Campbell et al.: Attraction and avoidance behaviors of fish in response to proximity of underwater vehicles 225 
Standardized count 
25 
Vehicle range (m) 
Figure 8 
Change in standardized fish counts observed at selected 
reef sites in the Florida Middle Grounds of the down mov- 
ers guild as a function of vehicle range (meters) and by 
vehicle type. Survey vehicles used in this experiment were 
an autonomous underwater vehicle (AUV), a remotely 
operated vehicle (ROV), and a towed vehicle (TV). Predic- 
tions were made by using the GAM predict function in the 
R package mgcv, and parameters were set in the models as 
follows: first transect transited, the relative vehicle alti- 
tude from the seafloor was set at the middle level (2 m), 
and habitat complexity was set at the average qualitative 
rating (2.5). The gray shaded areas represent the 95% con- 
fidence intervals. Surveys were conducted during August 
2014 and July and August 2015. 
relative abundance for the AUV and ROV but indicates a 
decreasing trend in relative abundance for the TV. 
From a qualitative perspective, relative abundance of 
the local reactive guild tended to change little in rela- 
tion to vehicle passage (Fig. 9). However, observations on 
camera indicate that individuals of this group tended to 
increase activity with a vehicle present but did not flee 
the sampling volume in front of the MOUSS platforms 
(i.e., locally reactive). For the local reactive guild, results 
from the TV model indicate significant effects on relative 
abundance for RVA, transect number, habitat complexity, 
and vehicle range. The TV model explained 11.6% of the 
deviance with an r” of 0.17 (Table 2). Results from the 
ROV model indicate significant effects on relative abun- 
dance for RVA, transect number, habitat complexity, and 
vehicle range. The ROV model explained 29.7% of the 
deviance with an r” of 0.42. The AUV had too small of a 
sample size to evaluate this group. Vehicle altitude could 
only be evaluated for the TV and was negatively related 
to relative abundance. 
Discussion 
Species-specific gear interactions were difficult to capture; 
however, various patterns of fish response to the vehicles 
allowed classification of species into behavioral guilds. 
Standardized count 
25 
0 
Vehicle range (m) 
Figure 9 
Change in standardized fish counts observed at selected 
reef sites in the Florida Middle Grounds of the local reac- 
tive guild as a function of vehicle range (meters) and by 
vehicle type. Survey vehicles used in this experiment were 
an autonomous underwater vehicle (AUV), a remotely 
operated vehicle (ROV), and a towed vehicle (TV). Predic- 
tions were made by using the GAM predict function in the 
R package mgcv, and parameters were set in the models as 
follows: first transect transited, the relative vehicle alti- 
tude from the seafloor was set at the middle level (2 m), 
and habitat complexity was set at the average qualitative 
rating (2.5). The gray shaded areas represent the 95% con- 
fidence intervals. Surveys were conducted during August 
2014 and July and August 2015. 
The use of GAMs proved to be an effective method to 
examine changes in relative abundance of fish as a func- 
tion of vehicle range and relative altitude, transect num- 
ber, and habitat complexity. The effect of each variable is 
dependent on the behavioral guild of interest, but vehi- 
cle range was a consistent predictor of relative changes 
in abundance regardless of vehicle and behavioral guild. 
In general, the best fitting models were associated with 
the strongest behavioral responses, such as the attraction 
behavior observed for the pelagic pursuers guild and the 
avoidance behavior observed for the down movers guild. 
Finally, although logistically difficult to conduct, the test- 
bed method we used to evaluate fish responses to sam- 
pling gear was effective and potentially will provide a path 
forward to conducting gear calibrations in the future. 
The consistency of vehicle range in the GAMs allowed 
the development of functional relationships between 
vehicle range and the change in relative abundance in the 
MOUSS platform sampling volume (Figs. 4-9). Although 
many studies have recognized or coarsely evaluated fish 
responses to sampling vehicles (Ralston et al., 1986; 
Richards, 1986; Koslow et al., 1995; Trenkel et al., 2004a, 
2004b; Lorance and Trenkel, 2006; Ryer et al., 2009), no 
studies have included estimation of a functional rela- 
tionship that predicts relative change in abundance to 
an approaching vehicle (i.e., far-field response). The test- 
bed approach proved to be a robust method of deriving 
