Underwood et al.: Behavior-dependent selectivity of Limanda ferrugmea in the mouth of a bottom trawl 
433 
Video analysis 
Analysis of the video footage was conducted in the 
laboratory by using Observer XT software, vers. 10.1 
(Noldus Information Technology, Wageningen, Nether- 
lands). A grid of 100 squares was placed over a 1080p 
high-definition monitor, and the use of that grid made 
it possible to provide information on where a fish was 
in relation to the gear (Fig. IB). Our approach was 
similar to that of Albert et al. (2003), but we increased 
the number of squares in the grid from 49 to 100 to 
more accurately record the location of individual fish 
in relation to the footgear. 
A square within the grid was selected from a list 
of randomly generated numbers and, while the video 
footage was playing, the behavior sequence of the first 
individual fish seen in that square was recorded. If the 
selected square included the trawl gear, then the next 
grid square on the list was selected. Only behaviors for 
individuals seen resting on the substrate were recorded 
because it was unclear whether a fish seen swimming 
into a frame had interacted with the sweeps or foot- 
gear. To reduce autocorrelation, observations were re- 
stricted to following a single fish in the video footage 
at any given time. After a sequence was analyzed, play- 
ing of the video footage was stopped, and the next grid 
square was selected from the list of randomly generat- 
ed numbers. The process was repeated until the footage 
ended or until it was impossible to identify individuals 
on or in the substrate because of reduced natural light 
or the presence of sand clouds. The video footage was 
reviewed a second time to identify segments greater 
than 30 s in duration that had not been evaluated pre- 
viously. The additional observations collected from this 
second round of analysis were added to the data set. 
Individual flatfishes were categorized as either yel- 
lowtail (identified by their pointed snout and small 
mouth; Collette and Klein-MacPhee, 2002) or as un- 
identified. The analysis of video footage was limited to 
yellowtail because of the dominance of this species in 
the footage, but the numbers of unidentified flatfishes 
were included in values for the “start density” category, 
which is described later. 
Categorical variables used for analysis (Table 2) 
were derived from similar behavioral studies (e.g., 
Walsh and Hickey, 1993; Albert et al., 2003; Piasente 
et ah, 2004; Ryer and Barnett, 2006). Location of an 
individual in relation to the footgear was recorded at 
the start of the observation and categorized into the 
following 3 groups. Individuals within 2 squares of and 
on either side of the center of the footgear were catego- 
rized as in the “middle” of the footgear. Individuals ob- 
served greater than 2 squares to the port side or star- 
board side of the center of the footgear were classified 
as “port” and “starboard”, respectively (Fig. IB). The 
orientation of an individual fish on or in the substrate 
was recorded at the start of each observation (i.e., be- 
fore the individual rose from the seabed), and swim- 
ming direction was recorded when a fish left the seabed 
(i.e., displayed initial behavior; Table 2). Previous gear 
experience was assumed to influence the orientation of 
an individual fish, and peripherally located individu- 
als (i.e., those not in the 4 middle squares, Fig. IB) 
that were facing inward (i.e., individuals on the port 
side facing starboard and vice versa) were recorded as 
“previously herded.” 
Fish length was estimated on the basis of the known 
dimensions of footgear components (one rockhopper disc 
and spacer together measured 30 cm in width) within 
the field of view that corresponded with the minimum 
legal size of yellowtail (30 cm). Measurements were 
taken when a single fish was close to the footgear, and 
each fish was then classified as being either larger or 
smaller than 30 cm. Individuals that were close to the 
reference length (-28-32 cm) or that were not visible 
or close to the footgear were grouped as “unmeasured.” 
Given that fish of different sizes swim at different 
levels within their swimming performance range, the 
choice of gait used by each fish was also recorded (Ta- 
ble 2; Webb, 1994; see review by Winger et al. [2010]). 
Responses of flatfishes to the footgear and sweeps 
had been classified into the 4 categories “pass under,” 
“hop,” “rise,” and “run” in previous studies (Ryer and 
Barnett, 2006; Ryer, 2008; Ryer et al., 2010; Table 2). 
We adopted this classification and added a fifth cat- 
egory, “slope.” After leaving the seabed, the swimming 
behavior of individual fish was classified into these 5 
categories of “initial behavioral response” (Table 2). 
Run and slope led to the initiation of herding by the 
footgear, and the behavioral responses of the other 3 
categories were seen as nonherding responses (Ryer 
et ah, 2010). If a subsequent change in the initial re- 
sponse of an individual was observed, then it was noted 
in “change in response” (Table 2) and the second behav- 
ioral response was recorded. The response of individu- 
als that maintained their initial behavioral response 
was recorded as “continued.” The capture outcome of 
each individual was recorded as “escaped” or “captured” 
and the method of escapement or capture was noted 
(i.e., “actively entered or sought escapement,” “overtak- 
en,” or “collided with the footgear”). 
The time, in seconds, from the point when an in- 
dividual left the seabed until it passed over or under 
the footgear was recorded as the residence time. Total 
flatfish densities, estimated as the number of station- 
ary and moving flatfishes in each video frame, were 
recorded at the start of each observation ( start density). 
After all video footage was analyzed, the behaviors of 
190 yellowtail were available for statistical analysis, 
representing approximately 1% of the total yellowtail 
catch from the 5 tows. 
Statistical analysis 
We concentrated on 4 main areas of analysis, look- 
ing at the influence of multiple variables on ori- 
entation (model 1: orientation=location), initial re- 
sponse (model 2: initial response=location+swimming 
direction+length+gait+start density+tow [random fac 
tor]), change in response (model 3: change in re 
