predation) with PER 
GPS data (time/latitude/longitude) we were able to 
obtain locations of each predation event. GoPro video 
footage was captured with a widescreen aspect ratio 
of 16:9, resolution 1920x1080 (1080p HD “Superview”), 
at 30 frames per second at the low light setting. Each 
camera produced on average approximately 12 to 20 
gigabytes of data per deployment depending on indi- 
vidual PER sampling time. Video foot- 
age was later viewed to confirm pre- 
dation events and to identify predator 
species. 
The relationship between survival 
of tethered smolt, exposure time, and 
environmental factors was modeled 
with a Cox proportional-hazards re- 
gression for censored data (Cox, 1972) 
by using the Olsurv package, vers. 0.2 
(Diez, 2013) in R statistical software, 
vers. 3.2.0 (R Core Team, 2015). Be- 
fore model construction, we examined 
correlation coefficients of candidate 
covariate pairs to identify collinear- 
ity and only included one variable of 
a pairwise comparison that had cor- 
relations greater than 0.7 (Dormann 
et al. , 2013). The candidate covariates 
for the model were total distance trav- 
eled (m), median light intensity (lux), 
median depth (m), standard deviation 
of depth (m), median water tempera- 
ture (° C), and median water velocity 
(ms -1 ). Akaike’s information criterion 
(AIC; Burnham and Anderson, 2002) 
was used to select the most parsimoni- 
ous model with the best fit to the data 
in a forward and backward step-wise 
fashion. Model residuals were exam- 
ined to evaluate the model fit. 
Results 
We conducted 216 PER deployments between late 
March and late May 2014. Of the 216 deployments, we 
recorded 33 total predation events (15%), 12 of which 
were captured on video by the GoPro camera. Through- 
out the study we were able to easily combine the timer 
data with the corresponding GPS data to produce ac- 
curate maps of PER pathways and predation event lo- 
cations within the study site (Fig. 2). 
Water conductivity and water velocity were collinear 
at r=— 0.75. Water conductivity was excluded, however, 
from the analysis because it was within the physiologi- 
cal range of both juvenile Chinook salmon and preda- 
tors and was assumed to have minimal impact on their 
ability to forage. AIC model selection indicated that 
water velocity and median depth best explained varia- 
tion in predation rate. The coefficient for water velocity 
was 2.3 and median depth was -0.7. The exponentiated 
coefficient for water velocity was 9.6 and median depth 
was 0.5. Exponentiated coefficients are interpretable as 
multiplicative effects on the hazard. For example, by 
holding the median depth constant, an additional me- 
ter per second increase in water velocity increases the 
minute-by-minute hazard of predation by a factor of 
9.6. Similarly, each increase in median depth decreases 
the hazard by a factor of 0.5. The likelihood-ratio [LR] 
