Campbell et al.: Attraction and avoidance behaviors of fish in response to proximity of underwater vehicles 217 
Patterson et al., 2009; Armstrong and Singh, 2012; Grasty, 
2014; Bacheler and Shertzer, 2015). Attractive aspects of 
optical sampling gears include that they 1) do not require 
extraction of fish, 2) have negligible impact on the habi- 
tat, 3) collect habitat data, and 4) often allow calculation of 
sampling volume or area and therefore can be used to pro- 
duce habitat-specific density estimates that can be scaled 
to calculate abundance (Royle et al., 2009; Trenkel and 
Lorance, 2011; Whitmarsh et al., 2017). 
An artifact of many fisheries survey methods is that 
the unit of measurement is an observation rate or catch 
rate and therefore cannot be used to estimate fish densi- 
ties (e.g., fish per unit area). Collection of rate data con- 
strains analysts to development of relative abundance 
indices (Williams et al., 2018) that are rarely calibrated 
against known fish densities and therefore cannot pro- 
vide data to calculate absolute abundance. In addition, 
gear bias can affect counts; for example, observations and 
counts of fish are negatively affected during scuba-diver 
surveys because of air bubbles from exhalation (Gray 
et al., 2016; Emslie et al., 2018). Using vehicles to conduct 
optical surveys is a popular sampling method because 
they can cover long distances while transiting, allowing 
estimation of area sampled and fish densities, but they 
too will have inherent sampling biases (Clarke et al., 
2009; Stierhoff et al., 2013; Thanopoulou et al., 2018). 
Estimates of fish densities are useful because they 
can be scaled over the area of known habitat to calcu- 
late absolute abundances for use in stock assessments 
(O’Connell and Carlile, 1993; Yoklavich et al., 2007). 
Although calculation of fish density might seem like a 
straightforward process, estimates can be affected by 
environmental conditions and fish behavior (Fig. 1). 
Attraction and avoidance 
functions 
Sighting function 
Fish density 
water-quality —> 
modifiers 
Distance 
Figure 1 
Diagram showing the theoretical effects of attraction and 
avoidance (dashed line above the sighting function) and 
sighting (solid line) functions on density estimation. The 
dashed line under the sighting function represents how 
fish density is theoretically modified by water quality, 
species, and size of the individuals observed. 
Therefore, it is critical to estimate sighting functions 
and attraction and avoidance functions when con- 
ducting distance sampling and strip-transect surveys 
(Sale and Sharp, 1983; Ensign et al., 1995; Cheal and 
Thompson, 1997). Importantly, the need for estimating 
such functions has been shown to be true even for spe- 
cies with minimal or no avoidance behaviors (Kulbicki 
and Sarramégna, 1999). 
As with any sampling gear, fish are likely to respond 
to the presence of stationary platforms and vehicles, 
and those responses have been recognized as a potential 
source of bias (Uzmann et al., 1977; Jagielo et al., 2008; 
Stoner et al., 2008). Each of these vehicles has its own 
characteristics related to movement speed, deployment 
altitude, acoustic signature, size, and visibility, all of 
which can introduce bias into the data collected (Koslow 
et al., 1995; Lorance and Trenkel, 2006; Stoner et al., 
2008). In addition, fish are apt to respond to novelties 
in the environment in different ways in accordance with 
survival and foraging needs (Olla et al., 1998) and per- 
haps out of curiosity. Regardless of the underlying moti- 
vation, species-specific responses to sampling gears need 
to be quantified in order to deal with underlying biases 
and to generate reliable count, density, and abundance 
estimates. 
Classes of vehicles commonly used in marine research 
include remotely operated vehicles (ROVs), autonomous 
underwater vehicles (AUVs), towed vehicles (TVs), and 
human occupied vehicles. Vehicles are used in sampling 
efforts predominately as part of line-transect methods, 
but the engineering specifications of vehicles and the exe- 
cution of the line-transect surveys differ. These vehicle- 
specific differences can result in measurement biases of 
unknown direction and magnitude, the result of which is 
a need for gear calibration (Clarke et al., 2009). Further, it 
is recognized that, within a vehicle class, there will likely 
be many variations and exceptions (Yoklavich et al., 2015). 
Potential stimuli that could elicit attraction to, or avoid- 
ance of, vehicles include transit speed and altitude, visual 
profiles, and acoustic signatures. Ideally the specific stim- 
uli causing a reaction can be identified, but more impor- 
tantly fish responses in general need to be evaluated and 
quantified as a first step in understanding how to develop 
gear-calibration methods. 
Because of the elevated interest in the use of vehicles to 
conduct surveys of high-relief, complex bottom types, the 
National Marine Fisheries Service (NMFS) initiated the 
Untrawlable Habitat Strategic Initiative (UHSD in 2014. 
Participants in the UHSI were tasked with designing a 
multitiered field experiment to evaluate the sampling 
efficiency of camera systems mounted on stationary plat- 
forms, ROVs, AUVs, and TVs used to count fish and 
invertebrates in a sampling area or volume. The analysis 
in this study was focused on the change in abundance of 
reef fish species in a sampling volume due to the passage 
of a mobile survey vehicle. Our intent was to develop a 
functional relationship between vehicle range from the 
sampling volume and relative change in fish abundance 
(Fig. 1). 
