Campbell et al.: Attraction and avoidance behaviors of fish in response to proximity of underwater vehicles 
Table 2 
223 
Generalized additive model run results by behavioral guild and survey vehicle, used to assess changes in the relative abun- 
dance of fish due to vehicle passage at sites in the Florida Middle Grounds during August 2014 and July and August 2015. 
Vehicles evaluated included an autonomous underwater vehicle (AUV), a towed vehicle (TV), and a remotely operated vehicle 
(ROV). P-values are provided for each variable included in the models: relative vehicle altitude above the seafloor (RVA), tran- 
sect number, habitat complexity, and vehicle range. Also given are the sample size (n), percentage of deviance explained, and 
coefficient of determination (r”). Asterisks (*) indicate significant variables (P<0.05). NA indicates a variable that was not used 
in a particular model run. 
Guild Vehicle 
Pelagic AUV 
pursuers TV 
ROV 
Benthic AUV 
pursuers TV 
ROV 
Down 
movers 
Lateral 
movers 
Local 
nm 
217 
527 
62 
279 
1798 
1919 
186 
744 
1921 
217 
1085 
1767 
Intercept 
0.10 
<2.0 x 10716 * 
NA 
0.55 
1.3 x 10° * 
0.20 * 
RVA 
8.8 x 107! * 
0.82 
NA 
2.0 x 107° * 
3.4 x 107! * 
3.9 x 107° * 
2.0 x 1074 * 
6.8 x 10°74 * 
<2.0 x 10716 
6.4 x 10° 
1.2 x 10° 
3.4 x 107? 
<2.0 x 10°16 
Transect 
number 
1.4 x 10> * 
2.2 x10 * 
NA 
0.55 
<2.0 x 10°16 * 
Lil 3 IO 
4.8x 107! * 
<2.0 x10716 * 
1.9 x 107" * 
5.0 x 1074 * 
Be Ue & 
2.0 x 1073 * 
<2.0 x 10°16 * 
Habitat 
complexity 
0.56 
1.3 x 10°? * 
0.46 
5.0 x 1074 * 
<2.0 x 10716 * 
<2.0 x 10716 * 
<2.0 x 10716 * 
<2.0 x 10°16 * 
<2.0 x 1071° * 
1.9 x 107° * 
<2.0 x 10716 * 
<2.0 x 1071° * 
<2.0 x 10°16 * 
Vehicle 
range (m) 
1.2 x 10°* * 
<2.0 x 10716 * 
6.7 x 10°°* 
6.2 x 10°? 
2.2 x 1071! * 
1:7 x 105° * 
0.76 
<2.0 x 10°16 * 
0.03 * 
0.42 
<2.0 x 10°16 * 
1.4x 10° * 
1.8 x 1077 * 
Deviance 
explained 
48.9% 
59.3% 
81.1% 
46.7% 
34.0% 
13.9% 
98.0% 
60.8% 
18.6% 
26.8% 
38.3% 
11.6% 
29.7% 
reactive 1112 NA 
Standardized count 
0 25 
Vehicle range (m) 
Figure 5 
Change in standardized fish counts observed at selected reef 
sites in the Florida Middle Grounds for the pelagic pursuers 
guild as a function of vehicle range (meters) and by vehicle 
type. Survey vehicles used in this experiment included an 
autonomous underwater vehicle (AUV), a remotely oper- 
ated vehicle (ROV), and a towed vehicle (TV). Predictions 
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 altitude 
from the seafloor was set at the middle level (2 m), and hab- 
itat complexity was set at the average qualitative rating 
(2.5). The gray shaded areas represent the 95% confidence 
intervals. Surveys were conducted during August 2014 and 
July and August 2015. 
From a qualitative perspective, the benthic pursuers 
group exhibited strong attraction to all of the vehicles 
tested but in general were not fast enough to keep up with 
a transiting vehicle. Therefore, members of this guild 
tended to only briefly chase vehicles following passage 
(Fig. 6). For the benthic pursuers guild, results from the 
AUV model indicate significant effects for RVA and habi- 
tat complexity, but vehicle range was marginally signifi- 
cant and transect number was not significant. The AUV 
model explained 46.7% of the deviance with an r” of 0.52 
(Table 2). The TV model had significant effects for RVA, 
vehicle range, transect number, and habitat complexity. 
The TV model explained 34.0% of the deviance with an r” 
of 0.22. The ROV model had significant effects for RVA, 
vehicle range, transect number, and habitat complexity. 
The ROV model explained 13.9% of the deviance with an r” 
of 0.10. For the benthic pursuers guild, relative abundance 
increased with decreasing vehicle range to the MOUSS 
sampling volume, peaking at a distance of around 0 m and 
remaining at elevated levels following vehicle passage 
(Fig. 6). In general, relative abundance of this group also 
increased as a function of transect number with varying 
results relative to vehicle type. For both slow-transiting 
vehicles (AUV and ROV), increasing RVA resulted in 
increased relative abundance; whereas, with the TV rela- 
tive abundance decreased with increasing RVA. 
From a qualitative perspective, the lateral movers group 
tended to flee laterally as the vehicle neared the MOUSS 
platform sampling volume (Fig. 7). For the lateral movers 
guild, results from the TV model indicate significant 
